In order to investigate loop conformations, a coiled-coil scaffold was constructed to provide a folding constraint for the particular loop of interest. Oakley et al. have previously studied designed peptide sequences for their ability form parallel and anti-parallel coiled-coils [32]. The ability of peptide pairs to form a coiled-coil was confirmed using CD. The placement of two Asn residues was important in distinguishing a parallel versus an anti-parallel coiled-coil. In this study we used their sequences along with EL2 of LPA₁ in a continuous sequence (LPA₁-CC-EL2) which formed an anti-parallel coiled-coil due to the placement of two Asn residues in positions a′ and d (Figure 2). EL2 was chosen for this study since it has a wide variety of structures across the different GPCR (Figure 1).

Circular dichroism spectra were collected in order to define an appropriate buffer for NMR studies, in which the LPA₁-CC-EL2 peptide exhibited the predominantly helical structure expected on the basis of the design. Once an appropriate buffer was selected, thermal denaturation was performed using 220 nm to follow denaturation of the α-helical secondary structure. Figure 4 shows a wavelength scan at 20 °C as well as thermal denaturation. Deconvolution of the secondary structure using the CDSSTR algorithm with reference set 7 [34] on Dichroweb [35–37] indicates the LPA₁-CC-EL2 peptide is 65% helical, 11% strand, 9% turns, and 14% disordered. Thermal denaturation data demonstrated that LPA₁-CC-EL2 unfolding was not completely cooperative, as a plateau was apparent at about 90 °C. The CD thermal denaturation of the LPA₁-CC-EL2 peptide (Figure 3) is consistent with the thermal denaturation studies performed by Oakley et al.[32] on the separate coiled-coil sequences as an equimolar mixture, which showed a single thermal denaturation below 80 °C (the highest temperature tested).

Sequence-specific chemical shift assignments for LPA₁-CC-EL2 were obtained from the interactive use of the backbone 3D NMR experiments listed in experimental methods. These chemical shifts were then used to obtain the sidechain assignments (aromatics unassigned). Chemical shifts have been deposited in the BioMagResBank under accession number 17993. The backbone nitrogen of E26 has gone unassigned; however, its sidechain chemical shifts were obtained using back correlations from K27. The TALOS program provided 119 Φ and Ψ dihedral angles from the chemical shift data. A total of 715 ¹H–¹H distance restraints (NOEs) from the 3D ¹⁵N-NOESY-HSQC and ¹³C-NOESY-HSQC experiments were employed in the CNS structure calculation. Four NOE peak intensity categories were used to classify intra ¹H–¹H upper distance limits; strong (<2.7 Å), medium (<3.3 Å), weak (<4.0 Å), and very weak (<4.5 Å). These two types of constraints were used in the initial CNS structure calculations. Once a calculation was completed, the NOEs and dihedrals were assessed based on violations. If a distance or dihedral was violated, it was redefined to the next weakest category for NOEs or 5° was added to the dihedral range. This continued until there were no NOE violations greater than 0.5 Å and no dihedral violated by greater than 10°. At this point another calculation was performed that defined 24 hydrogen bonds down the two helices to further stabilize their structures.

Initial structure calculations employed short and medium range NOEs to refine the two separate helical structures. Both TALOS and chemical shift analysis (data not shown) indicated helical structure from residues 3–31 and 50–78. Once these structures converged, the coiled-coil was initially defined by 10 distances observed in the ¹³C-NOESY-HSQC data. Specifically, very weak distances were defined between the methyl groups of L4 to L74, L18 to L60, L22 to L64, L25 to L53 and L29 to L57 (two distances for each pair). These distances represented the only unique chemical shift pairs in the predicted coiled-coil since most of the sidechain leucine interactions were overlapped. However, this proved to be sufficient to define the coiled-coil structure. One key feature in the design of this coiled-coil was the presence of N11 and N71. The positions of these residues are important to promote an anti-parallel coiled-coil. There were two NOEs observed in the ¹⁵N-NOESY-HSQC, one from the NH₂ group of N11 to the methyl of L74 and the other from the NH₂ group of N11 to the NH₂ group of N71. First, a calculation was performed that defined the distance for N11 to L74. In all the resulting structures from previous coiled-coil calculations the N11 and N71 residues were at positions a′ and d which indicated an anti-parallel coiled-coil. Also the sidechain NH₂ group of N11 pointed to the sidechain O of N71 which still placed the two NH₂ groups close enough to show a weak NOE, therefore another calculation contained this distance restraint.

