Ttg2DPae contains a large hydrophobic cavity that binds four acyl tails
Sequence analysis indicates that Ttg2DPae (PA4453) is the soluble periplasmic SBP component of the ABC transporter encoded by the ttg2 operon and a member of the Pfam family MlaC (PF05494). Interestingly, the available 3D structures for the MlaC family from Ralstonia solanacearum (PDB entry 2QGU), P. putida (PDB entries 4FCZ and 5UWB) and E. coli (PDB entry 5UWA) were all solved in complex with a ligand in their hydrophobic pocket, except for one structure from E. coli (PDB entry 6GKI) where the protein was delipidated. Electron densities for the ligand were in all these structures compatible with phospholipid moieties, supporting their predicted role as phospholipid transporters. A sequence alignment shows that some of the residues thought to be involved in phospholipid binding in the R. solanacearum Ttg2D structure are conserved in the P. aeruginosa ortholog (Fig. S1). Remarkably, the electron densities for P. putida Ttg2D revealed the presence of two diacyl lipids in its pocket27.
To investigate ligand binding at the molecular level, we determined by molecular replacement the crystallographic structure of the Ttg2DPae mature protein (without the signal peptide, aa 23–215) produced in E. coli at 2.53 Å resolution (PDB entry code 6HSY) (Fig. 1A). The structure was refined to a final Rwork and Rfree of 20.9 and 24.9%, respectively, and good validation scores (Supplementary Table S1). All residues but the last three C-terminal ones (plus the C-terminal expression tag) could be modeled. Ttg2DPae adopts a mixed α + β fold with a highly twisted anti-parallel β-sheet formed by five strands and surrounded by eight α-helices. It exhibits a “decanter” shaped structure never described before for any other protein family (Fig. 1A). The structure presents a highly hydrophobic cavity between the β-sheet and the helices that spans the whole protein and has a volume of 2979 Å3 and a depth of ~ 25 Å (Fig. 1B). After the first refinement stage (AutoBuild), without any ligand added, clear density was visible inside the cavity that could correspond to four acyl chains (Fig. S2). We therefore modeled inside the cavity two PG(16:0/cy17:0) (Fig. 1A), as mass spectrometry (MS) experiments suggested that this lipid was one of the most abundant among those found to bind Ttg2DPae when expressed in E. coli (see below). Real-space correlation coefficients of 0.9 for the lipids indicate a good fit to the 2mFo - DFc electron density. The four acyl tails are deeply inserted into the hydrophobic cavity, while the polar head groups are exposed to the solvent and make only few contacts with the protein (Fig. 1, A and C). This lack of specific recognition of the head group could explain why Ttg2DPae is able to bind different types of phospholipids. The presence of two diacyl lipids suggests that the protein could also be able to bind one tetra-acyl lipid, such as diphosphatidylglycerol (cardiolipin).
To investigate the mechanism of entry and release of the two lipids in the cavity of Ttg2DPae, we performed a normal mode analysis (NMA). NMA may be used to model the internal collective motions of a protein, relevant to ligand biding and function in general, typically described by a few low-frequency modes37. Figure 1D shows the collective motions along mode 7, the first non-trivial mode (modes 1 to 6 account for translational and rotational motions of the protein as a whole). Rather than "en bloc" relative motions of sub-domains, all secondary structures of the protein appear to move in a concerted manner, helix α4 and the core of the β-sheet being more rigid. This breathing-like motion increases in a concerted manner the volume of the cavity and its mouth area, and may allow the lipids to enter into or exit from the cavity. Inspection of the next 10 lowest-frequency normal modes shows similar concerted motions. The recent MalC structure with no lipid bound (PDB entry 6GKI)32 shows similar collective motions along all modes, with similar amplitudes, indicating that the cavity could open and close in the absence of lipid. The normal modes can be also used to compute atomic mean-square displacements, which can be in turn related to B-factors38. The NMA-derived and the observed (crystallographic) B-factors are closely correlated except in regions 75–95 and 180–200, which are involved in crystal contacts, and region 105–120, where the electron density is weaker (Fig. S3). This suggests that the normal modes provide a realistic description of the protein's flexibility.
