Research Article 863

Crystal structure of a colicin N fragment suggests a model for toxicity Ingrid R Vetter†*, Michael W Parker‡*, Alec D Tucker§, Jeremy H Lakey#, Franc Pattus¶ and Demetrius Tsernoglou¥

Background: Pore-forming colicins are water-soluble capable of Address: European Molecular Biology Laboratory, binding to and translocating through the cell envelope. They Postfach 10.2209, D-6900 Heidelberg, Germany. then undergo a transition to a transmembrane ion channel in the cytoplasmic Present addresses: †Max-Planck-Institute for membrane leading to bacterial death. Colicin N is the smallest pore-forming Molecular Physiology, Rheinlanddamm 201, 44139 colicin known to date (40 kDa instead of the more usual 60 kDa) and the crystal Dortmund, Germany, ‡St. Vincent’s Institute of structure of its membrane receptor, the porin OmpF, is already known. Medical Research, 41 Victoria Parade, Fitzroy, Victoria 3065, Australia, §Pfizer Central Research, Structural knowledge of colicin N is therefore important for a molecular Molecular Sciences Department, Sandwich, Kent, understanding of colicin toxicity and is relevant to toxic mechanisms in general. CT13 9NJ, UK, #Department of Biochemistry and Genetics, The Medical School, The University of Results: The crystal structure of colicin N reveals a novel receptor-binding Newcastle-upon-Tyne, NE2 4HH, UK, ¶Département Récepteurs et Protéines Membranaires, UPR 9050, domain containing a six-stranded antiparallel β sheet wrapped around the CNRS, ESBS, Rue Sébastien Brant, F-67400 α 63 Å long N-terminal helix of the pore-forming domain. The pore-forming Illkirch, France and ¥Department of Biochemical domain adopts a ten α-helix bundle that has been observed previously in the Sciences, Università di Roma, ‘La Sapienza’, pore-forming domains of colicin A, Ia and E1. The translocation domain, Piazzale Aldo Moro 5, I-00185 Rome, Italy. however, does not appear to adopt any regular structure. Models for receptor *Corresponding authors. binding and translocation through the outer membrane are proposed based E-mail: [email protected] on the structure and biochemical data. [email protected]

Conclusions: The colicin N–ompF system is now the structurally best-defined Key words: colicins, membrane translocation, porin, receptor binding, toxin structure translocation pathway. Knowledge of the colicin N structure, coupled with the structure of its receptor, OmpF, and previously published biochemical data, Received: 9 February 1998 limits the numerous possibilities of translocation and leads to a model in which Revisions requested: 25 March 1998 the translocation domain inserts itself through the porin pore, the Revisions received: 28 April 1998 Accepted: 27 May 1998 receptor-binding domain stays outside and the pore-forming domain translocates along the outer wall of the trimeric porin channel. Structure 15 July 1998, 6:863–874 http://biomednet.com/elecref/0969212600600863

© Current Biology Ltd ISSN 0969-2126

Introduction energy-dependent uptake of vitamin B12 and iron sidero- Colicins are -encoded bacteriocins, produced by phores [6,7]. The Tol system consists of six proteins: Escherichia coli, that kill sensitive E. coli strains and related TolA, TolB, TolQ, TolR, PAL and Orf2 [8,9]. Group B bacteria (for recent reviews see [1–5]). The cell envelope colicins (B, D, Ia, Ib, M, 5 and 10) are taken up via the Ton of gram negative bacteria, with its two membranes, pre- pathway, whereas the translocation of group A colicins (A, sents a serious obstacle for the uptake of macromolecules E1–9, K, N and cloacin DF13) involves interaction with of the size of colicins (40–60 kDa). To cross this barrier TolA and different combinations of the proteins from the and reach their target, colicins must first bind to their Tol family [10–12]. Additional proteins are also essential cognate receptor on the cell surface and then translocate for the translocation of some colicins: the porin OmpF for across the outer membrane. The pore-forming colicins (E1, colicins A, K and N; TolC for colicins E1 and 10; and A, B, N, Ia, Ib, K, 10 and 5) and colicin M, an inhibitor of OmpA for colicin K [13]. murein biosynthesis, insert into the inner membrane, whereas colicins with activity (E2–E9 and cloacin Colicins consist of distinctive protein modules, which pre- DF13) must enter the cytoplasm. Uptake of colicins sumably originated from the genetic recombination of through the outer membrane follows two alternative path- DNA fragments that encoded for different specialized func- ways mediated by the Ton or Tol systems. These two tions. The N-terminal domain specifies the translocation systems are also involved in the transport of bacteriophage route (Ton or Tol pathways), a central domain specifies the DNA into the cells. The Ton system consists of the pro- receptor, and a C-terminal domain carries the killing activ- teins TonB, ExbB and ExbD and is also involved in the ity. Colicins E1 and 10 contain a fourth domain, the TolC 864 Structure 1998, Vol 6 No 7

