Path of the polypeptide in bacteriorhodopsin (purple membrane/diffraction/protein folding)

Biophysics

Path of the polypeptide in bacteriorhodopsin

(purplemembrane/diffraction/proteinfolding)

D. M. ENGELMAN*, R. HENDERSON, A. D. MCLACHLAN, AND B. A.WALLACEt

Medical Research CouncilLaboratoryofMolecularBiology,HillsRoad,CambridgeCB22QH,England Communicatedby FredericM.Richards, January7,1980

ABSTRACT Anattempthasbeen made to fit the amino acid sequenceofbacteriorhodopsintothe three-dimensionaldensity map of the molecule. First, seven segmentsof thesequencewere selectedasbeing probabletransmembraneahelices. Then each of the5040possible ways offittingthesesevensegments into theseven regions of helical densityin themapwereevaluated basedon the criteria ofconnectivityof the nonhelical link re- gions, chargeneutralization, and total scattering density per helix. A single modelthatmaybeexperimentallytestedemerged asthe mostprobable.

Bacteriorhodopsin is a transmembrane protein found in Halobacter halobium. Lightenergyabsorbedbyretinal bound tothe protein is used topump protonsacrossthemembrane, and the protongradientisthen usedasanenergysource(for review seeref. 1). The molecules of this proteinexistnaturally inahighly ordered two-dimensional array. Alow-resolution map(2) of theelectron-scatteringdensityofpurple membrane shows that each bacteriorhodopsin molecule iscomposedof sevenrods ofdensityorientedperpendiculartothemembrane surface. Thesearebelievedtorepresenta-helical segments of the polypeptide that cross the membrane. The molecular boundaryofasinglebacteriorhodopsin moleculeis nowknown from observation of the same molecule in a different crystal environment(3).

The complete amino acid sequence of bacteriorhodopsin has been determined(4)and portions ofithave beenindependently confirmed (5,6). Ovchinnikovetal. (4)used theproteolytic cleavagepointsof the proteininthenativemembranetopro- poseanarrangementof thepolypeptideinthemembrane. This consisted ofseven a-helical segments ofpolypeptide,each of which containedbetween 26and 32 amino acids, with short nonhelicalsegmentsof upto 8aminoacidslinking them. Ad- jacenthelicesinthe sequence had opposite orientations in the membrane. Additional cleavagesites atthe NH2 and COOH terminideterminedby Walkeretal. (6) provide further con- straints onthepolypeptidearrangement.

A useful step towards a more detailed description of the structureofbacteriorhodopsin wouldresult from a determi- nationofthe way in which the sequence fits into the density map.There are, apriori, 5040(7!) ways of doing this. There wouldbe 10,080 ways if the orientation of the sequence and density map were not known. However, the correlation of electron diffraction with freeze fracture (7, 8) has established the orientation of the map with respect to the outer surface of the cell. The top of the model in our convention is the cyto- plasmicsurface. Similarly, digestion experiments on inside-out vesicles havedemonstrated that the carboxyl end of the se- quenceis onthecytoplasmic surface of the membrane (4, 9).

Inthis paper we examine the 5040 possible models by using threecriteria:thelengths of links between helix ends, the for- mation of ion pairs in the protein interior, and the electron

scattering powerof each helix. We rejectalarge numberof models as unrealistic andproposeasinglemodel as the most probable.

RESULTS AND DISCUSSION

Arrangementofpolypeptide in membrane

Anyconsiderationslimitingthewaysinwhich a model canbe constructeddependonhavingareliable idea of thearrange- mentof thepolypeptide segmentsacrossthe membrane.We must,therefore,make an estimate of the exactlengthsof helical segmentsandlinkregions connectingthem before examining waysoffitting the polypeptide segmentsof thesequenceinto thedensitymap.