The Ttg2D/MlaC fold: a new two-domain architecture and SBP structural cluster
All MlaC homologs of known structure have a highly superposable "decanter" shaped configuration (Fig. 1A), previously described as an "extended" NTF2 fold27 but never assigned a distinct structural classification. Thus, while the P. putida structure (PDB entry 4FCZ) appears in the CATH database as a single domain protein and unique structure of superfamily 3.10.450.710, belonging to fold topology 3.10.450 (Nuclear Transport Factor 2; Chain: A), the R. solanacearum structure (2QGU) is described as a two-domain structure with domains belonging to CATH-superfamilies 3.10.450.50 (NTF2-like) and 1.10.10.640. The latter superfamily belongs to the all-alpha fold topology 1.10.10 (Arc Repressor Mutant, subunit A) and has 2QGU as unique structure. To clarify the structural classification of Ttg2D/MlaC proteins we run a DALI search of the putative second domain (D2 in Fig. S1) of Ttg2DPae against the whole PDB. The results showed very good superposition with the small alpha domain of several AAA + proteins, the best match being with 6UKS chain C39 (Fig. S4). AAA (ATPases Associated with diverse cellular Activities) domains are formed by a large N-terminal domain adopting a Rossmann fold and a small C-terminal domain forming an alpha-helical bundle. They tend to adopt homo-hexameric ring complexes and hydrolyze ATP to perform activities that involve protein remodeling. The helical domain plays an important role in coupling the conformational changes resulting from ATP hydrolysis to the neighbor monomer within the AAA-ring and to the underlying protease ring40 and has a specific classification (other than AAA) both in the PFAM (PF17862) and CATH databases (1.10.8.60). Interestingly, the Ttg2C ortholog MlaD, which in E.coli interacts with MlaC32, forms also a homo-hexameric ring. We therefore conclude that currently known Ttg2D/MlaC structures combine two structural domains in an architecture not seen yet in other proteins, a NTF2-like domain and an AAA helical-bundle domain. In Ttg2DPae, the first domain is formed by two non-contiguous sequence segments: D1S1 (PDB residues 23–68), with three alpha-helices, and D1S2 (113–169), with five beta-strands. The second domain, with five helices, is also split in two non-contiguous sequence segments: D2S1 (69–112) and D2S2 (170–212) (Fig. 1A and S1).
Classical SBPs linked to ABC transporters are structurally similar and composed of two globular domains formed by discontinuous segments41, adopting either a Rossmann fold (CATH: 3.40.50) or a very similar CATH: 3.40.190. The 501 SBP structures available in the PDB have been classified in 7 clusters and several subclusters42. We superposed the Ttg2DPae structure to a representative of each of these subclusters. RMSD values, number of aligned residues and structural superpositions are shown in Fig. S5. The longest match aligns 47 residues with an RMSD of 3.97 Å (2PRS chain A, a 284-residue member of cluster A-I), while the best RMSD is 1.41 Å with 23 residues aligned (3MQ4 chain A, a 481-residue member of cluster B-V). These results clearly confirm that Ttg2D/MlaC family proteins do not belong to any previously known SBP structural cluster.
