The outer membrane of Gram-negative bacteria contains a large number of channel-forming proteins, porins, for the uptake of small nutrient molecules. Neisseria gonorrhoeae PorBIA (PorB of serotype A) are associated with disseminating diseases and mediate a rapid bacterial invasion into host cells in a phosphate-sensitive manner. To gain insights into this structure–function relationship we analysed PorBIA by X-ray crystallography in the presence of phosphate and ATP. The structure of PorBIA in the complex solved at a resolution of 3.3 Å (1 Å=0.1 nm) displays a surplus of positive charges inside the channel. ATP ligand-binding in the channel is co-ordinated by the positively charged residues of the channel interior. These residues ligate the aromatic, sugar and pyrophosphate moieties of the ligand. Two phosphate ions were observed in the structure, one of which clamped by two arginine residues (Arg92 and Arg124) localized at the extraplasmic channel exit. A short β-bulge in β2-strand together with the long L3 loop narrow the barrel diameter significantly and further support substrate specificity through hydrogen bond interactions. Interestingly the structure also comprised a small peptide as a remnant of a periplasmic protein which physically links porin molecules to the peptidoglycan network. To test the importance of Arg92 on bacterial invasion the residue was mutated. In vivo assays of bacteria carrying a R92S mutation confirmed the importance of this residue for host-cell invasion. Furthermore systematic sequence and structure comparisons of PorBIA from Neisseriaceae indicated Arg92 to be unique in disseminating N. gonorrhoeae thereby possibly distinguishing invasion-promoting porins from other neisserial porins.
The outer membrane of Gram-negative bacteria is strongly asymmetric and consists of integral membrane proteins of β-barrel architecture [OMPs (outer membrane proteins)], lipoproteins, lipopolysaccharides and simple phospholipids. In most cases this membrane constitutes the outermost protective layer of the microbe, with only some bacteria additionally covered by an S-layer or a dense layer of autotransporters (such as in certain Bartonella species) [1,2]. Physical linkage of the outer membrane to the bacterial PG (peptidoglycan) layer and preservation of bacterial morphology is mediated by a dense network of interactions. These contacts are mediated by integral OMPs carrying PG-binding domains such as OmpA, periplasmic connector proteins such as RmpN non-covalently bind OMP proteins and PG, or lipoproteins, most importantly the Brown lipoprotein [3–6].
Integral membrane proteins of the outer membrane serve a variety of different functions, most of which comprise the uptake of ions, small nutrient molecules and metal–siderophore complexes [7,8]. The most abundant protein of this membrane is the bacterial porin, a diffusion channel functionally responsible for the exchange of small metabolites with a ~600 Da molecular mass cut-off determined by the channel diameter [9,10]. Originally two types of porins were classified by structure and function as either 16-stranded unspecific or 18-stranded specific porins [9,11,12]. Previously 16-stranded porins were shown additionally to have a specificity towards small usually negatively charged molecules, such as malate or phosphate [13–15]. The crystal structures of the 16-stranded trimeric Omp32 from Comamonas acidovorans and the monomeric OprP from Pseudomonas aeruginosa in complex with substrate molecules provided the first rational basis for these functionalities [13,15]. A functionally related trimeric porin, PorB from Neisseria gonorrhoeae, was shown to bind ATP by means of biophysical techniques  and the published structure of meningococcal PorB in complex with p[NH]ppA (adenosine 5′-[β,γ-imido]triphosphate) provided further confirmation of the earlier biophysical investigations . More recently it was discovered that gonococcal PorBIA (PorB of serotype A) directly mediates invasion of host cells, which is prevented by the presence of small concentrations of inorganic phosphate, hinting to a second specific substrate-recognition site [17,18].