Fig. 1 showsa choice of helical segmentsratherdifferent from thatproposed by Ovchinnikovetal.(4).Indecidingthe exactpositionsof the ends of the helices,wehave endeavored toconstruct anarrangement conservatively, where theonly residues included in the helicalsegmentsarethose that fit strong criteria of hydrophobicity and inaccessibility to proteolytic cleavage. Theaccessibilitytoproteolytic cleavageisclearlya property of surfaceresidues;thehydrophobicityinthehelical segmentsispreferredonenergeticgroundsforresiduesthat are either buried inside theproteinorbelow the surface of thelipid bilayer.Correspondingly, theoccurrenceofahighdensityof chargedandpolarresidues is taken as astrongindication of a nonhelicallinkregion.Wehave also triedtoavoid assumptions that resultineitherunusually shortorlong helices becausethe knownmembrane thickness, the absence of substantialsurface projections,and thethree-dimensionaldensitymapsuggest that thehelices should becomparable in length. As aresult, the lengths of ournonhelical link regions are rather longer than may eventuallybe found. Forexample, theyare onaverage three residues longer than those proposed by Ovchinnikov et al.

(4).

Severalfeatures of Fig.1areimportant, someofwhich have alreadybeen noted (4). First,mostofthe charged residues (Asp, Glu, Lys,andArg)are onone ortheother surface.Ontheouter surface ofthemembrane,thereare sixcharged residues that areeither entirely accessibleor areclose enough to the end of ahelixsothat theirside chainscanreachthe solvent. Similarly, thereare 19charged residuesatthe cytoplasmic surface.

Second, thereare nine charged residuesinFig. 1 that are sufficientlyfar from either membrane surface to make direct interaction with water unlikely. These charged side chains wouldbe energetically very difficult to bury except as self- neutralizing ion pairs. Of these nine, four may naturally form ionpairswithinthe same helix because they are separated by either threeorfour residues in the sequence. These are (Arg-82 andAsp-85)and(Asp-211and Lys-215).Other ion pairs may

*Present address: Department of Molecular Biophysics and Bio- chemistry,Yale University,Box1937Yale Station, New Haven, CT 06520.

tPresentaddress: Department ofBiochemistry, College of Physicians and Surgeons, Columbia University,NewYork,NY10032.

Thepublication costs of this article were defrayed in part by page chargepayment.This article must therefore be hereby marked "ad- vertisement"inaccordancewith 18 U. S. C. §1734 solely to indicate this fact.

PLEGLW

LAAL

ILE

\SER SERKs>PRO

33 ALA3

_

Y

PHIILE YRO

ALA PHE THR

MET

TYR

A TR LEU

MET PRO

GLI

_ TYR GLY

LEU THR67 MET

VAL PRO

PHE

B

-~ALA

VAL THR

ALA LEU ALA H

,'LEU VLILE

CLVVALGL VAL LEU~~~~~~~~I

PO THR MET

THR ILE ALA

LUPHE ^{GLY THR}

AT

LIP

^)GLY

Sii

A YR GLY VAL

ARA)

^ALA VAL

TRP TR ( H Y

PROIALE^VAL

ASN T'YR-SER132

GIN75

CLV

ALI

SER THR pH[

LY PHE LEU

TYR LEU

SER MET

ALA ERTHR

TRPALA

LR PH

-

^w

B PRO 164 2

MAL VAL

THR SER PHI

VALD GLY

ASMSII

^A

VAL THR

VA VAL LLEU

TRPLEU

SER P

ALA PROTYR

VAL C

TPVAL

LE LE L

GLY SER192

<3GLY ALA

ALA THR PRO

ALA

<

^U

ALA PRO

25SER GL< E

ARG___ < GL ALA

GLY go

GLY

C D E F G

FIG. 1. Suggestedarrangementof thepolypeptide of bacteriorhodopsinacrossthemembrane. ThesevenhelicesarelabeledA-G, starting fromthe aminoterminus.Thesequenceisfrom Ovchinnikovetal. (4).Thehatchedmarksindicate the approximate location of the lipid hy- drocarbonregions. ⁰and0indicatepositive and negative charges, respectively.

existbetweenthepositively charged residues of helixF(Lys-171 andArg-174)and the otherwise isolatednegativechargeson

Asp-96ofhelixCand Asp-llS of helixD.Thesepairingscould neutralizefour^moreburiedcharges andwill beconsidered later.

Thus,Glu-203istheonlyotherdeeplyburiedsingle charge, andthis^mayevenform^asecond saltbridgewithArg-82. In fact,Arg-82istheonlyresiduethat Glu-203canreach.Another notable feature ofthearrangementofthepolypeptideinFig.

