Membrane Thickness Varies Around the Circumference of the Transmembrane Protein BtuB

Abstract

BtuB is a large outer-membrane β-barrel protein that belongs to a class of active transport proteins that are TonB-dependent. These TonB-dependent transporters are based upon a 22-stranded antiparallel β-barrel, which is notably asymmetric in its length. Here, site-directed spin labeling and simulated annealing were used to locate the membrane lipid interface surrounding BtuB when reconstituted into phosphatidylcholine bilayers. Positions on the outer facing surface of the β-barrel and the periplasmic turns were spin-labeled and distances from the label to the membrane interface estimated by progressive power saturation of the electron paramagnetic resonance spectra. These distances were then used as atom-to-plane distance restraints in a simulated annealing routine, to dock the protein to two independent planes and produce a model representing the average position of the lipid phosphorus atoms at each interface. The model is in good agreement with the experimental data; however, BtuB is mismatched to the bilayer thickness and the resulting planes representing the bilayer interface are not parallel. In the model, the membrane thickness varies by 11 Å around the circumference of the protein, indicating that BtuB distorts the bilayer interface so that it is thinnest on the short side of the protein β-barrel.

Introduction

The hydrophobic matching of membrane proteins to lipids is a well-accepted concept (1,2). Membrane thickness can affect the activity (3–5), orientation (6), stability (7), and the aggregation state of membrane proteins (8,9). In addition, hydrocarbon thickness has been shown to modulate the orientational order of β-barrel membrane proteins (10–12), their tilt within the bilayers (13), and their dynamics and/or structure (14). The energy required for small changes (2–3 Å) in bilayer thickness and minor membrane protein conformational or orientational changes is expected to be low relative to the energies associated with breaking or leaving hydrogen bonds unsatisfied, burying charged or highly polar side chains, or exposing hydrophobic side chains. As a result, the response of a membrane to a modest (∼5 Å) hydrophobic mismatch may include changes in bilayer hydrocarbon thickness (15,16) and relatively low energy protein conformational or orientational changes (17). At the time of this writing, relatively little is known regarding the uniformity of lipids around membrane proteins, although high-resolution NMR indicates that lipids in bicelles may solvate the protein-lipid interface in a somewhat heterogeneous manner (18).

BtuB belongs to a family of bacterial membrane transport proteins that are termed “TonB-dependent.” These transporters bind and move trace nutrients, such as forms of iron and vitamin B₁₂, into the periplasmic space, and they derive energy for transport from the inner membrane proton potential by coupling to the inner membrane protein TonB (19). Each member of this transporter family has a homologous structure, which is based upon a 22-stranded antiparallel β-barrel (see Fig. 1 a). Although the structures of these β-barrels have large hydrophobic surfaces, the structure of the BtuB barrel, as well as other members of the TonB-dependent family, is highly asymmetric. The length of the barrel is much shorter on the surface near β-strand 1 than it is on the opposite side of the protein. Near strand 1, regularly H-bonded β-strands are ∼25 Å in length, while on the opposite side of the barrel, near strand 11, the regular β-strands are 37 Å or greater in length. These lengths, projected onto the barrel axis, correspond to barrel heights of 19 Å and 28 Å, respectively (20), and this difference in strand length across the barrel suggests the lipid interface may not be uniform around the protein circumference.

Figure 1.

Model obtained by x-ray crystallography for BtuB (PDB ID: 1NQE), the Escherichia coli vitamin B₁₂ transporter (50). The transporter consists of a β-barrel formed from 22 antiparallel strands, where 136 residues on the N-terminal end of the transporter are folded within the interior of the barrel. (Shaded area) Phosphorus-to-phosphorus thicknesses of a POPC bilayer, having a distance of ∼37 Å (24,53). (Solid lines) Positions of the carbonyls in POPC. (Dashed line) Position of the membrane center. Shown on the model are the positions of Cα carbons for 21 sites that were individually spin-labeled in this study.

Based on the crystal structures of BtuB and the energetics of transfer from an aqueous to hydrophobic environment, estimates of the region of BtuB that is embedded in the membrane have been made (21,22). Using this approach, the barrel axis was found to be nearly parallel to the bilayer normal, with the periplasmic turns of BtuB localized near the periplasmic membrane interface. However, in these studies it was assumed that the two membrane interfaces were parallel. Molecular dynamics simulations have also been used to assess the intramembrane location of BtuB, and the results generally show good agreement with experimental data (23).