The last few rounds of calculations employed short and medium range distance restraints in the loop. Most of the loop residues did not result in any NOEs in either the ¹⁵N-NOESY-HSQC or ¹³C-NOESY-HSQC suggesting a complete lack of structure in these residues of the loop. However, there were short range (i to i and i to i ± 1) ¹⁵N-NOEs observed from residues Ala45 through Ala52. There were also three medium range ¹⁵N-NOEs involving Pro46 through Ala52 (Figure 4). The last calculation included these NOEs which produced an interesting conformation in this part of the loop (see Section 2.6). The final round of structure calculations consisted of 100 structures. Table 1 shows statistical information used for and resulting from the final calculation of the 58 lowest energy structures out of the 100 used in the final round of structure calculations. Of these 58 lowest energy structures, 36 representatives for the LPA₁-CC-EL2 were superimposed as shown in Figure 5.

The overall structure of LPA₁-CC-EL2 does not converge to a single family (backbone RMSD 4.01 Å) due to a lack of NOEs in the loop. However, each helix does show convergence (backbone RMSD 0.63 Å and 0.58 Å respectively) as does the coiled-coil (backbone RMSD 0.95 Å). The RMSD values for the bonds, angles, impropers, NOEs, and dihedrals are acceptably low (<3% of values for bonds, angles, impropers and NOEs, ~20% of value on dihedrals). Overall, >80% of the residues of the LPA₁-CC-EL2 family fall in favored or allowed regions of the Ramachandran plot indicating a well calculated structure. The set of 36 structures in Figure 5 were submitted to the Research Collaboratory for Structural Bioinformatics (RCSB) Protein Data Bank (PDB) with the RCSB ID code RCSB102692 and PDB ID code 2LQ4.

The results of these calculations as well as the CD wavelength spectra and thermal denaturation show that the peptide sequences from Oakley et al.[32] do indeed form a coiled-coil when connected by a linking peptide. These sequences, therefore, provide an excellent scaffold within which loop conformations can be investigated. These results show that the EL2 conformation of the LPA₁ receptor does not contain any well-defined secondary structure, analogous to EL2 in six of the class A GPCR members for which crystal structures are available (Figure 1). However, there was an interesting result in the conformation around residues P46 to A52, due to NOE contacts from the sidechains of residues P46 (β-hydrogen), Y48 (δ- and ɛ- hydrogens), and A52 (β-hydrogen) to the backbone amide hydrogens of residues L47, Y48, and D50 (Figure 4). Although the sidechain of coiled-coil residue A52 is involved in this network of contacts, the β-hydrogens of the tyrosine residue that occurs in full-length LPA₁ could make the same contacts, thus this conformation appears to be an intrinsic property of the LPA₁ EL2 loop sequence. Due to these NOE contacts, this part of the loop was always in a bent conformation (Figure 5); however it did not converge into a single conformation suggesting that this part of the loop is limited to a small amount of conformational space while the rest of the loop has a greater range of conformational freedom. This would represent the carboxy-terminal end of the loop transitioning into helix 5 of the LPA₁ receptor. The resulting structures from these calculations were superimposed into the corresponding helix4–EL2–helix5 of our LPA₁ receptor model in order to illustrate how this bend in EL2 fit onto the overall structure (Figure 6) and to provide starting structures for hybrid LPA₁ receptor model development. Of these 58 structures, only 2 showed any overlap of the loop with atoms of the receptor. The bend in the loop places the loop in an outward conformation that leaves the top of the receptor open, potentially to accommodate a structured segment within the N-terminus as observed in the recently characterized S1P₁ receptor [25]. The outward bend of EL2 is also observed in the recently characterized M2 muscarinic acetylcholine receptor although at the amino-terminal end of EL2 rather than the carboxy-terminal end (Figure 7).

The 58 lowest-energy LPA₁-CC-EL2 NMR structures were each superposed on the validated LPA₁ receptor model in order to produce a hybrid NMR/model structure by multi-template homology modeling (Figure 6). The EL1 portion of the LPA₁ receptor model had previously been remodeled on the basis of the NMR structure of our shorter coiled-coil EL1 mimetic of the S1P₄ receptor [31], which exhibits a short 3¹⁰ helical turn at the same location as a short helix noted in the recent S1P₁ crystal structure. Thus the 58 resulting models are hybrids of the NMR structure data for LPA₁-CC-EL2, S1P₄-EL1, and the original template LPA₁ receptor model. One of the resulting hybrid LPA₁ receptor model structures showed EL2 wrapped around EL3 and was eliminated from further consideration. Seventeen additional structures showed very close contacts (<1.6 Å) between atoms in the EL2 structure and residues R124^3.28 or Q125^3.29 at the extracellular end of TM3 (Figure 8), which have been demonstrated by previous modeling and mutagenesis to form required interactions with the natural agonist, LPA. These close interactions would prevent LPA interaction at these sites, so these 17 models were also eliminated from further consideration. The remaining 40 models are shown in Figure 8.

