Researchers are exploring whether these ubiquitous fluorinated molecules might worsen infections or hamper vaccine effectiveness.
Image credit: Shutterstock/Dmitry Naumov.
Edited by K. Christopher Garcia, Stanford University, Stanford, CA, and approved December 20, 2013 (received for review September 24, 2013)
Significance
Only a tiny fraction (<2%) of the unique structures in the protein database correspond to membrane proteins, and only a few of these are of eukaryotic origin, representing potential drug targets. The difficulties in structure determination of these proteins are due to two specific complications, which are unique for membrane proteins: first, low expression levels and, second, the necessity for detergent micelles, which are often destabilizing as they mimic the hydrophobic membrane environment only poorly. We prove that directed evolution has the potential to overcome these problems by determining several structures of evolved eukaryotic G protein–coupled receptor variants. High functional expression levels and superior receptor stability in harsh detergents allowed us to gain deeper insights into this important receptor family.
Abstract
Crystallography has advanced our understanding of G protein–coupled receptors, but low expression levels and instability in solution have limited structural insights to very few selected members of this large protein family. Using neurotensin receptor 1 (NTR1) as a proof of principle, we show that two directed evolution technologies that we recently developed have the potential to overcome these problems. We purified three neurotensin-bound NTR1 variants from Escherichia coli and determined their X-ray structures at up to 2.75 Å resolution using vapor diffusion crystallization experiments. A crystallized construct was pharmacologically characterized and exhibited ligand-dependent signaling, internalization, and wild-type–like agonist and antagonist affinities. Our structures are fully consistent with all biochemically defined ligand-contacting residues, and they represent an inactive NTR1 state at the cytosolic side. They exhibit significant differences to a previously determined NTR1 structure (Protein Data Bank ID code 4GRV) in the ligand-binding pocket and by the presence of the amphipathic helix 8. A comparison of helix 8 stability determinants between NTR1 and other crystallized G protein–coupled receptors suggests that the occupancy of the canonical position of the amphipathic helix is reduced to various extents in many receptors, and we have elucidated the sequence determinants for a stable helix 8. Our analysis also provides a structural rationale for the long-known effects of C-terminal palmitoylation reactions on G protein–coupled receptor signaling, receptor maturation, and desensitization.
Neurotensin is a 13-amino-acid peptide, which plays important roles in the pathogenesis of Parkinson’s disease, schizophrenia, antinociception, and hypothermia and in lung cancer progression (1⇓⇓–4). It is expressed throughout the central nervous system and in the gut, where it binds to at least three different neurotensin receptors (NTRs). NTR1 and NTR2 are class A G protein–coupled receptors (GPCRs) (5, 6), whereas NTR3 belongs to the sortilin family. Most of the effects of neurotensin are mediated through NTR1, where the peptide acts as an agonist, leading to GDP/GTP exchange within heterotrimeric G proteins and subsequently to the activation of phospholipase C and adenylyl cyclase, which produce second messengers in the cytosol (5, 7). Activated NTR1 is rapidly phosphorylated and internalizes by a β-arrestin– and clathrin-mediated process (8), which is crucial for desensitizing the receptor (9). Several lines of evidence suggest that internalization is also linked to G protein–independent NTR1 signaling (10, 11). To improve our mechanistic understanding of NTR1 and to gain additional insight into GPCR features such as helix 8 (H8), we were interested in obtaining a structure of this receptor in a physiologically relevant state.
To date, by far the most successful strategy for GPCR structure determination requires the replacement of the intracellular loop 3 by a fusion protein, as the intracellular domain is otherwise too small to provide crystal contacts. The fusion protein approach has provided a wealth of valuable structural data on GPCRs, but as it renders the crystallized constructs signaling-inactive, the most important functionality—the activation of G proteins—cannot be confirmed for these structures. This leads inevitably to a degree of uncertainty regarding the physiological relevance of intracellular structural aspects, and it also impedes the elucidation of signaling mechanisms, as functional assays and structure determination cannot be performed with the same GPCR constructs.
Crystallization in the absence of fusion proteins was so far mainly possible for rhodopsin (12), the A2A adenosine receptor (A2AR) (13), and the β1-adrenergic receptor (14). Together, they share a high stability, which is either given naturally (rhodopsin) or it is due to stabilizing mutations. High stability appeared to be crucial for crystallographic success, as it allowed the application of harsh short-chain detergents. These tend to form small micelles, which may explain why crystal contact formation can occur under these conditions despite the small extra- and intracellular domains of class A GPCRs.
Besides the stability requirement and/or the necessity of fusion proteins, structural studies of GPCRs have also been complicated by the need of eukaryotic expression systems [e.g., Spodoptera frugiperda (Sf9) insect cells], as prokaryotes exhibit generally low functional expression levels of wild-type GPCRs. However, prokaryotes such as Escherichia coli offer several advantages compared with insect cells, including quick genetic modification strategies, growth to high cell densities, fast doubling times, inexpensive media, absence of glycosylation, and robust handling. Furthermore, E. coli is well suited for producing fully isotope-labeled proteins—a crucial requirement for many NMR studies, which are limited to date.
