A Residue Quartet in the Extracellular Domain of the Prolactin Receptor Selectively Controls Mitogen-activated Protein Kinase Signaling*

Background: Discrimination of cytokine receptor signaling pathways is poorly understood. Results: Manipulation of extracellular prolactin receptor interface residues selectively affected activation of the MAPK pathway but not the STAT5 pathway. Conclusion: MAPK pathway activation correlated with apparent ligand·receptor complex lifetimes suggesting pathway-specific lifetime thresholds for activation. Significance: Discrimination of prolactin receptor signaling is controlled by the extracellular homodimer receptor interface. Cytokine receptors elicit several signaling pathways, but it is poorly understood how they select and discriminate between them. We have scrutinized the prolactin receptor as an archetype model of homodimeric cytokine receptors to address the role of the extracellular membrane proximal domain in signal transfer and pathway selection. Structure-guided manipulation of residues involved in the receptor dimerization interface identified one residue (position 170) that in cell-based assays profoundly altered pathway selectivity and species-specific bio-characteristics. Subsequent in vitro spectroscopic and nuclear magnetic resonance analyses revealed that this residue was part of a residue quartet responsible for specific local structural changes underlying these effects. This included alteration of a novel aromatic T-stack within the membrane proximal domain, which promoted selective signaling affecting primarily the MAPK (ERK1/2) pathway. Importantly, activation of the MAPK pathway correlated with in vitro stabilities of ternary ligand·receptor complexes, suggesting a threshold mean lifetime of the complex necessary to achieve maximal activation. No such dependence was observed for STAT5 signaling. Thus, this study establishes a residue quartet in the extracellular membrane proximal domain of homodimeric cytokine receptors as a key regulator of intracellular signaling discrimination.

Cytokine receptors elicit several signaling pathways, but it is poorly understood how they select and discriminate between them. We have scrutinized the prolactin receptor as an archetype model of homodimeric cytokine receptors to address the role of the extracellular membrane proximal domain in signal transfer and pathway selection. Structure-guided manipulation of residues involved in the receptor dimerization interface identified one residue (position 170) that in cell-based assays profoundly altered pathway selectivity and species-specific biocharacteristics. Subsequent in vitro spectroscopic and nuclear magnetic resonance analyses revealed that this residue was part of a residue quartet responsible for specific local structural changes underlying these effects. This included alteration of a novel aromatic T-stack within the membrane proximal domain, which promoted selective signaling affecting primarily the MAPK (ERK1/2) pathway. Importantly, activation of the MAPK pathway correlated with in vitro stabilities of ternary ligand⅐receptor complexes, suggesting a threshold mean lifetime of the complex necessary to achieve maximal activation. No such dependence was observed for STAT5 signaling. Thus, this study establishes a residue quartet in the extracellular membrane proximal domain of homodimeric cytokine receptors as a key regulator of intracellular signaling discrimination.
Cytokine receptors activate many different intracellular pathways, including JAK/STAT, MAPK, Src, and PI3K/Akt, resulting in a wide variety of cellular responses (1)(2)(3). Besides cytokines, various stimuli have been shown to trigger signaling by these receptors, including antibodies, small molecule agonists, or even activating mutations intrinsic to the receptors (see below). However, several studies have shown that the same cytokine receptor can elicit different cellular responses depending on the nature of these stimuli, which implies signaling discrimination between agonistic signals. Typical examples include the family of type I IFNs, whose numerous members exhibit markedly different antiviral versus antiproliferative properties despite acting through the same heterodimeric receptor. In-depth analysis of structures, energetics, and dynamics of ligand⅐receptor complexes revealed that this functional diversity was mainly driven by different complex lifetimes and ligand affinities more than by specific conformational differences (4). Recently, a polyclonal anti-idiotypic antibody directed against porcine growth hormone (GH) 4 was shown to trigger activation of the GH receptor (GHR) expressed in rat hepatocytes (5). However, only a partial response was observed as just the JAK/STAT, but not the MAPK (i.e. ERK1/2), pathway was activated. The molecular mechanism involved was not elucidated (5). Antibody-mediated activation of the thrombopoietin receptor also resulted in a different sig-naling profile compared with thrombopoietin stimulation, which translated into the promotion of cell proliferation at the expense of cell differentiation (6). Activation of the erythropoietin (EPO) receptor (EPOR) using an antibody binding to an epitope of the extracellular domain (ECD) distinct from the EPO-binding site stimulated in vitro erythropoiesis, but nevertheless displayed altered STAT signaling (7). In line with this, the constitutively active EPOR R129C variant (Arg-129 within the ECD replaced with Cys) was shown to activate mainly STAT5B, whereas EPO-triggered EPOR WT activates both A and B isoforms of STAT5 (8). These examples illustrate that in addition to binding circulating ligands, one function of cytokine receptor ECDs may be to participate in selective activation of intracellular signaling pathways. However, the structural and mechanistic features of the ECD regulating this process are poorly understood. Deciphering precise molecular mechanisms of signaling selectivity will no doubt result in substantial advances in our understanding of several diseases and allow the development of a new generation of drugs that exhibit well controlled signaling properties, including pathway selectivity.
