Shifting hydrogen bonds may produce flexible transmembrane helices

Abstract

The intricate functions of membrane proteins would not be possible without bends or breaks that are remarkably common in transmembrane helices. The frequent helix distortions are nevertheless surprising because backbone hydrogen bonds should be strong in an apolar membrane, potentially rigidifying helices. It is therefore mysterious how distortions can be generated by the evolutionary currency of random point mutations. Here we show that we can engineer a transition between distinct distorted helix conformations in bacteriorhodopsin with a single-point mutation. Moreover, we estimate the energetic cost of the conformational transitions to be smaller than 1 kcal/mol. We propose that the low energy of distortion is explained in part by the shifting of backbone hydrogen bonding partners. Consistent with this view, extensive backbone hydrogen bond shifts occur during helix conformational changes that accompany functional cycles. Our results explain how evolution has been able to liberally exploit transmembrane helix bending for the optimization of membrane protein structure, function, and dynamics.

There are many advantages of transmembrane helix kinks for membrane protein structure and function. They create weak points for movement during catalytic cycles (1, 2); they enable the precise positioning of key side chains (3); they can help recruit water to functional sites (4); and they can prevent off-pathway folding (5). It is nevertheless surprising that distortions are much more common in transmembrane than soluble protein helices (6, 7), because helices are more stable in the apolar membrane environment (8–11) where backbone hydrogen bonds are stronger (12, 13). It has therefore remained mysterious how distortions can possibly be generated by the evolutionary currency of random point mutations. Significant structural fluctuations have been seen in molecular dynamics simulations of isolated transmembrane helices (6, 14–16), but there have been no experimental measurements of transmembrane helix deformability to our knowledge.

To learn how hard it is to deform a transmembrane helix, we attempted to engineer an alternative helix conformation into a membrane protein so that the energetic differences could be assessed. We focused on the kink at Pro50 in Helix B of bacteriorhodopsin (bR) because it is relatively distant from the retinal chromophore, a useful probe of folding. The helix contains a relatively modest bend ranging from 11–19°, depending on how it is measured. Proline can play a significant role in creating helix distortions because it is incompatible with a regular helix (6, 7, 17, 18). Nevertheless, a P50A mutation does not remove the kink (19), indicating that other residues participate in bending the helix.

We reasoned that the P50A substitution might introduce an energetic cost to bend the helix because a more regular hydrogen bonding pattern could form if the helix straightened. The putative cost must be overcome by favorable side-chain interactions in the bent helix in order to maintain the distortion. It might therefore be possible to break the bend by removing a second key side chain, because there would then be insufficient stabilization to counteract the cost of bending.

By making a second mutation, T46A, we were able to free the helix to adopt several distinct, straighter conformations. However, the straightened helices included noncanonical i → i + 3 hydrogen bonds. We estimate that the energetic differences between these conformations are quite modest, which suggests that shifting of hydrogen bonds within transmembrane helices could provide a mechanism for imparting considerable flexibility into transmembrane helices. Consistent with this view, we find that significant conformational changes can occur in transmembrane helices during functional cycles, and that these structural changes are accommodated by frequent shifting to alternative backbone hydrogen bond patterns.

Results and Discussion

It is speculated that in the hydrophobic membrane environment, transmembrane helix structure can be maintained in the unfolded state (10, 20). In the case of unfolding of bacteriorhodopsin in micelles, approximately 65% of the helix structure remains intact (11, 21), and recent distance measurements by electron paramagnetic resonance throughout the protein are consistent with an unfolded state in sodium dodecyl sulfate (SDS) consisting of mostly helical structure with frayed ends (22). In the case of Helix B specifically, where P50 resides, circular dichroism spectra of a B helix peptide indicates that roughly 19 of the residues remain helical in SDS, whereas a nuclear magnetic resonance structure of a fragment of bR from residues 1–71 finds that the region from 39–62 within helix B remains helical (23). Thus, the current evidence supports a model in which the B helix loses tertiary contacts and frays at the ends in SDS (22), but the hydrophobic center of the helix surrounding P50 maintains its helical character.

