Journal of Structural Biology 128, 243–249 (1999)Article ID jsbi.1999.4180, available online at http://www.idealibrary.com on

Charting the Surfaces of the Purple Membrane

J. Bernard Heymann,* Daniel J. Muller,* Ehud M. Landau,† Jurg P. Rosenbusch,† Eva Pebay-Peyroula,‡Georg Buldt,§ and Andreas Engel*,1

*M. E. Muller Institute, and †Department of Microbiology, Biozentrum, University of Basel, Basel, CH-4056 Switzerland; ‡Institut deBiologie Structurale, 41 avenue des martyrs, F-38027 Grenoble Cedex 1, France; and §Structural Biology, Forschungszentrum Julich, IBI-2,

D-52425 Julich, Germany

Received July 19, 1999, and in revised form September 3, 1999

mc(tiTacmamAbdlmttdfdAar

e

ism

wtth(ar

obitfie9ambecwmpoG1f(ntalX1tca

va

2

rdr

The preponderance of structural data of the purpleembrane from X-ray diffraction (XRD), electron

rystallography (EC), and atomic force microscopyAFM) allows us to ask questions about the struc-ure of bacteriorhodopsin itself, as well as about thenformation derived from the different techniques.he transmembrane helices of bacteriorhodopsinre quite similar in both EC and XRD models. Inontrast, the loops at the surfaces of the purpleembrane show the highest variability between the

tomic models, comparable to the height varianceeasured by AFM. The excellent agreement of theFM topographs with the atomic models from XRDuilds confidence in the results. Small technicalifficulties in EC lead to poorer resolution of the

oop structures, although the combination of atomicodels with AFM surfaces allows clear interpreta-

ion of the extent and flexibility of the loop struc-ures. While XRD remains the premier technique toetermine very-high-resolution structures, EC of-ers a method to determine loop structures unhin-ered by three-dimensional crystal contacts, andFM provides information about surface structuresnd their flexibility under physiological conditions.1999 Academic Press

Key Words: bacteriorhodopsin; X-ray diffraction;lectron crystallography; atomic force microscopy.

INTRODUCTION

The surfaces of the purple membrane envelope thenternal photosensitive machinery of bacteriorhodop-in (BR)2 and constitute the interfaces to the environ-ent through which protons must pass and with

1 To whom correspondence should be addressed. Fax: 141 6167 2109. E-mail: aengel@ubaclu.unibas.ch.

2 Abbreviations used: AFM, atomic force microscopy; BR, bacte-iorhodopsin; EC, electron crystallography; MSD, mean-squareisplacement; RMSD, root-mean-square displacement; XRD, X-

tay diffraction.

243

hich soluble proteins interact. How well we resolvehese interfaces will determine our understanding ofheir functions. Information about the surfaces of BRave been derived from electron crystallography

EC), by X-ray diffraction (XRD) at high resolution,nd by atomic force microscopy (AFM) at mediumesolution (,6 Å).

BR is the light-driven proton pump forming highlyrdered two-dimensional lattices in the plasma mem-rane of Halobacterium salinarum. After pioneeringmage processing and 3D reconstruction methods,he Henderson group succeeded in producing therst atomic model of BR using EC of glucose-mbedded purple membrane (here called EC-MRC-5; see Table I) (Grigorieff et al., 1996; Henderson etl., 1990). Subsequently, the BR structure was deter-ined to higher resolution by EC of purple mem-

rane embedded in trehalose (EC-Kyo-99) (Kimurat al., 1997; Mitsuoka et al., 1999). Extensive X-rayrystallographic work yielded another four modelsith BR crystals produced in various ways: (a)icrocrystals of BR were grown in a novel lipid cubic

hase (Landau and Rosenbusch, 1996) and used tobtain the first XRD model and later refined (XRD-re-99) (Belrhali et al., 1999; Pebay-Peyroula et al.,997); (b) a second model was also derived using datarom microcrystals grown in a lipid cubic phaseXRD-UCI-98) (Luecke et al., 1998); (c) BR trimersucleated on an organic 2D crystal were used to seedhe growth of 3D crystals (XRD-Mun-98) (Essen etl., 1998); and (d) purple membrane vesicles fused atow temperature yielded 3D crystals suitable forRD (XRD-Nag-98) (Sato et al., 1999; Takeda et al.,998). In addition, AFM provided a detailed view ofhe surfaces of the purple membrane, as well as theonformational flexibility of surface loops (Muller etl., 1995a, 1999).Comparison of the BR structures derived from the

arious techniques gives an estimation of the valuend reliability of each source of information. While

he transmembrane helical structures from XRD

1047-8477/99 $35.00Copyright r 1999 by Academic Press

All rights of reproduction in any form reserved.

