.I. Mol. Bid. (1992) 225, 487494

Membrane Protein Structure Prediction

Hydrophobicity Analysis and the Positive-inside Rule

Gunnar von Heijne

Department of Molecular Biology Karolinska Institute Center for 8tructural Biochemistry

NOVT’Af. S-141 57 Huddinge. Suleden

(Received 14 October 1991; accepted 20 Jan,uary 1992)

A new strategy for predicting the topology of bacterial inner membrane proteins is proposed on the basis of hydrophobicity analysis, automatic generation of a set of possible topologies and ranking of these according to the positive-inside rule. A straightforward implementation with no attempts at optimization predict’s the correct topology for 23 out of 24 inner membrane proteins with experimentally determined topologies, and cborrectly identifies 135 transmembrane segments with only one overprediction.

Keywords: membrane protein: protein struct’urr; prediction

1. Introduction

The prediction of protein structure from sequence is a central problem in molecular biology. For globular proteins, even secondary structure predic- tion is fraught with difficulties and not very reliable (Kabsch & Sander. 1983); however, recent progress in the field of membrane protein assembly and folding has fuelled hopes that a solution to the folding problem may be within reach for t,his impor- tant class of proteins (von Heijne & Manoil, 1990).

The basic structural building-block in plasma membrane proteins of both prokaryotic and eukary- otic cells is the apolar, often slightly amphipathic, transmembrane a-helix (Jennings, 1989; Pattus, 1990), and the arguably most important event during the biogenesis of an integral membrane pro- tein is the insertion of its transmembrane segment(s) into the lipid bilayer. Once this has been accomplished, the basic topology of the molecule is defined. What t’hen remains is for the trans- membrane helices to assemble into a membrane- embedded helix bundle and for the polar segments exposed outside t#he bilayer to fold into their proper tertiary structure. It should be noted that the situa- tion may be very different for outer membrane proteins m Gram negative bacteria, where the cano- nical structural principle is that of a large, antipar- allel P-barrel rather than a helical bundle (Jeanteur et al., 1991; Weiss et al., 1991); this latter type of protein will not’ be considered further here.

Thus. at least for multi-spanning (polytopic) membrane prot.eins with most of their mass buried

within the bilayer in the form of transmembrane helices, the insertion event is decisive. For structure prediction, this means that a good part of the problem is solved if the transmembrane organiza- tion of the chain can be effectively calculated from the amino acid sequence. At the present state of the art. this is best attempted using some method of hydrophobicity analysis, where t,he amino acid sequence is scanned to locate segments rich in apolar residues. Typically, such an analysis will identrfy some segments of such high average hydro- phoblalty that they can be considered “certain” transmembrane segments, but will also produce one or more candidate segments of intermediate average hydrophobicity that cannot be confidently predicted as transmembrane.

Clearly, the ultimate hydrophobicity analysis method would be able to discriminate perfectly between transmembrane and non-membrane segrnents in any protein chain; this might be accom- plished by bett’er algorithms and better hydrophobi- city scales. Here, however, I suggest a complementary approach: to “bootstrap” the output from a standard hydrophobicity analysis by a subsequent step of charge-bias analysts t’hat ranks all possible structures on the basis of their confor- mity with the positive-inside rule; i.e. on the observation that positively charged amino acids are manifold more abundant in cytoplasmic, as compared to periplasmic, segments of integral membrane proteins (von Heijne, 1986; von Heijne & Gavel, 1988; Gavel et aE., 1991) and that) positively charged residues can be used to manipulate the

0022-“836/92/100487-08 $03.00/O 187

c 1992 Academic Press Limited

Figure 1. Sliding window used in the hydrophobicity analysis. For a given window-position, the hydrophobi- city value hi for each residue in the window is multiplied by the corresponding window-weight, wi, and the sum over the window is taken. wi is given by u+ = {i/S for I<i<n-q+l; (n-q+l)/S for (n-q+l)<i<(n+q+l): (%+2--Q/S for (n+q+l)<i<2n+l}. where S = (1 + n)’ -q2 is a normalization factor to make.