Porins from Gram-negative bacteria have been demonstrated to insert into mitochondrial membranes of mammalian cells and the yeast Saccharomyces cerevisiae [19,20]. Neisserial PorB was reported to follow the same mitochondrial import pathway as the mammalian 19-stranded porin or VDAC (voltage-dependent anion-selective channel) due to a similarity in the amphipatic β-strand pattern and the co-occurrence of a C-terminally located translocation motif . Escherichia coli porins such as OmpF and neisserial PorBIA were shown to use the TOM (translocase of the outer membrane) for import into the mitochondrial intermembrane space [20,21]. Subsequent delivery into the outer or inner membrane strongly depended on the origin of mitochondria. If studied in S. cerevisiae, porins from E. coli are selectively delivered to the outer membrane, whereas in higher eukaryotic cell lines PorB was observed to localize to the intermembrane space/inner membrane [20,22]. Although the physiological relevance of this inner membrane translocation shown under non-native conditions (e.g. overexpression of the protein in the cell) for bacterial pathogenesis is not fully understood, it is considered to be important during the induction of apoptosis [19,22]. In vivo experiments demonstrated that, following attachment of N. gonorrhoeae to epithelial cells, the outer membrane protein PorBIA is transferred into the host cell's cytoplasmic membrane  and further translocated into the mitochondrial inner membranes . These events can subsequently cause apoptosis of host cells through a breakdown of the mitochondrial membrane potential (ΔΨm), outer membrane rupture and release of cytochrome c into the cytoplasm, the typical fingerprint of mitochondrially induced apoptosis. Localization of porin channels to the mitochondrial membranes has a significant impact on their membrane potential-dependent conductance behaviour. Although the outer mitochondrial membrane is considered to lack a significant transmembrane potential (ΔΨm), the inner membrane maintains a ΔΨm in the range of 120 mV, which facilitates temporal closure or a permanent loss of conductivity of most bacterial porins [11,23]. This typically voltage-dependent gating is absent in PorB, which favours an open state in the presence of ATP even under conditions of increased transmembrane potential. The influence of ATP binding in the induction of apoptosis is further underlined by mutations of residues in the predicted ATP-binding site, which rendered the PorB mutant cell lines less susceptible to apoptosis . Notably PorB is required, but not sufficient, for the induction of apoptosis as overexpression of the protein alone does not lead to cellular demise [21,22]. Furthermore, PorB has also been demonstrated to interfere with apoptotic signalling; however, the mechanism of PorB's anti-apoptotic activity has yet to be investigated [22b,22c].
PorBIA from N. gonorrhoeae has been studied for many years and a large body of biophysical and physiological data has accumulated. To understand these results on the basis of a three-dimensional structure we crystallized the protein in the presence of ATP and phosphate. The structure analysis showed both ligands attached to the channel interior. Furthermore, a structural remnant peptide derived from gonococcal OmpA homologue protein III located in the periplasmic space confirmed the adjustment of PorB-like proteins to the PGs via protein–protein interactions. Site-directed mutation of one PorBIA-specific phosphate-binding site caused a strongly reduced invasion of gonococci into host cells suggesting these sites as critical for the PorBIA–receptor interaction.
N. gonorrhoeae strain N242 [PorBIAs Opa (opacity-associated) 27, Opa27.5, Opa28, Opa29 and Opa30]  and the MS11 derivatives N920 (strain MS11; PorBIAs P+ and Opa−)  and N917 (PorBIBs P+ and Opa−)  were described previously. Clinical isolates from patients with disseminated goncoccal infection from blood (strain 24871) and articular fluid (strains 20665, 18832 and 17149) were obtained from Dr Muhamed Taha (Institute Pasteur, Paris, France). Gonococci were grown on Oxoid GC agar base plates supplemented with 1% vitamin mix for 14–18 h at 37°C in 5% CO2 in a humidified atmosphere. Opa- and pili-negative phenotypes were monitored by colony morphology under a stereo microscope or by immunoblotting.
PorBIA purification from wild-type cells
Neisserial porin was purified as described previously  with several modifications. Briefly, overnight cultures of the N242 strain  were harvested and disrupted in 1 M sodium acetate, pH 4.4, containing 50 mM 2,3-disulfanylpropan-1-ol, by means of an Ultra-Turrax® disperser (Ika). 0.5 M CaCl2 (5.5 vol.) containing 5% (w/v) Zwittergent were added and dispersed at room temperature (23°C). Ice-cold ethanol was added to a concentration of 20% and the precipitated material was removed by centrifugation (10000 g for 60 min at 4°C). Proteins from supernatants were dialysed in 20 mM Tris/HCl (pH 8.0), 20 mM MgCl2 and 0.5% Zwittergent and loaded on to a Source Q ion-exchange column. Proteins were eluted with NaCl gradients. Eluates were further purified by size-exclusion using a Superdex HR200 (26/60) column in 20 mM Tris/HCl (pH 8), 150 mM NaCl, 2 mM MgCl2 and 0.05% Zwittergent 3-14.
Chang cells were grown to 80–90% confluence. Infections were performed in phosphate-free Hepes medium as described previously . Monolayers were washed once with Hepes medium to remove phosphate before infection. Bacteria were suspended in the Hepes medium and added to the cells at a MOI (multiplicity of infection) of 50. To synchronize the infections the cells were centrifuged at 600 g for 3 min at 37°C. Infections were stopped after a 1 h incubation at 37°C in 5% CO2 atmosphere by washing the cells three times with Hepes medium.
Gentamicin protection assay
Gentamicin protection assays were conducted as described previously . Briefly, cells were infected at 80–90% confluence with the indicated neisserial strain at a MOI of 50 for 1 h. To quantify the number of cell-associated bacteria the cells were lysed with 1% saponin for 7 min. Suitable dilutions were plated on Oxoid GC agar plates and the number of CFUs (colony forming units) were determined 24 h later. For quantification of viable intracellular bacteria the cells were incubated with 50 μg/ml gentamicin in Hepes medium for 2 h at 37°C and 5% CO2 prior to saponin lysis and re-plating.
