The evolution of transmembrane helix kinks and the structural diversity of G protein-coupled receptors

Abstract

One of the hallmarks of membrane protein structure is the high frequency of transmembrane helix kinks, which commonly occur at proline residues. Because the proline side chain usually precludes normal helix geometry, it is reasonable to expect that proline residues generate these kinks. We observe, however, that the three prolines in bacteriorhodopsin transmembrane helices can be changed to alanine with little structural consequences. This finding leads to a conundrum: if proline is not required for helix bending, why are prolines commonly present at bends in transmembrane helices? We propose an evolutionary hypothesis in which a mutation to proline initially induces the kink. The resulting packing defects are later repaired by further mutation, thereby locking the kink in the structure. Thus, most prolines in extant proteins can be removed without major structural consequences. We further propose that nonproline kinks are places where vestigial prolines were later removed during evolution. Consistent with this hypothesis, at 14 of 17 nonproline kinks in membrane proteins of known structure, we find prolines in homologous sequences. Our analysis allows us to predict kink positions with >90% reliability. Kink prediction indicates that different G protein-coupled receptor proteins have different kink patterns and therefore different structures.

Transmembrane helices contain many more bends than helices in water-soluble proteins. Indeed, the vast majority of transmembrane helices contain significant distortions from ideal helix geometry (1). For example, six of seven helices in rhodopsin contain significant irregularities (2). Some helices even contain multiple bends. These distortions may be tolerated because of the high stability of helices in a membrane environment (1, 3). Helix distortions are one mechanism for creating structural diversity from the relatively simple building blocks used in helix bundle membrane proteins.

Most transmembrane helix deformations (≈60%) occur at proline residues. In the middle of a helix, proline distorts ideal α-helix geometry because of steric conflicts with the preceding residue and the loss of a backbone H bond (4). These proline-induced distortions in membrane proteins may be functionally important. For example, they can create a weak point in a helix that is thought to facilitate movements required for the function of some transmembrane channels (5–7). Alternatively, prolines in transmembrane helices can be required to facilitate folding by the prevention of off-pathway intermediates (8).

Not all helix bends occur at proline. Indeed, ≈40% of all transmembrane helix distortions are at nonproline residues. The causes of nonproline distortions are less obvious. These structural alterations could be driven by local or long-range tertiary interactions. If local interactions induce distortions, it may be possible to identify particular sequence motifs that specify a bend or kink (9). So far, no such motifs have been identified, however. If tertiary interactions drive the helix distortion, the evolutionary path creating these nonproline kinks would appear to be quite complex given the large number of contacts made by each helix. Here, we present evidence that the evolutionary path to nonproline kinks involves a simple local deformation caused by the introduction of a proline, followed by packing adjustments and later removal of the proline, leaving the kink stabilized by tertiary interactions. These results lead to a simple method for the reliable prediction of nearly all kink positions in membrane proteins from sequence information alone.

Our method for kink prediction allowed us to examine structural diversity within the G protein-coupled receptor (GPCR) family. GPCRs represent one of the largest protein families and are the target of the majority of drugs on the market (10). Despite their importance, the only GPCR structure known is bovine rhodopsin (2). Thus, rhodopsin has served as a template for a large number of efforts to model other GPCRs (10). We find that different GPCR families have different patterns of helix kinking, however, indicating significant structural diversity within the GPCR family.

Methods

Bacteriorhodopsin (bR) mutants P91A and P186A were created and purified as described (11). The SDS unfolding experiments were performed as described (11). Both mutants were crystallized by using the bicelle method (12). Four microliters of a 4:1 protein/bicelle solution of 40% (wt/vol) 2.8:1 1,2 dimyristoyl-sn-glycerol-3-phosphocholine/3[3-cholamidopropyl)dimethylammonio]-2-hydroxy-1-propanesulfonate was mixed with 1.5 μl of a well solution, inverted over the well solution, and incubated at 37°C. P91A was crystallized by using a well solution containing 2 M NaH₂PO₄ (pH 3.7) and 0.09 M hexanediol. The well solution for P186A was 2 M NaH₂PO₄ (pH 4.5), 0.8 M NaH₂PO₄ (pH 3.7), and 0.09 M hexanediol. Crystals were transferred to a cryoprotectant solution of 4.0 M NaH₂PO₄ (pH 3.7) before freezing under a stream of liquid nitrogen. Data collection was performed at Advanced Light Source Beamline 8.2.2 at the Berkeley National Laboratory. X-ray diffraction data were processed and scaled by using the denzo package (13), and the structure was solved by molecular replacement using the 1PY6 structure (11) and refined by using cns with a twinning fraction of 0.50 (14). Five percent of the reflections were withheld for the calculation of R_free. The same reflections withheld in the WT refinement were used for both mutant refinements to eliminate bias. P91A was refined to an R factor of 21.5 and an R_free of 26.7, and P186A was refined to an R factor of 19.7 and an R_free of 25.5.