To exploit these advantages, we recently developed a directed evolution method for high functional GPCR expression levels in E. coli (15). In contrast to screening a few hundred mutants one by one, this strategy allows the simultaneous, competitive testing of >108 different protein variants for highest prokaryotic expression and functionality. Briefly, diverse libraries of NTR1 variants were either obtained synthetically (16, 17) or by error-prone PCR on the wild-type sequence (15). The libraries were ligated to a plasmid encoding an inducible promoter, which was subsequently used to transform E. coli. Selection pressure for high functional expression levels was applied by incubating the induced cells with fluorescently labeled neurotensin, which allowed enrichment of the best expressing cells by fluorescence-activated cell sorting (FACS). The outlined procedure was performed in cycles, leading to a gradual adaptation of the NTR1 population toward high functional expression levels, and additionally, it gave rise to an increase in thermostability for certain variants.
In a second technology, termed CHESS (cellular high-throughput encapsulation, solubilization and screening), we adapted this concept to directly evolve NTR1 variants for high thermostability in short-chain detergent micelles—a property that is not only beneficial for structural studies but also for in vitro drug screening (18). The crucial development of CHESS was to surround, simultaneously, every E. coli cell by a semipermeable polysaccharide capsule. This allows us to solubilize the receptor mutants with harsh short-chain detergents, each mutant inside its own encapsulated cell, all at once and in the same test tube. Both the solubilized receptors and their encoding plasmids are maintained within the same capsules. Long-term incubation under these conditions followed by labeling of the encapsulated solubilized receptors with fluorescent neurotensin and rounds of FACS enrichment ensured a strong selection pressure and a gradual adaption of the NTR1 population toward high stability in harsh short-chain detergents (18).
In this work, we present the crystal structures of three evolved NTR1 variants, which were either obtained by evolving high functional expression levels in E. coli or by directed evolution for stability in detergent micelles. In contrast to the majority of crystallized GPCRs, our NTR1 variants are devoid of bulky modifications at the cytoplasmic face and can thus remain signaling-active, which allows us to gain unique insights into the structure–function relationship of NTR1.
Results
Directed Evolution for High Expression Levels Enabled Structure Determination of NTR1-TM86V.
Directed evolution for high functional expression in E. coli yielded a population of well-expressed NTR1 mutants, which provided a basis for the identification of suitable variants for structural studies. We have chosen to use the variant NTR1-TM86V for crystallization, as it was the most thermostable mutant that was capable of catalyzing GDP/GTP exchange at the heterotrimeric G protein αi1β1γ1 (Gi) in an agonist-dependent way (see Fig. 2 C and D) (17). NTR1-TM86V harbors 11 point mutations (A86L, H103D, H105Y, A161V, R167L, R213L, V234L, I253A, H305R, F358V, and S362A) that confer the high expression levels in E. coli and its stability in detergent solution (Table S1). We observed that the long and putatively flexible intracellular loop 3 and the receptor termini are not required for Gi signaling and hence shortened them to aid crystallization (TM86V-ΔIC3A). The protein could be purified to homogeneity (Fig. S1) in the short-chain detergent nonyl-β-d-glucopyranoside, and it was crystallized by standard vapor diffusion techniques.
The crystal structure of TM86V-ΔIC3A at 3.26 Å [I/σ(I) = 2.0] revealed a canonical GPCR fold (Fig. 1A and Table S2) with seven transmembrane helices (TMs) and the prototypical amphipathic H8 (Fig. 1C). We observed strong electron density for the agonist neurotensin, confirming that the GPCR produced in E. coli reaches a functional conformation despite the absence of the eukaryotic translation and membrane insertion machinery (Fig. 1D, Table S3, and Fig. S2). The resolution was subsequently further improved by a change of the intracellular loop 3 deletion (TM86V-ΔIC3B), which resulted in an additional crystal contact in the same space group. TM86V-ΔIC3B was overall identical to TM86V-ΔIC3A (RMSDCα = 0.3 Å) and could be refined to a resolution of 2.75 Å (Table S2).
Fig. 1.
Structures of three evolved NTR1 variants determined devoid of fusion proteins. (A) The signaling-competent NTR1-TM86V-ΔIC3A (blue) bound to its natural agonist neurotensin (green). All selected mutations for increased expression levels in E. coli and high stability in detergent solution are depicted (orange). (B) Superposition of NTR1-TM86V-ΔIC3A (blue), NTR1-OGG7-ΔIC3A (green), and NTR1-HTGH4-ΔIC3A (orange). (C) Close-up view of the H8 region in NTR1-TM86V-ΔIC3B. Certain hydrophobic contacts of amino acids of the semiconserved H8 motif (beige) are depicted by dashed lines for clarity. The helix dipole of TM7 is illustrated by an arrow. The first of the two palmitoylation sites adjacent to the H8 C terminus is indicated. Note the absence of the palmitoyl moiety due to the prokaryotic expression. (D) Vacuum-electrostatic surface representation (PYMOL) of the neurotensin-binding pocket of TM86V-ΔIC3A. Parallel (Left) and perpendicular (Right) view to the membrane. TM5 is represented as a transparent tube in the Left panel for clarity. Neurotensin is a 13-amino-acid peptide in vivo, but only the C-terminal residues 8–13 were reported to be relevant for binding to NTR1. Strong electron density for these six amino acids was found and allowed us to model the ligand unambiguously (Fig. S2). In addition, relatively weak electron density for two N-terminal linker amino acids (Gly–Gly) of the peptide was observed in one complex of the asymmetric unit (modeled here).
