Crystal Structure of an Affinity-matured Prolactin Complexed to Its Dimerized Receptor Reveals the Topology of Hormone Binding Site 2*

We report the first crystal structure of a 1:2 hormone·receptor complex that involves prolactin (PRL) as the ligand, at 3.8-Å resolution. Stable ternary complexes were obtained by generating affinity-matured PRL variants harboring an N-terminal tail from ovine placental lactogen, a closely related PRL receptor (PRLR) ligand. This structure allows one to draw up an exhaustive inventory of the residues involved at the PRL·PRLR site 2 interface, consistent with all previously reported site-directed mutagenesis data. We propose, with this description, an interaction model involving three structural components of PRL site 2 (“three-pin plug”): the conserved glycine 129 of helix α3, the hydrogen bond network involving surrounding residues (glycine cavity), and the N terminus. The model provides a molecular basis for the properties of the different PRL analogs designed to date, including PRLR antagonists. Finally, comparison of our 1:2 PRL·PRLR2 structure with those of free PRL and its 1:1 complex indicates that the structure of PRL undergoes significant changes when binding the first, but not the second receptor. This suggests that the second PRLR moiety adapts to the 1:1 complex rather than the opposite. In conclusion, this structure will be a useful guiding tool for further investigations of the molecular mechanisms involved in PRLR dimerization and activation, as well as for the optimization of PRLR antagonists, an emerging class of compounds with high therapeutic potential against breast and prostate cancer.

Competitive prolactin (PRL) 5 receptor antagonists have recently emerged as a new class of molecules with high therapeutic potential for treating unresolved PRL-sensitive pathologies (1)(2)(3). The main indications include hyperprolactinemia resulting from dopamine-resistant prolactinomas and tumors affecting peripheral PRL-target tissues such as the prostate and the breast. In the latter cases, PRL receptor (PRLR) antagonists are currently viewed as one of the most promising approaches to counteract the undesirable growth promoting actions mediated by locally produced PRL (4 -7) or by constitutively active PRLR mutants (8).
PRL is a member of a long established hormone family, also including growth hormone (GH) and placental lactogen (PL), with which it shares high structural homology (9,10). The biochemical, structural, and dynamic properties of GH have been well characterized (11)(12)(13). The determination of the x-ray structure of human GH (hGH) bound to the dimerized extracellular domain (ECD) of its receptor (13) was a crucial step toward the understanding of GHR activation mechanism (14). This structure identified two asymmetric binding sites on hGH, called sites 1 and 2, which were proposed to trigger hormoneinduced sequential dimerization of the GH receptor, leading to activation (11). More recently, this model has been revisited based on the comparison of unliganded and liganded human GHR-ECD (12); it is now believed that GHR exists as an inactive dimer at the cell membrane, which is activated by hormoneinduced relative rotation of subunits within a dimeric receptor (12). In addition to the characterization of the structural changes occurring in GH and GHR upon their interaction, the GH⅐GHR 2 three-dimensional structures have also largely contributed to the development of GHR antagonists. This novel class of molecules consists of modified GH-core proteins, which harbor one steric mutation within binding site 2, namely the replacement of the helix ␣3 Gly with long and charged residues such as Lys or Arg. Initially introduced in the bovine hormone (G119R-bGH) (15), the Gly substitution in the human hormone was also shown to generate an antagonist (G120K/R-hGH), and the crystal structure of the hGH⅐hGHR 2 complex revealed that the Arg/Lys could actually impede the formation of a functional receptor dimer (14). Subsequently, these studies have resulted in the approval of a GHR antagonist, Somavert (Pegvisomant for injection), as a drug for the treatment of acromegaly (16).
In the absence of any three-dimensional structure of PRL and more importantly of PRL⅐PRLR complexes, we initially took advantage of the homology between PRL and GH systems to engineer the first PRLR antagonists. Based on the paradigms established from the GH⅐GHR interaction, our working hypothesis was that PRL also induced PRLR dimerization by interacting with two receptor molecules via two distinct binding sites. This assumption was supported by mutational studies (10,17,18) and by the observation that mutation of the conserved helix ␣3 glycine (Gly 129 ) also generated PRLR antagonists (19,20). However, despite obvious similarities, GH⅐GHR and PRL⅐PRLR also appeared to have their own distinctive features. For example, the G129R-hPRL antagonist was shown to exhibit significant residual agonistic properties in various PRLR-mediated bioassays (3), whereas, to the best of our knowledge, there are no similar observations for G120R-hGH in GH-responsive bioassays. In addition, although mutation of Gly 120 in GHR antagonists does not affect binding affinity for GHR at the cell surface, every mutation of Gly 129 tested so far in hPRL was shown to decrease 10-fold their affinity for cell-expressed PRLR, which is clearly detrimental to the efficacy of these antagonists (2,20).
Understanding the molecular features responsible for these hormone specificities has been repeatedly hampered by the lack of a crystal structure of the 1:2 PRL⅐PRLR 2 complex (21). Indeed, the first structural data reported for the PRL system involved PRLR-ECDs bound to non-PRL ligands: a 1:1 complex between hGH and the hPRLR (22), and a 1:2 complex between ovine PL (oPL) and the rat (r) PRLR (23). As for hPRL, its structure in solution was obtained by NMR, confirming its four antiparallel ␣-helix bundle fold (24). More recently, we reported the first set of x-ray coordinates for a PRL-core hormone and determined the structure of the pure antagonist Del1-9-G129R-hPRL (20), and the structure of this compound bound to the hPRLR-ECD (1:1 complex) was then determined (25).
