Overall architecture of LolA-PG and LolA-VC
Full length LolA from P. gingivalis (LolA-PG) and V. cholerae (LolA-VC), corresponding to the mature forms of the proteins, were expressed in E. coli. After protein purification they were crystallized, and data were collected. The structure of LolA-PG was solved using single anomalous dispersion (SAD) on selenomethionine (SeMet) labelled protein whereas the structure of LolA-VC was determined with molecular replacement using a model obtained from Alphafold22. The models were refined to 1.7 and 1.8 Å respectively. Data processing and refinement statistics are presented in Supplementary Table S1.
LolA from both P. gingivalis and V. cholerae consist of a large curved β-sheet comprising twelve β-strands (Fig. 2a,b). Eleven of the strands are antiparallel whereas the last strand, β12, runs parallel to β6 on the flank of the β-sheet. β12 is connected to β11, in the center of the sheet, via a 20 amino acid (aa) extended segment that runs on the convex side of the β-sheet.
The electron density for both LolA structures is generally good, but weaker in some loops, especially between strands β8 and β9 of LolA-VC which constitutes part of the opening of the binding cleft. The equivalent loop in LolA-PG is well ordered, probably due to stabilizing crystal contacts. Additional small differences are found in other loops or turns that connect the β-strands. The concave surface forms a mainly non-polar cleft which is shielded from solvent by two helices, α2 and α3 (a 310 helix). The N-terminal helix, α1, forms the floor of the cleft. The non-polar cleft can interact either with the periplasmic loop of LolC, when uncharged, or the acyl chains of a lipoprotein during protein transportation18,20. The helical segment that fills the hydrophobic cavity is comprised of 22 aa in both LolA-PG and LolA-VC but the different position of the helices results in different accessibility to the clefts. In LolA-PG the cleft is continuous and relatively open (36 Å3) as compared to the scattered cleft in Lol-VC which is occluded and only expose 0.15 Å3 (Fig. 2d,e).
Overall architecture of LolB-VC
The mature but non-acylated form of LolB from V. cholerae (LolB-VC), was overexpressed in E. coli, purified and crystallized. Diffraction data were collected, and the LolB-VC structure was determined using molecular replacement. The model was refined to 1.46 Å resolution (Supplementary Table S1). The fold of LolB-VC is similar to LolA, with a large curved β-sheet consisting of eleven β-strands that folds into a hydrophobic cavity filled with two short helices. In contrast to LolA, LolB-VC does not have the long-extended segment running on the convex side of the sheet nor a 12th β-strand (Fig. 2c). Furthermore, the helical segment that fills the hydrophobic cavity of LolB-VC is slightly longer, approximatively 30 aa, leaving 11 Å3 of the cleft accessible in this crystal structure (Fig. 2f).
Comparative structural analyses
LolA-VC is structurally very similar to other known LolA proteins, especially those from γ-proteobacteria. A DALI search23 using LolA-VC identified LolA from E. coli, Y. pestis and P. aeurginosa (PDB codes 1ua8, 4ki3 and 2w7q)16,17 as the closest structural relatives (Z-scores 25–22). When using LolA-PG as the search model, LolA from P. aeurginosa (PDB code 2w7q)16 was recognized as the structurally most similar (Z-score 19, sequence identity 13%). LolA from B. uniformis (LolA-BU) (PDB code 4mxt), which also originates from the bacteroidota phylum resulted in a Z-score of 18, albeit the sequence identity is higher, 24%. A pairwise structural alignment to compare LolA-VC and LolA-PG, using the structural alignment tool in the protein data bank (rcsb.org), resulted in an r.m.s.d. of 2.8 Å calculated on 176 aa and a sequence identity of 18%.
A sequence analysis was performed on the proteins described (Supplementary Fig. S1a). Within each phylum, proteobacteria (class γ-proteobacteria) and bacteroidota, the sequence similarity is high. When proteins from both phyla were analyzed together, only three identical aa are found: two phenylalanines (on β1 and β3 respectively) contributing to the hydrophobic environment in the binding cleft. The third conserved residue is a proline (Pro120 in LolA-PG and Pro105 in LolA-VC) that creates the bend between the α2 and α3 helices, that fill the hydrophobic cavity (Fig. 2a,b).