TM86V-ΔIC3A Exhibits the Functional Characteristics of a Typical GPCR.
To verify the physiological relevance of the initial structure of TM86V-ΔIC3A, we characterized the crystallized construct regarding ligand affinities, G protein activation, and neurotensin-dependent internalization. Ligand-binding assays on whole E. coli cells revealed that TM86V-ΔIC3A exhibits an apparent dissociation constant of 2.3 ± 0.4 nM for the agonist neurotensin (cf. wild-type NTR1, 2.8 ± 0.3 nM). In contrast to the agonist, the antagonist SR142948 had never been used as a ligand during directed evolution, but we still observed only a moderate increase in IC50 for TM86V-ΔIC3A (30 ± 2.4 nM) compared with the wild-type receptor (8.4 ± 0.9 nM) (Fig. 2 A and B and Fig. S3), which may be attributable to the point mutation F358V in NTR1-TM86V—a residue that was shown to be specifically involved in antagonist (but not agonist) binding (19).
Fig. 2.
Pharmacological characterizations of the crystallized NTR1 construct TM86V-ΔIC3A. (A) Neurotensin saturation-binding assay of wild-type NTR1 (circles) and TM86V-ΔIC3A (open squares). Note that Bmax levels are not representative for the expression levels of the different mutants, as 10-fold more cells were used for wild-type NTR1 to obtain a similar signal-to-noise ratio—that is, the normalized Bmax would be about 10-fold lower. (B) SR142948 antagonist competition binding experiment using wild-type NTR1 and TM86V-ΔIC3A. (C) GDP/[35S]GTPγS signaling assays of wild-type NTR1, TM86V-ΔIC3A, and TM86V-ΔIC3A L1673.50R in insect cell membranes. Equivalent amounts of active GPCR and reconstituted Gi were assayed in the presence (gray) or absence (black) of neurotensin. The signals correspond to the average of two signaling assays performed in parallel from two independent GPCR expressions, and the error bars represent SDs. (D) Pull-down experiment using immobilized Gi and solubilized GPCR from E. coli membranes. (E) Confocal imaging of living HEK293T cells expressing NTR1, TM86V-ΔIC3A, or TM86V-ΔIC3A L1673.50R-CT (reconstituted D/ERY motif and C terminus) after stimulation with fluorescent neurotensin8-13-HL647 for the indicated times.
To confirm interactions with G proteins, we measured GDP/GTP exchange in membranes containing TM86V-ΔIC3A and the reconstituted heterotrimeric G protein αi1β1γ1 (Gi) (Fig. 2C and Fig. S4). The crystallized GPCR construct exhibited a slightly increased basal GDP/GTP exchange catalysis at Gi compared with wild-type NTR1, which was further stimulated by the addition of agonist. Even though the maximal signaling level is reduced compared with wild-type NTR1, it appears that the crystallized construct is indeed able to bind to and activate Gi. To confirm these observations, we also demonstrated specific Gi binding of detergent-solubilized TM86V-ΔIC3A in a pull-down experiment using immobilized G protein on magnetic beads (Fig. 2D). Basal and agonist-dependent signaling of the crystallized construct TM86V-ΔIC3A was further increased by reverting the mutation R1673.50L [superscript according to Ballesteros–Weinstein (20)] in the highly conserved D/ERY motif (Fig. 2C and Fig. S5). Even though the reintroduction of R1673.50 resulted in significantly reduced expression levels in Sf9 insect cells (Fig. S4A), the thermostability remained almost unperturbed (Fig. S4D).
We also investigated β-arrestin2–dependent desensitization behaviors by confocal microscopy on living HEK293T cells, which coexpressed TM86V-ΔIC3A and β-arrestin2–YFP. Despite the lacking C terminus in the crystallized construct, we observed a weak internalization when bound to fluorescent neurotensin (Fig. 2E and Fig. S6). Furthermore, after reconstituting R1673.50 and the receptor C terminus, a pronounced cointernalization of β-arrestin and fluorescent neurotensin was observed, suggesting that this mutant can indeed interact with β-arrestin2 in a fashion similar to wild type (Fig. S6).
In summary, our pharmacological data clearly suggest that the crystallized NTR1 construct TM86V-ΔIC3A exhibits all essential core functions of a GPCR. Considering the simplicity of expression and genetic modification strategies in E. coli and the high stability of TM86V-ΔIC3A, the protein will likely serve as a valuable model system for future structural and functional studies.