Although titration of hPRL by hPRL-ECD in NMR experiments suggested changes in chemical shifts at high receptor/ hormone ratio (24), the lack of structural data for the 1:2 complex between PRL and its receptor has remained detrimental to our understanding of this hormonal system. For example, it has been impossible to this date to evaluate the putative structural changes occurring in the intermediate 1:1 hormone⅐receptor complex upon interaction with the second receptor. Also, the ligand-receptor interactions involving binding site 2, which is the targeted area for PRLR antagonists generation (3), remain structurally uncharacterized.
One of the recurrent problems encountered for the determination of the PRL⅐PRLR 2 structure is the poor stability of this ternary complex in vitro. Although the formation of 1:2 complexes has been demonstrated by various approaches, including gel filtration, native electrophoresis, and surface plasmon resonance, interaction at binding site 2 appears to be too transient to stabilize ternary complexes (20,24,26). To circumvent this problem, we aimed at generating an affinity-matured hPRL variant able to form stable 1:2 complexes. Because the structure of oPL⅐PRLR 2 suggested that the elongated N terminus of oPL could contribute to the binding of the second receptor moiety (23), we iteratively substituted the N-terminal residues of hPRL by their homologs from oPL. We here report that this strategy was successful in increasing significantly hPRL binding site 2 affinity and, thereby, stabilizing the PRL⅐PRLR 2 ternary complex. This allowed us to solve the crystal structure of the 1:2 complex involving the most potent affinity-matured hPRL and the rPRLR-ECD.

EXPERIMENTAL PROCEDURES
Site-directed Mutagenesis-N-terminal mutations (replacements and elongations) were performed using the QuikChange II mutagenesis kit from Stratagene (La Jolla, CA). Mutations were introduced iteratively using the pT7L-hPRL expression plasmid as initial template (27). Forward and reverse (complementary) primers as well as iterative templates are shown in Table 1. After transformation, Escherichia coli BL21(DE3) colonies were analyzed for their DNA content; plasmids were sequenced to verify the presence of the expected mutations. Sequences encoding the ECD of human or rat PRLR were inserted into the pQE-70 expression plasmid (Qiagen) containing a His 6 tag at the C-terminal end as described (20,24). Subcloning constraints led to the addition of 4 amino acids just before the His 6 tag (Gly 211 -Ser 212 -Arg 213 -Ser 214 for hPRLR-ECD and Arg 211 -Ser 212 -Arg 213 -Ser 214 for rPRLR-ECD).
Production of Recombinant Proteins-Recombinant hPRL (WT and variants) and PRLR ECDs were overexpressed in 0.5to 1-liter cultures of E. coli BL21(DE3) or M15(REP4), respectively, and purified as described previously (20). Briefly, pro-  Nter  hPRL  FOR 5Ј GA GAT ATA CAT ATG GCA CAG CAT CCA CCA TAC TGT CCC GGC GGG G 3Ј  REV 5Ј C CCC GCC GGG ACA G TA TGG TGG ATG CTG TGC CAT ATG TAT ATC TC 3Ј  T14P  hPRL  FOR 5Ј CGA TGC CAG GTG CCT CTT CGA GAC CTG TTT GAC C 3Ј  REV 5Ј GGT CAA ACA GGT CTC GAA GAG GCA CCT GGC ATC G 3Ј  PGGA  Nter  FOR 5Ј CAT CCA CCA TAC TGT CGA AAC CAG CCA GCC CGA TGC CAG GTG 3Ј  REV 5Ј CAC CTG GCA TCG GGC TGG CTG GTT TCG ACA GTA TGG TGG ATG 3Ј  PGGA-T14P  PGGA  FOR 5Ј CGA TGC CAG GTG CCT CTT CGA GAC CTG TTT GAC C 3Ј  REV 5Ј GGT CAA ACA GGT CTC GAA GAG GCA CCT GGC ATC G 3Ј  Full oPL  PGGA-T14P  FOR 5Ј GT CGA AAC CAG CCA GGC AAA TGC CAG ATC CCT CTT CGA GAC C 3Ј  REV 5Ј G GTC TCG AAG AGG GAT CTG GCA TTT GCG TGG CTG GTT TCG AC 3Ј teins were overexpressed as insoluble inclusion bodies, which were solubilized in 8 M urea (5 min at 55°C, then 2 h at room temperature) and refolded by continuous dialysis (72 h, 4°C) against 100 volumes of 25 mM NH 4 HCO 3 , pH 8.6. Solubilized proteins were then loaded onto a HiTrapQ anion-exchange column (Amersham Biosciences) equilibrated in 25 mM NH 4 HCO 3 , pH 8.6. Human PRL (WT and variants) and receptor ECDs were eluted along a NaCl gradient (0 -500 mM), and the major peak was collected, quantified, and kept frozen until use. Purity of the various hPRL variant/receptor ECD batches was Ͼ95% as judged from SDS-PAGE analysis.
Binding assays were performed at least in triplicate as previously described (20). Briefly, for site 1 characterization, the hPRL variants (concentrations ranging from 0.78 to 200 nM) were injected onto the PRLR-ECD surfaces. For sites 2ϩ3 characterization, the PRLR-ECD surface was first saturated with each hPRL variant (350 nM). Rat or human PRLR-ECDs (concentrations ranging from 23 nM to 28.6 M) were then injected for 8 min onto the pre-formed PRLR-ECD⅐PRL complexes, followed by a 5-min dissociation period. No nonspecific signal could be detected when injecting PRLR-ECD over the PRLR-ECD surface in the absence of hPRL.
The association and dissociation profiles were double-referenced using the Scrubber 2.0 software (BioLogic Software), i.e. both the signals from the reference surface (ethanolamine derivatized) and from blank experiments (using running buffer instead of protein) were subtracted. The steady-state SPR responses (R eq , experimental or extrapolated) were plotted against the concentration (C) of the PRL variant (site 1) or of the PRLR-ECD (site 2) and fitted using the following equation, R eq ϭ (R max * C)/(K d ϩ C), where K d is the equilibrium dissociation constant, and R max the maximal binding capacity of the surface. Kinetic parameters (k on and k off ) were determined using a nonlinear least squares algorithm implemented in the BIAevaluation 4.1 software (Biacore). Further details on the equations used for data fitting are provided as supplemental information.