2ntl -C~wi=l i=l

In the calculations reported here, n = 10 and q = 5

topology of such proteins (Boyd & Beckwith, 1990: Dalbey, 1990; von Heijne & Manoil, 1990). As shown below, this strategy leads to a considerable improvement in prediction accuracy. at least for bacterial inner membrane proteins.

2. Methods

(a) Collection of sequence data

Twenty-four bacterial inner membrane proteins of known sequence and with experimentally well-character- ized topologies were collected from the literature (Table 1). In most cases, the topology has been determined by fusion protein analysis (Manoil & Beckwith, 1986). Although this method seems to be a very reliable way of obtaining topological information, it, should be kept in mind that the 3-dimensional structure has been deter- mined for only 3 of the proteins listed in Table 1, namely bacteriorhodopsin, and the recation center L and M subunits.

(b) Hydrophobicity analysis

Hydrophobicity analysis was carried out using the GES-scale (Engelman et al., 1986) and a trapezoid sliding window composed of a central, ll-residue rectangular section and 2 flanking wedge-like sections, each 5 residues long (Fig.1). This window-shape was chosen to combine the favorable noise-reduction of a triangular window (Claverie & Daulmiere, 1991) with a physically more realistic rectangular window representing the central apolar part of a lipid bilayer. In this way, one obtains a physically reasonable, soft transition from the apolar interior to the polar surface of the membrane.

From the hydrophobicity profile, candidate trans- membrane segments were extracted automatically by a procedure that 1st identifies the highest peak, records its average hydrophobicity (H), and removes the part of the profile that corresponds to the 21 residues in this segment. This is repeated until either no stretch longer than 20 residues remains, or until no peak higher than the lower cutoff ((H) = 0.5) remains. All peaks with (H) 2 1.0 are considered “certain” transmembrane segments, and all with 95 I(H) < 1.0 are considered putative candidates.

-4 -3 -2 -I 0 I 2

<l-f4

Figure 2. Peak-height distributions derived from hydrophobicity analysis of 92 bacterial inner membrane proteins ( q ), 25 periplasmic Escherichia coli proteins (+). and the 135 transmembrane segments identified in the 24 bacterial inner membrane proteins discussed in t,he t,ext (ml.

These cutoff values were chosen on t,he basis of an initial analysis of the bimodal peak-height distribution for the whole sequence sample obtained with no cutoffs (Fig. 2: see von Heijne, 1986).

Finally, all possible topologies that include the certain transmembrane segments and either include or exclude each of the candidate segments were automatically gener- ated. a,nd the difference in the number of posnively charged amino acid residues (Arg and Lys) between the 2 sides of each structure was calculated. If the penultimate N-terminal amino acid was neither Arg, Lys, Leu. Phe nor He (i.e. residues that block f-Met removal in prokaryotes (Flinta et al., 1986)) an additional positive charge repre- senting the free N-terminal amino group was added to the 1st polar segment. Polar segments longer than 70 residues were not included in the charge-bias calculation, since t,heir content of charged residues does not seem to be dependent on their cytoplasmic or periplasmic location (von Heijne & Gavel, 1988). The possible topologies were then ranked in decreasing order of charge bias, and their orientation was predicted as the one with the more highly charged side facing the cytoplasm. A program called TOP-PRED (TOPology PREDiction program. writt!en in THINK Pascal for Macintosh computers) implementing these steps is available upon request.