Evolution of sequence and structural diversity of the MlaC family
A structural alignment of MlaC family proteins with known 3D structures (Fig. 2A) reveals that, despite sequence identities ranging from 63% for the P. putida protein to as low as 17% for the E. coli one, the RMSDs of the structural alignments are very low, ranging from 1.6 to 3.1 Å (188 to 185 Cα), respectively (Table S2). Clearly, secondary structure elements are highly or strictly conserved among all four proteins, despite substantial amino-acid variation (Fig. 2A and S1). However, the four proteins split into two groups: P. aeruginosa and P. putida Ttg2D have a hydrophobic cavity of 2979 − 2337 Å3 and can bind two diacyl lipids, while the R. solanacearum and E. coli proteins have a half-size cavity of 1444 − 1332 Å3 and bind only one diacyl lipid (Table S2). Surprisingly, although the different number of ligands had been already noticed when the structure of Ttg2D from P. putida was solved, cavity differences were never analyzed. Figure 1B illustrates the cavity difference between P. aeruginosa and R. solanacearum Ttg2D. The volume differences correlate with the different number of residues forming the cavities, from 55 down to 31 (Table S2). However, these residues, which are spread along the whole protein sequence (Fig. S1), are largely conserved in terms of position and, in most cases, in terms of identity or similarity, with a few substitutions such as V147/L, or V163/I or M directly affecting the volume. Some side-chain reorientations, like Y105, and small secondary structure displacements, like strands β3 and β4 or helix α6 shifted by ~ 2 Å (Fig. 2A and S6), also modulate the volume. Taken together these changes are, nevertheless, not sufficient to explain how the cavity volume can double. Helix α8 seems to be the main responsible for the difference between a two and a one diacyl-phospholipid cavity, not only because the helix is longer in the first case (Fig. S1), but also because it adopts a different conformation. Indeed, for the second group (R. solanacearum and E. coli, one diacyl lipid), this helix has a straight conformation, covers the α6 helix (Fig. 2A and S6) and does not participate in the cavity (Fig. 1B and S1), while in the first group (P. aeruginosa and P. putida, two diacyl lipids), the α8 helix is bent towards and over the α7 helix and greatly enlarges the cavity (with additional residues from β4 and β5 strands). This bend occurs at residue G195 with an angle of 40º and 64º in Ttg2D proteins from P. aeruginosa and P. putida, respectively (Fig. 2A). The helix of the first protein has an additional bend of 43º at K202. Glycine has a poor helix-forming propensity43 and tends to disrupt helices because of its high conformational flexibility. On the other hand, the Phe197 and Gln196 residues occupying the Gly195 position in the R. solanacearum and E. coli proteins, respectively (Fig. S1), have better helix-forming propensities and maintain the α8 helix straight. In addition, W196, exclusive of the pseudomonal structures, may also contribute to the influence of the α8 helix on the cavity's volume, since its bulky hydrophobic side chain, deeply inserted into a hydrophobic pocket on the concave side of the curvature, could stabilize the helix α8 bend (Fig. S6).
Components of the Mla system are broadly conserved in Gram-negative bacteria, except for the periplasmic MlaC that notoriously shows high inter-species sequence diversity (Fig. 2B). Interestingly, an alignment of 151 representative amino-acid sequences belonging to the MlaC family and identified across different Gram-negative species revealed that W196 is conserved not only in Pseudomonas species but also in a group of related sequences in other non-phylogenetically related gamma-proteobacteria (Fig. 2C). In this group of proteins that would hypothetically bind two diacyl phospholipids, other positions with distinct residues relative to other MlaC family members stand out, especially in two regions located between the central part and the C-terminal end of the protein (Fig. S6). Side-chain orientation and hydrophobicity of some residues in these regions could be also contributing to a tighter binding of the two diacyl phospholipids inside the ligand cavity. The presence of common protein sequence signatures in species that are not closely related indicates that horizontal gene transfer, mediated by recombination events between flanking conserved genes, could have contributed to MlaC family diversity.
Ttg2D Pae binds two diacyl glycerophospholipids and cardiolipin, representing a novel phospholipid trafficking mechanism among Gram-negative bacteria
Native MS was used to determine the lipids that bind specifically to recombinant Ttg2DPae produced in the cytoplasm of E. coli and the stoichiometry of the interaction in a cellular environment (Fig. 3). The native mass spectrum of Ttg2DPae shows two major charge states (m/z 2701, z = 9 and m/z 2430, z = 10) corresponding to a deconvoluted average mass of 24296 Da that matches the MW of the recombinant protein Ttg2DPae produced in E. coli without the first methionine residue plus bound ligands (Fig. 3A and S7). After isolation of the wide peak ion at m/z 2700 (z = 9) and gas-phase fragmentation with a transfer collision energy (CE) of 50 V, we detected the unbound protein (m/z 2536, z = 9) and a family of released phospholipids in the low mass range (Fig. 3B-C). Additional native MS experiments on isolated ion m/z 2700 (z = 9) using increasing transfer CE confirmed that at least a fraction of the bound Ttg2DPae population hosts two phospholipids, since lipid dissociation starts at a transfer CE of 35 V and goes through the single-bound species (Fig. S8). The results also show that a second fraction of the Ttg2D population hosts a single molecule, that remains bound up to a transfer CE of 50 V, which could correspond to cardiolipin as further MS experiments indicated. A tighter binding of cardiolipins could explain why these molecules were not detected in the gas-phase fragmentation experiments (Fig. 3B-C). We also note that at the transfer CE required for cardiolipin release these molecules are easily fragmented (see below).
