Structural Asymmetry in a Trimeric Na+/Betaine Symporter, BetP, from Corynebacterium glutamicum

doi:10.1016/j.jmb.2011.01.028 J. Mol. Biol. (2011) 407, 368–381

Contents lists available at www.sciencedirect.com

Journal of Molecular Biologyj ourna l homepage: ht tp : / /ees .e lsev ie r.com. jmb

Structural Asymmetry in a Trimeric Na+/BetaineSymporter, BetP, from Corynebacterium glutamicum

Ching-Ju Tsai1, Kamil Khafizov2, Jonna Hakulinen1, Lucy R. Forrest2,Reinhard Krämer3, Werner Kühlbrandt1 and Christine Ziegler1⁎1Department of Structural Biology, Max Planck Institute of Biophysics, 60438 Frankfurt am Main, Germany2Computational Structural Biology Group, Max Planck Institute of Biophysics, 60438 Frankfurt am Main, Germany3Institut of Biochemistry, University of Cologne, Zülpicher Str. 47, 50674 Köln, Germany

Received 10 June 2010;received in revised form10 December 2010;accepted 12 January 2011Available online31 January 2011

Edited by W. Baumeister

Keywords:automated rigid-body fitting;electron crystallography;membrane protein structure;Na+-coupled transport;osmotic stress

*Corresponding author. E-mail [email protected] address: C. -J. Tsai, Camb

Medical Research, Cambridge CB2 0Abbreviations used: 3D, three-dim

two-dimensional; NCS, noncrystalloEM, electron microscopy; LPR, lipidtransmembrane; L-CC, Laplacian-filcoefficient; CC, cross-correlation coeData Bank.

0022-2836/$ - see front matter © 2011 P

The Na+-coupled symporter BetP catalyzes the uptake of the compatiblesolute betaine in the soil bacterium Corynebacterium glutamicum. BetP alsosenses hyperosmotic stress and regulates its own activity in response tostress level. We determined a three-dimensional (3D) map (at 8 Å in-planeresolution) of a constitutively active mutant of BetP in a C. glutamicummembrane environment by electron cryomicroscopy of two-dimensionalcrystals. The map shows that the constitutively active mutant, which lacksthe C-terminal domain involved in osmosensing, is trimeric like wild-typeBetP. Recently, we reported the X-ray crystal structure of BetP at 3.35 Å, inwhich all three protomers displayed a substrate-occluded state. Rigid-bodyfitting of this trimeric structure to the 3Dmap identified the periplasmic andcytoplasmic sides of the membrane. Fitting of an X-ray monomer to theindividual protomer maps allowed assignment of transmembrane helicesand of the substrate pathway, and revealed differences in trimerarchitecture from the X-ray structure in the tilt angle of each protomerwith respect to the membrane. The three protomer maps showedpronounced differences around the substrate pathway, suggesting threedifferent conformations within the same trimer. Two of those protomermaps closely match those of the atomic structures of the outward-facing andinward-facing states of the hydantoin transporter Mhp1, suggesting that theBetP protomer conformations reflect key states of the transport cycle. Thus,the asymmetry in the two-dimensional maps may reflect cooperativity ofconformational changes within the BetP trimer, which potentially increasesthe rate of glycine betaine uptake.

© 2011 Published by Elsevier Ltd.

ress:

ridge Institute forXY, UK.ensional; 2D,graphic symmetry;/protein ratio; TM,tered correlationfficient; PDB, Protein

ublished by Elsevier Ltd.

Introduction

Under hyperosmotic stress, bacteria accumulate upto molar concentrations of organic molecules, such asbetaine, in order to prevent dehydration of thecytoplasm. Since these so-called osmolytes, or “com-patible” solutes, do not perturb protein function, suchhigh concentrations are not detrimental to cellularfunction.1 For a rapid response to changes in osmoticpressure, specific tightly regulated membrane

369Three-dimensional BetP Structure to 8 Å by EM

transport proteins are required to facilitate the uptakeor extrusion of these compatible solutes.BetP is the main betaine transporter in Corynebac-

terium glutamicum that belongs to the betaine/carnitine/choline transporter family.2–4 BetPrequires the cotransport of two Na+ per betainemolecule and can achieve extremely high substrateconcentration gradients across the membrane ofmore than 6 orders of magnitude.2 The cytoplasmic∼50-residue-long N-terminal and C-terminaldomains of BetP are highly charged.5 Truncationsat the N-terminal domain of more than 10 residuesincrease the osmolarity required for activation,while truncation of the C-terminal domain bymore than 23 residues results in constitutiveactivation, albeit at a reduced level of activity.6

The role of the C-terminal domain in activationincludes detection of the cytoplasmic K+ concentra-tion, which acts as a measure of hyperosmoticstress.6 An important factor here may be that the N-terminal and C-terminal domains interact both withone another and with cytoplasmic loops.7 Inaddition, BetP was shown to respond to stimulioriginating directly from the membrane, such aschanges in lipid composition.8 In the entirelynegatively charged lipid environment of C. glutami-cum, BetP shows a shift to higher osmolarityrequired for activation compared to activation inEscherichia coli, which contains only ∼30% negative-ly charged lipids.8

A projection map of BetP from two-dimensional(2D) crystals grown in the presence of E. coli polarlipids and cardiolipin revealed a trimeric architec-ture of BetP in the membrane. However, significantdifferences between the three protomers indicatedthat the trimer does not have exact 3-fold non-crystallographic symmetry (NCS).9 We recentlysolved the structure of a regulated N-terminallytruncated form of BetP (BetPΔN29) to 3.35 Åresolution by X-ray crystallography.10 In thesethree-dimensional (3D) crystals, the BetP trimeralso showed a break in 3-fold NCS around theosmosensing C-terminal domains, which formextended α-helices and are differently oriented ineach protomer. To separate the roles of differentconformational states within the trimer in regulationand transport, we have determined the 3D structureof a C-terminally truncated, constitutively activemutant of BetP (BetPΔC4511) at 8 Å in-planeresolution and 16 Å resolution perpendicular tothe membrane by electron cryomicroscopy of 2Dcrystals. The 3Dmap clearly shows that BetPΔC45 isan asymmetric trimer in the membrane, like thewild-type protein. Rigid-body fitting of individualprotomers in the electron microscopy (EM) maprevealed an altered trimer architecture in themembrane compared to that found in 3D crystals.Furthermore, the conformations of the three proto-mers differ significantly from one another. By

comparison with structural data obtained fortransporters sharing the same fold as BetP, at leasttwo of the three distinct protomer states within theBetPΔC45 trimer could be assigned to differentconformational states involved in the transport cycleof BetP.