3. Results

(a) The positive-inside rule holds for all exposed loops in bacterial inner membrane proteins

The positive-inside rule was originally suggested by the observation that positively charged amino acid residues (Arg and Lys) are much more ahun- dant in cytoplasmic as compared to periplasmic regions of bacterial inner membrane proteins (von Heijne, 1986), and has since been confirmed by a number of experimental as well as further stat’istlcal studies (von Heijne, 1989; Boyd & Beckwith, 1990; Dalbey, 1990; Nilsson & von Heijne, 1990; von Heijne & Manoil, 1990). However, to be able t’o make use of this rule for prediction purposes. it was

Membrane Protein Structure Prediction

Table 1

489

Membrane segments identi$ed by hydrophobicity and charge-bias analysis

I’rotrin Membrane Segment Charge segments categoryt biasf Reference

lkc~trriorhodopsin 22-42 57-77 96116

119-139 151-171 189-209 217-237

C’olA immunit\ 17-37 protein 72-92

104-124 143-163

(‘olE1 immunity G-26 protein 37-57

90-110 (‘yoA 9-29

4C-86 88-108

(‘,td< 16-36 57-77

105-125 143-163 194-214 235-255 265-285 286-306 313-333 342-362 380-400 G-441 456-476 494-514 539-559 590-6 10 611-631

( ‘yd ’ 35-55 71-91 97-l 17

1455165 184-204

19-39 40-60 78-98 11-31 36-56 85-105

107-127 133-153 160-180 208-228 231-2.51 264-284

CyW558 12-32 6G80 93-l 13

137-1,57 178-198

‘i-27 45-65 78-98

102~12% 145p I65 167YlRi 219-239 263- 283 291L311 315.-335 34%369 382-402

Phage Ml3 prowat 5-25 protein 44-64

C+

C+ t, +

C+ C+ C+

t+

C+

c+ c+ c+ c+

t+

c+ c+ c+

C+

c+ c+ c+

c+ c+ c+ t,-

c+ c+ Cf c+ c+

C+ c+ t-

C+

C+

C+ C+ c+ C+ C+

C+ C+ C+

c+ c+ C+ C+ C+ (‘+

C+ C+

C-t

C+ c+ Cf

C+ c + C+ c+ C+

C’ +

C+ (‘+ c+ C+

C+ tt

C+ Cf Cf Cf

-5, +7* Song & (Iramer ( 1991)

-IS* +6. -6,‘-4

Chepuri & Gennis (1990)

-2. 0. +3 -5*

+4*

+7*

+6*

+ :3*

+ lo*

+4*

+6*

Henderson et al. (1990)

Geli ct al. (1989)

Chepuri & Gennis (1990)

Chepuri & Gennis (1990)

(‘hepuri & Gennis (1990)

C’hepuri & Gennis (1990)

Fridkn (1989)

Kuhn et al. (1986)

+9, +17* (klamia & Manoil (1990)

Light-harvesting

complex LH 1 Light-harvesting

complex LH:! Keac%ion crntrr

complex L-subunit

I’hpT

PI* Rohrw & Kuhn (1990)

OX. -2 Moore $2 Miura (1987)

-I. +9* Munoa rf (11. (I!,!,])

+4* Ihws (19%)

+1* Ihws ( I9X:i)

+x*. t-" Michel rt /I/. (1986)

+9*. +:I

+fi* Schatz rt al. ( 19X9)

+?A*. +x .Ikiyarna & Ito (1987)

+ 1:3*

+27* Lloyd & Kadnrr (1990)

Membrane Protein Structure Prediction 491

Table 1 C’ontd.

Membrane Segment Charge Protein segments categoryt bias: Reference

299-319 (‘+ 38tk34ti c- + 351-371 C+ 405-425 C’ + 427447 (‘+

t’trr~F: 11-31 Cf -2* Drckers-Hebestreit, C 59-79 c + Altendorf (1986)

tr. certain; t, tentative; +, a membrane segment present in the known structure; -, a membrane segment not present in the known structure.

$The t,otai difference in the number of Arg+I,ys residues between thr 2 sides of the structure for all po&hlr topologies generated as described in Methods are given, with the value corresponding to the kuown topology marked with an asterisk. Positive values correspond to topologies where the S terminus is predicted to be cytoplasmic. negative values to topologies predicted to have the X t~rrminu~

import,ant to establish that it, holds for all segments of a transmembrane protein, and not just for the N-terminal or (‘-t,erminal ends, say.

