Introduction

The Gram-negative outer membrane is instrumental for virulence, antimicrobial resistance, and immune evasion. The outer membrane is an asymmetric lipid bilayer consisting of phospholipid on the inner leaflet, lipopolysaccharide (LPS) on the outer leaflet, integral outer membrane proteins (OMPs) and associated lipoproteins. Biogenesis of the outer membrane is completed by several multi-component proteinaceous machines. Central to this is the β-barrel assembly machine (Bam) complex, the essential protein machinery responsible for folding and insertion of OMPs into the outer membrane, which then act as transporters, porins, receptors, virulence factors, adhesins and enzymes [1, 2]. All OMPs are synthesised in the cytoplasm before being trafficked through the inner membrane to the periplasmic space by the Sec machinery. OMPs are prone to aggregation and must be maintained in an unfolded state during their translocation to the Bam complex [3]. Three quality control proteins that chaperone OMPs across the periplasm are SurA, Skp and DegP, the latter of which also serves as a protease to degrade misfolded OMPs in the periplasm. The double mutants ΔsurAΔskp and ΔsurAΔdegP are not viable, which suggests that Skp, SurA and DegP might be functionally redundant [4]. The primary chaperone pathway to the Bam complex is thought to be SurA, while Skp and DegP are suggested to form a secondary pathway that is amplified in stress conditions [5]. However, single molecule studies with purified OmpC and Skp or SurA demonstrate that the two proteins recognise different conformations of the OMP and that while Skp is capable of dispersing aggregated OmpC, SurA is not [6]. In addition, the absence of SurA increases degradation of OMPs that have stalled on the Bam complex, whereas loss of Skp in this scenario has the opposite effect, enhancing their assembly [7]. This is potentially due to the role of Skp in recognising and removing stalled OMP substrates from the Bam complex before being degraded alongside the misfolded OMP by the protease DegP [8]. Together, these studies challenge the long-standing assumption that these pathways are redundant and suggests that Skp and SurA have specialised roles in maintaining OMPs in a folding competent state during periplasmic transit.

Following transport across the periplasmic space, OMPs are received at the outer membrane by the Bam complex. In Escherichia coli, the Bam complex is composed of two essential subunits, the outer membrane β-barrel BamA and the lipoprotein BamD, and three non-essential accessory lipoproteins, BamB, BamC and BamE [9, 10]. BamA is the central component of the complex and consists of a 16-stranded β-barrel and five periplasmically localised POlypeptide-TRansport-Associated (POTRA) domains that function to dock the accessory subunits [11, 12]. The POTRA domains are hypothesised to interact with incoming unfolded OMP substrates and feed them into the assembly machinery [13]. BamD is a lipoprotein composed of five tetratricopeptide repeat (TPR) domains that interact with POTRA 5 of BamA [9, 14, 15]. The remaining components of the complex are the accessory lipoproteins, BamB, BamC and BamE, which interact with BamA and BamD to form a functionally coordinated ring structure under the BamA barrel on the periplasmic face of the outer membrane [16–20].

The Bam complex accessory lipoproteins each contribute to full activity of the complex through a shared redundant function of coordinating efficient BamA/D interaction in OMP engagement and folding. Beyond this shared function, there is some limited information about specific roles of these proteins in the cell [10, 21]. For example, as part of the Rcs stress response signalling pathway, the outer membrane lipoprotein RcsF is threaded through an OMP during folding by the Bam complex. This process specifically requires BamE to directly interact with and coordinate BamA and BamD to enable completion of OMP folding around RcsF [22–24]. Lethal jamming of the Bam complex in the absence of BamE can be prevented by BamB [23, 25]. BamB has also been shown to be important in folding of high-flux substrates [26]. However, further information about the specific roles of BamB, BamC and BamE are lacking [21].