aestto

C

rLcclwcbwsids

A

1d

cR

B

(B1vtlattlrcwtmhaMladscTcnrts

H

S

eII

EXXXX

ap

244 HEYMANN ET AL.

nd EC are in excellent agreement, significant differ-nces are manifested in the loop regions at theurfaces. Here we attempt to explain these varia-ions as resulting from a combination of characteris-ics of the technique used and the intrinsic flexibilityf the loop structures.

MATERIALS AND METHODS

oordinate MSD and RMSD

The six available models of BR derived from EC and XRD wereetrieved from the PDB (Table I). After superposition using theSQMAN program (Kleywegt, 1996) (Table II), the coordinates oforresponding atoms in the models were averaged to produce aombined structure and the mean-square displacement was calcu-ated as MSD 5 (1/n) Si51

n Dxi2 1 Dyi

2 1 Dzi2, for each model i,

here Dxi, Dyi, and Dzi are the deviations from the averageoordinates. For comparison with B factors, the MSD was scaledy 8 p2. The root-mean-square displacements of the Ca atomsere calculated as RMSD 5 ÎMSD. To compare images of the

tandard deviation in height with the atomic models, the variancen the loop Ca positions was calculated and projected in the zirection onto a 2D plane to provide a measure of height varianceimilar to that in the AFM images.

FM RMSD

Purple membranes were imaged by AFM (Muller et al., 1995b,999), the images were correlation averaged, and the standardeviation (5RMSD) for each pixel was calculated over all unit

TAExperimental Structures of Bacterio

ource (PDB-ID) Labela Temperatureb (K) Reso

2brd EC-MRC-95 1532at9 EC-Kyo-99 4.21qhj XRD-Gre-99 1001brr XRD-Mun-98 1001brx XRD-UCI-98 1001bm1 XRD-Nag-98 285

a The label contains the type of crystallography, an abbreviatilectron crystallography; XRD, X-ray diffraction; MRC, Medical Rnstitut de Biologie Structurale and Universite Joseph Fourier, Grvine, California; Nag, University of Nagoya, Japan.

b Temperature during data collection.c Only a global B factor.

TABLE IIRoot-Mean-Square Deviations (in Å) between the Ca

Atoms of the Bacteriorhodopsin Models

EC-MRC-95

EC-Kyo-99

XRD-Gre-99

XRD-Mun-98

XRD-UCI-98

C-Kyo-99 1.217RD-Gre-99 0.977 0.835RD-Mun-98 0.903 0.846 0.532RD-UCI-98 0.915 0.838 0.379 0.486RD-Nag-98 1.010 0.967 0.753 0.575 0.754

Note. The monomeric structure, EC-MRC-95 (Table I), was useds reference for orientation of the other five models with the

wrogram LSQMAN (Kleywegt, 1996).

ells (Schabert and Engel, 1994). All correlation averages andMSD maps were p3 symmetrized.

RESULTS AND DISCUSSION

-Factor Distribution and Coordinate Variance

The six models of BR (Table I) were superposedTable II) and compared with respect to isotropic-factor distribution and coordinate variance (Fig.). The ranges of B factors (as given in the PDB files)ary from small (EC-Kyo-99), intermediate (all XRD),o large (EC-MRC-95). Also shown in Fig. 1 (thickine) is the mean-square displacement (MSD) of alltomic coordinates represented on the same scale ashe B factors (see Materials and Methods for calcula-ions). The B factors from the atomic models are alless varied than the MSD, in general adhering to theule that the helical B-factors are lower than adja-ent loop B-factors. The MSD follows the same rule,here values for the helices are mostly lower than

he corresponding B factors and loop values areostly higher. The exceptions are EC-Kyo-99, which

as B factors that are always lower than the MSD,nd XRD-UCI-98, which tracks the minima of theSD curve in the helical regions. The termini and

oops BC and EF show the highest coordinate vari-nce, followed by loops AB and DE. The markedifferences between the model B factors suggestome systematic bias, and the lack of a qualityriterion makes assessment of their value difficult.his raises the concern that the reported B factorsarry too many influences associated with crystalature and quality, as well as data collection andefinement to be useful as absolute values. However,he relative values reveal some general trends intructural flexibility.