To this end. 24 bacterial inner membrane proteins of known sequence and with experimentally mapped topologies were analyzed (see Table 1). For each protein, the number of Arg and Lys residues in each cytoplasmic and periplasmic segment was recorded, and t,hr counts were pooled according to the position of the segment (counting from the r\; terminus) and its cytoplasmic or periplasmic loca- tion. As shown in Figure 3: the bias in the distribu- tion of’ Arg+ Lys is equally strong throughout, the sequences; i.e. all cytoplasmic loops have a similarly high content (- IS”/,) whereas all periplasmic loops have a low content (-SOL). It is thus clear that all parts of a multi-spanning inner membrane protein [*onform to the positive-inside rule.

(b) Thr positiwinxide rule can be used to improrr

thw prediction of transmembrane sepents

Since the positive-inside rule apparently applies to any part of a bacterial inner membrane protein. it should provide a strong crit,erion for testing whether a putative transmembrane topology is likely to be correct or not. One is thus led t’o consider the following procedure. First, make a list of all possible transmembrane segments in the pro- t)ein based on some standard method of hydrophobi- city analysis and a liberal cutoff criterion; second. decide which of these candidat’es are certain and

which must be regarded as tentative based on a

more stringent cutoff criterion; third, construct’ all possible transmembrane st’ructures always including thr cert,ain candidates but either including or excluding each of the tentative segment#s: and fourth, rank these structures according to their degree of bias in the dist’ribution of positively cshargrd residues. The rationale behind the last st,ep. is that any structure with an incorrect number of transmembrane segments would have one or more polar domains placed on the wrong side of the

membrane, and hence would be likely to have a smaller charge bias than the corre& structure.

As an example. consider the 1,acY protein. which is known to have a cytoplasmic N terminus followed by 12 transmembrane segments (Calamia B Manoil, 1990). The hgdrophobicity plot shown in Figure 4(a). predicts 11 certain segments and one putative candidate. Thus, two topologies can be generated, one excluding and one including the putative candi- date (Fig. 4(b)). The former has a bias in the distri- bution of positively charged residues of nine, the latt’er of 17, and it is clear from the Figure that the C-terminal part of the second. correct model has a much better correspondence with the positive-inside rule.

To test this strategy more thoroughly, hydropho- bicity analysis was carried out on the 24 proteins listed in Table 1 as described in -Methods, with a lower cutoff set to (H) = 0.5 and a higher cutoff set

Segment positlon

Figure 3. Mean number of Lys+ hrg in v)-toplasmic (stippled bars) and periplasmic (open bars) polar segments as a function of the position of the segment (1st cyto- plasmic segment. 2nd cytoplasmic segment. rt,c., counting from the N terminus) in a sample of 24 bac*twial inner membra,ne prot,eins with known topology.

2

I

A 0 $ v -I u v

1 li

U I

-2

-3

-41 I I 1

0 100 200

Posltion

(a)

300 400

A+ = 9

A+=17

2 3 5 4 2 2 2

(b)

Figure 4. (a) Hydrophobicity plot for the SecY protein. The upper and lower cutoffs are marked. A tentative transmembrane segment with a mean hydrophobicit falling between the 2 cutoffs is marked by an arrow. (b) Two possible topologies for the SecY protein based on the hydrophobicity plot. The putative transmembrane segment is shown in black. The number of Arg+Lys residues is shown next to each polar segment. IVote that t)hr correct, alternative (bottom, including the putative transmembrane segment) has a much higher charge-bias than the incorrect. one.

