Structural Basis for Hormone Recognition by the Human CRFR2α G Protein-coupled Receptor*

The mammalian corticotropin releasing factor (CRF)/urocortin (Ucn) peptide hormones include four structurally similar peptides, CRF, Ucn1, Ucn2, and Ucn3, that regulate stress responses, metabolism, and cardiovascular function by activating either of two related class B G protein-coupled receptors, CRFR1 and CRFR2. CRF and Ucn1 activate both receptors, whereas Ucn2 and Ucn3 are CRFR2-selective. The molecular basis for selectivity is unclear. Here, we show that the purified N-terminal extracellular domains (ECDs) of human CRFR1 and the CRFR2α isoform are sufficient to discriminate the peptides, and we present three crystal structures of the CRFR2α ECD bound to each of the Ucn peptides. The CRFR2α ECD forms the same fold observed for the CRFR1 and mouse CRFR2β ECDs but contains a unique N-terminal α-helix formed by its pseudo signal peptide. The CRFR2α ECD peptide-binding site architecture is similar to that of CRFR1, and binding of the α-helical Ucn peptides closely resembles CRF binding to CRFR1. Comparing the electrostatic surface potentials of the ECDs suggests a charge compatibility mechanism for ligand discrimination involving a single amino acid difference in the receptors (CRFR1 Glu104/CRFR2α Pro-100) at a site proximate to peptide residue 35 (Arg in CRF/Ucn1, Ala in Ucn2/3). CRFR1 Glu-104 acts as a selectivity filter preventing Ucn2/3 binding because the nonpolar Ala-35 is incompatible with the negatively charged Glu-104. The structures explain the mechanisms of ligand recognition and discrimination and provide a molecular template for the rational design of therapeutic agents selectively targeting these receptors.

The corticotropin releasing factor (CRF) 4 /urocortin (Ucn) family of peptides comprises four members in mammals in-cluding CRF, Ucn1, Ucn2, and Ucn3 that exert their actions through either of two cell surface G protein-coupled receptors (GPCRs), CRFR1 and CRFR2 (1)(2)(3). Distinct genes encode the four peptides and two receptors (4 -9). The peptides form ␣-helices that are structurally very similar (10), but they exhibit unique pharmacological profiles. CRF activates both receptors but is more potent at CRFR1. Ucn1 exhibits equal potency for both receptors, whereas Ucn2 and Ucn3 are selective for CRFR2. CRFR1 is highly expressed in the CNS and is also present in peripheral tissues. CRFR2 is expressed in the CNS and peripheral tissues including the heart, vasculature, skeletal muscle, and gastrointestinal tract. CRF plays a central role in coordinating endocrine, autonomic, and behavioral responses to stress via activation of CRFR1 in the brain. The Ucn peptides exert actions in central and peripheral tissues where they modulate cardiovascular, immune, gastrointestinal, and reproductive functions in response to stress (11). The CRFR2-selective Ucn2 and -3 have recently emerged as important metabolic regulators that link stress and energy homeostasis (12)(13)(14)(15). Urocortins can also affect tumor biology as a consequence of the CRFR2 role as a suppressor of vascularization (16). Exogenously administered Ucn1 or Ucn2 inhibited tumor growth in mice (17,18). The involvement of CRF/Ucn family peptides and their receptors in diverse pathophysiological states has generated intense interest in developing therapeutic agents that selectively target the receptors. CRFR1-selective antagonists have the potential to treat anxiety, depression, and related disorders (19), and CRFR2-selective agonists have the potential to treat heart failure (20), hypertension (21), and other disorders.
The CRF receptors belong to the class B/secretin family GPCRs that includes the receptors for parathyroid hormone (PTH), calcitonin, glucagon, glucagon-like peptide 1 (GLP1), and several other clinically important peptides (22). The receptors have a bipartite domain structure consisting of an N-terminal extracellular domain (ECD) of about 100 -150 amino acids linked to a 7-transmembrane helical bundle domain embedded in the plasma membrane. Peptide binding spans the two domains, with the C-terminal portion of the peptide binding to the ECD to impart high affinity and the N-terminal portion binding to the 7-transmembrane domain to activate the receptor. CRF/Ucn family peptides are 38 -41 amino acids in length and have a C-terminal amide group that is strictly required for bioactivity. The 26 -41 fragments constitute the minimal ECD binding region (23,24). Both the ECD and the 7-transmembrane domains of the CRF receptors contribute to selectivity (25)(26)(27)(28), but the structural basis for peptide discrimination is unclear.