eight and Variance of the PurpleMembrane Surface

The largest coordinate variances are associated

Ipsin from Halobacterium salinarum

(Å) B-factor average (Std Dev) (Å2) Reference

114.1 (94.6) Grigorieff et al. (1996)15.7 (8.2) Mitsuoka et al. (1999)35.4 (14.5) Belrhali et al. (1999)58.2 (17.4) Essen et al. (1998)25.1 (12.3) Luecke et al. (1998)22.0 (0)c Sato et al. (1999)

icating the institution, and the date of submission to PDB; EC,h Council, Cambridge, UK; Kyo, University of Kyoto, Japan; Gre,le, France; Mun, MPI Martinsried, Munich, Germany; UCI, UC

BLErhodo

lution

3.53.01.92.92.33.5

on indesearcrenob

ith the surface structures of BR. Imaging the

pva(sdcmnRo

achpmeootdEsi3laTlfnF

c

tR(ditclsDwHsb

h3Riil

iXaa1mrtmaXb

X8bp

245CHARTING PURPLE MEMBRANE SURFACES

urple membrane surface with AFM further pro-ides information on the topography and the vari-nce in the heights of the different loops of BRMuller et al., 1999). Figure 2 shows a collage of eachurface of the purple membrane, with each monomerisplaying a different surface property. Here weompare the RMSD (root-mean-square displace-ent) for coordinates in the z direction (membraneormal) with the height variance (also expressed asMSD) of the AFM images (see Materials and Meth-ds for calculations).On the cytoplasmic side, the peaks in AFM height

nd RMSD, as well as the coordinate RMSD, coin-ide with loop EF (Fig. 2), confirming that this is aighly flexible loop. In AFM images the EF looprotrudes ,8 Å from the membrane on the cytoplas-ic side, in good agreement with the atomic models

xcept for EC-Kyo-99, where the loop extends an-ther 4 Å above the surface (Fig. 3, top). The variancef the other five models in Ca positions of loop EF inhe z direction is 0.6 Å, compared to the standardeviation of 1.9 Å calculated from AFM images. TheF loops of both EC-Kyo-99 and EC-MRC-95 showignificant deviation from the XRD models, suggest-ng that the freedom of this loop is more restricted inD crystals. Similar features are observed for the ABoop, with the EC-Kyo-99 and EC-MRC-95 modelslso differing significantly from the XRD models.he higher stability of loops AB and CD than that of

oop EF has been shown by AFM imaging at higherorce (Muller et al., 1999), indicating that the coordi-ate RMSD for these loops may be overestimated inig. 2.On the extracellular side, the AFM height peak

FIG. 1. Isotropic B factors for all atoms taken from the bRD-UCI-98 and the average of the three monomers of XRD-Mun-p2 to compare with B factors; thick line). The solid lines are for Elack bars with single letters. The two-strand b-sheet in loop BC isresentation.

orresponds to the BC loop that rises by ,5 Å above c

he membrane in the AFM data (Fig. 3). The AFMMSD is much lower than the coordinate RMSD

Fig. 2, bottom), because the EC-MRC-95 modelisagrees with the other five models, where loop BCs seen as a b-sheet. When EC-MRC-95 is excluded,he AFM RMSD for loop BC (,1.2 Å) agrees with theoordinate RMSD (,0.7 Å). The stability of the BCoop compared to the EF loop on the cytoplasmicurface is thus confirmed. The AFM RMSD for loopE is slightly lower than the coordinate RMSD,hile loop FG appears significantly more flexible.owever, loop FG is buried deeply underneath the

urface, so that its apparently high variance mighte attributed to the proximity of loop BC.The N- and C-termini of the atomic models are

ighly variable (Fig. 3). The points of divergence lie–5 Å below the AFM surfaces, for the N-terminus at7 and the C-terminus at S226. In the AFM RMSD

mages, both termini are located at positions show-ng midrange variance, with the N-terminus nearoop BC and the C-terminus near loop EF (Fig. 2).