to (H) = 1.0 (i.e. t’he criterion for tentative trans- membrane segments was 0.5 _< (H) < 1.0). Fourteen prot’eins were predicted to contain only certain transmembrane segments with no additional tenta- tive candidat,es; all of these turned out to be correct and their orientation in the membrane was correctly predicted by the positive-inside rule in all cases. For the remaining ten proteins, the correct structure had the highest charge bias (highest difference hetween the number of periplasmic and cytoplasmic Arg+ Lys) in eight cases (Table 1). An extra. incor- rect certain transmembrane segment ((H) = 1.06) was predicted for the SecY protein; closer inspection revealed that this resulted from a “shoulder” on a higher, somewhat broad peak. Inclusion of this segment gave a topology with a charge bias of eight. and its exclusion gave a charge bias of 26. Thus.

Thus. using the positive-inside rule as :I J(‘r(‘en. t)he transmcmhranr topologies of 22 (or. wit)h some gootl-will. 23) out of 21 proteins were (*orrec+lJ predicted. atld all of the 135 transrnrm t)ranc> segments wTt‘re identified with onl?- oru’ ovtq)rerlicn- tion (I,tlf)).

(c) Possibilities for improvements in thv hydrophobicity scnlrCs

In the a,bove analysis. the standard (JES hydra- phobicity scale (Engelman rt nl.. 1986) based rriti- matrly on physico-chemical caonsiderations was used and no at,tempt at, optimization was made. With this scale and through application of t,he positivr- inside rule. 135 t’ransmembrnne segments were correctI\- identified in t.hr 24 proteins analvzrd. From t,hese data, we derived a new. statistically based scale by comparing the amino acid frequen- cies fi M in the cent,ral I 1 -residue stretch of t,ht> transmembrane segments with the amino acid frequencies f;” m the non-meml)raneous srgrnent#s (taking the membrane-embedded segments to hf. 17 residues long): Ai = In(fiM/fiA), Table 2. \Vith a11 upper cutoff of O$.i and a lower of - 0. I and t hts

Table 2 Hydrophobicity scak bused on 735 transmemhrane

segments from 24 bacterial inner ,membrane proteins

Memhranr Protein Strwtuw Prediction 493

same sliding window as above, this scale seems t’o perform slightfly better than the original GES-scale, with 18 out of the 24 proteins (including SecY) now being predicted to contain only certain trans- membrane segments (a.11 correct). Again. ranking based on the positive-inside rule produced the correcst topolop~ for iLlI but one (I,ep) of the remaining prot,elns.

4. Discussion

Two important conclusions can be drawn from t.he data present,ed here: first., the posit’ive-inside rule applies with equal strength to all parts of polytopic bacxterial transmembrane proteins, and. second. the positive-inside rule can very efficiently be used to rank possible predicted transmembranc topologies and thus in a sense t’o bootstrap the st.andard hydrophobicit? analysis.

The first, c*onclusion has certain mechanistic impli- rations, and suggests a mode of assembly of bacterial membrane proteins where locally formed “helical hairpins” composed of two neighboring hydrophobic- helices and a connecting loop containing few positively charged residues insert into the membrane more or less independently of t*he remaining parts of the polypeptide chain (Engelman & Steitz, 1981; von Heijne, 1986). A similar model is also suggested by recent experi- mental data on the t,opological consequences of deleting one or more transmembrane segment. (Yatnane et al., 1990; Kibi et al.. 1991: Ehrmann 8: Krckwit’h. 1991: McGovern et al.. 1991). This is in contrast t.0 it “linear” insertion model, where the most S-terminal transmembrane segment(s) insert first. and the rest of the chain simply follows the topological dictate of this first insertion (Vessels & Spiess. 1988: Hartmann et al., 1989). It may be. though. t’hat the latter model more accurately describes membrane protein assembly into the endo- plasmic reticulum membrane, which is t’hought to be obligatlorily co-translational and hence likely to proceed in an K to C-terminal direction (High et al.. 1991). It should be noted in this regard that the bias in i he dist,ribnt.ion of positively charged residues. although clearly t,here, is nevertheless less marked in eukary-otic plasma membrane proteins (von Heijne & Gavel. 198X).

