Structural and Thermodynamic Bases for the Design of Pure Prolactin Receptor Antagonists

Competitive antagonists of the human prolactin (hPRL) receptor are a novel class of molecules of potential therapeutic interest in the context of cancer. We recently developed the pure antagonist Del1-9-G129R-hPRL by deleting the nine N-terminal residues of G129R-hPRL, a first generation partial antagonist. We determined the crystallographic structure of Del1-9-G129R-hPRL, which revealed no major change compared with wild type hPRL, indicating that its pure antagonistic properties are intrinsically due to the mutations. To decipher the molecular bases of pure antagonism, we compared the biological, physicochemical, and structural properties of numerous hPRL variants harboring N-terminal or Gly129 mutations, alone or combined. The pure versus partial antagonistic properties of the multiple hPRL variants could not be correlated to differences in their affinities toward the hPRL receptor, especially at site 2 as determined by surface plasmon resonance. On the contrary, residual agonism of the hPRL variants was found to be inversely correlated to their thermodynamic stability, which was altered by all the Gly129 mutations but not by those involving the N terminus. We therefore propose that residual agonism can be abolished either by further disrupting hormone site 2-receptor contacts by N-terminal deletion, as in Del1-9-G129R-hPRL, or by stabilizing hPRL and constraining its intrinsic flexibility, as in G129V-hPRL.

Human (h) 4 prolactin receptor (PRLR) antagonists are a new class of potential drugs developed to target prolactin (PRL)sensitive pathologies that cannot be treated with current inhibitors of the production of hPRL by the pituitary (1). These include dopamine-resistant prolactinomas (i.e. pituitary tumors of PRL-secreting cells), as well as breast cancer, prostate cancer, and benign prostate hyperplasia, in which evidence for the tumor growth-promoting actions of autocrine PRL has been emerging within the past decade (2,3). In view of the ubiquitous expression pattern of the PRLR (4), these indications are not necessarily exhaustive and may be extended to other pathologies that remain to be identified (or better characterized) with respect to the involvement of locally produced PRL in their etiology.
Because of their mechanism of action involving competition with endogenous hPRL for receptor binding, competitive hPRLR antagonists need to be used in molar excess compared with endogenous hPRL. Therefore, the most promising compounds are anticipated to be those that are devoid of any residual agonistic properties, even at high concentration (5). We recently developed an antagonist exhibiting such unique properties, referred to as Del1-9-G129R-hPRL. In contrast to other antagonists developed to date (for reviews see Refs. 2, 6), Del1-9-G129R-hPRL was shown to be devoid of residual agonism in every cell or animal model in which it has been tested to date (5,(7)(8)(9)(10)(11).
As highlighted by its name, Del1-9-G129R-hPRL is a hPRL core protein containing two modifications: deletion of the nine N-terminal residues, and substitution of Gly 129 for an Arg (5). The rationale for engineering this variant was based on previous structure-function studies performed in the Goffin laboratory. First, as demonstrated for all members of the PRL/growth hormone (GH)/placental lactogen (PL) family (2), substitution of the conserved helix 3 Gly for an Arg was shown to drastically impair the agonistic properties of hPRL while maintaining its ability to bind to the PRLR (12)(13)(14). Second, the sole G129R mutation appeared to be insufficient to completely abolish the ability to activate the PRLR, as highlighted in sensitive cell bioassays (12,15) or in transgenic mice overexpressing this variant. 5 Third, mutations of the N terminus, which is interestingly the most divergent region within the PRL/GH/PL family, were shown to slightly modulate the properties of hPRL (16).
Although the combination of the N terminus deletion and the G129R mutation was successful in achieving the goal of generating a pure PRLR antagonist, the mechanism underlying these unique properties remained poorly understood. Although structural studies of the hGH⅐hGHR complex have clearly identified the helix 3 Gly pocket as a critical characteristic of binding site 2 (17,18), the roles of the structurally equivalent region (Gly 129 ), and even more of the N terminus, in receptor binding and activation have remained largely speculative for hPRL. Ultimately, this lack of information about the molecular bases of our best antagonist hampers its improvement through knowledge-assisted strategies.
The aim of this work was to elucidate the importance of these two hPRL modifications at the molecular level, using a combination of structural, biophysical, and biological approaches. First, to identify the structural characteristics underlying its pure antagonistic properties, we determined the three-dimensional structure of Del1-9-G129R-hPRL by x-ray diffraction. Second, we generated numerous hPRL variants harboring single or combined modifications at the N terminus (deletion/ elongation) and the Gly 129 residue (different substitutions), to be compared with the prototype G129R-hPRL and Del1-9-G129R-hPRL variants. These mutants were analyzed for their (residual) agonistic and antagonistic properties in cell bioassays, and for their thermodynamic stability and structural integrity. Finally, we designed an appropriate methodology to determine independently the affinities of binding sites 1 and 2 of all the variants by surface plasmon resonance.

Materials
Culture media, fetal calf serum (FCS), geneticin (G-418), trypsin, and glutamine were purchased from Invitrogen. Luciferin and cell lysis buffer were from Promega (Madison, WI), and luciferase activity was measured in relative light units using a Lumat LB 9501 (Berthold, Nashua, NH). IODO-GEN was purchased from Sigma, and carrier-free Na 125 I was obtained from GE Healthcare. Oligonucleotides were from Eurogentec (Liège, Belgium). All immobilization reagents used for surface plasmon resonance experiments were purchased from Biacore (Uppsala, Sweden). Optimization of crystallization conditions was performed in Linbro plates from Hampton. Chemicals were purchased from Sigma, VWR (Fontenay-sous-Bois, France), or Merck.

