Home Page Local Home Page Text Assignments Schedule Links

Bioinformatics for protein sequence, structure and function

Gunnar von Heijne

Department of Biochemistry

Stockholm University

S-106 91 Stockholm, Sweden

Fax: Int+46-8-15 36 79

E-mail: gunnar@dbb.su.se

The general structural principles evident in membrane proteins primarily reflect the need to provide a large external hydrophobic lipid interface, but also depend on requirements imposed by the mechanisms whereby proteins are inserted into lipid bilayers. In this review, an attempt is made to discuss membrane protein structure in the context of what is known about membrane protein assembly. Structure prediction techniques are briefly reviewed, and the interplay between structural information gleaned from various techniques based on site-directed mutagenesis and theoretical prediction method is emphasized.

Membrane proteins come in a variety of sizes and shapes, though the basic architectural principles apparent from the few available 3D structures are far less diverse than those of the globular proteins. This is a consequence of the requirement that membrane proteins either must bind to one leaflet of, or completely span a lipid bilayer while at the same time maintaining contact with the surrounding aqueous medium. Peripheral binding to one side of the bilayer can be mediated by amphiphilic a-helices oriented parallel to the bilayer surface and membrane-spanning structures can be built either from �-barrels or from bundles of transmembrane a-helices. Large extra-membraneous domains do not seem to be much influenced by their attachment to the bilayer, and can essentially be viewed as tethered globular proteins.

The structural differences between membrane and globular proteins do not only result from the different physico-chemical requirements imposed by their different environments, but also reflect fundamental differences in the mechanisms responsible for their folding and assembly. While globular proteins encounter only one kind of environment from the time of their synthesis until they are degraded in old age, membrane proteins are first exposed to aqueous surroundings and only later gain access to the milieu for which they are designed. The question of how the controlled integration of a membrane protein into the lipid bilayer takes place is still not fully worked out, and there are certainly aspects of membrane protein structure that will only be fully appreciated once this has been accomplished.

Both the structural and biogenetic aspects of membrane proteins have been reviewed before (Cowan and Rosenbusch, 1994; Lemmon and Engelman, 1994; von Heijne, 1994a; Manoil and Traxler, 1995; Popot and Saraste, 1995; Reithmeier, 1995; Spiess, 1995), and my ambition here is mainly to provide an update of what has happened in the field during the last few years. This includes some highly interesting new structures and some pioneering studies on helix-helix packing in a non-aqueous environment. The basic aspects of the initial step of insertion into the lipid bilayer are becoming increasingly clear, and there have also been some definite advances in the area of membrane protein structure prediction. For fuller treatments of many facets of the membrane protein assembly problem, the reader is referred to a recent book (von Heijne, 1996).

Membrane proteins are not only tricky to handle biochemically, they are also notoriously difficult to overexpress (Grisshammer and Tate, 1995) and seldom yield high quality 2D or 3D crystals. So far, all the known structures are of proteins that can be isolated in sufficient quantities from natural sources, though recent work on overexpression in E. coli (Miroux and Walker, 1996) and other systems (Tucker and Grisshammer, 1996) look promising.

Good 3D crystals seem to require the presence of rather extensive globular domains protruding from the membrane that can form the necessary crystal contacts. In the case of the bacterial cytochrome c oxidase, the globular domain was artificially enlarged by binding a monoclonal Fv fragment to the protein complex prior to crystallization (Iwata et al., 1995). The formation of 2D crystals, on the other hand, can be induced by protein-protein contacts in the plane of the membrane (K�hlbrandt, 1992), and such crystals may thus be easier to obtain when the globular domains are small. However, structure determination by electron crystallography of 2D crystals rarely give sufficient resolution for structure modeling on an atomic scale.

In this section, all the known high-resolution membrane protein structures as of August, 1996, will be briefly reviewed. The focus will be on structural rather than functional aspects.

