Identification of outer membrane proteins of Mycobacterium tuberculosis

SUMMARY

The cell wall of mycobacteria contains an unusual outer membrane of extremely low permeability. While Escherichia coli uses more than 60 proteins to functionalize its outer membrane, only two mycobacterial outer membrane proteins (OMPs) are known. The porin MspA of Mycobacterium smegmatis provided the proof of principle that integral mycobacterial OMPs share the β-barrel structure, the absence of hydrophobic α-helices and the presence of a signal peptide with OMPs of gram-negative bacteria. These properties were exploited in a multi-step bioinformatic approach to predict OMPs of M. tuberculosis. A secondary structure analysis was performed for 587 proteins of M. tuberculosis predicted to be exported. Scores were calculated for the β-strand content and the amphiphilicity of the β-strands. Reference OMPs of gram-negative bacteria defined threshold values for these parameters that were met by 144 proteins of unknown function of M. tuberculosis. Two of them were verified as OMPs of unknown functions by a novel two-step experimental approach. Rv1698 and Rv1973 were detected only in the total membrane fraction of M. bovis BCG in Western blot experiments, while proteinase K digestion of whole cells showed the surface accessibility of these proteins. These findings established that Rv1698 and Rv1973 are indeed localized in the outer membrane and tripled the number of known OMPs of M. tuberculosis. Significantly, these results provide evidence for the usefulness of the bioinformatic approach to predict mycobacterial OMPs and indicate that M. tuberculosis likely has many OMPs with β-barrel structure. Our findings pave the way to identify the set of proteins which functionalize the outer membrane of M. tuberculosis.

INTRODUCTION

The cell wall of mycobacteria is an intriguingly complex structure consisting of a great variety and a large amount of lipids^{1, 2}. Very long chain-fatty acids, the mycolic acids, are covalently bound to the arabinogalactan-peptidoglycan co-polymer and were proposed to form the inner layer of an asymmetric outer membrane while other lipids constitute the outer leaflet³. X-ray diffraction studies showed that the mycolic acids are indeed oriented in parallel and perpendicular to the plane of the cell envelope⁴. The presence of a second lipid bilayer outside of the cytoplasmic membrane in mycobacterial species has recently been visualized for the first time in a near native state by cryo-electron microscopy⁵. The discovery and the analysis of the porin MspA of Mycobacterium smegmatis provided the first conclusive evidence that functionally similar, but structurally completely different outer membrane proteins (OMPs) exist also in mycobacteria^6-9. Despite the well-documented importance of OMPs for the import of nutrients, secretion processes and host-pathogen interactions in gram-negative bacteria¹⁰, surprisingly few OMPs of mycobacteria are known. The only two well characterized examples of integral OMPs are the porin MspA of M. smegmatis and the channel-forming protein OmpA of M. tuberculosis^11-14. By contrast, E. coli uses more than 60 proteins to functionalize its outer membrane¹⁵, none of which has significant sequence similarity to any M. tuberculosis protein.

Traditionally, OMPs have been discovered by isolating the cell envelope and then separating the inner from the outer membrane in sucrose gradients^16-18. Due to the covalent linkages between the peptidoglycan, the arabinogalactan and the mycolic acid layer¹, it is difficult to mechanically lyse mycobacterial cells¹⁹. This is usually achieved only by harsh conditions, which inadvertently leads to mixing of components of both membranes. This has hampered localization experiments and identification of M. tuberculosis OMPs so far²⁰. Bioinformatic analysis of the genome of M. tuberculosis provides an alternative strategy, but OMPs are more difficult to identify by the amino acid sequence than inner membrane proteins, whose hydrophobic α-helices are predicted with accuracies exceeding 99%²¹. So far, all known OMPs are β-barrel proteins, which are characterized by a pattern of alternating hydrophobic and hydrophilic amino acids in the β-strands forming the β-barrel²². Such a pattern is recognizable in the protein sequences and has been exploited in recent years to develop programs for prediction of β-barrel proteins. For example, more than 10 previously unknown OMPs were predicted for E. coli²³. Notably, a consensus method performed better than each individual prediction method for a set of 20 β-barrel OMPs whose structures are known at atomic resolution²⁴. The success of these approaches motivated the application of one of these algorithms to predict OMPs of M. tuberculosis²⁵. However, the usefulness of this analysis is limited for several reasons: (i) Pajon et al. did not exclude proteins with hydrophobic α-helices and therefore the list of predicted OMPs contains a large number of inner membrane proteins (IMPs). (ii) One of the two variables chosen as predictors of putative β-barrel structures was based on the prevalence of a C-terminal phenylalanine²⁶, which is typical for OMPs of gram-negative bacteria but is not present in the two known mycobacterial OMPs MspA and OmpA^{6, 7, 11}. (iii) Proteins with sequence homology to known cytoplasmic and periplasmic proteins or lipoproteins, which are anchored in membranes by their lipid moieties and not by a transmembrane β-barrel²⁷, were not excluded from the list.

In this study, we have employed an algorithm entirely based on physical principles to predict OMPs of M. tuberculosis. In contrast to all other prediction methods so far, we provide a scoring system for the number and amphiphilicity of β-strands of a particular protein. By combining both parameters with biological knowledge we predicted 144 proteins as OMPs of Mtb. The subcellular localization of two of these proteins was determined by demonstrating their association with membranes and their surface accessibility to proteases. This alternative approach identified Rv1698 and Rv1973 as OMPs of M. tuberculosis and provided experimental evidence for the usefulness of the bioinformatic predictions.

MATERIAL AND METHODS

Genome-wide analysis to identify putative OMPs of M. tuberculosis

A FASTA file with 3991 protein sequences for the Mycobacterium tuberculosis H37Rv genome was obtained from ftp://ftp.ncbi.nih.gov/genbank/genomes/Bacteria/Mycobacterium_tuberculosis_H37Rv/[version: AL123456.2 date: 04/18/2006,²⁸]. All sequences were scanned for sequence similarity matches against the Pfam database of protein motifs (version 20.0²⁹) using hmmpfam^{30, 31}. To identify exported proteins of M. tuberculosis we predicted the presence of an N-terminal signal peptide using SignalP 3.0^{32, 33}. All proteins that had a Y_max score ≥ 0.51 were considered as exported proteins. These proteins were selected and the sequences corresponding to the signal peptide predicted by SignalP were removed. These shortened sequences should represent the mature proteins and were used for all further analyses. Next, the sequences were examined by TMHMM to predict transmembrane α-helices^{34, 35}. For all proteins that did not have a hydrophobic α-helix the secondary structure was predicted from the sequence using the Jnet algorithm³⁶ which gives the best performance among secondary structure prediction algorithms and achieves a 76.4% average accuracy on a large test set of proteins³⁷. Predicted β-strands of a minimum of five consecutive residues were registered. Next, we computed the amphiphilicity of these β-strands. To this end, the mean hydrophobicity of one side of a β-strand H_β(i) was calculated at position i in a sequence following Vogel and Jähnig³⁸ as H_β(i) = 1/5 × (h(i-4)+h(i-2)+h(i)+h(i+2)+h(i+4)), where h(i) is the hydrophobicity of the amino-acid at position i. Note that in a sequence of aminoacids from 1 to N, these values can only be computed for i = 5, ..., N - 4. The values of hydrophobicity for the amino-acids were taken from Sweet and Eisenberg³⁹. Given the average value of hydrophobicity over a whole sequence H_βm, we counted the zero crossings of the H_β - H_βm. We combined this measurement with the secondary structure prediction. The number of hydrophobicity crossings per residues in β-strand was defined as “amphiphilicity” of the β-strands of a protein and used as discriminating parameter. Higher values indicate the propensity to form a transmembrane β-barrel (Fig. 1). Further, the number of cysteines was counted, and the pI of the protein was computed using the piCalculator BioPerl module (http://www.bioperl.org/).

Fig. 1. Prediction of putative OMPs of M. tuberculosis H37Rv.

A. Strategy. The 3991 predicted proteins of M. tuberculosis H37Rv were analyzed for the presence of a signal peptide using the SignalP algorithm³². The 587 proteins containing a signal peptide were analyzed using the TMHMM algorithm³⁵ to recognize hydrophobic α-helices. Their β-strand content was predicted using Jnet³⁶. The amphiphilicity of the β-strands was computed using an algorithm from Vogel and Jähnig³⁸.

B. β-strand content and amphiphilicity of the predicted exported proteins. The minimal length of a transmembrane β-strand was set to five amino acids. The red diamonds represent 29 known OMPs from gram-negative bacteria and from mycobacteria (MspA, OmpA/Rv0899).

C. β-strand content and amphiphilicity of 144 putative outer membrane proteins. Only exported proteins are depicted that do not have a transmembrane α-helix and have β-strand content of at least 0.09 and amphiphilicity of at least 0.19. In addition, all proteins with similarities to proteins with other subcellular localizations are not shown. Rv1698 and Rv1973 were chosen to experimentally examine their subcellular localization. The purple, orange and green diamonds represent members of the PE-PGRS, Mce and PPE families, respectively.

Analysis of the secondary structure of selected proteins

The SignalP3.0 algorithm³² was accessed at http://www.cbs.dtu.dk/services/SignalP/ to confirm the presence of signal peptides for selected proteins of M. tuberculosis. The TMHMM program^{34, 35} and the ConPred II prediction server⁴⁰ were used for prediction of hydrophobic transmembrane α-helices. M. tuberculosis proteins were considered not to be integral inner membrane proteins when both methods did not predict a transmembrane α-helix. All programs were used with standard settings unless otherwise noted.

Bacterial strains and growth conditions

Mycobacterium smegmatis mc²155 was grown at 37°C in Middlebrook 7H9 liquid medium (Difco Laboratories) supplemented with 0.2% glycerol, 0.05% Tween 80 or on Middlebrook 7H10 agar (Difco Laboratories) supplemented with 0.2% glycerol unless indicated otherwise. For growth of Mycobacterium bovis BCG ATCC 27291 the enrichment OADC (BD Biosciences) was added to the Middlebrook 7H9 liquid medium. Escherichia coli DH5α and Escherichia coli Rosetta (Novagen) were used for cloning experiments and for overexpression of ompA, rv1698 and rv1973, respectively, and were routinely grown in LB medium at 37°C. The following antibiotics were used when required at the following concentrations: ampicillin (100 μg ml^-1 for E. coli), kanamycin (30 μg ml^-1 for E. coli; 10 μg ml^-1 for mycobacteria), hygromycin (200 μg ml^-1 for E. coli, 50 μg ml^-1 for mycobacteria).