CRFR1 and CRFR2 exhibit 68% amino acid sequence identity, with the most divergence occurring in the ECDs. CRFR1 exists as a single functional isoform, although several nonfunctional isoforms have been identified (29). CRFR2 has three functional full-length isoforms, ␣, ␤, and ␥, that differ only at the extreme N terminus of the receptor (3). In humans, the CRFR2␤ isoform is expressed in the brain, and the CRFR2␣ isoform predominates in the periphery. CRFR1 and CRFR2␤ are produced with an N-terminal signal peptide that is cleaved in the endoplasmic reticulum to yield the mature receptor (26,27). In contrast, the CRFR2␣ isoform contains a unique pseudo signal peptide that is not removed from the mature receptor (30). The function and structural feature of the pseudo signal peptide is an open question.
Recent structural advances have begun to elucidate the mechanisms of ligand recognition by the CRF receptors. There are two NMR structures, including the mouse CRFR2␤ ECD in complex with the non-selective CRF-like antagonist astressin (23) and the human CRFR1 ECD in complex with a high affinity CRF-like agonist (31). There are also three crystal structures of the human CRFR1 ECD, one in a ligand-free form and two in distinct CRF-bound states (24). These structures reveal the architecture of the ECD fold for hormone binding and the conformational changes induced by such binding. The ECD forms a short consensus repeat fold that consists of two ␤-sheets each with two anti-parallel ␤-strands held together by three conserved disulfide bonds. The CRFR1 ECD-CRF crystal structure clearly elucidated the mechanism of endogenous ligand binding. CRF-(26 -41) forms a continuous ␣-helix that occupies a shallow groove in the receptor surface formed at the interface of three loop regions. The bimolecular interaction is anchored at the peptide C terminus by a network of hydrogen bonds involving the amide group and the docking of the Met-38 side chain into a hydrophobic pocket on the receptor surface. Interestingly, the location of the peptide-binding site on the surface of the CRFR1 ECD differs significantly from that of other class B GPCRs (24). The structural advances expanded our understanding of ligand binding, but the structural basis for recognition and discrimination of the Ucn peptides remains unclear. In addition, the functions of the distinct N-terminal regions of the three CRFR2 isoforms and the structural features of the ␣ isoform pseudo signal peptide are unknown.
Here we demonstrate that the purified ECDs of hCRFR1 and the ␣ isoform of hCRFR2 are sufficient to discriminate CRF/Ucn family peptides, and we present three crystal structures of the CRFR2␣ ECD in complex with Ucn1, -2, or -3. The structures indicate that the pseudo signal peptide of CRFR2␣ forms an N-terminal ␣-helix that is similar to those present in other class B GPCRs, such as the receptors for PTH and GLP1, but absent in mCRFR2␤. Comparison of the structures with the hCRFR1 ECD-CRF complex reveals the mecha-nism of ligand selectivity to involve an electrostatic charge compatibility mechanism that is facilitated by a single amino acid difference in the CRFR1 and CRFR2 ECDs. The structures provide the first views of a CRFR2 ECD-Ucn peptide complex and will aid in the rational design of therapeutic agents selectively targeting these receptors.

EXPERIMENTAL PROCEDURES
Molecular Cloning-Gene fragments encoding residues 1-117 or 3-104 of the ␣ isoform of the human CRFR2 receptor were PCR-amplified from a cDNA clone obtained from the Missouri S&T cDNA resource center. The resulting fragments were digested with EcoRI and NotI restriction endonucleases and inserted into the previously described pETDuet1/MBP/DsbC vector (32) to encode a fusion protein consisting of bacterial maltose-binding protein (MBP) followed by the linker amino acid sequence NAAAEF and residues 1-117 or 3-104 of CRFR2. The constructs also contain a C-terminal His 6 tag. The 1-117 construct was used for functional studies, and the 3-104 construct was used for crystallization.
Protein Purification and Peptide Synthesis-The MBP-CRFR1 ECD-His 6 fusion protein, containing residues 24 -119 of human CRFR1 fused to the C terminus of MBP, was previously described (24). The MBP-CRFR2␣-ECD-His 6 proteins (1-117 or 3-104) were expressed in the Escherichia coli strain Origami B (DE3) (Novagen) with induction by isopropyl 1-thio-␤-D-galactopyranoside (0.2 mM) at 16°C overnight. The purification process was similar to that previously described for MBP-CRFR1 ECD (24) except that pH 8.0 buffers were used. In brief, cells were broken using a homogenizer at 10,000 p.s.i., and the protein was purified using nickel-chelating-Sepharose (GE Healthcare) and amylose columns (New England Biolabs), making use of the His 6 and MBP tags, respectively. Disulfide shuffling in vitro was carried out in a redox buffer of 1 mM reduced glutathione and 1 mM oxidized glutathione overnight at 20°C. Superdex 200 gel filtration chromatography (GE Healthcare) was used to separate the properly folded, monomeric protein from aggregated material. A final ion-exchange step using a 1-ml HiTrap Q column (GE Healthcare) yielded a highly purified sample. Peptides were custom synthesized and HPLC-purified by New England Peptide or Peptide 2.0. All peptides contained an additional tyrosine residue at their N terminus to permit quantitation by UV absorbance at 280 nm and a C-terminal amide group unless otherwise noted.