domain, located between the translocation and the receptor recently solved tripartite structure of colicin Ia at 3.0 Å domains [14]. TonB-dependent colicins contain a homolo- resolution [26] is compatible with such organization. The gous sequence of five amino acids, called the TonB box, in molecule is 210 Å long with the translocation and pore- their N-terminal translocation domain, which is also present forming domains separated from the receptor-binding in TonB-dependent receptors [15]. The Tol-dependent col- domain by a pair of helices, each 160 Å long. icins possess a glycine-rich and proline-rich region in their N-terminal translocation domains. The existence of a TolA Colicin N, which belongs to the group A colicins, possesses box has been proposed with the following motif: Asp–Gly– similar properties and functions to other pore-forming col- (Ser,Thr)–Gly–Trp [14]. But mutation of serine in this icins, but has a much shorter polypeptide chain (385 amino motif to phenylalanine in colicin E3 inhibits TolB binding acids) and a single receptor. It thus provides a simpler instead [12] and the motif is present in only one subset of model system than other colicins for pursuing structure– Tol-dependent colicins. Another TolA box with the motif function studies. Here, we present the crystal structure, at Glu–Gly–Gly–Gly–Ser has been proposed for protein III of 3.1 Å resolution, of a large fragment of colicin N (missing phage f1 [16]. There is experimental evidence that the the first 66 residues) that contains the pore-forming and N-terminal domains of both group A and group B colicins the receptor-binding domains. The pore-forming domain interact with components of the translocation system on the adopts the canonical ten helix bundle seen in other colicins inner side of the outer membrane; translocation through the [24–26], whereas the receptor-binding domain adopts a outer membrane therefore occurs during colicin uptake. novel fold. The crystal structure of the colicin N receptor, Mutations in the TonB box of colicins B and M impaired the protein ompF, which is one of the major unspecific uptake, but this was partially restored by a mutation in porins from E. coli, has been solved previously at 2.4 Å TonB [17]. Expression of the N-terminal domain of either resolution [37]. The structures have enabled us to look at colicin B or colicin E3 in the periplasm make E. coli strains possible interactions between the colicin receptor-binding tolerant to the corresponding colicins [12]. domain and its receptor, leading to new insights into the translocation mechanism of colicins. Pore-forming colicins induce voltage-dependent potass- ium and phosphate efflux across the plasma membrane of Results the target bacterium [18] and form voltage-gated ion chan- The pore-forming domain nels in artificial membranes [19–23]. The pore activity is The structure of colicin N is shown in Figure 1 and the carried by the C-terminal domain [20,23]. Near the car- corresponding topology diagram is presented in Figure 2. boxyl end of the pore-forming domain is a hydrophobic The pore-forming domain of colicin N consists of a bundle segment present in all the pore-forming colicins. The of ten α helices arranged in three layers forming the so- crystal structure of the pore-forming domain of colicin A, called ‘globin-like’ topology [38,39], also found in the pore- solved to 2.4 Å resolution, revealed that this hydrophobic forming apoptosis regulatory protein Bcl-xL [40]. This topol- segment forms a buried α-helical hairpin, surrounded by ogy is identical to that observed in the pore-forming an outer bundle of eight amphipathic helices [24]. A similar domains of other colicins (Figure 2; [24–26]). A character- fold has been identified recently in the crystal structures istic feature of the pore-forming domains of colicins is of the pore-forming domain of colicin E1 and the full- their hydrophobic core consisting of a buried helical hairpin, length colicin Ia [25,26]. A large amount of data have been formed by α helices 8 and 9, and surrounded by a cluster accumulated concerning the mechanism of membrane inser- of aromatic residues. Although the hairpin is present in all tion and pore formation by this domain [1–4,27–31]. A uni- pore-forming colicins, some colicins exhibit extensive dele- fying model has been proposed recently that reconciles tions in this region, for example colicins E1 and Ia. various proposals on the orientation of the hairpin with Colicin N has a deletion of one amino acid in helix 8 com- respect to the membrane [32] and segments of the poly- pared to colicin A (corresponding to Leu165 in the colicin A peptide chain have been identified that move across the pore-forming domain; see Figures 3 and 4). In addition, membrane in response to voltage [33,34]. there is a proline residue (Pro346) close to the site of dele- tion (which is an alanine in colicin A). These changes do Much less is known about the mechanism of the colicin not shorten the loop adjacent to this helix, but disturb the interaction with its receptor and the subsequent transloca- structure of the helix itself. The helix loses one helical tion of the toxin. The mechanism appears to follow the turn compared to colicin A, but the conformation of the general rule that significant unfolding should occur prior proline residue ensures that the loop between helices 8 and to, or during, translocation [35]. Engineering disulphide 9 nearly retains the same conformation in both colicins bridges within the pore-forming domain of colicin A slows (Figures 3 and 4; [41]). down translocation, but does not abolish it completely [36]. There is some experimental evidence that the colicin A remarkable feature of the hydrophobic core is its poor molecule spans the periplasmic space and contacts both packing arrangement highlighted by a tube-like channel the inner and outer membranes simultaneously [35]. The (Figure 5). The channel is created by virtue of the structural Research Article Colicin N structure Vetter et al. 865

Figure 1

(a) (b) (c)

N

C Structure

Schematic representation of colicin N. (a) The receptor-binding figures were generated using MOLSCRIPT [70]. (c) An electrostatic domain is coloured red and the pore-forming domain is coloured blue. surface representation of the receptor-binding domain highlighting the The hydrophobic helical hairpin is coloured yellow. The cluster of solvent-exposed cleft. Red indicates electronegative potential positively charged residues in the putative binding cleft is shown in (< –10 kT) and blue indicates positive potential (> 10 kT). This figure ball-and-stick representation. (b) A view of the receptor-binding was generated using GRASP [71]. domain viewed from the top of the molecule as shown in (a). These

alterations at the C-terminal end of helix 8 (compared to Figure 2 colicin A), which is six residues shorter than helix 9. The channel stretches from one side of the molecule almost 116 114 through to the other side where it is met by a large depres- 170 143 122 161 137 sion formed by the loop between helices 3 and 4. It runs through the middle of the hydrophobic hairpin and is filled β ββ β111 97 7 6 5 4 β β by four water molecules. The walls of the channel are 94 formed mostly by hydrophobic residues with the excep- 3 2 1 tions being the sidechains of amino acids Trp365, Tyr347, 148 131 128 108 101 Tyr309, Thr362, Ser305 and Trp270 and the mainchain 157 93 155 carbonyls of Leu342 and Ser343. These residues contribute 149 N to the hydrogen bonding network that stabilizes the water α 5287 234 236 molecules inside the channel. 1 275 283 289

The receptor-binding domain 230 240 271 294 326 331 370 372 The receptor-binding domain consists of a six-stranded β sheet wrapped around an α helix, which is an extension α α α α α α α α of α-helix 1 from the pore-forming domain (Figure 1). In 2 34678910 total, this helix is 63 Å long. The topology and secondary structure of the receptor-binding domain of colicin N is shown in Figure 2. Regions showing increased tempera- ture factors, besides the N terminus, are located around 214 216 251 260 209 315 343 352 385 residues 140, 150 and 164. These regions correspond to C β loop regions connecting strands. There is no significant Structure sequence similarity of the receptor-binding domains of colicin N to the other colicins, although OmpF also serves A topology diagram of the receptor-binding and pore-forming domains as a co-receptor of colicin A [10]. A search for similar of colicin N. 866 Structure 1998, Vol 6 No 7

Figure 3

A comparison of colicins N and A. A stereoview of the α-carbon trace of the 170 170 superimposed colicin N (red trace) and colicin A (yellow trace; [41]). This figure was produced using MOLSCRIPT [70]. 160 160 120 120

180 140 180 140

110 110 100 150 100 150 N N 130 130 190 190

330 330 290 290

230 230 280 280 370 200 370 320240 200 320240 340 300 340 300

220 220 360270 250 360270 250 210 380 210 380 310 310

350 350 260 260 C C Structure

topologies in other proteins was conducted using the AW 68.1 (HLA), for which the root mean square deviation program FSSP [39]. The highest structural similarity was (rmsd) of mainchain atoms is 2.9 Å. There is no significant found with the human class I histo-compatibility antigen sequence similarity between colicin N and HLA, however.

Figure 4

A stereoview comparison of the hydrophobic helical hairpin of colicin A (yellow; [41]) and colicin N (green). For clarity only the α-carbon 332 332 backbones of the mainchain are shown and 370 370 the oxygen atoms of sidechain residues are coloured red. This figure was produced using 337 337 MOLSCRIPT [70].

364 364

344 344 360 360

346 346 356 356

348 348

Structure Research Article Colicin N structure Vetter et al. 867

Figure 5

A stereoview showing the water-filled channel of colicin N running through the centre of the molecule. Some sidechains are labeled for better orientation. The water molecules are shown as red spheres. One water molecule, which sits in the funnel-shaped entrance to the channel, cannot be seen in this view. Other water molecules are located in the protein core, but have been omitted in this figure for clarity. The surface was calculated using the Connolly algorithm [72].

