β-Barrel proteins tether the outer membrane in many Gram-negative bacteria.

Gram-negative bacteria have a cell envelope that comprises an outer membrane (OM), a peptidoglycan (PG) layer and an inner membrane (IM)1. The OM and PG are load-bearing, selectively permeable structures that are stabilized by cooperative interactions between IM and OM proteins2,3. In Escherichia coli, Braun's lipoprotein (Lpp) forms the only covalent tether between the OM and PG and is crucial for cell envelope stability4; however, most other Gram-negative bacteria lack Lpp so it has been assumed that alternative mechanisms of OM stabilization are present5. We used a glycoproteomic analysis of PG to show that β-barrel OM proteins are covalently attached to PG in several Gram-negative species, including Coxiella burnetii, Agrobacterium tumefaciens and Legionella pneumophila. In C. burnetii, we found that four different types of covalent attachments occur between OM proteins and PG, with tethering of the β-barrel OM protein BbpA becoming most abundant in the stationary phase and tethering of the lipoprotein LimB similar throughout the cell cycle. Using a genetic approach, we demonstrate that the cell cycle-dependent tethering of BbpA is partly dependent on a developmentally regulated L,D-transpeptidase (Ldt). We use our findings to propose a model of Gram-negative cell envelope stabilization that includes cell cycle control and an expanded role for Ldts in covalently attaching surface proteins to PG.

The tripartite cell envelope of Gram-negative bacteria is composed of an OM, a PG layer, and an IM[1]. The OM is decorated with membrane-spanning β-barrel proteins that regulate flux of nutrients and hydrophilic antibiotics into the bacterial cell[6,7]. PG lies beneath the OM and is an elastic, mesh-like polymer that protects bacteria from osmotic lysis. It is composed of repeating disaccharide-pentapeptide subunits called muropeptides that consist of β−1,4-linked N-acetylglucosamine and N-acetylmuramic acid sugars substituted by peptide stems containing L-and D-amino acids (Fig. 1a). In Enterobacteriaceae, Vibrionaceae, and Pseudomonaceae, the cell envelope is stabilized by multiple IM and OM components including the highly abundant β-barrel OM protein OmpA, Peptidoglycan-associated lipoprotein (Pal), and Lpp. Both OmpA and Pal are OM proteins that non-covalently interact with PG via conserved PG binding motifs at their C-termini[8-10]. Lpp is an abundant 5.8 kDa helical lipoprotein that forms the only physical tether between the OM and PG. The triacylated N-terminus is embedded in the OM and the C-terminal lysine is covalently attached to meso-diaminopimelic acid (mDAP) of the PG stem peptide by Ldts[4,5,11-14]. Lpp functions as a molecular caliper to maintain the periplasmic distance between the OM and IM. Disruption of this spacing perturbs assembly of periplasmic-spanning structures as well as transduction of stress signals from the extracellular environment into the bacterial cell[15,16]. Lpp therefore plays a critical role in cell envelope structure and function in E. coli and closely related bacteria. Interestingly, most Gram-negative bacteria lack lpp orthologs, and the mechanism underpinning cell envelope stabilization in these organisms is an open question. Decades ago, PG-associated proteins were reported in the environmental pathogens C. burnetii and L. pneumophila, hinting at molecular determinants of OM stabilization in Lpp-deficient bacteria[17-21]. These organisms, along with Lpp- Agrobacterium tumefaciens, replicate in eukaryotic host cells to cause disease but also express unique physiologies that allow them to persist in environmental reservoirs[22-24]. These features make them attractive systems to investigate alternative mechanisms of cell envelope stabilization. For example, C. burnetii alternates between a replicating, exponential phase large cell variant (LCV) and a non-replicating, stationary phase small cell variant (SCV) during intracellular growth. The SCV has an unusually thick cell envelope, exhibits extreme resistance to physical and chemical disruption, and is hypothesized to be responsible for the long-term environmental stability of the pathogen[24]. Here, we use an automated glycoproteomic analysis of PG to demonstrate that OM β-barrel proteins are covalently attached to PG of C. burnetii and other ecologically diverse Lpp-deficient proteobacteria. In C. burneti, this attachment stabilizes the OM and is developmentally programmed.

Fig. 1.

