A Tetrameric Porin Limits the Cell Wall Permeability ofMycobacterium smegmatis *

Mycobacteria protect themselves with an outer lipid bilayer, which is the thickest biological membrane hitherto known and has an exceptionally low permeability rendering mycobacteria intrinsically resistant against many antibiotics. Pore proteins mediate the diffusion of hydrophilic nutrients across this membrane. Electron microscopy revealed that the outer membrane of Mycobacterium smegmatis contained about 1000 protein pores per μm2, which are about 50-fold fewer pores per μm2 than in Gram-negative bacteria. The projection structure of the major porin MspA of M. smegmatis was determined at 17 Å resolution. MspA forms a cone-like tetrameric complex of 10 nm in length with a single central pore. Thus, MspA is drastically different from the trimeric porins of Gram-negative bacteria and represents a new class of channel proteins. The formation of MspA micelles indicated that the ends of MspA have different hydrophobicities. Oriented insertion of MspA into membranes was demonstrated in lipid bilayer experiments, which revealed a strongly asymmetrical voltage gating of MspA channels at –30 mV. The length of MspA is sufficient to span the outer membrane and contributes in combination with the tapering end of the pore and the low number of pores to the low permeability of the cell wall of M. smegmatis for hydrophilic compounds.

About two million people die each year from tuberculosis (TB). 1 This number qualifies TB as the leading agent of death due to a single infectious disease (1). Treatment of TB caused by non-resistant strains of Mycobacterium tuberculosis is effective but is based on a regimen of up to four drugs over a period of six months (2). The major problem in TB therapy is the slow uptake of drugs across the mycobacterial cell wall, which mounts a formidable permeability barrier toward diffusion of hydrophobic and hydrophilic compounds (3). The diffusion of hydrophilic compounds across the mycobacterial cell wall is mediated by water-filled channel proteins (4). The transport function of these channel proteins is similar to that of the porins of Gram-negative bacteria, but the porin pathway of mycobacteria is at least 1000-fold less efficient than that of Escherichia coli (5,6). It is assumed that the combination of low cell wall permeability with the action of detoxifying proteins such as degrading enzymes or efflux pumps is responsible for the intrinsic resistance of mycobacteria to many antibiotics such as penicillins, cephalosporins, and tetracyclines (7). However, it is not known why the porin pathway in mycobacteria is so inefficient.
Two integral proteins with transport properties have been identified in mycobacterial cell walls: OmpATb from M. tuberculosis, which has channel activity in vitro but unexplored physiological functions (8), and MspA, which was first discovered in chloroform-methanol extracts from Mycobacterium smegmatis as an oligomeric channel protein composed of 20 kDa subunits (9). Enzyme-linked immunosorbent assays and immunogold labeling experiments demonstrated that MspA is localized in the cell wall (6). Uptake of glucose by an mspA deletion mutant was 4-fold slower compared with the wild type indicating that MspA is the major porin of M. smegmatis (6). MspA is the prototype of a family of four very similar porins of M. smegmatis, which did not show any homology to any other known protein (6). Up to now, MspA is the only mycobacterial porin, which can be purified in milligram quantities (10) and is, therefore, amenable to structural investigations.
The structure of mycobacterial porins would be of paramount importance both for understanding the nutrient transport across the mycobacterial cell wall and for treatment of TB with hydrophilic drugs but is also likely to be of fundamental scientific interest because porins reside in an asymmetric outer membrane with unique properties. (i) The membrane has a very low fluidity and does not melt up to 70°C in contrast to cytoplasmic membranes of other mesophilic organisms, which begin to melt at 20°C (11), (ii) it is about 10 nm thick, exceeding the thickness of all other known membranes by about 2.5-fold (3), (iii) it provides a very hydrophobic cell surface, which causes the bacteria to clump in an hydrophilic environment, and (iv) its fluidity decreases toward the periplasmatic side of the membrane (12) in contrast to that of the outer membrane of Gram-negative bacteria.
Here, we present the first structural analysis of a mycobacterial porin. The MspA complex is composed of four subunits and forms a single pore of extraordinary length. A hydrophobicity gradient along the surface of the MspA pore leads to an oriented membrane insertion in planar lipid bilayers. The structural properties of MspA and the low number of porins contribute to the low permeability of the cell wall of M. smegmatis compared with E. coli.