In this study, we provide evidence for specialised functions and show that changes to LPS structure lead to increased membrane fluidity and decreased Bam complex activity. We also identify that the cyclic form of enterobacterial common antigen (ECA), a periplasmic carbohydrate, is essential in the absence of the chaperone protein SurA. We demonstrate that the Bam accessory lipoproteins, including BamC, are essential in the absence of the peptidoglycan stem-peptide component meso-DAP, providing a strong phenotype for this poorly understood protein. Lastly, we show that loss of BamB causes the cells to over-initiate DNA replication, indicating potential coordination between OMP biogenesis and DNA replication. Together, our data indicate specialised roles for the Bam-associated proteins and highlights potential mechanisms for coordination of OMP biogenesis with processes such as outer membrane lipid biogenesis, peptidoglycan synthesis and DNA replication control.

Results

Loss of Bam-associated proteins leads to distinct phenotypes

We hypothesised that if the Bam-associated proteins have specialised roles in the cell it would be unlikely that they phenocopy each other. Therefore, we tested fitness of ΔbamB, ΔbamC, ΔbamE, ΔsurA, Δskp and ΔdegP mutants of E. coli K-12 BW25113 compared to the parent strain under various stresses. We arrayed the strains in 384-well format on solid agar plates containing each stress or compound. Endpoint pictures were taken to quantify colony fitness based on size by the image analysis software Iris [27]. Fitness was scored using the chemical genomics analysis software ChemGAPP Small by comparing the mean colony size for the mutant in each condition to the mean colony size of that mutant in the LB agar condition, which was normalised to a fitness score of 1 [28]. Fitness scores below 1 represent decreased fitness, as a function of colony size compared to growth on LB agar, and scores above 1 indicate increased fitness in that condition.

The fitness of each mutant across all conditions identified that the ΔbamB and ΔsurA mutants have the strongest fitness defects (Fig 1). Correlation between each mutant fitness profile pair was assessed by calculating the Pearson correlation coefficient. This demonstrated that ΔbamB and ΔsurA have a similar phenotypic profile with a correlation coefficient of 0.74, the highest of all pairs of mutants (Fig S1). Similar decreased fitness scores were observed for both strains in the presence of bacitracin, carbenicillin, erythromycin, sodium dodecyl-sulfate (SDS), osmotic stress conditions due to varied salt concentrations, and vancomycin (Fig 1). This is unsurprising as SurA is the major chaperone pathway for OMPs to reach the Bam complex resulting in low fitness of ΔbamB and ΔsurA when growing on LB at 37°C. However, they do not phenocopy in all conditions. For example, the ΔbamB strain grew better in the presence of the MreB inhibitor A22 with a fitness score of 1.59, but the ΔsurA mutant was less fit in this condition (score of 0.51). They also had opposing fitness scores in the presence of doxycycline (ΔbamB – 1.65, ΔsurA – 0.94) and hydroxyurea (ΔbamB – 0.68, ΔsurA – 1.14) (Fig 1). The ΔbamE mutant mildly correlated to ΔbamB with a coefficient of 0.51 (Fig S1). However, both mutants were sensitive to vancomycin (ΔbamB – 0.06, ΔbamE – 0.66) and bacitracin (ΔbamB – 0.02, ΔbamE – 0.58) two antibiotics that exceed the exclusion limit of outer membrane porins and to which sensitivity indicates damage to the permeability barrier. In contrast, ΔbamC has no strong phenotypes and correlated most strongly with the WT parent strain (Fig 1 and Fig S1). Similarly to ΔbamC, the Δskp mutant also has no strong phenotypes in our screen. The ΔdegP strain was negatively correlated with the other periplasmic chaperone pathway group mutants Δskp (-0.51) and ΔsurA (-0.37) and with ΔbamE (-0.68) due to opposing fitness scores in a range of stresses (Fig. 1 and Fig S1). Interpretation of this result is complicated by DegP not only functioning as a chaperone, but also as a protease that targets misfolded OMPs in the periplasm [29]. However, due to the role of DegP in degrading stalled OMPs that have been recognised by Skp, they might have been expected to correlate [8]. Together, the phenotypic profiling data indicate that while there are shared phenotypes between each knockout, we observe a broad range of unique, mutant specific phenotypes that suggests specialised roles beyond those observed previously.