Bacteriorhodopsin: the tour de force of electron crystallography

Bacteriorhodopsin, a light-driven proton pump from Halobacterium halobium (Lanyi, 1995), was the first integral membrane protein to be shown to consist of a bundle of transmembrane helices oriented roughly perpendicular to the membrane plane (Henderson and Unwin, 1975). 15 years later, the structure had been determined to sufficient resolution to allow the bulky aromatic side chains of the transmembrane helices to be built into the electron density (Henderson et al., 1990), and another 6 years of data collection and analysis finally has produced a density map where most side chains, both in the helices and the connecting loops, can be fitted (Grigorieff et al., 1996), Fig. 1A. Bacteriorhodopsin is not only one of the best structurally and functionally characterized integral membrane proteins, but has also served as the test-bed for the development of both hardware and software for electron crystallography.

The transmembrane helices are composed mainly of hydrophobic residues. A few charged residues are buried within the structure, serving important functions in the proton-relay system, Fig. 1B. Bacteriorhodopsin has been widely used as a starting structure for modeling many medically important G-protein coupled receptors, though it now appears that the relative disposition of the helices in these eukaryotic so-called 7TM proteins is sufficiently different from that in bacteriorhodopsin to make such modeling very difficult (Schertler et al., 1993).

The structure of the photosynthetic reaction center from Rhodopseudomonas viridis was solved by X-ray crystallography in 1995 (Deisenhofer et al., 1995), followed 2 years later by the closely related one from Rhodobacter sphaeroides (Allen et al., 1997; Yeates et al., 1997). The structure is considerably larger than bacteriorhodopsin, with a total of three membrane-bound and one peripheral protein subunits, 11 transmembrane helices, four bacteriochlorophylls, two bacteriopheophytins, two quinones, and a nonheme iron. The heart of the protein is the electron-conducting chain of co-factors held in place by the protein scaffold, Fig. 2A.

Since this was the first high resolution structure of a membrane protein available, it provided the first clear indications of how transmembrane helices pack (Rees et al., 2001a; Rees et al., 2001b). The atomic packing density in the interior of the protein is similar to what is found in globular proteins, as is the average hydrophobicity of the internal residues. Residues exposed to the lipid environment are more hydrophobic than the internal ones, just the opposite of what is the case for globular proteins. The exposed surface residues are more variable between homologous proteins from different species than are the internal residues.

Many of the transmembrane helices protrude quite a distance from the bilayer, and consequently have a central hydrophobic part flanked by more polar residues. Interestingly, Tyr and Trp residues tend to be concentrated near the presumed position of the lipid headgroup region Fig. 2B.

Another protein involved in photosynthesis, the plant light-harvesting chlorophyll a/b-protein complex, has been solved to high resolution, in this case by electron crystallography (K�hlbrandt and Wang, 1991; K�hlbrandt et al., 1994). This is a rather small protein with three transmembrane helices that provide binding sites for a large number of chlorophylls and two carotenoids, Fig. 3. Two of the helices form a pair and are highly tilted (320) relative to the membrane normal, whereas the third helix makes no direct contacts with the other two. The helix pair appears to be stabilized by two salt-bridges at one end, and by the two carotenoids that serve as crossbraces, almost spanning the entire bilayer.

The structure of the light-harvesting complex II, a part of the photosynthetic antennae, from Rhodospeudomonas acidophila was recently solved by X-ray crystallography (McDermott et al., 1995). This remarkable ring structure is composed of 9 a- and 9 �-subunits, each only contributing a single transmembrane helix, and 27 chlorophylls, Fig. 4. The a-subunits form an inner ring, separated from the outer ring of �-subunits by a layer of chlorophylls. There is thus very little contact between the a- and �-subunits, and the chlorophylls seem to be as important for keeping the structure together as are the proteins. The tightly interacting chlorophylls provide an ideal structure for charge delocalization and serve to funnel the electron resulting from light excitation towards the reaction center.

