The outer membrane protects Gram-negative bacteria against a harsh environment. At the same time, the embedded proteins fulfil a number of tasks that are crucial to the bacterial cell, such as solute and protein translocation, as well as signal transduction. Unlike membrane proteins from all other sources, integral outer membrane proteins do not consist of transmembrane α-helices, but instead fold into antiparallel β-barrels. Over recent years, the atomic structures of several outer membrane proteins, belonging to six families, have been determined. They include the OmpA membrane domain, the OmpX protein, phospholipase A, general porins (OmpF, PhoE), substrate-specific porins (LamB, ScrY) and the TonB-dependent iron siderophore transporters FhuA and FepA. These crystallographic studies have yielded invaluable insight into and decisively advanced the understanding of the functions of these intriguing proteins. Our review is aimed at discussing their common principles and peculiarities as well as open questions associated with them.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Gram-negative bacteria such as Escherichia coli are surrounded by two concentric lipid bilayer membranes, which confine the periplasmic space containing the murein sacculus. Both membranes contain proteins that assist in the passage of matter and information. Still, the two membranes differ markedly with respect to composition and function. (i) The lipid component of the inner (cytoplasmic) membrane is exclusively composed of phospholipids, mainly phosphatidylethanolamine (70–80%), phosphatidylglycerol and cardiolipin, equally distributed among the inner and outer leaflet. In contrast, the outer membrane is highly asymmetric, with its inner leaflet showing the same lipid composition as the cytoplasmic membrane and the outer leaflet consisting of lipopolysaccharides (or lipooligosaccharides in the case of Neisseriae). (ii) With respect to function, the major difference is that the outer membrane, because of the presence of pore-forming proteins (porins), is considerably more leaky than the cytoplasmic membrane. Remarkable advances in the understanding of the functions of outer membrane proteins have been made recently, which is mostly the result of a number of atomic structures solved by X-ray crystallography. In addition to the immediate relevance of these studies to the outer membrane protein field, the results may lead to conceptual models for cytoplasmic membrane proteins with similar functions but unknown structures. This review will summarize our knowledge about six families of integral outer membrane proteins from E. coli.
Outer membrane proteins
About 50% of the outer membrane mass consists of protein, either in the form of integral membrane proteins or as lipoproteins that are anchored to the membrane by means of N-terminally attached lipids. More than a dozen different outer membrane lipoproteins have been identified in E. coli (Blattner et al., 1997). A few integral membrane proteins, such as OmpA and the general porins, are expressed at high levels. Besides these, there are minor proteins whose synthesis in some cases is strongly induced when they are needed, such as porins (e.g. PhoE and LamB), TonB-dependent receptors (e.g. FhuA and FepA), components of several protein export systems, referred to as autotransporters and systems I to III (comprising the ABC transporters, the general secretion pathway, the type III secretion apparatus; Pugsley, 1993), proteins involved in the biogenesis of flagella and pili (Macnab, 1999; Soto and Hultgren, 1999), and enzymes (e.g. OmpT protease, Mangel et al., 1994; and phospholipase A, Dekker, 2000), whose function is still very much in the dark.
Exposure at the cell surface has led to the exploitation of outer membrane proteins by pathogenic agents such as bacteriophages and bacteriocins (Table 1). In fact, several outer membrane proteins were first described as bacteriophage receptors before their physiological function was discovered. The iron-siderophore transporter FhuA was first identified as a protein that is necessary for infection by phage T1 and, hence, was dubbed TonA (Tone protein A). Likewise, maltoporin was initially discovered as a receptor for phage λ, a finding that is still reflected by its present name, LamB. However, the evolutionary pressure did not result in the disappearance of those proteins but, instead, in a higher sequence variability of their surface-exposed regions and/or a sophisticated regulation of their synthesis, thus underscoring their vital importance for the bacterium.
Table 1. Structural and functional features of prototype outer membrane proteins from E. coli.
Small β-barrel membrane anchors
Small β-barrel membrane anchors
General (non- specific) porins
. PDB code, Brookhaven Protein Data Bank accession code.
The dielectric constant within a lipid bilayer is very low compared with that of the aqueous environment. Membrane proteins thus expose a hydrophobic surface to the lipid bilayer core, a property that distinguishes them from water-soluble proteins. This also implies that a maximum of hydrogen bonds of the protein segment located in the lipid bilayer is formed. Therefore, long before the first structure of a membrane protein was determined, it was predicted that only regular secondary structure elements (α-helices and β-sheet) could occur within the lipid bilayer in order to saturate the entire main-chain hydrogen-bonding potential. All donor and acceptor groups could be saturated either intrasegmentally, as in the case of α-helices, or intersegmentally by the formation of hydrogen bonds between adjacent β-strands (Rosenbusch, 1988).
High-resolution structures of membrane proteins, most of which became available only during the last few years, have indeed confirmed this prediction. All hitherto known integral membrane proteins consist of either α-helical bundles or β-pleated sheets in the form of a closed barrel. These architectures not only define two classes of membrane proteins, but also correlate with the location: while α-helical bundles are only found in the cytoplasmic membrane, β-barrels are restricted to the outer membrane. It is generally assumed that this differentiation originates from the biogenesis of outer membrane proteins, whose polypeptide chains have to cross the cytoplasmic membrane, where they would become stuck if they were too hydrophobic.