An Arg-Pro motif located on the turn between β2 and β3 is conserved in γ-proteobacterial LolA (Arg59, Pro60 in LolA-VC). There, the arginine side chain is facing the interior of the protein where it interacts with main chain carbonyls of α2 (Fig. 3a). Mutational studies in E. coli have shown that the arginine is important for lipoprotein transfer to LolB24. When the arginine was mutated to leucine the delivery of lipoprotein to LolB was negatively affected however the interaction between LolA (without bound lipoprotein) and LolB became stronger20. The recent structure of LolA-EC in complex with a triacylated peptide indeed conferred that the bound acyl chains and protein obtain different conformations when the arginine is mutated to leucine, and as a consequence the delivery of lipoprotein to LolB is altered20. The Arg-Pro motif is not present in LolA-PG or in LolA-BU. Instead, they both have a glycine located at the corresponding position (Gly75 in LolA-PG). The equivalent space of the arginine side chain is instead filled with the hydrophobic Leu41 in LolA-PG (Phe43 in LolA-BU). This is an interesting difference considering that a recipient LolB protein has not been identified in bacteroidota, hence the function of these LolA proteins must rely on other patterns of interaction (Fig. 3a,b). It should be added that when the sequence analysis was extended to include LolA also from other classes of proteobacteria it is established that the Arg-Pro motif is also found in β-proteobacteria (Neisseria meningitidis) but not ε-proteobacteria (Helicobacter pylori), hence the motif is not universal even among proteobacteria. Furthermore, there is a difference in the theoretical isoelectric point for the two proteins in this study, 4.8 for LolA-VC and 9.3 for LolA-PG calculated on the mature proteins. This may reflect differences in milieu and interaction partners. It is also interesting to speculate if LolA-PG and other bacteroidota LolA proteins really have evolved to bind diacylated substrates, as has been assumed since an lnt gene has not been identified. By comparing the Lol-PG binding cleft to LolA-EC (Fig. 3c) there are no obvious differences indicating that only two acyl chains should be accommodated by the bacteroidota protein. Indeed, it has been experimentally shown that the P. gingivalis protein PG1828 and the Bacteroides fragilis protein BF1333 are triacylated25,26, but there is to our knowledge no studies that describe if this is the most common lipidation in bacteroidota or not. As the sequence homology between the different phyla is very low it is possible that an enzyme that adds a third acyl chain, just like Lnt in γ-proteobacteria, does exist—but has not yet been identified. Hence, we hypothesize that some bacteroidota’s lipoproteins may be triacylated, just like in γ-proteobacteria.
LolB, is unlike LolA, itself a lipoprotein and is acylated at residue Cys27 and bound to the inner leaflet of the OM. There it receives lipoproteins from LolA and anchors them to the membrane. A Dali search using the LolB-VC structure unsurprisingly identified LolB from E. coli (LolB-EC) as the closest structural relative with a Z-score of 26 and an r.m.s.d of 1.4 Å with 31% sequence identity of 172 aligned residues17. Further down the list of other putative lipoprotein sorting proteins, most likely LolA proteins, are found, with Z-scores 12–13. For LolB-EC it has been shown that a leucine (Leu68) is important for its function to accept lipoprotein from LolA and target them to the membrane27, Indeed, LolB-VC exhibits a leucine (Leu63) in the same position, an indication of a conserved function. Interestingly, both LolB proteins have a proline positioned between helix α2 and α3 as previously described for LolA (Fig. 2c). An interesting difference between LolB-EC and LolB-VC is first of all the difference in theoretical pI, 8.7 for LolB-EC and 5.1 for LolB-VC. This may also be reflected by the more negative surface potential of the helices that fill the binding cavity of LolA-VC. Further, the main structural difference is that strands β9 and β10 each are two residues longer in LolB-VC (Supplementary Fig. S2).