Diffraction data were collected from single crystals at LS-CAT beamlines 21-ID-F and 21-ID-D (Advanced Photon Source), and the data were processed with HKL2000 (33). The structures were solved by molecular replacement using the PHASER program (34) in the CCP4 suite (35). COOT (36) was used for model rebuilding, and REFMAC5 (37) was used for refinement. The molecular replacement solution for the Ucn3-bound structure was obtained using two separate search models corresponding to MBP and the ECD from the human parathyroid hormone receptor (PTH1R) from PDB coordinate file 3H3G (38). The refined CRFR2␣ ECD model was then used as the ECD molecular replacement search model for the Ucn1-and Ucn2-bound structures. NCS restraints were used in all refinements as well as TLS refinement in the final stages. The data collection and refinement statistics are summarized in Table 1. Structure figures were prepared with PyMol (39). Electrostatic surface potential calculations were performed with APBS (40).
AlphaScreen Luminescent Proximity Assay for Peptide-Receptor Interaction-The interaction between peptides and the MBP-CRFR1-ECD-His 6 or MBP-CRFR2␣-ECD-His 6 proteins was analyzed using AlphaScreen technology (PerkinElmer Life Sciences) essentially as previously described (38). MBP receptor ECD-His 6 protein (75 nM) was incubated with N-terminal-biotinylated CRF-(12-41)NH 2 (50 nM) in the presence of increasing concentrations of unlabeled competitor peptides in a buffer of 50 mM MOPS (pH 7.4), 150 mM NaCl, 7 mg/ml fatty acid free BSA, and 5 g/ml each of streptavidin-coated donor beads and nickel-chelate-coated acceptor beads. The reactions were incubated at 22°C for 5 h to reach equilibrium, and the signal was recorded in 384-well microplates using an Envision 2104 plate reader (PerkinElmer Life Sciences). Nonlinear regression, as implemented in Prism 5.0 (GraphPad Software), was used to fit the data to a variable slope dose-response inhibition equation for determination of IC 50 values. Control experiments to ensure that signal inhibition by competitor peptides was specific were carried out using a biotin-Gly 6 -His 6 peptide (50 nM) in place of the MBP receptor and biotin-CRF to bring the donor and acceptor beads into proximity. In all cases the competitor peptides did not significantly diminish this control signal at the concentrations used in our experiments.

The N-terminal Extracellular Domains of CRFR1 and CRFR2␣ Are Sufficient to Determine the CRF/Ucn Family Peptide Selectivity Profiles Observed for the Full-length
Receptors-To facilitate structural and functional analyses of CRF/Ucn family peptide binding to the CRFR2 receptor and to allow comparison with the CRFR1 receptor, we expressed and purified the N-terminal, ECD of the ␣ isoform of human CRFR2 using the methodology we developed for purification of class B GPCR ECDs and previously applied to CRFR1 (24) and PTH1R (32,38,41). The CRFR2␣ ECD, consisting of residues 1-117 and containing the pseudo signal peptide sequence, was expressed in E. coli as a soluble fusion protein with MBP at its N terminus and a His 6 tag at its C terminus. Properly folded, monomeric fusion protein was obtained after in vitro disulfide shuffling in a redox buffer and purification by size exclusion and ion exchange chromatography. The purified sample migrated as a single, distinct band on non-denaturing (native) gel electrophoresis (supplemental Fig. 1).
The ability of the purified MBP-CRFR2␣ ECD-His 6 fusion protein and the previously described MBP-CRFR1 ECD-His 6 fusion protein (24) to bind CRF/Ucn family peptides was assessed using an AlphaScreen luminescent proximity assay (PerkinElmer Life Sciences). In this assay, N-terminal-biotinylated CRF-  attached to streptavidin-coated donor beads interacts with the MBP-ECD-His 6 fusion protein attached to nickel-chelate-coated acceptor beads to bring the two beads into proximity and generate a luminescent signal. Competition with unlabeled peptides can be used to estimate their binding affinity (38). CRF- (26 -41) and Ucn1, -2, and -3 peptides corresponding to the 26 -41 fragment of CRF ( Fig.  1A) were used as competitors because previous studies indicated that this region was sufficient for ECD binding (23,24). The CRFR1 ECD bound Ucn1 with an IC 50 value in the 500 nM to 1 M range and CRF with an ϳ10-fold lower affinity and did not exhibit detectable binding to Ucn2 nor Ucn3 (Fig.  1B). The CRFR2␣ ECD also bound Ucn1 with an IC 50 value in the 500 nM to 1 M range, but CRF binding was ϳ30-fold weaker, and Ucn2 and -3 exhibited detectable binding with IC 50 values similar to CRF (Fig. 1C). These results indicate that the CRFR2␣ ECD purified in this study is functional and that the isolated ECDs of CRFR1 and CRFR2␣ are sufficient to determine the ligand selectivity profiles that are observed for the full-length receptors in their natural membrane environment.