Interestingly, transmembrane β-barrels are not only found in outer membrane proteins (e.g. OmpA, OmpX, phospholipase A, general and substrate-specific porins and TonB-dependent receptors), but also in several microbial toxins that assemble on membranes to form oligomeric transmembrane channels, such as the heptameric pore-forming α-haemolysin from Staphylococcus aureus (Song et al., 1996). Other probable β-barrel membrane proteins include the toxin aerolysin (Parker et al., 1994) and the anthrax-protective antigen (Petosa et al., 1997), in which crystal structures of the proteins in solution were obtained, but the structure of their membrane-embedded forms are still unknown. β-Barrels are not restricted to membrane proteins, but are commonly observed in protein structures. They are classified in terms of two integral parameters: the number of strands in the β-sheet, n, and the shear number, S, a measure of the stagger of the strands in the β-sheet. These two main parameters determine the major geometrical features of β-barrels, such as the mean slope of the strands to the axis of the barrel (referred to as tilt angle), α, and the mean radius of the barrel, R (Murzin et al., 1994). In order to fulfil the hydrogen-bonding requirements along the whole β-strands, the sheets also have to be twisted and coiled, and the corresponding mathematical formulation has been described (Murzin et al., 1994). Based on these criteria, optimal shear numbers for least stressed β-barrels, S0, have been predicted for each number of strands, with S0 = n + 4 (Murzin et al., 1994). β-Barrels can have different strand connections or ‘topologies’. The simplest and most frequently observed of those topologies is the all-next-neighbour connection between adjacent strands. This is the architecture that has been found in all integral outer membrane proteins whose structures have been solved.
The OmpA protein occurs at about 100 000 copies cell−1, making it one of the major outer membrane proteins of E. coli. It plays a structural role in the integrity of the bacterial cell surface. It is composed of two domains: an N-terminal membrane-embedded domain of 170 amino acid residues, serving as a membrane anchor; and a C-terminal 155-residue domain, which is located in the periplasmic space and has been proposed to interact specifically with the peptidoglycan layer (Demot and Vanderleyden, 1994; Koebnik, 1995). Presumably, the physiological function of the OmpA protein is to provide this physical linkage between the outer membrane and the underlying peptidoglycan layer, a hypothesis that is substantiated by the observation that a double mutant in OmpA and the Lpp lipoprotein, another major outer membrane protein that interacts with the peptidoglycan, leads to spherical cells that can only survive under well-balanced osmotic conditions (Sonntag et al., 1978).
After many attempts to solve the structure of OmpA or that of its membrane domain, site-directed mutagenesis of predicted surface-exposed loops yielded proteins that formed well-ordered crystals suitable for X-ray crystallography (Pautsch and Schulz, 1998). The structure was found to be in perfect agreement with a previously established two-dimensional folding model, which, in fact, was the first one that proposed an antiparallel arrangement of transmembrane β-strands (Morona et al., 1984). Whereas the model correctly predicted the location of eight antiparallel β-strands that are connected by three short periplasmic turns and four relatively long surface-exposed loops, only the X-ray structure revealed the arrangement of side-chains within the barrel. A cluster of highly conserved charged residues (Lys-12, Glu-52, Arg-96, Arg-138 and Glu-140) was uncovered, which builds up a network of salt bridges and hydrogen bonds and may explain OmpA's extraordinarily high thermal stability. With a hydrophobic protein surface and a polar interior, the OmpA membrane domain can be envisaged as an inverse micelle (Pautsch and Schulz, 1998). The barrel interior accommodates several small, probably water-filled cavities (Fig. 1). This finding is in good agreement with a recent randomization mutagenesis study, in which it was found that the internal residues can tolerate a great number of sequence alterations without gross assembly defects of the proteins provided that the side-chain volume is not enlarged too much (Koebnik, 1999a). Despite the presence of cavities, no continuous transmembrane channel could be detected, thus questioning the relevance of reports of a possible function for OmpA in pore formation (Saint et al., 1993; Sugawara and Nikaido, 1994). Possibly, detection of pores in in vitro studies resulted from minor contaminations or misfolded molecules.
Based on its relatively small size and monomeric character, OmpA proved to be a well-suited model for the investigation of the folding of β-structured membrane proteins in vivo as well as in vitro. A surprising tolerance against mutational alterations was observed. It was possible to insert small peptides into all three periplasmic turns or into the surface-exposed loops without serious effects on membrane assembly (Freudl, 1989; Ried et al., 1994). Also, drastic shortening of all surface-exposed loops at the same time was perfectly tolerated (Koebnik, 1999b). Furthermore, side-chains that point towards the lipid bilayer could be replaced by other hydrophobic residues, and even several inward-pointing residues could be replaced at once and still gave properly folded molecules (Koebnik, 1999a). Finally, circular permutation of the polypeptide chain and even co-synthesis of protein fragments also failed to inactivate the protein (Koebnik and Krämer, 1995; Koebnik, 1996). In vitro folding studies identified a kinetic folding intermediate that was characterized spectroscopically (Surrey and Jähnig, 1995; Kleinschmidt and Tamm, 1996). From these studies, it was concluded that the unfolded polypeptide chain adopts its β-structure only after contact with an amphiphilic entity, such as a lipid bilayer, and only assembles into the membrane in a second step. The same scenario could hold true for the in vivo situation in which a membrane-attached assembly intermediate was identified (Freudl et al., 1986).