The strategy for site-directed mutagenesis of PorB was adapted from Bauer et al. . Two overlapping PCR fragments were amplified from the chromosomal DNA of the donor N920 strain with the oligonucleotides: ISO3, 5′-GCTTGCCGTCTGAATTACGCCCCGCCCTGCCACTCATGC-3′; PorBMutR92SF, 5′CGCGTCGGCAGTTTGAAC-3′; FJB4, 5′-GCTTGCCGTCTGAATTACGCCCCGCCCTGCCACTCATGC-3′; and PorBMutR92S, 5′-GTTCAAACTGCCGACGCG-3′. The resulting fragments were fused in a subsequent PCR with the oligonucleotides FJB4 and ISO3. Using the same procedure a DNA fragment was constructed using the oligonucleotides FJB4/ISO3 and chromosomal DNA of the N920 strain as a control. The recipient N917 strain was transformed with these fragments according to a protocol previously described  yielding the strain N2012 (PorBIAR92S, P+ and Opa−). Briefly, gonococci were suspended in 1 ml of transformation medium at an D550 of 0.32. After the addition of 20 ng of DNA the bacteria were grown at 37°C overnight on a GC plate. At 24 h later the bacteria were resuspended in PPM medium and plated on GC plates containing antibiotics (7 μg/ml erythromycin).
Crystallization and structure analysis of PorBIA
The buffer from the size-exclusion chromatography [20 mM Tris/HCl (pH 8.0), 150 mM NaCl, 2 mM MgCl2 and 0.05% Zwittergent 3-14] was exchanged into several detergent buffers, but crystals appeared only in the presence of LDAO [lauryldimethylamine N-oxide; 20 mM Tris/HCl (pH 8), 2 mM MgCl2, 1 mM ATP and 0.2% LDAO]. Crystallization was performed using the sitting-drop technique by mixing 400 nl of the protein solution with equal amounts of reservoir solution using the ‘honeybee robot’ (Genomic Solutions). The drops were incubated at 20°C and the images were recorded using a RockImager system (Formulatrix). The crystals appeared in a precipitant condition containing 30% (v/v) (±)-2-methyl-2,4-pentanediol and 100 mM phosphate buffer (pH 7.5) and these drops were reproduced as 1 μl+1 μl hanging drops. For crystal freezing in liquid nitrogen 10% PEG 400 was added as a cryoprotectant. Diffraction data were collected at the Swiss Light Source PX10 (Villigen, Switzerland). Diffraction images were recorded on a MarCCD Camera 225 (Mar Research) and images were processed using the XDS/XSCALE software . The structure was solved by MR (molecular replacement) using the Molrep with PDB code 3A2S as a search model . Rebuilding of the structure and structure refinement was performed using the programs Coot, Refmac and Phenix [31–33]. The quality of the structure was analysed by PROCHECK . All figures were generated using PyMOL (http://www.pymol.org).
Protein isolation, crystallization and structure determination
The major OMP PorB from N. gonorrhoeae strain N242 (henceforth PorBIA) was isolated and purified under non-denaturing conditions. Crystallization was performed in the presence of ATP and traces of phosphate (see the Experimental section for details). Crystals using LDAO as detergent appeared in space group C2, diffracting to a resolution of 3.2 Å (1 Å=0.1 nm) with one trimer in the asymmetric unit. The crystal packing of PorBIA was induced by intercalation of three extraplasmic faces of the porin via hydrophilic contacts and hydrophobic matching of the barrel walls to symmetry related porin trimers (neither type I nor type II membrane protein crystals). The solution of the structure was initially attempted by MR using the co-ordinates of Omp32 from C. acidovorans with a sequence identity of 22% (PDB code 2FGR) as a search model . Although the model used for MR was of high quality (resolution of 1.5 Å), the failure to trace the PorBIA model may be explained by the relatively high r.m.s.d (root mean square deviation) of 1.7 Å for the 271 aligned residues, the low identity and the limited resolution. Fortunately during the process of extensive rebuilding the co-ordinates of PorBNM (Neisseria meningitidis PorB) at a medium resolution became accessible (PDB code 3A2S; identity of 62% and r.m.s.d of 1.2 Å for 296 residues) and improved tracing of the model significantly . ATP and two phosphate ligands were introduced owing to differences in the electron densities on the basis of phases of the fully refined protein model. Moreover a short phospho-alanine peptide stretch representing part of the PIII [PGBD (neisserial PG-binding protein)] protein was modelled at the last stage of model completion and the entire structure comprising all cofactors was refined to R/Rfree factors of 0.22/0.26 (for details see Table 1).