MATERIALS AND METHODS
Purification of MspA-MspA was purified from M. smegmatis by selective extraction at 100°C using 0.5% n-octyl-polyoxyethylene (octyl-POE) as a detergent and a two-step chromatographic purification procedure as described (10). MspA eluted from the anion exchange column * This work was funded by the Deutsche Forschungsgemeinschaft (NI 412) and the Volkswagen-Stiftung (I/77 729). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
¶ To whom correspondence should be addressed. Tel.: 49-9131-8528989; Fax: 49-9131-8528082; E-mail: mnieder@biologie.unierlangen.de. 1 The abbreviations used are: TB, tuberculosis; POE, polyoxyethyl-at 0.57 M NaCl and was separated from another porin in the extracts, MspC, which eluted at 1.3 M NaCl as described (13). Purified MspA was stored in a 25-mM sodium phosphate buffer (pH 7.5) containing 0.5% octyl-POE (NaP-POE buffer). Subpicogram amounts of purified MspA showed a high channel-forming activity in planar lipid bilayer experiments, which were done as described (10). Cross-linking Experiments-Glutardialdehyde, the most commonly used cross-linking reagent, could not be used because MspA contains only a single lysine residue. Therefore, water-soluble carbodiimides were employed to cross-link carboxyl groups of MspA (14). To obtain a maximal yield of cross-linked protein the molar ratio of 1-ethyl-3-(3dimethylaminopropyl)carbodiimide (EDAC) and cystamine to MspA was optimized. In the final reaction, 1 g of MspA in 100 l (0.13 M) was incubated with 76 mM EDAC and 0.95 M cystamine for 90 min at room temperature. The reaction was buffered with 25 mM sodium phosphate at pH 7.0. The sample was dialyzed for 12 h against 400 ml of 25 mM sodium phosphate (pH 7.0). Four-fifths of the sample containing 800 ng of MspA was denatured by heating in 80% dimethyl sulfoxide to 100°C for 30 min and precipitated with acetone (9) before gel electrophoresis, whereas one-fifth of the sample containing 200 ng of MspA was directly loaded on the gel. Gel electrophoresis and staining with silver was done as described (10).
Electron Microscopy and Single Particle Analysis-Isolated and solubilized MspA was either dialyzed to remove the detergent or reconstituted into lipid vesicles prior to inspection in the electron microscope. MspA was dialyzed against water for at least 24 h, diluted to a protein concentration of ϳ0.1 mg/ml and applied to carbon-coated copper grids made hydrophilic by glow discharge. MspA samples and cell wall fragments were negatively stained with a solution of 0.2% unbuffered uranyl acetate for about 20 s and inspected in a Philips EM420 or CM12 transmission electron microscope at a primary magnification of 35,000fold. For documentation electron micrographs were taken and scanned using the CCD Scanner Flextight (Imacon, Kopenhagen Denmark). Reconstitution of solubilized MspA in lipid membranes was performed by dialysis as described (15) using dimyristoyl phosphatidylcholine (DMPC) as a lipid and 25 mM HEPES (pH 7.5) plus 3 mM NaN 3 as a buffer. The molar lipid to protein ratio was 10:1, and the dialysis temperature was 35°C.
For single-particle analysis, electron micrographs were digitized applying a pixel size of 0.45 nm at the specimen level. Correlation averaging and unbending were performed with the SEMPER image processing system (16) and principal component analysis with the EM system (17). Images of individual protein channels were centered, i.e. laterally aligned, by correlation. Rotational correction was performed after principal component analysis using two related eigenvectors representing the rotational misalignment of the centered images according to the procedure described elsewhere (18). These data consisting of 1757 images were again applied to principle component analysis and classified. Images contributing to ill defined classes were omitted. Finally, a set of 1067 images separated in three classes was used to calculated the individual class averages. This method is a bias-free approach because it does not rely on reference structures for alignment.
Lipid Bilayer Experiments-The channel activity of selected samples was measured in lipid bilayer experiments as described (10). In all lipid bilayer experiments, the aqueous phase contained 1 M KCl and 10 mM 2-(N-morpholino)ethanesulphonic acid (MES, pH 6.0). The temperature was kept at 20°C using a thermostat.
The single channel conductances of at least 100 pores reconstituted from a freshly diluted solution of purified MspA were checked with several diphytanoyl phosphatidylcholine membranes before multichannel experiments such as the analysis of the voltage gating of MspA were done. A lipid film was painted across the hole of a new Teflon cell. After formation of the lipid bilayer purified MspA was added at a final concentration of 26 pg/ml to the cis-side, which was connected to ground. The number of reconstituted channels was determined from the overall conductance increase of the membrane. When ϳ100 channels were reconstituted into the membrane, increasing positive and negative voltages were applied to the membrane and the membrane current was recorded.