Shortly after solving the OmpA structure, the Schulz laboratory succeeded with another three-dimensional structure of a small β-barrel membrane protein, the OmpX protein (Vogt and Schulz, 1999). OmpX belongs to a family of highly conserved proteins that appear to be important for virulence by neutralizing host defence mechanisms (Heffernan et al., 1994). The prototype OmpX protein, Ail from Yersinia enterocolitica, promotes adhesion to and entry into eukaryotic tissue culture cells. Salmonella typhimurium produces two related proteins, PagC and Rck, which are important for survival in macrophages and inhibition of the formation of the membrane attack complex of the complement system. lom is not normally encoded by the bacterial chromosome, but belongs to the genome of phage λ, from which it is expressed during lysogeny, thus conferring adhesion of Escherichia coli to human cells.
Although its basic architecture (eight antiparallel amphipathic β-strands, hydrophilic surface-exposed loops and periplasmic turns, cluster of internal salt bridges and internal cavities) is very similar to that of the outer membrane-spanning β-barrel of the OmpA protein, it is sufficiently different that an attempt to solve the structure by molecular replacement with OmpA failed. One main difference is a shear number of 8 instead of 10, as in the case of OmpA, which results in a less tilted arrangement of its β-strands (35° versus 42°). Furthermore, OmpX has a much more ellipsoidal cross-section with an axes ratio of 1.6:1.0 (OmpA, 1.25:1.0). Another peculiarity of the OmpX protein is an elongation of four of its β-strands far beyond the region of the polar head groups of the outer leaflet of the membrane. Such a structural motif, reminiscent of a ‘waving flag’ or a ‘fishing rod’, has been proposed to function in cell adhesion and invasion as well as in the inhibition of the complement system by binding to one of its essential proteins by the edge of this elongated β-sheet (Vogt and Schulz, 1999).
Phospholipase A (OMPLA) is involved in colicin release from E. coli and is implicated in the virulence of Campylobacter and Helicobacter strains. It represents the only outer membrane enzyme whose three-dimensional structure has been solved (Snijder et al., 1999). Again, the polypeptide chain folds into an all-next-neighbour antiparallel β-barrel, which in this case is composed of 12 transmembrane β-strands. Like OmpA and OmpX, the barrel interior is polar and contains an intricate hydrogen-bonding network, thus providing a rigid structure. A 12-stranded barrel with nearly circular cross-section could hardly be filled by amino acid side-chains; in accordance with this consideration, the protein adopts a half moon-shaped cross-section with a convex and a flat side. The two polypeptide termini as well as surface-exposed loops L1, L4 and L6 obstruct the barrel from outside, thus excluding a pore function, as was proved by black lipid bilayer experiments (Snijder et al., 1999).
Phospholipase A normally exists as a monomer in the outer membrane, but has the potential for reversible dimerization in the presence of its substrate and calcium ions, with the dimeric form being catalytically active (Dekker et al., 1997). In this case, the flat sides of two barrels pack against each other. Almost all interactions between the monomers, such as knob-into-hole and aromatic stacking interactions, take place in the hydrophobic membrane-embedded region, except for three main-chain hydrogen bonds formed at the polar–apolar membrane boundary. Interestingly, the monomeric protein carries a polar patch of three residues (Tyr-92, Gln-94 Ser-96) at its membrane-exposed surface; however, a hydrogen bond network among these residues allows them to partition into the hydrophobic membrane environment. Upon dimerization, this network is modified to involve another hydrogen bond between the side-chains of Gln-94 of both monomers. The strict conservation of this residue suggests that it plays an important role in functional dimerization (Snijder et al., 1999). Despite these minor changes, no gross structural rearrangements are observed upon dimerization.
Although the function of OMPLA in the hydrolysis of phospholipids has been demonstrated, it is still not completely understood why such an enzymatic function evolved in the outer membrane and what physiological role the enzyme plays. The arrangement of amino acids His-142, Ser144 and Asn-156 resembles that of classical serine hydrolases, and their involvement in the active site was established by chemical modification and site-directed mutagenesis experiments (Horrevoets et al., 1991; Brok et al., 1996; Snijder et al., 1999). The three-dimensional structure, however, shows that these residues are located on the exterior of the β-barrel, positioned just outside the surface-exposed polar–apolar membrane boundary where phospholipids are not normally present. As substrate-induced dimerization is a prerequisite for the catalytic activity of OMPLA, the question arises as to how the protein may then fulfil its enzymatic function. As activation of phospholipase A concurs with the perturbation of the bacterial membrane, it was proposed that the relief of lipid asymmetry and the concomitant presentation of substrate to the active site binding pocket governs dimerization and calcium binding. The dimeric enzyme then hydrolyses the substrate into lyso-phospholipids and fatty acids, which will further destabilize the membrane bilayer and may thus facilitate the release of colicins or virulence factors (Dekker, 2000).