|a, b, c (Å)||114.05, 111.470, 88.340|
|α, β, γ (O)||90, 102.07, 90|
|Resolution (Å)||38.00–3.31 (3.51–3.31)|
|Rsym or Rmerge||0.19 (0.67)|
|Resolution (Å)||38.00–3.31 (3.4–3.31)|
|Number of reflections||15040|
|Number of atoms|
|B-factors (all atoms)|
|Bond lengths (Å)||0.022|
|Bond angles (°)||1.94|
|Residues in favoured region (%)||855 (91.4)|
|Residues in allowed region (%)||78 (8.3)|
|Residues in outlier region (%)||2 (0.2)|
|a, b, c (Å)||114.05, 111.470, 88.340|
|α, β, γ (O)||90, 102.07, 90|
|Resolution (Å)||38.00–3.31 (3.51–3.31)|
|Rsym or Rmerge||0.19 (0.67)|
|Resolution (Å)||38.00–3.31 (3.4–3.31)|
|Number of reflections||15040|
|Number of atoms|
|B-factors (all atoms)|
|Bond lengths (Å)||0.022|
|Bond angles (°)||1.94|
|Residues in favoured region (%)||855 (91.4)|
|Residues in allowed region (%)||78 (8.3)|
|Residues in outlier region (%)||2 (0.2)|
Architecture of the PorB–PGBP complex
The protein trimer and monomer shows essentially the same architecture as described for the related trimeric porins including OmpF- and Capsulatus-like porins (Supplementary Figure S1 at http://www.BiochemJ.org/bj/449/bj4490631add.htm), all of which comprise 16 antiparallel β-strands with elongated loops facing the extraplasmic space and turns oriented towards the periplasm (Figure 1). Besides the longest loop, L3, folding inside the barrel, an additional archetypical property of trimeric porins is their large intermolecular interface (PorBIA covers 1150 Å2), which contributes to the unusually high thermal and chemical stability [9,35].
Crystal structure of PorBIA from N. gonorrhoeae
We purified PorBIA from N. gonorrhoeae membranes under native conditions, which resulted in co-crystallization of a peptide localized at the periplasmic side of the porin. This peptide is connected to the two adjacent symmetry related peptides and the PorBIA N-terminus (Figure 1 and Supplementary Figure S2 at http://www.BiochemJ.org/bj/449/bj4490631add.htm). These residues are part of the PGBP known as PIII. This protein has an OmpA-homology domain that forms 1:3 complexes with PorBIA (see Supplementary Figure S2 for the sequence alignment of PIII) . Since the resolution of the PorB–PIII complex structure was insufficient to trace the side chains of the peptide, these residues were modelled as a polyalanine sequence.
We compared PIII with an already known structure of a similar protein RmpM from N. meningitidis (PDB code 1R1M). This protein forms non-covalent contacts to the PG and an unknown OMP of the membrane (presumably PorBNM). Another two proteins that are similar in sequence are the periplasmic domain of OmpA from E. coli and the OmpA–MotB protein from Delftia acidovorans (Supplementary Figure S2). We used the PorBIA complex structure to model the putative hetero-oligomeric complex PorBIA–PIII shown in Supplementary Figure S3 (at http://www.BiochemJ.org/bj/449/bj4490631add.htm) .
Charge distribution in PorBIA
We compared the charge distribution of PorBIA with those of other sequence-related porins. As for all bacterial OMPs, these porins show the archetypical formation of aromatic girdles at the height of phospholipids; a rather small subset of porins shows charge girdles, additionally exposed to the counter charged lipid and LPS (lipopolysaccharide) head groups respectively (Figures 2A–2C) . This feature has first been reported for the bacterial porin Omp32 and is also clearly visible in the structure of PorBNM. Although the PorBIA complex shows several positively charged residues pointing outwards to the outer membrane, the overall expression of this girdle is much less pronounced (Figure 2A).
Comparison of molecular electrostatics including charge girdles and the pore interior using PorBNG, Omp32 and PorBNM structures
Analysing the net charges we found that for PorBNM the overall positive charge is +10 (+43/−33), whereas PorBIA shows a positive charge of +5 similar to Omp32 (+36/−31). Lysine residues almost exclusively contribute to the formation of the charge girdle (19 in PorBNM, 8 in PorBIA and 10 in Omp32). By contrast arginine residues were located more predominantly in the pore interior (see Figures 2D–2F). Given the involvement of a significant number of these positive charges in the lysine residue girdle, the charge distribution in the pore is less pronounced than might be expected from the calculated amino acid distribution.
The monomeric PorBIA architecture
All of the 16-stranded trimeric porin structures discovered so far follow the same structure principles as the basis for the architecture of the barrel wall enclosing the long L3 loop . This L3 loop folds into the barrel wall and strongly restricts the lateral diameter of the protein channel as can be also seen in the structure of PorBIA of N. gonorrhoeae (Figure 3). However, additional structural fingerprints (protrusions in loops or strands from bulky residues) can significantly contribute to the overall dimensions of 16-stranded bacterial porins. When comparing the superimposed porin structures Omp32, PorBIA and PorBNM, we observe that the L3 loop is more pronounced in Omp32 compared with the PorB species. Omp32 L3 also expresses a long additional extension oriented towards the extraplasmic space (Figure 3B). Generally the loop structures (L1–L8) are highly conserved between PorBIA and PorBNM (Figure 3). Major structural differences are only visible in the orientation of certain loops and the differing length of L4 and L5, whereas the fine structure of the barrel architecture is essentially preserved. The overall comparison between PorBIA and Omp32 shows significant structural differences due to the lower homology, insertions and deletions, and the distribution of charges (Figure 3A and Supplementary Figure S1).