Analytical Ultracentrifugation-Sedimentation velocity experiments were carried out at 25°C in an XL-I analytical ultracentrifuge (Beckman-Coulter) equipped with double UV and Rayleigh interference detection. Samples were prepared in 25 mM NH 4 HCO 3 , 150 mM NaCl, pH 8.6, and spun using an An60Ti rotor and 12-mm double-sector aluminum centerpieces. The partial specific volume of hPRL Full-oPL (0.736 ml.g Ϫ1 ), rPRLR-ECD (0.728 ml.g Ϫ1 ), and their complexes (0.732 ml.g Ϫ1 ) were estimated from their amino acid sequences using the software Sednterp 1.09 (available on-line from The Boston Biomedical Research Institute). The same software was used to estimate the buffer viscosity ( ϭ 1.002 cP) and density ( ϭ 1.0032 g.ml Ϫ1 ). hPRL Full-oPL (300 l at 20 M) and rPRLR-ECD (300 l at 20 M) were spun at 50,000 rpm, whereas the hPRL Full-oPL⅐rPRLR -ECD mixture (300 l at 5 M/15 M) was spun at 36,000 rpm. Absorbance and interference profiles were recorded every 3 min. Sedimentation coefficient distributions c(s) were determined using the software Sedfit 11.3 (28). Theoretical sedimentation coefficients of the complexes were calculated from the Prl complex PDB file using Hydropro 7c (29) with a hydrated radius of 3.1 Å for the atomic elements.
Crystallization and X-ray Diffraction Data Collection-Initial crystallization screening was performed in 96-well sitting drop crystallization plates (Greiner Bio-One) using a Cybi-Disk robot from Cybio. Crystallization screens were set up using several commercially available high throughput crystallization screening kits (Hampton Research). Small diamond-shaped crystals appeared after 1 month at 18°C in condition 25 of the MemFac crystallization kit. The crystallization condition was optimized manually in Linbro plates using the hanging drop method. Crystals of 40 ϫ 40 ϫ 80 m 3 final size were recorded at European Synchrotron Radiation Facility (ESRF, Grenoble, France) on ID14 beam line at 3.5-Å resolution. They belonged to space group P2 1 2 1 2 with unit cell dimensions of a ϭ 178.31 Å, b ϭ 59.61 Å, c ϭ 72.43 Å. Unfortunately all our efforts to refine the structure were not sufficient to reach acceptable statistics. This was probably due to the low completeness of the data set (80%), because crystals died very rapidly.
A last check of the original Greiner plates after 1 year showed large crystals in different conditions of both MemFac and Index crystallization kits. All of these were tested on the ID14-4 beam line at ESRF without addition of cryoprotectant before flashfreezing in liquid nitrogen. Half of the crystals were protein crystals, but only one did not display high mosaicity and could provide x-ray data. The crystallization reservoir was composed of 100 mM sodium citrate, pH 5.6, 100 mM lithium sulfate, 12% polyethylene glycol 4000, which corresponds to the MemFac no. 13 condition. A complete data set was recorded at 3.7-Å resolution. Despite several attempts to optimize manually the crystallization conditions, none of the newly recorded data sets diffracted better than this one. The recorded images were reduced, scaled, and merged with programs MOSFLM and SCALA (30,31). The intensities were then converted to the structural factor amplitudes with TRUNCATE (30,31). This new crystal belonged to space group P4 3 2 1 2 with unit cell dimensions of a ϭ b ϭ 92.12 Å, c ϭ 215.85 Å. This crystal is referred to as Prl complex in the text.
Phase Determination and Structure Refinement-The crystal structure of Prl complex was solved by molecular replacement using the program PHASER (32), with the 1:2 complex structure of oPL⅐rPRLR 2 (Protein Data Bank code 1F6F) as search model (23). The complete complex was easily positioned in one block in the unit cell, and the position of each domain was refined by rigid body refinement using CNSv1.1 (33).
Refinement of the structure at 3.8-Å resolution was carried out by multiple cycles of manual rebuilding using the program Coot (34) and refinement using CNSv1.1, resulting in a final model with a R factor of 28.1% and an R free factor of 38.2%. These values all together with poor geometry quality led us to proceed to a second structure refinement procedure. The structure issued from the CNS rigid body refinement was used as the first model. This second structure refinement was performed with Phenix (35). The model was build in O (36) with the lego algorithm using structure fragments from the high res-olution rebuild data base. Crystallographic refinement was accomplished using the maximum likelihood target with amplitudes. No I/sigma cutoff was applied as weak reflections with large experimental error estimates are automatically downweighted in the likelihood-based target function. The main improvement in map quality was due to the new bulk solvent procedure that is more robust than in CNS v1.1 due to the grid search to optimize the bulk-solvent parameters ksol and Bsol. Because the number of reflections was not enough to resolve the system, atomic displacement parameters were refined by groups of five residues and periodically reset to the average value.
After several rounds of manual model building in O and refinement in Phenix, the final R and free R factors were lowered to 25.2% and 32.3%, respectively. The geometry is of good quality with 88.9% and 8.9% of the residues in the most favored and allowed regions of the Ramachandran plot, respectively. These values are in accordance with the validation criteria expected at this resolution, especially for structures with only one molecule per unit cell (37).