Site-directed Mutagenesis
Expression plasmids encoding WT hPRL, G129R-hPRL, and Del1-9-G129R-hPRL were available from previous studies (5,12). They were used as templates for generating the other mutants of this study, using the QuikChange II mutagenesis kit from Stratagene (La Jolla, CA). Sequences of forward and reverse (complementary) primers are given in Table 1, with mutated codons underlined. The same primers were used for generating Nter-hPRL and Nter-G129R-hPRL plasmids using, respectively, WT hPRL and G129R-hPRL encoding plasmids as templates. For all steps, we strictly followed the recommendations of the manufacturer. After transformation, Escherichia coli BL21(DE3) colonies were analyzed for their DNA content; plasmids were sequenced to verify the presence of the expected mutations.
Plasmids encoding the extracellular domain (ECD) of human or rat (r) PRLR were generated by PCR amplification using plasmids containing the full-length receptor cDNA (19,20); sequences encoding residues 1-210 of WT PRLR were inserted at SpH1-BamHI sites into the pQE-70 expression plasmid containing a His 6 tag at the C-terminal end (Qiagen, Courtaboeuf, France). Subcloning constraints led to the addition of four amino acids just before the His 6 tag (Gly 211 -Ser 212 -Arg 213 -Ser 214 for hECD, and Arg 211 -Ser 212 -Arg 213 -Ser 214 for rECD) as described (21).

Production and Purification of hPRL Variants and hPRLR-ECD
Recombinant WT hPRL, hPRL variants, and PRLR ECDs were overexpressed in 0.5-1-liter cultures of E. coli BL21(DE3) and purified as described previously (22), with minor modifications. Briefly, when the A 600 of bacterial cultures reached 0.7-5 V. Rouet, C. Kayser, and V. Goffin, unpublished observations. 0.9, overexpression was induced using 2 mM isopropyl 1-thio-␤-D-galactopyranoside for 4 h (A 600 ϭ 2-2.5 after 4 h). Cells were broken using high pressure (French press). Proteins were overexpressed as insoluble inclusion bodies, which were solubilized in 8 M urea (5 min at 55°C and 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 centrifuged and then loaded onto a HiTrap Q anion exchange column (GE Healthcare) equilibrated in 25 mM NH 4 HCO 3 , pH 8.6. Prolactin (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.

Crystallization of Del1-9-G129R-hPRL 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). Crystals appeared as needle clusters after 1 week at 18°C in several conditions (29,36,41) of the MemFac crystallization kit.
The initial crystals were refined using standard techniques, which lead to isolated needles too thin to be collected. A second screening was performed by mixing 75% of the optimized crystallization solution with 25% of the crystal screen kits from Hampton. Larger needles appeared in five new conditions. After manual optimization in Linbro plates using the hanging drop method, only one led to diffracting crystals. The corresponding reservoir was composed of 100 mM Tris-HCl, pH 8.5, 675 mM K 2 HPO 4 , 45 mM (NH 4 ) 2 PO 4 , 50 mM LiSO 4 , 7.5% glycerol, 8% PEG 4000 . The drop was formed by mixing 1.5 l of protein at 10 mg/ml with 1.5 l of reservoir. The largest final crystal size was 40 ϫ 40 ϫ 200 m 3 . Best diffracting crystals were flash-frozen in liquid nitrogen without previous soaking. A complete data set at 2.6 Å resolution was obtained on the ID29 beam line at ESRF (Grenoble, France).
The diffraction images were reduced, scaled, and merged with programs MOSFLM and SCALA (23,24). The intensities were then converted to the structural factor amplitudes with TRUNCATE (23,24). A summary of the crystallographic data and refinement statistics is given in Table 2. The crystals belonged to space group I4 with unit cell dimensions of a ϭ b ϭ 122.59 Å, c ϭ 28.68 Å. The Matthews coefficient (2.25) and solvent content analysis (ϳ42%) indicated the presence of one antagonist molecule in the asymmetric unit.

Phase Determination and Structure Refinement
The crystal structure of the antagonist was solved by molecular replacement using the program PHASER (25)(26)(27), with the structure of oPL extracted from the 2:1 complex structure of oPL/rPRLR (Protein Data Bank code 1F6F) used as search model. A PHASER search with the WT hPRL structure resolved by NMR (Protein Data Bank code 1RW5) used as search model also yielded the right solution, but the PHASER statistics were not convincing. Refinement of the structure was carried out through multiple cycles of manual rebuilding using the program O (28) and refinement using Refmac5 (29), resulting in a final model with an R factor of 21.1% and an R free factor of 30.2%. The refined structure was validated by the program PROCHECK (30). Figure panels showing three-dimensional structures were generated using the PyMOL Molecular Graphics System (DeLano Scientific). The refined structure of the antagonist and that of WT hPRL (PDB code 1RW5) (21) were both subjected to all-atom normal mode analysis, using the web-based servers HingeMaster (31) and NOMAD-Ref (32).

Circular Dichroism
Far-UV CD spectra as well as temperature melts of all hPRL variants were recorded on a Jasco 810 spectropolarimeter. Temperature melts were performed in 0.01 M NaH 2 PO 4 , pH 7.4, using a water-jacketed cuvette of 1-cm light path length, from 37 to 95°C, and the change in signal followed at 222 nm. Far-UV CD spectra were run at 37°C from 250 to 190 nm, with a 1-cm path length. However, because of high absorbance from buffer, no signal below 197 to 195 nm could be interpreted. Protein concentrations were set at ϳ1-2 M for the temperature melts and far-UV CD spectra.
Unfolding was assumed to follow a two-state transition. Circular dichroism thermal scans were fitted to the Gibbs-Helmholz equation as described (33), using Gnuplot. Because of the high melting temperatures, which result in the lack of sufficient post-transition regions, it was assumed that the unfolded state was unchanged by the mutations. The slope of the post-transition region was therefore extrapolated for each mutant protein from the well determined post-transition of G129P-hPRL. The results of the fits are melting temperature (T m ) and change in enthalpy on unfolding at T m (⌬H (Tm) ).
where ͗I͘ h is the mean intensity of symmetry equivalent reflections. c ⌺͉F obs Ϫ F calc ͉/⌺F obs . The formula for R free is the same as that for R, except it is calculated with a portion of the structure factors that had not been used for refinement.