The general diffusion pores formed by porins allow the diffusion of hydrophilic molecules (< 600 Da) and show no particular substrate specificity, despite some selectivity for either cations or anions. Twenty years ago, the general porin OmpF was the first membrane protein to yield crystals of a size and order that were amenable to high-resolution structural analysis by X-ray crystallography (Garavito and Rosenbusch, 1980). However, it took more than 10 years before the atomic structure of the first bacterial porin from Rhodobacter capsulatus was elucidated (Weiss et al., 1991), soon followed by the structures of the osmotically regulated cation-selective OmpF porin and of the phosphate limitation-induced anion-selective PhoE porin, both from E. coli (Cowan et al., 1992). All these porins form homotrimers in the outer membrane. In each monomer, 16 β-strands span the outer membrane (Fig. 1). Unlike the other loops, the third loop, L3, is not exposed at the cell surface but folds back into the barrel, forming a constriction zone at half the height of the channel, giving it an hourglass-like shape. Therefore, this loop contributes significantly to the permeability properties, such as exclusion limit and ion selectivity, of the pore. Interestingly, this loop contains a sequence motif, PEFGG, that is highly conserved among enterobacterial porins (Jeanteur et al., 1991). At the constriction site, a strong transverse electrostatic field is caused by acidic residues in loop L3 and a cluster of basic residues at the opposite barrel wall (Fig. 2). It has been shown by chemical modification and site-directed mutagenesis that lysine and arginine residues contribute to the selectivity filter in the anion-selective porins PhoE of E. coli (Bauer et al., 1989; Benz et al., 1989) and the related Omp34 porin from Acidovorax delafieldii (Brunen and Engelhardt, 1995). Moreover, additional charges present in the channel and the channel mouth of PhoE are important for ion selectivity (Van Gelder et al., 1997a), and a corresponding influence of charges was observed with the cation-selective OmpF porin (Saint et al., 1996; Schirmer and Phale, 1999). Thus, it seems that the total charge constellation contributes to the formation of a specific electrical field, giving each porin its unique properties.
Porins are extremely sturdy proteins that can resist denaturation in the presence of 5 M guanidium hydrochloride or 2% SDS at 70°C. Recent studies showed that the latching loop L2 of OmpF, which bends over the wall of the adjacent monomer, contributes strongly to this exceptionally high stability (Phale et al., 1998). The hydrophobic trimer interface is also thought to add to the robustness of trimeric porins, as removing specific monomer–monomer interactions by site-directed mutagenesis results in a dramatic decrease in stability in the case of OmpF and PhoE (Van Gelder and Tommassen, 1996; Phale et al., 1998). PhoE trimers dissociate into folded monomers, which are extremely sensitive to heat denaturation (Van Gelder and Tommassen, 1996). Therefore, it seems unlikely that stability is generated by the hydrophobic interface itself. Probably, this interface drives the molecules to oligomerize in the periplasm before insertion into the outer membrane. In this lipophilic environment, hydrophobic interactions play a less significant role, and stability is ensured by other types of interactions.
Purified porins can be reconstituted into lipid bilayers where they form ion-permeable pores. Conductance measurements have shown that most porins exist in either open or closed states, depending on the transmembrane potential. The physiological significance of this voltage gating has been questioned as the critical voltage, Vc, above which the OmpF and PhoE channels close, is larger than the naturally occurring Donnan potential across the outer membrane. However, there is evidence that Vc is affected by several parameters, such as pH, ionic strength, pressure and the presence of polysaccharides, membrane-derived oligosaccharides or polycations, e.g. endogenous cadaverine (Delcour, 1997). Therefore, it seems that varying only one parameter (e.g. Donnan potential, osmotic pressure) in vitro might have resulted in higher values than are probably needed to close the channels under certain circumstances in vivo. It is clear from previous reports that charges present in porin channels are involved in voltage gating. Replacement of charged residues in porins resulted in a changed voltage sensitivity. Interestingly, the anion-selective PhoE and the cation-selective OmpF porin showed an opposite voltage dependence upon mutation of charged residues (Saint et al., 1996; Van Gelder et al., 1997a; Samartzidou and Delcour, 1998). Therefore, charged residues at the constriction zone were suggested to act as sensors for voltage gating.
The mechanism of voltage gating is still not resolved. From mutageneses and chemical modification studies, it is clear that charges present inside the porin channels are involved. A possible mechanism might be the screening–unscreening of charges. The entrance of molecules into the channel involves the removal of the hydration shell. Subsequent redistribution of water molecules (or counter ions) in the channel might result in unscreening of the charges in the channel and an increase in the strength of the electrical field. This might induce a reorientation of side-chains in the channel and/or locally restricted movements within loop L3, resulting in closing of the pore. Based on molecular dynamics (MD) simulations, movement of the complete L3 loop and subsequent blocking of the channel was suggested as a possible gating mechanism (Watanabe et al., 1997). However, as this loop has many interactions (salt bridge, hydrogen bond network) with the barrel wall, this idea seems to be very unlikely. Indeed, tethering the tip of loop L3 of OmpF or PhoE to the barrel wall still allowed complete closing of the channel, thus disproving gross movements of loop L3 (Eppens et al., 1997;Phale et al., 1997). On the other hand, MD perturbation studies have also suggested that at least part of loop L3 from R. capsulatus porin is flexible (Soares et al., 1995). This part may well correspond to the region just behind the conserved PEFGG sequence motif of E. coli porin. In another MD simulation of OmpF, precisely this segment of L3 was observed to be the most flexible region (Tieleman and Berendsen, 1998). Although the OmpF L3 loop could even be tethered to the barrel wall at several positions (Bainbridge et al., 1998a), this potentially flexible region was unfortunately overlooked, thus leaving the possibility of local movements open. Indeed, in support of this idea, substitution of the two glycine residues in the PEFGG sequence resulted in altered channel characteristics (Van Gelder et al., 1997b).