For the test of kink prediction, all helical membrane proteins solved by 2002 at a resolution of 2.7 Å or better were collected, and if two proteins had >25% sequence identity, the lower-resolution structure was rejected. This process left a set of 10 protein structures [1AIJ (15), 1AR1 (16), 1F88 (2), 1C3W (17), 1GUE (18), 1JB0 (19), 1FX8 (20), 1J4N (21), 1KZU (22), and 1K4D (23)].

To build multiple sequence alignments, homologous sequences were identified by searching release 66 of the Protein Information Resource database (http://pir.georgetown.edu) by using the fasta program (24) implemented in the GCG package, version 10.2, by the Genetics Computer Group, University of Wisconsin. Sequences with >25% sequence identity were included in a multiple sequence alignment performed by using the GCG program pileup.

To help identify kinks in the known structures and measure the bend angles, we calculated the angles between the helix axes formed by all pairs of five residue segments throughout each helix. To find the expected deviations that would be seen in a regular long helix we determined angles for the helix of the leucine zipper GCN-4 (2ZTA). For GCN-4, the mean angle was 7.1° with a SD of 0.7°. All reported kinks were >11°, which is 5.6 SDs from the mean value for a regular helix.

Results

Tertiary Interactions Drive Helix Kinks. bR contains three helices that are bent at proline residues (P50, P91, and P186). These prolines can be changed to alanine without significant alterations in the spectrum of bR or proton pumping activity (25, 26). In prior work, we found that the P50A mutant does not affect stability, and a crystal structure of P50A revealed that the kink remained intact (11). Thus, P50 is not required to induce the kink in helix B, suggesting that tertiary interactions drive the distortion of this helix. This surprising result motivated us to investigate the structures of P91A and P186A.

We first measured the thermodynamic stability of P91A and P186A by using an SDS unfolding assay (11, 27). As shown in Fig. 1, both P91A and P186A unfold at lower SDS concentrations than the WT protein. At an SDS concentration of 0.6 mole fraction, the corresponding change in free energy, ΔΔG_u, is –1.3 ± 0.3 kcal/mol for P91A and –0.9 ± 0.1 kcal/mol for P186A. Thus, unlike P50, both P91 and P186 contribute to the stability of the protein.

Fig. 1.

Unfolding curves for the WT and mutant proteins. SDS unfolding data are shown for WT (○), P91A (□), and P186A (⋄) proteins. The curves are fits to the data assuming a two-state unfolding model and a linear relationship between SDS concentration and unfolding free energy (11). Linear extrapolation of the WT curve back to zero denaturant yields an unfolding free energy of 13.0 ± 0.1 kcal/mol.

To examine the detailed structural consequences of the P91A and P186A mutations, we solved the crystal structures of both mutants at 2.1 and 2.3 Å, respectively. Fig. 2 shows WT/mutant structural superpositions of helices B, C, and F containing positions 50, 91, and 186. The overall structure of the helices is the same in the WT and all of the mutants.

Fig. 2.

Structural superpositions of WT and mutant transmembrane helices. C^α traces of the WT (Right) and mutant (Left) helices are shown, along with their optimal superpositions (Center). (A) Transmembrane helix B of WT bR and P50A. (B) Transmembrane helix C of WT bR and P91A. (C) Transmembrane helix F of WT bR and P186A.

The mutations cause local structural changes that are illustrated in Fig. 3, which compares the detailed structures, and in Fig. 4, which compares N_i+4···O_i distance plots. Although the P50A mutation results in the smallest change in stability, it leads to the most significant change in the structure (11). In particular, the distance between the O₄₆ and N₅₀ is 4.2 Å in the WT protein, but closes to 3.3 Å in the mutant, restoring the helical backbone hydrogen bond. Pro cannot make this H bond because there is no amide proton. As shown in Fig. 4A, the backbone hydrogen bonding distance is only slightly longer than average. Thus, the bend in the mutant helix is accommodated by the cumulative effects of small, correlated hydrogen bond stretches on one side of the helix. The fact that P50A is as stable as the WT protein suggests that, for a bend of this magnitude, the energetic gain of a backbone hydrogen bond compensates for the required stretching of other backbone H bonds to create the bend.

Fig. 3.