Surface Plasmon Resonance
Immobilization of PRLR ECD-As a first approach, the ECDs (human or rat) were covalently coupled in a random orientation, through their solvent-accessible primary amine groups, to the carboxymethylated dextran matrix of a CM5 sensor chip, using a Biacore 2000 instrument and the amine coupling kit (Biacore), according to manufacturer's instructions. Briefly, each flow cell, equilibrated at a flow rate of 5 l/min in phosphate-buffered saline (PBS, pH 7.4, supplemented with 0.005% Tween 20), was activated for 12 min with an NHS-EDC solution (50 mM N-hydroxysuccinimide and 200 mM N-ethyl-NЈ-(3 dimethylaminopropyl) carbodiimide), followed by an injection of the ECD of interest (100 nM) in 10 mM sodium acetate, pH 4.5. The surface was finally deactivated for 12 min with 1 M ethanolamine, pH 8.5. We routinely immobilized 500 -1500 resonance units (1 resonance unit ϳ1 pg⅐mm Ϫ2 ) of ECD on three flow cells out of four on the sensor chip. The fourth, treated only by NHS-EDC and ethanolamine, was used as a reference surface. In a second approach, the ECDs were covalently coupled in an oriented fashion onto a nitrilotriacetic acid-derivatized NTA sensor chip, as described (34). Briefly, each flow cell, equilibrated at a flow rate of 5 l/min in PBS supplemented with 0.005% Tween 20 and 50 M EDTA, was sequentially loaded with 500 M NiCl 2 for 4 min and activated with NHS-EDC for 2 min. The His 6 -tagged ECD (100 nM in PBS/Tween/ EDTA) was injected until 500 -2,000 resonance units were captured by the Ni-NTA moieties. The surface was finally deactivated by a 4-min ethanolamine injection, followed by a 2-min injection of EDTA 0.35 M. One flow cell was used as a reference surface, after activation by NiCl 2 and NHS-EDC, and deactivation by ethanolamine and EDTA.
Real Time Binding Assays-All the binding assays were performed at 25°C at a flow rate of 20 l/min, in running buffer (25 mM NH 4 HCO 3, pH 8.6, 150 mM NaCl, 0.005% Tween 20). For site 1 characterization, 5-8 different concentrations (ranging from 0.54 to 350 nM) of the hPRL variants were injected for 12 min onto the PRLR surfaces, followed by a 10-min dissociation period. At the end of each cycle, the sensor chip was regenerated by two 1-min injections of 2 M MgCl 2 (followed by a 1-min injection of EDTA 0.35 M for Ni-NTA chips only). For site 2 characterization, the PRLR surface was first saturated with each hPRL variant (350 nM). Five concentrations of ECD (ranging from 1.7 to 350 M) were then injected for 2 min on the ECD⅐PRL complexes, followed by a 5-min dissociation period. Control experiments showed the absence of detectable ECD-ECD interaction in the absence of hormone (not shown).
Data Analysis-All the association and dissociation profiles were double-referenced using the Scrubber 2.0 software. The site 1 binding curves were globally analyzed with a nonlinear least squares algorithm implemented in the BIAevaluation 4.1 software (Biacore), using single-exponential functions of time (Langmuir monovalent binding model). The site 2 binding curves were subjected to steady-state analysis. Kinetic parameters (k on and k off ) and equilibrium dissociation constants (K d ) were determined based on at least two experiments.

Cell-based Binding Assays
Binding affinities of hPRL variants were determined using cell homogenates of HEK 293 cells stably expressing the human PRLR (so-called HL-5 clone), following a procedure described previously (14). Briefly, hPRL was iodinated using IODO-GEN, and binding assays were performed overnight at room temperature using 150 -300 g of cell homogenate protein in the presence of 20,000 -30,000 cpm 125 I-labeled hPRL and increasing concentrations of unlabeled competitor (WT or mutated hPRL). Results presented are representative of at least three independent experiments performed in duplicate. The relative binding affinities of hPRL variants were calculated with respect to that of WT hPRL based on their IC 50 .

Cell-based Bioassays
The biological properties of hPRL variants were analyzed using two homologous bioassays that we recently developed for human lactogens (15). The transcriptional assay involved HEK 293 cells stably expressing the hPRLR and the lactogenic hormone-response element-luciferase reporter gene (HL-5 clone); it was used to determine the antagonistic properties by competing a fixed concentration (9 nM) of WT hPRL with increasing amounts of each variant (14,15,35). A proliferation assay was also performed to determine the residual agonistic activity of the antagonists, using Ba/F-03 cells stably expressing the human PRLR (referred to as Ba/F-LP cells); this assay has been shown to be much more sensitive than the HL-5 clone to reveal low level agonistic effects (5,15,36). These assays were run as follows.
Antagonism Bioassay (HL-5 Cells)-The HL-5 clone was routinely cultured in Dulbecco's modified Eagle's medium supplemented with 10% FCS, 2 mM glutamine, 50 units/ml penicillin, 50 g/ml streptomycin, and 700 g/ml G-418 (for clonal selection). The assay was performed in 96-well plates using 50,000 cells/100 l/well in medium containing only 0.5% FCS. Cells were allowed to adhere for 6 h and then 100 l of (2ϫ) hormones diluted in 0.5% FCS medium were added to each well. After overnight stimulation, cells were lysed (50 l lysis buffer; Promega) and then luciferase activity contained in 10 -20 l cell lysate was counted for 10 s. To avoid inter-assay variations, all variants to be compared were systematically tested in the same experiment, and data obtained in one experiment representative of at least three experiments performed in duplicate are shown. In antagonism experiments, a mix of 50 l of (4ϫ) hormone variants combined with 50 l of (4ϫ) WT hPRL (to obtain a final concentration of 9 nM) were added. Experimental data were plotted using the GraphPad software.
Agonism Bioassay (Ba/F3-LP cells)-Ba/F3-LP cells were maintained in RPMI 1640 medium supplemented with 10% heat-inactivated FCS, 2 mM glutamine, 50 units/ml penicillin, 50 g/ml streptomycin, 500 -1000 g/ml G-418, and 10 ng/ml WT hPRL instead of interleukin-3 (nontransfected Ba/F03 cells are dependent on that cytokine for survival/growth). For the assay, cells were starved for 6 h in 1-2% FCS RPMI medium (with additives) and then distributed in 96-well plates at a density of 50,000 cells/100 l/well in the same medium (excluding hormones). One hundred l of [2x] hormones were added to Physicochemical Properties of Pure PRL Receptor Antagonists NOVEMBER 9, 2007 • VOLUME 282 • NUMBER 45 each well. Cell proliferation was monitored after 2 days of hormonal stimulation using 10 -15 l of cell proliferation reagent WST-1 (Roche Applied Science). Experiments were performed at least three times in triplicate. Experimental data were plotted using the GraphPad software.