Interestingly, substrate-specific channels (see next paragraph) show no voltage gating, an observation that could result from a more balanced distribution of oppositely charged residues compared with the general porins or from a tighter fixation of its L3 constriction loop. We were indeed able to construct a voltage-dependent LamB channel by creating a charge segregation at the constriction zone; however, further mutagenesis studies failed to confirm this simple model. Moreover, voltage gating was even observed with the heptameric staphylococcal α-haemolysin, which contains a 14-stranded antiparallel β-barrel acting as a membrane pore, thus disproving the requirement for a transverse electric field or an internal loop for voltage gating (Bainbridge et al., 1998b).
Besides voltage-induced channel closure, fast flickering of the current traces was observed in patch-clamp experiments (Berrier et al., 1992). These spontaneous fluctuations may also reflect small conformational changes within loop L3. Accordingly, it has been demonstrated that the gating frequency in OmpC is affected by mutations that disrupt a putative hydrogen bond network or a salt bridge between the L3 loop and the barrel wall (Liu and Delcour, 1998).
Besides general diffusion pores that discriminate between solutes primarily on their size and charge, the outer membrane contains channels with specificity for certain substrates. The best-studied examples are the maltooligosaccharide-specific maltoporin LamB from E. coli and the sucrose-specific porin ScrY from S. typhimurium. The lamB gene is expressed as part of the mal regulon upon induction by maltose or maltodextrins. Its gene product forms ion-conducting channels when reconstituted into lipid bilayers (Benz et al., 1986). Liposome swelling assays showed that maltoporin has a high permeation rate for maltose and maltodextrins, while uptake of sucrose could hardly be detected (Luckey and Nikaido, 1980). Maltodextrins up to maltoheptaose could be transported, as was shown in in vivo and in vitro uptake experiments with radiolabelled sugars (Gehring et al., 1991; Charbit et al., 1998). In contrast, uptake of sucrose is mediated by the plasmid-borne sucrose regulon, which consists of the scrKYABR genes. Cells containing this plasmid are able to grow on this sugar as a sole carbon source. Accordingly, ScrY porin reconstituted into vesicles showed a high permeation rate for sucrose (Hardesty et al., 1991).
Recently, the three-dimensional structure of these two specific porins have been elucidated (Schirmer et al., 1995; Forst et al., 1998). Although they display only moderate sequence homology, the crystal structures can be superimposed. Both LamB and ScrY form homotrimers whose monomers consist of 18-stranded antiparallel β-barrels. As was found with the general porins, the third loop, L3, folds back inside the β-barrel. A peculiar feature of ScrY, compared with LamB, is the presence of a 70-residue-long N-terminal extension, which hangs out into the periplasm in the form of a parallel triple-stranded coiled-coil. By X-ray analyses of sugar-soaked LamB crystals, a substrate translocation pathway, consisting of a row of aromatic amino acids (greasy slide) that is lined up by polar residues (ionic track), has been discovered (Fig. 2) (Dutzler et al., 1995; Meyer et al., 1997). Sugar residues are in van der Waals' contact with the greasy slide by their hydrophobic face, while hydrogen bonds are formed between their hydroxyl groups and the ionic track residues. It has been proposed that movement of the substrate through the channel proceeds by continuous disruption and formation of these hydrogen bonds (Meyer and Schulz, 1997). Indeed, we could demonstrate the importance of hydrogen bond formation for sugar translocation by studying site-directed LamB mutants in vitro by current fluctuation and liposome swelling assays and by in vivo sugar uptake experiments (Dumas et al. 2000).
Most of the channel-lining residues are conserved between LamB and ScrY porins and can be structurally superimposed. However, comparison of the two structures revealed a remarkable difference at three positions at the constriction site. Arg-109 of LamB, which protrudes into the channel lumen, is replaced by Asn-192 pointing towards the barrel wall in ScrY. Tyr-118, which constricts the channel in LamB, is replaced by the shorter Asp-201. Finally, Asp-121 of LamB is replaced by a Phe-204. Transplanting these ScrY residues to LamB resulted in a wider channel, allowing transport of the more bulky sucrose molecule, which is not able to permeate through the wild-type LamB porin (own unpublished data). The opposite was observed when Asp-201 of ScrY was replaced by the corresponding LamB residue (Ulmke et al., 1999). This mutation led to a narrowing of the substrate range of ScrY to that resembling LamB. These experiments fit nicely with a three-dimensional structural analysis that demonstrated that sucrose enters LamB with its glucosyl moiety from the outside and, indeed, becomes stuck at the channel constriction because of its bulky fructosyl residue (Wang et al., 1997). It will be interesting to study how certain residues, most importantly those around the constriction zone and residues from the greasy slide, as well as from the ionic tracks, influence substrate specificity and transport kinetics. Well-established procedures for structural (X-ray analyses) and functional (in vitro-reconstitution) characterization of mutant proteins make LamB an ideal model for in-depth study.