The interaction between LolA and LolB
The transfer of acylated proteins between LolA and LolB is proposed to occur through a mouth-to-mouth interaction28 where the opening of their respective clefts dock and allow the acyl chains of the transported lipoprotein to slide from LolA to LolB (Fig. 1). It was originally described for the E. coli system that negatively charged residues predominantly reside on the LolA side, matched by positively charged residues on LolB29. An electrostatic surface representation of LolA-VC and LolB-VC indeed show that the residues lining the LolA-VC cleft is mainly negatively charged whereas LolB-VC has a ring of positively charged residues around the opening (Fig. 4a,b). We used isothermal titration calorimetry (ITC) to investigate if the recombinant proteins LolA-VC and LolB-VC can interact without the presence of lipoprotein as has been reported for E. coli. The measurements showed that the two proteins indeed interact, with a dissociation constant (Kd) measured to 30.2 µM (Table 1, Supplementary Fig. S3, Supplementary Table S2). This is similar to the Kd of 30.6 µM that previously was reported for the interaction between LolA and LolB of E. coli29. LolA-PG, for which a LolB partner has not been identified, showed no significant interaction to LolB-VC despite its predominantly negatively charged surface at the cleft opening (Fig. 4c, Supplementary Fig. S3, Table 1, Supplementary Table S2).
Interactions between LolA, LolB and polymyxin B
Polymyxins are cyclic cationic lipopeptide antibiotics (Fig. 5a) that through electrostatic attraction can interact with lipid A of the lipopolysaccharides. As a result, the OM is disrupted, and after passing through the periplasm, the IM can also be damaged30. In order to uncover the mechanism of polymyxin B transport from the OM to the IM, a molecular dynamics simulations study suggested that LolA and LolB are responsible for binding and transporting the molecule14. The idea was derived from the observation that polymyxin has a lipid tail that could theoretically bind LolA and LolB like the acyl chains of lipoproteins.
We investigated this hypothesis experimentally by ITC and could conclude that LolA from both V. cholerae and P. gingivalis bind polymyxin B. LolA-PG shows higher affinity for polymyxin with a Kd of 14 µM compared with 56 µM for LolA-VC. When such experiment was performed on LolB-VC, the results were not conclusive. For LolB the high Kd in millimolar range, in addition to solubility issues of polymyxin B, affected the reproducibility of the experiments and therefore the analysis (Table 1, Supplementary Table S2, Supplementary Fig. S4). Despite the relatively high affinity of LolA to polymyxin B, previous studies have shown that P. gingivalis is resistant to polymyxin B concentrations up to 200 µg/mL. In this regard a lipid A 4ʹ phosphatase (PGN_0524) plays a crucial role since it removes a phosphate group from lipid A and makes the surface less negatively charged and hence less prone to attract the cationic polymyxin31. In strains carrying this enzyme it is likely that very little polymyxin enters the periplasm where LolA is active. However, this opens the prospects for development of combinations of strong physical OM disruptors of P. gingivalis together with novel polymyxin-like compounds that can be used synergistically against P. gingivalis infections. This is of high importance since P. gingivalis also has increased minimum inhibitory concentration (MIC) values for antibiotics such as tetracyclines, macrolides, lincosamide and fluoroquinolones and there is a fear that resistance can be transferred from other bacteroidota and prevotella species32. Also V. cholerae exhibits different sensitivity to polymyxin depending on strain. Whereas some classical strains are very sensitive to the antibiotic (MIC 0.15 μg/mL), El Tor strains express three enzymes that add a glycine to Lipid A resulting in a bacterial surface less prone to bind polymyxin B and yielding a MIC of 100 μg/mL33.
Interactions between LolA, LolB and A22
It was discovered that lipoprotein trafficking in E. coli is disrupted by the small antibacterial compound MAC13243, or its hydrolysis product A2234 (Fig. 5b). In in vitro assays it was shown that these compounds induce release of lipoproteins from LolA20 whereas it is claimed that in vivo targets of these compounds are not LolA or LolB but instead the Mre system35. In this study we used ITC to clarify if A22 interacts with recombinant LolA or LolB, when they are not charged with lipoproteins. We found that LolA and LolB from V. cholerae have Kds in the millimolar range, that could not be measured with high confidence due to the low solubility of A22 (Supplementary Table S3, Fig. S5). On the contrary, LolA-PG has higher and more reproducible affinity for A22 and the Kd was measured to 680 µM. Hence, LolA and LolB from both γ-proteobacteria and LolA from bacteroidota bind A22 albeit with very low affinity, which is in line with previous studies.