Overexpression in E. coli and preparation of recombinant Rv1698, Rv1973 and OmpA

The T7 expression system was chosen to express rv1698, rv1973 and ompA in E. coli as previously described for the porin gene mspA of M. smegmatis ⁴¹. The genes were amplified from chromosomal DNA of M. tuberculosis H37Rv by PCR using the primers as listed in Table S2. The truncated rv1698, rv1973 and ompA genes lacking the first 90, 69, and 132 nucleotides were generated from plasmids pMN335 (rv1698), pMN339 (rv1973) and pML588 (OmpA) using corresponding primers (Table S2) by PCR, respectively. These genes were cloned into the vector pET28b+ (Novagen) or its derivate under control of the T7 promoter using corresponding restriction sites (Table 3). The recombinant rv1698 and rv1973 genes encoded an N-terminal His₆ tag, whereas recombinant ompA genes encoded a C-terminal His₆ tag. The resulting plasmids pML122 (rv1698), pML123 (rv1973) and pML591 (ompA) were transformed into E. coli Rosetta.

For protein expression and purification, 1 L culture was grown for each recombinant strain to OD₆₀₀=0.6-1.0, and induced with isopropyl-β-D-thiogalactopyranoside (IPTG) at a final concentration of 0.5 mM for 2 h at 37°C. The bacteria were harvested by centrifugation, resuspended in 20 ml lysis buffer (50 mM Tris-HCl, 100 mM NaCl, 0.5% Triton X-100, pH 8.0) and sonicated on ice using sonicator 3000 (Misonix). After sonication, DNAse (final con. 0.01mg/ml) and lysozyme (final con. 0.1mg/ml) were added, and incubated at RT for 20 min. Then, the mixture was centrifuged at 4,500xg for 15 min at 4 °C, and the pellet was resuspended in 20 ml lysis buffer, sonicated on ice, and centrifuged as above to separate inclusion bodies from cell debris. The inclusion body (pellet) was washed with 20 ml lysis buffer (without Triton X-100) three times and resuspended in 500 μl TS buffer (50 mM Tris-HCl, 100 mM NaCl, pH8.0), dispersed completely by sonication, and then dissolved drop wise into 8 ml TS buffer with 8.5 M urea. One fifth of suspended inclusion bodies (100 μl) was used for each column purification, mixed gently with 500 μl equilibration buffer (50 mM Tris-HCl, 100 mM NaCl and 0.5% OPOE, pH8.0) and put into ultrasonication bath (FS60H, Fisher Scientific) for 15-30 min to disperse proteins completely before loading. Ni²⁺ charged resin column (HIS-Select™ Spin Columns, Sigma) was equilibrated with 600 μl equilibration buffer and loaded with treated sample. Unbound protein was washed out three times using 600 μl wash buffer (50 mM Tris-HCl, 100 mM NaCl, 0.5% OPOE, 5 mM imidazole, pH8.0). Bound proteins were eluted from the column using 500 μl elution buffer (50 mM Tris-HCl, 100 mM NaCl, 0.5% OPOE, 500 mM imidazole, pH8.0).

The recombinant proteins of rRv1698_His and rRv1973_His were purified from inclusion bodies, whereas most of the rOmpA_His protein was solubilized from lysed cells of E. coli without any detergents. The rOmpA_His protein was purified by Ni²⁺-affinity chromatography as described above with minor modifications. The protein concentrations were determined using bicinchoninic acid (BCA; Pierce, Rockford, USA). The purified proteins were used as positive controls in Western blot and ELISA experiments.

Generation of polyclonal antisera against putative outer membrane proteins of M. tuberculosis

To generate a polyclonal antisera against Rv1698 and Rv1973, 200 μg purified recombinant protein per rabbit was injected using TiterMax® adjuvant (TiterMax). After 28 days, serum samples were taken and checked for protein-specific antibodies in ELISA experiments. All absorptions were 40-fold increased when 1 μg purified protein was incubated with serum compared to incubation with preimmunisation serum of the same animal. The terminal bleedings were done at day 35. The antisera were obtained from Rockland Immunochemicals, Inc. The anti-OmpA antiserum was kindly provided by Dr. Philip Draper.

Expression of putative outer membrane proteins of M. tuberculosis in M. bovis BCG

The full-length rv1698, rv1973 and ompA genes were generated from chromosomal DNA M. tuberculosis H37Rv by PCR, and used to replace the mspA fragment in pMN016 using restriction sites PacI and SwaI resulting under control of the p_smyc promoter (Table 1). In the resulting plasmids pMN035 (rv1698), pMN039 (rv1973) and pML003 (ompA), the p_smyc promoter was exchanged by p_imyc promoter, which was obtained from pMN013⁴², to yield the plasmids pMN335 (rv1698), pMN339 (rv1973) and pML588 (ompA) using the restriction sites PacI and PmeI. The M. bovis BCG strain was transformed with the expression vectors pMN335 (rv1698), pMN339 (rv1973) and pML588 (ompA) and carrying the M. tuberculosis genes in fusion with p_imyc promoter. The vectors pMN013 (p_imyc-mspA), pMN437 (p_smyc-mycgfp2+) and pML970 (p_smyc-phoA_HA) were used as positive and negative controls, respectively. The M. bovis BCG strains carrying the corresponding plasmids were streaked on 7H10 agar plates, and inoculated into 7H9 liquid medium until OD₆₀₀=0.8.

Table 1. Plasmids used in this work.

“Origin” means origin of replication. The annotations Hyg^R and Kan^R indicate that the plasmids confer resistance to hygromycin and kanamycin, respectively

Plasmid	Parent vector, relevant genotype and properties	Source or reference
pMN013	Pimyc-mspA; ColE1 origin; PAL5000 origin; HygR; 6000 bp	42
pMN016	psmyc-mspA; ColE1 origin; PAL5000 origin; HygR; 6164 bp	8
pMN406	pimyc-mycgfp2+; ColE1 origin; PAL5000 origin; HygR; 6059 bp	will be published elsewhere
pMN437	psmyc-mycgfp2+; ColE1 origin; PAL5000 origin; HygR; 6236 bp	will be published elsewhere
pET28b+	f1 origin, pBR322 origin, LacI; KanR; 5368 bp	Novagen
pMN035	Psmyc-rv1698; ColE1 origin; PAL5000 origin; HygR; 6484 bp	this study
pMN039	Psmyc-rv1973; ColE1 origin; PAL5000 origin; HygR; 6004 bp	this study
pMN335	Pimyc-rv1698; ColE1 origin; PAL5000 origin; HygR; 6307 bp	this study
pMN339	Pimyc-rv1973; ColE1 origin; PAL5000 origin; HygR; 5827 bp	this study
pML003	Psmyc-ompA; ColE1 origin; PAL5000 origin; HygR; 6534 bp	this study
pML122	PT7-hisrv1698; f1 origin, pBR322 origin, LacI; KanR; 6184 bp	this study
pML123	PT7-hisrv1973; f1 origin, pBR322 origin, LacI; KanR; 5736 bp	this study
pML440	pimyc-phoA; ColE1 origin; PAL5000 origin; HygR; 7707 bp	45
pML588	Pimyc-ompA; ColE1 origin; PAL5000 origin; HygR; 6357 bp	this study
pML591	PT7-ompAhis; f1 origin, pBR322 origin, LacI; KanR; 6129 bp	this study
pML970	pimyc-phoAHA; ColE1 origin; PAL5000 origin; HygR; 6895 bp	this study

Subcellular fractionation of M. bovis BCG

To separate membrane proteins from cytoplasmic proteins of M. bovis BCG we made use of an established protocol⁴³. Each strain was grown to OD₆₀₀=0.8 in 7H9 Middlebrook medium supplemented with 10% OADC, 0.2% glycerol and 0.05% Tween 80 with or without hygromycin and harvested by centrifugation. The cells were washed twice with PBS (80 mM Na₂HPO₄, 20 mM NaH₂PO₄, 100 mM NaCl, pH 7.2), resuspended in PBS and lysed by sonication 5 times for 20 sec on ice with 10 sec interval and 12 Watt output power. Unbroken cells were removed by low speed centrifugation at 4,000xg. The supernatant of cell lysates was subjected to ultra-centrifugation at 100,000xg for 1 h. The resulting supernatant (SN100) was isolated and the pellet was washed 3 times with PBS (P100). The P100 pellet consists of cell envelope material including inner and outer membrane proteins. The P100 pellet was resuspended in PBS buffer containing 2% SDS and extracted at 50°C for 2 h before performing ELISA and Western blot analysis. All samples were mixed with 4 fold protein loading buffer (160 mM Tris-Cl pH 7.0, 12% SDS, 32% glycerol, 0.4% Bromophenol blue) according to 1:3 ratio (v/v for supernatant, or v/w for pellet) and boiled for 5 min before loading on the 10% SDS-PAGE gel.

Subcellular fractions of M. tuberculosis

Subcellular fractions of M. tuberculosis H37Rv such as the cell wall (CW), total membrane (MEM) and SDS-soluble cell wall proteins (SCWP), cytosol and culture filtrate were obtained from Colorado State University (CSU) as part of the NIH (NIAID) Contract HHSN266200400091C entitled “Tuberculosis Vaccine Testing and Research Materials”. The protocols describing how these fractions were produced are available at http://www.cvmbs.colostate.edu/microbiology/tb/pdf/scf.pdf. Briefly, M. tuberculosis H37Rv was grown to late-log phase (day 14) in glycerol-alanine-salts (GAS) medium, washed with PBS (pH 7.4) and inactivated by gamma-irradiation. The cells are then suspended in PBS containing 8 mM EDTA, DNase, RNase and a proteinase inhibitor cocktail (PMSF, pepstatin A, and leupeptin), and broken at 4°C. Unbroken cells are removed by low speed centrifugation (3,000xg). The cell wall including the outer membrane is first isolated by 27,000xg centrifugation for 45 min. The cell wall pellet is collected (CW), suspended and dialyzed in 0.01M ammonium bicarbonate, and extracted at 50°C with 2% SDS for 2h to yield the SDS-soluble cell wall proteins (SCWP). The supernatant of the 27,000xg centrifugation consisting of soluble proteins and inner membrane is further separated by centrifugation at 100,000xg for four hours. The proteins in the 100,000xg supernatant consist mainly of water-soluble cytoplasmic and periplasmic proteins (cytosol), while the pellet primarily contains membrane proteins (MEM). The culture supernatant is harvested from the live cells by passing through a 0.2 micron filter. The culture filtrate (CFP) is concentrated by Amicon ultrafiltration using a membrane with a molecular weight cutoff of 5,000 Da.