The prolactin receptor (PRLR) and the GHR represent archetypes of the so-called homodimeric hematopoietic cytokine receptors (9), i.e. the simplest of type 1 cytokine receptors, made of two identical transmembrane chains exhibiting minimal structural complexity compared with other family members. Their ECDs contain a single cytokine receptor homology module made of two fibronectin type III domains, termed D1 (membrane-distal) and D2 (membrane-proximal) (Fig. 1A). The latter contains the conserved WS motif (Trp-Ser-Xaa-Trp-Ser sequence), a hallmark of type 1 cytokine receptors (10,11). Under physiological conditions, signaling requires the formation of ternary complexes involving one ligand interacting with both receptor chains via two asymmetrical binding sites, termed sites 1 and 2 (Fig. 1A). Cell-based studies have shown that at least parts of the GHR and PRLR population expressed in eukaryotic cells are pre-dimerized in the absence of ligand, suggesting that signaling activation involves ligand-induced conformational changes within the receptor homodimer rather than alteration of receptor stoichiometry (12)(13)(14). For the PRLR, we recently used nuclear magnetic resonance (NMR) spectroscopy to show that in the unbound state the two tryptophan residues of the WS motif adopt a T-stack conformation that stabilizes the receptor off-state, whereas in the ligandbound state these residues form a Trp-Arg ladder characteristic of the receptor on-state (15). Chemical shift perturbation studies following PRL binding to the D2 domain further point to structural rearrangements involving various residues, including specific sites within the receptor-receptor dimerization interface, also called site 3 (Fig. 1B).
In this work, we discovered that a residue quartet positioned at the receptor dimerization interface tunes the level of MAPK signaling while not affecting STAT5 signaling. This quartet includes position 170 that is either a Leu or a Phe in the human (h) and rat (r) PRLRs, respectively (Fig. 1B). Optical and NMR spectroscopic analyses of soluble receptor domains (ECD and D2, both WT and variants) revealed underlying specific local structural changes. These included alteration of a novel aro-matic T-stack between residues from the E-and B-strands of the D2 domain (see Fig. 1B), which promoted selective signaling affecting primarily the MAPK pathway. Furthermore, based on real time analysis of the PRL/PRLR-ECD interactions, in vitro stability of the ternary ligand⅐receptor complex appeared to correlate with both structural (T-stack) and functional (the level of MAPK activation) properties, suggesting the lifetime of the ternary complex to be deterministic for activation. Thus, using the PRLR as a prototypic homodimeric cytokine receptor, this study establishes the dimerization interface as a key regulator of intracellular signaling discrimination. These findings FIGURE 1. PRLR homodimer interface (site 3). A, representation of the 2:1 rPRLR⅐hPRL complex structure (PDB accession code 3NPZ) (25). The two receptors (rPRLR1 and rPRLR2) are shown in gray, and their D1 and D2 domains are labeled in light and dark blue, respectively, and hPRL is shown in pale cyan. The three intermolecular interaction sites are indicated in the structure (sites 1 and 2 between hPRL and the receptor and site 3 between the two receptors). B, magnification of site 3 with blue sticks (light blue on rPRLR1 and dark blue on rPRLR2) identifying interacting residues at the interface, which are at focus in the current work. The A to G strands are identified (see Ref. 15,25 for structural details).
may have a huge impact on future knowledge-based drug development aimed at providing pathway-selective treatment strategies.
Cell Cultures and Transfections-Routine culture conditions, cell transfections, generation of stable populations, and luciferase and proliferation assays have been extensively described (19). The iterative selection scheme used for Ba/F3 cell populations was recently described (20).
SureFire pSTAT5 and pERK1/2 Assays-All reagents were supplied in the SureFire pSTAT5 or pERK1/2 assay kits (PerkinElmer Life Sciences). Various amounts of fresh cell lysates (4 -8 g of protein content, ϳ1-2 ϫ 10 5 cells) were added per well of 384-well plates. Five l of acceptor mix (activation buffer, reaction buffer, and acceptor beads) were added to cell lysates. The plate was incubated under low light conditions for 2 h at room temperature and then 2 l of donor mix (dilution buffer and donor beads) were added. After 2 h of incubation, the plate was read on an EnVision TM device (Perkin-Elmer Life Sciences).
Cell Surface PRLR Binding-Recombinant hPRL was fluorescently labeled using the DyLight 650 NHS Ester kit (catalog 62266) from the ThermoScientific Kit (Pierce). We strictly followed the instructions of the manufacturer. Briefly, hPRL sus-pended in 50 mM sodium borate, pH 8.5 (2 g/ml), was mixed with fluorescent DyLight 650 dyes (1 vial) for 1 h at room temperature, and then the mix was dialyzed overnight in PBS to eliminate free dyes. After measuring A 280 (protein) and A 650 (fluorescent dye) using a Nanodrop, the degree of labeling of fluorescent hPRL (hPRL fluo ) was calculated as ϳ0.75 (mol of dye/mol of protein). For cell surface PRLR binding experiments, stable Ba/F3 populations were starved in 1% FCS medium for ϳ2 h, and then 2⅐10 6 live cells (per condition) were washed and resuspended in 500 l of RPMI 1640 medium, 0.5% BSA. Unlabeled hPRL (3 l from 2 g/ml stock), hPRL fluo (3 l from 2 g/ml stock), or both (3.5 molar excess of hPRL over hPRL fluo ) were added to each well. After 210 min of incubation at 4°C (to avoid receptor internalization) in the dark, cells were pelleted in 1 ml of cold PBS, 2% BSA and added 1 l of SYTOX Blue before FACS analysis (LSR Fortessa SORP, BD Biosciences).