Two Structures of Stability-Evolved NTR1 Variants.
NTR1-TM86V was obtained by evolving high functional expression in E. coli and subsequently by choosing and recombining the most thermostable mutations (16, 17). In contrast, the CHESS technology can directly generate detergent-stable NTR1 variants by an evolutionary process (18). As a proof of this principle, we were interested in confirming the structural integrity of these variants as well. NTR1-OGG7 and NTR1-HTGH4 were generated by CHESS and represent the most thermostable mutants obtained so far. They crystallized readily under various conditions, and the structures were refined to 3.1 Å (OGG7-ΔIC3A) and 3.57 Å (HTGH4-ΔIC3A), respectively. Despite significant sequence variations, OGG7-ΔIC3A and HTGH4-ΔIC3A are structurally nearly identical to TM86V-ΔIC3A (TM86V-ΔIC3A/OGG7-ΔIC3A RMSDCα = 0.4 Å; TM86V-ΔIC3A/HTGH4-ΔIC3A RMSDCα = 0.4 Å) (Fig. 1B and Table S1). This suggests that the ligand-guided selection pressure has favored or preserved the same conformational state in these evolved variants, independent of the particular kind of directed evolution (for functional expression or high stability in detergents).
Because TM86V-ΔIC3A is signaling-active (Fig. 2 C and D) and exhibiting wild-type–like ligand affinities (Fig. 2 A and B) and also desensitization characteristics (Fig. 2E and Fig. S6), it is likely that all our structures represent a naturally occurring conformation of NTR1. Taken together, our four structures of the three different NTR1 variants exemplify that two directed evolution methods in E. coli, which we have recently developed, are valuable tools for structural studies of GPCRs. Our technologies have been applied successfully to a number of other receptors (21), underlining the potential of Darwinian evolution in protein research.
Improved Interhelical Surface Complementarity May Contribute to Increased Thermostability in NTR1-TM86V.
We were interested in identifying the molecular causes of different thermostability characteristics among NTR1 mutants. When comparing the thermostabilities of NTR1-TM86V with one of its precursors, termed NTR1-D03 (15), we uncovered a pattern that sheds light on this issue. NTR1-D03 harbors all NTR1-TM86V mutations except A861.54L, I2535.54A, and F3587.42V. Despite only three amino acid differences, NTR1-D03 exhibited a very low thermostability in the short-chain detergent octyl-β-d-glycopyranoside, whereas NTR1-TM86V exhibited a high thermal denaturation point of 38 °C under these particularly harsh detergent conditions (Fig. 3A). Interestingly, the mutations cause only the replacement of hydrophobic amino acids with other hydrophobic residues. It is striking that the bulky wild-type amino acids at positions 253 and 358, where directed evolution favored a shortening of the hydrophobic side chains, would lead to obvious clashes for all common rotamers in silico (Fig. 3 C and D). At position 86, where the longer leucine was preferred over the shorter alanine, the in silico back-mutation would cause a loss of favorable van der Waals contacts between TM1 and TM2 (Fig. 3B). These observations suggest that improved interhelical surface complementarity contributes significantly to the high thermostability of NTR1-TM86V, and conversely, it may be speculated that optimal helix packing is not required for this particular state of wild-type NTR1 in nature.
Fig. 3.
Improved interhelical surface complementarity correlates with increased thermostability. (A) Thermostability assays of NTR1-D03 (gray) and NTR1-TM86V (orange) bound to neurotensin in the harsh detergent octyl-β-d-glucopyranoside. Note that the low stability of NTR1-D03 in this detergent did not permit an accurate determination of its thermal denaturation transition point. NTR1-D03 and NTR1-TM86V are identical except for three additional mutations in NTR1-TM86V, which must confer this thermostability difference. (B–D) The structure of TM86V-ΔIC3B illustrates the 3-dimensional context at these positions. In silico back-mutating the selected residues (orange) to the wild-type amino acids (gray) would either cause a reduction of favorable van der Waals contacts (green circles in B), or it would lead to steric clashes (red circles in C and D). For the wild-type residues in C and D, the most common rotamers based on the library of PYMOL are shown. (See Fig. S8 for additional rotamers.)
NTR1 Can Adopt a Prototypical Inactive State at the Cytosolic Domain.
In activated GPCR states, the cytosolic ends of TM5 and TM6 were described to be tilted outward relative to their inactive state (22). This is observed in the most prominent way in the structure of the β2-adrenergic receptor bound to GαSβ1γ2 (23). Even though TM86V-ΔIC3A is bound to its natural agonist and capable of triggering GDP/GTP exchange at Gi, the conformations of TM5 and TM6 that were trapped in the crystal are highly similar to dark-state rhodopsin, which represents an inactive or “closed” state (Fig. 4, Fig. S5, and Fig. S7). Our finding is in agreement with other agonist-bound GPCR structures that were crystallized in inactive states, and it provides further evidence that fully active states require the G protein for stabilization. Several structural studies on rhodopsin and on the β2-adrenergic receptor suggest that the observed closed conformation would occlude the G protein–binding site (22⇓⇓–25). Nevertheless, TM86V-ΔIC3A is able to functionally couple to G proteins to a certain degree (Fig. 2C), suggesting that the crystallized construct exhibits structural flexibility and allows a conformational change when bound to agonist. The evolved NTR1 thus shows characteristics consistent with a conformational equilibrium typical for GPCRs: In the absence of a G protein, energetically the most favorable arrangement at the intracellular side of TM86V-ΔIC3A is likely the inactive conformation that was trapped in the crystal. This may also be the case for wild-type NTR1 in the apo-state, as it exhibits very low basal signaling activity toward Gi (Fig. 2C).