Contrary to the HLA molecule, the β strands of the distribution of charged amino acids. During the structure colicin N receptor-binding domain resemble one half of a determination it was noted that heavy atom reagents were porin-like β barrel when viewed along the axis of central prone to bind in the cleft. α-helix 1 (Figure 1b). The relative arrangement of a β sheet wrapped around an α helix is also found in the pro- Contacts between the pore-forming and receptor-binding tease inhibitor cystatin [42], although the topology is dif- domains ferent. In cystatin, the α helix is connected to the opposite Besides the connecting α helix, there are few contacts end of the β sheet compared to colicin N. There is also between the receptor-binding and pore-forming domains. some superficial resemblance between the structures of There are two potential hydrogen bonding interactions: colicin N and pilin [43]. Although there is no detectable the mainchain amide of Lys129 to the mainchain oxygen sequence similarity between the two, both possess a long of Ile326 and the mainchain oxygen of Ser127 to the helix with a predominantly β-sheet-containing domain teth- mainchain amide of Gly328. In addition, there are six van ered to one end. But the β sheet domains are located at der Waals contacts. Thus, the interaction between the different ends of the helix, the number of strands differ, two domains is weak, probably allowing movements of the topology of the sheets is different, and pilin possesses the domains relative to each other in solution. Although a disulphide bridge. Intriguingly, however, both proteins the colicin N molecules in the crystal pack as head-to-tail translocate across the outer membrane of gram negative dimers with the possibility of introducing conformational bacterial cells, although in opposite directions. constraints on the inter-domain movement, it is not very likely that the weak interaction between the two domains The most prominent feature of the receptor-binding is an artifact; because the shapes of the two domains are domain is the existence of two large clefts formed by the so different, a fit yielding a sufficiently large contact twisted β sheet (Figure 1). One cleft is formed by the surface is impossible. The long helix connecting the two β sheet wrapping around the long helix and is located domains has one outstanding feature, namely a stretch of at the interface between the two secondary structures charged amino acids (Arg–Lys–Glu–Glu–Lys–Glu–Lys– (Figure 1b). This cleft is all hydrophobic and is particu- Asn–Glu–Lys–Glu) in the region between the two domains larly rich in aromatic residues. In the crystal structure, the (Arg184–Glu194). Colicin Ia also possesses a highly charged cleft is completely filled by the long helix. The other cleft stretch of sequence proceeding its pore-forming domain. is formed on the opposite side of the sheet to the first cleft The function of this stretch in either colicin remains to (Figure 1c). Unlike the first cleft, it is empty and fully be elucidated. solvent exposed. It contains a number of charged amino acid sidechains including two clusters of positively charged The translocation domain sidechains in the upper and lower parts of the cleft as well The translocation domain, located between residues 17 as two negatively charged residues in the centre of the and 66 [44], has a glycine/proline/serine/asparagine-rich cleft. It is tempting to assume that this cleft might be a sequence, which is likely to have poorly defined secondary ligand-binding pocket based on its shape and unusual structure. The translocation domain is absent in the 3.1 Å 868 Structure 1998, Vol 6 No 7

resolution structure, but the majority of it is present in A comparison of colicin receptor-binding and translocation the initial crystal form (see the Materials and methods domains section), which comprises residues 37–387. This region The receptor-binding domain of colicin N is a novel struc- includes residues 58–61, which has been shown recently to ture, quite unlike that found in the crystal structure of be essential for binding to TolA [45]. There is no evidence colicin Ia, which has been published recently [26]. But of any density for the translocation domain; this is in although colicin Ia belongs to the group B colicins and complete accord with circular dichroism and fluorescence bears no sequence similarity with colicin N outside the results, which demonstrate that this domain does not region of the pore-forming domains, there are a number of adopt any regular secondary structure [45,46]. interesting similarities in the structure and function of the two colicins. The receptor for colicin Ia is Cir, an outer Discussion that probably adopts a β-barrel topol- A comparison of colicin pore-forming domains ogy similar to that of OmpF [5]. The TonB translocation The structure of the pore-forming domain of colicin N machinery used by colicin Ia is analogous to the TolA is very similar to the pore-forming domain of colicin A system used by colicin N. Both TonB and TolA are thought (Figure 3; [24]) as expected from the high pairwise sequence to form extended rods that span the periplasm with an identity (56%) and very similar channel properties [22]. anchor in the periplasmic membrane. Accessory proteins The rmsd between the mainchain atoms of the pore- (TolQ/TolR and ExbB/ExbD) in each system show signif- forming domains of colicin N and the mainchain atoms of icant sequence similarity [3,8]. The C-terminal domains of the two monomers in the asymmetric unit of the colicin A TonB and TolA are thought to provide binding sites for crystal, denoted ‘A’ and ‘B’ respectively [41], is 1.31 Å colicin translocation domains [47]. and 1.25 Å (for these superpositions, residues 346–349 of colicin N and 162–166 of colicin A were omitted because In colicin Ia, the receptor-binding and pore-forming colicin N has a one residue deletion in this region). The domains are separated by a pair of 160 Å long helices. major differences in the superpositions (< 2 Å) were located This compares with the single long 63 Å helix in colicin N. at residues 215, 235 and 255 of colicin N compared to mol- The receptor-binding domains of the two colicins share a ecule B of colicin A and, additionally, at residues 285 and superficial resemblance in that they both consist of a 325 compared to molecule A of colicin A. The differences curved β sheet wrapped around a central, long α helix, in all these residues can be attributed to crystal contacts, which extends into the pore-forming domain to form the although differences in the first three residues might also first helix of this domain. In the case of colicin Ia, however, reflect genuine differences. the β sheet is composed of only two strands. Unlike colicin N, the much larger translocation domain of colicin Ia Although the sequence similarities of colicin N to the partially adopts regular secondary structure: three long pore-forming domains of colicins E1 and Ia are quite low helices. Interestingly, the TonB box is located in an (26% and 22%, respectively), they nevertheless exhibit extended coil of ~70 residues at the N terminus, which is very similar folds [25,26]. As well as the canonical ten about the same size as the entire colicin N translocation helix bundle, the first helix is found to be extended to domain. Thus, in both colicins the precise region thought form a bridge to the receptor-binding domain present in to interact with the translocation machinery adopts no the intact toxin. Other common features include the inter- structure and this flexibility maybe a key part of the nal cavities surrounding the hydrophobic hairpin and recognition process. patches of positive charge on the surface of the molecule. The internal cavities might be a key to structural rearrange- Implications for the interaction of colicin N with OmpF ments required for membrane insertion, whereas the Colicin N has only one receptor, the matrix porin OmpF. patches of positive charge may be used for initial docking This is in contrast to colicin A, which requires two differ- of the toxin molecule onto the membrane [41]. A signifi- ent receptors, OmpF and BtuB, both located in the outer cant difference between colicin N and colicins E1 and Ia [48]. Colicin N requires its receptor for full is a shorter hydrophobic helical hairpin in colicins E1 and activity in vivo and the related porins, PhoE and OmpC, Ia. In contrast to colicin N, colicin Ia has three additional are associated with greatly reduced cellular sensitivities to helices from the translocation domain, which are packed colicin N [49,50]. Colicin N can bind to the three porins against the ten helices of the pore-forming domain [26]. with similar affinity in vitro indicating that the complexes Because the structures of the pore-forming domains of all with PhoE and OmpC are poorly competent for transloca- the colicins are very similar, it is highly likely that their tion in vivo [51]. The OmpF protein has been closely inves- mechanism of membrane insertion and channel forma- tigated for its involvement in the reception and translo- tion are also broadly similar. Detailed proposals for the cation steps of colicin N entry [50,52–54] and its crystal mechanisms have been put forward and the interested structure is known at 2.4 Å resolution [37]. Although all reader is directed towards recent papers and reviews on the domains of colicin N are mainly basic, the OmpF extra- the subject [1–5,25,32,41]. cellular loops are mainly acidic. Several regions of OmpF Research Article Colicin N structure Vetter et al. 869