Determination of the Del1-9-G129R-hPRL Three-dimensional Structure
The totality of the crystallized sequence has been positioned in the electron density. The refined structure is composed of 1567 protein atoms and 48 water molecules. The secondary structure corresponds to the classical long chain cytokine fold consisting of four main ␣-helices (Leu 15 to His 46 , Gln 77 to Arg 103 , Glu 110 to Val 137 , and Glu 161 to Ile 194 ) arranged in an up-up-down-down topology with two long overhand loops termed 1 and 3 (Fig. 1A). A short 3 10 helix is present from His 59 to Ser 62 (helix 1Ј), and two additional small ␣-helices are located from Lys 69 to Gln 74 (helix 1Љ) and from Gly 152 to Gln 157 (helix 3Ј). Helix 3Ј was not observed in the NMR structure of WT hPRL, and helix 1 is four residues longer in the crystal structure of the antagonist, possibly reflecting the more dynamic state of the soluble protein.
There is one disulfide bond involving residues 58 and 174, which links loop 1 to helix 4. The disulfide bond between Cys 191 and Cys 199 described in both the crystallographic structure of hPL (PDB code 1Z7C; Ref. 37) and in the NMR structure of WT hPRL (PDB code 1RW5; Ref. 21) is not present in the antagonist. Instead, the C-terminal Cys is in close contact with Ser 11 of a symmetry-related molecule (Fig. 1E). The natural residue at position 11 in WT hPRL is a Cys, which is involved in a disulfide bond with Cys 4 . In the antagonist, as the N-terminal nine first FIGURE 1. Comparison of hPRL and Del1-9-G129R-hPRL three-dimensional structures. A, crystallographic structure of Del1-9-G129R-hPRL (antagonist), colored as a function of temperature factors (red/high, blue/low). B, NMR structure of hPRL (agonist) including the 20 models described (21). C, superimposition of C-␣ atoms of Del1-9-G129R-hPRL (green) and hPRL (blue, the lowest energy model is presented) was performed using PyMOL; it highlights the absence of major structural changes in the antagonist. The situation of binding sites 1 and 2 and the position Gly 129 (or Arg in the antagonist) are also indicated. D, analysis of Del1-9-G129R-hPRL structure as a function of residue number. The accessibility has been multiplied by minus 1 for clarity. The r.m.s.d. values are calculated between Del1-9-G129R-hPRL and WT-hPRL structures. E, representation of the packing of Del1-9-G129R-hPRL in the I4 space group, which illustrates the following: (i) the dynamic regions are all pointing into the solvent channel, and (ii) the N termini are in proximity of the C termini of symmetry related molecules.
residues have been omitted, Cys 11 has been mutated to Ser (to avoid a free Cys; Ref. 5), and residues 10 -14 are thus free to be oriented differently. This loop packs on the C-terminal part of a symmetry related molecule. Accordingly, these structural features move in a coordinated manner in the channel formed by the I4 packing (Fig. 1E).
The structure of the antagonist, colored according to temperature factors, is shown in Fig. 1A. The core of the structure composed of the four classical ␣-helices is quite stable, with a mean B factor calculated on C-␣ of 28.6 Å 2 , to be compared with the mean B factor of the whole structure of 33.8 Å 2 ( Table  2). The most dynamic parts are composed of the N and C termini of the protein, of loop 2 (located between helix 2 and helix 3), and of the long loop-helix-loophelix located between helices 1 and 2. In particular, the first part of the latter (loop 1) and helix 1Љ are very dynamic, with B values up to 57 Å 2 . This region of the hormone is poorly defined in most of PRL/ GH/PL structures found in the PDB (also when determined by NMR), and it has been hypothesized to adopt a defined structure only when the hormone is receptor-bound (17). In our case, this is also partly the case because of the location of this loop in the crystal packing, as loop 1 and helix 1Љ point inside the I4 space group channel (Fig. 1E). Fig. 1D represents the B factor and the accessibility (multiplied by minus one for clarity) in the crystal as a function of residue number. The two curves evolve in parallel, with the sole exception of the N and C termini of the protein. This is because of the reasons listed above, i.e. that those two regions are surrounded by solvent and are in close contact; therefore their respective movements are correlated.