β-barrel OM proteins are covalently attached to PG and form a tight tether in C. burnetii.

a. Schematic of the canonical Gram-negative muropeptide structure consisting of β−1–4 linked N-acetylglucosamine-N-acetylmuramic acid sugars [GlcNAc-MurNAc (GM)] attached to pentapeptide stems composed of L-alanine (A), γ-D-glutamic acid (E), meso-diaminopimelic acid (m), D-alanine (A), and D-alanine (A). The GM-AEm tripeptide moiety is outlined b. β-barrel OM proteins covalently attached to PG were identified in several proteobacteria. The number of predicted Ldts in each organism is shown. c. Schematic of the four covalently attached peptides identified in PG harvested from C. burnetii. Peptides are attached to the mDAP residue of the GM-AEm tripeptide moiety (red). d. Molecular dynamics simulation of C. burnetti OM protein-tethered models. The distance in angstroms between the centers-of-mass of the PG and the phosphorus atoms of the inner leaflet of the OM was measured for three independent 400 ns simulations of BbpA, ala-BbpA, BbpB, LimB, and E. coli Lpp. Distances are shown as averages with standard deviations.

First, to identify elements that mediate OM-PG tethering, we harvested PG from 8 proteobacterial species grown to stationary phase and analyzed PG using an unbiased tandem mass spectrometry approach[25]. A general search of the PG MS/MS data using species-specific proteomic databases identified the signature lysyl-arginyl dipeptide from Lpp covalently attached to PG in E. coli[26]. No peptides of appreciable length were found covalently attached to PG in Campylobacter jejuni, Neisseria gonorrhoeae, Helicobacter pylori, or Myxococcus xanthus, whereas peptides from 8-stranded or 16-stranded β-barrel OM proteins were covalently attached to PG in C. burnetii, L. pneumophila, and A. tumefaciens. Six β-barrel OM proteins were attached to PG in A. tumefaciens. Up to three possible β-barrel OM proteins were attached to PG in L. pneumophila. Two β-barrel OM proteins and a unique lipoprotein (LimB[27]) were attached to PG in C. burnetii [(hereafter, Bbp for β-barrel protein; BbpA (CBU0307), BbpB (CBU0311)] (Fig. 1b, c, Extended Data Fig. S1, 2, 3, 4). With a few exceptions, these β-barrel proteins exhibit little sequence similarity; however, invariant N-terminal residues are observed in a species-specific manner at the predicted signal peptide cleavage site. L. pneumophila and C. burnetii β-barrel proteins have conserved N-terminal glycyl residues whereas A. tumefaciens β-barrel proteins have conserved alanyl residues. These residues are also predicted to localize to the periplasm and were subsequently identified as the tethering residues to PG (Extended Data Fig. S1, 5). Using a complementary and modified search approach, we also identified alternative sites of PG attachment on −1 alanyl residues preceding the signal peptide cleavage site of C. burnetii BbpA (hereafter ala-BbpA) and A. tumefaciens ATU1020 and/or ATU1021, thereby representing two distinct linkages to the same protein(s) (Fig. 1c, Extended Data Fig. S1, 2, 3). Our proteomics search also identified that the C. burnetii-specific lipoprotein LimB is covalently attached to the mDAP residue of PG via an internal lysine residue (Lys21). This PG attachment is presumed to occur via the ε-amino group of lysine, similar to Lpp (Fig. 1c, Extended Data Fig. S1, 2)[27,28]. Non-covalent interactions between OM proteins and PG may stabilize the cell envelope in bacteria where proteins were not found covalently attached to PG[8-10]. Alternatively, covalent interactions between lipoproteins or other OM proteins that have tryptic cleavage sites close to the PG attachment site may preclude their identification. Collectively, these studies identified multiple proteins covalently bound to PG in many Gram-negative proteobacteria that do not encode an lpp ortholog.

Extended Data Fig. 1

PG modifications identified on proteobacterial OM proteins

Proteins from A. tumefaciens, C. burnetii, and L. pneumophila are covalently attached to PG. Proteins identified in MS/MS data that contained a PG modification scoring higher than the first decoy and containing greater than 40% of the expected band y- ions were established as cut-off criteria. Predicted signal peptide cleavage sites of OM proteins covalently attached to PG from C. burnetii, L. pneumophila, and A. tumefaciens are shown. Amino acid numbering is based on the predicted signal peptide cleavage site. Peptide sequences that were found covalently attached to PG are bolded. Residuces with a PG tripeptide modification are also highlighted in red.

Extended Data Fig. 2

Automated spectrum assignment identifies PG modifications on C. burnetii OM proteins

Representative MS/MS spectra for C. burnetii BbpA, ala-BbpA, BbpB β-barrel proteins and the lipoprotein LimB, covalently attached to mDAP (m) residues of PG. Spectra are shown as annotated by Byonic with manual annotations corresponding to internal fragments in black. J[+72.0848] has been replaced by m to represent mDAP. The PG tripeptide (AEm) is highlighted in red in the HCD spectra showing covalent attachment of PG to BbpA, ala-BbpA, and BbpB. The LimB Lys21 residue with PG tripeptide modification is highlighted in red in the ETD spectra showing covalent attachment of PG to LimB.