In 3D crystals some of the surface loops arenvolved in crystallographic contacts. In the cases ofRD-UCI-98 and XRD-Gre-99, the BR trimers arerranged in a head-to-tail packing, with loops BCnd AB forming the main contacts (Luecke et al.,998; Pebay-Peyroula et al., 1997). In contrast, theodels for XRD-Mun-98 and XRD-Nag-98 were de-

ived from crystals with a head-to-head and tail-to-ail packing (Essen et al., 1998; Sato et al., 1999). Theain interactions are between opposing BC loops

nd N-termini on the extracellular side, while forRD-Nag-98 the contacts on the cytoplasmic side areetween the EF loops (for XRD-Mun-98 this is not

rhodopsin models, EC-MRC-95, EC-Kyo-99, XRD-Gre-99, andcompared to the mean-square displacement of all atoms (scaled byels and the broken lines for XRD models. Helices are indicated asndicated. The curves were smoothed over a 10-residue window for

acterio98 areC modalso i

lear). The agreement between EC-Kyo-99 and the

dsarcs

FIG. 2. The cytoplasmic (top) and extracellular (bottom) surfacesifferent surface property (note that the views of the surfaces are frourface loops are shown as backbone tracings colored according to thend the height maxima (light gray areas) correspond to the most proepresent the coordinate RMSD (i.e., the variation between the atomic

of the purple membrane, with each bacteriorhodopsin monomer displaying am outside the membrane as you would see it in an AFM experiment). Thebackbone coordinate RMSD. The grayscale images were determined byAFMminent loops EF (cytoplasmic) and BC (extracellular). The colored imagesmodels) and the AFM height RMSD (i.e., flexible parts on the surface). The

oordinate RMSD for each surface was calculated by summing the RMSD calculated for all loop atoms in the z direction and rendering it at ,6 Å toimulate theAFM resolution.

246

Xcsv(3

tTsg

rcffast,

247CHARTING PURPLE MEMBRANE SURFACES

RD models for loop BC (Fig. 3) indicates thatrystal contacts do not change its conformationignificantly. However, the EF loop in the EC modelsaries significantly from that in the XRD modelsFig. 3), suggesting that it may be less flexible due to

FIG. 3. The AFM cytoplasmic (top) and extracellular (bottomepresentations of the six atomic models. The largest deviations arontacts (particularly the BC and EF loops). The double letters indrom the origin as defined in the EC-MRC-95 model. The AFM sollowing measurements as guide: The purple membrane thicknessnd extracellular surfaces were positioned to give this separatiourface); The cytoplasmic side height varies by 8 Å and the extracwo surfaces of 43 Å (i.e., 27 Å for the cytoplasmic side and 16 Å fxray/dino).

D crystal packing. R

Comparing the different surface structures of BR,he prominent protrusions are the loops BC and EF.he good height correspondence between the AFMurface and the atomic models in the EF loop sug-ests the measurement of real flexibility in the AFM

ces of the purple membrane (shown in purple) overlaid over tubein the EC models where the surfaces are not restricted by crystal

oops and numbers indicate distances along the membrane normalwere positioned by eye along the membrane normal, using the

easured by AFM as 56 Å and the highest points of the cytoplasmic35 Å for the cytoplasmic surface plus 21 Å for the extracellularside height by 5 Å, giving a separation of the lowest points of theextracellular side) (rendered in DINO:http://www.bioz.unibas.ch/

) surfae seenicate lurfaceswas m

n (i.e.,ellularor the

MSD image. EPR spectroscopy of spin-labeled cys-

tr1fdncArfabtlsp

bwldiwlm(9dofl

aAwerai1wanlmsc

bmpttmi

lttcovt

cwCt(

B

D

E

G

H

K

K

K

L

L

M

M

M

248 HEYMANN ET AL.

eine mutants also suggested a high mobility ofesidues 158 to 165 in the EF loop (Pfeiffer et al.,999). The flexibility in this loop may be importantor function, as the observed movement of helix Furing the photocycle (Dencher et al., 1989; Subrama-iam et al., 1997; 1999; Vonck, 1996) would require aonformational change in loop EF. Loops BC andB are much more stable than loop EF, and theiroles in providing 3D crystal contacts should there-ore not be a surprise. The equivalents of loops CDnd EF and the C-terminus in visual rhodopsin haveeen implicated in the binding of its G-protein,ransducin (Konig et al., 1989). The correspondingoops and the C-terminus cover about half of theurface of BR, with the EF loop as the most flexibleart of it.The surfaces of the purple membrane as revealed