Results and Discussion

BetP is functional after 2D crystallization

A deletion mutant of BetP truncated by 45residues in its C-terminal domain (BetPΔC45) wasused for the structure determination of membrane-reconstituted BetP by electron cryomicroscopy of 2Dcrystals. The C-terminal truncation in BetPΔC45results in the deregulation of BetP in cells orproteoliposomes,6 rendering it constitutively activefor betaine uptake even in the absence of osmoticstress, although its maximum activity is 25% of theoptimal wild-type activity.6 Two-dimensional crys-tals were obtained by detergent dialysis9,12 both inthe presence of E. coli lipid/cardiolipin and in nativeC. glutamicum lipids (Fig. 1a), respectively. Incuba-tion of the 2D crystals with the substrate at betaineconcentrations higher than 10 mM led to distortionsin the 2D crystal lattice; therefore, crystals weregrown in the absence of substrate betaine. Toconfirm that BetPΔC45 retains its functionalityafter 2D crystallization, we measured its [14C]betaine uptake activity.13 Closed proteoliposomeswere obtained by fusion of the 2D crystal sheetsgrown in C. glutamicum lipids with preformed E. coliliposomes. BetPΔC45 transports betaine in fused 2Dcrystals although the rate of transport by BetPΔC45under these conditions is ∼15% less than that byBetPΔC45 reconstituted in proteoliposomes (Fig. 1b)at the same lipid/protein ratio (LPR). This deviationlikely originates from subtle differences in the LPRof the two systems, since, for example, the vesiclesmay become leaky when the protein content is high.Indeed, uptake rates diminish with decreasing LPRin both proteoliposomes and fused 2D crystals (Fig.1b). Thus, in spite of these small differences inmaximal turnover, BetPΔC45 appears to be fullyfunctional after 2D crystallization. We thereforeproceeded to determine the structure of BetΔDC45by testing two different lipid conditions, namelyeither an E. coli lipid/cardiolipin mixture or a C.glutamicum lipid extract.

Two-dimensional crystals of BetPΔC45 grown inan E. coli lipid/cardiolipin mixture

Well-ordered 2D crystals of BetPΔC45 occurredvery rarely (in b1% of crystallization trials) in thepresence of a 58:42 mixture of E. coli lipid/

Fig. 2. Projection structure to 7.5 Å resolution of the 2Dcrystals of BetPΔC45 crystallized in the presence of a 58:42mixture of E. coli lipid/cardiolipin. The map reveals fourtrimers of BetP in a rectangular unit cell of 186.6 Å×167.0Å.The trimers (orange broken outline) are related by P22121symmetry. Protomers 1 and 2 have similar conformations,while protomer 3 appears to be in a different state. The redarrow indicates additional crystal contacts (red densities)between two protomers in symmetry-related trimers,adjacent to TM1 (blue densities) from each protomer.

Fig. 1. Functionality of BetPΔC45 after 2D crystalliza-tion. (a) Electron micrograph of BetPΔC45 2D crystalsheets stained with 1% uranyl acetate. (b) Dependence ofbetaine uptake rate by BetPΔC45 on protein concentrationin proteoliposomes. Uptake rates at 0.64 Osm/kg areshown for three different LPRs (from 30:1 to 10:1) in 2Dcrystals (gray bars) and proteoliposomes (black bars).Error bars represent the standard deviation from three tofive individual measurements. Inset: Micrograph offreeze-fractured 2D crystals of BetPΔC45 after fusion topreformed E. coli lipid liposomes at a final LPR of 20:1.

Table 1. Electron crystallographic data for BetPΔC45 in2D crystals

Plane group p121_bCell dimensions

a, b, c (Å) 92.1, 155.2, 150α, β, γ (°) 90, 90, 90

Number of imagesa 72Number of structure factors (IQ1–IQ4) 2677Resolution limit for merging (Å) 8.0Effective resolution of 3D set (Å)b

In-plane 8.0Perpendicular to the membranec 16.0

Completeness (%)0–50° 870–90° 73

Overall weighted R-factor (%) 29.4Overall weighted phase residuals (°)d 17.9

a Nineteen at 0°, 5 at 10°, 22 at 20°, 14 at 30°, and 19 at 40–60°(nominal tilts).

b As calculated from a point-spread function of theexperimental data.

c As calculated by the point-spread function.d From the program LATLINEK.

370 Three-dimensional BetP Structure to 8 Å by EM

cardiolipin and under conditions that were previ-ously successful for the 2D crystallization of wild-type BetP.9 The trimers (orange broken line) arerelated by P22121 symmetry, and a projection mapcould be calculated by merging three independentlattices to 7.5 Å resolution (Fig. 2). The map revealsfour trimers of BetP in a rectangular unit cell of186.6 Å×167.0 Å. These trimers resemble the wild-type BetP trimer reported previously.9 Therefore,we can conclude that the asymmetric trimerpreviously observed for the regulated wild-typeprotein is also observed for the deregulatedBetPΔC45 under similar conditions. However, in

this crystal of BetPΔC45, an additional density (Fig.2, in red, red arrow) appears in the projection mapbetween protomers 1 of neighboring trimers, whichwas not observed in crystals of the wild-typeprotein. Unfortunately, the yield of well-diffracting2D crystals in E. coli lipid and cardiolipin was notsufficiently high for 3D data collection.

Fig. 3. Data collection on 2Dcrystals by electron crystallogra-phy. (a) Plot of combined phaseerror to 7.5 Å merging 72 lattices.The number and the size of theboxes correspond to the phase errorafter averaging and rounding to 0°and 180° for each measurement (1,b8°; 2, b14°; 3, b20°; 4, b30°, where45° is random). Values over 30° areindicated only with decreasing boxsize. The resolution limit was set to7.5 Å. (b) Lattice line data shown asplots of amplitudes (bottom) andphases (top) along the z⁎-axis forthe (4,1) and (3,5) reflections. Thelattice lines were produced byweighted least-squares fitting, andresulting errors are shown.