All Gram-negative bacteria contain several high-molecular-weight outer membrane proteins (TonB-dependent receptors, e.g. FhuA and FepA), which are involved in the uptake of large substrates, such as iron-siderophore complexes or vitamin B12. Intriguingly, some of these receptors (e.g. FecA of E. coli and PupB of Pseudomonas putida) contain an N-terminal extension that enables them, together with a protein of the cytoplasmic membrane, to signal the presence of substrates from the cell surface into the cytoplasm (Braun, 1995). For their function, these proteins depend on the electrochemical potential of the cytoplasmic membrane and a proteinaceous, energy-transducing module, the TonB–ExbBD complex. By analogy with the porins and based on genetic and biochemical studies, it was assumed that these large proteins are built up from huge β-barrels that are gated by a flexible loop (Rutz et al., 1992; Killmann et al., 1993; Jiang et al., 1997). Here, solving the three-dimensional structure revealed an unexpected structural organization (Ferguson et al., 1998; Locher et al., 1998; Buchanan et al., 1999). Both FhuA and FepA form a C-terminal 22-stranded β-barrel and an N-terminal plug domain, which is located inside the barrel and thus obstructs the channel interior. The plug domain is tightly attached to the barrel interior by nine salt bridges and more than 60 hydrogen bonds (FhuA; Locher et al., 1998). For energetic reasons, it appears highly unlikely that the entire domain is removed from the barrel interior in order to form a translocation pathway for substrates. Instead, it seems that, upon action of the energized TonB protein, conformational changes will lead to the opening of a translocation pathway that might connect some of the pre-existing cavities (Fig. 1).
The plug-and-barrel organization is so far unique, and it would be of interest to determine whether these domains represent independent folding units. A deletion mutant that lacks the entire N-terminal plug domain (FhuAΔ5–160) was shown to insert into the outer membrane and to form large pores, as judged by the TonB-independent uptake of substances that are too large to pass through the general diffusion pores, such as maltotetraose, maltopentaose, ferrichrome, as well as several antibiotics including albomycin, vancomycin and bacitracin (Braun et al., 1999). Unfortunately, no in vitro data are available on pore formation of this protein variant. It would be interesting to compare the atomic structure of the plugless protein with the wild-type FhuA barrel, which has a bean-shaped cross-section (35 × 25 Å). Possibly, this cross-section is stabilized by the inserted plug domain; therefore, it is conceivable that the deletion mutant could display another β-sheet geometry. Furthermore, it is possible that, in the absence of any internal stabilization, the lateral pressure of the lipid bilayer would induce a different geometry from that one would observe after crystallization of the protein in the presence of detergents; therefore, in vitro results have to be interpreted with caution.
The FhuA protein was crystallized with and without its substrate ferrichrome. Ferrichrome binds on top of the plug domain in a binding pocket that is mainly padded by aromatic residues of the plug domain (Tyr-116) as well as the barrel domain (Tyr-244, Tyr-315). In addition, Arg-81 appears to be a crucial residue for ferrichrome binding, and it shifts by about 1.5 Å towards the substrate. This local change is propagated exclusively across the plug domain and is markedly amplified towards the periplasmic channel mouth, resulting in an unwinding of an α-helix (residues 22–30) and movement of the corresponding peptide segment to the opposite barrel wall (with a Cα distance of Trp-22 of 17 Å) (Ferguson et al., 1998; Locher et al., 1998). This drastic conformational change at the periplasmic surface of the protein probably allows the TonB protein to recognize substrate-loaded receptor molecules. The action of TonB might then induce a conformational change in the receptor protein that liberates the substrate from its binding site and generates a translocation pathway across the protein. Interestingly, ferrichrome uptake by the plugless mutant FhuAΔ5–160 strain is strongly enhanced by the action of TonB. This indicates that ferrichrome might still bind to a partially disrupted binding site and that TonB is also able to induce a conformational change in these substrate-loaded molecules. It is unclear whether these results reflect the wild-type situation, i.e. if not only the plug domain but also the barrel wall is subject to a conformational change. It appears possible that lack of internal stabilizing interactions in the plugless mutant might have created a permanently open pore with a low-affinity binding site, from which bound substrates might be displaced by a smaller conformational change than in the wild-type protein.
Another interesting FhuA mutant carries a small deletion in the surface-exposed loop L4 (FhuAΔ322–355). In a tonB background, this mutant displayed a phenotype similar to that of FhuAΔ5–160, i.e. it allowed the transmembrane diffusion of large substrates (Killmann et al., 1993). Furthermore, black lipid bilayer measurements demonstrated the formation of ion-conducting pores by this mutant protein that were electrophysiologically similar to pores that were created upon interaction of wild-type FhuA with bacteriophage T5 (Bonhivers et al., 1996). This finding was interpreted as evidence for the formation of a large hollow transmembrane channel. Possibly, the disruption of the surface-exposed protuberance leads to a folding defect in the L4 mutant, resulting in a protein variant that does not carry the plug domain inside the barrel. In order to test this idea, a peptide segment of 16 residues was inserted at the boundary between both protein domains (FhuA161–16Δ322–355; own unpublished data). When spheroplasts that had expressed the protein were treated with subtilisin, the protein was cleaved near the insertion site as well as in the spatial vicinity of the shortened surface loop. The latter cleavage was also observed with the parental loop deletion mutant, thus indicating that shortening of loop L4 might have distorted the conformation of the neighbouring loops. More strikingly, however, the plug domain was not digested at all, a finding that suggests that this protein domain was still assembled within the barrel domain and did not become displaced by the loop deletion. Alternatively, the plug domain could be resistant to cleavage by the protease used; solving the atomic structure will give an answer to this question.