Fluorescence Spectroscopy-Conformational stability was measured from GdmCl-induced changes in intrinsic tryptophan and tyrosine fluorescence with excitation at 280 nm, protein concentrations of 1-3 M (rECD and variants) or 5 M (hD2 and variants) in a 0.5-cm quartz cuvette, 10 M NaH 2 PO 4 / Na 2 HPO 4 , pH 7.4, 25°C. The concentration of GdmCl was measured by refractive index. The fluorescence emission intensity at unfolded max was plotted as a function of increasing GdmCl concentration, normalized to fraction of unfolded population, and fitted to a two state unfolding as described (21).
Surface Plasmon Resonance-rPRLR-ECD WT and rPRLR-ECD L170F were immobilized covalently onto nitrilotriacetic acid-derivatized sensor chips, as described (23)(24)(25). Surface densities of 3,000 -4,000 resonance units (resonance unit Ϸ pg mm Ϫ2 ) were obtained. Binding assays were performed in trip-licate on a BIAcore 2000 instrument equilibrated at 25°C. The PRLR-ECD surfaces were first saturated with hPRL (250 nM), and then rPRLR-ECDs (0.4 -50 M) were injected for 3 min at 20 l/min onto the preformed PRLR-ECD⅐PRL complexes, followed by a 5-min dissociation period. The site 2 ϩ 3 association and dissociation profiles were double referenced using the Scrubber 2.0 software (Bio-Logic Software) (i.e. both the signals from the reference surface (ethanolamine-derivatized) and the signals from blank experiments (using running buffer instead of FIGURE 2. Bioactivity of rPRLR interface variants in HEK293 and Ba/F3 cells. A, HEK293 cells transiently transfected with expression vectors encoding rPRLR WT or rPRLR variants were starved and then stimulated with 1 g/ml hPRL (15 min). Total PRLR (U5 mAb), phosphorylated and total STAT5 (from the same gel), and phosphorylated and total ERK1/2 (from a different gel) were analyzed by immunoblotting. B, HEK293 cells were transiently co-transfected using expression vectors encoding rPRLR WT or rPRLR variants, and the LHRE-luciferase reporter plasmid. For each variant, the luciferase values obtained at varying PRL concentrations were normalized to the unstimulated condition for comparison (means Ϯ S.E., n ϭ 3 in triplicate). C, stable populations of Ba/F3 cells expressing rPRLR WT or rPRLR variants were obtained and selected as described in the text and Table 1. Proliferation of populations 2 in response to a range of hPRL concentrations was measured (WST-1 reagent) in three independent experiments performed in triplicate (mean Ϯ S.E.). D, starved cells were treated using 1 g/ml hPRL (15 min), then analyzed for phosphorylated and total STAT5 (upper), and ERK1/2 (lower) by immunoblotting (two noncontiguous parts from the same immunoblot are shown and separated by a dividing line). E, densitometric analysis of STAT5 and ERK1/2 activation was averaged from five separate immunoblotting experiments. The phosphorylated/total ratio obtained in Ba/F3-rPRLR WT cells stimulated with hPRL was set to 100% (mean Ϯ S.E.). F, basal and stimulated levels of STAT5 and ERK1/2 activation were determined using semi-quantitative SureFireா homogeneous assays (AlphaScreenா technology, n ϭ 3). E and F, stimulated conditions were compared by one-way ANOVA. p values are represented by asterisks, using *, p Ͻ 0.05; **, p Ͻ 0.01. protein) were subtracted). The site 2 ϩ 3 steady-state SPR responses (R eq ) were plotted against the concentration of rPRLR-ECD (site 2) and fitted using Equation 1, where K d is the equilibrium dissociation constant and R max is the maximal binding capacity of the surface.
Statistics-Statistical analyses were performed using GraphPad Prism software. As indicated in the figure legends, we used one-way ANOVA followed by Dunnett's post hoc test or two-way ANOVA followed by Bonferroni post hoc test. p values are represented using 1 (p Ͻ 0.05), two (p Ͻ 0.01), or three (p Ͻ 0.001) symbols (*).

PRLR Interface Residues Control
Signaling Selectivity-Using oriented SPR to measure ternary complex stabilities (24), we recently identified Lys-168 and Phe-170 as hot spot residues of the rPRLR dimerization site ( Fig. 1B) (25). To investigate the role of these inter-molecular contacts for signal transfer, alanine substitutions were generated in the full-length receptor (rPRLR K168A and rPRLR F170A ), and their capacity to activate the canonical STAT5 pathway in transiently transfected HEK293 fibroblasts was analyzed ( Fig. 2A). No STAT5 phosphorylation was detected in the absence of PRL, ruling out ligand-independent activity of these receptor variants. When stimulated with 1 g/ml hPRL for 15 min, all variants retained the ability to induce strong STAT5 phosphorylation similarly to that of the rPRLR WT ( Fig. 2A). Using the classical LHRE-luciferase reporter gene assay to monitor STAT5 transcriptional activity (16), we obtained overlapping curves in PRL dose-response experiments irrespective of the receptor variant (Fig.  2B). These observations demonstrated that activation of the STAT5 pathway was not altered by these interface substitutions. However, in contrast to the STAT5 pathway, the ERK1/2 pathway was not activated by hPRL stimulation in HEK293 cells ( Fig. 2A, bottom), in agreement with our previous observations (19,20,26).