Western blot analysis

Proteins from bacterial extracts (such as the cell wall, total membrane and SDS-soluble cell wall proteins), cell lysates as obtained after sonication, supernatants and SDS-extracted pellets after ultracentrifugation of lysates, and purified proteins, were separated on a SDS-containing 10% polyacrylamide gel. The proteins were transferred to a PVDF membrane using a standard protocol ⁴⁴ and were detected by the rabbit antiserum against MspA (pAK #813)⁶, OmpA, Rv1698, Rv1973 and GFP (Sigma), respectively. Horseradish peroxidase coupled to an anti-rabbit antibody oxidized Luminol (ECL plus kit, Amersham) whose chemoluminescence was detected by EpiChemi³ Darkroom (UVP BioImaging system).

Enzyme-linked immunosorbent assay with whole cells of M. bovis BCG

To examine the accessibility of a protein on the cell surface of M. bovis BCG, an enzyme-linked immunosorbent assay with whole cells was employed as described earlier for M. smegmatis⁷. Cells were grown to an OD₆₀₀ of about 0.8, harvested by centrifugation, washed twice in PBS containing 0.05% Tween 80 and resuspended in 50 mM NaHCO₃ (pH 9.6) to yield a cell concentration of about 10⁹ cells/ml. Aliquots of 100 μl of cell suspension or dilutions thereof were transferred into wells of microtiter plates (NUNC-Immuno™ MaxiSorp™ Surface, Nalge Nunc International). At the same time, lysed cells as obtained after sonication, supernatants and SDS-extracted pellets after ultracentrifugation of lysates were also coated in the wells. After incubation overnight at 4 °C, wells were washed two times with 200 μl TBST buffer containing 50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 mM MgCl₂ and 0.05 % Tween 80. Remaining protein binding sites were blocked with 200 μl 5 % powdered skim milk in TBST for 1 h at room temperature. Polyclonal antibodies were diluted 1:2000 and incubated with cells for 1.5 h at room temperature. The wells were washed 3 times with 200 μl TBST. Horseradish peroxidase conjugated secondary antibody (Sigma) was diluted 1:10000 in TBST and incubated at room temperature for 1 h. After 4 washing steps with 200 μl TBST, 50 μl of O-phenylenediamine substrate (Sigma) was added per well. To harvest the cells after each incubation and during the washing steps, special swing-out insets were used for centrifugation (3000 rpm for 5 min). The supernatant was removed by carefully decanting the plates. All incubations were carried out at room temperature. Incubation of the sample with substrate occurred in the dark until a color change to yellow occurred. Then, the reaction was stopped by addition of 50 μl 1 M H₂SO₄. The plate was centrifuged at 3000 rpm for 5 min and the supernatant was transferred into a new 96-well plate for measuring absorption at 490 nm with a microplate reader (Synergy HT, Bio-TEK Instrument Inc, USA). The lysed cells, supernatants and pellets after sonication, and purified recombinant proteins were also coated on the well and treated in the same way as described above.

Protease accessibility assay

PhoA was used as a control for the protease accessibility assay. To this end, the M. smegmatis phoA gene was amplified from the vector pML440 ⁴⁵ by PCR using the primers phoASD_01 and phoA-HA (Table S1). The SphI digested PCR fragment was cloned into the backbone of pMN406 obtained by digestion with SphI and SwaI to yield the vector pML970 expressing a PhoA_HA fusion protein under the control of the mycobacterial promoter p_imyc (Table 1). To examine the surface accessibility of GFP, PhoA_HA, Rv1698 and Rv1973, protease accessibility experiments were performed as described previously⁴⁶ with minor modifications. M. bovis BCG strains carrying the plasmids pMN437 (p_smyc-mycgfp2+), pML970 (p_imyc-phoA_HA), pMN335 (p_imyc-rv1698), pMN339 (p_imyc-rv1973) and wt M. bovis BCG were grown in 50 ml of Middlebrook 7H9/OADC medium and harvested as the cultures reached an OD₆₀₀ of 4.0. The cells were washed once with TBS buffer (50 mM Tris-HCl pH 7.2, 150 mM NaCl, 3 mM KCl) and then re-suspended in 1 ml of the same buffer. Two aliquots of 200 μl were taken and Proteinase K (Sigma-Aldrich) was added to one of the aliquots to a final concentration of 100 μg/ml. After 30 min incubation at 37 °C the reaction was stopped by adding complete EDTA free protease inhibitor cocktail (Roche) from a 7-fold stock solution. The samples were immediately centrifuged, washed twice in 500 μl TBS and re-suspended in 150 μl TBS containing 1% SDS. The suspensions were kept at 40°C with shaking (850 rpm) for 30 min, and 50 μl of 4x protein loading buffer was added into each sample. All samples were boiled for 10 min, centrifuged to remove insoluble debris and 20 μl of the protein extract were separated on a 10% polyacrylamide gel and analysed by western blotting using the appropriate antibodies and standard protocols⁴⁴.

RESULTS

Prediction of exported proteins of Mycobacterium tuberculosis

The easiest way to identify putative OMPs of M. tuberculosis would be to find homologs of OMPs known from gram-negative bacteria. However, an extensive search did not yield any homologs except for OmpA, which showed a weak similarity in the periplasmic C-terminal domain^{11, 47}. The apparent difference of integral OMPs may reflect the different chemical environments in the outer membranes of mycobacteria and gram-negative bacteria as previously noted⁴⁸. This assumption was confirmed by the crystal structure of the porin MspA of M. smegmatis⁹. In an alternative approach, we, therefore, followed a four-step strategy based on secondary structure analysis to identify putative OMPs of M. tuberculosis as outlined in Fig. 1A. The vast majority of bacterial OMPs have a canonical N-terminal signal sequence (positive charges, hydrophobic α-helix, signal peptidase cleavage site), which targets proteins to the Sec system for translocation across the inner membrane⁴⁹. Therefore, the first step was to identify all M. tuberculosis proteins with a signal peptide using the SignalP algorithm³². This algorithm predicted 587 out of 3991 annotated proteins of M. tuberculosis H37Rv to be exported by the Sec translocase (Table S2). Notably, OmpA (Rv0899), a pore-forming OMP, which is required for growth of M. tuberculosis at low pH and for survival in mice¹², was not in this list, because its N-terminus does not fit well the definitions of classical signal peptide as noted earlier¹¹ and confirmed recently¹⁴.

Identification of inner membrane proteins of Mycobacterium tuberculosis

The final subcellular location of exported proteins in gram-negative bacteria depends on their physical and structural properties and could be (i) the inner membrane, (ii) the periplasm, (iii) the outer membrane or (iv) the extracellular space (medium)⁵⁰. Proteins with at least one hydrophobic α-helix after cleavage of the signal peptide are assumed to be inner membrane proteins because an hydrophobic α-helix acts as a stop-transfer sequence and anchors the protein in the inner membrane of gram-negative bacteria⁵¹. Hence, the next step was to identify inner membrane proteins in the list of the 587 exported proteins of M. tuberculosis. We used the TMHMM algorithm³⁵, which recognizes hydrophobic α-helices with near 100% reliability, to identify 134 IMPs (Table S1) within the group of 587 exported proteins of M. tuberculosis. These proteins constitute a subgroup of the 787 inner membrane proteins of M. tuberculosis as provided by the PEDANT database (http://pedant.gsf.de). It is concluded that more than 80% of all IMPs of M. tuberculosis do not appear to have a canonical signal peptide. The majority of the 134 inner membrane proteins with signal peptide showed sequence similarities to inner membrane proteins known from other bacteria. In order to identify putative OMPs we were interested in the 453 remaining proteins without certain transmembrane helix.

Prediction of outer membrane proteins of Mycobacterium tuberculosis

All known integral OMPs of gram-negative bacteria have a β-barrel structure with a hydrophobic surface^{22, 52}. The β-barrel structure of MspA⁹ and its localization in the outer membrane of mycobacteria⁷ indicate that this also applies to mycobacterial OMPs. Amino acids in β-strands that are part of a membrane-spanning β-barrel have alternating hydrophilic and hydrophobic residues^{9, 53, 54}. This pattern can be recognized in the protein sequence and has been exploited by a number of algorithms to detect OMPs in gram-negative bacteria^{24, 55, 56}. However, this approach is only useful after exclusion of integral inner membrane proteins, which sometimes also follow such a pattern in extramembrane domains²³. To test the usefulness of the Jnet algorithm³⁶, the secondary structure of 29 known OMPs of gram-negative bacteria were predicted (Table 1). β-Barrel proteins such as OmpC and OmpF of E. coli consist entirely of β-strands as shown in the crystal structures⁵³ and yielded β-strand scores between 0.18 and 0.28. All reference proteins with the exception of TolC have a β-strand score of at least 0.1 (Fig. 1C, Table S2). The amphiphilicity of these β-strands was calculated using an algorithm which was specifically developed for detection of OMPs³⁸. Values close to one indicate that their β-strands show a perfect pattern of alternating hydrophilic and hydrophobic residues, whereas a value of zero means that all residues in each beta-strand are either hydrophobic of hydrophilic. Eleven of these proteins including MspA of M. smegmatis and OmpA of M. tuberculosis have an amphiphilicity score exceeding 0.5 (Fig. 1C) indicating that more than half of the amino acids in their predicted β-strands consisted of alternating hydrophilic and hydrophobic residues. The porin of Rhodobacter capsulatus⁵⁷ set the lower limit with an amphiphilicity score of 0.19. Therefore, we conservatively set threshold values of 0.09 for the β-strand score and of 0.19 for the amphiphilicity for putative OMPs of M. tuberculosis. These thresholds were met by 299 out of the 453 proteins without certain transmembrane helix (Table S2). Then, the following proteins were eliminated: (i) 72 proteins annotated as lipoproteins, which are not integral membrane proteins but are anchored solely by an acyl chain into the membrane²⁷, (ii) 74 proteins with similarities to known periplasmic and cytoplasmic proteins, and (iii) 9 secreted proteins. This reduced the number of putative OMPs to 144 (Fig. 1C, Table 2). Most of these putative OMPs (94/144) do not have any homologs of known function and are therefore annotated as unknown proteins. In addition, 19 Mce, 11 PE/PGRS and 10 PPE proteins have secondary structures similar to those of OMPs (Fig. 1C, Table 2). Nine proteins that have only one β-strand consisting of five or more residues are marked with a star in Table 2. These proteins are unlikely to form a β-barrel and to be integral OMPs.