Cytochrome c oxidase: the biggest yet

Given the paucity of membrane protein structures, the almost simultaneous appearance of the structures of the bacterial and mitochondrial cytochrome c oxidase complexes was a delightful coincidence (Iwata et al., 1995; Tsukihara et al., 1995; Tsukihara et al., 1996). The bacterial complex contains three central subunits, while the mitochondrial one has no less than 13 subunits, 10 of which have one or more transmembrane helix, pushing the total number of helices to 28, Fig. 5. In addition, the two catalytically important Cu-sites, two hemes, and 8 phospholipids have been located in the structure.

Analysis of the structure of the mitochondrial oxidase has suggested a rather detailed consensus structure that seems to be shared between the cytochrome c oxidase, the bacterial photosynthetic reaction center, and bacteriorhodopsin, and that may thus be generally valid for helix bundle membrane proteins (Elofsson et al., 1996): a central, helical region of about 20 � rich in aliphatic residues and Phe, an "aromatic belt" composed of Trp and Tyr similar to but less prominent than that seen in the �-barrel porin structures (see next section), a region near the helix ends with amidated and charged residues, and helix breaking Pro residues just outside the helices.

The porins form passive diffusion pores in the outer membrane of Gram-negative bacteria, and are built on a �-barrel architecture. The structures of five porins are known: those from Rhodobacter capsulatus (Weiss et al., 1991b), Rhodopseudomonas blastica (Kreusch et al., 1994), and the Escherichia coli OmpF, PhoE, and LamB porins (Cowan et al., 1992; Schirmer et al., 1995). In all cases, the structure is composed of a large anti-parallel �-barrel with short loops facing the periplasm and larger loops protruding outside the outer membrane, Fig. 6A. Substrate specificity and translocation kinetics depend on a segment of the chain that loops into the barrel, Fig. 6B, thus modulating its properties (Weiss et al., 1991a; Schirmer et al., 1995).

Not surprisingly, the outer surface of the barrel is composed of hydrophobic residues, while residues on the inside may be both hydrophobic and hydrophilic. Since the membrane-spanning segments of the chain are �-strands, this means that only every second residue is facing the lipids and hence needs to be hydrophobic. The strands in porin from R. capsulatus are tilted by about 300- 600 relative to the membrane normal, and are 6-17 residues long (Weiss et al., 1991a). The long hydrophobic segments typical of the helix bundle membrane proteins are absent from the porins, and in fact such segments, if introduced by molecular genetic techniques into an outer membrane protein, become trapped in the inner membrane during translocation (MacIntyre et al., 1998), thus preventing assembly into the outer membrane.

A notable feature of the porin structures are the belts of aromatic residues that point out towards the lipid headgroup region on either side of the bilayer, Fig. 6A. The functional significance of these aromatic belts is not known, though it has been speculated that they may serve to keep the molecule in a fixed orientation relative to the bilayer (Weiss et al., 1991a). As noted above, a similar clustering of aromatic residues near the lipid headgroup region, Trp and Tyr in particular, is seen also in the helix bundles proteins.

Prostaglandin H2 synthase binds to the lumenal leaflet of the endoplasmic reticulum (ER) membrane, and is thus classified as a peripheral membrane protein. In contrast to the peripherally bound globular subunits of, e.g., the photosynthetic reaction center and cytochrome c oxidase, its membrane binding is not mediated by other proteins but is the result of direct protein-lipid interactions.

The structure of prostaglandin synthase strongly suggests that binding to the bilayer is through four amphiphilic a-helices which, in contrast to most helices in globular proteins, has a series of solvent-exposed hydrophobic residues flanked by basic residues (Picot and Garavito, 1994; Picot et al., 1994), Fig. 7. These hydrophobic residues are postulated to insert into the bilayer, while the flanking charged residues may interact with the phospholipid headgroups. The amphiphilic helices are thus thought to lie more or less flat on the membrane surface, a mode of binding that has also been seen in some small lytic peptides such as magainin (Bechinger et al., 1993).

An interesting class of integral membrane proteins are the pore-forming toxins (Parker and Pattus, 1993). These molecules manage the remarkable feat of first folding into a globular, water-soluble structure that can convert into a membrane-bound form with ion channel activity upon exposure to a lipid bilayer under the proper conditions.