Comparison of Del1-9-G129R-hPRL and WT hPRL Structures
Two structures of PRL exist in the PDB, 1N9D (38) and 1RW5 (21) (Fig. 1B). Both are unbound, natural PRL of human origin, produced in bacteria and resolved by NMR. Those two structures are significantly different with a root mean square deviation (r.m.s.d.) value of 3.8 Å, as noted previously (21). Our structure is the first PRL-core structure resolved by crystallography. It was fitted to both NMR structures, with r.m.s.d. values on C-␣ of 5.2 and 1.53 Å, respectively. When limited to the four-helix bundle core, the r.m.s.d. fit improved to 4.08 and 0.90 Å, respectively. As we (21) and Hodsdon and co-workers (39) recently agreed that 1RW5 is more accurate than 1N9D, this strongly suggests high structural similarity between the antagonist and the WT hormone (Fig. 1C). We plotted the r.m.s.d. between our structure and 1RW5 as a function of residue number (Fig. 1D). Fig.  1D also shows temperature factors and solvent accessibility of Del1-9-G129R-hPRL in the crystal packing. The three curves are synchronized, indicating that the only noticeable differences between the WT hPRL NMR structure and the antagonist crystallographic one are localized in accessible dynamic regions. This further argues that the mutations performed in FIGURE 2. Structural comparison of the Gly 129 pocket in hPRL and Del1-9-G129R-hPRL. A, in WT hPRL, Gly 129 is surrounded by long side chain residues (Arg 21 , Leu 25 , Arg 125 , and Glu 128 ), which form the "site 2 cup." The N terminus (N-ter) is folded on the edge of this cup. B, two modifications harbored by the antagonist modify the conformational characteristics of this region; the N-terminal truncation removes a stretch of residues assumed to modulate site 2 properties, and Arg 129 (in red) now fills the bottom of the cup. C, superimposition of the Gly 129 regions from WT hPRL and Del1-9-G129R-hPRL. This panel shows that some of the long side chain residues that delineate the cup in WT hPRL (white residues) adopt a different conformation in the antagonist (green residues) because of the presence of the Arg 129 (in red). Residues presented in yellow and orange belong to symmetry-related molecules. The N terminus is on the edge of the cup and is represented as a green tube. NOVEMBER 9, 2007 • VOLUME 282 • NUMBER 45 binding site 2 do not generate major structural rearrangement (Fig. 1C).

Physicochemical Properties of Pure PRL Receptor Antagonists
A detailed comparison of the two structures focused on the Gly 129 pocket is presented on Fig. 2. This highlights the cup formed by Gly 129 and the large residues surrounding it in WT hPRL and the absence of a cavity because of the Arg mutation in the antagonist. It also shows the N terminus on the edge of the Gly pocket in WT hPRL, and its absence in the antagonist mutant. Residues involved in the formation of the site 2 binding pocket are part of both helix 1 and of helix 3 (Fig. 2C). There are correlated movements of Leu 18 , Leu 132 , and Glu 128 which position at van der Waals distances of Arg 129 . Leu 25 is not affected by the presence of the mutated residue. The rearrangements of Arg 125 , Arg 21 , and Gln 136 are not because of the presence of Arg 129 , but because of two symmetry-related molecules. Sym-Glu 161 forms a hydrogen bond with Gln 136 and with the carbonyl of the N-terminal methionine (initiation codon). As can be seen in the WT structure (Fig. 1B), the N terminus is undefined because of the lack of distance information. In the antagonist structure, the presence of the symmetry molecule forces the N terminus to reorient toward the Arg 129 cup. Despite that, the backbone of residues 11-14 is superimposed on the corresponding region of the NMR structure (Fig. 1C). In conclusion we can argue that the replacement of Gly 129 by an Arg does not affect the global structure of hPRL and generates only small displacements of residues surrounding locally the arginine.

Rationale for the Design of hPRL Variants
To investigate the individual contribution of the two modifications in Del1-9-G129R-hPRL (N terminus and Gly 129 ) to the ability of hPRL to activate the hPRLR, we generated several mutants in addition to those already available. All mutants analyzed in this study are shown in Fig. 3. Two N-terminal variants were generated to mimic those of the two other members of the hormone family. Deletion of the nine first residues (Del1-9-hPRL) mimics the N terminus of hGH as described previously (16). As for the elongated Nter-hPRL variant, it contains the sequence AQHPPY at its N terminus instead of the natural LPI sequence of hPRL; in ovine PL, this AQHPPY motif was shown to contact the PRLR and to contribute to a large extent of the energy of site 2 interaction (40).
With respect to the Gly 129 mutations, seven new substitution variants were generated and compared with the original G129R variant (12): Pro, Asp, Asn, Val, Leu, Tyr, and Phe. These residues were chosen to explore the influence of different parameters such as volume (Pro Ͻ Asp Ͻ Asn Ͻ Val Ͻ Leu Ͻ Tyr Ͻ Phe Ͻ Arg), charge (negative (Asp), positive (Arg), or none (Pro, Asn, Val, Leu, Tyr, Phe)), and polarity (Asp, Asn, Tyr). N-terminal variants and Gly 129 mutations were combined in Del1-9-G129R-hPRL and in Nter-G129R-hPRL variants.

Circular Dichroism
None of the mutations induced any major structural alteration, as far-UV circular dichroism analysis revealed similar spectral properties for all variants compared with WT hPRL, which is shown on Fig. 4A. All, except for G129N, had a double trough ratio of 222 / 208 larger than one and identical to WT hPRL, with an average of 1.05 Ϯ 0.01 as expected (12,16). G129N, however, Sequences of the N-terminal/helix 1 and Gly 129 /helix 3 regions are shown for the three natural hormones of the PRL/GH/PL family (numbering according to hPRL sequence). The N terminus was modified in hPRL or in G129R-hPRL to mimic either hGH (9 residues truncation in "Del1-9" variants) or oPL (elongation/mutation of LPI into AQHPPY sequence in "Nter" variants). Other Gly 129 variants contain a single substitution of that residue, as indicated. showed an increased contribution to the ellipticity at 222 nm, indicating either a more pronounced coiled-coil structure or other subtle structural changes involving aromatic residues (Fig. 4). This was not investigated further.
The heat denaturation of all hPRL variants was then followed by measuring the change in helical ellipticity at 222 nm with temperature. As shown in Table 3, N-terminal mutations (truncation or elongation) did not induce significant changes of T m , whether these modifications were introduced in the WT or the G129R contexts. In contrast, the Gly 129 substitutions led to a decrease in T m in two cases with 2 K for G129R-hPRL and 10 K for G129P-hPRL (Fig. 4B), the latter as expected for a proline substitution at a mid-helical position. All other Gly 129 substitutions led to a large and significant increase in T m ; the values for the G129N-hPRL (Fig. 4B), G129D-hPRL, and G129F-hPRL (not shown) mutants could be determined by extrapolating post-transition curves, whereas this was not possible for variants G129L-hPRL (Fig. 4B), G129V-hPRL, and G129Y-hPRL (not shown), indicating that their T m values were even higher than 368 K. For these last three variants, three assumptions were made to estimate their T m values as follows: 1) the unfolded state was unchanged by the mutation; 2) the percentage of signal change at 222 nm was expected to be the same as the average measured for other variants (32 Ϯ 2%); and 3) the slope of the post-transition region was unchanged compared with WT. Errors on these estimates were given as 4 K. All data are summarized in Table 3.