In the work described here, we examine the lipid-protein interface of the outer membrane transporter BtuB when reconstituted into bilayers of palmitoyloleoylphosphatidylcholine (POPC). Although POPC does not mimic the heterogeneous and asymmetric composition of the Escherichia coli outer membrane, it has a hydrophobic thickness of 27 Å (24), which is similar to the 30 Å hydrocarbon thickness of bilayers formed from the lipopolysaccharide component in the outer membrane (25). Moreover, the inner leaflet of the outer-membrane consists of lipids having primarily 16- and 18-carbon-length acyl chains (26). As a result, the hydrophobic matching in POPC is expected to be similar to that in the native outer membrane where BtuB resides.

In this study, an electron paramagnetic resonance (EPR)-based method termed “site-directed spin-labeling” (27–29) is used to estimate distances relative to the bilayer interface for the spin-labeled side chain R1 (Fig. 1 b) when it is incorporated into 21 outward facing sites in BtuB. The software package Xplor-NIH (30), along with EPR-derived bilayer depth restraints and periplasmic loop restraints, were used to perform simulated annealing and determine the position of two planes that represent the bilayer interface surrounding BtuB. The results indicate that the interior and exterior lipid interfaces are not symmetrically arranged around the protein. The two planes that define the interface are not perpendicular to the barrel axis and are tilted relative to each other so that bilayer thickness varies by as much as 11 Å from one side of the protein to the other. Although the use of a plane is a crude approximation for the intersection of the bilayer with a membrane protein, the data indicate that a mismatch between the POPC and BtuB hydrophobic regions occurs around most of the BtuB circumference. The functional consequences of such a mismatch are discussed.

Materials and Methods

Materials

DL-dithiothreitol was purchased from Sigma (St. Louis, MO), sarkosyl was obtained from Fisher Chemical (Pittsburgh, PA), and phenylmethanesulfonyl fluoride was purchased from Boehringer-Mannheim (Indianapolis, IN). The detergent n-Octyl-β-D-glucopyranoside, ANAGRADE, was purchased from Anatrace (Maumee, OH) and POPC (1-palmitoyl-2-oleoyl-sn-glycero- 3-phosphocholine) (16:0, 18:1) was purchased from Avanti Polar Lipids (Alabaster, AL). The sulfhydryl-reactive spin-labeled methanethiosulfonate, MTSL (S-(1-oxy-2,2,5,6-tetramethylpyrroline-3-methyl) methanethiosulfonate), was obtained from Toronto Research Chemicals (Ontario, Canada).

Methods

Expression, labeling with MTSL, purification, and reconstitution of BtuB

The expression, labeling with MTSL, and purification of the BtuB mutants were performed as previously described (31). BtuB was reconstituted by dialysis from mixed micelles of protein and lipid, as detailed elsewhere (31). The reconstituted lipid vesicles were suspended in a buffer containing 10 mM HEPES (N-2-hydroxyethylpiperazine-n′-2-ethanesulfonic acid), 130 mM NaCl, 0.25 mM NaN₃, and 1 μM EDTA, pH 6.5, and were concentrated for EPR spectroscopy using a Beckman Airfuge (Beckman Coulter, Brea, CA).

Electron paramagnetic resonance

EPR spectra were obtained using a Varian E-line 102 series X-band spectrometer (Palo Alto, CA) equipped with a loop-gap resonator (Medical Advances, Milwaukee, WI). Spectra were acquired and analyzed using LABVIEW software that was generously provided by Drs. Christian Altenbach and Wayne Hubbell (University of California, Los Angeles). All spectra were recorded from samples prepared in glass capillary tubes with an 0.8 mm i.d. (VitroCom, Mountain Lakes, NJ) using 2.0 mW of incident microwave power and a 100 kHz modulation amplitude of 1.0 G. Unless otherwise indicated, all spectra were 100-Gauss field sweeps and were recorded at room temperature. For membrane depth measurements, samples were loaded into TPX capillaries and EPR spectra from the spin-labeled BtuB mutants were power-saturated in the presence and absence of a secondary paramagnetic species.