1 isthat the^acarbon oflysine41 isnearthesurface,approxi-

mately19Afrom the membranecenter.Theretinal attached

tolysine 41(ref. 10)willextend^noclosertothecenterthan9 Aandisnearertotheinnermembranesurfacethanproposed previously (2). This position isin conflict with reports from neutronanalysis(11).

Applicationofcriteria for correlationofsequencewith three-dimensional densitymap

Connectivity ofPolypeptideMust Be Possible. The first criterionthatweusetoeliminateunrealistic fits of the helices tothedensitiesisthatthelengthsofthepolypeptides linking theends of the^ahelicesmustbe sufficienttoconnectthe ends ofthe features in thethree-dimensional map. Toapplythis criterionsystematically,weusethenumberingshowninFig.

2torefertothe rodsof helixdensityinthemap,togetherwith theletteringofFig. 1 torefertothehelixassigned fromthe

sequence,givingthelettersinorderof the numbered helices 1-7.

Theconnectivity criterion isthenapplied by considering each linkineachof the5040possiblemodels andeliminatingany

model in which one or moreimpossible connections occur.

Table¹givesthe estimated numberof residues allowedineach linkbetween connectedhelicesin thesequencesand themax-

imallengthofthe link if the chainis inafullyextendedcon-

formation(3.6Aperresidue).Alternate links lieontheinside and outside surfaces. Table 2 gives themeasured center-to- centerdistance between the ends of the rodsinthe three-di- mensional^map.Notethat thedistances between the endsofthe pairsofrodsaredifferentonthetwosurfaces of the membrane because of the different tilts of the helices. If theconnecting pathisobstructed byaninterveninghelix,weadd 5Atothe

FIG. 2. Convention usedfor numbering the sevenregions of presumedhelicaldensityin theprojection (fromref.12)and(Inset) the three-dimensionalmap(fromref.2).Theareasused in thedensity calculationareshown.

VA GLY

Inside

Outside

MET ,~GIv

(LYSJ

VAL PHELI TYR LEU THR

GLY LEU GLI

MET LEU ALA

GLYTHR

LEU AU

TRPLEII ILE TRP

PRO6

GLY 6 THR ILE GLN ALA GLU

A

Table 1. Numbers of residues in links between helices

Helix Side of Maximumlength

pair membrane Residues oflink, A*

A-B In lot 36.0

B-C Out 9t 32.4

C-D In 5 18.0

D-E Out 4t 14.4

E-F In 8 28.8

F-G Out 6 21.6

*Maximumlengthassumed as3.6Aperresidue fully extended.

tThenumbers of residues in these links are slightlydifferent from those inFig. 1,where A-B=9,B-C=11,and D-E=2.The four residuesbetween Dand E are computationally necessary because wehave assumed that links must span the center-to-center distance between helices. The twolinks A-B and B-Careverylong and changesinthemonly affect structurescontaininglinks greater than 32A.

measured lengths in Table 2 to allow for the longer linkages required;ifthepathisobstructedbyseveral intervening helices and is extralong, we add 10 A to the measured length of the linkage. The connectivity test is applied by a simple computer programand reduces the number of feasible models to 405.

These 405 are then ranked inorder of increasing total link length, models with longer link lengths being considered less likely. At this stage the numberofpossiblemodels is inten- tionally large. If we used shorter link lengths (e.g., ref. 4), the number would be substantially smaller. Also, in some of the models(144), thepolypeptidesforming two different links cross over oneanother onthe same side of the membrane. Such a topography has not been found in any ofthe five known protein structures that contain groups of fourparallelhelices, so we consider these crossed models less likely. In any crossover, at least one of the links would need to take a longer path.

BuriedChargesMust Form Ion Pairs. As mentioned above, theisolatedchargesofAsp-96,Asp-115, Lys-171, and Arg-174 areconsidered to form ion pairs with charges on other helices.