Figure 6

The proposed model for colicin N receptor-binding and translocation. membrane and binding to OmpF via its receptor-binding domain. The The trimeric porin, OmpF, is shown in the outer membrane and Tol translocation domain threads through the porin pore to interact with proteins are shown projecting into the periplasm from the inner TolA in the periplasm. The pore-forming domain subsequently membrane. For simplicity, the receptor-binding domain of colicin N is translocates to the inner membrane. The closed form of the colicin N shown as a cusp-shaped object and the pore-forming domain is channel is shown in the last panel. Application of a transnegative indicated by the cylindrical rods that represent the helices of that voltage would cause the helices lying on the surface of the membrane domain. The colicin N molecule is shown approaching the target to insert and form the open channel.

have been identified that are necessary for the receptor- 285 of OmpF, and the translocation domain reaches down binding and the translocation steps of colicin N. Point to, or past, residue 119. The solvent-exposed cleft in the mutations of Glu284 and Gly285, which are located on the colicin N receptor-binding domain forms interactions with outer rim of the porin channel, indicate that these residues the porin and there is a complementary interaction between are necessary for colicin N binding, but not for transloca- the positively charged surface of the colicin and the nega- tion [53]. Another colicin N-resistant point mutant is tively charged surface provided by the porin loops. There is Gly119→Asp, which is located in a loop within the porin experimental evidence that the translocation and the pore- pore, suggesting that the binding site is either deep inside forming domains contribute significantly to the binding the pore or that the pore represents part of the transloca- affinity: the receptor-binding domain (R) of colicin N alone tion pathway [53]. It is clear from looking at the size and or attached to the translocation domain (RT) have been shape of the colicin N molecule that the receptor-binding overexpressed, purified and their affinity for OmpF, both in domain cannot possibly bind to residue 119. Even if one vivo and in vitro, were compared to the full-length colicin N postulated conformational changes of OmpF in the region and the fragment studied here (RP; [46]). All the two- of residue 119, the domain cannot bind to residue 119 and domain fragments display a weaker affinity for OmpF than residues 284 and 285 simultaneously. The crystal structure full-length colicin N, suggesting that molecular recognition of the OmpF mutant, Gly119→Asp, is known and shows is a property of the receptor domain but affinity is influ- only minor conformational changes [55]. Indeed, the muta- enced by the entire molecule. This model also predicts tion subdivides the existing pore into two smaller pores, that the OmpF pore would close on colicin N binding, suggesting that the effect of the mutation is to block the which agrees with experiments in which the isolated passage of some part of colicin N. R domain is found to block OmpF channels (J.H.L. and F.P., unpublished observations). These results suggest only one model of binding (Figure 6). In this model the colicin N receptor-binding domain sits The translocation process and channel formation like a plug above the pore of OmpF, with the pore-forming The Tol machinery is located at adhesion zones between domain interacting with the outer rim and residues 284 and the two bacterial membranes. TolA is anchored through 870 Structure 1998, Vol 6 No 7

a single transmembrane stretch to the cytoplasmic mem- helices 1–3 could form a single helix long enough for the brane. The C-terminal domain of ~120 residues is pre- toxin to insert in the cytoplasmic membrane and the dicted to sit at the end of an enormously long helix receptor-binding domain can, if required, remain bound that projects into the periplasm [8]. It is the C-terminal to the outside of the cell. (Because TolA is located at domain that interacts with the translocation domain of adhesion zones, it is likely that colicin N translocates at group A colicins [14,56]. Because the C-terminal domain these sites. Thus, the required separation between the of TolA resides in the periplasm, the translocation domain receptor-binding and pore-forming domains could be less of colicin N must somehow traverse the outer membrane. for colicin N than for other colicins, which translocate at Furthermore, the proximity of the receptor-binding domain different sites.) The N-terminal half of helix 1 is suffi- to the TolA binding site, which is located at residues ciently long to span a membrane and it is amphiphilic so 59–62 [45], also provides a limitation on the distance the hydrophobic face could conceivably interact with the between TolA and the receptor-binding domain of < 90 Å. hydrophobic core of the outer membrane. These struc- There appear to be three possible distinct routes. The tural rearrangements are supported by calorimetry exper- first is along the outside of the porin barrel, but this iments, which reveal thermodynamics consistent with large would require direct interaction with the lipid bilayer, structural changes occurring to colicin N on binding to which is energetically unfeasible. Second, the domain OmpF [51] even though the receptor-binding and pore- could slip through the porin pore or between a transient forming domains are very stable [46]. The model is also pore created between the monomers of the porin trimer. reminiscent of the model for colicin Ia, which is also pos- The third possibility requires destablization of the tulated to span the periplasmic space by virtue of the monomer interfaces, for which there is no experimental unusually long helices that bridge the receptor-binding evidence. Indeed, the porin trimer is very stable and can and pore-forming domains in that molecule [26]. only be dissociated in SDS at temperatures > 75°C [57]. Because the translocation domain appears to form a highly The mechanism by which the pore-forming domain tra- flexible coil, it is easy to envisage that the 66 residue verses the outer membrane is puzzling. It doesn’t seem domain could poke through the porin pore so as to make likely that it can pass through the porin pore. First, the contact with the C-terminal domain of TolA, which is pore would be plugged by the translocation and receptor- located underneath (Figure 6). This is also in agreement binding domains. Second, to fit through the pore, the with the model presented above in which the transloca- toxin would require complete unfolding. Although the tion domain was proposed to interact with the OmpF porin pore is sufficiently wide to allow the passage of the pore in the initial receptor-binding step. unstructured translocation domain, it is not wide enough to allow the passage of fully folded receptor-binding or The final step of the translocation process would be the pore-forming domains. There is no obvious source to passage of other parts of the toxin across the outer mem- promote complete unfolding of these domains and previ- brane. Proteolysis experiments with colicin A suggest ous studies have shown that both receptor-binding and that the receptor-binding domain remains attached to pore-forming domains are very stable towards denaturants OmpF after channel formation in the inner membrane [46]. Furthermore, colicin A mutants in which disulphide [35]. Engineering disulphide bonds between helices 1 and bonds have been engineered between various helices of 2 and between helices 3 and 10 of the pore-forming the pore-forming domain can still be translocated through domain of colicin A blocked the in vivo activity of the the outer membrane [36,58]. There are two alternative toxin without affecting pore formation in artificial mem- possibilities: transient destabilization of the porin trimer to branes. These covalent linkages and others in the pore- create a pore large enough for the pore-forming domain to forming domain (between helices 2 and 10 and between slip through; or translocation across the outer surface of helices 3 and 9) did not stop the translocation process the trimer. The energy costs required for either pathway across the outer membrane. These observations lead to could be provided by the toxin binding to the receptor and the suggestion that helices 1–3 must span the periplasm the translocation domain ‘switching’ on the translocation of E. coli when the pore-forming domain is inserted into machinery upon binding to TolA. Hybrids of OmpF, and the inner membrane [58]. There is presently no evidence the related porin OmpC, which bind colicin N but are not as to whether colicin N also spans both membranes and active in colicin N uptake, have been used to map the region the only proteolysis studies on colicin N to date lend no of OmpF required for translocation [52]. The region was support to the suggestion [54]. If colicin N did span the localised on the surface of the porin trimer directly below membrane, helix 1 is probably not long enough to span the loop containing residues 284 and 285, which have been the entire periplasmic space in order to allow insertion of implicated in binding (see above), and largely separated the pore-forming domain of colicin N whilst part of the from the pore lumen by an internal long loop (loop L3). molecule is still bound to OmpF. It has been shown, These data support passage across the outer surface of the however, that helices 4–10 are sufficient for channel for- trimer. Presumably, the Tol translocation machinery medi- mation [59]. We suggest, by analogy to colicin A, that ates this passage across the outer membrane. Research Article Colicin N structure Vetter et al. 871