Pairwise comparison of monomeric PorB/Omp32-like porin structures
We recognized another important structure element of PorBIA in the β2-strand as a protrusion of four residues with side chains pointing into the barrel interior. This element, also described in the literature as a β-bulge, is stabilized via backbone interactions of the side chains of Asp36 and Ser39 (Figure 3).
Ligand binding in PorBIA
It has long been described that bacterial porins of the PorB type specifically bind and translocate ATP molecules when studied in artificial planar membranes . In the crystal structure of PorBIA the ATP ligand was modelled into the difference electron density map as described (Figure 4C). ATP can be divided into three different chemical entities: the adenine ring, the ribose sugar and three condensed phosphate groups (Figure 4A). Our model shows that the ATP is ligated by residues of the channel forming a wall together with the residues of loop L3, most of which are primarily positively charged. The adenosine moiety is ligated by Lys62, Arg75, Lys98 and Arg124 via an extended hydrogen bond/salt bridge network. In agreement with a role for Lys98 in ATP binding, a Gln98 mutant of PorBIA showed a strongly reduced capacity to bind ATP  (Figure 4A). In the structure of PorB from N. meningitidis this binding pattern is less pronounced and only two residues (Arg130 and Lys100) are in proximity to the adenosine moiety .
ATP and phosphate binding site in PorBNG
In PorBIA the ribose moiety of ATP is bound only via the backbone interaction through Gly38 and hydrogen bonds between the two hydroxo groups and Lys9 (Figures 4A and 4C). When we compare this to N. meningitidis PorB we find that the molecule of a non-hydrolysable ATP analogue p[NH]ppA, used for determining PorBNM structure, is slightly shifted towards the extraplasmic barrel entrance and the interaction between Lys9 (4 Å) is not as strongly expressed as in the PorBIA structure. Instead, the ribose is co-ordinated by the Arg77 side chain, which is involved in binding of the adenosine moiety in PorBIA. Finally, the triphosphate group contact is mediated by Lys40 of PorBIA, which co-ordinates both the β- and γ-phosphate, whereas the γ-phosphate is co-ligated additionally by Glu58 and Glu76 (Figure 4A).
The analysis of the conserved residues involved in ATP of PorBIA binding reveal that most residues can be identified in the related species N. meningitidis, N. sicca, N. cinerea and N. lactamica (see the alignment in Figure 5A). Although there is a significant consensus between the binding modes discovered in PorBIA and PorBNM not all of these residues are identical. This may explain why different Neisseria species also differ in capacity to bind ATP nucleotides .
Alignment of six related PorB sequences and analysis of the protein models
Structural differences in PorB-like channels in Neisseria
We modelled the structures of six PorB homologues from previously characterized neisserial strains on the basis of PorBNM (PDB code 3A2S), the structure solved at the highest resolution, with the aim of investigating the functional heterogeneity between the different strains. Protein alignments visualized heterologous parts, in particular in the loop structures. Residues forming these loop structures were generally not well conserved and deletions and insertions were detected when comparing the six sequences with a long extension in L6 of N. sicca (PorBNS) being the most evident (Figure 5). N. sicca, N. lactamica and N. cinerea are commensal strains, unlike N. meningitidis and N. gonorrhoeae, which are the only two human pathogens from the genus Neisseria.
Pathophysiology of PorBIA
Evidence from physiological data showed the influence of even traces of phosphate molecules in the buffer on the capacity of PorBIA-expressing N. gonorrhoeae to invade host cells [17,18]. In the crystal structure we identified two phosphate molecules, one of which is located at the extraplasmic entrance of the PorBIA channel (Figure 4A). This ligand is bound with Arg92 and Arg124. Although Arg124 is conserved between all PorB sequences (for a selection see the alignment in Figure 5A), only PorBIA porins display arginine or histidine residues at position 92, whereas all PorBIB porins possess a conserved serine residue at this position (Figure 6E and Supplementary Figure S6 at http://www.BiochemJ.org/bj/449/bj4490631add.htm).
Critical role for Arg92 in PorBIA for the phosphate-sensitive invasion of gonococci
Since only PorBIA derivatives mediate the invasion of gonococci in a phosphate-sensitive manner [17,18], the exclusive presence of Arg92/His92 in PorBIA (Supplementary Figure S6) provided strong evidence for its role in invasion. To further support a role of Arg92/His92 in PorBIA in invasion, we tested four clinical isolates from disseminated strains isolated from blood or synovial fluid of patients for phosphate-sensitive invasion. The sequenced PorBs from these isolates belong to different subtypes, thus excluding a clonal origin of these strains. All isolates, irrespective of arginine (strains 24871, 20665, 18832) or histidine (strain 17149) residues at position 92, showed phosphate-sensitive invasion (Figure 6A). We then exchanged Arg92 of PorBIA from the N920 strain for a serine residue, the amino acid invariantly found in all PorBIB subtypes (Supplementary Figure S6), creating the mutant strain N2012. The mutation did not affect bacterial growth (Figure 6B), but it reduced adherence to Chang conjunctiva cells by approximately 50% and almost completely prevented invasion (Figures 6C and 6D). These results demonstrate that Arg92/His92 is critically involved in the invasion of PorBIA-expressing strains.