We previously reported that stably transfected Ba/F3 cells express low receptor density and thus constitute a much more sensitive bioassay to discriminate bioactivity of mutated PRLs or PRLRs (27,28). To generate stable cell populations expressing rPRLR K168A or rPRLR F170A , we followed the three-step selection scheme recently described (20) with survival data summarized in Table 1. All receptor plasmids were successfully stabilized in Ba/F3 cells upon G418 addition, keeping IL-3 as the survival/growth factor (selection 1). When hPRL was substituted for IL-3 (selection 2), all cell populations could be propagated in routine culture conditions, confirming that interface rPRLR variants triggered sufficient levels of intracellular signaling to ensure some degree of cell survival/proliferation. However, hPRL dose-response proliferation assays performed in low serum conditions (Fig. 2C) revealed that cells expressing rPRLR F170A or rPRLR K168A exhibited ϳ5to ϳ17-fold higher EC 50 values compared with Ba/F3-rPRLR WT cells, respectively ( Table 2). When stimulated with 1 g/ml hPRL (corresponding to the plateau in proliferation assays for all receptor variants, see Fig. 2C), the interface variants induced similar levels of STAT5 phosphorylation compared with rPRLR WT (Fig. 2D), suggesting that PRLRs present at the cell surface were functional. We therefore hypothesized that differential intracellular signaling could explain the observed proliferative differences. In contrast to HEK293 cells, Ba/F3 cells expressing hPRLR WT exhibited strong activation of the ERK1/2 pathway in response to hPRL stimulation (19,20,26), although other PRLR pathways (STAT3, Akt, and Src) displayed much more discrete responses (data not shown). Interestingly, Ba/F3 cells expressing rPRLR F170A or rPRLR K168A exhibited markedly lower levels of ERK1/2 phosphorylation compared with rPRLR WT (Fig. 2D). Densitometric quantification of five independent immunoblotting experiments (Fig. 2E) were fully confirmed by semi-quantitative SureFire homogeneous assays using AlphaScreen technology (Fig. 2F). These results demonstrated that PRLR interface residues selectively control activation of the MAPK pathway. This was further confirmed by the experiments described below.
Residue 170 Directs the PRLR Species-specific Signaling Profile in Response to Ligands of Various Origins-As shown in Table 1, rPRLR WT was able to promote Ba/F3 cell survival across PRL withdrawal (selection 3). Intriguingly, this property was lost in both rPRLR interface variants (Table 1). Similarly, cells expressing hPRLR WT did not survive in the absence of hPRL supply, which emphasized a marked difference between receptor species. We suspected that the receptor-specific capacity to support cell survival might reflect a different sensitivity to bovine (b) PRL present in the culture medium via the   addition of serum. To address this point, we compared doseresponse proliferation assays performed in low serum conditions (1% FCS) using hPRL versus bPRL (Table 2). First, these assays revealed a higher sensitivity of rPRLR WT to bPRL compared with the rPRLR variants, providing a strong rationale for the survival of Ba/F3-rPRLR WT cells in hPRL-free growth medium (see Table 1). Second, comparison of the EC 50 values of proliferation curves involving hPRL versus bPRL revealed that in contrast to rPRLR K168A , the growth-promoting properties of rPRLR F170A were highly dependent on the ligand origin ( Table  2, right column). As shown below, investigation of rPRLR F170A signaling properties confirmed that this variant was indeed much less responsive than rPRLR WT to bPRL and, notably, to the homologous rPRL. Based on these findings, we aimed to further investigate the importance of the residue located at position 170, Leu in the hPRLR versus Phe in the rPRLR. Although the data described above indicated that the F170A substitution conferred human receptor-like properties to the rPRLR, we found that the L170F substitution conferred rat bio-characteristics to the hPRLR. Indeed, Ba/F3-hPRLR L170F cells (i) survived and proliferated in growth medium without hPRL supplementation (Fig. 3A), (ii) exhibited detectable STAT5 signaling in response to serum (Fig. 3B, lanes 6 and 7), and (iii) lost the species-specific activation profiles of STAT5 and ERK1/2 in response to PRLs of var-ious origins (Fig. 3B, lanes 8 -10). None of these properties were observed for cells expressing hPRLR WT (Fig. 3, A and B, lanes  1-5). The basal activation of STAT5 signaling observed in Ba/F3-hPRLR L170F cells in the presence of serum (no hPRL added) was abolished by treating them with the pure hPRLR antagonist Del1-9-G129R-hPRL (Fig. 3C), confirming a PRLRmediated mechanism (18). These data collectively showed that the residue at position 170 directly controls the species-specific properties of the PRLRs.
hPRLR WT ) displayed marked species preferences toward the ligand and reduced ERK1/2 activation, although STAT5 activation was unaffected. This underlines the key role of the residue at position 170 in the selective control of PRLR signaling properties.