A summary of the crystallographic data and refinement statistics is given in Table 2. An example of the map quality is presented in supplemental Fig. S1. The refined structure was validated using the program MolProbity (38). The characterization of the secondary structure elements was performed using programs Molscript (39) and Stride (40). The figure panels showing three-dimensional structures were generated using the PyMOL Molecular Graphics System. 6

Rationale for the Design of hPRL Variants
We focused our mutational strategy on the N-terminal tail of hPRL based on two experimental observations. First, the N terminus of oPL was shown to account for a large part of the energy of interaction with the second PRLR-ECD (23), indicating that it represents an important feature of binding site 2. Second, deletion of the nine N-terminal residues in G129R-hPRL abolished the residual agonistic activity of this partial antagonist, further highlighting the functional role of the PRL N terminus in the receptor activation process (2,20).
Interestingly, the N-terminal tail is the most divergent region among PRL/GH/PL hormones (10), and non-primate PLs dis-play the longest N terminus (three additional residues compared with PRLs). Sequence comparison shows that three main features distinguish the N-terminal sequences of hPRL versus oPL (Fig. 1). The first difference is the presence of a proline right before the helix ␣1 of oPL (Pro 14 , hPRL numbering); variant hPRL T14P was generated to test its potential role. The second difference is the AQHPPY motif located before the first cysteine (Cys 4PRL ), which elongates the N terminus of oPL by three residues compared with hPRL. Because this AQHPPY motif was identified as an important feature of binding site 2 of oPL (23), the hPRL Nter variant was generated by substituting AQHPPY for LPI. The third difference is the short stretch forming the loop constrained on its extremities by the Cys 4PRL -Cys 11PRL disulfide bond. With the exception of a positive charge in position 10 (Arg PRL /Lys oPL ), this sequence is not conserved and shifts a proline from position 5 in hPRL (PGGA) to position 8 in oPL (RNQP). Variant hPRL PGGA substitutes RNQP for PGGA in hPRL Nter , and hPRL PGGA,T14P also includes the T14P replacement. Finally, the so-called hPRL Full-oPL variant was obtained by substituting the entire oPL N terminus (17 residues) for the native hPRL N-terminal sequence. With respect to hPRL PGGA,T14P , hPRL Full-oPL includes three additional replacements: Ala 9 3 Gly, Arg 10 3 Lys, and Val 13 3 Ile.

Structural and Functional Characterization of hPRL Variants
The five hPRL variants were produced and purified with similar yields as hPRL. CD experiments showed that the secondary structure and the thermodynamic parameters for heat denaturation of the different variants were close to those of hPRL (see supplemental Table 1), indicating that the N-terminal mutations did not modify significantly the folding or stability of the hormone. N-terminal mutations also had only a minimal impact on the global affinity for membrane receptors deter-  mined by cell-based radio receptor assays, with at most a 2-fold increase in the affinity of elongated variants for both rat and human PRLR (see supplemental Table 2). Accordingly, the biological activity of hPRL variants was not significantly affected, with at most a 4-fold bioactivity decrease for variant hPRL T14P in the bioassay involving the human receptor (see supplemental Table 3).

Identification of Affinity-matured hPRL Variants and Characterization of Their Complexes
To decipher more specifically the impact of the mutations on site 1 and/or sites 2ϩ3 binding, we used the SPR methodology that we recently set up to monitor sequentially the interaction of hPRL with two separate PRLR ECDs (20). Table 3 summarizes the equilibrium dissociation constants (K d ) of the complexes formed by the different N-terminal variants with the rat or the human PRLR-ECD. For the rPRLR-ECD, site 1 affinities of hPRL T14P , hPRL Nter , and hPRL PGGA were close to that of hPRL WT , whereas those of hPRL PGGA,T14P and hPRL Full-oPL were, respectively, 3-fold and 8-fold higher, mostly due to a 5-fold decrease of the dissociation rate k off , from 10 Ϫ3 s Ϫ1 for hPRL WT to 2 ϫ 10 Ϫ4 s Ϫ1 for hPRL Full-oPL (Fig. 2). As for their sites 2 plus 3 affinities, with the exception of hPRL T14P , significant increases were observed for all variants, from 3-fold for hPRL Nter and hPRL PGGA to 30-fold for hPRL PGGA,T14P and hPRL Full-oPL (Table 3, Fig. 2, and supplemental Figs. S2 and S3). These very significant improvements could be attributed both to a stabilization of the site 2 interaction and to a faster formation of ECD-ECD contacts (site 3), leading to a significant increase of the half-life of the 1:2 complex formed by hPRL Full-oPL with respect to hPRL WT (see supplemental "Results"). For the human PRLR-ECD, only marginal site 1 or sites 2ϩ3 affinity improvements were observed, 3-fold and 2-fold, respectively, for hPRL Full-oPL (Table 3).
These promising results prompted us to investigate whether hPRL Full-oPL and rPRLR-ECD formed stable 1:2 complexes in solution. A mixture of hPRL Full-oPL and rPRLR-ECD (ratio 1/2) indeed appeared to elute mostly as a 1:2 complex in size-exclusion chromatography (data not shown). Analytical ultracentrifugation experiments were conducted to further characterize the hPRL Full-oPL ⅐rPRLR-ECD complex (Fig. 3). Both hPRL Full-oPL and rPRLR-ECD sedimented as single species compatible with a monomeric form, with respective sedimentation coefficients of 2.4 Ϯ 0.2 S and 2.5 Ϯ 0.1 S. The hPRL Full-oPL ⅐rPRLR-ECD mixture (1/3 ratio), was then analyzed, revealing, apart from residual free species, a larger species with a sedimentation coefficient of 5.0 Ϯ 0.3 S (compatible with a 1:2 complex).

Global Description of the Prl Complex Structure and Comparison with Other Hormone⅐Receptor 1:2 Complexes
Having shown that hPRL Full-oPL and rPRLR⅐ECD could indeed form relatively stable 1:2 complexes, co-crystallization trials were carried out leading to the resolution of the 1:2 hPRL Full-oPL /(r(PRLR1⅐PRLR2)-ECD) complex structure referred to as Prl complex in the text.