Fig. 4.
View from the cytosol onto the superposition of TM86V-ΔIC3A (blue), dark-state bovine rhodopsin (green, PDB ID code 1U19), and β2-adrenergic receptor bound to GαSβ1γ2 (salmon, PDB ID code 3SN6; GαSβ1γ2 is omitted).
Structural Comparison of the Evolved NTR1 Variants to NTR1-GW5.
The observation of a prototypical inactive NTR1 state represents one of the unique features that distinguishes the structures presented in this work from the structure of the NTR1 variant GW5 [Protein Data Bank (PDB) ID code 4GRV] (Fig. 5A) (26). The mutations present in NTR1-GW5 were identified by alanine-scanning mutagenesis, and the protein required expression in Sf9 insect cells, fusion to T4 lysozyme replacing intracellular loop 3 (GW5-T4L), and crystallization in the presence of ligand in lipidic cubic phase (26). The crystallized construct GW5-T4L exhibits a 200-fold increased IC50 value for the antagonist SR48692, and it is signaling-inactive in the presence and absence of the fusion protein. The authors suggested nevertheless that the structure represents an active-like conformation, based on a partial outward tilt of the intracellular end of TM6 and on the observation of a hydrogen bond between R1673.50 and N2575.58. Similar features had previously been found in other GPCR structures, which represent most likely active states (27).
Fig. 5.
Comparison of neurotensin-bound TM86V-ΔIC3B and GW5-T4L. (A) Superposition of TM86V-ΔIC3B (blue) and GW5-T4L (red), view from the intracellular side. The fused T4 lysozyme of GW5 replacing IC3 is omitted for clarity. Black arrows highlight the two different C-terminal conformations and the alternative states of TM6. (B) View along the inner leaflet of the membrane, including a part of the fused T4 lysozyme of GW5. (C and D) Comparison of the ligand-binding pockets, focusing on the interactions of EC3 with neurotensin (green). (C) The 2FO-FC omit map of TM86V-ΔIC3B (contoured at a σ level of 1.2) suggests a single α-helical turn of ECL3 in close proximity to the ligand. (D) In GW5-T4L (26) the loop contains no secondary structural element and it was modeled more distant to the peptide agonist with a cis-peptide following Asp336. Side chains of Ser335 and Gln338 were modeled up to Cβ only.
White et al. (26) suggested that an unusual elongation of TM7 may cause the observed lack of signaling. Indeed, a comparison with our structures reveals that TM7 is extended in GW5-T4L by a peptide segment that corresponds to the amphipathic H8 (Fig. 5 A and B). A canonical H8 would clash into a neighboring molecule in the lipidic cubic phase crystal lattice of GW5-T4L. Instead, the H8 segment resides at the center of the cytosolic domain, where it blocks the prototypical inactive position of TM6 and thus also the putative G protein–binding pocket—an arrangement that has never been observed for other GPCRs.
In contrast, all our structures suggest a canonical H8 (Fig. 1B), and one of the two TM86V-ΔIC3A molecules in the asymmetric unit exhibits no crystal contacts at H8. Furthermore, as described above, TM6 is positioned in a prototypical inactive conformation when bound to agonist. The observed outward tilt of TM6 in GW5-T4L could thus alternatively be explained by the unusual contacts of TM7 to TM6.
Although the cytosolic regions of the evolved NTR1 variants described here are very different from GW5-T4L, at the extracellular side, only one major discrepancy can be observed (Fig. 5 C and D). The 2FO-FC omit map of TM86V-ΔIC3B suggests a single α-helical turn of ECL3 with several ligand contacts (including a salt bridge between D336 and R9 of neurotensin). The same arrangement was found in all structures of the evolved mutants, and it is in agreement with published mutagenesis data (28, 29). In GW5-T4L, on the other hand, the loop contains no secondary structural element and it was modeled significantly more distant to neurotensin with an unusual cis-peptide bond following D336 (Fig. 5D).
NTR1-Specific Determinants of Reduced H8 Stability.
Practically all high-resolution GPCR structures exhibited an amphipathic H8 following TM7. Its presence thus appeared to be a general feature of GPCRs, but surprisingly, the recently determined structures of proteinase-activated receptor 1 (PAR1), chemokine receptor 4 (CXCR4), and NTR1 (GW5-T4L) do not exhibit H8. How relevant are these findings physiologically?