If, as proposed for colicin A, the receptor-binding domain its chymotryptic fragment have been described elsewhere [44]. Crystals remains bound to OmpF, we have suggested a model that of the chymotryptic fragment of colicin N were grown by the hanging drop method at room temperature. The protein concentration was allows the translocation and the pore-forming domains to 10 mg/ml and the reservoir solution consisted of 50 mM phosphate translocate to the periplasmic space. It explains the obser- buffer (pH 5.6–7.0) and 60% (w/v) saturated ammonium sulphate. Small vation that on initial binding to the receptor, the transloca- crystals appeared after 1–2 days and grew to their full size of up to × × tion domain of colicin N is prone to proteolysis although it 0.6 mm 0.6 mm 0.6 mm in 1–2 weeks. is resistant to proteolysis upon translocation [54]. On Data collection and processing initial binding, the translocation domain would be accessi- The full-length colicin N-derived crystals belong to the space group ble to proteases, whereas upon translocation it would have P4132. A low-resolution data set was collected to 6 Å resolution (R = 8.9% and 97% complete). The crystals obtained from the chy- passed across the membrane. Our current model for colicin sym motryptic fragment belong to the cubic spacegroup I4132 with unit cell toxicity is summarized in Figure 6. parameters of a = b = c = 187.3 Å and with one molecule per asym- metric unit. The best native data set (to 3.1 Å resolution) was collected Biological implications on the X-31 beamline at the Deutsche Elektronen Synchrotron (DESY) Colicins are a distinctive class of antibiotics produced by at the European Molecular Biology Laboratory Outstation, Hamburg, Germany. The wavelength was set to 1.009 Å and data were collected various strains of Escherichia coli. When induced through using a Hendrix-Lentfer imaging plate scanner. Processing was per- the SOS system, they are secreted into the extracellular formed using the CCP4 program suite [61] and the data statistics medium with the aid of a lysis protein. Following secre- are presented in Table 1. Diffraction data for the heavy-atom search tion, the toxin molecules bind to metabolite receptors on (Table 1) were collected on a multiwire area detector (Nicolet/Xentron- the outer membrane of sensitive cells, thereby par- ics, Madison, USA) or on a FAST area detector system (Enraf-Nonius, Delft, The Netherlands) using a rotating anode X-ray generator. The asitizing these bacterial transport systems and using data sets were processed using the XDS software suite [62]. them as entry ports. The translocation across the peri- plasm is aided by some of the target cell’s transport Preparation of heavy-atom derivatives and Patterson search machinery. Subsequent killing of the bacterial cell is Attempts to solve the colicin N structure by molecular replacement methods using the crystal structure of the colicin A pore-forming domain achieved by a variety of means, including DNase activ- as a model failed. No solution for the function could be found, ity, RNase activity or pore formation. Pore-forming col- even though the rotation function showed significant peaks. For the icins have proved to be powerful model systems for heavy-atom derivative searches, crystals were transferred to a synthetic investigating the mechanisms of protein translocation, mother liquor containing 70% ammonium sulphate and 60 mM phos- phate buffer pH 6.4 or 70% ammonium sulphate and 50 mM acetate channel formation and function. buffer pH 5.6. Various heavy-atom reagents were added in concentra- tions of 2–30 mM and the crystals were soaked in these solutions for Colicin N is the shortest of the known colicins, but still 1–8 days (Table 1). More than 40 different reagents were tested. The retains all the functionalities of its relations. Because of best heavy-atom derivative was obtained by soaking crystals for 2 days in saturated K PtI . This was the first derivative solved using the program its relative simplicity, it is an ideal system for studying 2 6 GROPAT [63]. The correct solution was the third best solution in the colicin translocation mechanisms. In addition, the entry ranked program output. Because this derivative had a significant anom- route into the target cell used by colicin N is the same as alous signal, other derivative sites could be located via difference Fourier that used by filamentous bacteriophages and hence work calculations using the (at that time apparently single) Pt site. At this on colicin N is also likely to shed light on phage infection. stage, five derivatives turned out to be useful for the initial phasing (Table 1). The first MIR map was calculated with an initial figure-of-merit We have solved the crystal structure of a large fragment of 0.8 for 3,180 reflections (20–4.5 Å). In the initial map, the α helices of of colicin N which includes the receptor-binding and the pore-forming domain were already visible as thick rods. In the course pore-forming domains. The pore-forming domain adopts of the refinement, more heavy atom sites were successively included. In the canonical globin-like fold of other colicin pore- forming domains, whereas the receptor-binding domain adopts a novel fold. Based on the structure and the previ- Table 1 ously determined crystal structure of its receptor, OmpF, Data collection statistics. it has been possible to produce a plausible model of how colicin N traverses two membranes to form a voltage- Resolution Unique Completeness Rsym* Number gated ion channel. Data (Å) reflections (%) (%) MFID† of sites