Table 2. Putative OMPs of M. tuberculosis.

The length represents the theoretical number of amino acids of the mature protein after removal of the predicted signal peptide. The signal peptide score Y_max was taken from SignalP 3.0. All further calculations were done with the sequence of the mature protein. The number of β-strands with a length of five or more amino acids was predicted by using the Jnet algorithm. Amphiphilicity is defined as the fraction of alternating hydrophilic and hydrophobic residues in β-strands. This set of putative OMPs was obtained by applying cutoff values of >0.09 for the β-strand percentage and of >0.19 for the amphiphilicity score. Proteins with only one β-strand consisting of 5 or more residues are marked with a star. A core set of 32 putative OMPs was defined by an amphiphilicity score of >0.39 and an isoelectric point (pI) of <6.0. These proteins are highlighted in grey

No.	Rv#	ID	GeneSymbol	Description	Length	SP score	β-strands	% β-strand	Amphiphilicity	Cysteines	pI
1	Rv0088	CAA98924	none	HYPOTHETICAL PROTEIN	196	0.8	5	0.24	0.56	0	10.75
2	Rv0116c	CAA17310	none	POSSIBLE CONSERVED MEMBRANE PROTEIN	222	0.8	7	0.4	0.34	1	7.89
3	Rv0152c	CAE55246	PE2	PE FAMILY PROTEIN	485	0.8	8	0.14	0.27	1	6.4
4	Rv0164	CAE55251	TB18.5	CONSERVED HYPOTHETICAL PROTEIN TB18.5	131	0.6	4	0.28	0.42	0	4.31
5	Rv0169	CAE55253	mce1A	MCE-FAMILY PROTEIN MCE1A	414	0.7	8	0.21	0.63	2	4.59
6	Rv0170	CAB09753	mce1B	MCE-FAMILY PROTEIN MCE1B	313	0.6	7	0.19	0.48	2	8.21
7	Rv0171	CAB09754	mce1C	MCE-FAMILY PROTEIN MCE1C	476	0.7	5	0.11	0.67	2	4.9
8	Rv0172	CAB09755	mce1D	MCE-FAMILY PROTEIN MCE1D	497	0.7	8	0.17	0.41	2	4.61
9	Rv0174	CAB09741	mce1F	MCE-FAMILY PROTEIN MCE1F	489	0.9	6	0.17	0.48	8	4.94
10	Rv0225	CAB06992	none	POSSIBLE CONSERVED PROTEIN	380	0.8	7	0.19	0.44	3	10.14
11	Rv0241c	CAA17333	none	CONSERVED HYPOTHETICAL PROTEIN	257	0.6	4	0.22	0.26	0	10.18
12	Rv0257*	CAE55259	none	CONSERVED HYPOTHETICAL PROTEIN	104	0.6	1	0.18	0.3	3	8.99
13	Rv0295c	CAA17370	none	CONSERVED HYPOTHETICAL PROTEIN	239	0.6	4	0.13	0.82	1	5.61
14	Rv0309	CAB09582	none	POSSIBLE CONSERVED EXPORTED PROTEIN	183	0.9	4	0.23	0.32	3	8.47
15	Rv0320	CAB09604	none	POSSIBLE CONSERVED EXPORTED PROTEIN PROBABLE CONSERVED MEMBRANE PROTEIN	195	0.6	5	0.19	0.46	2	6.51
16	Rv0403c	CAB06594	mmpS1	MMPS1 PROBABLE CONSERVED MEMBRANE PROTEIN	120	0.8	4	0.38	0.21	2	6.62
17	Rv0451c	CAA17408	mmpS4	MMPS4	112	0.7	5	0.38	0.39	2	4.75
18	Rv0455c*	CAA17411	none	CONSERVED HYPOTHETICAL PROTEIN PROBABLE CONSERVED MEMBRANE PROTEIN	117	0.9	1	0.2	0.6	2	6.5
19	Rv0506	CAB00932	mmpS2	MMPS2	123	0.8	5	0.33	0.59	2	7.51
20	Rv0584	CAA17455	none	POSSIBLE CONSERVED EXPORTED PROTEIN	844	0.8	28	0.28	0.33	3	4.76
21	Rv0589	CAE55301	mce2A	MCE-FAMILY PROTEIN MCE2A	365	0.6	4	0.23	0.62	2	4.58
22	Rv0590	CAB09962	mce2B	MCE-FAMILY PROTEIN MCE2B	240	0.6	8	0.25	0.49	2	6.72
23	Rv0592	CAB09960	mce2D	MCE-FAMILY PROTEIN MCE2D	472	0.8	5	0.16	0.65	2	4.63
24	Rv0594	CAB09959	mce2F	MCE-FAMILY PROTEIN MCE2F	478	0.8	9	0.2	0.35	8	5.24
25	Rv0614	CAB09971	none	CONSERVED HYPOTHETICAL PROTEIN	287	0.5	4	0.14	0.2	2	6.95
26	Rv0677c	CAA17460	mmpS5	POSSIBLE CONSERVED MEMBRANE PROTEIN MMPS5	108	0.7	4	0.33	0.5	2	4.11
27	Rv0679c	CAA17462	none	CONSERVED HYPOTHETICAL THREONINE RICH PROTEIN	122	0.8	6	0.41	0.44	2	4.06
28	Rv0755c	CAE55320	PPE12	PPE FAMILY PROTEIN	613	0.6	7	0.15	0.39	0	4.22
29	Rv0774c	CAB02386	none	PROBABLE CONSERVED EXPORTED PROTEIN	273	0.6	3	0.1	0.21	2	5.97
30	Rv0799c	CAB09574	none	CONSERVED HYPOTHETICAL PROTEIN	304	0.6	8	0.25	0.49	2	5.18
31	Rv0817c	CAA17623	none	PROBABLE CONSERVED EXPORTED PROTEIN	242	0.8	4	0.19	0.59	0	5.61
32	Rv0875c	CAA97382	none	POSSIBLE CONSERVED EXPORTED PROTEIN	137	0.8	4	0.3	0.42	2	6.34
33	Rv0878c	CAE55333	PPE13	PPE FAMILY PROTEIN	407	0.6	4	0.1	0.24	1	6.25
34	Rv0888	CAA97395	none	PROBABLE EXPORTED PROTEIN	458	0.5	13	0.24	0.41	6	4.63
35	Rv0906	CAA97381	none	CONSERVED HYPOTHETICAL PROTEIN	335	0.7	6	0.18	0.29	2	6.18
36	Rv0980c	CAE55345	PE_PGRS18	PE/PGRS FAMILY PROTEIN	423	0.7	13	0.29	0.22	1	4.06
37	Rv0988	CAA17587	none	POSSIBLE CONSERVED EXPORTED PROTEIN	353	0.8	7	0.2	0.55	0	4.64
38	Rv0999	CAB08148	none	HYPOTHETICAL PROTEIN	212	0.7	8	0.27	0.24	4	6.5
39	Rv1006	CAB08141	none	HYPOTHETICAL PROTEIN	538	0.9	8	0.17	0.58	4	4.97
40	Rv1067c	CAE55351	PE_PGRS19	PE/PGRS FAMILY PROTEIN	636	0.9	7	0.13	0.22	0	5.12
41	Rv1081c	CAA17197	none	PROBABLE CONSERVED MEMBRANE PROTEIN	101	0.6	6	0.5	0.83	2	8.2
42	Rv1087	CAE55354	PE_PGRS21	PE/PGRS FAMILY PROTEIN	736	0.6	6	0.1	0.23	0	4.05
43	Rv1091	CAE55359	PE_PGRS22	PE/PGRS FAMILY PROTEIN	822	0.7	9	0.12	0.23	0	4.11
44	Rv1135c	CAE55364	PPE16	PPE FAMILY PROTEIN LOW MOLECULAR WEIGHT T-CELL ANTIGEN	589	0.8	11	0.14	0.31	0	3.96
45	Rv1174c	CAA15851	TB8.4	TB8.4	81	0.8	3	0.23	0.25	2	4.37
46	Rv1184c	CAA15861	none	POSSIBLE EXPORTED PROTEIN	333	0.6	5	0.11	0.56	0	4.56
47	Rv1209	CAB07832	none	CONSERVED HYPOTHETICAL PROTEIN	99	0.7	2	0.11	1	0	4.84
48	Rv1268c	CAB00911	none	HYPOTHETICAL PROTEIN	199	0.9	4	0.21	0.45	1	4.17
49	Rv1325c	CAE55377	PE_PGRS24	PE/PGRS FAMILY PROTEIN	572	0.6	10	0.17	0.21	0	3.96
50	Rv1339	CAA99972	none	CONSERVED HYPOTHETICAL PROTEIN	248	0.5	5	0.19	0.25	5	5.41
51	Rv1351	CAA99984	none	HYPOTHETICAL PROTEIN	79	0.6	3	0.24	0.44	4	11.66
52	Rv1375	CAB02636	none	CONSERVED HYPOTHETICAL PROTEIN	422	0.6	8	0.18	0.42	6	4.78
53	Rv1382	CAB02643	none	PROBABLE EXPORT OR MEMBRANE PROTEIN	131	0.6	2	0.19	0.43	0	4.95
54	Rv1386*	CAE55383	PE15	PE FAMILY PROTEIN	70	0.6	1	0.23	0.78	1	4.38
55	Rv1419	CAB02167	none	HYPOTHETICAL PROTEIN	123	0.6	3	0.2	0.54	6	4.08
56	Rv1433	CAB09251	none	POSSIBLE CONSERVED EXPORTED PROTEIN	250	0.6	7	0.28	0.43	1	5.05
57	Rv1477	CAA16005	none	HYPOTHETICAL INVASION PROTEIN	432	0.9	5	0.14	0.33	1	6.82
58	Rv1478	CAA16006	none	HYPOTHETICAL INVASION PROTEIN	209	0.7	4	0.19	0.45	1	9.88
59	Rv1488	CAB02038	none	POSSIBLE EXPORTED CONSERVED PROTEIN	346	0.6	4	0.15	0.33	0	6.16
60	Rv1548c	CAE55401	PPE21	PPE FAMILY PROTEIN	650	0.8	9	0.15	0.25	0	4.18
61	Rv1566c	CAB09071	none	Possible inv protein	201	0.9	3	0.17	0.5	0	10.38
62	Rv1669	CAA17609	none	HYPOTHETICAL PROTEIN	93	0.5	2	0.29	0.31	3	8.5
63	Rv1698	CAB10955	none	CONSERVED HYPOTHETICAL PROTEIN	285	0.7	6	0.19	0.55	0	4.83
64	Rv1784	CAA17706	none	CONSERVED HYPOTHETICAL PROTEIN	923	0.6	14	0.16	0.29	7	5.17
65	Rv1800	CAE55427	PPE28	PPE FAMILY PROTEIN	623	0.6	10	0.15	0.26	2	4.22
66	Rv1803c	CAE55430	PE_PGRS32	PE/PGRS FAMILY PROTEIN	606	0.7	7	0.14	0.2	0	3.73
67	Rv1804c	CAA17725	none	CONSERVED HYPOTHETICAL PROTEIN	85	0.8	3	0.24	0.59	4	6.76