Extended Data Fig. 3

Automated spectrum assignment identifies PG modifications on β-barrel proteins from A. tumefaciens

Representative MS/MS spectra for A. tumefaciens β-barrel proteins covalently attached to mDAP (m) residues of PG. Residues with PG tripeptide modification are highlighted in red. Spectra are shown as annotated by Byonic using an unbiased search approach.

Extended Data Fig. 4

Automated spectrum assignment identifies PG modifications on β-barrel proteins from L. pneumophila

Representative MS/MS spectra for L. pneumophila Major Outer Membrane Protein (MOMP) covalently attached to mDAP (m) residues of PG. Residues with PG tripeptide modification are highlighted in red. Spectra are shown as annotated by Byonic using an unbiased search approach.

Extended Data Fig. 5

Structural topology of C. burnetii BbpA and BbpB

Predicted periplasmic, transmembrane, and extracellular domains of BbpA and BbpB using PRED-TMBB. BbpA and BbpB are depicted with periplasmic N-terminal regions. Similar topology is predicted in β-barrel proteins from A. tumefaciens and L. pneumophila that are covalently attached to PG.

Computational modeling and molecular dynamics simulation (MD) can be used to model molecular interactions and structural dynamics within the bacterial cell envelope[29]. In C. burnetii, the OM and PG layers are difficult to resolve by cryo-EM due to the pleomorphic nature of the bacterium and the unusually thick cell envelope of the SCV[30]. Therefore, we generated atomistic models of OM-PG tethering mediated by BbpA, BbpB, or LimB and used 400-ns MD simulations to estimate the distances between the OM and PG[2,29]. Interestingly, the OM-PG distance for all C. burnetii β-barrel and LimB tethers was consistently smaller than the OM-PG distance for E. coli Lpp (Fig. 1d, Extended Data Fig. S6, 7). These data suggest that β-barrel tethering confers unique transenvelope architecture and that critical periplasmic-spanning protein complexes, such as type IV secretion systems and OM signaling complexes of C. burnetii, have co-evolved with the architecture of the β-barrel tethered envelope[15,16].

Extended Data Fig. 6

Extended Data Figure S6. Structural modeling of C. burnetii OM proteins

a. The protein sequences of BbpA, BbpB, and LimB are shown without N-terminal signal peptides. The disordered N-terminal domains of BbpA and BbpB were excluded from structural models but were included in subsequent molecular dynamics simulations (blue). Residues bound to PG in MD simulations are highlighted in red. b. The mature LimB was modeled as a random coil lacking any appreciable secondary structure. The three acyl tails at the N-terminus are shown in grey as is Lys21. c. Contact map predicted from coevolution analysis of BbpA. Contacts are shown in shades of blue (darker blue = higher probability) and contacts from E. coli OmpA (PDB entry 1QJP) are shown in gray. d. Predicted contacts mapped onto the final Rosetta model of BbpA. Contacts are color coded by Cα-Cα distance: green (<5 Å), yellow (5–10 Å), red (>10 Å).

Extended Data Fig. 7

β-barrels form a tight tether between the OM and PG

Structural model of BbpA (red) in the C. burnetii OM. Structures of the cell envelope are colored as follows: inner leaflet of OM (grey), lipid A of LPS (orange), core oligosaccharides (yellow), glycan chains of PG (blue) peptide stems of PG (green). A similar model was generated for BbpB. b. Molecular dynamics simulation of C. burnetti OM-PG protein-tethered models. The distance in angstroms between the phosphorus atoms of the inner leaflet of the OM and PG layers was measured for three runs for BbpA, ala-BbpA, BbpB, LimB, and Lpp from E. coli. The solid lines are running averages. Distances measured in angstroms for run 1, run 2, and run 3 are shown.