The known structures are all of the water-soluble form, and are only suggestive of how the membrane-inserted structure may look. Nevertheless, there are good biophysical and biochemical evidence for both helix bundle and �-barrel structures (Parker and Pattus, 1993; Parker et al., 1994), though just how similar these structures may be to the membrane proteins described above is not known. As an example, the structure of the soluble form of colicin A (Parker et al., 2001) is shown in Fig. 8. The molecule consists of two layers of a-helices surrounding a central pair of very hydrophobic helices - a "helical hairpin". The initial binding to the lipid bilayer is thought to involve surface-exposed basic residues near the tight turn between the two hydrophobic helices (Lakey et al., 1994). Upon binding, the whole structure converts into a molten-globule like state (van der Goot et al., 1991) allowing the central helical hairpin to associate with the lipids (Lakey et al., 1993). How the ion channel is formed after these initial events is not known.

The difficulties encountered with high-resolution structure determination of membrane proteins has prompted the development of alternative approaches based mainly on molecular genetics, sometimes coupled with various biophysical techniques. The basic idea is to obtain bits and pieces of structural information related to helix-helix packing that can then be used to constrain model-building exercises.

Glycophorin A (GpA) is a single-spanning membrane protein found on red blood cells. It forms unusually stable homo-dimers that will not dissociate even in SDS, and that can thus be readily identified by denaturing SDS-PAGE. Competition of dimer formation with synthetic peptides (Lemmon et al., 1992a) and the construction of various chimeric proteins have shown that the transmembrane helix is both necessary and sufficient for dimerization.

These properties have made it possible to study helix packing in the GpA dimer by saturation and site-directed mutagenesis. In a first series of experiments, some 280 random mutations throughout the transmembrane helix were analyzed, and it was found that only 7 residues were critical for dimer formation, Fig. 9A. These residues all line up on one side of the helix, suggesting that the helix dimer is a right-handed supercoil (Lemmon et al., 1992b; Treutlein et al., 1992), Fig. 9B. Further studies demonstrated that all the other residues in the helix could be simultaneously changed to leucine with minimal effects on dimer formation (Lemmon et al., 1994).

The central feature of the helix-helix interface is a Gly-Val-Xxx-Xxx-Gly-Val motif in each helix. Apparently, the glycines allow a very close approach between the two helices, while the valines provide a large contact area on either side of the Gly-Gly contact. The N-terminal Ile-Leu pair is somewhat less critical, and can be replace by certain other combinations of hydrophobic residues (Mingarro et al., 1996).

The initial mutational analysis of the GpA dimer was based on a very large number of random mutations, and it has recently been shown that a more rapid way to identify the critical interface residues is to make a first screen using alanine insertions rather than substitutions (Mingarro et al., 1996). The basis for this approach is that the insertion of a residue into a transmembrane helix will displace the residues on the N-terminal side of the insertion by 1000 relative to those on the C-terminal side, thus effectively disrupting helix-helix packing interfaces. Insertions that have no effect on GpA dimerization will thus allow the ends of the interface motif to be identified before saturation mutagenesis is applied. An Ala-insertion scan of helix III in lactose permease (see next section) suggests that insertion mutagenesis may be a generally efficient way to obtain a first, rough idea of where the structurally and functionally critical residues reside in transmembrane helices (Braun et al., 1996).

Mutational analysis of the homo-pentameric channel protein phospholamban that forms SDS-resistant pentamers has recently been carried out (Arkin et al., 1994; Adams et al., 1995; Arkin et al., 1995; Arkin et al., 1996; Ludlam et al., 1996; Simmerman et al., 1996), much along the lines established for GpA, and a 3D model has been proposed, Fig. 10. In this case, the packing in the helix-helix interfaces is reminiscent of the well-characterized leucine-zipper motif found in globular proteins.