The mass of [M + H] + ions at m/z 664.5, 704.5, 718.5 and 730.5 released from Ttg2DPae confirmed the identity of some ligands as phosphatidylethanolamines (PE) with different hydrocarbon chains (PE C30:0, PE C33:1, PE C34:1 and PE C35:2 respectively) (Fig. 3C). Subsequent direct MS analysis under denaturing conditions in both positive and negative ion modes, identified additional species of PE and of other phospholipid classes that are also components of the bacterial membrane, such as phosphatidylglycerol (PG), phosphatidylcholine (PC) and phosphatidylserine (PS) (Fig. 3D and S9). With these methods, PG C33:1, PG C34:1, PE C33:1 and PE C35:1 came out as most abundant (annotation of lipid species based on the most abundant fatty acids in E. coli is given in Table S3). The observed distribution of phospholipids bound to recombinant Ttg2DPae relates to their relative abundances in E. coli BL21 (the recombinant protein-expression host) as determined by lipidome analyses (Fig. S10), and it correlates well with the reported phospholipid composition of E. coli under comparable conditions44, 45.
LC-MS analysis under denaturing conditions in positive ion mode of the complexes produced in E. coli shows lipid species of m/z 690–800 and m/z 1400–1500 that could correspond to phospholipids and cardiolipins, respectively (Fig. S9D). In addition, direct MS analysis in negative ion mode indicates the presence of glycerophospholipids and shows peaks at m/z 1050–1160 that could correspond to E. coli cardiolipin species having lost one fatty acid (Fig. 3D-E). During the fragmentation processes of phospholipids in negative ion mode, upon low energy collisional activation, ions corresponding to the loss of fatty acids are the most abundant46. For example, when the most abundant cardiolipin species in E. coli, i.e. CL C68:2 with MW 1405.0 Da (Fig. S10), loses one of the fatty acids at position sn-2, it shows a prominent fragment ion at m/z 1147 or 1121 depending on the fatty acid species in that position (Fig. 3D-E).
In light of these results, we performed additional native MS experiments to investigate whether delipidated Ttg2DPae can bind cardiolipin in vitro. In Fig. 4, representative native mass spectra of a reaction mixture containing 26.25 µM delipidated protein and 245 µM cardiolipin CL(18:0)4 (1:9 molar ratio) in a buffered aqueous solution, are shown. The protein–cardiolipin complex and the unbound protein were detected around the 10 + and 9 + charge states (Figs. 4 and S11). Notably, the dissociation energy needed to release cardiolipin was higher than that required to release diacyl glycerophospholipids in the native complex (transfer CE of 60 V vs 35 V, see Figs. S11A and S8, respectively). This correlates with the previous observation that a "P + 2PL" population in the native spectra in Fig. S8 loss its cargo only at 60 V. At the transfer CE required for cardiolipin release this molecule is in fact easily fragmented (Fig. S11B).
Ttg2D Pae binds two diacyl glycerophospholipids or cardiolipin in the periplasm of P. aeruginosa
Given that the cytoplasm of E. coli is clearly not the natural environment of Ttg2DPae, we decided to produce and purify this protein directly from the periplasmic space of P. aeruginosa. For this purpose, protein Ttg2DPae tagged with six C-terminal histidine residues and containing its own N-terminal secretion signal was expressed in the genetic background of a P. aeruginosa PAO1 Δttg2D mutant using a new derivative of a broad-host-range cloning vector (Fig. 5A). Restoration of the colistin susceptibility phenotype after transformation (MIC of 0.25 µg/ml vs. 0.0625 µg/ml in the non-transformed mutant) was used as evidence that the protein being expressed in the mutant strain was functional. Under these conditions, the mature Ttg2DPae could be purified from the periplasm in sufficient amount and purity (Fig. 5B inset) to determine its phospholipid cargo by MS.