Residue 170 Controls PRLR Turnover in Ba/F3 Cells-Because changes in signaling intensity may be linked to receptor turnover, we subsequently investigated whether manipulation of position 170 affected the amount of receptors available at the cell membrane. First, we performed FACS analyses using hPRL fluo (see "Experimental Procedures") to assess cell surface expression of the PRLR variants. Fluorescent signal specificity was assessed by competing hPRL fluo with 3.5-fold excess unlabeled hPRL (Fig. 5A). Interestingly, much lower cell surface hPRL fluo binding in Ba/F3-hPRLR L170F cells compared with Ba/F3-hPRLR WT cells was observed, although it was the opposite for Ba/F3-rPRLR F170A cells compared with Ba/F3-rPRLR WT cells (Fig. 5B). Because lower cell surface binding was not correlated with lower PRLR affinity (see below), but rather the reverse, cell surface expression and degradation may be intimately linked to receptor activation and thus indirectly to position 170. We therefore investigated the degradation profiles of the different receptor variants using immunoblotting with different antibodies directed against the hPRLR-ECD (1A2B1) or -ICD (H300) or the rPRLR-ECD (U5) (Fig. 5, C and  D). In routine culture conditions, the degradation patterns of hPRLR WT and hPRLR L170F appeared to be similar. The smear at ϳ60 -70 kDa was recognized by the H300 antibody ( Fig. 5C) but not by the 1A2B1 mAb (data not shown), indicating that it corresponded to the ICD of the receptor. Of note, the band corresponding to full-length hPRLR (ϳ90 kDa) was much less intense for hPRLR L170F than for hPRLR WT (Fig. 5C), suggesting that L170F promoted faster receptor degradation. Accordingly, densitometric analysis of two independent experiments confirmed the lower full-length/degraded receptor ratio of hPRLR L170F compared with hPRLR WT (data not shown). In agreement with this, the F170A substitution in the rPRLR had the opposite effect as it clearly prevented degradation of the full-length receptor (Fig. 5C). Finally, PRLR degradation profiles were not altered by PRL stimulation (Fig. 5D) suggesting that they are intrinsically linked to receptor species and specifically to the residue present at position 170.
Longer Persistence Time of the Ternary Complex Parallels Efficient ERK1/2 Activation-We have previously shown using our established SPR setup that the F170A substitution was detrimental to ternary hPRL⅐rPRLR-ECD complex formation due to ϳ10-fold reduced affinity at sites 2 ϩ 3 (25). We hypothesized that this feature could be correlated with the reduced ability of rPRLR F170A to activate the ERK1/2 pathway observed in this study. We therefore investigated whether the L170F substitution that increased the ability of the hPRLR to activate ERK1/2 also increased the site 2 ϩ 3 affinity of the ternary complex. Indeed, SPR experiments showed that hPRLR-ECD L170F had an increased affinity for the preformed 1:1 hPRL⅐rPRLR-ECD complex by Ͼ2-fold (n ϭ 3) irrespective of the nature of the immobilized ECD (WT or L170F) ( Table 3). This increase in affinity could be correlated to an ϳ2-fold increase in the lifetime of the complex (Fig. 6). Although the gain in affinity provided by the L170F substitution was less pronounced than the loss of affinity resulting from the F170A substitution, these results strongly suggested that the longer the lifetime of the ternary complex, the higher the level of ERK1/2 activation. Thus, by modulating complex lifetimes, the chemical structure of the residue at position 170 appears to play a key role in regulating intracellular signaling properties.
Residue Quartet Modulates Structure and Stability of the ECD-To understand why PRLR signaling was so affected by single substitutions in the D2 domain, we first analyzed the structures and stabilities of the ECD variants in the rPRLR background. Previously, the overall fold of the rPRLR-ECD interface variants (K168A and F170A) were evaluated from their far-UV circular dichroism (CD) spectra (25). Because of the presence of aromatic residues that are spatially close to each other in the ECD, this gives rise to a through-space coupling of their electric dipoles. This results in complex and characteristic CD spectra with a characteristic splitting. Such coupling, known as exciton coupling, occurs in the D2 domain and stems from residues Trp-191 and Trp-194 of the WS motif. Thus, the CD spectrum of rPRLR-ECD has a positive ellipticity at 230 nm, which obscures the signature from the dominant ␤-strand structure (15). Interestingly, both rPRLR-ECD K168A and rPRLR-ECD F170A gave rise to a smaller contribution from aromatic exciton couplings, suggesting the following: (i) the substitutions interfered with the organization of the WS motif; (ii) a contribution from additional aromatic exciton couplings of an opposite sign at the same wavelength appeared, or (iii) couplings of a similar sign disappeared. To address whether substitutions affected protein stabilities, we measured the conformational stabilities of rPRLR-ECD WT versus rPRLR-ECD K168A and rPRLR-ECD F170A using fluorescence spectroscopy ( Fig. 7A and Table 4). The conformational stability of rPRLR-ECD WT was large, ⌬G N-D H2O ϭ Ϫ32 Ϯ 5 kJ mol Ϫ1 , suggesting a stable scaffold. The rPRLR-ECD responded remarkably differently to the substitutions. Whereas, the stability of rPRLR-ECD K168A was not significantly altered compared to the WT receptor, rPRLR-ECD F170A was severely destabilized by 20 kJ mol Ϫ1 (Table 4).
To obtain site-specific insight into the possible conformational changes of the variants and to the extent of the involvement of the WS motif, we transferred the substitutions into the   Table 3), is shown for all complexes investigated (n ϭ 2 experiments) as follows: C, hPRLR-ECD WT /hPRLR-ECD WT (blue) and hPRLR-ECD L170F /hPRLR-ECD WT (red); D, hPRLR-ECD WT /hPRLR-ECD L170F (blue) and hPRLR-ECD L170F /hPRLR-ECD L170F (red).