A growing number of porin structures have been determined over the years with two closely related PorBNM and Omp32 structures being most similar to PorBIA [13,14] (see Supplementary Figure S1 for a representation of the sequence relationship between all 16-stranded porin structures). Although this plot is based on pairwise sequence comparisons of porins, it also reflects the structure deviation between the species [r.m.s.d. (root mean square deviation) for PorBNM is 1.2 Å and for Omp32 is 1.7 Å]. Structural deviations in the porin group relative to the PorB subfamily become more evident if the unspecific porins of the OmpF group from E. coli are compared with PorBIA (r.m.s.d. of 2.7 Å for 279 and 274 aligned Cα atoms respectively, Supplementary Figure S1) [38,39].
In analogy with a handful of OMPs PorBIA was purified from membranes under native conditions (for statistics of OMP structures determined for native and refolded proteins see ). One advantage of this approach is the co-crystallization of natively attached molecules such as LPS or small protein ‘contaminations’. Consequently the PorBIA crystal structure revealed additional electron density through the presence of peptides originating from a PG-binding protein. The physical connection between the outer membrane and the PG layer is important to maintain the overall stability and shape of Gram-negative bacteria. Proteins mediating PG binding to the outer membrane can be encoded by a single gene comprising the OMP domain and the PG-binding domains, such as observed in the OmpA-family members or the P100 protein from Thermus thermophilus . Other proteins carrying PG-binding domains are located in the periplasmic space and simply connect the outer membrane and PG (e.g. Omp32 protein). In the crystal structure of the PorBIA homologue Omp32 a peptide at a similar location was already observed in electron difference densities attached to Omp32, which was isolated as complex from wild-type membranes. This peptide comprised 54 residues as part of the larger MotB protein with homology with an OmpA-like domain (Figures 1 and 2A, and Supplementary Figure S2). On the basis of SDS gels analysis the stoichiometry of this complex is 1:1 ; however, there is a repetitive element of nine residues (including a highly conserved tryptophan residue) at the N-terminus of MotB from D. acidovorans which would in principle also allow the formation of a 3:1 (porin–MotB) complex (see Supplementary Figure S2A). This protein fragment was sequenced and further analysed by bioinformatics techniques showing similar PG motifs as for proteins of the OmpA family (see Supplementary Figure S2B). Although the resolution of this complex was 1.5 Å, only eight N-terminal residues directly connected with the periplasmic OMP barrel side were ordered and allowed tracing in the electron density due to the high resolution of the Omp32 electron density map . These residues are connected to the C-terminal PGBD domain via a long proline-rich linker of approximately 40 residues predicted to be disordered (Supplementary Figure S3). Since the N-terminal sequence of PIII does not show a clear similarity to MotB from D. acidovorans (see Supplementary Figure S2B), we can only speculate that residues of the very N-terminus of PIII are involved in the physical linkage of the PG and the PorBIA complex. Therefore the model described in the present study may indicate a reasonable structure arrangement and extension of this putative complex architecture physically linking the outer membrane to the peptidoglycan layer (Supplementary Figure S3A).
Charge girdle of PorBIA
The characteristic of a residue girdle of positively charged amino acids in PorBIA (in addition to the aromatic girdles) is much less pronounced in comparison with related proteins (Supplementary Figure 2A). So far the functional importance of this girdle has not been determined, but, inferred from analogy to the aromatic girdle architecture, it may contribute to the stabilization of the porin trimer in the outer membrane by counter balancing negatively charged LPS molecules. Since this charge girdle is visible in a subset of porins related to Neisseria species, it is possible that it resembles an additional adaptive mechanism for the individual bacterial membrane composition or adaption to environmental conditions. This girdle may also comprise a specific function for bacterial–host contacts, e.g. in mitochondrial inner membranes enriched in negatively charged cardiolipin [43,44].