The hydrodynamic characteristics of the Prl complex were calculated from the crystal structure, giving a hydrodynamic radius of 3.8 nm and a sedimentation coefficient of 4.99 S, which is in excellent agreement with the analytical ultracentrifugation experimental value (5.0 Ϯ 0.3 S). The dimensions and shape of the complex are therefore consistent with that of the complex in solution.
As expected, the structure is globally similar to the 1:2 x-ray structures available for the other members of the hormone/ receptor family reported in the PDB (1HWG: hGH⅐hGHR 2 ; 3HHR: hGH⅐hGHR 2 ; 1F6F: oPL⅐rPRLR 2 ; and 1KF9: hGH⅐ hGHR 2 ). The hormone consists of four core ␣-helices and two long overhand loops, which correspond to the classic longchain cytokine fold. The two receptors are each divided in two fibronectin type-III domains termed D1 and D2. The D1 (0 -98) and D2 (104 -205) domains are mainly composed of seven ␤-strands that form a sandwich of two anti-parallel ␤-sheets (ABE and DCFG, see Fig. 4). In D1 domains, most of the strands The native N terminus of hPRL was iteratively modified to mimic that of oPL (elongation/mutation). Numbering corresponds to hPRL sequence, numbering for elongated mutants starts from Ϫ2 as indicated. The initial methionine Met (Ϫ3) , found in recombinant proteins, is not indicated. are shorter than in D2, or split in two strands, but the structure is stabilized by the presence of the two disulfide bonds (Cys 13 -Cys 23 and Cys 52 -Cys 63 ). In the D1 domain of PRLR1, the N terminus (Met 0PRLR1 -Lys 6PRLR1 ) seems to be highly flexible. In the 1:2 oPL⅐rPRLR 2 structure complex, it is bent at residue Lys 6PRLR1 , just at the end of strand AЈ, and thus oriented toward loop 1 PRL . In the Prl complex structure it forms an additional ␤-strand AЈ (Pro 4PRLR1 -Glu 8PRLR1 ) (Fig. 4) aligned with strand A, which occupies a space between the two small strands BЈ and GЈ pushing them apart but without forming a ␤-sheet. This particular N terminus conformation is also observed in the 1:1 hPRL⅐hPRLR complex (3D45) (25).

TABLE 3 Equilibrium dissociation constants of the complexes between the hPRL variants and PRLR-ECD (rat or human)
The only few interactions between domains D1 and D2 involve the interdomain ␤-strand with loops L1 and L6 forming a hinge around which the two domains can rotate. In PRLR1, this hinge is bordered on one side by interactions between Arg 13PRLR1 and both Glu 102PRLR1 and Tyr 190PRLR1 (just above the conserved WSXWS 195 sequence) and on the other side by contacts between Trp 139PRLR1 and both Trp 72PRLR1 and Lys 17PRLR1 . The three latter residues are all part of binding site 1. In PRLR2, only some of these interactions are conserved. Arg 13PRLR2 is slightly far-off but Glu 102PRLR2 and Tyr 190PRLR2 are still at van der Waals distances. On the other side of the hinge, as loop L5 is not seen in the electronic density, the contacts are completely different. They are replaced by van der Waals' contacts between Tyr 99PRLR2 , which is part of binding site 2, with His 188PRLR2 . On both receptors, the movement between D1 and D2 seems to be governed by residues already involved in hormone-receptor interactions. Consequently, the difference in sequence between the two hormones hPRL and oPL at residues involved in the two binding sites is presumably sufficient to apply a difference in the orientation of the receptors domains. This is in accordance with a conformational change following hormone binding. The hormone interacts with the two receptors in an asymmetric manner. Binding site 1 is a flat surface of ϳ1180 Å 2 , which is slightly larger than in the oPL⅐PRLR 2 complex (ϳ1000 Å 2 ), but smaller than in hGH⅐hGHR 2 complex (ϳ1300 Å 2 ). It is closely similar to that in the recently determined structure of the 1:1 hPRL⅐hPRLR complex (25), although 7 of 28 residues involved in the binding interface differ between human and rat PRLR (supplemental Fig. S4). Only three substitutions at positions 46, 73, and 74 appear to lead to local modifications of the interaction network sufficient to explain the shift of region Thr 52PRL -Asn 56PRL . This region, which is part of loop 1 PRL (Tyr 44 -Asn 56 ), was shown for hGH to adopt a defined structure only when the receptor is bound (13). Only Thr 52PRL -Asn 56PRL is visible in the Prl complex electron density. It adopts a different conformation in the complex than in free PRL, but without forming well defined secondary structure elements.
Finally, the Prl complex globally shows the same receptor-receptor interface (also called stem-stem) as described in the 1:2 complex structure involving rPRLR with oPL (1F6F). It is mainly governed by van der Waals' contacts. Nevertheless, due to tighter packing in this region, an additional loop has been built in Prl complex (Glu 112PRLR1 -Lys 120PRLR1 ) (Fig. 4). Functional analysis of this region is ongoing based on this structural data.

Binding Site 2
Our SPR experiments showed that binding of PRLR to hPRL Full-oPL site 2 was significantly less tight than at binding site 1. Accordingly, the surface of the PRL⅐PRLR2 interface was ϳ700 Å 2 , close to that reported in 1:2 complexes involving the two other ligands (oPL and hGH). In our Prl complex structure, binding site 2 can be divided into two structural regions, which share residue Asp 96PRL (Fig. 5, A and C): the glycine cavity and the N terminus.