Of all membrane proteins that have been subjected to mutational analysis, lactose permease (LacY) from E. coli has no competitor (Kaback et al., 1994; Kaback, 1996). Extensive site-directed and cysteine-scanning mutagenesis studies in which all of the 417 residues in the permease have been mutagenized have revealed that only four residues - Glu269 (helix VIII), Arg302 (helix IX), His322 (helix X), and Glu325 (helix X) - are irreplaceable with respect to active lactose transport. Such a small number of essential residues may appear surprising at first sight, but is in fact similar to what has been found by extensive mutagenesis of another E. coli inner membrane protein, diacylglycerol kinase (Wen et al., 1996).

This library of single and double Cys mutants has made possible a number of sophisticated biophysical approaches aimed at obtaining structural information. Attachment to defined cysteines of fluorescent probes capable of exhibiting excimer fluorescence if located within 3.5� of each other has allowed neighboring residues in the structure to be identified (Jung et al., 1993). Neighboring residues have also been identified by engineering divalent metal binding sites (bis- or tris-His residues) within the permease (He et al., 1995a; He et al., 1995b; Jung et al., 1995) and by using engineered Cu-binding sites to effect site-directed chemical cleavage (Wu et al., 1995). More long-range distance measurements have been made by a combination of engineered metal-binding sites and spin labeling of defined cysteines (Voss et al., 1995). Using site-directed mutagenesis of functionally critical residues followed by genetic selection of second-site revertants, specific ion-pairs have been identified that also provide structural constraints (Lee et al., 1992a; Dunten et al., 1993; Lee et al., 1993). Collectively, these studies have allowed a rough model of the packing of helices VII to XI to be proposed (Kaback et al., 1994; Kaback, 1996).

Related methods that have been used to obtain structural information on other membrane proteins include the engineering of double-Cys mutants that may be oxidized to form disulfide bonds if properly located (Pakula and Simon, 1992; Whitley et al., 1993; Lee et al., 1994; Lee and Hazelbauer, 1995), and the use of lipid- and water-soluble probes to quench the signal from ESR spin-labels attached to defined cysteines (Altenbach et al., 2001; Altenbach et al., 1990; Altenbach et al., 1994). Also, a large body of data related to ligand binding residues in the family of G-protein coupled receptors have been used to extract structural information (Schwartz et al., 1995).

Segments of eukaryotic membrane proteins that are translocated across the ER membrane (see below) can become modified by N-linked glycosylation on Asn-X-Thr/Ser acceptor sites. The transfer of the oligosaccharide moiety onto the nascent chain is catalyzed by oligosaccharyl transferase (OST), itself an oligomeric integral membrane protein with its catalytic site located in the lumen of the ER (Kelleher et al., 1992; Silberstein et al., 1992).

Detailed mutagenesis studies demonstrated that potential Asn-X-Thr/Ser acceptor sites can only be glycosylated if they are located at least 10-12 residues away from the nearest transmembrane segment (Nilsson and von Heijne, 1993), suggesting that the OST active site is located some distance away from the membrane surface. Further studies using artificial poly-Leu transmembrane segments showed that the location of the end of a hydrophobic transmembrane segment relative to the OST active site can be determined with high precision by measuring the degree of glycosylation of potential acceptor sites located at different distances form the hydrophobic segment (Nilsson et al., 1994). The OST active site can thus be used as a point of reference, in essence providing a "molecular ruler" that can be used to study transmembrane helices.

This approach has been used to determine the point at which a transmembrane helix exits the membrane, i.e., the approximate location of the helix relative to the lipid headgroup region (our unpublished data). So far, the number of residues between the known end of the transmembrane segment in the H-subunit of the photosynthetic reaction center and a potential acceptor site required for half-maximal glycosylation of that site has been determined. The location of other transmembrane helices relative to the lipid headgroup region can now be determined from the position of the half-maximally glycosylated acceptor site by comparison to the reference H-subunit helix.

In summary, techniques based on mutagenesis, although labor-intensive and yielding only bits and pieces of structural information, may prove to be the only way to guide modeling efforts when good crystals cannot be obtained.