The native mass spectrum in Fig. 5B shows a charge-state distribution that corresponds to the ligand-bound mature Ttg2DPae complexes (MW of intact complex: 23621 Da; MW of His-tagged protein without the signal peptide: 22140 Da). The wide peak widths, besides poor desolvatation in aqueous buffer in the native MS instrumental conditions, suggests also the coexistence of multiple species of similar mass, supporting the presence of different classes of phospholipids bound to the protein. Gas phase dissociation mass spectra at varying transfer CE of isolated ions around m/z = 2625 (z = 9) indicated the presence of two populations at peaks m/z = 2626 (z = 9) and m/z = 2619 (z = 9), which are in agreement with the binding of two phospholipids and one cardiolipin, respectively (Fig. 5C). Further MS experiments under denaturing conditions in negative ion mode allowed the identification of two main phospholipid classes (Fig. S12 and Table S4). The obtained distribution is in agreement with relative abundances in P. aeruginosa PAO1 as determined by a lipidome analysis (Fig. S10), and it correlates well with the reported phospholipid composition in this species47.
The Ttg2 system provides P. aeruginosa with a mechanism of resistance to membrane-damaging agents
As expected, the P. aeruginosa Δttg2D mutant exhibited a debilitated outer membrane leading to increased susceptibility to several membrane damaging agents (Fig. 6), as demonstrated by the 1-N-phenylnaphthylamine (NPN) assay. Indeed, an enhancement in NPN uptake was observed in the mutant in the presence of the permeabilizer agents EDTA and colistin (Fig. 6, A and B). In line with this, the Δttg2D mutant is significantly more susceptible to the action of polymyxins (lipid-mediated uptake), but also of antibiotics that use both the lipid- and porin-mediated pathways to penetrate the cell, including fluoroquinolones, tetracyclines and chloramphenicol (Fig. 6C). With regard to polymyxin antibiotics, the ttg2D transposon insertion mutant was eight-fold more susceptible to colistin than the PAO1 wild-type, a colistin-susceptible reference strain (Table S5). In general, the mutation did not significantly affect the resistance phenotype displayed by the PAO1 strain to the beta-lactam antibiotics or aminoglycosides tested. The susceptibility phenotypes due to deletion of ttg2D could be fully or partially reverted by complementation with the cloned ttg2D gene or the full operon ttg2 in the replicative broad-range vector pBBR1MCS-5 (Fig. 6 and Table S5), confirming the link between the gene and the phenotypes. We have also confirmed that insertional mutations in each of the other components of the ttg2 operon (ttg2A, ttg2B, ttg2C) and vacJ (mlaA ortholog) lead to an increased susceptibility to antibiotics in the same way as for the Δttg2D mutant (Table S5). The Δttg2D mutant is also significantly susceptible to the toxic effect of the organic solvent xylene (Fig. 6D) and it is four-fold more susceptible to the chelating agent EDTA (MIC = 0.5 mM) than the parental wild-type PAO1. However, no difference was observed between the mutant and wild-type cells in their susceptibility to SDS, obtaining for both strains a MIC value of 0.8%. Finally, disruption of the ttg2D gene resulted in an approximately two-fold reduction in biofilm formation and greatly increased the activity of EDTA against P. aeruginosa biofilms at a subinhibitory concentration of 0.05 mM (Fig. 6E).
The Ttg2 system is associated to P. aeruginosa's intrinsic resistance to low antibiotic concentrations
The susceptibility of Ttg2-defective mutants to antibiotics was further studied in strains with different genetic backgrounds. To this end, the full ttg2 operon was mutated in the clinical MDR P. aeruginosa strains C17, PAER-10821 and LESB58, which had shown different patterns of resistance to several antibiotic classes, specifically, polymyxins, fluoroquinolones and tetracyclines (Table 1). In particular, PAER-10821 and LESB58 are P. aeruginosa strains with low-level resistance to colistin. The generation of mutants with disrupted gene functions in MDR bacteria is troublesome because the antibiotics commonly used in the laboratory are no longer useful for selection of gene knockouts. In addition, the loci mutated in this case is involved in a general mechanism of resistance to antimicrobial agents and mutant strains are therefore expected to be generally susceptible and thus potentially lost during the selection steps. For this reason we have adapted a mutagenesis system based on the homing endonuclease I-SceI48, 49 to construct targeted, non-polar, unmarked gene deletions in MDR P. aeruginosa strains (see material and methods, text S1 and Fig. S13 for details). With this modified mutagenesis strategy we have obtained and validated unmarked deletion mutants of the selected MDR strains lacking the full ttg2 operon (Fig. S13). Complemented strains were also obtained by transformation of mutant strains with a replicative plasmid containing the full ttg2 operon and its expression in the complemented clones was confirmed by RT-PCR (Fig. S13). All these strains were tested for their susceptibilities to different classes of antibiotics (Table 1).