The Glycine Cavity
The cavity is made of two individual components: Gly 129PRL itself, which forms the bottom of the pocket, and large surrounding amino acid residues, which form the walls. Upon binding, this cavity is filled with Trp 72PRLR2 (equivalent to Trp 104GHR2 ). This structural feature is a hallmark of the PRL/GH binding site 2 (Fig. 5, A and B), and its functional importance is highlighted by the fact that all substitutions of Gly 129PRL and its GH and oPL homologs prevent receptor docking, leading to receptor antagonists (3,14,20). The walls of the cavity involve residues of ␣-helices 1 and 3 that interact with each other and/or with residues of PRLR2 to form a hydrogen bond network that locks the interaction. The comparison of the three complexes indicates that most of the interacting residues are topologically equivalent, although not strictly sequence conserved (Figs. 5A and 6). Despite the fact that oPL⅐PRLR 2 and PRL⅐PRLR 2 complexes are more similar to each other than to the GH⅐GHR 2 complex with respect to primary and tertiary structures, the hydrogen bond network is maintained in the three 1:2 complexes. This network cannot be described at the atomic level in the Prl complex due to the low resolution of the structure. However, it allows the determination that Asp 17PRL and Arg 21PRL are in close proximity with Asp 96PRLR2 and Thr 98PRLR2 , and thus might be involved in tight interactions. These  interactions are conserved in the GH complex, between Arg 8GH and Ser 124GHR2 , Asn 12GH , and Asp 126GHR2 . In the oPL complex, Ser 20oPL and Arg 24oPL are not in contact with each other, but the interaction between Arg 24oPL and both Asp 96PRLR2 and Thr 98PRLR2 are maintained. This side of the cavity wall is thus made of four residues involved in conserved interactions in the three complexes although residue pairs are different. There is one additional interaction in the Prl complex between Asp 17PRL and Tyr 99PRLR2 , which is observed neither in the oPL complex (due to the shorter side chain of Ser 20oPL ) nor in the GH complex (due to steric hindrance between Arg 8GH and Glu 127GHR2 ).
The other side of the wall is less conserved. In the oPL complex there is a hydrogen bond between Lys 124oPL and Glu 18PRLR that could also exist in the Prl complex between Arg 125PRL and Glu 18PRLR as the two residues are in a pre-disposed orientation. But in the GH complex, the equivalent hormone residue (Asp 116GH ) interacts with Trp 104GHR2 instead of Glu 44GHR2 . A direct interaction with Trp 72PRLR2 is also present in the oPL complex, but it involves the carbonyl of Val 127PL . Additional interactions strengthening site 2 were also noticed in the GH complex only. These involve a hydrogen bond between Asn 12GH and Arg 43GHR2 and between Glu 119GH and Ser 102GHR2 (beyond the central tryptophan residue), a stacking contact between Arg 16GH and Trp 169GHR2 (Loop L5), and a van der Waals' interaction between Tyr 103GH (loop between ␣-helices 2 and 3) and Ile 165GHR2 (loop L5) (not shown on Fig. 5B). Due to the absence of electronic density at the level of loop L5 in PRLR2, and to the very different structure of the loop between ␣-helices 2 and 3 in the hormones, these interactions were not observed in the Prl complex. This may partly account for the markedly lower affinity of the second receptor monitored by SPR in the hPRL⅐rPRLR 2 complex (11 M) compared with the hGH⅐hGHR 2 complex (3.8 nM) (41).
In summary, hydrogen bonds surround the tryptophan in both hGH⅐hGHR 2 and oPL⅐rPRLR 2 complexes. In the Prl complex, the equivalent residues are in close proximity compatible with tight interactions on one side of the tryptophan, and a little more spaced out on the other side even if their respective positions could predispose them for interacting.

The N Terminus
Comparison with the oPL⅐rPRLR 2 Complex-The second structural region of binding site 2 involves the N terminus of the hormone (Fig. 5,  C and D). Because the affinity-matured hPRL Full-oPL that we crystallized harbored the oPL N terminus, one would have expected the PRL and oPL complexes to exhibit the same interactions at this level, which is not exactly the case (Fig. 5C). There are three hydrogen bonds in the oPL⅐rPRLR 2 complex, between Tyr 3PRL and Asp 96PRLR2 , between Pro 92PRLR2 carbonyl and Ala 1oPL main chain, and between Cys 14PL (equivalent to Cys 11PRL ) and Glu 43PRLR2 . The latter cannot occur in the Prl complex due to the very different position of Glu 43PRLR2 . Furthermore no density could be observed at the level of the N-terminal Ala of hPRL Full-oPL . Consequently, only Tyr 3PRL is close enough to interact with the receptor (residue Asp 96PRLR2 ) like in the oPL complex.
Comparison with the hGH⅐GHR 2 Complex-The N terminus of hGH is also involved in the binding to GHR 2 , although only hydrophobic interactions appear to be involved (Fig. 5D). Pro 2GH is at van der Waals' distance from Pro 106GHR2 , as are Phe 1GH from Arg 71GHR2 , and Cys 122GHR2 and Ile 4GH from Ser 124GHR2 . The latter interaction is conserved in the two structures involving PRLR (I1e 3PRL -Asp 96PRLR2 ). It is noteworthy that, although the N terminus of hPRL (ϩ9 residues) and oPL (ϩ12) is much longer than that of hGH (Fig. 5E), most of the extra residues are involved in the small N-terminal loop constrained on each extremity by the disulfide bond between Cys 4 FIGURE 5. Important binding site 2 contact residues in the complexed hPRL, oPL, and hGH hormones. A and B correspond to the glycine binding pocket. C and D correspond to the hormone N terminus binding interface. In A and C, the superimposition of hPRL Full-oPL ⅐rPRLR 2 and oPL⅐rPRLR 2 complex structures are represented. In B and D the hGH⅐hGHR 2 complex is shown. E, alignment of the N-terminal sequences of the three hormones. Color codes are as indicated. and Cys 11 . Because this loop is oriented away from the receptor (Fig. 5C), the part of the N terminus really available for interaction with the receptor is of similar length in hGH and hPRL, and elongated by only three residues in oPL and in the affinitymatured mutant hPRL Full-oPL . Clearly, the substitution of the natural hPRL N terminus by that of oPL stabilized the 1:2 hPRL Full-oPL ⅐rPRLR 2 complex by increasing both site 1 and site 2 affinities (Table 3). The addition of a single hydrogen bond is likely responsible for the stabilization of PRLR2 docking. However, although all hPRL mutants but hPRL T14P harbored Gln (Ϫ1) and Tyr 3 in their sequences, significant increase of site 2 affinity was observed only for hPRL PGGA-T14P and hPRL Full-oPL . This suggests that the entire N-terminal sequence of oPL is required for the N terminus to adopt a suitable folding for the stabilization of the 1:2 complex. Determining the crystal structure of wild-type hPRL complexed to a PRLR-ECD dimer would help in understanding the actual network of interactions generated by the native N terminus hPRL sequence.