Class A GPCRs exhibit relatively small intracellular domains, and it is thus apparent that the presence or absence of the complete H8 is a major factor determining the characteristics of the cytosolic interface. Multiple lines of evidence suggest an important functional role for this protein segment including G protein coupling and β-arrestin activation (30⇓⇓⇓⇓⇓–36). Our finding of a canonical H8 in the evolved NTR1 variants now shows that a GPCR, which was previously crystallized without H8 being formed, can exhibit a canonical H8 structure (Fig. 1C).
To understand structural key features that are critical for the presence (or absence) of the canonical H8, we compared our structures to A2AR (PDB ID code 4EIY) (Fig. 6 A and B): A2AR likely exemplifies one of the most stable H8 arrangements, as the amphipathic helix shows large contacts to TM1, TM2, IC1, and TM7 and because all reported A2AR structures exhibited a canonical H8 irrespective of the presence or absence of crystal contacts in this region and despite a variety of crystallization conditions.
Fig. 6.
Key interactions of the H8 region in A2AR and NTR1. (A–C) Depicted are the cytosolic ends of TM1, TM2, TM7, and H8 of A2AR (A; PDB ID code 4EIY), TM86V-ΔIC3B (B), and GW5-T4L (C; PDB ID code 4GRV) viewed parallel to the membrane (Left) or from the intracellular side (Right). The yellow arrow in the GW5-T4L structure corresponds to the approximate position of H8 in TM86V-ΔIC3B.
Both A2AR and NTR1 encode the semiconserved H8 motif F(R/K)xx(F/L)xxx(L/F) (Fig. 7). A common feature of our structures and of A2AR is the location of the positively charged guanidinium group of R8.51 (Ballesteros–Weinstein numbering, 8.50 = F376 in NTR1) at the negative helix-dipole at the C terminus of TM7 (Fig. 6 A and B). This interaction likely contributes to the stabilization of the helix break between TM7 and H8, which is not encoded per se, as helix-destabilizing residues (prolines or glycines) are absent in the peptide segment connecting the two helices. Nonetheless, R(K)8.51 is conserved among class A GPCRs (81%) and similar interactions can be found in the majority of published GPCR structures. Another similarity between the crystal structures of A2AR and NTR1 is the absence of palmitoylation membrane anchors, in the case of A2AR because of the absence of cysteines adjacent to the H8 C terminus and in the case of NTR1-TM86V because of expression in a prokaryotic system.
Fig. 7.
Sequence alignment representing the end of TM7 and H8. The sequences are numbered according to Ballesteros–Weinstein (residue 8.50 chosen as F376 of NTR1). The NPxxY and F(R/K)xx(F/L)xxx(L/F) motifs are highlighted (green) and putative palmitoylation sites [experimentally confirmed in NTR1 (44, 49)] are depicted (yellow).
Clearly distinct interactions in NTR1 and A2AR are observed for the most conserved residue F8.50 of the H8 motif. In A2AR, F8.50 is entirely surrounded by a hydrophobic pocket (Fig. 6A). Most prominently, the conserved Y7.53 of the NPxxY motif at H7 exhibits π–π stacking interactions to F8.50, typical for a receptor in the inactive state (13). On the opposite side of its aromatic side chain, F8.50 engages in van der Waals contacts to the poorly conserved L37 in IC1 (sequential numbering of A2AR used for this residue). The aromatic ring of F8.50 is thus sandwiched between Y7.53 and L37. Furthermore, several hydrophobic residues of TM1, TM2, TM7, and H8 contact the CH groups of the aromatic system of F8.50 (not depicted in Fig. 7), and in addition, F8.54 of the H8 motif covers the hydrophobic pocket of F8.50. F8.54 is accommodated between the poorly conserved A7.54 and L1.52.
In our NTR1 structures, F8.50 mediates only weak interhelical interactions and no π–π stacking to Y7.53, as it is only partially inserted into the pocket between TM1, TM2, and TM7 (Fig. 6B). This is due to the following three reasons: First, F8.54 fails to cover the pocket, because it cannot obtain an analogous rotamer conformation to A2AR; it would clash into the longer side chain of N7.54 (A7.54 in A2AR), and V1.52 would not be long enough to stabilize the A2AR-like rotamer (L1.52 in A2AR). Second, the pocket itself is considerably different, as the weakly conserved residues at TM1, TM2, and TM7 contacting the CH groups of the aromatic system of F8.50 in A2AR are less hydrophobic in NTR1 (not depicted in Fig. 6B). And third, because IC1 of NTR1 is longer and presumably flexible (disordered in all our structures and in GW5-T4L), it does not provide a hydrophobic residue like L37 in A2AR to sandwich F8.50 from the intracellular side (Fig. 6 B and C). Additionally, the absence of a structured IC1 in NTR1 causes also another lack of interactions to H8, as the loop mediates not only the L37 to F8.50 contacts in A2AR but also extensive interactions with nonconserved H8 residues (Fig. 6A).