Comparisons of β-barrel structures
With high-resolution structures of six protein families at hand, one can look for common and unique features. All these integral membrane proteins are embedded in the outer membrane in the form of antiparallel all-next-neighbour β-barrels with an even number of β-strands that may vary between eight and 22. Despite this regular structure, cross-sections of the barrel domains show considerable variation: from almost circular (OmpA) or more ellipsoid (OmpX) to bean-shaped (FhuA) or even half moon-shaped (OMPLA) cross-sections. The shape of the barrels is probably determined by their interior, i.e. by the inward-pointing side-chains as well as by peptide segments (as in the case of porins) or protein domains (e.g. the plug domain of TonB-dependent receptors) that fold back inside the barrel domain.
The length of transmembrane β-strands varies between six, the minimum that is needed to span the membrane bilayer, and 25 amino acid residues (Table 2). With an average length of 12.3 residues, they are considerably longer than β-strands from water-soluble globular proteins. Whereas the periplasmic end is always located near the polar head groups of the inner leaflet of the membrane, the β-strands may extend far behind the lipopolysaccharide (LPS) core region at the other end. β-Strands frequently contain aromatic residues whose side-chains point to the polar head groups. These aromatic girdles have been proposed to anchor the protein in the membrane (Schulz, 1992). In all cases, the C-terminal β-strand ends with an aromatic residue, although mutagenesis studies have shown that this is not an absolute prerequisite for membrane assembly (Struyvéet al., 1991; Koebnik, 1999a). Cysteine residues are absent from transmembrane β-strands. In a few cases, the regular hydrogen-bonding network of the β-barrel is distorted by the appearance of small non-regular bulges of one or two residues or by prolines; these irregularities are usually found near the ends of the β-strands.
Table 2. Statistical comparison of bacterial outer membrane β-barrel proteins.
At the periplasmic side, β-strands are connected by short peptide segments that are often referred to as turns, although they do not always form hydrogen-bonded reverse turns in the classical sense. Their length varies between one and 12 residues. At the opposite side, the strands are connected by long hydrophilic loops of two to 46 residues that sometimes contain short segments of regular secondary structure or disulphide bridges (loops L1 of maltoporin, L4 and L11 of FhuA, L7 of FepA). These loops sometimes fold back into the barrel interior, as is the case with loop L3 of the porins.
As a rule, the protein surface within the plane of the membrane is covered by non-polar side-chains, a finding that is in accordance with the low dielectric constant of the environment. Aromatic (29.3%) and hydrophobic (49.2%) residues account for more than three-quarters of the membrane-embedded protein surface, with a concomitant discrimination against charged residues (2.8%) (Table 2). Surprisingly, the nearly isohydrophobic amino acid isoleucine (5.7% of lipid-contacting residues) is very under-represented in comparison with leucine (16.6%), although studies with globular proteins have indicated that β-branched amino acids are superior for the formation of β-structure (Regan, 1994). Other constraints, such as tRNA availability, might be responsible for this phenomenon. Furthermore, although the general trend holds true for all the proteins discussed here, there are surprising differences even within one protein family; for example, whereas FhuA exposes 14 tyrosine residues and only one tryptophan residue to the membrane, the situation is reversed with FepA (four Tyr and 12 Trp).
One interesting exception to this rule is the polar patch at the surface of monomeric phospholipase A, which is involved in dimerization of the enzyme. Interestingly, the interacting faces of the trimeric porins also have a different amino acid composition to the rest of the membrane-embedded domains (Seshadri et al., 1998). Thus, certain motifs appear to be unique for protein oligomerization within the plane of the membrane. Another surprising example is given by the TonB-dependent receptors, which contain a few charged residues at the lipid-contacting protein surface. Here, these residues are located at the extremities of transmembrane β-strands, near the polar head groups. The possibility that charged residues can be tolerated at such positions was shown with the OmpA protein (Koebnik, 1999a). As in α-helical transmembrane segments, this phenomenon might be explained by the so-called snorkel effect, i.e. the long side-chains of these residues might reach up along the transmembrane strand to allow the terminal charged moiety to reside in the lipid head group region, while the Cα atom of the residue is positioned well below the membrane–water interface (Monnéet al., 1998). Another possibility is that such polar spots might be used for docking of another membrane protein, a hypothesis that was put forward for maltoporin (Meyer et al., 1997).
Interestingly, the interior of the β-barrels, which contain either channels or cavities, is much more hydrophilic than the protein surface, and even the moderately densely packed interiors of the OmpA and OmpX barrels contain numerous polar residues that form an extended and conserved network of salt bridges and hydrogen bonds within the barrel, a finding that makes these two proteins appear somewhat like inverse micelles.
A final point of interest that has to be taken into account for modelling approaches is the finding that the polypeptide chain of integral outer membrane proteins might not always cross the membrane plane in the form of amphipathic β-strands, but that other topologies are possible, as exemplified by the plug domain of the TonB-dependent receptors. By folding into such composite structures, surface exposure of peptide segments or even protein domains is possible without lipid-contacting stretches that bracket these regions. A conceptually similar situation could apply to the protein autotransporters, which are predicted to form C-terminal β-barrels that govern the translocation of an N-terminal protein domain (Klausner et al., 1993).