68	Rv1813c*	CAB09499	none	CONSERVED HYPOTHETICAL PROTEIN	110	0.8	1	0.15	0.6	4	8.96
69	Rv1815	CAB01480	none	CONSERVED HYPOTHETICAL PROTEIN	193	0.9	5	0.29	0.31	4	5.78
70	Rv1890c	CAB10062	none	HYPOTHETICAL PROTEIN	200	0.7	3	0.17	0.38	0	10.22
71	Rv1906c	CAB10031	none	CONSERVED HYPOTHETICAL PROTEIN	123	0.8	3	0.16	0.27	2	4.65
72	Rv1910c	CAB10054	none	PROBABLE EXPORTED PROTEIN	171	0.8	2	0.13	0.33	2	6.08
73	Rv1914c	CAB10028	none	HYPOTHETICAL PROTEIN	106	0.6	4	0.26	0.67	1	10.16
74	Rv1955	CAB06506	none	HYPOTHETICAL PROTEIN	140	0.5	2	0.11	0.67	0	11.06
75	Rv1966	CAE55443	mce3A	MCE-FAMILY PROTEIN MCE3A	376	0.6	4	0.12	0.59	4	5.13
76	Rv1967	CAA17840	mce3B	MCE-FAMILY PROTEIN MCE3B	313	0.6	7	0.23	0.67	2	9.24
77	Rv1968	CAA17841	mce3C	MCE-FAMILY PROTEIN MCE3C	383	0.6	7	0.15	0.66	0	7.74
78	Rv1969	CAA17842	mce3D	MCE-FAMILY PROTEIN MCE3D	391	1	7	0.16	0.44	2	5.02
79	Rv1971	CAA17844	mce3F	MCE-FAMILY PROTEIN MCE3F	412	0.8	4	0.13	0.48	4	4.7
80	Rv1973	CAA17846	none	POSSIBLE CONSERVED MCE ASSOCIATED MEMBRANE PROTEIN	136	0.6	4	0.24	0.65	0	6.52
81	Rv1974	CAA17847	none	PROBABLE CONSERVED MEMBRANE PROTEIN	88	0.6	3	0.23	0.65	2	8.19
82	Rv1975	CAA17848	none	CONSERVED HYPOTHETICAL PROTEIN	193	0.8	4	0.21	0.43	2	4.28
83	Rv2075c	CAA98220	none	Possible hypothetical exported or envelope protein	456	0.8	6	0.1	0.21	14	6.12
84	Rv2112c	CAB10703	none	CONSERVED HYPOTHETICAL PROTEIN	519	0.7	7	0.15	0.51	4	5.5
85	Rv2223c	CAA94646	none	Probable exported protease	485	0.9	8	0.16	0.36	10	4.54
86	Rv2224c	CAA94647	none	Probable exported protease	478	0.8	11	0.19	0.38	10	5
87	Rv2232	CAA94666.2	none	CONSERVED HYPOTHETICAL PROTEIN	266	0.5	7	0.21	0.33	1	6.17
88	Rv2251	CAA17288	none	POSSIBLE FLAVOPROTEIN	455	0.6	10	0.2	0.29	10	6.39
89	Rv2253	CAA17290	none	Possible secreted unknown protein	138	1	4	0.27	0.38	4	5.67
90	Rv2264c	CAA17301	none	conserved hypothetical proline rich protein	569	0.7	10	0.16	0.25	3	4.76
91	Rv2307c	CAB00991	none	CONSERVED HYPOTHETICAL PROTEIN	256	0.7	5	0.14	0.32	1	6.52
92	Rv2353c	CAE55477	PPE39	PPE FAMILY PROTEIN	322	0.7	6	0.25	0.27	0	3.64
93	Rv2356c	CAE55478	PPE40	PPE FAMILY PROTEIN LOW MOLECULAR WEIGHT ANTIGEN CFP2 (LOW MOLECULAR WEIGHT PROTEIN ANTIGEN 2)	585	0.7	7	0.14	0.56	0	3.6
94	Rv2376c	CAB08476	cfp2	(CFP-2)	145	0.9	2	0.14	0.4	0	4.94
95	Rv2396	CAE55484	PE_PGRS41	PE/PGRS FAMILY PROTEIN	327	0.6	7	0.17	0.2	1	3.89
96	Rv2525c	CAB06183	none	CONSERVED HYPOTHETICAL PROTEIN	204	0.8	5	0.22	0.61	1	9.42
97	Rv2565	CAB01050	none	CONSERVED HYPOTHETICAL PROTEIN	556	0.5	19	0.22	0.49	3	5.34
98	Rv2597	CAB01278	none	PROBABLE MEMBRANE PROTEIN	182	0.8	3	0.23	0.37	1	4.43
99	Rv2599	CAB01276	none	PROBABLE CONSERVED MEMBRANE PROTEIN	112	0.9	2	0.25	0.83	2	9.33
100	Rv2668	CAB02339	none	POSSIBLE EXPORTED ALANINE AND VALINE RICH PROTEIN	151	0.8	4	0.3	0.43	1	4.77
101	Rv2672	CAB02326	none	POSSIBLE SECRETED PROTEASE	492	0.8	8	0.14	0.5	10	4.65
102	Rv2741	CAE55516	PE_PGRS47	PE/PGRS FAMILY PROTEIN	494	0.7	7	0.16	0.36	0	4.32
103	Rv2799	CAB03647	none	PROBABLE MEMBRANE PROTEIN	181	0.7	4	0.2	0.33	4	5.12
104	Rv2840c*	CAB03669	none	CONSERVED HYPOTHETICAL PROTEIN	77	0.7	1	0.13	0.33	1	10.11
105	Rv2891	CAA98367	none	CONSERVED HYPOTHETICAL PROTEIN	212	0.9	3	0.14	0.64	3	11.12
106	Rv2956	CAB05420	none	CONSERVED HYPOTHETICAL PROTEIN	219	0.6	3	0.16	0.88	3	5.26
107	Rv2980	CAB05432	none	POSSIBLE CONSERVED SECRETED PROTEIN LOW MOLECULAR WEIGHT PROTEIN ANTIGEN 6	148	0.6	2	0.18	0.44	2	7.25
108	Rv3004*	CAA16089	cfp6	(CFP-6)	76	0.6	1	0.22	0.83	0	11.83
109	Rv3033	CAA16118	none	HYPOTHETICAL PROTEIN	155	0.9	4	0.25	0.53	4	4.82
110	Rv3036c	CAA16121	TB22.2	PROBABLE CONSERVED SECRETED PROTEIN TB22.2	203	0.8	3	0.16	0.24	3	4.78
111	Rv3096	CAB08388	none	CONSERVED HYPOTHETICAL PROTEIN	351	0.8	6	0.14	0.41	2	9.38
112	Rv3159c	CAE55561	PPE53	PPE FAMILY PROTEIN	554	0.8	13	0.17	0.28	0	3.93
113	Rv3196	CAA16661	none	CONSERVED HYPOTHETICAL PROTEIN	271	0.5	4	0.11	0.29	5	10.54
114	Rv3209	CAB08307	none	CONSERVED HYPOTHETICAL THREONIN AND PROLINE RICH PROTEIN	142	0.6	6	0.36	0.27	2	9.78
115	Rv3212	CAB08304	none	CONSERVED HYPOTHETICAL ALANINE VALINE RICH PROTEIN	369	0.6	14	0.31	0.46	4	8.28
116	Rv3224A*	CAE55570	none	CONSERVED HYPOTHETICAL PROTEIN	40	0.6	1	0.23	0.5	0	9.45
117	Rv3224B*	CAE55571	none	CONSERVED HYPOTHETICAL PROTEIN	64	0.6	1	0.19	0.71	1	11.16
118	Rv3267	CAB07086	none	CONSERVED HYPOTHETICAL PROTEIN (CPSA-RELATED PROTEIN)	473	0.9	11	0.2	0.51	2	4.67
119	Rv3333c	CAA17105	none	HYPOTHETICAL PROLINE RICH PROTEIN	248	0.5	2	0.11	0.36	2	6.95
120	Rv3351c	CAA15736	none	CONSERVED HYPOTHETICAL PROTEIN	247	0.5	4	0.19	0.65	3	9.96
121	Rv3388	CAE55593	PE_PGRS52	PE/PGRS FAMILY PROTEIN	697	0.8	8	0.16	0.29	0	4.68
122	Rv3484	CAB08707	cpsA	POSSIBLE CONSERVED PROTEIN CPSA	467	0.6	8	0.14	0.41	2	4.51
123	Rv3492c	CAB08715	none	CONSERVED HYPOTHETICAL MCE ASSOCIATED PROTEIN	138	0.6	4	0.25	0.41	1	9.3
124	Rv3494c	CAA17731	mce4F	MCE-FAMILY PROTEIN MCE4F	531	0.7	5	0.12	0.51	4	5.15
125	Rv3496c	CAA17733	mce4D	MCE-FAMILY PROTEIN MCE4D	429	0.8	11	0.25	0.47	2	4.31
126	Rv3497c	CAA17734	mce4C	MCE-FAMILY PROTEIN MCE4C	322	0.5	6	0.21	0.26	0	8.74
127	Rv3498c	CAA17735	mce4B	MCE-FAMILY PROTEIN MCE4B	314	0.6	7	0.2	0.56	2	7.5
128	Rv3499c	CAE55602	mce4A	MCE-FAMILY PROTEIN MCE4A	364	0.8	2	0.12	0.63	2	6.29
129	Rv3533c	CAE55610	PPE62	PPE FAMILY PROTEIN	546	0.8	7	0.12	0.2	0	5.51
130	Rv3547	CAB05059	none	CONSERVED HYPOTHETICAL PROTEIN	124	0.6	2	0.23	0.5	1	9.33
131	Rv3558	CAE55613	PPE64	PPE FAMILY PROTEIN	523	0.7	10	0.15	0.36	0	4.06
132	Rv3572	CAB07146	none	HYPOTHETICAL PROTEIN	151	0.9	3	0.25	0.33	4	4.55
133	Rv3587c	CAA17856	none	PROBABLE CONSERVED MEMBRANE PROTEIN	228	0.5	4	0.17	0.38	4	5.13
134	Rv3627c	CAB08829	none	CONSERVED HYPOTHETICAL PROTEIN	432	0.8	7	0.15	0.31	3	5.81
135	Rv3683	CAA18005	none	CONSERVED HYPOTHETICAL PROTEIN	292	0.6	5	0.16	0.4	4	8.95
136	Rv3693	CAA18015	none	POSSIBLE CONSERVED MEMBRANE PROTEIN	393	0.6	4	0.11	0.6	1	11.81
137	Rv3705c	CAA18027	none	CONSERVED HYPOTHETICAL PROTEIN	188	0.7	5	0.23	0.41	4	4.56
138	Rv3749c	CAA18071	none	CONSERVED HYPOTHETICAL PROTEIN	146	0.6	2	0.1	0.56	0	4.86
139	Rv3796	CAE55641	none	CONSERVED HYPOTHETICAL PROTEIN	331	0.5	8	0.23	0.38	1	5.83
140	Rv3802c	CAA17866	none	PROBABLE CONSERVED MEMBRANE PROTEIN	296	0.9	5	0.15	0.33	4	5.51
141	Rv3878	CAA17970	none	CONSERVED HYPOTHETICAL ALANINE RICH PROTEIN	247	0.5	4	0.14	0.22	0	3.87
142	Rv3908	CAB08093	none	CONSERVED HYPOTHETICAL PROTEIN	226	0.6	3	0.15	0.64	0	7.38
143	Rv3909	CAB08092	none	CONSERVED HYPOTHETICAL PROTEIN	762	0.6	14	0.18	0.61	2	5.66
144	Rv3916c*	CAA16229	none	CONSERVED HYPOTHETICAL PROTEIN	240	0.7	1	0.1	0.6	6	4.57