371Three-dimensional BetP Structure to 8 Å by EM

Three-dimensional map of BetP inC. glutamicum lipids

The 2D crystals of BetPΔC45 grown in thepresence of primarily negatively charged C. gluta-micum lipids reproducibly yielded images withstructure factors to 8 Å and thus were used tocalculate a 3D density map. The crystals belong tothe p121-b symmetry plane group, with unit celldimensions of 90.5 Å×152.1 Å and γ=90.8°.Seventy-two images of crystalline areas, obtainedfrom specimens tilted by up to 50° under an electron

Fig. 4. Three-dimensional map of BetPΔC45 at 8 Å×16 Å rthree protomers of one trimer are labeled as protomers 1, 2, anside view in (b). Three-dimensional densities are contoured aindicated by the broken line pictured in (a) of the three protom2, and 3 contoured at 1.3σ from the top of the membrane. The pThe overall shape of protomer 1 is outlined by the broken linecomparison. A red circle indicates the position of the putative sa monomeric X-ray structure of BetPΔN29 (PDB entry 2WIT),repeat 2 is shown in green, and the peripheral helices TM1 anthe domains to which they belong (i.e., where B signifies the buconnecting B and H). Thus, TM3–TM4 is labeled as B1–B2; TMTM10–TM11 is labeled as H3–H4; and TM7 and TM12 are labelcomputed to 8 Å in all three directions, which is the reason thwith the membrane plane.

microscope, were processed and merged to 8 Åresolution with an overall phase residual of 17.9°(Table 1). The phases and amplitudes from fittedlattice lines (Fig. 3a and b) were combined tocalculate the 3D map at an in-plane resolution of8 Å (Fig. 4a and b). The resolution perpendicular tothe membrane plane (assigned to the z-axis) wascalculated to be 16 Å from the point-spread functionof the experimental data. This reduced resolution inthe z-direction reflects the cone of missing data thatbecomes prominent for images at high-tilt angles. Toreduce the noise in the z-direction originating from

esolution. (a) View of one unit cell (92.5 Å×155.2 Å). Thed 3. The broken line indicates the sectioning plane for thet 1.3σ. (b) Side view of a cross section through the mapers. (c) View of the density of individual BetP protomers 1,rotomers are aligned in the same orientation as protomer 1., which has been superimposed on protomers 2 and 3 forubstrate pathway. (d) An 8-Å density map constructed forsuperimposed on the structure. Repeat 1 is shown in red,d TM2 are shown in gray. Helices are labeled according tondle, H signifies the hash domain, and A signifies the arms8–TM9 is labeled as B3–B4; TM5–TM6 is labeled as H1–H2;ed as A1 and A2, respectively. Note that the density here isat densities can be observed for the helices that lie parallel

Fig.

4(legend

onprevious

page)

372 Three-dimensional BetP Structure to 8 Å by EM

373Three-dimensional BetP Structure to 8 Å by EM

overweighting, we scaled down the amplitudes byapplying a B-factor of −300 Å2. At a cutoff of 1.3σ,the side view of the resulting map showed a single-layered density with a moderate amount of noiseperpendicular to the membrane (Fig. 4b).The three protomers (labeled 1, 2, and 3) are of

similar overall size and shape (Fig. 4c, blue outline)and resemble the X-ray structure of a BetP monomerwhen displayed at a comparable resolution (i.e., 8 Åin all three dimensions) (Fig. 4d). A pore of ∼15 Ådiameter is clearly visible in the center of each EMprotomer, as well as in the X-ray density (Fig. 4c andd, red circle). At 3.35 Å resolution, this region isoccupied by bulky aromatic side chains and byresidues located in extramembranous loops,10 nei-ther of which can be resolved at 8 Å.Many of the densities of the EM protomer maps

appear to be difficult to reconcile with canonicaltransmembrane (TM) helices, which might at firstglance appear to reflect poor-quality images or data.However, the low phase residual of 17.9° obtainedfor the relatively large number of images usedargues against this (Table 1). In fact, these featuresreflect the complex overall fold of the protein andthe arrangement of the helices therein. The BetP foldcomprises a repeat of closely intertwined structur-ally related segments of five TM helices each (in Fig.4d, the first repeat is in red, and the second repeat isin green). BetP shares this fold with transporters offive sequence-unrelated transporter families.10 Thetwo TM5 repeats are related by a 2-fold pseudo-symmetry axis running parallel with the membraneplane, such that they have opposite topologies withrespect to the membrane. In the context of thestructure, the first two helices of each repeat form afour-helix bundle (Fig. 4d, B1–B4), while an outerscaffold of helices surrounds the bundle. Within thescaffold, the third and fourth helices of each repeatform V-shaped elements that together create the so-called hash domain (Fig. 4d, H1–H4), and the lastflexible helix of each repeat is assigned as arms (Fig.4d, A1–A2). Such close intertwining of the helicescauses their densities to merge, as is clearly seen forthe four-helix bundle in the X-ray structure (Fig. 4d).Although such noncanonical densities are thereforeexpected for this fold and at this resolution, theirassignment to specific helices is nevertheless achallenge; thus, to reduce the likelihood of mis-assignment, we carried out an automated fitting ofthe atomic X-ray structure to the EM density map.

Automated fitting of the trimeric X-ray structureto the EM map

Automated rigid-body fitting was performedusing Situs14,15 in order to model the symmetric X-ray trimer into the EM density of the BetPΔC45trimer. This involved an exhaustive six-dimensionalsearch with ∼1010 attempted fits. To achieve better

discrimination of correct fits between experimentalmaps and calculated maps, we took contourinformation into account using Laplacian filtering.14

The Laplacian-filtered correlation coefficient (L-CC)has a theoretical range of 0–1, although the absolutevalues tend to be significantly lower than standardcross-correlation coefficients (CCs). Nevertheless,seemingly small differences in L-CC can allowbetter discrimination of the fits than is possiblewith the standard CC.14 For the trimer, the sevenhighest-ranking fits deviate from one another onlyby small differences in tilt angle with respect to themembrane plane (the average difference betweenthem is 3.9±2.3°) and, in all cases, the X-raystructure overlays well with the EM density. Thefit with the highest rank is shown in Fig. 5a and b.The L-CC scores of these top seven fits were between0.090 and 0.085, and the corresponding CCs were0.619–0.612. Fits in which the trimer is inverted withrespect to the membrane plane were ranked lower(L-CCb0.080 and CCb0.590) and therefore exclud-ed. Thus, we assign the top view of the 3D map ofBetPΔC45 shown in Figs. 4 and 5 as from theperiplasm.