As illustrated in Fig. 1A, Helix B will remain at least somewhat distorted in the unfolded state of bacteriorhodopin because the proline blocks canonical helix formation. In the P50A mutant, however, the helix is free to adopt a canonical helix in the unfolded state, but it must become bent in the folded structure. In P50A, the helix bend is not a simple breakage of a single hydrogen bond at residue 50, because the hydrogen bond actually remains intact in the P50A structure (19). Rather, the bend is created by subtle alterations involving many backbone atoms and side chains. Thus, to the extent that there is an energetic penalty for bending, it is paid for in subtle ways throughout the helix.

Fig. 1.

Detecting energetic cost of helix bending. (A) A cartoon depicting the equilibrium between folded and unfolded bacteriorhodopsin molecules. Upon unfolding, Helix B containing P50 is expected to remain at least somewhat distorted (or more flexible) than the same helix containing a P50A mutation. In the case of P50A, the helix is free to adopt a helical structure in the unfolded state, and needs to become bent for the protein to fold properly. (B) The experimental scheme. Helix B contains a kink centered around P50. The P50A mutation does not destroy the bend, but we envision a bending cost is imposed on the helix (relative to the unfolded state not depicted here), represented by the spring. To detect possible bending energy we made side-chain substitutions to alanine in either the wild-type background or the P50A background. A generic side chain X is represented by the red ball and stick and the mutation to Ala is represented by a stick. We then compare unfolding free-energy differences to obtain ΔΔG_U(X → A) (the change in unfolding for the single X → A mutations in the wild-type, P50, background compared to the wild type) and ΔΔG_U(P50A/X → A) (the change in unfolding free energies for the double mutations of X → A in the P50A background compared to the wild-type). If there is no coupling between the sites, the side-chain substitution should have the same effect in both the wild-type and P50A backgrounds. A difference indicates an energetic penalty. (C) The observed unfolding free-energy differences. The changes in unfolding free energy for various side-chain, X, substitutions to alanine in either the P50A background (black) or M56A background (red) are plotted against the same substitutions in the wild-type background. The only exception is for the side chain Y57, which was changed to an F because the A substitution was too destabilizing. The black line represents a least squares fit to all the black points. The red line is a least-squares fit to all the red points. The error bars reflect the standard deviation of two or more independent measurements and the fitting uncertainty from each experiment. Two mutants, Y57F and V49A, produced anomalous unfolding curves which were analyzed as described in SI Text. These results are shown in open diamonds and were not included in the curve fits.

If there is indeed a higher helix bending cost in P50A relative to wild type that is compensated by other side chains in the helix upon folding, we would expect to see energetic coupling effects between the P50A mutant and other side chains in the Helix B. We therefore compared the effects on protein stability of side-chain substitutions in Helix B in both the wild-type and P50A backgrounds.

The P50A Mutation Produces Long-Range Energetic Coupling in Helix B.

A plot of the unfolding free energy contributions of 14 side chains in the P50A versus the wild-type background, shown in Fig. 1B, reveals long-range energetic coupling with remarkable uniformity. In particular, the points fall roughly on a straight line with a slope of 0.75 ± 0.05. These results indicate that energetic contribution of a side chain in the P50A background is reduced by approximately 25% compared with that in the wild-type background. As a control, we performed the same experiment in a different mutant background, M56A, targeting a residue not obviously involved in helix bending. As shown in Fig. 1B, the slope of the line is 1.02 ± 0.02, reflecting no energetic coupling between M56A and other residues in the same helix. Thus, the P50A mutation has long-range energetic consequences throughout the helix.

Why are the coupling effects so uniform? In particular, regardless of the energetic importance of the side chain, its contribution in P50A is reduced by approximately 25%. As outlined in SI Text, this can be explained in terms of a highly simplified model in which we envision the energetic contributions of each of the side chains represented by springs of various strengths. We can then envision that the P50A mutation generates a new straightening force that is resisted by all the springs. The new counteracting force will be distributed among the springs according to their strengths. By analogy, one can imagine a right-handed man holding on to a chin-up bar. If a new weight is now attached to his belt, the additional force will be distributed more toward his stronger right arm than his left. These new forces will ultimately play out in the reduced energetic contribution of each side chain as we observed. We recognize that the situation in the protein is much more complex than this conceptual model, but we believe the basic principle may be operating here.

Freeing the Helix to Adopt a New Conformation.