The results reported here for the CRFR1 ECD differ from our previous study where we observed competition of the Al-phaScreen signal by micromolar concentrations of Ucn2 and -3, suggesting their binding to the CRFR1 ECD (24). However, in subsequent experiments we discovered that an insufficient BSA concentration (0.1 mg/ml) in the reaction buffer permitted nonspecific inhibition of the signal to occur at high competitor peptide concentrations. 5 In the experiments reported here, we included a high BSA concentration (7 mg/ml) in the reactions, which prevented nonspecific inhibition of the signal as determined by control experiments carried out as described under "Experimental Procedures." Crystallization and Structure Determination of the CRFR2␣ ECD in Complex with Ucn Peptides-We sought to determine crystal structures of the CRFR2␣ ECD in complex with the Ucn peptides to elucidate the structural bases for their recognition and discrimination by CRFR2␣. Complexes of the MBP-CRFR2␣ ECD-His 6 fusion protein bound to synthetic fragments of Ucn1, -2, or -3 corresponding to residues 26 -41 were prepared and subjected to crystallization screening. Initial trials with the fusion protein containing residues 1-117 failed to yield any crystals, but shortening the N and C termini of the ECD to include only residues 3-104 yielded a fusion protein that readily crystallized in complex with Ucn1, -2, or -3. Diffraction data were collected for the three crystal forms, and the structures were solved by molecular replacement. The data collection and refinement statistics are listed in Table 1. The Ucn1-bound structure was refined to R work /R free values of 23.9/27.3% at 2.75 Å resolution. The Ucn2-bound structure was refined to R work /R free values of 22.9/26.8% at 2.72 Å, and the Ucn3-bound structure was refined to R work /R free values of 22.8/27.7% at 2.50 Å resolution.
The Ucn1-bound crystal form contained four fusion protein molecules (A, B, C, and D) in the asymmetric unit (ASU), each of which was bound to a Ucn1 peptide (E, F, G, and H). The Ucn2-bound crystal form also contained four fusion protein molecules in the ASU, but only molecules C and D had observable electron density for the peptide. Thus, molecules A and B are in a ligand-free state. The Ucn-3 bound crystal form contained two fusion protein molecules in the ASU, with molecule B bound to Ucn3 and molecule A in the ligand-

MBP-CRFR2␣-ECD-Ucn1
MBP  Table 1 lists the ASU contents of the three crystal forms and the average B-factor values for each of the domains. In the following sections we focus our descriptions on the molecules that had the best electron density maps and the lowest B-factors. The CRFR2␣ Pseudo Signal Peptide Forms an ␣-Helix That Is Linked to the Core Short Consensus Repeat Fold of the ECD through a Disulfide Bond-The crystal structures reveal that the CRFR2␣ ECD forms the expected short consensus repeat fold comprised of two ␤-sheets, connecting loops, and three disulfide bonds ( Fig. 2A). ECD residues that are conserved among all class B GPCRs play the same structural roles in CRFR2␣ that were observed for other class B receptors (Fig. 2B). Interestingly, the CRFR2␣ pseudo signal peptide forms a long N-terminal ␣-helix that is linked to ␤2 through the C14-C50 disulfide bond. The helix makes extensive contacts with residues in the ␤1-␤2 sheet but has no contact with loop 2 and only minor contact with loop 3 mediated by the side chains of Leu-6 and Tyr-67. This helix is absent in CRFR2␤ (Fig. 2C) and is significantly shortened in CRFR1 (Fig. 2D). The amino acid sequences of the ␣ and ␤ isoforms of human CRFR2 differ only in this N-terminal region, and CRFR1 and CRFR2 differ most substantially in this region (Fig. 2E). As a result of the pseudo signal peptide, CRFR2␣ resembles other class B GPCRs such as PTH1R, gastric inhibitory peptide receptor, and GLP1 receptor, which all contain a long N-terminal ␣-helix (32,42,43).