In summary, the shape complementarity (37) between H8 and the receptor is significantly worse in NTR1 (Sc = 0.642) than in A2AR (Sc = 0.81) and the buried surface area is strongly reduced (NTR1, 222 Å2; A2AR, 303 Å2). As described above, these differences are due to alternative amino acids at poorly conserved positions in IC1, TM1, TM2, and TM7, including the residues constituting the pocket of F8.50, and they imply that the canonical H8 arrangement in NTR1 is significantly less stable than in A2AR.
Discussion
Ligand and Palmitoylation Dependence of the Canonical H8 State.
In this work, we present the agonist-bound structures of the three NTR1 variants TM86V, OGG7, and HTGH4, which were generated by directed evolution for high functional expression and for stability in short-chain detergents. In contrast to most other crystallized GPCR constructs so far, TM86V-ΔIC3A not only exhibited wild-type–like ligand-binding properties, it was also able to signal to Gi to some extent. Moreover, when expressed in eukaryotic cells, the classical features of receptor desensitization and internalization were detected, suggesting that the structure derived from this construct resembles a physiologically relevant state. The canonical inactive-like positioning of TM6 is distinct from the outward tilted helix in GW5-T4L, and it is only permitted because of the presence of a canonical H8 that does not occlude this position and the putative G protein–binding pocket. We observed elevated B-factors in the H8 region for all our structures (Fig. S9), and in addition, we found comparatively weak contacts to TM1 and TM7 and a lack of interactions to IC1 and TM2 (Fig. 6B). Considering these observations and the absence of H8 in GW5-T4L, it is tempting to speculate that the canonical H8 of NTR1 is of lower stability than that of the prototypical version in A2AR and/or only partially occupied under certain conditions.
PAR1 and CXCR4 may represent even more extreme cases in this regard, as none of their crystal structures exhibited H8 (38, 39). These absences can be explained by the fact that they are not only different at the nonconserved positions, which cause the reduced H8 stability in NTR, but also by the observation that these two receptors lack parts of the rather conserved H8 motif (Fig. 7). This correlation points to a sequence-specific origin, and thus a naturally evolved feature of physiological relevance.
H8 dynamics have previously been investigated for a number of GPCRs (34, 40, 41). Among all published GPCR structures that include the amphipathic helix, it can be observed that alterations of poorly conserved residues cause a variety of subtly deviating H8 arrangements. The resulting stability differences of H8 likely reflect an evolutionary adaptation of every receptor to the particular requirement on its amphipathic helix. Importantly, the most conserved interaction (the stacking of F8.50 and Y7.53) was described to be disrupted upon agonist binding, as Y7.53 flips toward the G protein–binding cavity upon receptor activation (42). The current body of high-resolution structural data therefore suggests that a reduction of forces that keeps H8 in its canonical arrangement is a common theme of GPCR activation.
Besides intramolecular interactions, another important parameter influences the stability of the amphipathic helix: the number and positions of palmitoyl anchors at its C terminus. Palmitoylation is known to be a reversible and dynamic protein modification that can be cell-cycle–dependent (43) and developmentally regulated (44). In GPCRs, the palmitoylation state was reported to affect G protein signaling, receptor maturation, membrane delivery, phosphorylation efficiency, and desensitization (44, 45). Our finding of an unstable canonical H8 in nonpalmitoylated NTR1 implies that the occupancy of the canonical H8 state of this receptor—and potentially also of other GPCRs—may depend crucially on the palmitoylation state. Considering that the presence or absence of H8 certainly represents an important source of binding specificity to cytosolic interaction partners, it is conceivable that palmitoylation/depalmitoylation events exert their physiological effects in many cases via modulating the stability and dynamics of H8.
Potential of Directed Evolution for Membrane Protein Research.
Most membrane proteins are unsuitable for high-resolution structure determination, because of difficulties in overexpression, instability in detergent solution, or both. To date, the most successful approaches to circumvent these problems rely on trial-and-error procedures, like homology screens or alanine scans, which involve expression, stability, and purification tests of individual proteins in high-throughput formats. Miniaturization has indeed advanced membrane protein structural biology significantly in recent years, but given the resources it takes and the still striking underrepresentation of structural data in the PDB, it is apparent that alternative approaches are needed.
Loss of functionality and low sequence identity to the protein of interest (e.g., by using a bacterial homolog) are frequently accepted as necessary evils on the way to the structure. We have shown in this work that a fundamentally different approach was successfully applied to generate several crystallizable GPCR variants with high sequence identity to the protein of interest (93.2–97.5%) (15⇓⇓–18). Instead of screening mutants or homologs one by one, our method exploited the power of evolution on populations of more than a hundred million GPCR variants at once. Analogous to natural evolution, directed evolution amplified favorable GPCR traits through the alternation of random mutagenesis and selection pressure, allowing a gradual adaptation of the characteristics of the whole GPCR population toward the selected phenotype—it tailored an array of GPCR variants with suitable properties for structural biology independent of previous structural knowledge.