TABLE 3 Equilibrium dissociation constants, K d (in M), for the interactions between PRLR-ECD in solution and immobilized PRLR-ECD⅐hPRL 1:1 complex
The scheme in the upper left corner describes the experimental setting (see Refs. 22-24 for details). The experiment was performed using hPRLR and rPRLR ECDs (WT and variants), as indicated. $ See Fig. 6 for more information. # Data are from Van Agthoven et al. (25). * Increased versus decreased affinity of the variants is symbolized by (1) versus (2), respectively. hPRLR-D2 scaffold for which NMR assignments are available (which is not the case for rPRLR-D2) (15). To match the alanine and rat/human substitutions characterized in functional assays, we created hPRLR-D2 K168A , hPRLR-D2 L170A , and hPRLR-D2 L170F . These substitutions led to changes in the CD spectrum similarly to the rPRLR-ECD variants (25) with a pronounced effect on the aromatic exciton band at 230 nm (Fig. 8A). However, whereas hPRLR-D2 K168A and hPRLR-D2 L170A both led to a reduction in ellipticity compared with hPRLR-D2 WT , hPRLR-D2 L170F led to an increase at 230 nm. Inspection of the structural environment around Lys-168 and Leu-170 in the threedimensional structures identified two aromatic residues, Tyr-122 and Trp-124 (Fig. 8B), which could contribute to the observed change in the CD spectrum. Substitution of Tyr-122 with alanine led to a significant decrease of the positive ellipticity at 230 nm (Fig. 8A), confirming a contribution from this residue to the aromatic exciton couplings observed in the CD spectrum.
To explore in detail the involvement of these aromatic residues and to locate whether the site-specific changes imposed by the interface substitutions (residues 168 and 170) also involved the WS motif, we measured the changes in NMR chemical shifts of the indoles and of backbone amides of the three hPRLR-D2 variants in comparison with hPRLR-D2 WT . Such changes report on both chemical and conformational changes   imposed by the substitutions. The indole NH region of the spectra clearly showed that Trp-191 and Trp-194 of the WS motif were unaffected by the substitutions (Fig. 8C). Instead, the chemical shifts of Trp-124 indole changed significantly as suggested from the CD data (Fig. 8A), implicating Trp-124 to be related to the changes observed in the CD spectra. Thus, the CD and NMR analyses confirmed conformational changes of Tyr-122 and Trp-124 in response to the hot spot interface substitutions.
To locate the regions affected by the substitutions, we then compared the backbone chemical shifts of the three variants to those of hPRLR-D2 WT (Fig. 9). As expected from the indole analysis, no significant changes were seen in the WS motif (Fig.  9, marked by the red bar) in accordance with the fact that none of the variants exhibited ligand-independent activity, and hence that the off-state of the receptor was in principle preserved (15). From the combined changes in chemical shifts, residues whose chemical shifts were affected more than one S.D. from the trimmed average were highlighted on the structure of hPRLR-D2 WT (Fig. 9, right panels). For all three variants, changes were observed at the site of the substituted residue and locally within 5 Å from this. Moreover, all three substitutions resulted in chemical shift changes for residues of the A-and B-strands (in particular of Trp-124 and Tyr-122) and the loop connecting them (Fig. 9, green underlines). Interestingly, the C-terminal membrane proximal residues were also affected in all variants.
Because the chemical shifts from the side chain of Trp-124 were significantly affected in the variants, particularly in hPRLR-D2 K168A (Fig. 8C), and because the ellipticity at 230 nm for this variant almost disappeared (Fig. 8A), this suggested the presence of an exciton coupling with the opposite sign. Thus, removal of the lysine side chain in hPRLR-D2 K168A would allow room for alterations in the interactions between Tyr-122 and Trp-124, where a rotation of Tyr-122 would lead to the mirror image and hence exciton couplings of the opposite sign. In contrast, in hPRLR-D2 L170F the side chain of Trp-124 was mostly unaffected, but an increase in ellipticity at 230 nm was observed. This suggests a stabilization of the interaction between Tyr-122 and Trp-124. By the same reasoning, the L170A substitution would then lead to a destabilization of their interaction and hence a lowering of the 230 nm intensity. This is indeed what was observed and correlates well with the strong destabilization seen in the rPRLR-ECD upon substitution of Phe-170 by alanine (Table 4). We subsequently measured the stabilities by chemical denaturation of hPRLR-D2 L170F and hPRLR-D2 L170A and compared them with hPRLR-D2 WT (Fig.  7B and Table 4) (15). None of these substitutions significantly destabilized the human D2 domain. For hPRLR-D2 L170A the slope of the transition (m), which reports on the degree of newly exposed surface area, was smaller than for hPRLR-D2 WT , and the midpoint of the transition (C m ) larger. Thus, fewer surfaces were exposed, suggesting that the folded state had some local destabilization around the mutation site, which was compensated for by an increase in entropy, resulting in the same global stability.
In summary, the interface substitutions in both rPRLR and hPRLR strongly affected the local structural geometry in a quar-tet of highly conserved residues consisting of Tyr-122, Trp-124, Lys-168, and Leu/Phe-170 (Fig. 8B). This quartet connects the E-and B-strands of the receptor ECD. Consistently, structural effects were noticed in the A-strand, B-strand, the AB-loop, and in the C-terminal of the D2 domain, and these latter two parts of the receptor ECD are notably closest to the membrane. Remarkably, the substitution-specific effects on ERK1/2 signaling nicely paralleled the structural alterations of this quartet.