Structural Basis for Ucn1 Binding to the CRFR2␣ ECD-The structure of the CRFR2␣ ECD-Ucn1 complex indicates that Ucn1 forms a continuous ␣-helix that binds at the interface of loops 2, 3, and 4 (Fig. 3A), as expected based on previous structural studies of CRFR1 and mCRFR2␤ (23,24). The pseudo signal peptide does not contribute to peptide recognition because it is too far away from the binding site to contact the ligand. Electron density was observed for residues 26Ј-41Ј of Ucn1 (Fig. 3B) (peptide residues are denoted with a prime (Ј) to distinguish them from receptor residues). The density A, a ribbon diagram shows the 2.5 Å resolution structure of the human CRFR2␣ ECD from molecule A of the Ucn3-bound crystal form. This molecule did not have a bound Ucn3 peptide and is, thus, in the ligand-free state. MBP is not shown for clarity. B, CRFR2␣ ECD shows selected side chains. The residues with carbon atoms colored cyan are strictly conserved among all class B GPCRs. Red dashes indicate hydrogen bonds. C, shown is a comparison of the CRFR2␣ ECD with the mouse CRFR2␤ ECD from the ligandfree NMR solution structure (PDB code 2JNC). D, shown is a comparison of the CRFR2␣ ECD with the human CRFR1 ECD from the crystal structure of CRFR1 bound to CRF (PDB code 3EHU). For clarity, the CRF peptide is not shown. E, amino acid sequence alignment of human CRFR2␣, CRFR2␤, and CRFR1 is shown. Secondary structural elements of CRFR2␣ are shown above the sequence. The residue numbering on top corresponds to CRFR2␣, and the numbering on the bottom corresponds to CRFR1. The disulfide bond connectivity is indicated below the sequences in blue numbering. The signal peptide sequences of CRFR2␤ and CRFR1 are not shown. DECEMBER 17, 2010 • VOLUME 285 • NUMBER 51 was better at the peptide C terminus than at the N terminus. The peptide N-terminal region exhibits higher B-factors than the C-terminal region, consistent with increased mobility of the N terminus and the C terminus serving as the "anchor point" of the interaction, as previously observed for the binding of CRF to the CRFR1 ECD (24).

Ucn1-, 2-, and 3-bound CRFR2␣ ECD Structures
The minimal pharmacophore appears to span residue Asn-34Ј to the C-terminal amide group. The Asn-34Ј side chain forms hydrogen bonds with the Tyr-73 side chain hydroxyl and the Phe-68 backbone carbonyl (Fig. 3D). The Phe-38Ј phenyl ring is buried in a hydrophobic pocket of the receptor that is formed by Tyr-95, Ile-47, Pro-65, Phe-68, Tyr-73, and the Cys-64 -Cys-98 disulfide (Fig. 3, C and D). Ile-37Ј and Val-41Ј also form hydrophobic contacts with the receptor. At the C terminus, a series of hydrogen bonds form critical contacts. The C-terminal amide nitrogen forms an intramolecular hydrogen bond with the backbone carbonyl of Phe-38Ј to stabilize the helical conformation of the peptide and an intermo-lecular hydrogen bond with the backbone carbonyl of Ile-93 (Fig. 3D). In addition, the amide carbonyl group forms an intermolecular hydrogen bond with the backbone amide nitrogen of Ile-93. This pattern of hydrogen bonds was previously observed for the C-terminal amide groups of CRF and PTH interacting with their respective receptors (24,32). Last, the electronegative peptide C terminus is capped by the positively charged Lys-92 of the receptor.
To validate the interactions observed in the crystal structure, we examined the ability of alanine-scan mutants of Ucn1 to bind to the CRFR2␣ ECD using the AlphaScreen assay. Alteration of residues Ile-37Ј, Phe-38Ј, or Val-41Ј to alanine resulted in a peptide with diminished ability to inhibit the association of biotin-CRF and the receptor (Fig. 3E). In addition, substituting a carboxylate for the C-terminal amide group resulted in drastically reduced competition of the binding signal. These results indicate that the side chains of Ile-37Ј, Phe-38Ј, and Val-41Ј and the C-terminal amide group are important CRFR2␣ ECD binding determinants, consistent with the crystal structure.
Structural Bases for Ucn2 and Ucn3 Binding to the CRFR2␣ ECD-The Ucn2-and Ucn3-bound structures reveal that Ucn2 and Ucn3 bind to the CRFR2␣ ECD with binding modes very similar to Ucn1 despite their significantly lower affinity for the receptor. Electron density was observed for residues 26Ј-41Ј for both Ucn2 (Fig. 4A) and Ucn3 (Fig. 5A). Residue Asn-34Ј and the C-terminal amide groups of Ucn2 and Ucn3 form the same network of hydrogen bonds observed for Ucn1 (Figs. 4B and 5B). Ucn2 and Ucn3 both contain a hydrophobic residue at position 38Ј (Leu in Ucn2, Met in Ucn3) that occupies the hydrophobic pocket in the receptor surface, similar to Phe-38Ј of Ucn1 (Figs. 4, B and C, and 5, B and C). The most significant differences between Ucn2/3 and Ucn1 are at residues Ala-35Ј and Ala-39Ј in Ucn2/3, which are Arg-35Ј and Asp-39Ј in Ucn1 (Figs. 4C and 5C).