Docking studies were performed for the ligands shown in Figure 9 using each of the 40 hybrid LPA₁ receptor models from Figure 8 as docking targets. These studies were performed in order to determine if specific members of the NMR ensemble were compatible with known features of the full-length LPA₁ receptor. LPA 18:1 is expected to interact with residues R124^3.28 and Q125^3.29 at the extracellular end of TM3, as mutations of either of these residues to alanine produces receptors that are either only weakly activated or are not activated by LPA 18:1 [12]. The corresponding sites in the closely-related S1P₁ receptor, R120^3.28 and E121^3.29, have been demonstrated to interact with the phosphate and ammonium groups of a co-crystallized antagonist [25]. Thirty-one models were eliminated from further consideration due to high-ranking poses of LPA with phosphate groups placed in the lower end of the binding pocket, distant from both R124^3.28 and Q125^3.29. Such a result suggests that hybridization of that particular loop geometry has altered the headgroup recognition pocket, and therefore the resulting hybrid does not represent a biologically meaningful model of the full-length receptor. The remaining nine models in which the top-ranked LPA pose exhibited a headgroup position near R124^3.28 and Q125^3.29 are shown in Figure 10. Figure 11 illustrates the consistency of complexes docked into the hybrid model generated from chain 3 in PDB entry 2LQ4 with the known antagonist pharmacology. The active antagonists, while all having a buried carboxylate functional group, have that polar functional group in a relatively polar subpocket formed by S3.39, S7.46, and N7.45 (panel B). In contrast, the inactive compound, NSC 47091, buries two carboxylic acid functional groups in relatively hydrophobic pockets, including S5.37 as the only polar sidechain (panel C).

The present study demonstrates that the sequence found in the LPA₁ EL2 is largely flexible and unstructured, even when the termini are held close together by a coiled-coil scaffold. Two characteristics that might reduce the conformational space available to this loop in the full-length receptor are disulfide bonds and contacts with other extracellular segments including EL1, EL3, or the amino terminus. The recently crystallized S1P₁ structure has an EL2 identical in length which exhibits 48% residue identity and 57% homology to the LPA₁ EL2, and is the only class A GPCR crystallized to date that lacks an interloop disulfide bond between cysteine residues in EL1 and EL2. S1P₁ instead exhibits an intraloop disulfide bond between two cysteine residues in EL2 and a second intraloop disulfide bond between two cysteine residues in EL3. A superposition of the NMR ensemble on the S1P₁ EL2 is shown in Figure 12. Like S1P₁, the LPA₁ receptor lacks cysteine in EL1, and exhibits conservation of the cysteine residues involved in the internal EL2 disulfide bond (C184 and C191 in LPA₁) [38]. However, LPA₁ additionally contains a third cysteine residue in both EL2 (C186) and EL3 that is not found in S1P₁. The additional cysteine residues prevent direct inference of matching disulfide bonds in S1P₁ and LPA₁. However, it is likely that one or more of the three LPA₁ EL2 cysteine residues participates in intraloop or interloop disulfide bonds in the context of the full-length receptor that would provide an additional conformational restriction relative to the LPA₁-CC-EL2 peptide characterized with cysteine residues in reduced form. However, it is not clear at this time if the LPA₁ EL2 structure in the context of the full-length LPA₁ receptor should closely match that of S1P₁ due to the differences in number of cysteine residues noted. The NMR ensemble of LPA1-CC-EL2 showed no consistent Cys–Cys close contacts that suggest an intrinsic disulfide pairing preference.