Power saturation was used to obtain a ΔP_1/2 value and to determine a collision parameter (Π) as described previously (32). These parameters provide a relative measure of the collision frequency between R1 and either air (20% O₂) or the paramagnetic nickel chelate Ni(II)EDDA. From the values of ΔP_1/2^O2 or ΔP_1/2^NiEDDA a depth parameter, Φ, can be determined according to

$$\Phi =\text{ln}\left(\Delta {\mathit{P}}_{1/2}^{\text{O}2}/\Delta {\mathit{P}}_{1/2}^{\text{NiE}\mathrm{D}\mathrm{D}\text{A}}\right).$$

The depth parameter, Φ, varies with the immersion depth of a nitroxide within the bilayer, and both spin-labeled lipids and proteins of known structures have been used to empirically calibrate Φ (33,34). For membrane proteins, Φ appears to uniformly vary with membrane depth, and is not significantly influenced by local protein polarity.

As discussed previously (34), the depth parameter, Φ, is found to have the behavior

$$\Phi =A\text{Tanh}\left[B\left(x-C\right)\right]+D,$$ (1)

where x represents the distance of the label from a plane defined by the average position of the lipid phosphorus atoms (positive values of x are inside and negative values are outside the bilayer), A and D set the bulk values of Φ in water and hydrocarbon, C determines the position of the inflection point of the curve, and B determines the slope of the curve. In this case, we used values of A = 3.4, B = 0.11, C = 8.56, and D = 1.7 to convert values of Φ into position. These values were set based upon previous work using spin labels and integral membrane proteins, and peripheral membrane-binding proteins (34,35). The value of D was found to differ slightly from those used previously, because the effective concentration of Ni(II)EDDA is less in POPC than it is in POPC/POPS bilayers.

The EPR depth restraints provided distances from a plane defined by the lipid phosphorus atoms to the nitrogen atoms in 21 R1 labeled sites in BtuB. A simulated annealing protocol using Xplor-NIH, similar to one used previously (36), was used to determine the location of BtuB in a POPC bilayer. A starting structure for the docking was generated by performing an annealing where all the backbone atoms were fixed, side-chain motion was allowed, and all spin-label X₁, X₂ dihedral angles were restrained to −60° using standard dihedral angle restraint parameters. This was done to generate reasonable spin-label side-chain conformations based upon the most likely rotomers for the R1 side chain (37). This initial annealing yielded average X₁, X₂ values of −51° ± 14° and −74° ± 12° for the R1 side chain. The lowest energy structure of 100 structures was then used as the starting structure for further annealing. Two types of atom-to-plane restraints were incorporated into the simulated annealing. In addition to the intrabilayer depths for the MTSL nitrogen atoms to the membrane interface, all periplasmic loop backbone amide hydrogen and carbonyl oxygen atoms were constrained to be no deeper than 8 Å from the average lipid phosphorus position. This distance was chosen based upon previous work indicating that water can penetrate to this depth in the bilayer (38,39). Discovery Studio Visualizer (Accelrys, San Diego, CA) and PyMOL (Schrodinger, New York) were used for molecular visualization and analysis.

Results

EPR spectra from BtuB show a high degree of variability that reflects differences in local structure

Shown in Fig. 2 are X-band EPR spectra for 21 spin-labeled sites that are either positioned on the outward surface of the BtuB barrel (Fig. 2, a–d, corresponding to strands 7, 12, 17, and 3, respectively), or attached to the periplasmic turns of BtuB (Fig. 2 e). Unlike the spectra obtained for R1 on the surface of helical proteins (40), the spectra shown in Fig. 2, a–d, exhibit significant broadening and arise from spin labels that are moderately to highly restricted in their motion. For example, G170R1, which is located near the center of the β-barrel, is near the rigid limit of nitroxide motion, whereas residue L260R1, which is located in the third periplasmic turn, results from an R1 side chain executing a moderate degree of motion. Unlike labels on solvent-exposed surfaces of helical proteins, which do not strongly interact with neighboring residues, spectra on exposed surfaces of β-sheets and β-barrels may be strongly influenced by interactions with hydrogen-bonded and non-hydrogen-bonded neighbors, and by the local twist, splay, and dynamics of the β-strand (14,41).

Figure 2.

X-band EPR spectra for spin labels at 21 solvent-exposed or exterior-facing β-barrel sites for purified BtuB reconstituted into POPC bilayers. EPR spectra for: (a) sites in strand 7, (b) sites in strand 12, (c) sites in strand 17, (d) sites in strand 3, and (e) sites located in the periplasmic turns of BtuB. Spectra are 100-Gauss scans.