Automated rigid-body fitting of the X-raystructure to protomers in the EM map

In none of the high-ranking trimer fits did themore perpendicular helices fit the map equally wellin all three protomers simultaneously (e.g., whileprotomer 1 fits the density nicely, protomer 3protrudes out of the membrane) (Fig. 5b). Toaccount for those differences in protomer fits, wecarried out automated rigid-body fitting of the X-rayprotomer structure to each of the three densities inthe 3D EM map. Automated fitting to the protomermaps was more challenging than for the wholetrimer due to the reduced number of constraints. Inparticular, there was some ambiguity in the verticalpositioning of these fits due to the limited resolutionin this direction (see Materials and Methods).Nevertheless, for each protomer, we obtained onehigh-ranking fit (i.e., within the top 10) that wasconsistent with the general orientation in the trimerfit.The perpendicular helices in these protomer fits

matched the regions of strong density better than inthe trimer fits (Fig. 5c and d). For example, withoutfurther adjustment, peripheral helices TM1, TM6,and TM11 matched reasonably well with a set ofthree rod-shaped densities at the outermost cornerof each protomer (Fig. 5c and d). The intensity ofthese densities is consistent with the perpendicularorientation of the peripheral helices in the X-raystructure with respect to the membrane plane. Basedon these assignments, the central pore within eachprotomer of the 3D EM map (Fig. 5c, red circle inprotomer 1) does indeed correspond to the substrate

Fig. 5. Automated rigid-body fittings of the X-ray structure of BetPΔN29 to the 3D EM map of BetPΔC45. The EMdensity (blue) is contoured at 1.5σ. (a) Fit of the trimeric X-ray structure to the density of a trimer in the 3D EM map(viewed from the top). (b) View of protomer 1 (blue) and protomer 3 (green) from the plane of the membrane with theperiplasm towards the top; the approximate membrane boundaries are indicated by parallel lines. Protomer 2 is shown inorange. (c) Fits of the protomeric X-ray structure of BetP to the density of individual EM protomer maps (viewed from thetop) and (d) from within the plane of the membrane (with the periplasm towards the top). Helices in protomers 1–3 arecolor coded, and individual helices are numbered. The position of the substrate pathway is indicated with a red circle inprotomer 1. Black arrows in the side views of (b) and (d) highlight the change in distance between the protomers on thecytoplasmic side due to a change in the tilt angle of the protomers with respect to the membrane plane.

374 Three-dimensional BetP Structure to 8 Å by EM

pathway10 (Fig. 4c, red circle), which is lined byTM3, TM4, TM5, TM8, and TM10 in the X-raystructure. In addition, the strong off-center densityadjacent to the central pore is occupied by the four-helix bundle comprising TM3, TM4, TM8, and TM9;these helices are also oriented perpendicular to themembrane plane in the X-ray structure and thereforeshow up strongly in the EMmap. Helices in the BetPX-ray structure—such as TM2, TM7, and theamphipathic helix H7 that run parallel with oradopt a shallow angle with the membrane plane—were only partly resolved in the 3D EM map, asmentioned previously. Finally, the densities for thetilted helices TM5 and TM10 (Fig. 5c), which connect

the hash domain to the four-helix bundle, are onlypartially resolved for similar reasons.The improvement from trimer fit to protomer fit

reflects differences in tilt (10–15° for protomers 1 and2, and ∼5° for protomer 3) relative to the trimer fit(Fig. 5b and d, black arrow). Nevertheless, signifi-cant differences between the three protomers arestill clearly visible in, for example, the strongdensities adjacent to the central pore, and thesedifferences are observable at a wide range of σ levels(Fig. 6a).To test the extent of similarity between the

protomer maps beyond their differences in tiltangle, we used the fitted protomer models to

Fig. 6. Analysis of differences in densities in the three protomer maps. (a) Top view of a BetP trimer displayed at sixdifferent σ levels (1.4, 2, 2.5, 3, 3.5, and 4) showing that the differences between individual protomers are not caused bynoise in the map. (b) Top view of the averaged EMmodel. Important features in the off-center density and at the perimeterare averaged out in comparison to the original EMmap (left). Averaging could only be performed based on the individualprotomer fits, as this allowed us to compensate for the different tilt angles with respect to the membrane plane.

Table 2. L-CCs of atomic structures fitted to the protomermaps of BetPΔC45

Maps

Models

X-ray Protomer 1a Protomer 2 Protomer 3

Protomer 1 0.066b 0.092 (0.090)c 0.086 0.078Protomer 2 0.125 0.104 (0.119) 0.142 0.129Protomer 3 0.072 0.081 (0.079) 0.072 0.093

a For protomer 1, the scores of a second model are also shown(in parentheses).

b The fit of the X-ray structure for this map was adjusted in theaxis normal to the membrane (see Materials and Methods).

c The L-CC for the highest-scoring model for each map isshown in boldface. For reference, the standard correlationcoefficients (without Laplacian filter) for these fits are in therange of 0.43–0.49. Note that correlation coefficients for differentmaps cannot be compared due to differences in the intensities ofdensity.

375Three-dimensional BetP Structure to 8 Å by EM

average the EM density, resulting in a symmetrictrimer (Fig. 6b). This process reinforced the densitiesof the scaffold helices TM5 and TM10, but weakenedthe strong off-center densities and the densities onthe outer rim of the trimer, indicating that theprotomers of BetPΔC45 are at least similar in thecentral region of each protomer and in the proto-mer–protomer interface. This result is consistentwith the observation that the most pronounceddifferences between protomer maps are around thecentral pore and in the adjacent four-helix bundle(Fig. 4c and d). Therefore, we conclude that theasymmetric trimer of BetP reflects different confor-mational states of BetP.

Optimization of BetP models to the threeprotomer maps

To qualitatively assess the extent to which thethree protomers differ, we adapted the models ofBetP to make them more consistent with thedensities in each protomer. For this, we used anautomated flexible-fitting method involving rigid-body sampling of individual helix segments, fol-lowed by L-CC scoring (Materials and Methods).We focused our efforts on TM5 and TM12, whichline one side of the central pore and are part of thescaffold, and on TM3 and TM8 from the four-helixbundle, which line on the other side (Fig. 5c). Thisprocess resulted in models that have higher (i.e.,improved) L-CC scores compared to the rigid-bodyfittings of the X-ray crystal structure (Table 2), andthat clearly differ from one another structurally (see,

e.g., TM3 in Fig. 7a and b). Furthermore, when eachof the protomer models was scored against the othertwo EM protomer maps, the scores were consistent-ly lower (Table 2), consistent with the differencesbetween the models arising from optimization to aparticular map and providing support for theproposal that the three protomers adopt differentconformations.

Comparison with conformational changes in arelated transporter

The conformational changes that BetP undergoesmay well be similar to those proposed for

Fig. 7. Different conformations of BetP described by automated flexible fitting to individual protomers of BetP andcompared with two conformations of Mhp1. (a and b) The densities of BetP protomers 1 and 2 at 1.3σ, viewed from theperiplasmic side and overlaid with the models generated by flexible fitting of TM3 (red), TM5 (yellow), TM8 (cyan), andTM12 (blue). The models are shown in spiral representation. The four-helix bundle is indicated by a red broken circle.TM4 (orange) and TM9 (dark cyan) of the bundle, and TM6 (yellow), TM7 (lime), TM10 (slate), and TM11 (blue) of thescaffold were not allowed to move during the flexible-fitting protocol. (c and d) Maps constructed at 8 Å resolution for (c)the outward-facing state of Mhp1 (PDB entry 2JLN) and for (d) the inward-facing state of Mhp1 (PDB entry 2X79), withthe corresponding structures overlaid, and viewed in a similar orientation as the BetP protomers. The position of the four-helix bundle (red broken circle) consisting of TM1 (red), TM2 (orange), TM6 (cyan), and TM7 (dark cyan) changes relativeto the scaffold helices TM3–4 (yellow), TM5 (lime), TM8 (slate), and TM10 (blue).