The second extracellular loop of LPA₁ was inserted between helical segments known to form an antiparallel coiled-coil to form a synthetic protein sequence, termed LPA₁-CC-EL2. LPA₁-CC-EL2 DNA was prepared by PCR using 7 overlapping oligonucleotides and two terminal primers, and incorporated into pET-32 Ek/LIC (EMD Millipore) using ligation-independent cloning to produce the plasmid containing a hexahistidine tag. The resulting expression vector was confirmed with DNA sequencing (Molecular Resource Center, UTHSC, Memphis, TN) and then transformed into E. coli BL21(DE3) cells for protein expression. Cell cultures were grown in minimal media containing 1 L of ₁₃C₆-glucose as the sole source of ¹³C and 1.0 g/L of ¹⁵NH₄Cl as the sole source of ¹⁵N. This growth was performed as previously described, with no alterations [31]. Cells were grown at 37 °C and protein expression induced with 0.1 mM IPTG at an OD₆₀₀ of 0.6. The cells were pelleted by centrifugation and stored at −80°C. The cell pellet was lysed using the CelLytic B Plus Kit (Sigma-Aldrich, MI, USA) and loaded directly onto a Ni-NTA affinity column (GE Healthcare, New Jersey, NJ, USA), washed with 40 mM imidazole in 20 mM sodium phosphate buffer, pH 7.4, containing 500 mM sodium chloride followed by elution with 500 mM imidazole in the same buffer. The desired peptide (LPA₁-CC-EL2) was then cleaved from the His-tag using recombinant enterokinase (EMD BioSciences) and purified by HPLC using a reverse-phase C18 preparative column. Anion-exchange chromatography was then used to separate the desired protein from nonspecific cleavage products, followed by confirmation using ESI-mass spectroscopy and sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE). The final protein sample was concentrated to either 0.1 mM (for CD thermal denaturation) or 0.5 mM (for NMR studies), and the solvent exchanged to 50 mM Tris-HCl, 1.0 mM DTT, 1 mM sodium azide, pH 6.0 and 10% D₂O.

All CD data were acquired on an Aviv 410 spectropolarimeter using a 1 mm cell. Wavelength scans were collected for 0.1 mg/mL samples at 1 nm increments over a wavelength range of 190–260 nm at room temperature. Thermal denaturation experiments were performed over a temperature range from 20 °C to 110 °C at a wavelength of 220 nm. Ellipticities (φ) measured during thermal denaturation were converted to % unfolded at each temperature (T) based on the assumption that the structure was fully folded at the lowest temperature (Tmin) and fully unfolded at the highest temperature (Tmax) using Equation 1 [39]:

All NMR data were acquired at 25 °C on a Varian VNMRS 500 MHz spectrometer using a 5 mm triple-resonance ¹H, ¹³C, ¹⁵N probe with z-axis gradients. The sequential backbone assignments were accomplished through the interpretation of the experiments: HNCA, HN(CO)CA, HNCACB, CBCANH, HNCO, HN(CA)CO and ¹⁵N-NOESY-HSQC. Sidechain assignments were obtained through the H(CCO)NH-TOCSY, (H)C(CO)NH-TOCSY and HCCH-TOCSY experiments. NOE data was interpreted from the ¹⁵N-NOESY-HSQC and ¹³C-NOESY-HSQC experiments, both with a 150 ms mixing time. Data was processed using NMRPipe [40] software and analyzed with PIPP [41].

NMR chemical shift values for ¹⁵N, ¹H^α, ¹³C^α, ¹³C^β, and ¹³C′ were used in the TALOS program [42] for predictions of Φ and Ψ dihedral angles. NOE distances were obtained from the 3D ¹⁵N-NOESY-HSQC and ¹³C-NOESY-HSQC experiments. Dihedral angle and NOE restraints were used as input for the structure calculation with the crystallography and NMR system (CNS version 1.1) program [43]. A series of structure calculations ended when a family of structures was obtained in which no NOE distance was violated by >0.5 Å, no dihedral angle was violated by >10°, and low energies and good convergence were obtained.

Hybrid LPA₁ receptor models were developed using a previously validated model of the LPA₁ receptor [9,12,44] and the NMR structures of LPA₁-CC-EL2 as input for multi-template homology modeling by the MOE software [45]. The hybrid models were generated using residues Leu29-Gln51 from the NMR structures of LPA₁-CC-EL2 as the basis for modeling residues Ser183-Ser205 and the previously published LPA₁ receptor model as the basis for modeling the remainder of the structure. The resulting models did not include residues 1–35 and 332–364 (N- and C-termini).

LPA 18:1 and several compounds exhibiting varying potencies as LPA receptor antagonists (H2L5765834 K_i = 48 nM; H2L5105099 K_i = 50 nM; H2L7724589 K_i = 311 nM; H2L5226501 maximal inhibition of 59% at 30 μM and NSC47091 inactive, structures shown in Figure 9) [46,47] were docked into each hybrid LPA₁ receptor model using Autodock Vina. The docking box was centered between the sidechains of W271^6.48 and S304^7.46 and extended 40 Å along the long axis of the model and 22 Å along each of the other two axes. Top-ranked poses obtained using searches with exhaustiveness set to 16 were analyzed for consistency with known relative potencies and mutagenesis results.