These mayneutralizeoneanother directly in a simple C-F and

Table2. Measured distances(A)between helices in structure Inside surface Outside surface

Helixpair (top) (bottom)

1-2 2-3 3-4 4-5 5-6 6-7 1-7 2-6 2-7 3-6 4-6 1-3 1-6 2-4 2-5 3-5 3-7 4-7 5-7 1-4 1-5

11.5 10.0 10.0 12.0 11.5 9.5 11.0 11.5 9.5 10.5 12.0 20.5 (+5) 19.0 (+5) 20.0 (+5) 23.5 (+5) 17.5 (+5) 15.5 (+5) 21.5 (+5) 19.5 (+5) 30.0(+10) 30.0(+10)

10.0 8.0 11.0 9.0 8.5 10.0 13.5 14.0 14.0 13.0 14.5 17.0 (+5)

17.0 (+5)

18.5 (+5)

19.5 (+5)

12.5 21.5 (+5)

24.0(+10)

20.0 (+5)

27.0(+10)

25.0(+10)

Middle 11.51 12.01 12.51 10.5 I 10.01 10.01 9.5I 12.01 10.51 13.0I 15.01 23.0 18.0 21.5 21.0 17.01 19.0 25.0 19.0 32.0 28.0

D-Finteraction as describedearlier(andseeTable3)or may formpairswith otherchargesfrom the intrahelixpairs(Arg-82 andAsp-85) or(Asp-211 and Lys-215). There aretherefore manypossiblecombinations of neutralizing pairs amongthese eightburiedcharges.

Ahydrogen-bonded Asp-ArgorAsp-Lysionpairbetween twohelicescanonlyform if the axes of the two helices in the middledepthof the membrane lie closer than18.5A and17.0

A,respectively. Table2indicates the pairs of helicesthatsatisfy thiscondition and allowanunobstructedbridgebetween ion pairs(pairings between helices1and 6,2and 4,2and5,3and 7,and5and7 areconsidered tobe obstructedbyother hel- ices).

The vertical separation between charges, alongthe helixaxes, isalso an important factor.We considered that the nineburied residues could take partin ionpairs betweentwocompensating charges on helices with axes within 18.5 A of oneanother, provided that the paired amino acids were less than eight residues apart vertically. Pairings betweenresidues more than eight apart vertically wereallowedonly for helixaxispairs within 13.5A ofoneanother. Thus, pairs 3-5 and4-6were forbiddenfor the interactionsAsp-85andLys-171andAsp-115 and Arg-82, which requirealongverticaljump. Pairingsbe- tweenresiduesmorethan11apartverticallywereconsidered impossible.

Acomputerprogramtoexplore possibleionpairsbetween different heliceswasusedto setupatable for each model. The pairingswithinhelicesbetween Arg-82 and Asp-85 and Asp-211 and Lys-215wereassumedtobe feasibleineverymodel. We thenappliedthe additional strong condition thatanacceptable modelmustallow everyoneofthe four isolatedchargesAsp-96, Asp-llS, Lys-172,and Arg-174tobeneutralizedbypairingit off withauniqueexclusive partner from among thenineburied charges.This requirement reduced the number ofacceptable models from405 to 172.

Allof the surviving models allow the neutralization of at least eightburiedcharges. Somemodels alsoallow the pairing of Glu-203(to Arg-82), making possiblethe neutralization of all nineburied charges.

Table3lists the best35ofthesemodels, whichcomefrom the top quarterof theoriginalranked list of 405 models. Models 36-41have been added forspecialreasons, described later.

Thesimplest way to satisfy the strong condition is to place helixFclose to bothCandD(indicated by finTable 3), but this conditionis notessential. Models with helix CnearG(ginTable 3) allowanArg-82-Glu-203 pair.Somemodels can only satisfy the strongcondition by pairing Asp-115 with Arg-82 because helixDisclose to C but not to F and G. This pairing is unfa- vorable and may indeed be impossible, depending on the pre- cisevertical registration between helices, so that such structures (dinTable3)have been placedinaspecial group at the bottom ofthe table (models 30-35).

Correlation of Electron Scattering Cross Section with DensityinMap.The projection map of purple membrane (12) isveryaccurately determined, as seen objectively by the en- tirely independent mapinthe orthorhombic form of purple membrane(3). Thus,itshould be possible to use the differences inintegrated scattering density in each helix, together with the differencesinthe chemical composition of each helix, to pro- vide additional constraints on possible fits of the sequence to the map. This has been done in Tables 4 and 5. Table 4 gives the integrated density of the regions of the projection map of the p3structure(12) corresponding to each helix. Fig. 2 shows theareasused. Theboundary of the protein was chosen to have an areaof68%of the asymmetricunit.With this area,the total massofthe protein would be75%of thewhole, assuming that

Iindicates thepairsofhelices thatmayform ionpairs.The(+5

A)

and(+10A) distances, whereindicated,are usedtoaugmentthe measureddistancesasdescribedinthetext.