376 Three-dimensional BetP Structure to 8 Å by EM

Mhp1,16,17 as the two proteins share the same corefold. The proposed transport mechanism derivedfrom structures and simulations of Mhp1 involvesa rigid-body movement of the bundle and thehash domain (Fig. 4d) relative to each other,similar to the rocking-bundle mechanism firstsuggested by Forrest et al. in which the bundleof LeuT tilts relative to the scaffold.18 However,other mechanisms, such as flexing of the firsthelices of each repeat in the four-helix bundle,have been proposed.19,20 To further understand

the differences in individual protomers in the caseof BetP, we compared their conformational differ-ences (Fig. 7a and b) with those observed fordifferent structures of Mhp116 at a resolution of8 Å (Fig. 7c and d).The differences in conformation between an

outward-facing open state [Protein Data Bank(PDB) entry 2JLN] and an inward-facing openstate (PDB entry 2X79) of Mhp1 are clearlydetectable after the conversion of the X-ray struc-tures into 8-Å-resolution maps (Fig. 7c and d).

Fig. 8. Different conformational states within a trimer of BetP. Flexible-fitting models viewed from within themembrane. The accessible substrate pathway in the center of each protomer is shown as a red surface. Tryptophanresidues from TM8 (W362, W366, W373, W374, and W377) are shown as green spheres.

377Three-dimensional BetP Structure to 8 Å by EM

Specifically, differences in the relative position of thefour-helix bundle consisting of TM1–TM2 and TM6–TM7 in Mhp1 (Fig. 7c and d, red broken circle) andthe scaffold helices (TM3, TM5, TM8, and TM10; Fig.7c and d) can be observed. In the outward-facingstate, the bundle helices (e.g., TM1 and TM6) aremore distant from the scaffold (e.g., TM3 and TM10)than in the inward-facing state, when viewed fromthe periplasm. The density in the region of thebundle also alters between these two states, chang-ing from a solid density in the outward-facing openstate of Mhp1 (Fig. 7c) to a ring shape in the inward-facing state of Mhp1 (Fig. 7d). Importantly, theobserved differences in those regions of the BetP EMmaps are similar to those visible in the mapscalculated for Mhp1.Thus, we are confident that conformational

changes observed in transporters with the fold ofBetP are indeed detectable in the protomer maps ofBetP. However, a definitive assignment of theprotomers to specific conformational states remainsa challenge, even given different models fromflexible fitting, because the differences in the L-CCscores of those models are relatively small. Instead,we rely on the comparison to Mhp1, based on whichwe suggest that protomer 1 adopts an outward-facing conformation and protomer 2 adopts aninward-facing conformation (Fig. 7a and b). Asubsequent comparison of the pathways in themodels created by the flexible fitting of the X-raystructure to these two protomer maps is consistentwith that assignment (Fig. 8). Protomer 3 is rathermore difficult to assign but may also resemble aninward-facing state (Fig. 8).We note that at the current resolution, it is not

possible to determine whether or not any of thesestates is substrate bound. Betaine was not addedduring crystallization; in any case, both the openstate and the occluded state can exist in both apo

form and holo form during the transport cycle of asymporter such as BetP. In conclusion, although werefrain from a definitive assignment of the threeprotomers, the presence of different conformationalstates seems to be the most plausible explanation forthe asymmetry in the BetPΔC45 trimer.

Functional relevance of structural asymmetry inthe BetP trimer

Oligomerization of secondary transporters isgenerally assumed to occur for stability reasons,although specific evidence that it is (or it is not)functionally relevant is only available for a fewtransporters.21–24 Such questions are very difficultto answer, as demonstrated by the controversy overmitochondrial ADP/ATP carriers.23 In a crystalenvironment, if all protomers in an oligomer areable to function independently, every protomer canbe expected either to adopt the same low-energystate or to show an average overall accessible state;both of these situations should result in a symmet-ric oligomer in the EM map. Indeed, most X-ray orEM structures of secondary transporters consist ofsymmetric oligomers. Exceptions are the structureof the small multidrug transporter EmrE from E.coli, which forms an asymmetric homodimer in 2Dand 3D crystals,25,26 and the structure of themultidrug exporter AcrB from E. coli, which formsan asymmetric homotrimer in 3D crystals.27,28 InAcrB, the asymmetric oligomer state is thought tobe functional, whereas the functional relevance ofthe asymmetric structure of EmrE remainscontroversial.29 Questions therefore arise: Why doBetPΔC45 and wild-type BetP trimers in 2Dcrystals exhibit significant asymmetry9 while thetrimeric X-ray structure of BetPΔN29 appears moresymmetric?10 Which of these arrangements isphysiologically more relevant?

378 Three-dimensional BetP Structure to 8 Å by EM

One possible explanation for the observed differ-ences between the X-ray structure and the EMstructure of BetP is that they are due to thedifferent constructs of BetP used. While the X-raystructure was determined from an N-terminallytruncated mutant, the EM map was obtained froma C-terminally truncated BetP. Although both ofthese forms are functional active, they showdifferences in transport and regulatory properties.The N-terminal and C-terminal domains play animportant role in the function of BetP:5–8,11 trunca-tion of the C-terminal domain renders BetPconstitutively active at a lower level,6 whiletruncation of the N-terminal domain shifts theactivation threshold of BetP to higher osmolalitiesso that the transporter becomes less sensitive toosmotic stress.5,10 Moreover, mutual N-terminaland C-terminal interactions are important forbetaine transport.7 However, a trimer containingtwo essentially identical protomers was observed ina projection map of BetPΔC45 2D crystals when thecrystals were grown in a different, less polar lipidenvironment12 (Fig. 2). Based on the rigid-bodyfittings of the X-ray structure to the 3D EM map(Fig. 5), it appears that, in this crystal packing,contacts between trimers are maintained by thefull-length N-terminal domain from a single proto-mer of each trimer. This is analogous to thesituation observed in the symmetric X-ray structureof BetPΔN29, where only one of the C-terminaldomains within the trimer is involved in crystalcontacts.10 By contrast, crystal contacts between N-terminal domains are not observed in the 2Dcrystals of wild-type BetP, nor crystal contacts ofBetPΔC45 in the presence of the C. glutamicumlipids shown here. In BetPΔC45 crystals, the C.glutamicum lipids most likely promote a particularcrystal packing (p121) in which adjacent trimers areoriented towards opposite sides of the membraneand, thus, their terminal domains cannot interact.In these crystals, BetPΔC45 forms an asymmetrictrimer with each protomer in a distinct conforma-tion, like the wild-type protein.A dependence on the environment also explains