C. burnetii, A. tumefaciens, and L. pneumophila encode numerous Ldts (≥10) (Extended Data Fig. S8)[31,32]. In E. coli, six Ldts carry out distinct yet partially redundant roles during cell wall remodeling: two enzymes catalyze the formation of non-canonical L,D cross-linking of PG and four fulfill an auxiliary role in covalently attaching Lpp to PG[13,14,33]. Several genes belonging to the Ldt family are highly up-regulated in C. burnetii upon transition to stationary phase[31]. PG-bound BbpA is considerably more abundant during stationary phase while PG-bound LimB is present at similar levels during both exponential and stationary phase (Fig. 2a, b). We therefore sought to determine if the C. burnetii stationary phase-specific Ldts serve a similar functional role in covalently attaching β-barrel OM proteins to PG. To this end, three stationary phase ldts were deleted (Extended Data Fig. S8) and strains characterized. A higher molecular weight PG-bound isoform of BbpA was identified in all strains except the Δldt2 mutant (Fig. 3a). Consistent with these results, extracted ion chromatograms (XIC) representative of PG-bound BbpA and PG-bound BbpB, but not ala-BbpA, were also identified in PG from each strain except the Δldt2 mutant (Fig. 3b, Extended Data Fig. S9). Pronounced OM vesiculation was observed in the Δldt2 mutant, a phenotype analogous to the E. coli Δlpp mutant (Fig. 3c, d)[34]. Additionally, the mutant displayed a delayed lag phase, suggesting diminished OM tethering negatively impacts metabolic fitness and the onset of replication (Extended Data Fig. S10)[35]. Increased vesiculation was also observed in the Δldt3 mutant without a significant decrease in OM-tethering by BbpA and BbpB (Extended Data Fig. S9). However, this does not exclude the possibility that Ldt3 plays a role in OM stabilization independent of these proteins. Collectively, these data indicate that ldt2 is the gene involved in covalent attachment of BbpA and BbpB to PG, which in turn stabilizes the OM during stationary phase. A model depicting these results is shown in Fig. 4. The conserved N-terminal glycyl-glycyl-prolyl-aspartyl motif of BbpA and BbpB is likely important for acceptor substrate recognition by Ldt2. However, the presence of alanyl-linked BbpA (ala-BbpA) in the Δldt2 mutant suggests an enzyme with acceptor substrate specificity different from Ldt2 cross-links BbpA to PG via the −1 alanyl residue preceding the signal peptide cleavage site. C. burnetii transitions from a typical exponential phase Gram-negative cell into a hardy spore-like form during stationary phase[30,31]. Thus, increased β-barrel tethering by Ldts during stationary phase is likely critical for enhanced stability of C. burnetii and potentially other bacteria that employ this mechanism of OM stabilization.

Extended Data Fig. 8

Predicted or annotated Ldts in Proteobacteria used in this study

Predicted or annotated Ldts in E. coli (lpp+), C. burnetii, L. pneumophila, A. tumefaciens, M. xanthus, N. gonorrhoeae, H. pylori, and C. jejuni. The latter seven organisms lack lpp homologs. The C. burnetii ldt genes that were successfully inactivated in this study are denoted with an asterisk. C. burnetii ldts upregulated during stationary phase are in bold. The following ldts were previously annotated as enhanced entry genes (enh); cbu0053, cbu0318, cbu1122, cbu1138, and cbu1394.

Fig. 2.

β-barrel tethering in C. burnetii is abundant in stationary phase.

a. Immunoblots of whole cell lysate from exponential phase (large cell variant) and stationary phase (small cell variant) C. burnetii, lysozyme-digested or undigested, and probed with BbpA (top) or LimB (bottom) specific antibodies. Molecular weight markers are shown to left of the blot. b. Quantification of immunoblots from panel a. The plot depicts means ± standard error of the mean of lysozyme mobilized BbpA or LimB signal expressed as a percentage of total BbpA or LimB signal of n=3 independent experiments. Statistical significance was calculated using an unpaired two-sided Student’s t test, **P = 0.000016; NS, not significant.

Fig. 3.

The L,D transpeptidase ldt2 is required for covalent attachment of BbpA and BbpB to PG.

a. Immunoblots of stationary phase whole cell lysate from wild-type (WT), Δldt, and complement strains, lysozyme-digested or undigested, and probed with BbpA (top) or LimB (bottom) specific antibodies. Data are representative of n=3 independent experiments. Molecular weight markers are shown to left of the blot. b. Extracted Ion Chromatograms (XIC) of precursor masses (m/z) corresponding to PG-BbpA (AEmGGPDYVPAPS, m/z = 906.909, z = 2), PG-ala-BbpA (AEmAGGPDYVPAPS, m/z = 942.427, z = 2), PG-BbpB (AEmGGPDIPM, m/z = 769.843, z = 2) identified in PG from WT, Δldt2, and Δldt2 complement strains. c. WT and Δldt deletion mutants were grown in ACCM-D for 14 days and imaged by cryo-EM. OM vesiculation was quantified by imaging ≥50 bacterial cells per strain and expressed as the percentage of bacteria with vesicles attached to the OM. d. Ultrastructure of WT and the Δldt2 deletion mutant imaged by cryo-EM. Arrows denote OM vesiculation.