For predi&m purposes, a combination of hydro- phobicity analysis and charge-bias analysis clearl> improves the reault,s. almost to the point where veq little experiment.al work would have t.o be done to coonfirm a predicted t,opologg. In this study. the topology of 9.5(& of the prot.eins was correctly predi&ed I)!- a fully automatic method. even when the initial hydrophobicity analysis method had not been consc~iou+,- optimized. Thus, the topology of bacterial inner membrane proteins seems to be rather straightforward to predict direc+ from their sequence. setting the stage for attempts t,o develop methods for modeling their fIlli three-dimensional structure.

This work was supported hv grants from the Swedish Natural Sciences Research &unc:il and thr Bwrdish Board for Technical Development.

References

Akiyama. Y. & Ito. K. (1987). Topology analysis of’ the SrcY protein. an integral membrane protrin involved in protein rlxport in Eschrrichitr roli. h’.MBO ,I. 6. 3465 -3470.

Bibi. IX.. Verner. (i.. (‘hang. (‘. 1.. & Kabwk. H. R. (1991). Organization and stability of a polytopic membrane protein-deletion analysis of’ the Iac+ose permease of EarhfwZchiu coli. Pror Sat. ;Icrrd. Sri.. l’.S.A 88. i2’il-727s.

Rilgin. X.. Lee. .I. 1.. Zhu. H.. J)alt)r!-. Ii. & van Heijnr, (;. (1990). Mapping of c.atal~ticalI~ important domains in Eschrrichin roli leadel. prl~t~tiase, EURO ,J. 9. %il’i m”iP?.

Hoyti. I). d l+ckwith. ,I. (1990). The role of c,harped amino acids in the localieatiorl of ~rc*rt~trd and memhrane proteins. Cell. 62. 1031 IOXX.

(‘alamia. .J. k Manoil. (‘. (lR!)O). ],a(. lwrmcas(z of Esrhvrichicz co/i&topology and srquen~e elements promoting membrane insertion I’rvc .V///. .-Iwcl. Sci.. l‘.S.d. 87. -4937-4941.

(‘hrpuri. \‘. & (irnnis, R. 13. (1990). Tbr use of gene fusions to d&ermine the topology of all of the sub- units of t,hr cytochrome o terminal oxidase wmplex of Ewherichin coli. J. Biol. C’h~m. 265. 12978P1ZM6.

(‘la.verir. .1.-M. & Daulmiere, (‘. (1991). Smoot.hing profiles with sliding windows: better to wear A hat! /‘.-1f?If)S 7. 113-115.

I)albe>-. R,. E. ( 1990). Positively c*hargtad wsidues are important determinants of tncmbranr proWin topo- logy. Twnda Biochem. Sci. 15. %54-?.5i.

T)rc~kers-H~bestreit. (:. & r\Itendorf. K. (1986). r\ccrssibilit> of F, suhunit)s from KsrhPrirhirr coli ATI- synthase. Ew. ./. HiochPv. 161. ~~T,-~Sl.

Drrna. (:. (l!P&). Structure and func+ionnl organization of light-harvesting complexes and phot.oc~hrnricaI reaction c,enters in membranes of l~hototrophic~ bacteria. N icrobiol. Hev. 49. Z-70.

Eckert. R. & Reck, C. F. (1989). Topology of the trans- poson TnlO-encoded tetracycline resistance protein within the inner membrane of Escherichia coli. J Bid Chum. 264. 11633-11670.

Ehrmann. 1cl. & Reckwith. *J. (1991). I’ropw insertion of a complex mt\mhranr prot.rin in the absrnw 01‘ its amino-terminal export. signal. .I. /1;0/ (‘him. 266. 165‘3W 1651’i. 1. I.<

Engelman. I). 11 8: St,ritz. T. A. (1981 i. The spontaneous insertion of proteins int,o and across membranes: The helical hairpin hypothesis. (‘~11. 23. -411-4Z

E:ngcelman. I). 11.. Steitz. T. A. Xr C:oldtnan. ,1. (19X6). Identifying nonpolar transbilayrr hrlicbrh in amino wid srquenc~es of mrmhrane proteins. .~l nrclc. Rw. Biophys. Bicyhys. C’h.Pm. 15. 4”l -8X1.