DISCUSSION
Despite the core knowledge obtained over recent years on cytokine receptor activation (3,29), the molecular mechanisms and the structural details linking ligand binding to intracellular signaling events are still poorly understood. More enigmatic is to comprehend ligand-specific signaling profiles through the exact same receptor. Using the PRLR as a prototypic homodimeric cytokine receptor, we here demonstrate that the receptor interface selectively controls activation of downstream signaling pathways. Specifically, alanine substitution of two interface residues (Lys-168, Leu/Phe-170) left the STAT5 pathway unaffected, although signaling through ERK1/2 was significantly decreased (Fig. 2, D-F). This underlines the critical role of these residues in signaling pathway discrimination. In contrast to position 168, position 170 is occupied by different amino acids in the rat (Phe) versus the human (Leu) PRLR. Remarkably, we showed the following: (i) this position conferred the species with intrinsically different levels of ERK1/2 activation (Figs. 2E versus 3D); (ii) the very same position also conferred hormone species specificity to the receptors (Fig. 4); and (iii) these intrinsic functional properties could be transferred from one receptor species to the other (and vice versa) by changing only the chemistry at this position. Importantly, the effects of substitutions on ERK1/2 activation were inversely correlated to the PRL binding capacity at the cell surface (Fig. 5B) as well as receptor turnover (Fig. 5, C and D), preventing biased interpretation of the results. Thus, together these data identify a key role of position 170 in the control of intrinsic species-specific signaling properties of the PRLR.
A parallel between our findings on the PRLR and those recently reported by Waters and co-workers (30) for the GHR can be made. This group reported that mutations introduced within the FЈGЈ loop of the GHR-ECD also altered MAPK signaling without affecting the STAT5 pathway (30). Because the regions and residues involved in both receptors are distinct, we aimed to understand the structural mechanisms involved in the PRLR. To that end, NMR and other spectroscopic data were collected on WT and variants of the D2 domains and ECDs from human and rat PRLRs, respectively. The structural data suggested at least two different ways of modulating intracellular signaling. The first involved destabilization of the ECD/D2 scaffold as seen severely for rPRLR-ECD F170A and more locally for hPRLR-D2 L170A . The second and most predominant way was through alteration of the local structure around Tyr-122 and Trp-124 as observed for the K168A and L170F substitutions. From solving the NMR structure of hPRLR-D2 WT , we observed that Lys-168 and Tyr-122 interacted strongly, e.g. they had a number of mutual strong nuclear Overhauser effects (NOEs) (15). In contrast, we did not observe NOEs from these residues to the ring of Trp-124, indicating that this ring points away from the quartet. Thus, the Lys-168 and Tyr-122 interaction ensures a fixed orientation of the aromatic ring of Tyr-122. When the lysine side chain is removed in hPRLR-D2 K168A , the two aromatic rings of Tyr-122 and Trp-124 are free to reorient. This reorientation manifests in diagnostic changes in the CD spectrum revealing exciton couplings with negative ellipticity in hPRLR-D2 K168A , canceling the positive one originating from the WS motif. This structure coincides with decreased MAPK signaling (alanine substitutions). When Leu-170 is replaced by a Phe, the interaction between Tyr-122 and Trp-124 is instead stabilized, which is concluded from the increase in the positive exciton coupling. This structure coincides with increased MAPK signaling (human/rat substitutions). Remarkably, alanine substitution at position 170 destabilizes either the entire scaffold (rPRLR) or the local structure (hPRLR), leading to a destabilization of the Tyr-122/Trp-124 interaction. Again, this destabilization coincides with decreased MAPK signaling. The intrinsic structure and interaction network involving this residue quartet (Tyr-122, Trp-124, Lys-168, and Leu/Phe-170) thus seems to tightly regulate the maximal intensity of MAPK signaling triggered by ligand-bound PRLR variants. In the ternary PRL⅐receptor complex (Fig. 1B), Trp-124 is only involved in the receptor interface from PRLR2 (25); therefore, it is likely that the regulatory mechanism involving the residue quartet only applies to that receptor.
Signaling is tuned by the affinities between the ligand and the receptor complex. These affinities, or more righteously the offrates (k off ), determine the mean lifetime of a given complex (t1 ⁄ 2 ϭ ln2/k off ; mean lifetime ϭ 1/k off ), which must be long enough to trigger signaling. The two signaling pathways monitored here differ significantly in the number of steps involved. It is well known that the JAK2 tyrosine kinase is constitutively bound to the PRLR-ICD (31), and that STAT5 recruitment occurs very rapidly once the receptor is tyrosine-phosphorylated (32). In contrast, MAPK signaling requires multiple protein/protein interactions in the upstream cascade (Lyn/Fyn, PLC␥, Grb2, SHC, SOS, Ras, Raf, MEK, and ERK) (30,33). Accordingly, the time for half-maximal stimulation of STAT5 by the PRLR was recently established as ϳ3 times shorter compared with ERK1/2 (ϳ3 min versus ϳ10 min) with a small lag phase for ERK1/2 activation (33). This indicates either that signal relay takes time or that a longer lifetime of the functional PRL⅐PRLR ternary complex is needed to elicit maximal ERK1/2 activation compared with STAT5 activation, or both. The PRLR interface substitutions selectively altered activation of the ERK1/2 pathway. To demonstrate that this results from different lifetimes of the ternary complexes, precise measurements of the off-rates of PRLR2 interaction (sites 2 ϩ 3) would ideally be needed. However, these dissociation rates are in all cases higher than the limit accessible to SPR measurements (Ͼ0.25 s Ϫ1 ) (23)(24)(25). If we assume that the difference in K d values between the variants and the WT receptors (Table 3) originate exclusively from variations in k off (with k on being unaffected by the substitutions), we tentatively set an upper ϳ10-fold limit for the differences in the apparent mean lifetime of the ternary complexes involving rPRLR K168A or rPRLR F170A compared with rPRLR WT . In line with this, we would therefore expect that hPRLR L170F , which exhibited a stronger ERK1/2 phosphorylation than hPRLR WT , would have a lower K d value at sites 2 ϩ 3 and hence result in a longer complex lifetime compared with WT. This was indeed confirmed ( Table 3).