Importantly, the evolutionary system allowed us to determine structures of GPCRs produced in E. coli, thus establishing a prokaryote as a novel and robust host for quantitative, functional, and very rapid GPCR overexpression (15, 21). As E. coli is well suited for producing isotope-labeled proteins, we also provide the basis for an array of NMR studies that were not feasible for this class of membrane proteins so far. Furthermore, the high stability of functional GPCRs generated by directed evolution will facilitate high-throughput ligand screening in vitro, and thus likely contribute to the discovery of new drugs.
Materials and Methods
Construct Design and Expression for Crystallization.
All NTR1 variants were expressed in E. coli using an isopropyl-β-D-thiogalactopyranoside–inducible pBR322-derived vector, which was derived from a plasmid originally obtained as a kind gift from R. Grisshammer (National Institute of Neurological Disorders and Stroke, National Institutes of Health, Rockville, MD) (46⇓–48). They were N- and C-terminally truncated at G50 and G390, respectively, and linked via human rhinovirus 3C protease sites to maltose-binding protein (N terminal) and thioredoxin (C terminal). Amino acids V280-I295 were deleted in the constructs ΔIC3A and E273-T290 in ΔIC3B. Directed evolution of NTR1 was performed as previously described (17, 18). Full details are given in SI Text, and a list of all evolved mutations is given in Table S1.
Purification and Crystallization.
Whole E. coli cells were solubilized in 50 mM Hepes pH 8, 10% (vol/vol) glycerol, 200 mM NaCl, protease inhibitor tablets (Roche), and 0.6% 3-[(3-cholamidopropyl)-dimethylammonio]-1-propane sulfonate (CHAPS), 0.12% cholesteryl hemisuccinate tris salt (CHS), and 1.6% (wt/vol) decyl-β-d-maltopyranoside. All NTR1 variants were purified based on ligand affinity, cation exchange, and size exclusion in nonyl-β-d-glucopyranoside, and they were crystallized in standard vapor diffusion experiments using various mixtures of glucoside detergents and cholesterol hemisuccinate as additives. (See SI Text for details.) The reservoir solutions of the different NTR1 crystallization conditions varied significantly regarding buffering compound (acetate pH 5.5 or glycine pH 9.4), salt (500 mM or 2 M NaCl or 0.2 M CaCl2), and PEG 600 concentrations [20% (vol/vol)–26% (vol/vol)]. (See SI Text for details.) Diffraction data were collected from one single crystal per protein at the Swiss Light Source, and the structure was determined by molecular replacement. (See SI Text for details.)
Functional Assays.
Ligand affinity measurements were performed on whole E. coli cells using either 3H-neurotensin or 3H-neurotensin and the unlabeled NTR1 antagonist SR142948 for competition experiments. (See SI Text for details.) Signaling assays were performed with purified Gi protein (expression in Sf9 insect cells) composed of Gαi1, Gβ1, and Gγ1 and a defined amount of active NTR1, TM86V-ΔIC3A, or TM86V-ΔIC3A-L1673.50R on urea-washed membranes. (See SI Text for details.) Pull-down experiments were performed with purified Gi and solubilized E. coli membranes containing the expressed TM86V-ΔIC3A. (See SI Text for details.) Fluorescence microscopy was performed on living HEK293T cells that were transiently transfected with NTR1 variants and β-arrestin2–YFP. (See SI Text for details.)
Acknowledgments
We thank Prof. Raimund Dutzler for comments on crystallization procedures, Peer Mittl for support during structure determination, Mattia De Luigi and Christian Schori for insights regarding GPCR and G protein purification, and Beat Blattmann and Céline Stutz [National Center of Competence in Research (NCCR) crystallization facility] for their efforts during initial crystallization screening. This work was funded by the NCCR Structural Biology (Schweizerischer Nationalfonds).
Footnotes
- ↵1To whom correspondence should be addressed. E-mail: plueckthun{at}bioc.uzh.ch.
Author contributions: P.E., M.H., C.K., D.J.S., and A.P. designed research; P.E., M.H., C.K., A.B., P.H., S.B., K.M.S., D.J.S., and M.S. performed research; P.E., M.H., C.K., A.B., P.H., K.M.S., D.J.S., M.S., and A.P. analyzed data; and P.E., M.H., C.K., and A.P. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
Data deposition: The atomic coordinates have been deposited in the Protein Data Bank, www.pdb.org [PDB ID codes 4BUO (TM86V-ΔIC3B), 3ZEV (TM86V-ΔIC3A), 4BV0 (OGG7-ΔIC3A), and 4BWB (HTGH4-ΔIC3)].
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1317903111/-/DCSupplemental.
References
Structure of signaling-competent neurotensin GPCR
Article Classifications
- Biological Sciences
- Biochemistry
Sign up for Article Alerts
Jump to section
You May Also be Interested in
Recent experiments and simulations are starting to answer some fundamental questions about how life came to be.
Image credit: Shutterstock/Radoslaw Lecyk.
Archaeologists have long tried to define the transition between the two time periods.
Image credit: Ceri Shipton.
Adam Mastroianni and Daniel Gilbert explore why conversations almost never end when people want them to.
A study finds that giant pandas roll in horse manure to increase their cold tolerance.
Image credit: Fuwen Wei.