Other characteristics of outer membrane proteins

In addition to the secondary structure, other properties also appear to be characteristic of OMPs of gram-negative bacteria and might be exploited as predictive parameters for unknown OMPs. Most OMPs of gram-negative bacteria have a low isoelectric point (pI) due to the excess of acidic residues⁵⁸. The computed isoelectric points of the reference OMPs are all below 6.0 (Table S2) except for Omp32 of Delftia acidovorans which relies on basic amino acids for its function as a channel for organic acids⁵⁹. Importantly, both MspA of M. smegmatis and OmpA of M. tuberculosis have an acidic isoelectric point (Table S2) indicating that this criterion is a valuable parameter to identify OMPs in mycobacteria as well. 32 of the 144 predicted OMPs of M. tuberculosis have a pI below 6.0 and an amphiphilicity score higher than 0.39. They may represent especially interesting candidates and are highlighted in grey in Table 2.

We explored the possibility of recognizing OMPs if they contain a protein domain characteristic of known OMPs. For this purpose we used the Pfam database which recognizes protein domains based on sequence alignments^{29, 60}. None of the seven Pfam domains of OMPs of gram-negative bacteria were detected above the default cut-off of E-value 0.01 in proteins of M. tuberculosis. Significantly, both MspA of M. smegmatis and OmpA of M. tuberculosis do not have yet a Pfam domain. Hence, the current version of Pfam cannot be used to identify OMPs of M. tuberculosis.

Novel strategy for the experimental verification of the localization of proteins in bacterial outer membranes

To experimentally examine whether the predictions indeed identify outer membrane proteins, we chose two conserved hypothetical proteins, Rv1698 and Rv1973, that both have high scores for β-strand content and amphiphilicity (Table 2). Since it is often difficult to distinguish between inner and outer membrane proteins by subcellular fractionation experiments due to covalent linkages between the peptidoglycan, the arabinogalactan and the mycolic acids in mycobacteria and the concomitant mixing of membranes components, we resorted to a different strategy. The first step was to determine whether these proteins are associated with membranes. The next step was to demonstrate the surface accessibility of these proteins, because OMPs often expose extracellular loops. The confirmation of membrane association in combination with surface accessibility is experimental evidence for localization of a particular protein in the outer membrane because cytoplasmic, inner membrane and periplasmic proteins are inaccessible to reagents which are not able to penetrate the outer membrane. This holds true even if a protein is attached to cell wall components other than the outer membrane, e. g. the peptidoglycan, because it has to penetrate the outer membrane to become surface-accessible and is, thus, an OMP by definition. Examples of such proteins in gram-negative bacteria are TolC- and OmpA-like proteins²².

Membrane association of outer membrane proteins in M. bovis BCG and in M. tuberculosis

To examine the membrane association of the two selected putative OMPs, subcellular fractions of wild-type M. tuberculosis obtained as research materials from Colorado State University (CSU) were used: the cell membrane fraction which contains both inner and outer membrane proteins (MEM), the cell wall fraction (CW), the SDS-soluble cell wall proteins (SCWP), the water-soluble proteins (cytosol) and the culture filtrate (CFP). Since some proteins may not be expressed under standard growth conditions in M. tuberculosis and might therefore not be detectable in these fractions, we also expressed those proteins in M. bovis BCG. The cell envelope fraction (P100) and the fraction containing the water-soluble proteins were prepared by centrifuging a lysate of M. bovis BCG at 100,000xg. The ultracentrifugation pellet (P100) should contain the cell envelope including inner and outer membranes and the associated membrane proteins (total membrane). Then, the presence of three reference proteins in SDS extracts of these subcellular fractions was examined in Western blot experiments. OmpA of M. tuberculosis¹¹ and the porin MspA of M. smegmatis^{7, 9} were shown to be integral OMPs which are accessible at the cell surface and were chosen as reference outer membrane proteins. The green fluorescent protein GFP was used as a marker for cytoplasmic proteins. OmpA is naturally expressed in wild-type M. bovis BCG and was detected only in the total membrane fraction and not in the SN100 fraction (Fig. 2A). This demonstrated that OmpA is associated with membranes of M. bovis BCG. OmpA was also detected in the fraction of M. tuberculosis containing the SDS-soluble cell wall proteins (Fig. 2A). This is consistent with previous findings that OmpA is a cell wall protein of M. tuberculosis¹⁹. The detection of OmpA in the cell membrane fraction of M. tuberculosis (Fig. 2A) which should contain mainly inner membrane proteins is an example of the aforementioned difficulties in separating inner and outer membranes of mycobacteria in fractionation experiments.

Fig. 2. Membrane association of the outer membrane proteins marker proteins OmpA and MspA in M. bovis BCG and M. tuberculosis.

Subcellular fractions of M. tuberculosis (from CSU) and of M. bovis BCG were prepared and extracted with 2% SDS. A total of 150 μg protein per of M. tuberculosis fraction was analyzed in an SDS-polyacrylamide (10%) gel and blotted on a PVDF membrane for OmpA (A), MspA (B), and GFP (C). The same cell equivalents were loaded for each M. bovis BCG sample. The proteins were detected using specific antisera and an anti-rabbit antibody-horseradish peroxidase (HRP) conjugate.

A. Lanes are (from left to right): M: MagicMark™ XP Western standard (Invitrogen), lysed cells of M. bovis BCG, supernatant SN100 and pellet P100 of lysed cells of M. bovis BCG. M. tuberculosis (Mtb) cell membrane extracts (MEM) and SDS-soluble cell wall proteins (SCWP) of Mtb. Recombinant OmpA_His without the putative signal peptide (44 N-terminal amino acids) was purified from E. coli and was used as a reference. Note that the native proteins of M. tuberculosis and M. bovis BCG have a lower electrophoretic mobility than their recombinant protein because the N-terminus of OmpA is not cleaved¹⁴.

B. Lanes are (from left to right): M: MagicMark™ XP Western standard (Invitrogen), lysed cells of wild-type M. bovis BCG and of M. bovis BCG overexpressing mspA from the plasmid pMN013, supernatant SN100 and pellet P100 of lysed cells of M. bovis BCG overexpressing mspA from the plasmid pMN013. Recombinant MspA without the signal peptide was purified from E. coli and was used as a reference.

C. Lanes are (from left to right): M: MagicMark™ XP Western standard (Invitrogen), lysed cells of wild-type M. bovis BCG and of M. bovis BCG overexpressing mycgfp2+ from the plasmid pMN437, supernatant SN100 and pellet P100 of lysed cells of M. bovis BCG overexpressing mycgfp2+. The 60 kDa band in the supernatant fraction represents the GFP dimer, which did not dissociate because the samples were not boiled before loading of the gel.

MspA-like porins are not present in M. bovis BCG and M. tuberculosis. Therefore, we used a recombinant M. bovis BCG strain which expressed MspA from the plasmid pMN013 (Table 1). The Western blot clearly shows that the oligomeric form of MspA is completely associated with the total membrane fraction P100 (Fig. 2B) while only a tiny fraction of monomeric MspA was visible in the soluble fraction. This is consistent with previous experiments which demonstrate that MspA is a membrane protein^{6, 61}.

By contrast, GFP was almost only detected in the fraction containing water-soluble proteins after ultracentrifugation of lysed cells of a recombinant M. bovis BCG strain expressing gfp from plasmid pMN437 (Fig. 2C).

In conclusion these results demonstrated that the membrane fraction of M. bovis BCG was separated from the supernatant containing the soluble proteins by ultra-centrifugation, and that the outer membrane marker proteins OmpA and MspA are associated with the total membrane fraction. Hence, we showed that a simple ultracentrifugation step at 100,000xg is sufficient to separate membrane-bound proteins form soluble proteins in mycobacteria.