Flinta. t‘.. Pwsson, R.. ,Jiirnvall. H. & \-on Hrijne. (G. (1986). Sequence determinants of cytosolic X-terminal protein processing!. L:crr. .J. Hiockrm. 154. 193 196.

Fridin. H. (1989). Studies on the c~ytoc~hrornr B-558 sub- unit of HacilZ?Ls su6tilis succ~inate:yuino~~r oxidorr- duct,ase. Phi) t’hesis. Lund IrniversitJ-. C;wedrn.

Froshauer. S.. Green. G. X.. Boyd. I).. Mc(:overn, K. & Rrckw-ith. .I. (1988). (>metic ana.lysis of t.hr membrane insertion and t.opolog?- of MaIF. a cyto-

plasmic membrane protein of Exherichicl co/i. J, .Ilo/ Riol. 200. WI 511.

(:avc~l. E’.. Steppuhn. J . Herrmann, R. & van Heijntx. ( :. (1991). The posit,&-inside rule applies to thvlakoit] membrane proeeins. FEBA’ L&w. 282, 11 46.

(ieli. \‘.. Katy. I).. l’attus. F. & Lazdunski. (‘. (1989). Topology and fun&ion of the integral membrane protein conferring immunity to colicin A. Tklol.

Microbid. 3. 679. 687. Hartmann, E., Rapoport, T. A. 8r Lodish, H. F. (1989).

I’rrdic*ting t)hr orient’ation of eukaryotic membrane proteins. Pror. ,Vat. A cad. sci., 1’JS.d. 86. r,786X790.

Henderson. R.. Baldwin, J. M.. (‘eska, T. A.. Zemlin. F.. Beckmann. E. & Downing, K. H. (1990). 4 model for the structure of bacteriorhodopsin based on high rrsolut,ion electron cqo-mivroscsopy. -1. No/. Hiol. 213. 899-929.

High. S.. Flint. X. & Dobberstein, K. (1991). Rtyuirrments for the membrane insertion of signal- anchor type proteins. J. Call. Riol. 113. 2.5 34.

Jeant,rur. I).. Lakey. J. H. & Pattus, F. (1991). The ba&erial porin superfamily-sequence alignment and structure prediction. Mol. Microhiol. 5, %]53%2164.

.lennings, M. I,. (1989). Topography of membrane pro- teins. A~anu. Krv. Biochem. 58, 999.-1oP7.

Ka1~sc.h. M’. & Sander. C. (1983). How good are predic- tions of’ protein secondary structure? FERS Ltittux. 155. 179-182.

Kuhn. A.. Kreil. (:. & Wickner. W. (1986). Both hydro- phobic~ domains of Ml3 provoat are required to initiate membrane insertion. ENRO. .I. 5.

3681~ 3685. T~loytl. A. I>. & Kadnrr. R. ,J. (1990). Topology of the

h’sch~richia roli &pT sugar-phosphate transporter analyzed by using TnphoA fusions. ./. Bactrriol. 172. I6H8~1693.

Manoil. (‘. & Beckwith, J. (1986). A genetic approach to analyzing membrane protein topology. Science. 233. 1403-140x.

Mv(:ovrrn. K.. Ehrmann. M. $ Heekwith. .J. (1991). I)ec*oding signals for membrane protein assemhll using alkaline phosphatase fusions. EA’RO ./. 10. 2773 %7X”.