Binding Studies on the Full-length Transmembrane PRLR
The relative affinities of all hPRL mutants for the human receptor at the cell surface were estimated based on their relative IC 50 values (see displacement curves in Fig. 5). Truncation of the N terminus did not affect binding affinity, as shown by the IC 50 value of Del1-9-hPRL (1.1 Ϯ 0.4 nM), which is similar to that of WT hPRL (0.8 Ϯ 0.3 nM) (16). Elongation of the N terminus (Nter-hPRL) led to a modest (2-fold) increase in affinity (0.4 Ϯ 0.1 nM) (Fig. 5A). Similar effects were observed for the G129R-containing variants (Fig. 5B). The affinity of G129R-hPRL, which is ϳ1 order of magnitude lower than WT hPRL (IC 50 ϭ 19 Ϯ 5 nM), was not affected by N-terminal deletion (IC 50 ϭ 15.8 Ϯ 0.2 nM for Del1-9-G129R-hPRL), confirming our previous studies (5,14). Again, N-terminal elongation (Nter-G129R-hPRL) led to a modest 2-fold increase in receptor affinity (IC 50 ϭ 8 Ϯ 2 nM). None of the other substitutions at position 129 significantly modified the affinity for the human receptor compared with G129R-hPRL (data not shown).

In Vitro Binding Studies on the PRLR Extracellular Domain
Classical binding studies using radiolabeled ligands and cellexpressed receptors provide global averaged information on the affinity of the hormone-receptor complex. To decipher more specifically the impact of the mutations on site 1 and/or site 2 binding, we set up an SPR methodology that enables the sequential monitoring of the interaction of hPRL with two separate PRLR ECDs, adapted from recent studies of the GH-GHR interaction (41,42).
Two different strategies of immobilization of the human PRLR-ECD were investigated as follows: a classical random coupling through solvent-accessible primary amine groups (mainly from lysine side chains; data not shown), and an oriented coupling method relying on the capture of the C-terminal His 6 -tagged ECDs on a Ni-NTA surface, followed by an amide cross-linking (34) (Fig.  6). Although the random coupling method yielded surfaces with ϳ20% of active ECDs (i.e. ECDs able to bind hPRL), the oriented coupling method consistently gave better results, with 30 -35% of immobilized active ECDs. Furthermore, although both strategies resulted in site 1 association and dissociation profiles that could be very satisfyingly fitted

Physicochemical Properties of Pure PRL Receptor Antagonists
hPRLR-ECD was ϳ6 nM (Fig. 6A), with an association rate (k on ) close to 10 5 M Ϫ1 ⅐s Ϫ1 and a dissociation rate (k off ) of about 5 ϫ 10 Ϫ4 s Ϫ1 (Table 4). We also determined the K d value for the interaction between hPRLR-ECD and site 2 of hPRL, which was ϳ30 M, i.e. 10,000 times higher than that for site 1 (Fig. 6C and Table 4). As the interaction is very transient (the half-life of the 2:1 complex is just about 1 s), the association and dissociation rates could not be determined, and only steady-state analysis of the curves could be performed. N-terminal elongation (Nter-hPRL) or deletion (Del1-9-hPRL) did not significantly affect k on and k off values of site 1, whereas the affinity for site 2 was unaffected by N-terminal elongation and only slightly reduced (3-fold) by N-terminal truncation (Table 4). In contrast, Gly 129 mutations all induced a slight but significant increase (1.5-2.5-fold) in k off of site 1, with no significant change in k on . Most strikingly, all the substitutions at position 129 totally abolished site 2 binding, as exemplified by the G129P and G129R mutants for which no signal could be detected for hECD concentrations as high as 350 M, therefore giving a minimal threshold of 5 mM for the K d of site 2 for these mutants (Fig. 6D).

Correlations between Bioactivity and Thermodynamic Parameters
We systematically investigated whether the residual agonistic properties of the different PRL variants could be correlated to their physicochemical features, such as their thermodynamic stability, their site 1 binding parameters, and the volume and helix propensity of the residue in position 129. The only significant relation we evidenced was the inverse logarithmic correlation between the residual agonism and the global stability (T m ) of the Gly 129 variants (Fig. 8).