Rv1698 is associated with membranes

Rv1698 is annotated as a conserved hypothetical protein and has a β-strand score of 0.19 and an amphiphilicity score of 0.55 (Table 2). These are values similar to many other known OMPs (Table 2). Furthermore, Rv1698 had no cysteines and had a low pI which are known characteristics of OMPs of gram-negative bacteria. Further computational analysis confirmed the predictions that Rv1698 has a signal peptide from amino acids 1 through 25 (SignalP 3.0³²), and has no hydrophobic α-helix in addition to that in the putative signal peptide (TMHMM³⁴). To detect Rv1698 in subcellular fractions, we produced specific antibodies in rabbits and mice which were immunized with the recombinant Rv1698 protein purified from E. coli. A Western blot demonstrated that the antibody specifically recognized a ≈35 kDa protein in SDS extracts of the cell wall fraction (CW, SCWP) and in the total membrane fraction of wild-type M. tuberculosis H37Rv, but not in the cytosol fraction and in the culture filtrate (Fig. 3A). A protein of a similar size was recognized in samples of recombinant Rv1698 purified from E. coli (Fig. 3A) and in recombinant M. bovis BCG, which overexpresses Rv1698 (Fig. 3B). Further, ELISA showed a significant signal increase for cell lysates of M. bovis BCG strain when rv1698 was overexpressed from the plasmid pMN335 compared to wt (not shown). These results indicate that the antibodies specifically recognize the Rv1698 protein. Importantly, Rv1698 was associated only with the membrane fraction P100 of an M. bovis BCG strain overexpressing rv1698, but not with the supernatant SN100 containing water-soluble proteins. These results clearly show that Rv1698 is a membrane protein.

Fig. 3. Expression and membrane association of Rv1698 in mycobacteria.

A. Subcellular localization of Rv1698 in M. tuberculosis. Subcellular fractions (obtained from CSU) were extracted with 2% SDS. A total of 50 μg protein of each fraction was analyzed in an SDS-polyacrylamide (10%) gel. The samples were blotted on a PVDF membrane and detected with an Rv1698-specific antibody. M: MagicMark™ XP Western standard (Invitrogen); SCWP: SDS-soluble cell wall proteins. Recombinant Rv1698_His without the putative signal peptide (25 N-terminal amino acids) was purified from E. coli and was used as a reference.

B. Expression and membrane association of Rv1698 in M. bovis BCG. The supernatant SN100 and the pellet P100 of lysed cells of M. bovis BCG overexpressing rv1698 from the plasmid pMN335 were prepared by ultracentrifugation at 100,000xg. These samples were extracted with 2% SDS and the same cell equivalents were analyzed in an SDS-polyacrylamide (10%) gel. The gel was blotted on a PVDF membrane and detected with an anti-Rv1698 serum. The Rv1698 monomer is marked with a star. M: MagicMark™ XP Western standard (Invitrogen).

Rv1973 is associated with membranes

To further validate the bioinformatic predictions, we chose to determine the subcellular localization of Rv1973 as another putative OMP with no homologs of known function. This protein has high scores for both β-strand content (0.24) and amphiphilicity (0.65) (Table 2). It is annotated as a “possible conserved Mce associated membrane protein”. A detailed computational analysis confirmed the predictions that Rv1973 has a signal peptide from amino acids 1 through 18 (SignalP 3.0³²), and has no hydrophobic α-helix in addition to that in the putative signal peptide (TMHMM³⁴). To detect Rv1973 in subcellular fractions, we produced specific antibodies in rabbits which were immunized with the recombinant Rv1973protein purified from E. coli. A Western blot demonstrated that the antiserum recognized the purified Rv1973_His protein of ≈15 kDa (Fig. 4). This apparent molecular mass of monomeric Rv1973 is identical to the theoretical mass for Rv1973 after cleavage of the predicted signal peptide. No protein was recognized in wild-type M. bovis BCG, which does not contain the rv1973 gene (Fig. 4). By contrast, a 15 kDa protein was detected in lysates of a recombinant M. bovis BCG strain overexpressing rv1973 from the plasmid pMN339 (Table 1). This demonstrated that the antiserum specifically recognized Rv1973 in M. bovis BCG. Then, subcellular fractions of an M. bovis BCG strain overexpressing rv1973 from the plasmid pMN339 were examined using this antiserum. Rv1973 was associated with the total membrane fraction P100 but not with the supernatant SN100 containing water-soluble proteins. No signal was obtained for the subcellular fractions of wild-type M. tuberculosis. This indicates that rv1973 is not expressed in wt M. tuberculosis under the growth concentrations used for preparation of these fractions to levels which are detectable in this Western blot experiment.

Fig. 4. Membrane association of Rv1973 in M. bovis BCG and M. tuberculosis.

Subcellular fractions of M. tuberculosis (from CSU) and of M. bovis BCG were extracted with 2% SDS. A total of 150 μg protein per fraction of M. tuberculosis was analyzed in an SDS-polyacrylamide (10%) gel. The same cell equivalents were loaded for each M. bovis BCG sample. The samples were blotted on a PVDF membrane and detected with an anti-Rv1973 serum. Lanes are (from left to right): M: MagicMark™ XP Western standard (Invitrogen), lysed cells of M. bovis BCG, supernatant SN100 and pellet P100 of lysed cells of M. bovis BCG overexpressing rv1973 from the plasmid pMN339, Mtb cell membrane extracts (MEM). Recombinant Rv1973_His without the putative signal peptide (lacking the first 18 N-terminal amino acids) was purified from E. coli and used as a reference.

Whole cell ELISA experiments to demonstrate surface accessibility of marker outer membrane proteins

It was shown previously that the porin MspA could be detected by antibodies on the surface of M. smegmatis cells in enzyme-linked immunosorbent assays (ELISA)⁷. Therefore, we examined whether this method was also useful in detecting OMPs in M. bovis BCG. The ELISA with whole cells of wild-type M. bovis BCG and a recombinant strain of M. bovis BCG overexpressing the native ompA gene showed that more than 80% of OmpA was associated with the total membrane fraction P100 (Fig. S1A) in agreement with the Western blot analysis (Fig. 2A). Significantly, the signal was higher for whole cells of M. bovis BCG overexpressing OmpA compared to wt M. bovis BCG. This demonstrated that OmpA is surface accessible in M. bovis BCG. These results are consistent with the localization of OmpA in the outer membrane of M. tuberculosis as shown recently¹⁴. Similarily, more than 90% of MspA was associated with the total membrane fraction P100 (Fig. S2B) consistent with the Western blot analysis. However, MspA was not detected in whole cells of the recombinant M. bovis BCG strain in contrast to M. smegmatis for unknown reasons. It can be concluded that whole cell ELISA can be used to detect some but not all OMPs.

Rv1698 and Rv1973 proteins are surface-accessible in M. bovis BCG

As an alternative method to demonstrate surface accessibility of proteins, we employed protease accessibility experiments as described previously for a mycobacterial surface protein⁴⁶. Whole cells of wt M. bovis BCG were incubated with and without proteinase K. The cytoplasmic GFP and the periplasmic PhoA_HA proteins were clearly protected from proteinase K in M. bovis BCG cells (Fig. 5A and 5B). This was not due to the resistance to digestion as shown by the complete degradation of PhoA_HA in cell lysates (Fig. 5B). The finding that PhoA is not surface-accessible is consistent with our previous observation that the activity of PhoA depends on the presence of porins in the outer membrane of M. smegmatis⁴⁵ and the localization of PhoA in the periplasm of Gram-negative bacteria⁶². These results also demonstrated that the outer membrane of M. bovis BCG as a permeability barrier to proteinase K is still intact under these experimental conditions. By contrast, the outer membrane marker OmpA is significantly degraded by proteinase K in whole cells of M. bovis BCG (Fig. 5C). Importantly, both Rv1698 and Rv1973 were completely degraded by proteinase K in recombinant M. bovis BCG strains in contrast to the control samples without proteinase K (Fig. 5D and 5E). Thus, we have shown that both Rv1698 and Rv1973 are membrane proteins and are detectable on the cell surface. This demonstrates that Rv1698 and Rv1973 are indeed outer membrane proteins of M. tuberculosis.

Fig. 5. Surface accessibility of OmpA, Rv1698 and Rv1973 in M. bovis BCG.

Proteinase K accessibility experiments were performed using wt M. bovis BCG (OmpA), and M. bovis BCG strains carrying the plasmids pMN437 (p_smyc-mycgfp2+), pML970 (p_smyc-phoA_HA), pMN335 (p_imyc-rv1698), and pMN339 (p_imyc-rv1973). The strains were incubated with (+) or without (-) proteinase K at 37°C for 30 min. Extracts of whole cells with 1% SDS and cell lysates were separated on an SDS-polyacrylamide (10%) gel. Immunoblots were performed using anti-GFP (A), anti-HA (B), anti-OmpA (C), anti-Rv1698 (D) and anti-Rv1973 (E) antibodies. The MagicMark™ XP western standard (lanes M) was used as reference. Note the apparent molecular mass of PhoA_HA is significantly larger than the theoretical molecular mass of 52.5 kDa for the mature protein lacking the signal peptide. This is probably due to acylation of PhoA as shown previously⁸⁹.

DISCUSSION

Our aim was to identify integral OMPs of M. tuberculosis, which do not have any homology to OMPs in gram-negative bacteria except for OmpA¹¹. The porin MspA of M. smegmatis provided the proof of principle that integral mycobacterial OMPs share similar structural features with OMPs of gram-negative bacteria: The presence of a signal peptide, a β-barrel structure, and the absence of hydrophobic α-helices in the mature protein⁹. These features were exploited in a straightforward bioinformatic analysis consisting of multiple steps as depicted in Fig. 1A.

Prediction of exported M. tuberculosis proteins

SignalP 3.0 identified 587 proteins of M. tuberculosis with a signal peptide (Table S2) which, as in other bacteria, targets proteins to the general export system SecYEG for translocation across the cytoplasmic membrane⁴⁹. This is a relatively large number in comparison to 452 and approximately 300 exported proteins predicted by SignalP for the bacterial model organisms E. coli (www.cf.ac.uk/biosi/staff/ehrmann/tools/ecce/ecce.htm) and B. subtilis⁶³, respectively, and probably reflects the complexity of the mycobacterial cell envelope. It should be noted that this difference is not caused by different genome sizes which are very similar for all three species ^64-66. The proteins of M. tuberculosis predicted to be exported include 260 proteins of unknown function and 87 out of 168 proteins of the PE- and PPE protein families. Two-dimensional gel electrophoresis and MS identified 257 secreted proteins in the culture filtrate of M. tuberculosis⁶⁷. 121 of these secreted proteins are included in our list of exported proteins (Table S2).