The complex lifetime concept was previously proposed as a mechanism for the functional diversity of the multiple type I IFNs (4). For the homodimeric cytokine receptor family, previous studies involving ligands and receptors of the PRL/GH family have supported that a minimal time of homodimer persistence was needed to elicit cellular responses (34 -37). As it was subsequently demonstrated that both the GHR (12) and the PRLR (13,14) are pre-dimerized at the cell membrane, this should translate into minimal lifetime of ternary complex. Although this concept was built on the analysis of global cell responses (e.g. cell proliferation), we suggest a model where the complex lifetime of homodimeric cytokine type 1 receptors discriminates downstream signaling pathways (Fig. 10). The proposed model defines two different ranges of receptor signaling Residues with chemical shift differences higher than the trimmed average plus 1 S.D. are marked in red on the structure of unbound hPRLR-D2 WT (PDB accession code 2LFG). Residues more than 5 Å of the substituted residue are shown in red sticks. The substituted residues are shown as red space-filling atoms; the green bars indicate regions with chemical shift changes above average plus 1 S.D. and include the A-B strand and loop as well as the C terminus. sensitivity. One range, which we term the dynamic range, will not respond to an increase in the complex lifetime as maximum activation has already been reached. In the second range called the regulatory range, any change to the lifetime will impact signaling. With respect to the PRLR (rat or human), the lifetimes of the WT complexes are in the dynamic range for STAT5 activation, although they are within the regulatory range for MAPK activation. This model integrates the intrinsically higher MAPK activation by the rat compared with the human PRLR as well as the modulation (increased for hPRLR or decreased for rPRLR) of MAPK signaling by the interface hot spot variants. Likewise, because STAT5 activation is far within the dynamic range for both species, no effects were observed by the substitutions. This model may also explain why MAPK activation is lost first when unconventional receptor agonists are used (5).
The results presented in this study establish the dimerization interface as a key regulator of signaling pathway discrimination. For the PRLR, we propose a structural mechanism involving a residue quartet among which the residue occupying position 170 appears to play a major regulatory role. The regulation by the four residues seems to originate from the PRLR2 receptor in the ternary complex. Although this residue quartet is not conserved among homodimeric type 1 cytokine receptors (38), it is striking that the residue topologically equivalent to position 170 in the GHR, Tyr-200, is one of the four hot spot residues defining the central core of the GHR dimerization interface (39). This clearly suggests common molecular mechanisms. Whether GHR Tyr-200 also specifically tunes MAPK signaling is unknown because the functional importance of this residue was revealed using a c-fos luciferase reporter construct that integrates both STAT5 and ERK1/2 signaling (40). With respect to the PRLR, it is noteworthy that three of the four residues constituting the quartet were fully conserved across evolution, whereas the regulatory position 170 harbors either a Phe (rat, mouse, cow, sheep, red deer, and possum) or a Leu (primates, pig, dog, and rabbit) (41). This raises the question of the potential physiological benefit of having intrinsically higher levels of ERK1/2 activation in some species versus others. Finally, the proposed relationship between complex lifetime, signaling potency, and discrimination may also have pathophysiological relevance. In tumor contexts, the high glycolytic activity and proton extrusion by cancer cells is known to acidify the extracellular milieu (42). Knowing that PRL binding to the PRLR is strongly pH-dependent (43)(44)(45), the reduction of complex lifetime due to tumor acidification should reduce signaling to such an extent that it ultimately brings STAT5 activation into the regulatory range (Fig. 10). In good agreement, a recent study showed that acidosis, pH 6.5, disrupted PRLR signaling in various breast cancer cell lines, providing a mechanistic rationale for the loss of STAT5 phosphorylation frequently observed in aggressive breast cancers (42). Whether STAT5 and ERK1/2 pathways exhibit differential sensitivity to progressive acidosis remains unknown, but such results may further corroborate our model (Fig. 10).
Finally, from a decomposition of function, structure, and affinities of PRLRs and interface variants, we propose that within homodimeric type 1 cytokine receptor ECDs, so-called signaling hot spot residues control the extent and duration of the structural propagations that are necessary to disseminate the signal from the ECD to the ICD. One set of these residues is part of the dimerization interface of the ECDs, and they play key roles in determining the off-rates (k off ) and hence the complex lifetime. The identification of such residues in different cytokine receptors will have huge potential. Generally, in light of our findings they will significantly affect future structure-directed drug development strategies aimed at providing pathway-selective treatment strategies.