DISCUSSION
Since the discovery that transgenic mice expressing the G119R mutant of bovine GH exhibited a dwarf phenotype (43), replacement of the helix 3 Gly by an Arg (or Lys) has been considered as the most direct and efficient way to design receptor antagonists in the PRL/GH family. Soon after this pioneering study, the same strategy was used by Wells and co-workers (18) for generating competitive antagonists of hGH (G120R), by our group (13) for hPRL (G129R), and by Gertler and co-workers (44) for bovine PL (G133K). The mechanism involved for GHR antagonists was elucidated from the crystal structure of the hGH⅐hGHR 2 complex (17). The small side chain of the helix 3 Gly maintains a cleft (so-called "cup") at the helix 1/helix 3 interface (site 2), into which Trp 104 of the receptor docks upon binding; consequently, when a larger side chain is substituted for the natural Gly, docking of the Trp is impaired, leading to a nonfunctional interaction with the second GHR of the GH⅐GHR 2 complex. The same mechanism is assumed to be involved in the PRL-PRLR interaction. This is supported by the role of the homologous Trp (residue 72) at site 2 in the oPL⅐rPRLR 2 complex (40), and by the steric hindrance resulting from the Gly 3 Arg substitution in Del1-9-G129R-hPRL (compare Fig. 2, A and B).
Although all PRL/GH/PL variants mutated at the helix 3 Gly were shown to exhibit antagonistic properties against the PRLR in various in vitro assays, several reports also provided evidence that they retained the ability to activate this receptor to some extent. We have demonstrated that the detection of residual activity often requires the use of sensitive bioassays, i.e. cell systems in which a measurable biological response is observed at low hormone concentrations (2,15,45). As an example, G120K-hGH and G129R-hPRL exhibit no activity in the low sensitivity HL-5 assay (14,46) or in signaling studies (immunoblots) (35), although they are able to stimulate the proliferation of Ba/F-LP cells and of rat Nb2 cells (the two most sensitive bioassays for lactogens) to mid-and sub-maximal level, respectively (12,46,47). It would be misleading to neglect the residual activity of these mutants otherwise referred to as antagonists. First, our ongoing studies indicate that transgenic mice overexpressing G129R-hPRL exhibit phenotypes that are reminiscent of those reported for PRL-transgenic mice, 5 supporting the idea that residual agonism predominates over antagonism in vivo. Second, based on their competitive mechanism of action, PRLR antagonists must be used in molar excess compared with endogenous PRL, i.e. at concentrations at which G129R-hPRL exhibits residual agonism. Obviously, such a situation  (Table 3) have been plotted against their maximal residual agonistic potency as determined using the Ba/F-LP assay expressed as ln of maximal hPRL activity (Table 5). Agonism is inversely correlated to protein stability (R 2 ϭ 0.87).
would be highly detrimental to drug efficiency in a therapeutic context (1).
Abolition of residual agonism has thus become a priority in our research and development process of potent PRLR antagonists. As mentioned in the Introduction, deletion of the nine N-terminal residues in G129R-hPRL recently allowed us to achieve this goal (5). This deletion, which distinguishes the partial agonist G129R-hPRL from the pure antagonist Del1-9-G129R-hPRL, suggested a role for the N terminus in PRLR activation. However, the mechanism involved has remained speculative (5). Indeed, we showed that native hPRL activity was unaffected by N-terminal mutations, whether it involved elongation (Nter-hPRL; Fig. 7C) or truncation (Del1-9-hPRL; see Ref 16). However, when the same N-terminal mutations were performed in the G129R context, the consequences for the biological properties were emphasized (Fig. 7C). Insertion of the AQHPPY motif boosted the maximal agonistic activity of Nter-G129R-hPRL to twice the level of G129R-hPRL, whereas deletion of the N terminus knocked down residual agonism (Del1-9-G129R-hPRL), confirming our previous observations (5, 9 -11). It should be noted that the GHR antagonist G120R-hGH is also a PRLR antagonist (48). As a GH-core protein, it lacks the nine residues homologous to the hPRL N-terminal tail (Fig. 3). Most interestingly, we had previously shown that G120R-hGH is devoid of residual agonism toward the human PRLR (15), further indicating that the N terminus is an essential feature for residual agonism in hPRLR antagonists.
An explanation for the role of the N terminus of hPRL can be proposed on the basis of our new results. The three-dimensional structure of Del1-9-G129R-hPRL (Fig. 1), compared with the WT hPRL structure, showed no major conformational change other than the two mutated features (21). This rules out the possibility that the unique properties of Del1-9-G129R-hPRL result from significant distal alterations. This conclusion is also supported by the fact that site 1 affinity was unaffected by the N-terminal deletion, both in the WT and in the G129R contexts (Table 4). Moreover, global protein stabilities (T m values) of Del1-9-hPRL and Del1-9-G129R-hPRL were also unchanged compared with their N-terminal unmodified counterparts (Table 3). Taken together, these observations suggest that the role of the N terminus is restricted to site 2 binding and that it could be directly involved in the interaction with the second hPRLR chain. This hypothesis is in good agreement with the information gained from the crystal structure of oPL-PRLR 2 , which demonstrated that the hormone N terminus contributed largely to the binding energy at site 2 (40).
Alterations of the hPRL N terminus appear to have a significant impact on biological activity only when site 2 is impaired through a mutation at position 129. Indeed, in the WT hPRL context, the nine-residue N-terminal deletion only led to a slight decrease of site 2 affinity (3-fold as measured by SPR; see Table 4). As for the N-terminal AQHPPY motif insertion, it had no detectable effect in vitro, as measured by SPR (Table 4), although it only slightly, but reproducibly, enhanced global affinity measured using cell lysates (Fig. 5). These limited effects could be explained by the fact that, in the WT hPRL, binding of the second receptor to the hormone site 2 relies principally on the anchoring of Trp 72 PRLR into the Gly cup ( Fig. 2A), inde-pendently of the N terminus. In contrast, when hPRL site 2 is impaired by the bulky Arg 129 side chain (Fig. 2B), the N terminus might still be able to provide enough residual interaction energy to allow a limited second receptor recruitment or to induce a limited reorientation of the second receptor within a pre-formed hPRLR dimer (49,50) leading to a residual agonistic activation. The efficacy of this process could be further enhanced when additional residues leading to more interaction energy are added with the AQHPPY motif insertion, resulting in a clear iterative effect on agonistic activity between Del1-9-G129R-hPRL, G129R-hPRL, and Nter-G129R hPRL ( Fig. 5 and  7C). Unfortunately, we could not correlate the difference in agonistic activity of G129R-hPRL and Del1-9-G129R-hPRL to differences in their affinities at site 2, as the latter were beyond the threshold of detectable interactions for both mutants (Fig. 6D).