Extended Data Fig. 9

The L,D transpeptidase ldt2 is required for covalent attachment of BbpA and BbpB to PG

XIC analysis was performed on PG that was extracted from WT and Δldt mutant strains and analyzed by MS/MS. XIC’s of precursor masses (m/z) corresponding to PG-bound BbpA (AEmGGPDYVPAPS, m/z = 906.909, z = 2), ala-BbpA (AEmAGGPDYVPAPS, m/z = 942.427, z = 2), and BbpB (AEmGGPDIPM, m/z = 769.843, z = 2) are shown.

Extended Data Fig. 10

Onset of replication is delayed in the Δldt2 deletion mutant

Growth of wild-type (WT) C. burnetii and a Δldt2 mutant strain in ACCM-D was assessed using qPCR to quantitate genome equivalents (GE) during a 21-day incubation. The results are expressed as the means from n=3 independent experiments. Error bars indicate the standard error of the mean, and asterisks indicate a statistically significant difference from WT C. burnetii. **P at 3, 6 and 9 days post-inoculation = 0.0046, 0.0027, and 0.0001, respectively. Statistical significance was calculated using a two-sided Student’s t test.

Fig. 4.

Schematic showing Ldt2-dependent and -independent covalent attachments of OM proteins to C. burnetii PG.

The cell-cycle independent PG-bound isoform of LimB was identified in all ldt knockout strains, indicating PG-bound LimB is formed independent of ld1, ldt2, ldt3 (Fig. 3a). Approximately 40% of total LimB is covalently attached to PG throughout the cell cycle (Fig. 2a, b). LimB is a Coxiella-specific lipoprotein with a cysteine and histidine rich C-terminus that binds metal cations[27]. It is hypothesized to serve as a surface receptor for metal acquisition in Coxiella. Our data demonstrate LimB also serves a structural role in the C. burnetii cell envelope by tethering the lipoprotein to the PG layer. How covalent attachment of LimB to PG impacts its metal binding activity is unknown. The site of PG attachment (Lys21) is distinct from the metal binding C-terminal domain (His26-His45), so both PG and metal binding could occur concurrently. Alternatively, metal-binding could be specific to the fraction that remains unbound to PG.

Our findings reveal that β-barrel OM proteins function to tether the OM in selected proteobacteria. β-barrel OM proteins are among the most abundant, variable, and immunogenic proteins found on the bacterial surface[7]. They modulate bacterial physiology, host-pathogen interactions, and transport small molecules across the OM, including hydrophilic antibiotics such as β-lactams and fluoroquinolones[36]. Moreover, bacteria that alternate between host and terrestrial environments and survive extreme fluctuations in pH, temperature, and osmolarity, are hypothesized to have multiple OM tethers[5]. Here, we report that multiple β-barrel OM proteins are covalently attached to PG during stationary phase in various environmental pathogens. The identification of Ldt-mediated attachment of β-barrel proteins to PG in C. burnetii serves to broaden our understanding of the diversity of substrates recognized by these important antibiotic targets. Carbapenems inhibit Ldts, so determining the catalytic and cell-cycle specificities of Ldts in Gram-negative pathogens will have important implications for the treatment of bacterial infections[33,37-39]. The increased abundance of β-barrel tethering in C. burnetii during stationary phase points to a model in which a structural determinant has a role in the formation of growth-arrested proteobacterial cells.

The physiological importance of β-barrel tethering in these environmental bacteria remains unknown. But it is tempting to speculate that this mechanism contributes to cell-envelope stabilization during growth-arrest and long-term survival of these organisms in the environment. In fact, most of earth’s microbial biomass exists in a growth-arrested state, yet very little is known about cell envelope structure, maintenance, and modification during this physiological condition[40]. We hope that these findings will usher in further research into Gram-negative cell envelope structure and physiology in a more ecologically diverse assemblage of bacteria.

Methods

Bacterial strains and growth conditions.

C. burnetii Nine Mile phase II (clone 4, RSA439) was used in this study. C. burnetii Δldt1, Δldt2, Δldt3, and complement strains, were created using previously reported methods[41]. Exponential phase large cell variants (LCVs) and stationary phase small cell variants (SCVs) of C. burnetii were generated by culture in ACCM-D for 6 and 14 days, respectively[42]. C. burnetii peptidoglycan was purified from 3.5 liters of log phase culture (LCV, ~5 ×107 genome equivalents/mL) or 500 mL of stationary phase culture (SCV, ~1 ×109 genome equivalents/mL) and digested with mutanolysin to solubilize muropeptides as previously described[31]. Muropeptides were resuspended in a final volume of 200 μL.