Charge distribution in PorBIA channel
The results of the present study indicate that the charge distribution in the pore of PorBIA is less pronounced than was expected from the calculated amino acid distribution. Porin channels generally express a variety of three-dimensional charge profiles in the channel interior, which, in combination with the channel diameter, form the structural filter for the selectivity of these channels. Originally all 16-stranded porins were described to be unspecific. Later, however, due to the discovery of ‘unusual’ porins the pore diameter of the protein together with the charge distribution appeared to be most relevant for binding specificity of ligands independently of a 16- or 18-stranded architecture. PorB-like channels, including also the distant relative Omp32, carry a surplus of positive charges and show an anion selective behaviour which clearly discriminates anions over cations by a factor of 20 for Omp32 [23,45]. PorBIA carries a surplus of positive charges inside the channel with a rather heterogeneous distribution and a selectivity filter ratio of 1.8 cation/anion which increases up to 8 in the presence of nucleotides . PorBNM shows a different and unusually strong asymmetric distribution of charges with two opposite patch-like arrangements of positive and negative charges guiding the channel transversion and yielding a weak anion selectivity (Figures 2D–2F) [14,46]. Substrates of these porins are negatively charged (ATP and malate) and therefore aligned through the proximity of positively charged patches located on the inner wall of the barrel (see also above and Supplementary Figure S4 at http://www.BiochemJ.org/bj/449/bj4490631add.htm for a comparative view on porin channel diameters). Counter charged surface patches in porin channels are another typical observation of earlier structures such as OmpF or the Capsulatus porin. These porins, however, clearly showed a strongly increased lateral diameter at their constriction site. It remains questionable if the effect of charges on to a putative substrate can be as strong as for the PorB-like structures of small diameter [4,38,47]. Amphipathic or uncharged molecules may be aligned by oppositely charged electric fields or channels with further reduced diameters and a hydrophobic interior such as FadL (long-chain fatty-acid transporter), whereas charged molecules are attracted by counter charges on the porin surface .
The β-bulge of PorBIA
PorBIA exhibits a long L3 loop and a β-bulge in the β2-strand that resembles the motif present in PorBNM. A similar motif in the β2-strand has also been discovered earlier for the related Omp32 porin, but is, in this case, more pronounced through the presence of six residues folding into the barrel lumen (Figure 3A) . In Omp32 the β-bulge is stabilized by residues located at similar positions, but the aspartate residue of PorBIA is replaced by a serine leading to the serine–serine stabilization via hydrogen bonds. Interestingly in both PorBNM and PorBIA structures this β2-element not only contributes to a narrowing of the pore, but also makes a direct contact between the backbone atoms of Gly38 and the ATP ligand (see below). Owing to the extended β-bulge in Omp32 the ATP ligand is presumably not a favourable substrate for this protein channel. More interestingly, in the Omp32–malate complex the elongated β-bulge again makes direct contact with another negatively charged substrate molecule, the malate ligand via an hydrogen bond between Thr41 and the carboxy group of malate. Given the rather small total pore size of Omp32 it is more likely to bind smaller organic substrate molecules such as malate, succinate or fumarate. In summary, the β2-strand invagination provides a rationale for binding of ligands and appears to be a more important structural characteristic than previously thought.
The pore size of the PorB family of porins
The β-bulge appears to be a specific motif of the PorB/Omp32 porin family and is so far not visible in any other monomeric or trimeric porin structure discovered until present (see Supplementary Figure S1). This β-bulge, located at the same vertical height position as L3, leads to a second although smaller restricting element of the pore interior and a further reduced pore diameter for porins such as Omp32, PorBNM or PorBIA (Figure 3A and Supplementary Figure S4). In a comparison of five 16-stranded porins it becomes more obvious that the pore size may be adjusted during evolution to fulfil the specific substrate translocation requirements of the outer membrane. The smallest pore size is observed in a crystal structure determined for OprP from P. aeruginosa, a channel which is reported to passively transport phosphate over the outer membrane (see Supplementary Figure S4) . Another porin with a significantly decreased pore size is the Omp32 channel from D. acidovorans which is known to transfer small organic acids such as malate, fumarate or succinate . By contrast, the smallest diameter of PorB structures is significantly increased in comparison with Omp32, presumably due to the transport of the ATP molecules (see Supplementary Figure S4). Finally, OmpF as an unspecific porin responsible for the transport of small molecules including antibiotics possesses the largest pore diameter .
A narrowed pore size in 16-stranded bacterial porins is also reminiscent of the 18-stranded porins LamB (maltoporin) and ScrY (sucrose porin), which form a part of bacterial uptake systems and can specifically bind and transfer sugar molecules into the bacterial periplasm [52,53]. In fact in a superposition of 16-stranded PorBIA and 18-stranded maltoporin with bound maltotetraose the two ligands are almost perfectly aligned to each other in space (see Supplementary Figure S5A) . Likewise the 18-stranded OprD from P. aeruginosa was discovered to guide phosphate molecules along a ladder-like arginine arrangement similar to the 16-stranded OprP . The narrowed pore size in porins together with a specific charge distribution might be general prerequisites for the enhanced and guided translocation of specific molecules. If these molecules carry charged ligands such as malate, phosphate or ATP the influence of the electric field across the channel is significantly enhanced if the channel diameter is smaller and substrate selectivity and flux is more strongly controlled (see Figures 2D–2F).