The Ligand-free and Ligand-bound Conformations of the CRFR2␣ ECD Do Not Differ Substantially-Our previous crystal structures of the ligand-free and CRF-bound states of the human CRFR1 ECD indicated that loops 2 and 4 undergo clamp-like conformational changes upon CRF binding that probably help anchor the peptide C terminus (24) (Fig. 6A). Similarly, NMR studies of the mouse CRFR2␤ ECD revealed that loops 2 and 4 exhibit conformational flexibility in the absence of ligand and become ordered upon peptide binding (23). We examined if similar changes occur in the human CRFR2␣ ECD by comparing the two ECD molecules that were present in the ASU of the Ucn3 crystal form; molecule A was in the ligand-free state, whereas molecule B was bound to Ucn3. In contrast to the observations for hCRFR1 and mCRFR2␤, there were no substantial conformational differences between the ligand-free and Ucn-3-bound states of the hCRFR2␣ ECD (Fig. 6B). This difference appears to be determined by the presence of an Ile residue at position 93 in hCRFR2␣ instead of the Val residue present at the equivalent positions in hCRFR1 (Val-97) and mCRFR2␤. The additional methyl group contributed by Ile-93 in hCRFR2␣ would clash with the side chain of the conserved Asp residue in loop 2 (Asp-45 in hCRFR2␣) if similar clamp-like motions occurred (Fig. 6, A and B).
Structural Basis for Ligand Selectivity-Structural elements that determine the selectivity of Ucn2/3 for CRFR2 were examined by comparing the CRFR2␣ ECD-Ucn3 structure with our previous structure of the CRFR1 ECD-CRF complex (24). The Ucn3-bound structure was chosen for comparison because Ucn3 shares more sequence identity with CRF than Ucn2 (Fig. 1A), and the Ucn3 structure was determined to a higher resolution. Structural overlay of the two complexes reveals that the peptides bind to their respective receptors with nearly identical binding modes (Fig. 7A). CRF and Ucn3  (Ucn2). B, a detailed view of the peptide-receptor interface with selected side chains is shown as sticks. The receptor is green, and the peptide is salmon. C, structural overlay of the Ucn2-and Ucn1bound structures is shown. The ECD from the Ucn2-bound structure is green, and the ECD from the Ucn1-bound structure is dark forest green.  (Ucn3). B, a detailed view of the peptide-receptor interface with selected side chains is shown as sticks. The receptor is green, and the peptide is orange. C, structural overlay of the Ucn3-and Ucn1bound structures is shown. The ECD from the Ucn3-bound structure is green, and the ECD from the Ucn1-bound structure is dark forest green.
share Asn-34Ј, Leu-37Ј, Met-38Ј, and Ile-41Ј, which form identical contacts in the two structures. The architectures of the peptide-binding sites are very similar, with only six residues located at the periphery of the binding sites differing between CRFR1 and CRFR2␣ (Figs. 7A and 2E).
Comparing the electrostatic surface potentials of the two receptor ECDs suggests that the single CRFR1 Glu-104/ CRFR2␣ Pro-100 difference determines the selectivity profiles of the receptors by a charge compatibility mechanism. Residue Glu-104 creates a negative area of the CRFR1 receptor surface adjacent to peptide residue 35Ј (Fig. 7B). The positively charged Arg-35Ј of CRF/Ucn1 is electrostatically compatible with Glu-104, whereas the hydrophobic Ala-35Ј of Ucn2/3 would serve as a repelling force. In contrast, residue Pro-100 creates a hydrophobic surface on the CRFR2␣ receptor adjacent to peptide residue 35Ј (Fig. 7C) that would be compatible with the nonpolar Ala-35Ј of Ucn2/3 as well as the aliphatic portion of the Arg-35Ј side chain of CRF/Ucn1. Thus, we propose that CRFR1 residue Glu-104 is the "selectivity filter" that prevents Ucn2/3 binding.