RESULTS
Stoichiometry of MspA-The mspA gene encodes a 20-kDa protein, but the isolated MspA oligomer had an apparent molecular mass of about 100 kDa in denaturing polyacrylamide gels. Matrix-assisted laser desorption ionization mass spectrometry confirmed that the purified MspA protein was composed of identical 20 kDa monomers, but the number of subunits or the molecular mass of the oligomer remained to be elucidated (9). To determine the stoichiometric composition of the functional MspA oligomer, we varied the purification procedure by using different detergents and organic solvents and applied mass spectroscopy using both matrix-assisted laser desorption ionization and electrospray ionization, but no signals with a mass-to-charge ratio larger than that of the MspA monomer were detected. In another approach, MspA was crosslinked via its accessible carboxyl groups using a water-soluble carbodiimide. MspA was significantly modified by the crosslinker as evidenced by the reduction of the electrophoretic mobility of MspA oligomers (Fig. 1, lane 4). This might be explained by the loss of negative surface charges because crosslinking between MspA oligomers is unlikely at a protein concentration of 130 nM (2.6 g/ml), which is 20-to 1500-fold lower compared with other protocols used to obtain intermolecular cross-links (19). Denaturation of cross-linked MspA released four protein species (Fig. 1, lane 5) with apparent molecular masses of 21, 42, 64, and 83 kDa, i.e. multiples of the MspA monomer (20 kDa). Protein bands exceeding the apparent molecular mass of the native protein of 90 kDa were not observed neither in silver-stained gels nor by immunodetection of identical samples using an MspA antiserum (not shown). Molecular masses of 76 and 93 kDa were determined for oligomeric MspA by analytical gel filtration using both silica-and dextran-based column materials, respectively (not shown). The results of all these experiments were consistent with an MspA tetramer.
Pores in the Cell Wall of M. smegmatis-The structure of the M. smegmatis porins was investigated in their native environment, i.e. in isolated cell walls, by electron microscopy. In negatively stained preparations uranyl acetate-filled pores were clearly visible (Fig. 2A). The pore structures appeared randomly distributed at a density of 1010 Ϯ 310 pores per m 2 (determined from eight cell wall fragments containing 484 pores). Thus, the density of porins is about 15-fold less than that in the outer membrane of Gram-negative bacteria, in which more than 15,000 porin trimers per m 2 in a two-dimensionally crystalline arrangement were observed (20). The cell wall of an mspA mutant contained 3-fold fewer pores than the wild type indicating that most of the pores were MspA channels (not shown). This is in agreement with MspA being the major porin of M. smegmatis as evidenced by the permeability deficiency of an ⌬mspA mutant toward hydrophilic compounds and the reduced channel activity of detergent cell extracts of the mutant in lipid bilayer experiments (6). Because the other porins of M. smegmatis differ from MspA only by a few amino acids (6), these channels are most likely structurally indistinguishable in electron micrographs at the given resolution of 1.7 nm.
Channel Structure of MspA-It is obvious that the porins of M. smegmatis, and here MspA in particular, are radically different from trimeric porins of Gram-negative bacteria: MspA possesses only one channel per molecular complex (Fig. 2). Principal component analysis of 1757 individual pore images extracted from several micrographs of cell wall fragments revealed a 4-fold symmetric particle (Fig. 2, B and C) that forms a central channel. Rotational alignment of the projections was performed with a bias-free approach using two related eigenvectors for orientational correction (18). The result is in perfect agreement with the tetrameric stoichiometry as deduced from the cross-linking and gel filtration experiments, and it excludes other compositions such as trimers or pentamers.
Classification yielded three sets of molecular images: particles with a pronounced 4-fold symmetric projection structure (Fig. 2, B and C) and one class with poorly defined symmetry (Fig. 2D). The projections are consistent with a cylindrical molecule being partially embedded in stain and, by this way, enhancing the structure on one end or the other of the complex. The inner and outer diameter of class I molecules are 2.5 and 10 nm, respectively, and decreased to about 2.2 and 8 nm for class III molecules.