The main feature of PRL site 2 is clearly the Gly 129 pocket, as highlighted by the fact that G129R mutation induced at least a 150-fold decrease in site 2 binding affinity in vitro (SPR assays) and reduced the agonistic activity of hPRL by more than 98% as deduced from the luciferase assay (HL-5) (14). This major role prompted us to investigate this region in more detail as, surprisingly, residues other than Arg have never been tested as substitutes for Gly 129 , with the aim of generating pure PRLR antagonists. Indeed, based on the guess that a large and charged side chain residue should be the most effective in disrupting receptor binding, this prototype mutation was always assumed by us and others to be the best within the Gly pocket. To challenge this hypothesis, we performed various substitutions covering a range of physicochemical properties of natural amino acids (see "Results"). All Gly 129 mutants exhibited similar antagonism in the HL-5 assay (IC 50 varying by less than 3-fold; see Fig. 7B), confirming that any residue other than Gly at position 129 transforms hPRL into an antagonist. However, we were surprised to observe that all substitutions (except G129P) led to lower residual agonism than G129R (Fig. 7D). In other words, Arg is actually one of the less efficient substitutions of Gly 129 when attempting to generate a pure antagonist. Val, Leu, and Phe were in this respect the substitutes leading to the lowest residual agonism.
To understand why the Arg substitution was not optimal and whether contacts could persist between the site 2 of G129R-hPRL and the PRLR, we superimposed our three-dimensional structure of Del1-9-G129R-hPRL (which harbors the G129R mutation) on the only currently available structure providing information on the site 2-PRLR interaction, namely the oPL⅐rPRLR 2 ternary complex (PDB code 1F6F) (40). This model (Fig. 9)  Trp side chain could rotate upon G129R-hPRL site 2 binding, thereby disrupting the hydrogen bonds at one side of the cup (Gln 122 and Arg 125 ), while potentially preserving those at the other side (Asp 17 and Arg 21 ). As a result of the high flexibility of the Arg side chain, either the positive charge of Arg 129 G129R-hPRL could interact with N-⑀1 from Trp 72 PRLR or its guanidine group could stack with the aromatic moiety of Trp 72 PRLR . Such hypothetical interactions could help G129R-hPRL maintain a minimum of contacts at site 2 despite the Gly 3 Arg substitution. In Del1-9-G129R-hPRL, the same rearrangements would occur, but these weak contacts alone would probably be insufficient to induce any residual agonism in the absence of interactions involving the N terminus of hPRL (see above).
We then attempted to elucidate why every Gly 129 substitution but G129P resulted in a lower level of residual agonism than G129R. This observation could not be related to differences in the site 1 binding properties of the Gly 129 variants (site 2 binding being undetectable even for the best agonist G129P-hPRL; see Table 4) nor to the volume or helix propensity of the residue at position 129. Strikingly, we observed that the thermodynamic stability of all Gly 129 mutants (but G129P) was significantly improved compared with G129R-hPRL. Even more interesting, we found that residual agonism and protein stability were inversely correlated, as illustrated in Fig. 8. This unexpected observation raised two questions. First, why does the small-to-large replacement of Gly 129 (except Pro) lead to enhanced stability? Second, why does enhanced stability reduce residual agonism? It is well established that glycine residues allow for more regions in the / space to be accessed than for any other residue. Substitutions of Gly 129 therefore potentially constrain the dynamics of the hPRL scaffold, which, as reflected by the diverse set of mutations leading to increased stability, clearly appears to be non-energy-optimized. The vicinity of Gly 129 appears to be highly flexible, as shown by the increased amide hydrogen exchange rates of adjacent residues (39). Furthermore, normal mode analysis of WT hPRL and Del1-9-G129R-hPRL structures, as well as temperature factors (Fig.  1A), predicted the presence of a bending region close to Gly 129 . The amplitudes of these movements, as well as the center of mass, could very well be significantly constrained and shifted, respectively, upon mutation. In good agreement, Keeler et al. (38) observed from NMR relaxation data, high dynamics (nuclear Overhauser effect Ϸ0.5) at position Glu 128 in hPRL. Finally, when superimposing helices 3 from all the structures of GH (PDB codes 1HUW and 1HGU) and GH⅐GHR complexes (PDB codes 1A22, 1HWH, 1HWG, and 3HHR) available in the PDB, a clear difference in the bending of the helix appeared between unbounded and complexed structures (not shown). One can suggest that this movement is restrained/prohibited depending on the Gly 129 mutants and, conversely, that the helix is already properly bent in G129P-hPRL, which could explain the higher residual agonism of this mutant despite the intrinsically hindering effect of introducing a Pro within the cup.
We suggest from these observations and from the stability changes we determined that the intrinsic dynamics of hPRL is a prerequisite for its fully functional interaction with PRLR; global stabilization of hPRL by substitution of Gly 129 by any residue except Pro might either lock site 2 in an unproductive state leaving site 1 unaffected, or interfere with the coupling of site 1 and site 2 suggested by Brooks and co-workers (51). For several other signaling proteins such as GH (52), insulin (53), keratinocyte growth factor (54), and fibroblast growth factor (55), similar positive or negative correlations between the stability of specific regions and their functional properties have been underlined, showing the prominence of global or local structural flexibility in the fine-tuning of their activity.
In conclusion, we report here the first PRL core structure solved by crystallography. This structure of the pure antagonist Del1-9-G129R-hPRL revealed no major structural change compared with WT hPRL. Evidence obtained from our structural, physicochemical, and biological results from this protein and from a diverse set of mutant proteins allow us to propose the following. (i) The glycine pocket is the major component of hPRL site 2. (ii) The N terminus of hPRL is an intrinsic component of site 2, whose importance only becomes clearly apparent when the glycine pocket is impaired because of Gly 129 substitution. (iii) Structural flexibility of hPRL is required for fully functional site 2 binding. In line with these conclusions, we show that complete abolition of residual agonism can be achieved by two very different strategies, either by totally disrupting hormone site 2-receptor contacts, as in Del1-9-G129R-hPRL, or by stabilizing the hormone and constraining its intrinsic flexibility, as in G129V-hPRL.