Table 1
Antibiotic susceptibility profile of P. aeruginosa MDR strains lacking the full ttg2 operon.
Antibiotic | MIC† in µg/ml |
LESB58 | C17 | PAER-10821 |
WT | Δttg2 | WT | Δttg2 | WT | Δttg2 |
Polypeptides |
Colistin | 4 | 0.125* | 2 | 0.125* | 32 | 32 |
Fluoroquinolones |
Ciprofloxacin | 2 | 1 | 256 | 64* | 256 | 128 |
Levofloxacin | 8 | 2* | 256 | 256 | 256 | 128 |
Ofloxacin | 16 | 4* | > 32 | > 32 | > 32 | > 32 |
Norfloxacin | 8 | 4 | > 256 | > 256 | > 256 | 256 |
Tetracyclines |
Tetracycline | 16 | 8 | 32 | 16 | 32 | 8* |
Minocycline | 32 | 8* | 16 | 8 | 32 | 8* |
Tigecycline | 16 | 8 | 64 | 8* | 32 | 8* |
Chloramphenicol |
Chloramphenicol | 32 | 32 | 128 | 64 | 128 | 64 |
Sulfonamides |
Trimethoprim-sulphamethoxazole | 16 | 8 | > 64 | > 64 | > 64 | 64 |
Aminoglycosides |
Tobramycin | 8 | 2* | 64 | 128 | 128 | > 128 |
Amikacin | 64 | 64 | 8 | 32* | 32 | 32 |
Gentamicin | 32 | 16 | > 128 | > 128 | > 128 | > 128 |
Kanamycin | > 512 | > 512 | 256 | 512 | 512 | 512 |
Streptomycin | > 64 | > 64 | > 64 | > 64 | > 64 | > 64 |
Carbapenems (beta-lactam) |
Imipenem | 2 | 2 | 32 | 32 | 32 | 64 |
Meropenem | 2 | 2 | 32 | 16 | 16 | 16 |
Cephalosporins (beta-lactam) |
Ceftazidime | 256 | 256 | 64 | 128 | 16 | 32 |
Penicillins (beta-lactam) |
Piperacillin | 256 | 256 | 256 | > 256 | 128 | 256 |
Piperacillin-tazobactam | 128 | 128 | 256 | > 256 | 64 | 64 |
Ticarcillin | > 256 | > 256 | > 256 | 256 | 64 | 128 |
Ticarcillin-clavulanic acid | > 32 | > 32 | > 32 | > 32 | > 32 | > 32 |
† Minimum inhibitory concentration (MIC) determined by the broth microdilution method. MICs were confirmed by two or three independent replicates. MIC differences greater than 2-fold with respect to the corresponding wild type strain were considered significant (indicated with an asterisk). |
The three ttg2 mutants were significantly more sensitive (between 4- and 64-fold) than the corresponding wild-type bacteria to colistin, fluoroquinolones, and tetracycline analogues, but not to the other antibiotic classes (Table 1). The mutant susceptibility phenotypes could be reverted by providing an intact copy of the entire PAO1 ttg2 operon (PA4456-PA4452) in a replicative plasmid, except for colistin. The lack of complementation of the colistin susceptibility phenotype could be due to the effect of the antibiotic erythromycin (used as a selection marker for complemented strains) on the expression of global regulators that may influence colistin susceptibility50, 51 or to the overexpression of the ttg2 operon components (two- to eight-fold with respect to wild type, see Fig. S13) that may also affect the distribution of phospholipids in the OM. Surprisingly, the susceptibility to amikacin significantly decreased for the C17 mutant and an opposite effect was observed for the LESB58 mutant and tobramycin, suggesting a genetic-background component in the effect of the ttg2 mutation on the susceptibility to these antibiotics.