Functional Basis for Ligand Selectivity-In a previous study, Mazur et al. (28) assessed the ability of full-length CRF/Ucn peptides altered at positions 35Ј and 39Ј to stimulate cAMP accumulation in cells expressing either CRFR1 or CRFR2 and demonstrated that alanine at positions 35Ј and 39Ј contributed to CRFR2 selectivity (28). Our crystal structures provide a molecular rationale for why position 35Ј and, to a lesser extent position 39Ј, determine receptor selectivity. To further examine positions 35Ј and 39Ј, we assessed the CRFR1 and CRFR2␣ ECD binding ability of hybrid Ucn1/3 peptides containing reciprocal swap mutations at positions 35Ј and 39Ј. The single R35ЈA mutation was sufficient to drastically reduce the binding of Ucn1 to CRFR1 (Fig. 8A) and also decreased its  A, structural overlay of the CRFR2␣ ECD-Ucn3 complex (2.5 Å resolution) with the CRFR1 ECD-CRF complex (1.9 Å resolution; PDB code 3EHU) is shown. The CRFR2␣ ECD-Ucn3 complex is that of molecules B and C, respectively. The ECDs are shown as ribbon diagrams with side chains that form the peptide-binding site shown as sticks. The peptides are shown as coils with selected side chains as sticks. The six peptide-binding site residues that differ between CRFR1 and CRFR2␣ are labeled in color-coded format. CRFR1 residue Tyr-73 exhibited no side chain electron density in the 3EHU structure and was, thus, trimmed back to the ␤ carbon atom in 3EHU. For clarity, here the Tyr-73 side chain was added in the most sensible rotamer conformation. B, the electrostatic surface potential of the CRFR1 ECD was calculated with APBS assuming a solvent of 150 mM salt. A semitransparent molecular surface colored according to charge is shown over the CRFR1 ribbon diagram. The color ramp is from Ϫ5 (red) to ϩ5 (blue) kT/e. The side chains of the six peptide-binding site residues that differ between CRFR1 and CRFR2␣ are shown as sticks and labeled. CRF is shown as a yellow coil. To permit better comparison with CRFR2␣, residues 105-106 of CRFR1, which were not visible in the electron density maps for 3EHU, were added to the ECD in the same conformation observed for residues 101-102 of CRFR2␣. These residues are strictly conserved between CRFR1 and CRFR2␣ (Fig. 2E). C, shown is the electrostatic surface potential of the CRFR2␣ ECD, depicted as in panel B. Ucn3 is shown as an orange coil. affinity for CRFR2␣ (Fig. 8B). The R35ЈA/D39ЈA Ucn1 double mutant did not exhibit detectable CRFR1 binding and bound to CRFR2␣ with an affinity comparable with Ucn3 (Fig. 8, A  and B). Similarly, the single A35ЈR mutation was able to impart CRFR1 binding ability to Ucn3 (Fig. 8C) and slightly increase its affinity for CRFR2␣ (Fig. 8D). Inclusion of the A39ЈD mutation in combination with A35ЈR further increased the affinity of Ucn3 for both receptors, although not to the level of WT Ucn1 (Fig. 8, C and D). These results indicate that the residues at peptide positions 35Ј and, to a lesser extent 39Ј, are sufficient to determine receptor specificity via effects on ECD binding, and they complement the results reported for the full-length receptor (28). Together these results support the proposed electrostatic compatibility mechanism mediated by the CRFR1 residue Glu-104 selectivity filter.

DISCUSSION
The CRF/Ucn family of peptides and their two receptors constitute an important biological system that controls central stress/anxiety responses, links stress and metabolic energy homeostasis, and regulates cardiovascular, immune, gastrointestinal, and reproductive functions. The three crystal structures presented here show for the first time how the Ucn peptides are recognized by the CRFR2 receptor and reveal the structural basis for ligand discrimination by the CRFR1 and CRFR2 ECDs. Peptide binding assays demonstrated that the isolated ECDs of CRFR1 and CRFR2 are sufficient to discriminate the peptides with the same rank order as the full-length receptors, indicating that the ECDs act as selectivity filters that determine ligand specificity in addition to their role in enabling high affinity ligand binding. Thus, it appears that the ECDs provide an important "filtering" step in the ligand binding process, although the 7-transmembrane domains can also contribute to ligand selectivity through interactions with the N-terminal region of the peptides (25,28).
The CRFR2␣ ECD formed the same fold previously observed for the hCRFR1 and mCRFR2␤ ECDs with the exception of the long N-terminal ␣-helix formed by the pseudo signal peptide. This helix renders the CRFR2␣ ECD similar in appearance to those of other class B GPCRs, such as PTH1R, gastric inhibitory peptide receptor, and GLP1 receptor, which also contain a long N-terminal helix (32,42,43). In these other receptors, the N-terminal helix forms part of the peptide-binding site. In contrast, the CRFR2␣ ECD helix does not affect ligand binding because the ECD peptide-binding site in the CRF receptors is significantly shifted away from the Nterminal helix as compared with other class B GPCRs (24). If the pseudo signal peptide were cleaved from CRFR2␣, Cys-50 would be left unpaired, but the peptide-binding site would remain intact. Indeed, before it was understood that CRFR2␣ contains a pseudo signal peptide, Klose et al. (44) demonstrated that a recombinant form of the rat CRFR2␣ ECD that lacked the presumed signal peptide contained an unpaired Cys-50 residue and retained the ability to bind Ucn1. Interestingly, the ␥ isoform of CRFR2 would have five cysteine residues if the putative signal peptide were cleaved and six if it were not removed. Thus, CRFR2␥ may also have a pseudo signal peptide that is linked to the core short consensus repeat fold of the ECD through a disulfide bond.