Sequence comparisons of members of all outer membrane protein families show that the transmembrane β-strands are highly conserved and that the surface-exposed regions display the highest variability, not only in sequence but also in length. This might be a consequence of the recognition of these regions as receptors by phages or bacteriocins. For instance, five enterobacterial OmpA proteins show both 54% amino acid identity within the periplasmic turns and surface-exposed loops and 74% identity within the β-sheets (Braun and Cole, 1984). Similarly, 71% of the turn residues and 76% of the loop residues are identical between four phospholipase A proteins, whereas 83% of their β-strand residues are strictly conserved (Brok et al., 1994). The same trend is observed with four enterobacterial FhuA proteins, which show 52% amino acid identity within the periplasmic turns, 32% identity within the surface-exposed loops and 56% identity within the β-sheet (Killmann et al., 1998).
Separate comparisons of inward- and outward-pointing residues of the β-barrel show that the interior is more strongly conserved than the protein surface, a feature shared with water-soluble proteins (OmpA: 77% versus 70%; OMPLA: 85% versus 82%; FhuA: 61% versus 51%; OmpX homologues: 33% versus 14%; percentage identity of inward-pointing residues given first). A closer analysis reveals interesting differences within the β-sheet region: whereas the membrane-embedded region of four FhuA proteins is conserved at 65% identity (57% of outward-pointing and 73% of inward-pointing residues), the surface-exposed protrusion is much less conserved (42% identity, regardless of the side-chain orientation) and approaches the level of conservation in the loop residues. This difference is easy to understand, as the β-sheet does not form any contact with the internal plug domain outside the membrane plane, which certainly imposes an additional constraint on the inward-pointing barrel residues.
What can we learn from the new structures with regard to structure predictions that might be applied to uncharacterized outer membrane proteins? With respect to their membrane topology, the structures demonstrate the power of non-crystallographic approaches, such as spectroscopic analyses (circular dichroism and Raman), searching for potential amphipathic β-strands along the amino acid sequence, mapping of changed amino acid residues in phage-resistant mutants, comparative sequence analyses and peptide insertion mutageneses (for the introduction of epitopes or protease cleavage sites). With OmpA, a relatively small protein, the membrane topology was correctly predicted by a combination of these approaches (Morona et al., 1984). A satisfactory model of phosphoporin (PhoE), a close homologue of OmpF, was also obtained by these methods (Van der Ley et al., 1986). However, the success of these approaches was not that overwhelming with LamB and FhuA. These two proteins illustrate the pitfalls of all predictions: they were very much inspired by apparently analogous structures, which led to the wrong prediction of 16 strands in the case of LamB (guided by OmpF) (Schirmer and Cowan, 1993), and also to the proposed existence of a gating loop in FhuA instead of an autonomous plug domain (guided by the L3 loop of porins) (Killmann et al., 1993). Predictions failed to identify functionally important features, such as the transverse electric field inside the general porins, the greasy slide and ionic tracks in LamB and ScrY or the plug domain in FhuA and FepA. On the other hand, the atomic structures demonstrate nicely how topological knowledge can be used for better or more efficient production of X-ray-suited crystals. In the case of FhuA, a His-tag was introduced into the largest (and correctly predicted) loop to facilitate the purification of the protein; and with OmpA and OmpX, in anticipation that unfavorable protein–protein contacts would limit crystal growth, the same knowledge was used for a site-directed mutagenesis of the predicted surface-exposed loops. In summary, predictions can only give an idea about the general folding (topology) of a protein, but do not reveal the structural details that are necessary for a mechanistic understanding of protein function.
Solving the structure of members of six outer membrane protein families revealed a common architectural principle. It is amazing how this theme was varied by nature to come up with proteins that fulfil a variety of completely different functions, such as diffusion pores, substrate-specific transporters, signal transduction and enzymatic activity. It would be of interest to solve the structures of many more membrane proteins, such as TolC homologues from the type I secretion systems (Koronakis et al., 1997), secretins of type II and type III secretion machines (Genin and Boucher, 1994), autotransporters (Klausner et al., 1993), usher proteins that are involved in pilus assembly (Soto and Hultgren, 1999), neisserial opacity Opa and Opc proteins that are responsible for host cell interactions (Bhat et al., 1991; Merker et al., 1997), the long-chain fatty acid-specific and nucleoside-specific transporters FadL and Tsx (Fsihi et al., 1993; Kumar and Black, 1993) and another enzyme, the OmpT protease and its homologues (Mangel et al., 1994). Another point of interest is whether eukaryotic organelles also contain β-barrel membrane proteins; according to predictions, the mitochondrial VDAC channel appears to be a strong candidate (Mannella, 1992). These are certainly hard nuts to crack, and we are curious to see what surprises will come up. Years after the rule that membrane proteins are always built up from transmembrane α-helices was first shown to be incorrect, one should remain open-minded: it is NOT our intention to create a new dogma that ALL integral outer membrane proteins MUST be β-barrels. It might be instructive to envisage alternative folds, such as β-helices or mixed α/β-structures (Montal, 1996).
Present Addresses: †Martin-Luther-University, Institute of Genetics, D-06120 Halle (Saale), Germany.
Present Addresses: ‡California Institute of Technology, Division of Chemistry, Pasadena, CA 91125, USA.