Prediction of inner membrane proteins of M. tuberculosis

Our list of exported proteins of M. tuberculosis contains 134 proteins with a hydrophobic α-helix in addition to that in the signal peptide as predicted by TMHMM 2.0. All of these proteins are in the list of 787 possible transmembrane proteins provided by the PEDANT database (pedant.gsf.de) with the exception of Rv0956 and Rv2267. Manual re-analysis confirmed that both mature proteins indeed contain a transmembrane α-helix. These results also showed that the vast majority of inner membrane proteins of M. tuberculosis do not have a signal peptide. This is consistent with the findings for other bacteria that most IMPs are targeted to the inner membrane by the signal peptide recognition particle SRP and a non-cleavable signal anchor helix, which is more hydrophobic than the signal peptide⁶⁸.

Prediction of outer membrane proteins of M. tuberculosis

It is obvious that the Jnet algorithm underestimates the β-strand contents of OMPs. For example the β-strand contents of MspA and OmpF as derived from their three-dimensional structures exceed 60%^{9, 53}, but were predicted to be 27% and 23%, respectively, by Jnet (Table 1). This appears to affect all proteins similarly because the reference OMPs are still among the proteins with the highest β-strand content (Fig. 1). Thus, it is concluded that the Jnet-derived β-strand percentage is valid as a relative parameter for ranking purposes. However, some of the reference proteins have a very low β-strand content (Fig 1B) and thus our threshold for this parameter is not very discriminative. By contrast, the β-strand amphiphilicity offers a better discriminating power since the reference proteins have on average a higher value than the average Mtb proteins (Fig. 1B). Out of the 154 eliminated proteins, 139 lacked sufficiently amphiphilic β-strands and 69 did not meet the threshold for β-strand content. 54 of these proteins did not meet both criteria.

In contrast to all other predictors of bacterial OMPs^{23, 25, 55, 69}, the algorithm in our approach was not trained with a specific set of known transmembrane β-barrel proteins. This avoided any bias resulting from over-fitting to a data set of OMPs of gram-negative bacteria, which, except for basic structural parameters, is not appropriate for identifying OMPs of M. tuberculosis.

Three different predictors of β-barrel proteins trained to detect OMPs of gram-negative bacteria were previously used to identify putative OMPs of M. tuberculosis²⁵. Parameters which were used to distinguish OMPs from other proteins included a C-terminal sequence motif comprising phenylalanine at the last position in their prediction algorithm, which is typical for porins of gram-negative bacteria²⁶. However, none of the currently identified integral OMPs of mycobacteria (MspA, OmpA, Rv1698 and Rv1973) contain this C-terminal phenylalanine or sequence motif²⁵. Due to the inappropriate choice of predictive parameters and the use of algorithms that were probably over-fitted to an inappropriate training set, it is not surprising that only 21 proteins out of 144 proteins in our list of putative OMPs were also predicted by Pajon et al. Their list of predicted OMPs includes a large number of obviously false positives such as cytoplasmic proteins (kinase Rv0014c (MT0017), the DNA-binding protein HsdS (Rv2761c, MT2831), the transcriptional regulator CopG (Rv1398c, MT1442)), inner membrane proteins such as the sugar permease UgpA (Rv2835c, MT2901), the efflux protein EfpA (Rv2846c, MT2912) and the ammonium transporter AmtB (Rv2920c, MT2988) and lipoproteins such as LpqN (Rv0583c, MT0611) and LpqG (Rv3623, MT3725), which are not transmembrane β-barrel proteins but are anchored by a fatty acid residue attached to an N-terminal cysteine of the processed protein⁷⁰. It should be noted that most of the cell wall proteins as identified by subcellular fractionation and mass spectroscopy²⁰ are enzymes with locations other than the OM. These examples highlight the crucial importance of combining biological knowledge with bioinformatic methods to compile a useful list of putative OMPs.

However, despite considerable efforts to exclude proteins with subcellular localizations other than the OM, it is almost certain that our list of 144 putative OMPs (Table 2) also contains false positives such as periplasmic or secreted proteins. Indeed, 39 out of the 144 putative OMPs were classified as secreted proteins based on a proteomic analysis of proteins in the culture filtrate of M. tuberculosis⁶⁷. Careful subcellular localization experiments are required to determine which of these proteins are truly secreted and which might be present in the supernatant due to cell lysis^{71, 72} or shedding of membrane vesicles as was observed for gram-negative bacteria^{73, 74}. Further, it is apparent that our approach will miss the following classes of OMPs if they exist in mycobacteria: (i) OMPs in which the OM domain represents only a minor part of a protein with large or multiple extracellular or periplasmic domains, (ii) OMPs that do not contain a classical signal peptide detected by SignalP such as OmpA, which has been shown to be an OMP¹⁴ and fulfills all other criteria defined above as characteristic of OMPs such as high β-strand content, high amphiphilicity and low pI (Table 1), and (iii) outer membrane lipoproteins that use additional fatty acid residues as OM anchors such as OprM of Pseudomonas aeruginosa⁷⁵.

The mycobacterial cell entry proteins

The mce1A gene of M. tuberculosis was identified because its expression enabled Escherichia coli to enter epithelial cells⁷⁶. This gene is part of an operon comprising yrbEA, yrbEB, five mce genes (mceA, mceB, mceC, mceD, mceF) and a gene encoding a lipoprotein. Four copies of this operon are present in the M. tuberculosis genome (mce1-4)⁷⁷. Interest in these operons has been sparked by findings that mce deletion mutants affected the virulence of M. tuberculosis in mice^78-81. Our analysis predicts all Mce proteins as putative OMPs (Table 2). The only exception is Mce2C (Rv0591) which did not meet the SignalP cutoff value and is, therefore, not in the list of exported proteins. A manual re-examination showed that Mce2C may have a signal peptide and has a secondary structure compatible with that of an OMP. Thus, it is predicted that all mce genes code for OMPs of M. tuberculosis. Based on the observations that the mce genes are in the same operon as genes encoding a YrbEA/YrbEB ABC transporter we propose that the Mce proteins form an OM complex which connects to the inner membrane ABC transporters. This hypothesis is currently under investigation in our laboratory.

The PE/PPE proteins

The list of putative OMPs of M. tuberculosis includes 2 PE, 9 PE-PGRS and 10 PPE proteins (Table 2). These proteins belong to large families comprising proteins named for conserved proline and glutamate residues near their N termini^{82, 83}. In addition to these motifs, the family members share homologous N-terminal domains of approximately 110 amino acids for PE proteins and 180 amino acids for PPE proteins⁸³. It has been proposed that the PE/PPE proteins contribute to antigenic variation of M. tuberculosis. Consistent with this function is the accumulating evidence that at least some of these proteins are accessible at the cell surface^{46, 84, 85}. In this regard, it is striking that Rv1386 (PE15), which consists solely of a PE domain (70 amino acids without the predicted signal peptide), has high scores for both β-strand content and amphiphilicity (Table 2). This is consistent with the recent finding that the PE domain might be a cell wall anchor⁸⁶. However, the vast majority of the PE/PGRS proteins are not predicted as OMPs. This might be a problem of the algorithm which may not be able to detect secondary structural elements in these very unusual sequences. Alternatively, the PE/PGRS proteins may have different functions and subcellular localizations. The formation of protein complexes between paired PE/PPE proteins⁸⁷ is another finding which attests to the functional diversity of PE proteins.

Experimental evidence for the localization of proteins in mycobacterial outer membranes

In this study, we have developed a new experimental approach to detect mycobacterial OMPs. A single 100,000xg centrifugation of lysed cells of M. bovis BCG is sufficient to separate membrane proteins including OmpA and MspA from soluble proteins such as GFP. Surface accessibility in whole cells of M. bovis BCG was demonstrated for OmpA by both ELISA⁷ and protease experiments⁴⁶. This demonstrated unequivocally that OmpA is anchored in the outer membrane of M. bovis BCG consistent with its localization in the cell wall fraction of M. tuberculosis (Fig. 1A) and its function as a pore-forming protein^{11, 14}. The combination of establishing the association with membranes and the surface accessibility provides an alternative method to demonstrate whether the protein of interest is an OMP. This method is considerably less laborious compared to the recently improved protocol for subcellular fractionation of mycobacteria that consists of two cell lysis steps and at least six centrifugations¹⁹. An additional advantage is that it avoids the problem of mixing of inner membrane and cell wall fractions which can render localization experiments useless for certain proteins²⁰. This was a problem in particular for M. bovis BCG where substantial amounts of the inner membrane marker protein NADH oxidase were found in all subcellular fractions¹⁹. An obvious limitation of our approach is that it cannot be applied to OMPs which do not have surface-exposed loops or whose surface-exposed loops are not accessible. It appears that protease accessibility experiments are superior to antibody-based methods to detect surface proteins such as immunogold staining or whole cell ELISA experiments. This may be due to the unspecificity of cleavage by proteinase K which is not restricted to a particular position in the loops of OMPs in contrast to antibodies which need to bind to epitopes consisting of 5 to 10 amino acids⁸⁸.

Conclusions

The OMPs of mycobacteria remain largely unidentified. We have developed a new bioinformatic approach to predict OMPs of M. tuberculosis and shown the usefulness of this approach by demonstrating that two proteins of unknown functions are indeed OMPs of M. tuberculosis as predicted. This tripled the number of known OMPs of M. tuberculosis. Significantly, these studies indicated that M. tuberculosis likely has many OMPs with β-barrel structure and provided an alternative experimental approach how the localization of a protein in mycobacterial outer membranes could be demonstrated. Obviously, it is a challenging endeavour to identify the functions of the two newly discovered OMPs and of other yet unknown OMPs. However, our findings pave the way towards the identification of the set of OMPs of M. tuberculosis and other mycobacteria.

Abbreviations

OM

outer membrane

OMP

outer membrane protein

IM

inner membrane

IMP

inner membrane protein

wt

wild-type