Culture conditions used to propagate the remaining bacterial strains used in this study are described below. L. pneumophila strain JR32 was inoculated into AYE broth from a 2-day heavy patch and incubated at 37°C with shaking until stationary phase was reached. A. tumefaciens C58 was grown to stationary phase in AB minimal media at pH 5.5[43]. C. jejuni NCTC 11168 was grown in Muller-Hinton broth, supplemented with 20 mM L-serine and 10 μg/ml vancomycin, under standard microaerobic conditions at 42°C in 10% v/v O2, 5% v/v CO2, 85% v/v N2 for approximately 18 hours to reach early stationary phase. M. xanthus DK1622 was grown in CTT broth (10 g casitone, 10 mM Tris pH 8.0, 8 mM MgSO4, 1 mM K2HPO4 pH 8.0 per liter). Cultures were incubated at 32°C with shaking (200 rpm) and harvested at logarithmic phase. H. pylori 26695 was grown in BHI broth (Oxoid) with 10% fetal bovine serum (Eurobio) and addition of 1x antibiotic cocktail [500x antibiotic cocktail stock contains, polymyxin B - 155 mg/L, vancomycin - 6.25 g/L, trimethoprim - 3.125 g/L, amphotericin B - 1.25 g/L, DMSO - 2.5% (v/v)]. Cultures were incubated at 37°C in 6% O2, 10% CO2, 84% N2 until they reached late exponential phase (OD = 0.75) or coccoid phase (72 hours). N. gonorrhoeae MS11 was first streaked on chocolate agar (E & O Laboratories Limited, Scotland) and incubated overnight at 37°C in 5% CO2. Isolated colonies were used to inoculate 30 mL BHI broth supplemented with 0.5% (w/v) yeast extract (Oxoid, UK) and 10 mM NaHCO3 and the cultures incubated overnight at 37°C with shaking (180 rpm). The culture was used to inoculate 500 mL of BHI broth which was cultured overnight until stationary phase was reached. Peptidoglycan extraction and purification were done as previously described[31].

LC-MS/MS data acquisition.

For the quantification of PG-bound proteins from C. burnetii ldt knockout and complement strains, samples were analysed online using an Ultimate 3000 RSLC nano LC LC system (Dionex) coupled to an LTQ Orbitrap Elite Hybrid mass spectrometer (Thermo) equipped with an EasySpray ion source. Desalted PG samples (200 ng) were separated using an EasySpray C18 capillary column (150 μm × 150 mm, 2 μm particle size - PN ES806). PG-bound peptides were eluted using water with 0.1% (v/v) formic acid (buffer A) and 80% acetonitrile with 0.1% (v/v) formic acid (buffer B) at a flow rate of 1.5 μL/min with the following gradient: 3% buffer B for 5 minutes, 50% buffer B for 15 minutes, 95% buffer B for 5 minutes, then 3% buffer B for 5 minutes. The mass spectrometer operated in a standard data-independent acquisition mode controlled by Xcalibur 3.0 and LTQ tuneplus 2.7. The instrument was operated with a cycle of one MS (in Orbitrap) acquired at a resolution of 60,000 from an m/z range of 375 to 1600. The top 10 most abundant multiply charged ions (2+ and higher) were subjected to higher energy collisional dissociation (HCD) fragmentation (isolation window 2.50 m/z, normalised collision energy = 30, activation time 0.1 ms). An FTMS ACG target value of 1×106 and HCD target value of 50,000 were used. Dynamic exclusion was enabled with a repeat duration of 30 seconds, an exclusion list of 500, and an exclusion duration of 45 seconds. A lock mass using dodecamethylcyclohexasiloxane (m/z = 445.12003) was enabled for all experiments.

For identification of proteins covalently attached to PG, data were collected using an Orbitrap-Fusion Tribrid mass spectrometer (Thermo-Electron) equipped with an EASY-Spray Ion Source and an Easy-nLC 1000 liquid chromatography system. Nano-LC was conducted using PepMap C18 columns (the analytical column: particle size 2 μm, length 25 cm, inner diameter 75 μm, the trap column: particle size 3 μm, length 2 cm, inner diameter 75 μm), operating at a 300 nL/min flow rate with mobile phases A: water, 0.1% (v/v) formic acid and B: 80%/20% (v/v) acetonitrile /water, 0.1% formic acid. The analytical column temperature was set at 50°C. After a 5 μL sample injection, MS/MS data were acquired during a 75 min step gradient: 0 to 40% B for 64 min, 40 to 100% B for 6 min, held at 100% B for 5 min. MS was operated in standard data dependent acquisition mode. First, the Orbitrap FTMS was used to acquire a full MS scan of all ions with an m/z range from 200–1400 at a resolution of 120,000 using dodecamethylcyclohexasiloxane (m/z = 445.12003) as an internal calibrant. The top five most intense precursor ions were subjected to both electron transfer dissociation (ETD) and HCD with precursors isolated by the quadrupole mass filter and scanned by the Orbitrap at a resolution of 60,000. For HCD, the precursor was excited at the collision energy at 25%. For ETD, the charge-dependent reaction time was utilized.