The presence of possible helix bending cost in the P50A mutant suggests that it might be possible to free the helix to adopt an alternative conformation by the removal of a side chain that is important in bending the helix. The largest contributor to stability in Helix B is T46, and possibly Y57. A T46A mutation lowers stability by approximately 3.0 kcal/mol. A Y57F mutation lowers stability by approximately 3.9 kcal/mol based on the first unfolding transition, but its total contribution to the stability of the protein is difficult to assess because of the presence of a low signal for a second transition (SI Text). We therefore decided to test whether mutations at T46 and Y57 in the P50A mutant background would free the helix to adopt an alternative conformation by determining structures of the single and double mutants. The structure of P50A was determined previously (19) and is very similar to the wild-type structure with only small changes near the site of the mutation. For both Y57F and the double mutant, P50A/Y57F, we observed only modest changes in the structure of the protein (SI Text). Thus, Y57 does not appear to be a critical residue driving the kink. T46 is different, however. While the backbone structure of the T46A single mutant is very similar to the wild-type protein, the double mutant shows a significant straightening of Helix B (Fig. 2A and Movie S1). The T46 is hydrogen-bonded to D96 and forms one of the strongest interhelical hydrogen bonds we measured in prior work (24). The straightening of the helix in the P50A/T46A double mutant suggests that this strong interhelical hydrogen bond helps to effectively pin the helix in a bent conformation.

Fig. 2.

Changes in Helix B structure imparted by mutations. (A) Superposition of Helix B structures for wild-type bacteriorhodopsin and variants. C_α traces are shown. The wild-type structure is shown in black, P50A is shown in green, T46A is shown in blue, and P50A/T46A is shown in red. For each structure, both molecules A and B in the unit cell are shown, but are indistinguishable for all but the P50A/T46A double mutant. Only the residues in the N-half of the kinked helix (residues 39–46) were employed for the superposition to highlight the change in bending. (B) A plot of the backbone O-N distances between residues i and i + 4 throughout helix B for bacteriorhodopsin variants. The wild-type is shown in black, the T46A mutant in blue, the P50A mutant in green, and the P50A/T46A double mutant in red. For all the variants but P50A/T46A, the average distances are shown for the two chains in the asymmetric unit of the crystal, because the structures were so similar. The error bars reflect the standard deviation of the distances between molecules A and B. Because the A and B molecules are quite different in the P50A/T46A double mutant, the distances for the two molecules are shown separately. (C and D) Detailed hydrogen bonding patterns in the structure P50A/T46A seen for molecule A (C) and molecule B (D) in the asymmetric unit of the crystal structure. Backbone O-N distances between residues i and i + 4 for a canonical α-helix, and other side-chain hydrogen bonds are shown in black. Backbone O-N distances between residues i and i + 3 are shown in red. Residues with severely lengthened i → i + 4 O-N distances are highlighted with the yellow stars.

Noncanonical Hydrogen-Bond Patterns in the New Helix Conformations.

In the P50A/T46A double mutant, Helix B is noticeably straightened (Fig. 2A and Movie S1), but the structural change cannot be described by a simple straightening of Helix B into a more regular helix. Instead, Helix B adopts a new yet still distorted structure, with remarkable alterations at the atomic level. Fig. 2B plots the backbone O-N distances for residues i to i + 4 throughout Helix B. For the single mutants P50A and T46A, there are only small differences in the hydrogen bond distances compared to the wild-type protein for both molecules in the crystal asymmetric unit. On the other hand, there are large changes for the P50A/T46A double mutant compared to the wild-type structure as well as between the two molecules in the asymmetric unit of the crystal. Most notably, the canonical i → i + 4 α-helical hydrogen bonds are stretched or broken for residues 38, 39, 40, and 42 in molecule A and for residues 38, 39, 41, and 46 in molecule B. As illustrated in Fig. 2C, the broken hydrogen bond at residue 42 in molecule A is readily explained by a new ordered water molecule that bridges the side chain of D96 and the carbonyl oxygen of F42. However, as shown in Fig. 2D, the backbone hydrogen bond reforms in molecule B and the water molecule disappears. While this leaves the D96 side chain without an observable hydrogen bond partner, the D96 side chain appears to form a stabilizing electrostatic interaction with the aromatic ring of F219 (25). But how can so many of the other hydrogen bonds break without severe energetic consequences?