Materials and methods Native 3.1 10,644 93.5 8.3 – – K PtI 4.3 3,869 92.9 22.2 24.4 2 Protein purification and crystallization 2 6 Purification of colicin N was carried out as previously described [60]. K2PtCl4 4.3 3,774 90.6 17.0 21.9 3 Initial crystals were obtained from full-length colicin N. These crystals UO2(OAc)2 3.4 6,572 84.0 14.0 12.6 1 were grown by the hanging drop method at room temperature using KOsO4 4.1 4,082 86.3 13.1 12.5 1 reservoir conditions of 50 mM phosphate pH 7.0, 40% (w/v) ammonium KAu(CN) 3.0 8,237 72.0 22.4 15.8 2 sulphate and 0–5% PEG 6000. These crystals only diffracted to just 2 Σ Σ 〈 〉 Σ Σ 〈 〉 beyond 6 Å resolution, however. These crystals took several months to *Rmerge = hkl i |Ii – I |/ hkl i | I |, where Ii is the intensity for the ith appear and analysis of the crystals showed that a proteolytic cleavage measurement of an equivalent reflection with indices h,k,l. †MFID Σ Σ had occurred after residue 36. Attempts were then made to crystallize a (mean fractional isomorphous difference) = hkl |FPH–FP|/ hklFP, where proteolytic fragment of the toxin. The production and characterization of FPH refers to derivative data and Fp to native data. 872 Structure 1998, Vol 6 No 7

Table 2 model. This procedure was repeated several times, combining the model phases with the heavy-atom phases in every step using the Final refinement statistics. program SIGMAA [65] so that the danger of model bias was mini- mized. During this procedure, the electron density of the receptor- Parameter Value binding domain improved considerably, so that density modification methods (e.g. solvent flattening) were not needed. For the first 24 Resolution range (Å) 23.0–3.1 N-terminal amino acids no density was visible, although the density for Number of reflections used 10,644 the rest of the molecule was well-defined and unambiguous. Total number of atoms (excluding H) 4,772 For model refinement, both positional refinement and simulated anneal- R (23–3.1 Å)* (%) 17.3 crys ing refinement were applied using X-PLOR [66]. The final crystallo- R (%) 31.6 free graphic R-factor after refinement with X-PLOR was 19.9% (for 9,099 2 Average B-factors (Å ) 34.6 reflections > 2 sigma, 6.0-3.1 Å). The final refinement was performed Average B-factor for mainchain atoms (Å2) 29.4 using the program TNT [67] with all the data from 23 Å to 3.1 Å, con- Average B-factor for sidechain atoms (Å2) 39.7 taining 10,966 unique reflections. The temperature factors were refined Rmsd for bond lengths (Å) 0.007 with correlation restraints because of the insufficient resolution of the Rmsd for angles (°) 1.4 data. The crystallographic R-factor after the TNT refinement was 17.3%, the free R-factor was 31.6% (10% of reflections used for the test set), Rmsd for B-factors (Å2) 3.9 and the model exhibited excellent geometry (Table 2). The final model Σ Σ *Rcrys = hkl ||Fobs|–|Fcalc|| / hkl ||Fobs|, where Fobs denotes the observed includes 59 solvent molecules. The final map is of good quality (Figure 7). structure factor amplitude and Fcalc denotes the structure factor All heavy atoms bind to reasonable chemical sites and hence corrobo- amplitude calculated from the model. 10% of reflections were used to rate the structure. The final structure was analysed using PROCHECK calculate Rfree. [68]. The Ramachandran plot shows that all non-glycine residues fall in either the allowed (92.8%) or additional allowed regions (7.2%). Other stereochemical parameters such as peptide-bond planarity, α-carbon the end, it turned out that 21 of the tested compounds had yielded useful tetrahedral distortions and non-bonded interactions are all inside or derivatives, although only 18 were used in the final phase calculation. better than the allowed ranges according to PROCHECK. In addition to (Only the best five derivatives are shown in Table 1 because the others the PROCHECK tests, the veracity of the model was assessed using only made a marginal contribution to the phasing). In total, there were 38 the 3D_1D profile analysis program package [69]. The three-dimen- sites of which 19 were different. sional profile scores were better than 0.3 in all regions of the protein, corroborating the correctness of the model. The final model was used to Structure refinement solve the structure of the full-length-derived colicin N crystals by mol- In the first step of the model building, the pore-forming domain of ecular replacement. After rigid body refinement, the free R-factor was colicin N was modelled using the structure of the homologous colicin A 29.4%. If the data from these crystals were phased only with the pore- domain [41] as a starting structure. This was helpful in speeding up forming domain, then the free R-factor was only 37.0%, but the density the refinement process because the electron density maps calculated for the secondary structural elements of the receptor-binding domain from the MIR phases were not readily interpretable. Modelling was could clearly be seen. At no stage was there any evidence for density performed using the program O [64]. The phases resulting from this corresponding to the translocation domain, which is in full agreement initial model were combined with the heavy-atom phases and an with circular dichroism results, which show that this domain does not improved map was calculated, followed by an adjustment of the adopt any regular structure [45,46].

Figure 7

The quality of the final colicin N model. Stereoview showing an omit map of residues 338–352. The residues in this region were omitted from the model and the map was calculated after a round of simulated annealing. The map is contoured at a level of 1.2σ. Part of the α-carbon trace of colicin A (green) is superimposed [41]. This figure was produced using BOBSCRIPT [73]. Research Article Colicin N structure Vetter et al. 873