y EC, XRD, and AFM show a significant agreementith respect to the positions and variability of the

oops of BR. At the same time details from theifferent data sets show how reliable the informations. All the XRD models agree on the loop structures,hich fit very well with the AFM surfaces. The EC

oop models vary from good agreement with XRDodels (loop BC of EC-Kyo-99) to poor agreement

loops AB and EF in both EC-MRC-95 and EC-Kyo-9, and loop BC in EC-MRC-95). The poor correspon-ence of B factors for the various models casts doubtn their use as good indicators of the absoluteexibility of loop structures.In contrast, loop variability among the various

tomic models coincides with flexibility indicated byFM, establishing it as an excellent technique withhich to study loop conformational variation (Muller

t al., 1999). Alternative loop conformations wereecently demonstrated by comparison of the trigonalnd orthorhombic crystal forms of BR by AFMmaging under identical conditions (Muller et al.,999). In the orthorhombic crystal, BR forms dimershere the EF loop is stabilized, the AB loop shifted,nd the FG loop destabilized compared to the trigo-al crystal. Thus, the structure and flexibility of

oops are affected by adjacent molecules in theembrane, strongly indicating that such surface

tructures are best studied by AFM under physiologi-al conditions in their native packing.

CONCLUSION

Can one trust the surface features as determinedy the various techniques? There is a good agree-ent between AFM (which probes the surface under

hysiological conditions) and XRD (where the pro-ein resides in a crystal, typically at low tempera-ures, Table I). The slight disagreement with ECodels is a matter of resolution and some decrease

n data quality at the membrane surfaces. Neverthe-

ess, the concern that 3D crystal packing may limithe conformational flexibility of loops remains, whilehe surfaces in 2D crystals are not hindered by 3Drystal contacts. EC and AFM thus provide anpportunity to study the surface conformationalariability and flexibility with fewer artificial restric-ions.

We thank Professor D. Oesterhelt and Dr. S. Strelkov for usefulomments and A. Philippsen for use of his program, Dino. Thisork was supported by the European Union (Grant EC BIO4-T960472 toA.E. and EU-Biomed Grants PL950990 and PL970415

o E.M.L. and J.P.R.) and by the Swiss National FoundationSPP-Biotechnology Grants 5002-46092 and 5002-55179).

REFERENCES

elrhali, H., Nollert, P., Royant, A., Menzel, C., Rosenbusch, J. P.,Landau, E. M., and Pebay-Peyroula, E. (1999) Protein, lipid andwater organization in bacteriorhodopsin: A molecular view ofthe purple membrane at 1.9 Å resolution, Structure 7, 909–917.encher, N. A., Dresselhaus, D., Zaccai, G., and Buldt, G. (1989)Structural changes in bacteriorhodopsin during proton translo-cation revealed by neutron diffraction, Proc. Natl. Acad. Sci.USA 86, 7876–7879.ssen, L., Siegert, R., Lehmann, W. D., and Oesterhelt, D. (1998)Lipid patches in membrane protein oligomers: Crystal struc-ture of the bacteriorhodopsin–lipid complex, Proc. Natl. Acad.Sci. USA 95, 11673–11678.rigorieff, N., Ceska, T. A., Downing, K. H., Baldwin, J. M., andHenderson, R. (1996) Electron-crystallographic refinement ofthe structure of bacteriorhodopsin, J. Mol. Biol. 259, 393–421.enderson, R., Baldwin, J. M., Ceska, T. A., Zemlin, F., Beck-mann, E., and Downing, K. H. (1990) Model for the structure ofbacteriorhodopsin based on high-resolution electron cryo-microscopy, J. Mol. Biol. 213, 899–929.imura, Y., Vassylyev, D. G., Miyazawa, A., Kidera, A., Mat-sushima, M., Mitsuoka, K., Murata, K., Hirai, T., and Fujiyoshi,Y. (1997) Surface of bacteriorhodopsin revealed by high-resolution electron crystallography, Nature 389, 206–211.leywegt, G. J. (1996) Use of non-crystallographic symmetry inprotein structure refinement, Acta Crystallogr. Sect. D 52,842–857.onig, B., Arendt, A., McDowell, J. H., Kahlert, M., Hargrave,P. A., and Hofmann, K. P. (1989) Three cytoplasmic loops ofrhodopsin interact with transducin, Proc. Natl. Acad. Sci. USA86, 6878–6882.