the previously reported mirror-symmetric trimers inthe 2D crystals of BetPΔC45 grown in a 58:42mixture of E. coli lipid/cardiolipin.12 Apparently,this less negatively charged lipid environmentplaces fewer constraints on the BetPΔC45 trimer,allowing the protomers to adopt different confor-mational states. Thus, the superposition of trimersduring EM data processing led to a mirrorsymmetry12 similar to that obtained by averagingthe different conformational states (Fig. 6b), pre-sumably because the outward-facing and inward-facing states are not in a fixed order in this lipidenvironment. In summary then, crystal contactsformed by N-terminal or C-terminal domains, aswell as the lipid environment, appear to restrict the

conformational states of BetP under such specificconditions. We therefore conclude that the threedistinct protomer conformations in the asymmetrictrimer truly represent the physiologically relevantstructure of BetP.A similar dependence of transporter conformation

on crystal contacts was observed for the multidrugefflux pump AcrB, which also forms eithersymmetric30 or asymmetric27,28 trimers in 3Dcrystals. The asymmetric form is likewise thoughtto represent the physiologically relevant state ofAcrB, consistent with a transport mechanism in-volving conformational coupling.31 The differentforms of AcrB depend on the space group of thecrystal, supporting the premise that crystal contactsmay be an important factor also for BetP.The asymmetry in the BetP trimer implies that

oligomerization is important for the function ofbetaine transport. Moreover, it strongly suggests aconformational coupling between the protomers inthe membrane. Conformational coupling couldsimultaneously facilitate the conversion of individ-ual protomers from an outward-facing state into aninward-facing state by reducing the energetic barrierto one or more of the rate-limiting steps in thetransport cycle. Higher-resolution data will berequired to attain atomically detailed models ofthe three conformations of BetP and to conclusivelyassign each protomer to a given conformation in thetransport cycle. At the same time, investigations ofthe function of the BetP monomer are underway inorder to determine in what ways the catalytic andregulatory cycles take advantage of the trimericstate.

Materials and Methods

Protein expression, purification, and 2D crystallization

The plasmid pASK-IBA5betPΔC45 containing anN-terminal StrepII tag was constructed as describedpreviously11 and transformed into E. coli C43 cells.32

Protein expression and membrane preparation wereperformed as described previously.9 Membranes weresolubilized in 1.5% (wt/vol) dodecyl maltoside (Glycon)for 30 min on ice, and the insoluble material wasremoved by centrifugation (140,000g, 45 min). Thesupernatant was loaded onto 4 ml of Streptactin resin(IBA GmbH) and washed with 25 ml of 50 mM Tris–HCl(pH 7.5), 200 mM NaCl, 10% glycerol, and 0.04%dodecyl maltoside. BetPΔC45 was eluted with 5 mMdesthiobiotin. C. glutamicum lipid extracts in 0.15%decylmaltoside were mixed with purified protein at anLPR of 0.15 (wt/wt) and placed into an amini Slide-A-Lyzer 10K dialysis device (Pierce, Rockford, IL) anddialyzed at 30 °C for 3 weeks against 200–500 ml ofdialysis buffer [50 mM Tris–HCl (pH 7.5), 200 mM NaCl,5% glycerol, 5% 2-methy-2,4-pentanediol, 4 mM CaCl2,and 3 mM NaN3].

379Three-dimensional BetP Structure to 8 Å by EM

Betaine uptake measurements: Functionality in 2Dcrystals

Uptake of [14C]betaine by BetPΔC45 was measured inE. coli polar lipid proteoliposomes and 2D crystals.Proteoliposomes were prepared in accordance withSchiller et al.8 Two-dimensional crystals were fused withliposomes composed of E. coli polar lipids (Avanti) in thepresence of 100 mM potassium Pi (pH 7.5). For thispurpose, 2D crystals were mixed at the indicated ratiosand extruded 20 times through a polycarbonate filter witha pore size of 400 nm before they were frozen into liquidnitrogen. After gentle thawing, the mixture was incubatedfor 1 h at 20 °C. The freeze–thaw cycle was repeated threetimes. The mixture was extruded (15 times) before beingused for transport assays. Proteoliposomes and 2Dcrystals were adjusted to a final concentration of 3–4 μgof BetP with 50 mM Pi (pH 7.5) and 25 mM NaClcontaining also 15 μM [14C]betaine (0.1 mCi/ml) and0.5 μM valinomycin. To establish hyperosmotic condi-tions, we added proline to the external buffer (0.1–1.0 Osm/kg). After different time intervals of 5–20 s,samples were filtered rapidly through 0.22-μm GSnitrocellulose filters (Millipore Corp., Germany), andradioactivity was determined by liquid scintillationcounting. For the comparison of different LPRs in Fig.1b, the osmolality of the external buffer was adjusted to640 mOsm/kg by addition of proline, and uptake wasmeasured at different time intervals of up to 20 s. Themeasurements were repeated five times. Control experi-ments with wild-type BetP in proteoliposomes werecarried out under the same conditions.

Electron crystallography

Two-dimensional crystals were stained with 1% uranylacetate for screening by EM.33 Frozen–hydrated speci-mens for data collection were prepared by backinjection34

with 10% glucose and transferred into a JEOL 3000SFFelectron microscope equipped with a liquid-helium-cooled top-entry stage and a field emission gun operatingat 300 kV. Images were taken on Kodak SO-163 electronemulsion films (at a magnification of 45,000× or 53,000×)in spot-scan mode at a total electron dose of 10–20 e/Å2.

Image processing

Micrographs were selected by optical diffraction, andareas of 6000×6000 pixels were digitized on a Zeiss SCAIscanner with a 7-μm pixel size. The Medical ResearchCouncil image processing software package35 was used toextract structure factor amplitudes and phases from 72individual images, after correcting for lattice distortionsand the effects of the contrast transfer function. Image datafor each tilt angle were merged in-plane with space groupp121_b. The tilt geometry and phase origin of each imagewere refined at 12 Å resolution at the optimal signal-to-noise ratio. Image amplitudes were scaled with an averagenegative temperature factor B=−300 Å2 to compensate forthe resolution-dependent degradation of amplitudes. A3D map was constructed with the CCP4i package,36 andthe map was visualized with the graphics programs Cootand PyMOL.37 NCS-averaged maps were calculated in

Coot and in the CCP4i package using the protomermodels (see the text below).