Accession numbers 23. Ghosh, P., Mel, S.F. & Stroud, R.M. (1993). A carboxy-terminal The atomic coordinates of colicin N have been deposited in the fragment of colicin Ia forms ion channels. J. Membr. Biol. 134, 85-92. Brookhaven Protein Data Bank with the accession code 1a87. 24. Parker, M.W., Pattus, F., Tucker, A.D. & Tsernoglou, D. (1989). Structure of the membrane-pore-forming fragment of colicin A. Nature 337, 93-96. Acknowledgements 25. Elkins, P., Bunker, A., Cramer, W.A. & Stauffacher, C.V. (1997). A We thank the staff at the EMBL Outstation, DESY synchrotron, Hamburg, mechanism for toxin insertion into membranes is suggested by the Germany for help with data collection. M.W.P. is an Australian Research crystal structure of the channel-forming domain of colicin E1. Council Senior Research Fellow and acknowledges support from the National Structure 5, 443-458. Health & Medical Research Council. J.H.L. thanks the Wellcome Trust for 26. Wiener, M., Freymann, D., Ghosh, P. & Stroud, R.M. (1997). Crystal financial support. structure of colicin Ia. Nature 385, 461-464. 27. van der Goot, F.G., Gonzalez-Mañas, J.M., Lakey, J.H. & Pattus, F. References (1991). A ‘molten-globule’ membrane-insertion intermediate of the 1. Braun, V., Pilsl, H. & Gross, P. (1994). Colicins: structures, modes of pore-forming domain of colicin A. Nature 354, 408-410. action, transfer through membranes, and evolution. Arch. Microbiol. 28. Merrill, A.R., Palmer, L.R. & Szabo, A.G. (1993). Acrylamide quenching 161, 199-206. of the intrinsic fluorescence of tryptophan residues genetically 2. Lakey, J.H., van der Goot, F.G. & Pattus, F. (1994). All in the family: engineered into the soluble colicin E1 channel peptide. Structural the toxic activity of pore-forming colicins. Toxicology 87, 85-108. characterization of the insertion-competent state. Biochemistry 32, 3. Cramer, W.A., et al., & Stauffacher, C.V. (1995). Structure-function of 6974-6981. the channel-forming colicins. Annu. Rev. Biophys. Biomol. Struct. 24, 29. Lakey, J.H., Duche, D., Gonzalez-Mañas, J.M., Baty, D. & Pattus, F. 611-641. (1993). Fluorescence energy transfer distance measurements. The 4. Vetter, I.R., Parker, M.W., Pattus, F. & Tsernoglou, D.T. (1996). hydrophobic helical hairpin of colicin A in the membrane bound state. Insights into membrane insertion based on studies of colicins. In J. Mol. Biol. 230, 1055-1067. Protein Toxin Structure. (Parker, M.W., ed.), pp. 5-23, R.G. Landes 30. Shin, Y.K., Levinthal, C., Levinthal, F. & Hubbell, W.L. (1993). Company, Austin, Texas. Colicin E1 binding to membranes: time-resolved studies of spin- 5. Gouaux, E. (1997). The long and short of colicin action - the molecular labeled mutants. Science 259, 960-963. basis for the biological activity of channel-forming colicins. Structure 5, 31. Evans, L., Goble, M.L., Hales, K.A. & Lakey, J.H. (1996). Different 313-317. sensitivities to acid denaturation within a family of proteins - 6. Braun, V. (1989). The structurally related exbB and tolQ are implications for acid unfolding and membrane translocation. interchangeable in conferring tonB-dependent colicin, bacteriophage, Biochemistry 35, 13180-13185. and albomycin sensitivity. J. Bacteriol. 171, 6387-6390. 32. Kienker, P.K., Qiu, X.Q., Slatin, S.L., Finkelstein, A. & Jakes, K.S. 7. Postle, K. (1993). TonB protein and energy transduction between (1997). Transmembrane insertion of the colicin Ia hydrophobic hairpin. membranes. J. Bioenerg. Biomembr. 25, 591-601. J. Membr. Biol. 157, 27-37. 8. Webster, R.E. (1991). The tol products and the import of 33. Slatin, S.L., Qiu, X.Q., Jakes, K.S. & Finkelstein, A. (1994). macromolecules into Escherichia coli. Mol. Microbiol. 5, 1005-1011. Identification of a translocated protein segment in a voltage- 9. Vianney, A., et al., & Webster, R.E. (1996). Characterization of the dependent channel. Nature 371, 960-963. tol/pal region of E. coli K12. J. Bacteriol. 178, 4031-4038. 34. Qiu, X.Q., Jakes, K.S., Kienker, P.K., Finkelstein, A. & Slatin, S.L. 10. Bénédetti, H., et al., & Lazdunski, C. (1989). Comparison of the (1996). Major transmembrane movement associated with colicin Ia uptake systems for the entry of various BtuB group colicins into channel gating. J. Gen. Physiol. 107, 313-328. Escherichia coli. J. Gen. Microbiol. 135, 3413-3420. 35. Bénédetti, H., Lloubès, R., Lazdunski, C. & Letellier, C. (1992). 11. Schendel, S.L., Click, E.M., Webster, R.E. & Cramer, W.A. (1997). Colicin A unfolds during its translocation in Escherichia coli cells and The TolA protein interacts with colicin E1 differently than with other spans the whole cell envelope when its pore has formed. EMBO J. 11, group A colicins. J. Bacteriol. 179, 3683-3690. 441-447. 12. Bouveret, E., Rigal, A., Lazdunski, C. & Benedetti, H. (1997). The 36. Duché, D., Baty, D., Chartier, M. & Letellier, L. (1994). Unfolding of N-terminal domain of colicin E3 interacts with Tolb which is involved in colicin A during its translocation through the Escherichia coli envelope the colicin translocation step. Mol. Microbiol. 23, 909-920. as demonstrated by disulfide bond engineering. J. Biol. Chem. 269, 13. Pugsley, A.P. (1985). Escherichia coli K12 strains for use in the 24820-24825. identification and characterization of colicins. J. Gen. Microbiol. 131, 37. Cowan, S.W., et al., & Rosenbusch, J.P. (1992). Crystal structures 369-376. explain functional properties of two E. coli porins. Nature 358, 727-733. 14. Pilsl, H. & Braun, V. (1995). Novel colicin 10: assignment of four 38. Holm L. & Sander C. (1993). Structural alignment of globins, domains to TonB- and TolC-dependent uptake via the Tsx receptor phycocyanins and colicin A. FEBS Lett. 315, 301-306. and to pore formation. Mol. Microbiol. 16, 57-67. 39. Holm L. & Sander C. (1994). The FSSP database of structurally 15. Schramm, E., Mende, J., Braun, V. & Kamp, R.M. (1987). Nucleotide aligned protein fold families. Nucleic Acids Res. 22, 3600-3609. sequence of the colicin B activity gene cba: consensus pentapeptide 40. Minn, A.J., et al., & Thompson, C.B. (1997). Bcl-xL forms an ion among TonB-dependent colicins and receptors. J. Bacteriol. 169, channel in synthetic lipid membranes. Nature 385, 353-357. 3350-3357. 41. Parker, M.W., Postma, J.P.M., Pattus, F., Tucker, A.D. & Tsernoglou, D. 16. Boeke, J.D., Model, P. & Zinder, N.D. (1982). Effects of bacteriophage (1992). Refined structure of the pore-forming domain of colicin A at f1 gene III protein on the host cell membrane. Mol. Gen. Genet. 186, 2.4 Å resolution. J. Mol. Biol. 224, 639-657. 185-192. 42. Bode, W., et al., & Turk, V. (1988). The 2.0 Å X-ray crystal structure of 17. Mende, J. & Braun, V. (1990). Import-defective colicin B derivatives chicken egg white cystatin and its possible mode of interaction with mutated in the TonB box. Mol. Microbiol. 4, 1523-1533. cysteine proteinases. EMBO J. 7, 2593-2599. 18. Guihard, G., Bénédetti, H., Besnard, M. & Letellier, L. (1993). 43. Parge, H.E., et al., & Tainer, J.A. (1995). Structure of the fibre-forming Phosphate efflux through the channels formed by colicins and phage protein pilin at 2.6 Å resolution. Nature 378, 32-38. T5 in Escherichia coli cells is responsible for the fall in cytoplasmic 44. El Kouhen, R., et al., & Pagès, J-M. (1993). Characterization of the ATP. J. Biol. Chem. 268, 17775-17780. receptor and translocation domains of colicin N. Eur. J. Biochem. 214, 19. Schein, S.J., Kagan, B.L. & Finkelstein, A. (1978). Colicin K acts by 635-639. forming voltage-dependent channels in phospholipid bilayer 45. Raggett, E.M., Bainbridge, G., Evans, L.J.A., Cooper, A. & Lakey, J.H. membranes. Nature 276, 159-163. (1998). Discovery of critical TolA-binding residues in the bactericidal 20. Martinez, M.C., Lazdunski, C. & Pattus, F. (1983). Isolation, molecular toxin colicin N: a biophysical approach. Mol. Microbiol., in press. and functional properties of the C-terminal domain of colicin A. 46. Evans, L.J.A., Labeit, S., Cooper, A., Bond, L.H. & Lakey, J.H. (1996). EMBO J. 2, 1501-1507. The central domain of colicin N possesses the receptor recognition 21. Slatin, S.L. (1988). Colicin E1 in planar lipid bilayers. Int. J. Biochem. site but not the binding affinity of the whole toxin. Biochemistry 35, 20, 737-744. 15143-15148. 22. Wilmsen, H.U., Pugsley, A.P. & Pattus, F. (1990). Colicin N forms 47. Lazdunski, C. (1995). Colicin import and pore formation: a system for voltage- and pH-dependent channels in planar lipid bilayer studying protein transport across membranes? Mol. Microbiol. 16, membranes. Eur. Biophys. J. 18, 149-158. 1059-1066. 874 Structure 1998, Vol 6 No 7