The Three-pin Plug Interaction Hypothesis
As described above, three features can be distinguished within the two regions that constitute the binding site 2: the helix ␣3 glycine 129, the hydrogen bond network involving surrounding residues (glycine cavity) and the N terminus. We attempted to rank the importance of these three components in light of the available mutational data. Clearly, the glycine residue is the major feature of hPRL binding site 2, as highlighted by the fact that any mutation of Gly 129 tested to date generated antagonists (20). In contrast, mutations of the two other components failed to generate antagonists, and at best weakened agonistic properties. Truncation of up to the 13 N-terminal residues of hPRL or elongation of three residues (hPRL Nter ) affected only very marginally its affinity for the full-length transmembrane PRLR or the hPRLR-ECD, as well as its bioactivity in PRLR-mediated cell assays (20,42). With respect to the hydrogen bond network, we focused on Arg 21 , which is involved in the only hydrogen bond that was strictly conserved among the three hormone families. We generated R21A and R21W mutants, both of which led to a significant decrease of agonism (10-fold or more) in the Ba/F-hPRLR and HL-5 cellbased bioassays (data not shown), confirming the functional importance of the hydrogen bond between Arg 21PRL and Thr 98PRLR2 .
Despite the fact that mutation G129R induced at least a 150fold decrease in site 2 binding affinity (that was actually undetectable by SPR (20)), this sole mutation was not sufficient to obtain a pure antagonist (19,43). This suggests that when the glycine pocket is hindered, one or both other regions can ensure a limited level of receptor triggering. The involvement of the N terminus was experimentally demonstrated, because its deletion (Del1-9-G129R-hPRL) knocked down residual agonism (2), whereas insertion of the AQHPPY motif (G129R-hPRL Nter ) boosted the residual agonistic activity to twice the level of G129R-hPRL (20). This could suggest the N terminus provides enough residual interaction energy to allow a limited recruitment of the second PRLR, or to induce a limited reorientation of the latter within a pre-formed hPRLR dimer. To test the effect of the hydrogen bond network in residual agonism, we combined R21A and G129R mutations. The double mutant displayed similar residual agonism, but slightly improved antagonism compared with G129R-hPRL (Ͼ2 fold, data not shown), which partly agrees with our hypothesis. However, the effect of this double mutation is difficult to interpret precisely, because the replacing residue (Arg 129 ) may partly compensate the loss of Arg 21 by establishing hydrogen bonds, which could attenuate the intrinsic effect of mutation R21A alone.
In summary, we can propose that two of the three components of site 2 are needed to achieve detectable agonism, whereas mutation of Gly 129PRL is mandatory to generate antagonists. Mutation of the hydrogen network appeared to be more effective in agonists, while the importance of the N terminus was only apparent in antagonists.
Finally, as already observed for the oPL complex, it is noteworthy that the regions of the PRLR domain D1 involved in interactions with PRL binding sites 1 and 2 are very similar, although individual binding residues are not strictly identical (Fig. 6). In the GH complex, this symmetry is extended to the receptor domain D2. This could not be assessed in both oPL and PRL complexes due to the absence of loop L5 PRLR2 in the electron density. There are much less mutational data available to be correlated with our structural observations for the receptor than for the ligand. In addition, the residues that were identified as important for hormone binding could not be assigned to a particular binding site, because functional assays were performed using cell-based bioassays (dimerized membrane receptor) and not SPR (monomeric ECD). Mutation of any of the four cysteines of D1 (44), Arg 13PRLR , Glu 18PRLR , and Phe 64PRLR (45) were shown to decrease the affinity for PRL by 300-fold or more. It is likely that the loss of binding observed for the cysteine mutants results from structural alterations of D1, because these cysteines are involved in intramolecular disulfide bonds. This is probably also the case for the mutation of Phe 64PRLR , because this residue is clearly outside both binding sites, but is within the core of D1 and in contact with Phe 20PRLR , which is close to the two disulfide bonds. Arg 13PRLR participates in maintaining the hinge through interactions involving Tyr 190PRLR (see above), therefore its mutation could also alter the global folding of the extracellular domain. Finally, Glu 18PRLR is the only residue of this short list to be involved in direct interactions with both binding site 1 and site 2 of PRL. In oPL, this residue is also involved in the hydrogen bond network with the hormone residues surrounding the glycine pocket (the second site 2 component that we identified).

Comparison of Free and Bound Hormone Structures
From the numerous hGH structures that already exist for the free hormone (PDB codes 1HUW and 1HGU), 1:1 (PDB codes 1A22, 1HWH, and 1BP3), and 1:2 complexes (PDB codes 1HWG, 1KF9, and 3HHR), it has been proposed (46) that GH structure is modified upon binding to GHR 1 , leading to a conformation that allows recruitment of GHR 2 . Our newly determined 1:2 structure provides the clues to address this issue for the PRL⅐PRLR 2 complex. To that end, we superimposed free and bound hormones, choosing helix ␣2 as a superimposition region, because it is the only ␣-helix not involved in receptor contacts.