To obtain further structural information MspA was purified, reconstituted in the presence of DMPC (15), and analyzed by electron microscopy. Depending on the lipid-to-protein ratio a great variety of MspA supramolecular structures were observed: planar lipid membranes and lipid vesicles with randomly inserted MspA complexes, densely packed protein sheets, and spherical and, as depicted in Fig. 3A, linear MspA aggregates. Linear aggregates were obtained in top view and in side view projections (Fig. 3A). The latter show that the MspA pore complexes were arranged in two rows, intercalating like the teeth of a zipper. For image analysis, 20 individual MspA zippers were unbent in the computer, aligned by correlation methods, and averaged (Fig. 3, B and C). The side views confirmed the cone-like structure of MspA as deduced from the top view averages and matched their dimensions, with an outer diameter of about 10 nm, one central open hole of 2.5 nm at one end converging to about 2.2 nm toward the other end (Fig. 4). The channel protein is ϳ10 nm long. About 5.5 nm of the molecule were exposed to the polar environment and is well embedded in stain, thus, representing the more hydrophilic part of MspA. The remaining very hydrophobic portion of MspA is responsible for the assembly of the zipper-like structures.
MspA in Protein Micelles and Lipid Vesicles-To examine whether the apparent asymmetric surface hydrophobicity of the MspA complex is reflected in detergent-depleted solutions and in the lipid environment, octyl-POE was either removed or exchanged for DMPC by dialysis (15). Samples of purified and detergent-solubilized MspA that were extensively dialyzed against buffer or pure water did not yield insoluble precipitates. Instead they showed "protein micelles" in the electron microscope with a diameter of about 35 nm (Fig. 5A). Obviously, the hydrophobic portions of MspA molecules were hidden in the center of the micelle while the apparently more voluminous hydrophilic regions were exposed to the polar environment. The formation of micelles indicated an oriented interaction between individual MspA complexes. The thickness of the "micelle wall" is about 10 nm and agrees with the length of the MspA oligomer (Fig. 4).
Reconstitution of MspA oligomers together with DMPC resulted in membranes and lipid vesicles showing the typical stain-filled pores in electron micrographs (Fig. 5B). Images of flattened vesicles revealed side-on views of the MspA complexes at the vesicle edge and illustrate their apparently unidirectional insertion into the membrane. About 4 -5 nm of MspA protruded from the membrane surface and might represent the hydrophilic portion of the protein in agreement with the arrangement of the complexes in the zipper-like aggregates (Fig. 4).
Voltage-dependent Channel Closure of MspA-The apparently unidirectional insertion of MspA complexes in lipid vesicles (Fig. 5B) should be reflected in functional measurements. This was analyzed in lipid bilayer experiments in which MspA was added only to the cis-side of the membrane. After reconstitution of about 100 MspA channels into a DiphPC membrane, the polarity of the membrane potential was reversed and the potential was increased stepwise. No channel closure was observed at membrane potentials of 10 and 20 mV, regardless of whether the potential was positive or negative (Fig. 6). However, the membrane current decreased exponentially at a membrane potential of Ϫ30 mV, whereas no change of the current was detected up to a potential of ϩ40 mV. At Ϫ30 mV about 60% of the reconstituted channels switched rapidly to a closed conformation. The ratio of closed channels increased to 80%, when the potential was risen to Ϫ50 mV (Fig. 6). Addition of the protein to both sides of the membrane resulted in a symmetric response to the applied voltage (data not shown). These results indicated an asymmetrical orientation of MspA in the membrane and confirmed the assumption that the presence of a particular hydrophobic end of MspA determines the insertion into and the interaction with the lipid membrane.

MspA Is a Channel Protein with Unique Structural
Properties-The major porin of M. smegmatis MspA is composed of four monomers forming one central channel of 10 nm in length and ϳ2.5 nm of inner diameter. These structural features are unique and clearly distinguish the MspA pore and related mycobacterial porins from all other known transmembrane channel proteins. Pore proteins located in the outer membrane of Gram-negative bacteria form a single channel defined by one ␤-barrel per monomer. They occur as monomers such as the porin OmpG of E. coli (21) and the E. coli siderophore receptors FhuA (22) and FepA (23) or as trimers with three pores per molecule, e.g. OmpF of E. coli and related porins (24). Known tetrameric channel proteins belong to the ␣-type (25) and are located in the cytoplasmic membrane, e.g. the potassium channel (26). Another tetramer is formed by the black widow spider neurotoxin ␣-latrotoxin. However, this is a hydrophilic propeller-shaped 520-kDa channel, which only slowly inserts into lipid membranes in the absence of any receptor (27). Moreover, only a part of ␣-latrotoxin acquires amphilicity upon oligomerization and is able to insert into membranes. By contrast, MspA is likely to be embedded over a considerable portion of its length of 10 nm in the mycobacterial outer membrane if the latter has indeed a thickness of 10 nm as proposed earlier (3,28). We found evidence for a strongly hydrophobic portion of about 5 nm in length and assume that at least the outermost domain of ϳ2 nm thickness, which forms the 4-fold symmetry and is readily accessible to uranyl acetate, is exposed on the membrane surface in vivo. If the other part of MspA is lipidembedded in vivo, MspA may possess a longer membranespanning domain than any other bacterial channel protein.