Automated rigid-body fitting of X-ray structures to EMmaps

We carried out an automated rigid-body fitting of thetrimeric X-ray structure to the trimeric EMmap, as well asof a protomer of the X-ray structure to each of the threeprotomeric EMmaps. The fitting was performed using theCoLoRes tool from the Situs package.14,15 The structureused was the 3.35-Å-resolution X-ray crystal structure ofBetPΔN29 (PDB entry 2WIT) with a 45-residue C-terminaltruncation, with two missing residues (Asp273 andPro274) added using Modeller 9v2.38 CoLoRes performeda six-dimensional (three translational degrees of freedomand three rotational degrees of freedom) exhaustive search(accelerated by fast Fourier transform) of the best fitbetween an experimentally determined map and a mapderived from a model atomic structure. The resolution ofthe derived mapwas set to 8 Å in the membrane plane and16 Å normal to the membrane. The grid spacing was 2.3 Å,and the rotational sampling step size was 10°. To enhancethe “contour” information in the experimental andcalculated maps, we used Laplacian filtering.14 BothL-CC and CC scores are sensitive to the intensity anddimensions of a given map, so absolute values cannot becompared between maps. After ranking by L-CC, the 10best fits were then locally refined using an off-latticePowell optimization.39 In the case of protomer 1, thehighest-ranking fit of the X-ray protomer to the EM mapwas shifted by 10.7 Å along the axis normal to themembrane to match its average position in the top seventrimer fits; this fit was then rescored. This adjustmentallowed for a better comparison with the trimer fit and isjustified given the low resolution of the data along this axis,which is poorest in this particular protomermap. The L-CCscores of the final fits to protomers 1, 2, and 3 were 0.066,0.125, and 0.072, respectively.

Flexible fitting of individual helices to EM maps

Starting from the rigid-body protomer fits, the confor-mation of six TM helix segments was optimized based onthe EM density in each of the three protomers, using anautomated flexible-fitting approach. These segments are3c, 3p, 5, 8c, 8p, and 12p. For each segment, 104 putativeconformations were generated using the programsegsam.40 To generate each conformation, we randomlyperturbed all torsion angles in the flexible regions flankingthe helix segments, resulting in a rigid-body displacementof the segment. Breaks in the flanking loops were closedusing the CCD algorithm. Each conformation was thenL-CC rescored alongwith the rest of the structure using theCoLaCor tool.14 The conformation with the highest L-CCvalue was then selected, although the top five conforma-tions were typically similar. The highest-ranking confor-mations for all six segments were then combined with theremainder of the X-ray structure into a complete model forthat protomer (two alternate models in the case ofprotomer 2). To reduce side-chain steric clashes, weenergy minimized this model using the steepest-descentalgorithm for 1000 steps with the OPLS force field in

380 Three-dimensional BetP Structure to 8 Å by EM

GROMACS.41 The final L-CC scores were then obtainedfor each energy-minimized protomer model in itscorresponding EM protomer map. The DaliLite programwas used for structural alignments.42

Accession numbers

The EM density was deposited in the EM data bankwith accession code EMD-1756. The flexible-fitting modelswere deposited in the Protein Model Databank† withaccession code PM0076442.

Acknowledgements

We thankWinfriedHaase for help with the freeze–fracture experiments, Deryck Mills for assistancewith electronmicroscope operation, Janet Vonck andIngeborg Schmidt-Krey for help with image proces-sing, and Willy Wriggers for assistance with Situs.Caroline Koshy calculated the 8-Å maps of Mhp1.SusanneMorbach andVera Ott contributed valuablediscussions to the BetP transport mechanism. Thiswork was supported by the International MaxPlanck Research School Frankfurt.

References

1. Kempf, B. & Bremer, E. (1998). Uptake and synthesisof compatible solutes as microbial stress responses tohigh-osmolality environments. Arch. Microbiol. 170,319–330.

2. Farwick, M., Siewe, R. M. & Krämer, R. (1995). Betaineuptake after hyperosmotic shift in Corynebacteriumglutamicum. J. Bacteriol. 177, 4690–4695.

3. Morbach, S. & Krämer, R. (2003). Impact of transportprocesses in the osmotic response of Corynebacteriumglutamicum. J. Biotechnol. 104, 69–75.

4. Peter, H., Weil, B., Burkovski, A., Krämer, R. &Morbach, S. (1998). Corynebacterium glutamicum isequipped with four secondary carriers for compatiblesolutes: identification, sequencing, and characteriza-tion of the proline/ectoine uptake system, ProP, andthe ectoine/proline/betaine carrier, EctP. J. Bacteriol.180, 6005–6012.

5. Peter, H., Burkovski, A. & Krämer, R. (1998). Osmo-sensing by N- and C-terminal extensions of the betaineuptake system BetP of Corynebacterium glutamicum. J.Biol. Chem. 273, 2567–2574.

6. Schiller, D., Krämer, R. & Morbach, S. (2004). Cationspecificity of osmosensing by the betaine carrier BetPof Corynebacterium glutamicum. FEBS Lett. 563,108–112.

7. Ott, V., Koch, J., Späte, K., Morbach, S. & Krämer, R.(2008). Regulatory properties and interaction of the C-and N-terminal domains of BetP, an osmoregulated

†http://mi.caspur.it/PMDB/

betaine transporter from Corynebacterium glutamicum.Biochemistry, 47, 12208–12218.

8. Schiller, D., Ott, V., Krämer, R. & Morbach, S. (2006).Influence of membrane composition on osmosensingby the betaine carrier BetP from Corynebacteriumglutamicum. J. Biol. Chem. 281, 7737–7746.

9. Ziegler, C., Morbach, S., Schiller, D., Krämer, R.,Tziatzios, C., Schubert, D. & Kühlbrandt, W.(2004). Projection structure and oligomeric state ofthe osmoregulated sodium/betaine symporter BetPof Corynebacterium glutamicum. J. Mol. Biol. 337,1137–1147.

10. Ressl, S., Terwisscha van Scheltinga, A. C., Vonrhein,C., Ott, V. & Ziegler, C. (2009). Molecular basis oftransport and regulation in the Na(+)/betaine sym-porter BetP. Nature, 458, 47–52.

11. Schiller, D., Rübenhagen, R., Krämer, R. & Morbach,S. (2004). The C-terminal domain of the betaine carrierBetP of Corynebacterium glutamicum is directly in-volved in sensing K+ as an osmotic stimulus.Biochemistry, 43, 5583–5591.

12. Tsai, C. J., Ejsing, C. S., Shevchenko, A. & Ziegler, C.(2007). The role of lipids and salts in two-dimensionalcrystallization of the glycine-betaine transporter BetPfrom Corynebacterium glutamicum. J. Struct. Biol. 160,275–286.