The CRFR2␣ ECD peptide-binding site architecture is very similar to that of CRFR1, with only six residues at the periphery of the binding sites differing between the receptors (Fig.  7). This result was not completely expected because the NMR structure of the mCRFR2␤ ECD in complex with astressin (23) indicated a very different loop 3 conformation than was observed for the CRFR1 ECD (24). The structures reported  DECEMBER 17, 2010 • VOLUME 285 • NUMBER 51

Ucn1-, 2-, and 3-bound CRFR2␣ ECD Structures
here clearly indicate that the CRFR2 ECD loop 3 conformation is in fact quite similar to CRFR1. The binding of the Ucn peptides to the CRFR2␣ ECD closely resembles the binding of CRF to the CRFR1 ECD. Comparing the electrostatic surface potentials of the two ECDs suggested that Glu-104 of CRFR1, which is replaced by Pro-100 in CRFR2␣, acts as a selectivity filter preventing the binding of Ucn2/3 by a charge compatibility mechanism. Ucn2 and -3 contain the nonpolar residue Ala-35Ј instead of the positively charged Arg-35Ј present in CRF/Ucn1. The uncharged, hydrophobic Ala-35Ј presumably provides a repelling force to the negatively charged Glu-104, whereas the positively charged Arg-35Ј of CRF/Ucn1 provides an attractive force. Presumably, Arg-35Ј does not prevent CRFR2 binding because the aliphatic portion of the side chain is compatible with Pro-100. We previously proposed such a mechanism based on our structure of the CRFR1 ECD-CRF complex and peptide binding data that showed that the Arg-35Ј side chain of CRF was absolutely required for CRFR1 ECD binding even though the electron density maps for our structure (PDB code 3EHU) did not provide evidence for an intermolecular salt bridge between CRF Arg-35Ј and CRFR1 Glu-104 (24). The recently reported NMR solution structure of the hCRFR1 ECD bound to a high affinity CRF-like agonist showed that Arg-35Ј of the peptide is in proximity to Glu-104 of the receptor (31).
In support of our proposed selectivity mechanism, our ECD binding data for the hybrid Ucn1/3 swap peptides (Fig. 8) confirmed the importance of peptide residue 35Ј for determining selectivity. However, the data also indicated that peptide residue 39Ј cooperates with residue 35Ј to contribute to selectivity and affinity, especially in the context of converting Ucn3 into a non-selective peptide. This point is highlighted by our data showing that the Ucn3 A35ЈR/A39ЈD double mutant exhibited higher affinity for both ECDs than the A35ЈR single mutant (Fig. 8, C and D). One possible explanation is that an intramolecular salt bridge between Arg-35Ј and Asp-39Ј stabilizes the helical conformation of the peptide and, thus, increases its affinity. In the case of CRFR1 binding, a competition between Asp-39Ј of the peptide and Glu-104 of CRFR1 for the Arg-35Ј side chain might exist. For binding to CRFR2, the formation of an intramolecular Arg-35Ј-Asp-39Ј salt bridge would presumably facilitate the exposure of the aliphatic portion of the Arg-35Ј side chain to Pro-100. The electron density maps for the Ucn1-CRFR2␣ ECD structure were not well defined for the side chains of Arg-35Ј and Asp-39Ј, similar to our CRF-CRFR1 ECD structure lacking side chain density for CRF Arg-35Ј and Glu-39Ј; thus, we do not have crystallographic evidence supporting Ucn1 Arg-35Ј-Asp-39Ј interactions. Nonetheless, the peptide binding data clearly support a role for functional coupling between peptide residues 35Ј and 39Ј. Moreover, our results are entirely consistent with those reported in the literature for the full-length peptides acting at intact receptors in cells. Mazur et al. (28) demonstrated that substituting alanine at positions 35Ј and 39Ј of Ucn1 contributed to significantly reduced potency at CRFR1 while having little effect on potency at CRFR2, further supporting the charge compatibility mechanism for ligand selectivity as observed from our ECD structures.
The structures reported here provide us with more complete views of the mechanisms of ligand recognition used by the CRF receptors and increase our understanding of the molecular basis for peptide selectivity. From a translational standpoint, the structures provide additional templates to guide rational drug design efforts. The similarity of the CRFR1 and CRFR2 ECD peptide-binding site architectures highlights the challenges associated with developing drugs that selectively target one receptor ECD over the other. Fortunately, the distinct electrostatic surface potentials arising from the CRFR1 Glu-104/CRFR2␣ Pro-100 difference provides a possible route to developing drugs that selectively target the ECDs. The fact that the endogenous ligands Ucn2/3 can selectively target CRFR2 based on this charge difference bodes well for our ability to exploit this mechanism for therapeutic purposes.