Both the channel-tunnel TolC and the junctional pore complex of cyanobacteria are longer (14 and 32 nm, respectively), but their membrane-spanning domains do not exceed 4 nm (29,30).
The Surface Hydrophobicity Gradient Is Responsible for the Oriented Interaction of the MspA Pore with Lipid Membranes-The formation of protein micelles and the MspA zippers illustrate the existence of a hydrophobicity gradient along the MspA pore with voluminous hydrophilic domains at one end and a strongly hydrophobic portion at the other. This appears to be another new characteristic of MspA because protein micelles have not yet been observed for other bacterial porins. The assumption that the hydrophobicity gradient of MspA would lead to an oriented interaction with hydrophobic surfaces was supported by the highly asymmetric insertion of MspA into planar lipid membranes. This property also explains the oriented adsorption of the MspA pore onto hydrophobic carbon surfaces (31).
The mycolic acids in the mycobacterial cell wall form tight crystalline arrays (32), which exhibit a drastically decreasing fluidity toward its periplasmic end (12). The asymmetrical surface hydrophobicity and membrane insertion might reflect the orientation of the MspA pore in the asymmetrical mycolic acid layer and might be important in vivo for gating. Porins of Mycobacterium chelonae and Mycobacterium phlei close asymmetrically at rather low negative voltages of Ϫ40 and Ϫ20 mV, respectively, in artificial membranes (33,34) indicating that voltage gating might be a general property of porins from fast-growing mycobacteria. Because the existence of a Donnan potential across the mycobacterial cell wall similar to that observed in E. coli (35) was not experimentally demonstrated yet, it is not clear whether voltage gating of MspA in lipid bilayer experiments has any significance in vivo. However, the critical voltage of MspA, above which channels close, is only one third of the 90 mV, which is required to close the porin OmpF of E. coli (36). Moreover, there is growing evidence that a number of factors such as a low pH value (37) or the presence of polycations (38) or saccharides (39) reduce the critical voltage of porins from Gram-negative bacteria. Taking into account that charged lipids are only present in the outer layer of the mycolic acid membrane (40) it is tempting to speculate that a Donnan potential exists in mycobacteria and voltage gating of porins is exploited by mycobacteria to protect themselves from noxious molecules. However, this only partially accounts for the 50,000-fold lower permeability of the cell wall of M. smegmatis compared with E. coli for glucose (6). Molecular dynamics calculations of the diffusion of ions through the porin OmpF have shown that the diffusion constants are reduced to about 50% of their value in bulk solutions (41). Furthermore, diffusion rates through pore proteins decrease with the size of the substrate (42). Both effects are certainly more pronounced in the 2.5-fold longer MspA channel compared with the E. coli porins. Thus, the extraordinary length of the MspA pore hampers diffusion of hydrophilic compounds across the outer membrane of M. smegmatis further. A significant contribution might also originate from the putative periplasmatic end of the MspA channel, which was less filled with stain ( Fig. 4) indicating either an extremely hydrophobic channel interior or a channel constriction. Both properties would exclude uranyl acetate from staining and would further reduce the pore permeability for polar molecules. In conclusion, the length, the cone-like structure of the MspA pore with its less permissive end and the low number of pores may explain why the outer membrane of M. smegmatis has such an extraordinary low permeability for hydrophilic compounds. Similar mechanisms likely restrict the porin pathway in M. tuberculosis because the thickness of the outer membrane determines the minimal length of the porins and appears to be similar for mycobacteria (28) and the number of porins appears to be low in M. tuberculosis (43). and negative voltages (lower traces) were applied to the membrane when ϳ100 channels were reconstituted. The membrane current was recorded.