Power saturation data suggest that the BtuB barrel is not uniformly solvated by the lipid interface

Each of the 21 EPR spectra in Fig. 2 were power-saturated in the presence and absence of oxygen and Ni(II)EDDA, and the resulting ΔP_1/2 values and corresponding depth parameters, calculated according to Eq. 1, are shown in Table 1. As expected, labels that should be located near the center of the lipid bilayer have the most positive depth parameters, whereas those furthest out on the exterior loops or on the periplasmic turns have the most negative values. There is a substantial uncertainty in the depth parameters for deeply buried sites, because the values of ΔP_1/2 obtained in the presence of Ni(II)EDDA are small at these sites relative to the error in the measurement (see Table 1). From each of the ΔP_1/2 values obtained for the barrel sites (spectra in Fig. 2, a–d), the collision parameters (Π) for both O₂ and Ni(II)EDDA were calculated and are plotted in Fig. 3. The parameter Π is directly proportional to the frequency of collisions between R1 and the secondary paramagnetic reagent (32). Also shown is a boundary that roughly defines the hydrophobic-hydrophilic membrane interface. This boundary is based upon previous work carried out using spin-labeled bacteriorhodopsin and spin-labeled lipids (42). Roughly, when Π^NiEDDA/Π^O2 is greater (or less) than 1, the spin label is outside (or inside) the hydrocarbon region of the bilayer.

Table 1.

Power saturation and depth parameters for R1-labeled sites on BtuB in POPC

Residue	ΔP1/2O2	ΔP1/2NiEDDA	Φ	Depth (Å)∗
β-strand 3
170	4.55	0.36	2.54 ± 1.4	10.9 (+4.4, −3.7)
174	2.32	4.29	−0.62 ± 0.15	1.03 ± 0.7
176	2.7	12.2	−1.51 ± 0.09	−7.57 (+1.1, −2.3)
β-strand 7
265	7.21	0.13	4.02 ± 3.8	16.1 (+3.9, −12.1)
267	5.95	0.29	3.02 ± 1.7	12.3 (+7.7, −4.7)
269	3.96	1.5	0.97 ± 0.34	6.58 (+0.8, −1.0)
271	2.36	3.6	−0.42 ± 0.16	1.92 ± 0.8
273	3.27	10.3	−1.15 ± 0.08	−2.49 ± 0.8
275	3.65	16.1	−1.48 ± 0.06	−6.89 (+1.1, −1.6)
β-strand 12
367	5.72	1.9	1.10 ± 0.27	6.95 ± 0.7
369	5.97	1.95	1.12 ± 0.26	6.99 ± 0.7
371	5.18	1.3	1.38 ± 0.39	7.71 ± 1.1
β-strand 17
483	3.01	4.06	−0.30 ± 0.14	2.43 ± 0.5
485	5.6	11.0	−0.68 ± 0.06	0.687 ± 0.3
487	3.65	19.2	−1.66 ± 0.06	−14.72 (+4.2, −5.3)
Periplasmic turns
162	1.82	8.37	−1.52 ± 0.13	−7.82 (+2.3, −5.2)
260	1.36	1.62	-0.18 ± 0.34	2.91 ± 1.4
349	1.36	0.87	0.45 ± 0.59	5.06 ± 1.6
473	1.26	6.12	−1.59 ± 0.18	−9.91 ± (+4.4, −5.1)
512	1.37	1.43	−0.046 ± 0.38	3.4 ± 1.4
556	1.47	0.71	0.73 ± 0.72	5.9 ± 2.2

Individual values of ΔP_1/2^O2 and ΔP_1/2^NiEDDA are based upon one-to-three measurements and have errors of ∼0.2 mW and 0.5 mW, respectively. Errors in Φ are determined by standard error propagation in the calculation of Φ from the P_1/2 values. Large uncertainties in several Φ-values are due to small changes in P_1/2^NiEDDA relative to the error in the measurement.

^∗The variation in the average depth (upper and lower range) is based upon the uncertainty in Φ and Eq. 1.

Figure 3.

Collision parameters for oxygen and Ni(II) EDDA obtained for the 15 labeled sites on the surface of the BtuB β-barrel. (Dashed line) Approximate position of the aqueous/lipid boundary defined by these collision parameters.