48. Baty, D., et al., & Lazdunski, C. (1990). Uptake across the cell envelope and insertion into the inner membrane of ion channel-forming colicins in E. coli. Biochimie 72, 123-130. 49. Tommassen, J., Pugsley, A.P., Korteland, J., Verbakel, J. & Lugtenberg, B. (1984). Gene encoding a hybrid OmpF-PhoE pore protein in the outer membrane of Escherichia coli K12. Mol. Gen. Genet. 197, 503-508. 50. Bourdineaud, J.-P., Fierobe, H.P., Lazdunski, C. & Pagès, J.M. (1990). Involvement of OmpF during reception and translocation steps of colicin N entry. Mol. Microbiol. 4,1737-1743. 51. Evans, L.J.A., Cooper, A. & Lakey, J.H. (1996). Direct measurement of the association of a protein with a family of membrane receptors. J. Mol. Biol. 255, 559-563. 52. Fourel, D., Hikita, C., Bolla, J.-M., Mizushima, S. & Pagès, J.-M. (1990). Characterization of OmpF domains involved in Escherichia coli K-12 sensitivity to colicins A and N. J. Bacteriol. 172, 3675-3680. 53. Fourel, D., Mizushima, S., Bernadac, A. & Pagès, J.-M. (1993). Specific regions of Escherichia coli OmpF protein involved in antigenic and colicin receptor sites and in stable trimerization. J. Bacteriol. 175, 2754-2757. 54. El Kouhen, R. & Pagès, J.-M. (1996). Dynamic aspects of colicin N translocation through the Escherichia coli outer membrane. J. Bacteriol. 178, 5316-5319. 55. Jeanteur, D., et al., & Pagès, J.M. (1994). Structural and functional alterations of a colicin-resistant mutant of OmpF-porin from Escherichia coli. Proc. Natl Acad. Sci. USA 91, 10675-10679. 56. Bénédetti, H., Lazdunski, C. & Lloubès, R. (1991). Protein import into Escherichia coli: colicins A and E1 interact with a component of their translocation system. EMBO J. 10, 1989-1995. 57. Fourel, D., Bernadac, A. & Pagès, J.M. (1994). Involvement of exposed loops for the trimeric stability and for membrane insertion of the Escherichia coli OmpF porin. Eur. J. Biochem. 222, 625-630. 58. Duché, D., et al., & Baty, D. (1996). Membrane topology of the colicin A pore-forming domain analyzed by disulfide bond engineering. J. Biol. Chem. 271, 15401-15406. 59. Baty, D., Lakey, J., Pattus, F. & Lazdunski, C. (1990). A 136 amino acid residue COOH-terminal fragment of colicin A is endowed with ionophoric activity. Eur. J. Biochem. 189, 409-413. 60. Massotte, D. & Pattus, F. (1989). Colicin N and its thermolytic fragment induce phospholipid vesicle fusion. FEBS Lett. 257, 447-450. 61. Collaborative Computational Project No. 4 (1994). The CCP4 suite: programs for protein crystallography. Acta Crystallogr. D 50, 760-763. 62. Kabsch, W. (1988). Evaluation of single crystal X-ray diffraction data from a photon sensitive detector. J. Appl. Crystallogr. 21, 916-924. 63. Jones, E.Y., Walker, N.P.C. & Stuart, D.I. (1991). Methodology employed for the structure determination of tumour necrosis factor, a case of high non-crystallographic symmetry. Acta Crystallogr. A 47, 753-770. 64. Jones, T.A., Zou, J.-Y., Cowan, S.W. & Kjeldgaard, M. (1991). Improved methods for building protein models in electron density maps and the location of error in these models. Acta Crystallogr. A 42, 110-119. 65. Read, R.J. (1986). Improved Fourier coefficients for maps using phases from partial structures with errors. Acta Crystallogr. A 42, 140-149. 66. Brünger, A.T. (1992). X-PLOR, version 3.1: a system for X-ray crystallography and NMR. Yale University Press, New Haven, CT. 67. Tronrud, D.E., Ten Eyck, L.F. & Matthews, B.W. (1987). An efficient general-purpose least-squares refinement program for macromolecular structures. Acta Crystallogr. A 43, 489-501. 68. Laskowski, R.A., MacArthur, M.W., Moss, D.S. & Thornton, J.M. (1993). PROCHECK - a program to check the stereochemical quality of proteins. J. Appl. Crystallogr. 26, 283-291. 69. Lüthy, R., Bowie, J.U. & Eisenberg, D. (1992). Assessment of protein models with three-dimensional profiles. Nature 356, 83-85. 70. Kraulis, P.J. (1991). MOLSCRIPT - a program to produce both detailed and schematic plots of protein structures. J. Appl. Crystallogr. 24, 946-950. 71. Nicholls, A., Sharp, K.A. & Honig, B. (1991). Protein folding and association: insights from the interfacial and thermodynamic properties of hydrocarbons. Proteins 11, 281-293. 72. Connolly, M.L. (1985). Computation of molecular volume. J. Am. Chem. Soc. 107, 1118-1124. 73. Esnouf, R.M. (1997). An extensively modified version of Molscript that includes greatly enhanced coloring capabilities. J. Mol. Graphics 15, 132-134.