Further examination reveals that with one exception, the backbone hydrogen bonds have not disappeared but have shifted to make new, or improved, i → i + 3 hydrogen bonds (Fig. 2 C and D). Segments of 3₁₀ helices and π-helices have been observed in transmembrane helices (26–29) and the conversion of α-helices to 3₁₀ helices is thought to be a relatively low-energy transition, particularly in an apolar environment (30). But the hydrogen bonding patterns in the P50A/T46A mutant are not consistent with conversion to a regular 3₁₀ helix and by the criterion used in the program Dictionary of Protein Secondary Structure (31), the structure remains α-helical. Thus, the new conformation is not a canonical 3₁₀ or α-helix.

Noncanonical Hydrogen Bonds Are Parts of a Continuum of Helix Conformations.

Hildebrand et al. made the interesting observation that backbone hydrogen bonds in membrane proteins more commonly show bifurcation between the i → i + 3 and i → i + 4 types (32). Moreover, to assess how commonly the i → i + 3 hydrogen bonds found in transmembrane helices are parts of noncanonical helices as opposed to regular 3₁₀ helix segments, we analyzed a database of transmembrane helices from 41 nonhomologous, high-resolution membrane protein structures described previously (33). Of the 743 i → i + 3 backbone hydrogen bonds identified, only 11% were in contiguous patterns composed of three or more i → i + 3 hydrogen bonds, leaving 89% that cannot be part of a 3₁₀ helix. These results suggest that hydrogen bond shifts are part of a continuum of helical conformations rather than wholesale conversions to new helix types.

Small Energetic Cost of Helix Distortions.

The fact that the P50A/T46A mutant helix adopts a somewhat distorted structure over a more ideal α-helix suggests that the hydrogen bond shifts are relatively low-energy transitions. To obtain an actual experimental estimate of the energetic consequences of the engineered helix change we engineered in bR, we constructed the thermodynamic cycle illustrated in Fig. 3. In essence, we can measure the free energy contributions of the P50A and T46A mutations with and without helix bending. The result suggests that bending the helix only costs approximately 0.6 kcal/mol in the context of the protein. Moreover, the fact that we see multiple helix conformations in the P50A/T46A double mutant indicates that transmembrane helices are relatively flexible when freed from other constraints. Indeed, if bending the helix costs only 0.6 kcal/mol, the conformation seen in the wild-type protein could be explored by thermal motions.

Fig. 3.

Measuring the energetic difference between helix conformations. We have measured the contribution of two mutations, P50A and T46A, to stability with and without the inclusion of a conformational change in the helix. ΔG_AT and ΔG_PA are the free-energy differences between the wild-type and the T46A mutant or P50A mutant, respectively. The ΔG_conf is the free-energy difference between two helix conformation states. The blue arrows indicate the arbitrarily chosen direction of the reactions. The free-energy differences obtained from the unfolding free energies are indicated for each reaction. By comparing the horizontal and vertical reactions we can extract ΔG_conf , highlighted in red.

Rampant Hydrogen Bond Shifting Seen in Conformational Changes.

If alternative hydrogen bonding patterns provide low-energy pathways for helix bending that are accessible by thermal motions, we might also expect to see hydrogen bond shifts during the conformational changes that occur in catalytic cycles. We therefore examined the backbone hydrogen bonding patterns that occur in the sarcoplasmic Ca²⁺-ATPase, for which there are many structures representing distinct stages in the transport cycle (34). As shown in Fig. 4 A–E, the backbone hydrogen bonding patterns in the transmembrane helices change significantly during pumping. These changes in hydrogen bonding patterns translate into clear conformational changes in the helices. Fig. 4F and Movie S2 show a particular helix straightens and kinks during pumping with extensive hydrogen-bond shifting. These hydrogen bond shifts are not conversions to alternative regular helices but rather isolated shifts to accommodate local structural deformations.

Fig. 4.

Helix flexing by hydrogen-bond shifting during conformational changes. (A–E) The backbone hydrogen-bond patterns in the TM region of (A) Ca₂E1∶ATP (1T5S), (B) [H_n]E2P∶P∶ATP (1WPG), (C) E2P molecule A (2ZBE), (D) E2P molecule B (2ZBE) and (E) [H_n]E2-P∶ATP (2ZBG) were compared with the backbone hydrogen-bond patterns in the TM region of the ligand-free form, (1SU4). Backbone hydrogen bonds that are found in the reference model, Ca₂E1 , but not in the comparing model are labeled with open black diamonds. Backbone hydrogen bonds which are found in the comparing model but not in the reference model are labeled with open red squares. The TM regions are labeled with solid green rectangles. The five structures used all had resolutions of 2.7 Å or better. (F) An example of the resulting helix alterations for the TM helix spanning residues 762 to 780. The C_α atoms of residues 762–770 in the N-half of the helix from each comparing model are aligned with those from the reference model. The hydrogen bonds shown and the color scheme are the same as in A–E.