From Fig. 3, the six labels in strand 7 (residues 265, 267, 269, 271, 273, and 275) appear to be equally divided between aqueous and lipid exposure, which implies that the lipid interface intersects the strand in a manner as depicted in Fig. 1 a. However, there are indications in these data that the interface is not uniform around the protein. For example, three labels in strands 12 (residues 367, 369, and 371) and 17 (residues 483, 485, and 487) are located at roughly the same position along the length of the BtuB barrel; nonetheless, the three labels in strand 12 all have lipid exposure, whereas the three labels in strand 17 have an aqueous exposure.

Simulated annealing produces a view of the asymmetric lipid interface surrounding BtuB

The data in Table 1 and Fig. 3 suggest that the lipid interface around the protein might not be uniform, and we used the simulated annealing software package Xplor-NIH to dock BtuB to two planes. The intersection of these two planes with the protein surface represents the location of the lipid phosphorus atoms on each bilayer surface adjacent to the protein. The docking routine was similar to one used previously for synaptotagmin 1 bound to the membrane surface (see Methods) (36). For the depth parameters shown in Table 1, a distance range between the nitrogen on R1 and the average position of the lipid phosphorus atoms was determined based upon the error in Φ and Eq. 1 (the restraints used are listed in Table S1 in the Supporting Material). Initially the distances to the periplasmic turns were used as restraints for the periplasmic interface and the other restraints with the exception of 170, 265, 267, and 269 were assigned to the extracellular interface. All restraints used in the initial annealing calculations involved residues that were clearly closer to one bilayer surface than the other. The exceptions were initially excluded because they were close to the center of the bilayer. The initial calculations indicated that 265 was closer to the periplasmic surface and 170, 267, and 269 were closer to the external surface. The annealing was repeated with the preceding four residues assigned to be closest to the bilayer surface indicated by the initial annealing (265 closest to the periplasmic surface; 170, 267, 269 closest to the external surface).

Periplasmic loop backbone atoms capable of hydrogen bonding (carbonyl oxygen and amide hydrogen) should not be located in the bilayer hydrocarbon region and we incorporated atom-plane restraints to keep all periplasmic loop carbonyl oxygens and amide hydrogens in a region where water was available for hydrogen bonding (see Methods). The addition of these restraints made a small change in the location of the periplasmic bilayer phosphorus plane, moving it 1 Å further from the bilayer center. The effect of side-chain flexibility during the final part of the simulated annealing was also explored and three conditions were tested:

• 1.All side chains were rigid throughout the annealing.
• 2.All side chains were unrestrained during the final part of the annealing except that X₁, X₂ were fixed near −60°.
• 3.All side chains were unrestrained during the final part of the annealing.

No significant difference was seen among these three conditions, and the final condition was used for all results presented here.

Separate simulated annealing calculations were done to obtain external and periplasmic bilayer phosphorus atom plane orientations and 100 orientations were calculated for each phosphorus atom plane. The agreement between the measured and calculated distances for the lowest energy fit is shown in Fig. 4, and indicates that there is good agreement between the measurement and fit when the interface is defined by a plane intersecting the protein barrel. The fit for the lowest energy structure docked to two planes is shown in Fig. 5. In this model, we measured the shortest distance from the barrel center to the plane and the angle between the barrel long axis and the plane to assess variation among the calculated orientations. Here the barrel is defined by the eight residues that are closest to the periplasmic end of all of the 22-barrel β-strands (because the shortest strand is eight residues).

Figure 4.

Comparison of distances between spin-label nitrogen atoms and bilayer phosphorus planes obtained experimentally from the EPR power saturation technique and calculated via simulated annealing. The error bars for the experimentally determined distances are based upon the uncertainty in values of Φ given in Table 1 and Eq. 1.

Figure 5.

Location of BtuB relative to POPC bilayer phosphorus planes. Phosphorus planes (gray with white borders) were determined by using atom-to-plane restraints in simulated annealing calculations. Strands 19 and 8 (magenta) are indicated (left and right), respectively. Also shown is the location of planes (green) defined by the backbone atoms of the external facing W and Y residues near the bilayer aqueous-hydrocarbon interfaces. All external W are included, but only external Y below the external (top) phosphorus plane are included.