MS/MS data analysis.

Protein Metrics Byos version 3.5–4.0 was used to identify and compute extracted ion chromatograms (XIC) for PG-bound proteins. Identifications were made using Byonic module version 3.5.3 and XIC’s were computed using Byologic version 3.5.4. A FASTA database containing the peptide sequences for β-barrel proteins from C. burnetii BbpA (CBU0307), ala-BbpA (ala-CBU0307), BbpB (CBU0311) was assembled with the PG tripeptide monomer (GM-AEm) concatenated to the N-terminus of each mature sequence (signal peptide removed). A fixed modification of +72.0848 was added to J (an unused letter which Byonic sets to 100.0000 Da) to represent mDAP (mass = 172.0848). A variable modification of 480.1961 (GlcNAc-MurNAc) was enabled on the peptide N-terminus. Search tolerances of 10 ppm for precursor mass and 20 ppm for fragment mass tolerance were set. The manual score cut off was set to 0 and a protein false discovery rate of 1%. An m/z window of 20 ppm, Apex search window of 2 minutes, and XIC area window of 1 minute were set for XIC extraction using Byologic. Peptides without a PG modification were discarded. All remaining identified peptides where then manually inspected to ensure XIC integration was correctly computed.

Identification of alternative sites of PG attachment.

Unbiased searches were performed using Byonic module 3.6.0 against bacterial proteomes with variable modifications of 480.196 (GlcNAc-MurNAc) + 372.1645 (AE-mDAP) or 438.185 (GlcN-MurNAc) + 372.1645 (AE-mDAP) permitted once per peptide on any residue within the peptide. Searches were performed using non-specific cleavage parameters. Precursor mass tolerance was set at 8 ppm and fragment mass tolerance was set to 20 ppm for ETD and HCD fragmentation. A score cut-off of 200 was applied and peptides containing modifications scoring above this were examined and spectra were extracted.

Immunoblots.

Bacterial pellets were harvested from exponential and stationary phase bacteria and resuspended in phosphate buffered saline (PBS) containing 0.5% Triton-X. Lysozyme was added at 100 μg/ml and cell suspensions were incubated at 37°C for 2 hours. Lysozyme-treated samples were separated by SDS-PAGE using 12% Criterion XT Bis-Tris precast gels (Bio-Rad) and transferred to an Immobilon-P polyvinylidene difluoride membrane (MilliporeSigma). Membranes were blocked in Tris-buffered saline containing 0.1% Tween 20 (TBST) and 5% milk, then incubated with α-BbpA or α-LimB antibody. The α-BbpB antibody exhibits cross-reactivity with BbpA; therefore, immunoblot results were not included. Immunoblots were washed with TBST and incubated with SuperSignal West Femto maximum sensitivity substrate (ThermoFisher). Immunoblots were analyzed on an Azure c600 imaging system (Azure Biosystems) and immunoreactive bands were quantitated using Image Studio Lite (Li-Cor Biotechnology). The α-BbpA monoclonal antibody (K64) and α-LimB polyclonal antibody were kind gifts from James Samuel (Texas A & M University) and Micheal Minnick (University of Montana), respectively.

Cryo-EM.

Stationary phase C. burnetii samples were frozen on a glow discharged Quantifoil™ R 2/2 grid using a Leica EM GP plunge freezer (Leica Microsystems, Wetzlar, Germany) set to a relative humidity of 100%. Grids were mounted in an autogrid assembly and cryo 2D images collected using an FEI Titan Krios 300kV FEG-TEM and an FEI Falcon II CMOS direct electron detection camera. Low dose images of 13e/A2 were captured at 6 μm defocus with a nominal magnification of 29 K which corresponds to a pixel size of 2.87 Å using a 70 μm objective aperture. OM vesiculation was quantified by imaging ≥50 bacterial cells per strain and is expressed as the percentage of bacteria with vesicles clearly attached to the bacterial OM.

Structural modeling.