Table 3. The 41 bestmodels Total

link Density

Helix length, Cross- Link Ion corre- Model position A over types pairing lation

1 AGFEDCB 59.5

2 ABCDEFG 60.0

3 BCDEGFA 62.5

4 BCDEFGA 62.5

5 GFCBADE 63.0

6 EFGABCD 63.0

7 BCGFEDA 64.0

8 EDCBAGF 64.0

9 GFCABDE 64.5

10 EDCABGF 65.5

11 ACDEGFB 66.5

12 ACDEFGB 66.5

13 ACGFEDB 68.0

14 EFGBACD 69.0

15 BCDGEFA 69.5

16 BCDFEGA 70.0

17 BGFEDCA 71.5

18 AGFDECB 72.5

19 ECBAGFD 73.5

20 AEDFGCB 73.5

21 ACDGEFB 73.5

22 ABCGFDE 73.5

23 GDEABCF 74.0

24 ACDFEGB 74.0

25 BCFGEDA 74.5

26 BACDEFG 74.5

27 AFGEDCB 74.5

28 ABCFGDE 75.5

29 GCBAEDF 76.0

30 ADEFGCB 59.5

31 DCBAGFE 61.0

32 GEDABCF 62.0

33 GEDBACF 68.0

34 BDEFGCA 71.5

35 ADEGFCB 72.0

36 DCABGFE 76.5

37 DEBAGFC 77.5

38 DCBGAFE 79.5

39 ABECDFG 80.0

40 DCFGABE 82.0

41 DCGBAFE 83.0

v6 v6 v6 v6 V6 v5s

v6 v6

v5s v5sV5S V5S v5s

v*sm

X v*sl

X v*sm

v5m

X v5m

X vsl

v51

X V3S21

v51

X v*sm

X v3s2m

v51 v51 v51

X v51

v*sl v6 v5s

V5S

v*sm v5m v5m

v4sl

v51

v3sml

*~sl

v3s2m

v3s2l The latercolumns of the tablegivesomedetailed4 features of different models. Totallinklengthwa;

theaugmenteddistances in Table 2.Xindicates thi

over onthesameside of the membrane.The linktypesdescribe the length of eachof the linkregions:vmeansthataconnection isvery

short (8-12 A), s stands for short (12.5-14.0 A),m for medium (14.5-17.5 A),and¹forlong(18.0-24.0 A).Intheion-pairing column, findicates that helix F is closetoboth C andD,sothatthefour iso- latedchargescould all neutralizeoneanotherdirectly;gmeansthat helicesC andGareclose, allowinganArg-82-Glu-203 pair;dmeans

thathelix D isonlyclosetoC.Models in which helices D and Aare

inpositionsof lowscattering densityareindicatedin the lastcolumn.

Models 1-29comefrom thetopquarterofthe ranked list of 405 possibleconnectedmodels. Models 30-35arealso in thetopquarter, but havehelixDdistant fromFandGandcanonlyneutralizeAsp-115 by makinganunusually longion-pairconnection toArg-82.Models 36-39 haveDinlowdensityand lie between thetop25%andtop35%

intherankedlist.Models 40and 41 have bothAandD inlowdensity, butlieonlybetween the top35%andtop50%intherankedlist.Thus,

v,f,g,D,andA^areall favorabledescriptions;model1istheonlyone

withall ofthese.

fg f f

g

f fg

g g

f

g

f

g g

fg f

g

fg fg

DA

D

Table 4. Helix densities in map Integrated Absolute

Area on density integrated Relative map, above lipid, density, scattering Helix A2 contours xA2 contours xA2 strength

1 126 215 1106 0.955

2 111 302 1193 1.030

3 107 310 1200 1.036

4 109 294 1185 1.023

5 94 224 1115 0.963

6 104 255 1146 0.989

7 102 272 1162 1.004

The total area within eachcontour level was measured and assigned to theseparate helices by using the boundaries shown in Fig. 2. The absolute zero level of -9.98 contours was obtained by comparing the meanprojected protein and lipid densities with values calculated from their known atomic compositions. The calculated protein/lipid electron-scattering density ratio is 1.30. Absolute integrated densities (column 4) were obtained by adding a constant amount to each helix.