Binding of ATP
The two binding scaffolds of ATP in PorBIA and p[NH]ppA in N. meningitidis PorB differ in the nature of interactions between protein and ligand. This may be in part due to the chemical difference of the two ligands, although this should not affect the adenosine and ribose moiety. The chemical structure may, however, interfere with the conformational flexibility of the phosphate group. On the other hand, the additional phosphate molecule near the adenine moiety may influence the geometry of ATP and slightly induce a variable position of the molecule inside the channel.
The Omp32 porin is structurally less similar than porins of the neisserial clade; including a β-bulge in strand β2. By contrast, most residues are not conserved between PorBIA and Omp32 except Arg75 (Arg75 in Omp32), Arg92 and Arg124 (Arg133 in Omp32). Arg75 and Arg133 in Omp32 are interesting with respect to their influence on binding of the malate ligand, which is located in the very centre of the pore. These three residues form the central motif for ligand interactions. Interestingly, a second ligand molecule in Omp32 collected under high sulfate concentrations, a sulfate ion, is placed exactly at the α-phosphate position of the ATP molecule and co-ordinated by an Arg38 in Omp32 which is located nearby Lys40 in PorB (Figures 4B and 5B). Although these coincidences may give rise to the speculation that Omp32 could also bind ATP like PorB, this seems rather unlikely owing to the extended β-bulge of the structure, which in an overlay of PorBIA and Omp32 clashes with the triphosphate molecule and gives this pore an especially narrow structure (Figure 4B).
Phosphate binding and connection to pathogenicity
The crystal structure of PorBIA shows two phosphate molecules. One of them is located at the extraplasmic side of the PorBIA channel and is held by Arg92 and Arg124. Arg92 or His92 is found only in PorBIA and not in other homologous PorB proteins (Figure 5A). We currently do not know the role of Arg92/His92 in the invasion process. It is possible that this region (which is accessible from the outside, see Supplementary Figure S7 at http://www.BiochemJ.org/bj/449/bj4490631add.htm) is directly involved in the recognition of host cell receptors like the SREC (scavenger receptor of endothelial cells)-I, which we previously identified as an invasion receptor for disseminating gonococci . Another possibility is a role for Arg92 in the property of the pore channel. That the altered nutrient supply leads to a reduced survival of the PorBIA-R92S mutants is, however, unlikely. PorB porins are the only pores in the outer membrane of gonococci and a functional loss would ultimately result in a toxic effect for the bacteria and a reduced growth rate. This could be excluded since the PorBIA-R92S mutant strain N2012 grew with the same kinetics as the wild-type strain (Figure 6A).
The results of the present study now provide a structural basis for the long-standing observation that phosphate in millimolar concentrations prevents the invasion of gonococcal strains expressing PorB of the A serotypes [17,18]. Since these bacteria are frequently isolated from patients with disseminated gonococcal disease [57–59] the identification of the phosphate-binding structure in these porins may also have therapeutic implications. If this region of PorBIA is involved in the interaction with SREC, agents binding to this part of the porin may work to prevent rapid invasion and dissemination of gonococci. We currently do not know whether the nucleotide-binding domain of PorB is also part of, or overlaps with, a receptor-binding domain of PorBIA. However, nucleotides have also been shown to have a strong inhibitory effect on invasion by PorBIA-expressing gonococci . Detailed analyses based on the current structure of PorBIA will help to further elucidate the function of this important pathogenicity factor.
multiplicity of infection
outer membrane protein
neisserial PG-binding protein
PorB of serotype A
Neisseria meningitidis PorB
root mean square deviation
scavenger receptor of endothelial cells
Kornelius Zeth performed the structural analysis; Vera Kozjak, Robert Hurwitz and Oliver Kepp purified PorB; Vera Kozjak-Pavlovic and Michaela Fraunholz carried out the site-directed mutagenesis and invasion assays; Martin Fraunholz aligned the PorB sequences; and Kornelius Zeth and Thomas Rudel designed the study and wrote the paper.
We thank Dr Reinhard Albrecht for crystallization of the protein and Monika Götz (Biocenter University of Würzburg, Würzburg, Germany) for technical support. We thank Dr Muhamed Taha (Institute Pasteur Paris, Paris, France) for gonococcal isolates from patients with disseminated infection and all the people from the Max Planck Beamline PXII (Swiss Light Source, Villigen, Switzerland) for their support and the maintenance of the beamline.
This work was supported by the Max Planck Society, the German Science Foundation [grant numbers ZE522/5-1 (to K.Z.), SPP1131, ERA-Net 0315435B and RU631/7-1 (to T.R.), KO3882/1-1 (to V.K.-P.) and SFB630 (to V.K.-P. and T.R.)].
Present address: Unidad de Biofisica (CSIC-UPV/EHU), Barrio Sarriena s/n, 48940, Leioa, Vizcaya, Spain.
Present address: INSERM U848, 39 rue Camille Desmoulins, 94805 Villejuif, France.
The atomic co-ordinates and structure factors have been deposited in the PDB under accession code 4AUI.