13. Rübenhagen, R., Rönsch, H., Jung, H., Krämer, R. &Morbach, S. (2000). Osmosensor and osmoregulatorproperties of the betaine carrier BetP from Corynebac-terium glutamicum in proteoliposomes. J. Biol. Chem.275, 735–741.

14. Chacón, P. & Wriggers, W. (2002). Multi-resolutioncontour-based fitting of macromolecular structures. J.Mol. Biol. 317, 375–384.

15. Wriggers, W., Milligan, R. A. & McCammon, J. A.(1999). Situs: a package for docking crystal structuresinto low-resolution maps from electron microscopy. J.Struct. Biol. 125, 185–195.

16. Shimamura, T., Weyand, S., Beckstein, O., Rutherford,N. G., Hadden, J. M., Sharples, D. et al. (2010).Molecular basis of alternating access membranetransport by the sodium-hydantoin transporterMhp1. Science, 328, 470–473.

17. Weyand, S., Shimamura, T., Yajima, S., Suzuki, S.,Mirza, O., Krusong, K. et al. (2008). Structure andmolecular mechanism of nucleobase-cation-sympor-ter-1 family transporter. Science, 322, 709–713.

18. Forrest, L. R., Zhang, Y. W., Jacobs, M. T., Gesmonde,J., Xie, L., Honig, B. & Rudnick, G. (2008). Amechanism for alternating access in neurotransmittertransporters. Proc. Natl Acad. Sci. 105, 10338–10343.

19. Krishnamurthy, H., Piscitelli, C. L. & Gouaux, E.(2009). Unlocking the molecular secrets of sodium-coupled transporters. Nature, 459, 347–355.

20. Zhao, Y., Terry, D., Shi, L., Weinstein, H., Blanchard,S. C. & Javitch, J. A. (2010). Single-molecule dynamicsof gating in a neurotransmitter transporter homo-logue. Nature, 465, 188–193.

21. Rimón, A., Tzubery, T. & Padan, E. (2007). Monomersof the NhaA Na+/H+ antiporter of Escherichia coli arefully functional yet dimers are beneficial underextreme stress conditions at alkaline pH in thepresence of Na+ or Li+. J. Biol. Chem. 282, 26810–26821.

22. Bamber, L., Harding, M., Monné, M., Slotboomm, D. J.& Kunji, E. R. (2007). The yeast mitochondrial ADP/

381Three-dimensional BetP Structure to 8 Å by EM

ATP carrier functions as a protomer in mitochondrialmembranes. Proc. Natl Acad. Sci. 104, 10830–10834.

23. Kunji, E. R. & Crichton, P. G. (2010). Mitochondrialcarriers function as monomers. Biochim. Biophys. Acta,1797, 817–831.

24. Zheng, H., Taraska, J., Merz, A. J. & Gonen, T. (2010).The prototypical H+/galactose symporter GalPassembles into functional trimers. J. Mol. Biol. 396,593–601.

25. Ubarretxena-Belandia, I., Baldwin, J. M., Schuldiner,S. & Tate, C. G. (2003). Three-dimensional structure ofthe bacterial multidrug transporter EmrE shows it isan asymmetric homodimer. EMBO J. 22, 6175–6178.

26. Chen, Y. J., Pornillos, O., Lieu, S., Ma, C., Chen, A. P. &Chang, G. (2007). X-ray structure of EmrE supportsdual topology model. Proc. Natl Acad. Sci. 104,18999–19004.

27. Seeger, M. A., Schiefner, A., Eicher, T., Verrey, F.,Diederichs, K. & Pos, K. M. (2006). Structuralasymmetry of AcrB trimer suggests a peristalticpump mechanism. Science, 313, 1295–1298.

28. Murakami, S., Nakashima, R., Yamashita, E.,Matsumoto, T. & Yamaguchi, A. (2006). Crystalstructures of a multidrug transporter reveal a func-tionally rotating mechanism. Nature, 443, 173–179.

29. Schuldiner, S. (2007). When biochemistry meetsstructural biology: the cautionary tale of EmrE. TrendsBiochem. Sci. 32, 252–258.

30. Murakami, S., Nakashima, R., Yamashita, E. &Yamaguchi, A. (2002). Crystal structure of bacterialmultidrug efflux transporter AcrB. Nature, 419,587–593.

31. Takatsuka, Y. & Nikaido, H. (2009). Covalently linkedtrimer of the AcrB multidrug efflux pump providessupport for the functional rotating mechanism. J.Bacteriol. 191, 1729–1737.

32. Dumon-Seignovert, L., Cariot, G. & Vuillard, L.(2004). The toxicity of recombinant proteins in

Escherichia coli: a comparison of overexpression inBL21(DE3), C41(DE3), and C43(DE3). Protein Expres-sion Purif. 37, 203–206.

33. Kühlbrandt, W. (1982). Discrimination of protein andnucleic acids by electron microscopy using contrastvariation. Ultramicroscopy, 7, 221–232.

34. Wang, D. N. & Kühlbrandt, W. (1991). High-resolu-tion electron crystallography of light-harvesting chlo-rophyll a/b–protein complex in three different media.J. Mol. Biol. 217, 691–699.

35. Crowther, R. A., Henderson, R. & Smith, J. M. (1996).MRC image processing programs. J. Struct. Biol. 116,9–16.

36. Potterton, E., Briggs, P., Turkenburg, M. & Dobson, E.(2003). A graphical user interface to the CCP4program suite. Acta Crystallogr. Sect. D, 59, 1131–1137.

37. DeLano, W. L. (2002). The PyMOL Molecular GraphicsSystem. DeLano Scientific, San Carlos, CA. http://pymol.sourceforge.net/.

38. Šali, A. & Blundell, T. L. (1993). Comparative proteinmodelling by satisfaction of spatial restraints. J. Mol.Biol. 234, 779–815.

39. Jones, T. A., Zou, J. Y., Cowan, S. W. &Kjeldgaard, M. (1991). Improved methods forbuilding protein models in electron density mapsand the location of errors in these models. ActaCrystallogr. 47, 110–119.

40. Zhu, J., Xie, L. & Honig, B. (2006). Structuralrefinement of protein segments containing secondarystructure elements: local sampling, knowledge-basedpotentials, and clustering. Proteins, 65, 463–479.

41. Van Der Spoel, D., Lindahl, E., Hess, B., Groenhof,G., Mark, A. E. & Berendsen, H. J. C. (2005).GROMACS: fast, flexible, and free. J. Comput. Chem.26, 1701–1718.

42. Holm, L. & Park, J. (2000). DaliLite workbench forprotein structure comparison. Bioinformatics, 16,566–567.