This was estimated to be1h of the scattering expected from an area oflipid equal to the total projected area of the protein (753A2).The total area ofthe asymmetric unit (lipid plus protein) is 1102A2.

f proteinhasadensity1.4 timesthat of lipid and that the thick-

fg nessof theproteinand lipidregionsarethesame(13). The helix

f areasreflect the largerareassubtendedby tilted helices. The

fg relativedensitiesgiveninTable4arecorrected for the absence

f oftheF(0,0)termintheFouriermap. Thiswasestimatedto

g be 9.98contours(from thescale ofcontourinterval ofFig.2)

f by using the calculated electron-scattering density ratio of f proteinand lipid of1.30.The densities for helices1and5are g DA significantly weakerthan those forallother features.Thisisalso fg obviousbyinspectionof the^map(Fig. 2). Helix5issmaller than fg helix6or7bothinpeak heightandindiameter,evenwhenit gd isassignedamaximumarea. Helix 1isby farthesmallestin d peak

height

but hasa

bigger

^areathan helix5.Thelower

peak

d

heights

of helices2, 3,and4

compared

tohelices6and7^are gd

compensated

for

by

their

larger

^areasduetothegreater

tilting

gd ofthehelices whenviewed inthree-dimensions (Fig. 2 Inset and ref.2).

Table5 givesthecalculated electron-scatteringdensityof

d theatomsineachhelix, fromthetheoreticalscattering

d DA sections(14)forH,C, N,0,and Satoms.It isclearthathelix f DA Dissignificantlyweaker than the others.This isnot due toany

arbitrary decision about the numberofresidues ineachhelix, butmainlytotheintrinsic aminoacidcompositionof the hel-

d DA ices. Helix Dcontainsfiveglycine residues andnolarge hy- gd DA drophobic side chainsatall,instrikingcontrasttomanyofthe

descriptionofthe others,resulting in alowscattering power peramino acid.The scomputedfrom points at which helixD beginsand ends are alsowell defined, at twolinkscross and itwould be very difficultto make the helixmuchlonger.

Table5. Electron-scatteringdensityineach helix from atomiccomposition

No. ofamino Totalscattering Scatteringpower

Helix acids power peraminoacid

A 24 236.5 9.85

B 25 247.3 9.89

C 25 267.3 10.69

D 24 204.0 8.50

E 24 254.3 10.60

F 25 257.5 10.30

G 27 259.5 9.61

Fromthe helicalassignments fromFig.1, thetotalscatteringpower ofeach helixwascalculated. Theweightsgiventhedifferentatoms were:C=N=0=1,H=0.25, andS= 1.5.

No

G

FIG. 3. The preferred model viewed from the inside of the cell.

Retinal is attachedtoLys-41of helix Bnearthe insidesurface,but theorientationshownisunproven. The buriedchargesaredrawn facingtowards the interior of theprotein.The thickconnectinglines between helices indicate thepathofthepolypeptide.

Toequaltheaveragescatteringof the otherhelices,Dwould

havetobe lengthened by fiveaminoacids; helixDisthe best candidate foraweaklyscatteringhelix. Also lower thanaverage,

however, ishelixA, whichwemadeaslongaspossibleinan

efforttominimizethe differencesⁱⁿthelengthsof eachhelix, which mightbiasourcomparison.Theremainingfive helices have significantly higherscatteringpowercomparedtothese

two, andnonestandsoutasbeingthestrongest.

Wethereforeapplyathirdcriteriontoourmodels. HelixD shouldnotoccupypositionswithstrongfeaturesinthemap.