Conclusion

Our results indicate that transmembrane helices are quite flexible. Thus, the introduction of kinks appears to be well within the realm of simple evolutionary steps, and helix distortions can be readily accessed during conformational changes. We propose that flexibility may at least in part be explained by backbone hydrogen bonding donors and acceptors shifting to different partners. The introduction of a Pro residue in a helix is one dramatic way to distort helices, and our results indicate that the structure of transmembrane helices can be altered readily by a single-point mutation because the energy cost for helix bending is not extremely high. In the absence of P50, Helix B in bR can be shifted to distinct conformations by a single mutation (T46A). We had earlier proposed that most kinks are introduced by proline mutations which can then be removed once the kink is fixed in the structure (3). This hypothesis was supported by our finding that at the vast majority of kinks we found prolines in homologous proteins, suggesting an evolutionary connection. More recent work with much larger databases suggests that these so-called vestigial prolines are less frequent than we had observed (6, 7, 33). Moreover, we found that the introduction of Pro into a transmembrane helix is generally very deleterious and is likely to happen only infrequently (17). Another possibility is that prolines are accommodated favorably after the kink structure is generated by other mutational pathways, which appear to be well within the realm of simple evolutionary steps. Our results indicate that membrane protein structure is much more malleable than we might have imagined, a feature that has apparently been essential for optimizing membrane protein structure and function.

Materials and Methods

Preparation of Mutant bR Proteins.

All mutant bR proteins were prepared as described previously (19, 35).

bR_f-to-bO_u Unfolding Assays.

The bR_f-to-bO_u unfolding assays of wild-type and mutant bR were performed with the addition of excess all-trans retinal added as described (36).

X-Ray Crystallography.

The mutant bR proteins were crystallized by vapor diffusion using the bicelle method (37, 38). Purple membrane in water at > 10 mg/mL was mixed with a 40% (w/v) 2.8∶1 DMPC/CHAPSO bicellar solution at a volume ratio of 4∶1 and equilibrated on ice for more than 1 h in the dark. The drops contained 2 μL of the protein/bicelle solution mixed with 0.75 μL of well solution composed of approximately 3 M sodium phosphate (pH 3.6–4.0), 3.5% triethylene glycerol, and 0.03 M 1,6-hexanediol. After grown in the dark for 2–14 d at 37 °C, crystals were washed in the well solution, dipped into a liquid perfluoropolyether cryoprotectant, and frozen in liquid nitrogen. Diffraction data were collected at the Advanced Photon Source beamline 24-ID-C and were integrated and scaled by using the DENZO/SCALEPACK program package (39). Structures were phased by molecular replacement with PHASER (40) using the wild-type bR model 1PY6. T46A and P50A/T46A were refined at resolutions of 2.47 Å and 2.37 Å using CNS (41) and COOT (42) software with the twinning operator (-h, -k, h + l) applied to the reciprocal lattice at twin fractions of 0.31 and 0.50, respectively. Five percent of the reflections, which were the same as those omitted in the refinement of the wild-type structure 1PY6, were selected to calculate R_free for both of the mutants. Composite omit maps for T46A and P50A/T46A were computed by using detwinned reflection data (see SI Text). T46A was detwinned as described (43), but for detwinning the highly twinned P50A/T46A crystal. we used the method of Redinbo and Yeates (44), employing coordinates from the wild-type structure (1PY6) with residues 46–50 deleted. The Y57F and P50A/Y57F were refined at resolutions of 2.06 Å and 2.40 Å using REFMAC (40) and COOT (42) software, respectively. Five percent of the reflections, which were the same as those omitted in the refinement of the wild-type structure 1XJI, were selected to calculate R_free for both of the mutants. Data collection and refinement statistics details are given in SI Text.

Backbone Hydrogen Bonds.

Backbone hydrogen bonds of T46A and P50A/T46A mutant bR were identified using HBPLUS (45) and COOT (42).