Sequences used for structural modeling of BbpA (Q83EL2), BbpB (Q83EK8), and LimB (B5QSB9) are shown (Extended Data Fig. 6). For structural modeling of BbpA, HHpred from the Bioinformatics Toolkit was used to search for suitable structural templates[44,45]. The top hit (only 20% identity) was the 1.65 Å X-ray crystal structure of OmpA from E. coli (PDB entry 1QJP)[46]. Because of the low sequence identity, coevolutionary information was also used to facilitate structural modeling. Sequence homologs were searched and coevolution analysis was performed using the GREMLIN web server to identify alternative templates, and to generate residue-residue contact restraints for Rosetta structural modeling[47,48]. Briefly, GREMLIN used hhblits to perform an iterative hidden Markov model (HMM)-based search of the UniProt20 database for homologous sequences to BbpA[49,50]. The resulting multiple sequence alignment (MSA) was filtered to ensure that no two sequences were more than 90% identical and all sequences covered at least 75% of the query sequence. The MSA was further modified to remove columns that contained more than 25% gaps. The final MSA generated by GREMLIN included 3,376 homologous sequences to BbpA, which corresponds to an effective number (neff) of 2,440.6 sequences after filtering at an 80% sequence identity threshold. Accurate structural models can be generated by incorporating coevolutionary information provided that the quantity Nf is greater than 64 (Nf is defined as neff divided by the square root of the sequence length)[51]. For BbpA, this quantity was 175.7. The top X-ray structure identified using this approach was also OmpA from E. coli (PDB entry 1QJP). Thus, the coevolutionary and HHpred search approaches are in agreement. An initial threaded model was made on the basis of residue-residue contact predictions using the program map_align[51]. The threaded model was used in combination with coevolution-derived restraints (100,000 map_align models) and the final model was selected on the basis of the Rosetta score and overall cylindricity of the model.

For structural modeling of BbpB, a preliminary search with GREMLIN did not identify enough sequence homologs to enable a coevolution analysis. Thus, we used HHpred to search for suitable templates of BbpB to generate sequence alignments for threading. We then used RosettaCM to generate models with several of the top-scoring X-ray structures individually as templates[52]. Membrane span files were generated with Octopus, and membrane weights were applied during Rosetta modeling[53]. At least 5,000 models were generated for each template. The final BbpB model used for molecular dynamics simulation was selected from 50,000 models on the basis of its Rosetta energy and the positions of the extracellular loops[54]. This model was based on the 2.55 Å resolution X-ray crystal structure of the Neisserial surface protein A (NspA) from Neisseria meningitidis (PDB entry 1P4T, 19% identical to BbpB) as the template.

Molecular dynamics simulations.

To simulate the proteins in their native environment, a model of the C. burnetii Nine Mile phase II stationary phase, small cell variant outer membrane was constructed. The outer leaflet of the model is composed of 33 lipopolysaccharides, with lipid A modeled based on the chemical structure reported in[55] and core oligosaccharide added to lipid A based on the structure reported in[56]. The following distribution of lipids for the inner leaflet was chosen based on the phospholipid analysis reported in[57]: 27% phosphatidylethanolamine, 25% cardiolipin (PVCL2), 11% Lyso-phosphatidylethanolamine, 4% Lyso-phosphotidylglycerol, and 2% phosphotidylglycerol. For Lpp, an E. coli outer membrane model described previously was used[2]. A previously constructed model of the E. coli cell wall was used for all simulations. We note that this model includes only 4–3 cross-linked peptides, which varies slightly from the C. burnetii small cell variant that employs a mixture of 4–3 and 3–3 cross-linked peptides[2,31,58]. All systems were constructed using VMD[59]. For BbpA and BbpB, the N-terminal glycine was bonded to a free diaminopimelate residue in the cell wall. For BbpA, a third system was built in which an alanine was added to the N-terminus prior to cross-linking it to the cell wall. For LimB, Lys21 was covalently bonded to a diaminopimelate residue on the cell wall. After initial equilibration, each of these systems were simulated in triplicate for 400 ns, giving a total of 6 μs.

MD simulation parameters.

All simulations were carried out using NAMD[60]. A time step of 2 fs was used, with short-range non-bonded interactions evaluated every time step and long-range electrostatics evaluated every other time step using the particle-mesh Ewald method[61]. Short-range interactions were cut off at 12 Å with a force-based switching function starting at 10 Å. Because the cell wall was built in a strained state, as expected for living cells, the area was held constant at the equilibrated area of the OM-protein system[2]. A temperature of 310 K was maintained using Langevin dynamics with a damping constant of 1.0 ps−1; a pressure of 1 atm was maintained using a Langevin piston. The CHARMM36m force field for proteins and CHARMM36 force field for lipids were used[62,63].

Statistical analysis.

For all experiments, three technical replicates were used with n defined as the number of independent experiments performed. All data are presented with error bars defined as the mean ± standard error of the mean. Statistical analyses were performed using an unpaired two-sided Student’s t test to calculate P values with Prism software (GraphPad Software, Inc., La Jolla, CA). False-discovery rates were calculated based off of a q value of 1% using the Benjamini, Krieger, Yekutieli method.