Models1,17,and27inTable3allsatisfy thiscriterion. Inad- dition, models36-41 areincludedbecausetheyhavehelixD in weak density in spiteof apoorer connectivity orcharge distribution.IfhelixAisalso constrainedtobeinweakdensity, then models 17, 36, and 37are eliminated. Thus, the most

probable model, which has the shortest totallinklength of all 5040, ^is model 1,shownschematically inFig. 3. The choice

seems tobeindependent of the finerdetails ofour criteria;

model1always ranked best whenwetriedawidevarietyof otherweightingschemes for bothconnectivityandcharge.

Thepreferredmodel

Thismodelpossessesanumber of featureswebelieve desirable foramembrane-embeddedprotonpump. Thechargedamino acidsinhelices C, D, F,andGlineupalong single surfaces that face each other towards thecenterof the molecule, possibly forming a hydrophilic proton channel, whereas the hydro- phobic, uncharged faces of these helices face outwardtowards thelipid molecules. Atentative atomicmodel of thisstructure

suggeststhat manyof thepotentialionpairsarestereochemi- cally possibleand thattheyform one or more networkswithin the core of the proteinthrough which protons could jump.The retinal is likely to lie near the surface in theregionbetween helicesB,C, and Gand, therefore, the retinal ring maybe close tothe buried negativecharge ofAsp-96. Undesirablefeatures, such as a crossing-overof two chainsforminglinkregionsbe- tweendifferent helix pairs,also do not occur inmodel1.Finally, the preferred model possesses a topology that would allow folding duringbiosynthesisinto aglobularmolecule. All co- valently connectedhelicesareadjacent directlyaftersynthesis, andnohelix laterintervenesbetweentwopreviously existing helices so that nostructural rearrangement would be necessary inthe later stages ofpolypeptide synthesis.

Thus, we believe modelItobepreferredonseveralgrounds.

Onthe otherhand,allourarguments donotconstituteproof, andit remainsonly themostprobableof a number ofpossible models for thefoldingofthebacteriorhodopsinmoleculeinthe membrane.

B.A.W.is aFellow of theJaneCoffin Childs Memorial Fundfor MedicalResearch. Thisinvestigationwasaidedbyagrantfrom the JaneCoffin Childs Memorial Fund for Medical Research. D.M.E.is gratefultotheJohnSimonGuggenheim Foundation for theirsupport and tothe National ScienceFoundation (PCM 78-10361) and the National Institutes of Health(GM22778and HL14111).

1. Stoeckenius,W., Lozier,R. &Bogomolni,R. A.(1979) Biochim.

Biophys.Acta 505,215-278.

2. Henderson,R. &Unwin,P. N.T.(1975)Nature(London) 257, 28-32.

3. Michel,H.,Oesterhelt, D. &Henderson,R.(1980)Proc.Natl.

Acad.Sci.USA 77,338-342.

4. Ovchinnikov,Y.,Adbulaev,N., Feigira,M.,Kiselev,A. &Lob- anov,N.(1979)FEBSLett. 100,219-224.

5. Gerber, G. E.,Anderegg, R. J., Herlihy, W. C., Gray, C. P., Biemann,K. &Khorana,H.G.(1979)Proc.Natl.Acad.Sci.USA 76,227-231.

6. Walker,J. E.,Came,A.F. &Schmitt,H.(1979)Nature(London) 278,653-654.

7. Henderson,R.,Jubb,J. S. &Whytock,S.(1978)J.Mol. Biol. 123, 259-274.

8. Hayward,S. B.,Grano,D. A.,Glaeser, R. M. & Fisher,K. A.

(1978)Proc.Natl.Acad.Sci. USA 75,4320-4324.

9. Gerber,G.,Gray, C.,Wildernauer,D.&Khorana,H.G.(1977) Proc.Natl.Acad.Sci.USA 74,5426-5430.

10. Bridgen,J.&Walker, I. D.(1976)Biochemistry 15,792-798.

11. King, G. I.,Stoeckenius,W., Crespi,H. &Schoenbom,P. P.(1979) J.Mol. Biol. 130,395-404.

12. Unwin,P.N.T. &Henderson,R.(1975)J.Mol. Biol. 94,425- 440.

13. Henderson,R.(1975)J.Mol. Biol. 93, 125-141.

14. Lonsdale,K.,ed. (1968) International TablesforX-rayCrys- tallography (Kynoch, Birmingham, England), Vol. 3, pp.

216-226.

F