There was little difference among the 10 lowest energy orientations for each plane (see Table 2). The planes defining the position of the membrane interfaces are not parallel, and are clearly further apart on one side of the BtuB barrel than the other. The distances between phosphorus planes measured from several regions of the barrel are listed in Table 3. The transmembrane distance between phosphorus atom planes is shortest near the point where strand 19 crosses the bilayer, and is ∼27 Å. The furthest separation is found near strand 8, and is ∼38 Å. Previous work using x-ray diffraction indicates that the separation between POPC transbilayer phosphorus planes is 38 Å (24), which is almost a perfect fit for the longer side of the barrel. However, the 27 Å separation on the shorter side indicates that the lipid at this protein interface is highly distorted. To accommodate the hydrophobic mismatch on this surface, there must either be a larger fraction of gauche conformers and/or some lipid interdigitation.

Table 2.

Uncertainties in bilayer phosphorus plane positions

	Distance from barrel center to bilayer plane(Å)	Angle between barrel axis and bilayer plane
External bilayer plane	18.4 ± 0.6	80.6° ± 0.6
Periplasmic bilayer plane	14.4 ± 0.3	86.0° ± 0.4

The uncertainty in the position of the phosphorus planes is based upon the standard deviation of distances measured for the 10 lowest energy orientations obtained from the simulated annealing. The barrel center and axis is defined using the first eight residues on each strand closest to the periplasmic interface.

Table 3.

Position of planes defined by the lipid phosphorus atoms and BtuB tryptophan/tyrosine (WY) side chains

Distances between planes (Ångstroms)
	External P to periplasmic P	External WY to periplasmic WY	External P to external WY	Periplasmic P to periplasmic WY
Barrel center	32.7	17.5	6.8	8.4
Near strand 8	27	24
Near stand 19	38	20
Angle between planes (degrees)	14.7	12.7	1.2	1.3

Distances near stands 8 and 19 were measured at the barrel surface along a line parallel to the BtuB axis where the phosphorus atom planes are at maximum and minimum separations, respectively.

Discussion

Membrane proteins are generally believed to be matched to bilayers of a defined thickness (43); however, cases of apparent mismatch in native membranes have been observed (44), suggesting that mismatch may be a normal condition for at least some membrane proteins. In this work, a model has been generated for the lipid interface surrounding the outer membrane vitamin B₁₂ transporter BtuB. In BtuB, the highly asymmetric length of the barrel β-strands suggests that the lipid interface may not be symmetric around the protein, and the data presented provide evidence for this asymmetry. The results of simulated annealing using point-to-plane depth restraints indicate that the average phosphorus-to-phosphorus distance across a POPC bilayer differs by as much as 11 Å around the circumference of the protein. Furthermore, the difference in membrane thickness surrounding the protein, and the tilt in the planes defining the lipid interface, roughly follow the asymmetry in the structure of BtuB. The results also suggest that BtuB may be mismatched to the hydrophobic thickness of its native membrane.

Tryptophan (W) and tyrosine (Y) residues on membrane protein surfaces are typically positioned near the membrane water-hydrocarbon interface (45). For BtuB, the extracellular and periplasmic planes defined by the average backbone atom positions of these aromatic residues were determined and are shown in Fig. 5 (green planes). These WY planes are nearly parallel to, but displaced from, the experimentally determined phospholipid phosphorus planes (see Table 3). The displacement is roughly 7–8 Å and is consistent with the structure of a POPC bilayer, where a 4 Å distance has been reported for the separation between phosphorus and carbonyl-glycerol planes (which should lie near the WY planes) (24). Thus, the asymmetry observed here in the POPC interface surrounding BtuB is roughly consistent with the placement of aromatic residues around the protein.

In pure POPC bilayers, the transbilayer separation between phosphorus planes is 37.6 Å and the bilayer hydrocarbon thickness is 27.1 Å, ∼10 Å thinner (24). However, BtuB presents a hydrophobic surface which is, on average, shorter than that of pure POPC. For the experimentally determined phosphorus planes shown in Fig. 5, the distance between planes at the barrel center is 32.7 Å (or ∼5 Å shorter than that of pure POPC). If we assume that the bilayer hydrophobic thickness is ∼10 Å shorter than this distance, the result is shorter than that for pure POPC, but in reasonable agreement with estimates of the hydrophobic thickness of BtuB, which ranges from 18 to 23 Å based on crystal structures and aqueous-to-hydrocarbon transfer energies (46,47). As shown in Fig. 5, transbilayer lipid distances vary across the protein. The presence of BtuB decreases the thickness of a pure POPC bilayer by 5 Å near strands 2 and 13, which are locations where the transbilayer distance between phosphorus planes is equivalent to that measured at the β-barrel center. While there is little or no protein-POPC hydrophobic mismatch near strand 8 on the long end of the barrel, there is a 10 Å mismatch with POPC near strand 19. Thus, much of the protein surface appears to be mismatched to the thickness of a POPC bilayer.