Comparison N _i+4···O_i distances. (A) Transmembrane helix B of WT bR and the P50A mutant. (B) Transmembrane helix C of WT bR and the P91A mutant. (C) Transmembrane helix F of WT bR and the P186A mutant.

Fig. 4.

Local structural comparison between WT and mutant structures. (A) Comparison of P50A with WT. The backbone hydrogen bond between Ala-50–N and Thr-46–O is made in P50A (3.3-Å distance), whereas the WT cannot form the hydrogen bond because of the presence of the proline ring (4.2-Å distance between Pro-50–N and Thr-46–O). (B) Comparison of P91A with WT. Ala-91–N makes a possible weak N _i+3 to O _i hydrogen bond with Phe-88–O (3.6-Å distance). Ala-91–N is too far from the Thr-87 carbonyl oxygen to make a regular backbone hydrogen bond (4.3 Å). WT Pro-91–N is slightly further from both Thr-87–O (4.6 Å) and Phe-88–O (4.0 Å). (C) Comparison of P186A with WT. Neither WT nor P186A forms regular backbone hydrogen bonds. In both proteins, Tyr-185–N makes a N _i+3 to O _i hydrogen bond with W182-O (3.1 Å).

In contrast to P50A, the helical backbone N_i+4···O_i H bond is not restored in the P91A and P186A mutants. In the WT protein the N₉₁-O₈₇ distance is 4.6 Å and in P91A it is 4.3 Å. P91A forms a weak N_i+3 to O_i hydrogen bond with the F88 carbonyl oxygen (3.6-Å distance in contrast to 4.0-Å distance in the WT). Like P91A, the kink in the P186A mutant remains intact, and the N₁₈₆-O₁₈₂ backbone H bond still does not form (3.9-Å distance in the mutant versus 4.3 Å in the WT structure). In both the WT and mutant structures, the carbonyl oxygen of position 182 is close enough to the amide nitrogen of position 185 to form a N_i+3 to O_i backbone H bond in WT bR and P186A (3.1-Å distance in both the WT and mutant structures) as seen in 3₁₀ helices (9). Thus, in both P91A and P186A, the mutations introduce a new amide hydrogen with no hydrogen bonding partner. This unsatisfied hydrogen bond donor may at least partially explain the reduced stability of the mutant protein.

These results demonstrate that the structures of the proline mutants are similar to the WT protein in near atomic detail. Thus, the earlier results with P50A are not an aberration, but rather the norm. Proline is not required for helix kinking, indicating that in many cases, the kinks are driven by tertiary interactions. Why then are prolines so common at transmembrane helix distortions?

An Evolutionary Hypothesis. As illustrated in Fig. 5, we propose that over the course of evolution, kinks initially occur in transmembrane helices by mutation to a proline residue. These prolines force helix distortions because of side-chain steric constraints, but they may be tolerated or have adaptive value at certain positions (28) and are therefore retained in future generations. Over evolutionary time, additional mutations accumulate that act to stabilize the kinked form of the helix (29), leading to a modern protein containing a proline kink. At this point, the proline is no longer required to maintain the kink. Thus, when we replace the prolines at kinks in bR, we see no change in the structure. Similarly, a proline at a kinked position may be replaced over the course of evolution, leaving a nonproline kink.

Fig. 5.

Evolutionary hypothesis for the generation of kinks. (A) A primordial membrane protein with a straight transmembrane helix. (B) A proline mutation occurs in the middle of a transmembrane helix, causing a kink. (C) Further mutations occur that optimize packing around the distorted helix. (D) Proline is changed to a nonproline residue. The kink remains intact, supported by tertiary contacts.

A Test of the Hypothesis and Kink Prediction. If our hypothesis is true, we would expect to see remnants of the evolutionary pathway in sequence alignments. In particular, at positions of nonproline kinks, we should see prolines at equivalent positions in related sequences. Fig. 6 shows the frequency of prolines at positions in sequence alignments of the M subunit of the Rhodobacter sphaeroides photosynthetic reaction center (PRC-M) and the A subunit of cytochrome C oxidase (COX-A). The structures of PRC [1AIJ (15)] and COX are known [1AR1 (16)]. As shown in Fig. 6, two of the five helices in PRC-M and 7 of the 12 helices in COX-A contain significant distortions. Seven of the 12 kinks are at proline positions in the proteins of known structure. These prolines appear to be reasonably, although not completely, conserved. It is possible that in the homologous proteins without prolines at these positions, the helices are kinked, but this remains to be seen. It is interesting to note, however, that a nonproline kink in the L subunit of PRC is known to occur at the same position as a proline kink in the homologous M subunit (30). Significantly, at the four of the five positions where nonproline kinks occur, a peak of prolines is found in the alignment at that position. Moreover, we did not find significant peaks of prolines (>10% of the sequences) where kinks do not occur.