andau, E. M., and Rosenbusch, J. P. (1996) Lipidic cubic phases:A novel concept for the crystallization of membrane proteins,Proc. Natl. Acad. Sci. USA 93, 14532–14535.

uecke, H., Richter, H. T., and Lanyi, J. K. (1998) Proton transferpathways in bacteriorhodopsin at 2.3 angstrom resolution,Science 280, 1934.itsuoka, K., Hirai, T., Murata, K., Miyazawa, A., Kidera, A.,Kimura, Y., and Fujiyoshi, Y. (1999) The structure of bacterior-hodopsin at 3.0 Å resolution based on electron crystallography:Implication of the charge distribution, J. Mol. Biol. 286, 861.uller, D. J., Buldt, G., and Engel, A. (1995a) Force-inducedconformational change of bacteriorhodopsin, J. Mol. Biol. 249,239–243.uller, D. J., Sass, H.-J., Muller, S., Buldt, G., and Engel, A.(1999) Surface structures of native bacteriorhodopsin dependon the molecular packing arrangement in the membrane, J.

Mol. Biol. 285, 1903–1909.

M

P

P

S

S

S

S

T

V

249CHARTING PURPLE MEMBRANE SURFACES

uller, D. J., Schabert, F. A., Buldt, G., and Engel, A. (1995b)Imaging purple membrances in aqeous solutions at sub-nanometer resolution by atomic force microscopy, Biophys. J.68, 1681–1686.

ebay-Peyroula, E., Rummel, G., Rosenbusch, J. P., and Landau,E. M. (1997) X-ray structure of bacteriorhodopsin at 2.5 ang-stroms from microcrystals grown in lipidic cubic phases, Science277, 1676–1681.

feiffer, M., Rink, T., Gerwert, K., Oesterhelt, D., and Steinhoff,H. J. (1999) Site-directed spin-labeling reveals the orientationof the amino acid side-chains in the E-F loop of bacteriorhodop-sin, J. Mol. Biol. 287, 163–71.

ato, H., Takeda, K., Tani, K., Hino, T., Okada, T., Nakasako, M.,Kamiya, N., and Kouyama, T. (1999) Specific lipid–proteininteractions in a novel honeycomb lattice structure of bacterio-rhodopsin, Acta Crystallogr. Sect. D 55, 1251–1256.

chabert, F. A., and Engel, A. (1994) Reproducible acquisition of

Escherichia coli porin surface topographs by atomic forcemicroscopy, Biophys. J. 67, 2394–2403.

ubramaniam, S., Faruqi, A. R., Oesterhelt, D., and Henderson,R. (1997) Electron diffraction studies of light-induced conforma-tional changes in the Leu-93 = Ala bacteriorhodopsin mutant,Proc. Natl. Acad. Sci. USA 94, 1767–1772.

ubramaniam, S., Lindahl, M., Bullough, P., Faruqi, A. R., Tittor,J., Oesterhelt, D., Brown, L., Lanyi, J., and Henderson, R.(1999) Protein conformational changes in the bacteriorhodopsinphotocycle, J. Mol. Biol. 287, 145–161.

akeda, K., Sato, H., Hino, T., Kono, M., Fukuda, K., Sakurai, I.,Okada, T., and Kouyama, T. (1998) A novel three-dimensionalcrystal of bacteriorhodopsin obtained by successive fusion of thevesicular assemblies, J. Mol. Biol. 283, 463–74.

onck, J. (1996) A three-dimensional difference map of the Nintermediate in the bacteriorhodopsin photocycle: Part of the F

helix tilts in the M to N transition, Biochemistry 35, 5870–5878.