Coarse-grained molecular-dynamics simulations have been used to assemble bilayers around transmembrane proteins and to produce models for the structure of the lipid-protein interface (23,48). This coarse-grained approach was used to generate a database containing models of the lipid interface for a number of membrane proteins, including two high-resolution apo-BtuB crystal structures (48). One of these crystal structures was obtained in detergent (in surfo) and a second obtained in a lipidic phase (in meso) (49,50). The in meso structure yields a result for the lipid interface that is similar to the experimental result presented here: the bilayer lipid thickness immediately adjacent to in meso BtuB varies around the protein circumference and follows the asymmetry in β-sheet lengths around the β-barrel (see Fig. S2 A). However, the in surfo structure does not yield the same bilayer thickness asymmetry that is observed with the in meso structure, even though these two crystal structures have very similar β-strand structures in the membrane-embedded regions of the barrel (see Fig. S2 B). The differences in these two models might result from different side-chain configurations that are seen between the two crystal structures, but it is not immediately clear which structure (if either) represents the more appropriate BtuB bilayer structure.

The results presented here represent an experimental demonstration of differences in bilayer thickness around an integral membrane protein. Although the implications of this mismatch for BtuB structure and function are not presently known, previous work suggests a few possibilities.

First, a hydrophobic mismatch may lead to lipid sequestration. A study of the β-barrel protein, OmpF, reconstituted into PC bilayers with different acyl chain lengths, found that PC that matched the OmpF hydrophobic thickness bound most strongly to OmpF and that binding affinity decreased as the lipid-protein hydrophobic mismatch increased (51). In the Escherichia coli outer membrane, the external leaflet is primarily lipopolysaccharide, which has primarily 12-and 14-carbon acyl chains but also has some 16-carbon chains. The internal leaflet is composed largely of phosphatidylethanolamine and phosphatidylglycerol with 16- and 18-carbon acyl chains but also some 14-carbon chains (26). As a result, BtuB might sequester and segregate lipids around its circumference based on chain length to optimize lipid-protein hydrophobic matching.

Second, the unfavorable energetics associated with a hydrophobic mismatch might drive the system to minimize protein contact with the lipid bilayer (5), and in the process promote protein-protein association that is functionally important. BtuB appears to interact with OmpF, which together act as a receptor for certain colicins (52), and BtuB has been observed to copurify with OmpF (G. E. Fanucci and D. S. Cafiso, unpublished). These interactions may be driven by hydrophobic mismatch.

Finally, hydrophobic mismatch may help align and order BtuB in the outer membrane. Vibrational spectroscopy has shown that when bilayer hydrophobic thickness exceeds β-barrel hydrophobic length, the average alignment of the protein relative to the bilayer normal (or order) increases (10,12). Moreover, protein order decreases as membrane thickness becomes less than the protein hydrophobic thickness (13). The hydrophobic thickness of the bacterial outer membrane bilayer is likely larger than that of BtuB (21,25,26,53), and this might help order the protein and promote interactions that are functional, such as those with the transperiplasmic protein TonB. It is unlikely that the observed hydrophobic mismatch will cause a substantial change in the BtuB barrel backbone structure due to the considerable energy required for this type of change (5,17).

The coupling of protein function to bilayer physical properties has been demonstrated for many integral membrane proteins (see, for example, (54)). In the same way that membrane charge density and bilayer curvature strain energy are normal properties of native membranes (55,56), hydrophobic mismatch appears to be a natural feature of some membrane proteins (44). The work carried out here indicates that BtuB is likely to be hydrophobically mismatched even in its native membrane. In part, this is due to the shape of the BtuB barrel, which is a highly conserved feature of TonB-dependent transporters. This suggests that the asymmetric shape of the TonB-dependent transporter barrel may have some functional importance. Eukaryotic cell membranes are known to be laterally heterogeneous, having regions of different composition sometimes referred to as rafts. Recent work indicates that the bacterial inner membrane is also laterally heterogeneous (57). Lipid sequestration and protein-protein association, driven by hydrophobic mismatch, might be important forces in establishing a lateral organization and heterogeneity to the bacterial outer membrane.