Michrl. H ., IVryer. K. A.. Gruenberg, H., Dunger, I.. Orsterhelt, D. B Lottspeich, F. (1986). The ‘light’ and .medium’ subunits of the photosynthetic reaca- tion ventre from tlhodopseudomonacs kdis: isolation of the genes. nucleotide and amino avid seyuenc.r. F;MKO .I. 5. 1149~1158.

l\loorv. K. E:. R Miura. S. (19X71. .\ small hytlrc~l~hohit~ domain anc*hors Ieadt?r pr])t,ittascl t,o the (.yto])lastllic. tnrmbranr of Escl~~rirhirr ~oli. J Iiiol C’he,~/ 262. 8XO6~~881:1.

Mrmoa. I’. .I Miller. K. 1V.. ISeers. It.. (:raharrl. Xl. & \$‘I]. H ( ‘. ( 199 I ). Membrane topology of /fschrrichirr co/;

prolipoprot,ein signal peptidase (signal pt’])tidasr II ). .I. Hid. C’hPm. 266. 17667 1767”.

Nilssolt. 1. &I. 8r van Heijne. (;. (1990). Fine-tuning thr topology of a polytopic, mrmbrane ])rotrin. l<ol~~ of positix-taly and nrgativrly c.hargetl residuc~s. c ‘r/l. 62. 1135 1111.

I’attus. F. (1990). Mrmbranr prot,eitr struc%urv t ‘err. Opi,/ion (‘41. Biol. 2, 68lm 68.Y

Rohrer. *I. 8: Kuhn. A. (1990). The furtvtion of’ a leader prptidr in transloczat)ing caharged amino acsyl resitluvs across a mrmbranr. Srirwr. 250, 141 8 1121

Schatz. 1’. .] liiggs. 1’. I).. .Jaq. L\.. Fath. hl. .I. & Brvkwith. ,I. (1989). Thr sr~E gene t~nc~odvs an integral membrane protein required for protein export in Eschrrichicr r~ol;. (:etrP.s lk1dop 3.

IO3;i 1Ol-c. Song, H. 1’. & (‘ranter. L\‘. A. (1991). ,Ilrmbranr

Topography of’ (‘oNI gent’ producnts- the immrmit~ protein .I. IZnctrriol. 173, 835-2948.

van Hrijntx. (:. (1986). ‘I? IP distribution of positivt)l? caharged residues in bacterial inner mc~rnbratx~ ]m,~ teins (*orrelates with thtb tram-membranr topology. ENI~O ./. 5, 3021 -3027.

van Heijne. (:. (1989). C’ontrol of topology ant1 moth. of assrtnbl?. of a polytopiv membrane protein by posi- tivrl! charged rrsidurs. Srrt~r~ ~lonrlor/ 1, 341. 15tibkr,x.

van Hri,jncs. (:. & (:aveI, Y. (19%). ‘I’c,])ogrnic. signals in integral mC~mt)rattt~ proteins. E/c/.. ./. fjioChr,rr/. 174. 671 6iX.

van Heijnr. (:. & Manoil. C’. (1990). Membrane proteins --from sequence to structure. Protein En,g. 4.

109- I I”. MTeiss. M. S.. Kreuscsh, A.. Schiltz. E., Nested. I’.. \Yt~lt~r.

iv.. \Vrckrssrr. J. & Schulz. (:. E. (1991). ‘l’hv strucm turn of porin from Rl~odobfrct~~r rupxulatcf at I .X,i resolution. PEES Lrtters. 280. 379 382.

Vessels, H. 1’. & Spiess. M. (1988). Insertion of a multi- spanning membrane protein occurs sequentially and requires only one signal sequence. (‘ell, 55, 61- 70.

Yamant,. K.. Akivama. \‘.. Ito. K. & Mizushim:i. S. (1990). ;\ positively vhargrd regiotl is a dett~tmirtatrt of the orientat,ion of c~ytoplastnic membrane lnvteins in Esrhrrichin coli. ./. Niol. C’hun. 265. “1 I66 11 171.

Edifrd by K. IT. J!!atthru~.s