Fig. 6.

Correspondence between kink positions and the presence of prolines at the same position in related proteins. The number of prolines in a set of aligned sequences is plotted versus the sequence position. Only the transmembrane helix regions are plotted. C^α traces of the kinked helices are shown, and centers of the kinks are indicated in red. Proline kinks are labeled with black lettering, and nonproline kinks are labeled with red lettering. For each bend, the angle of helix axis deviation is indicated. (A) Proline occurrence in 40 aligned sequences of the PRC-M and the structure of the R. sphaeroides PRC [1AIJ (15)]. (B) Proline occurrence in 80 A-subunit COX sequences and the structure of COX-A [1AR1 (16)]. Similar plots for other structures and sequence relatives are given in Fig. 8.

We further examined eight other unrelated membrane protein structures solved at 2.7 Å or better, leading to a total of 39 kinks or other distortions (see Fig. 8, which is published as supporting information on the PNAS web site). Of the 39 total kinks, 22 occur at a proline residue and 17 occur at nonproline residues. We find that >10% of the sequences in an alignment contain proline at 14 of the 17 nonproline kinks. Moreover, no peaks of prolines (>10% of the total number of sequences) were found at nonkinked positions. Thus, >90% (36 of 39) of transmembrane kinks can be predicted from sequence information alone, by finding a proline in the sequence of interest, or by finding a peak of proline occurrences in a set of homologous sequences, with no false positives.

Kink Patterns in GPCRs. With a reliable method for kink prediction, it is possible to examine whether kinks can be used to introduce structural divergence and consequent functional variation within a protein family. Although all GPCRs have seven helices and transmit signals to G proteins, they interact with diverse ligands from small molecules to proteins and should therefore exhibit significant structural diversity (31). GPCR sequences have been divided into different classes based on sequence similarity. We constructed sequence alignments by using template sequences from three different classes: rhodopsin from class A, human secretin receptor from class B, and human metabotropic glutamate receptor 8 (MGR8) from class C. All sequences used in the alignments exhibit >25% sequence identity to the template sequence. For each set we then plotted the number of prolines for each position in the alignment, shown in Fig. 7. As above, we predict a kink to occur when >10% of the sequences contain proline at a position. By this criterion, we predict six kinks in the rhodopsin (class A) structures, which correspond precisely with the kinks in the known structure (see Fig. 7 Top). Although helices A, B, C, E, F, and G are kinked in rhodopsin, we predict that D, E, and F are kinked in the secretin receptor (class B) structure. Moreover, whereas helix D is straight in rhodopsin, we predict a highly distorted D helix for the secretin receptor, with a total of three kinks. For MGR8 (class C), we predict many more helical distortions than seen for rhodopsin. We predict that all of the helices in MGR8 are kinked, and three of them contain multiple kinks. These different kink patterns are indicative of considerable structural diversity within the GPCR protein family.

Fig. 7.

Predicted kink patterns in different GPCRs. The number of prolines in a set of aligned sequences is plotted versus the sequence position. Only the transmembrane helix regions are plotted. (Top) Proline occurrence in 80 aligned rhodopsin sequences. (Middle) Proline occurrences in 58 aligned human secretin receptor sequences. (Bottom) Proline occurrences in 20 aligned MGR8 sequences.

Discussion

Our results are consistent with a proline origin of nonproline kinks. We cannot rule out the possibility that kinks are first induced by tertiary interactions, and then followed by a favorable proline replacement at the site of the kink. Given the many tertiary contact adjustments that would need to be made to impart a kink, however, we believe our hypothesis is the simplest and therefore most likely scenario. Perhaps more importantly, our results provide a remarkably simple and reliable method for predicting the positions of kinks in transmembrane proteins. By identifying peaks of prolines in sequence alignments, we can predict 36 of all 39 kink positions (92%) and 14 of 17 nonproline kink positions (82%) without any false positives. Moreover, the approximate direction of the kink can be inferred from the proline peak positions, i.e., away from the missing backbone H bond at proline (32). At this point we cannot predict the magnitude of the kink, however, which may depend on both local and long-range interactions (29). Our prediction of variable kink positions in different GPCR proteins indicates that caution must be applied when modeling GPCRs using a rhodopsin template. Our kink predictions should serve as a useful guide for how to adjust structural models for GPCRs based on the rhodopsin structure. Thus, we believe our simple algorithm will prove to be extremely valuable for the prediction of membrane protein structure.