Concerning the GH family, as already mentioned, the structures of hGH in the 1:1 and 1:2 complexes are accurately superimposed (data not shown). In contrast, free hGH shows clear differences in the bending of the three ␣-helices involved in GHR binding. A clear inclination is observed for the upper part of ␣1, correlated with the same movement of the lower part of ␣4 despite the absence of direct interaction of those regions with the receptors. This movement seems to be induced by the displacement of the small helix ␣1Ј, due to receptor binding at site 1. On the opposite side of the hormone, there is a clear correlated bending of helices ␣1 and ␣3 in the regions involved in binding site 2.
Most of these observations can be transposed to the PRL family. Despite the low resolution of our Prl complex structure, the electron density is clear enough to allow a precise backbone comparison. The structures of PRL in the 1:1 and 1:2 complexes are perfectly superimposed, except for the 18 N-terminal residues, as expected. On the contrary, only helix ␣2 can be superimposed when comparing free and complexed PRL. The difference in the bending of the three other ␣-helices is much more pronounced than in the GH family. Most interestingly, the three ␣-helices bend in opposite ways in PRL and GH upon binding (Fig. 7).
Finally, we compared the three hormones (hPRL, oPL, and hGH) within their complexes with PRLR (supplemental Fig. S5). Although hPRL and oPL are complexed with two rat PRLR-ECDs, whereas hGH is bound to only one human PRLR-ECD, the comparison is relevant as both PRL and GH adopt the same structure in their respective 1:1 and 1:2 complexes (at the exception of the N terminus). Helix ␣2 from the three complexes was easily superimposed. Most of the differences between the three hormones are localized in the non-interacting regions with the exception of region 52-56 PRL (38 -47 GH ) from loop 1, which corresponds to the additional binding site 1 region, called ␣1Ј, in the PRL⅐PRLR 2 complex when compared with the GH⅐GHR 2 complex. This region, which moves to accommodate loop L2 PRLR , corresponds to the most noticeable difference in GH between the hGH⅐hGHR 2 and hGH⅐hPRLR 2 complexes (46) confirming it is specific to PRLR binding.
The curvature of helix 3 is completely different between the three hormones, but the region involved in the glycine cavity is closely superimposable. This suggests that it is PRLR2 that accommodates the hormone and not the opposite. This would explain why the structure of the hormone is perfectly superimposed in all the 1:1 and 1:2 complexes available for both the PRL and GH families, with the exception of the N terminus for the PRL family, which confirms the importance of this region in the stability of the 1:2 complex. The 1:2 complex structure between GH and PRLR would be necessary to verify this hypothesis.
When comparing the intramolecular hydrogen bonds among the different PRL forms (free or liganded), we found that the conserved helix-interacting residues correspond or are close to ␣-helix hinges (Fig. 7). One additional conserved interaction involves a residue from loop 1. This link between Thr 60PRL and Asp 178PRL reinforces the attachment of loop 1 with helix ␣4 in addition to the disulfide bond between Cys 58PRL and Cys 174PRL . The hinge positions are conserved in the GH family except for helix ␣3 where the hinge is localized two residues further. Consequently, movements of ␣-helix half pieces are coordinated. When PRLR interacts with PRL site 1, it imposes the final structure of segment 178 -199 PRL from helix ␣4 and of region 52-56 PRL from loop 1. The movement of this second region induces the correlated bending of the "upper part" of the hormone (28 -44 PRL with 161-178 PRL and with 110 -125 PRL ; see Fig. 7). As for the movement of segment 178 - 199 PRL , it drags along the "lower part" of the hormone (regions 15-28 PRL and 125-136 PRL ). Region 15-28 PRL reaches its final position only when the second PRLR-ECD interacts at binding site 2.
The intrinsic dynamics of the hormone, favored by the presence of the two long loops, seems to be a prerequisite for its fully functional interaction with the receptor, as hypothesized by Jomain et al. (20) and by the antagonistic properties of the ⌬41-52-hPRL (47). As evidenced by the superimposition of the three different hormones in complex with PRLR receptor, the binding of receptor 1 seems to impose the final structure of the hormone, at least at the level of binding site 1. The second receptor then appears to accommodate itself to the 1:1 complex and the N terminus of the hormone locks the 1:2 complex in its final conformation.

CONCLUSION
In this study, we report the first x-ray structure of a 1:2 PRL⅐PRLR 2 complex that involves PRL as the hormone ligand. This structure provides different valuable insights. First, it localizes precisely the two hormone/receptor interfaces (binding sites), allowing to interpret from a structural perspective the site-directed mutagenesis data that has been gathered along the years to determine the hot-spots of each of the two hPRL binding sites. The structure is globally in good agreement with these experimental data, and provides a more exhaustive picture of the residues involved in both receptor/hormone interfaces. In particular, it provides structural support for the key functional role of the glycine 129 pocket in binding site 2, and further clarifies the functional role of the N terminus of hPRL. Second, it allows us to better understand the molecular basis of the properties of the different hormone analogs that have been designed to date, including receptor antagonists, such as Del1-9-G129R-hPRL (2,20). Third, it suggests that, as described for the GH⅐GHR system, the binding of the second PRLR molecule to the 1:1 PRL⅐PRLR complex does not modify the structure of the hormone, suggesting that the receptor adapts to the 1:1 complex rather than the opposite. Altogether, this structure will also be useful for further improving current PRLR antagonists or for designing new ones that may act by novel molecular mechanisms. It will also be a precious tool in attempting to understand the mechanism by which the natural PRLR variant comprising the I146L mutation in its extracellular domain acquires constitutive activity (8).