Figure 1

TraDIS identifies unique synthetic lethal interactions for Bam-associated proteins

Considering that the Bam-associated proteins have specialised function, we sought to find their genetic interaction networks by using Transposon-Directed Insertion site Sequencing (TraDIS). When TraDIS libraries are generated in single gene deletion mutant backgrounds, the insertion frequencies of genes/regions compared to wild-type cells identifies genes that become more important for survival in the mutant backgrounds. These genes are termed conditionally essential, and help to identify co-dependencies of processes that the candidate gene might be involved in. The method can also identify mutants that are more fit in the given genetic background or condition [30–33]. Therefore, the parent strain (E. coli BW25113) and ΔbamB, ΔbamC, ΔbamE, ΔsurA, Δskp, or ΔdegP mutants were transformed with a mini Tn5-kanamycin transposon and recovered on LB agar at 37°C. Transposon mutants were then pooled and transposon insertion sites identified by DNA sequencing as described previously [32]. The BioTraDIS pipeline was used for data analysis to classify genes as either essential or non-essential. Essential genes were then compared to the parent BW25113 TraDIS library to identify genes that became essential in the knockout strains (conditionally essential) (Table S1) [34]. For quality control, two independent replicates of each transposon library were sequenced over several sequencing runs. Individual sequencing runs had a high correlation coefficient of ≥0.93 between samples and generated >6 million TraDIS reads and >500,000 unique insertion sites for each strain (Fig S2 and S3). Transposon insertion sites were evenly mapped throughout the genome, with the exception of an increased density around the origin of replication as expected due to gene dosage (Fig S4) [35].

These experiments are designed to identify novel genetic interactions, but this unbiased approach will also uncover known synthetic lethal gene pairs that allow benchmarking of our data. We confirmed known genetic interactions between surA and skp [4], degP and surA [4], bamB and degP [26] and bamE and bamB [25, 36] (Fig S5). The ΔbamB, ΔbamC and ΔbamE TraDIS libraries allowed identification of 93, 92 and 29 conditionally essential genes, respectively (Fig 2A and Table S1). This is particularly surprising as there are no known roles or dependencies for bamC and we did not observe strong phenotypes for ΔbamC in our phenotypic screen (Fig 1). Unfortunately, 54 of the genes identified as conditionally essential in the ΔbamC background are of unknown function, therefore providing no hypotheses for the role of BamC (Table S1). We then compared the conditionally essential gene lists to probe if they are involved in similar processes or contribute to the same function within the cell. In the case that all three Bam accessory proteins solely contribute to Bam function, we would expect to find significant overlap between mutants. However, we found many unique conditionally essential genes for each (62, 57 and 5 for ΔbamB, ΔbamC and ΔbamE, respectively), supporting the hypothesis that each has specialised roles in the cell (Fig 2A).

Figure 2

For the OMP chaperone pathway mutants we identified 247, 87 and 35 conditionally essential genes for the ΔsurA, ΔdegP and Δskp mutants, respectively (Fig 2A). The ΔsurA mutant had many strong phenotypes when probed against stresses (Fig 1), which could explain the large number of conditionally essential genes compared to the other OMP chaperone mutants (Fig 2A). Of the 87 conditionally essential genes in the ΔdegP dataset, only 15 were also in the Δskp background, despite DegP and Skp functioning together to degrade stalled OMP substrates [8]. This difference is also potentially due to the dual role of DegP as both a chaperone and a protease [29]. These results suggest that while there is functional redundancy between the SurA and DegP/Skp chaperone pathways, the function of the chaperones under specific chemical and genetic conditions is likely specialised (Fig 1 and 2A).To determine the functions and associated pathways of the conditionally essential genes identified in the mutant backgrounds, we completed GO and KEGG analyses. Enrichment analysis of KEGG pathways for the ΔbamC and Δskp datasets resulted in no significant enrichment of any one category. However, genes involved in “Lipopolysaccharide biosynthesis” and “Biosynthesis of nucleotide sugars” were enriched in the ΔsurA, ΔdegP and ΔbamE conditionally essential gene lists and “O-antigen nucleotide sugar biosynthesis” was enriched in the ΔsurA background (Fig 2B and Table S2). The most enriched KEGG pathway in the ΔbamB dataset was “Aminoacyl-tRNA biosynthesis”, however these genes all encode tRNAs and are on average 75 bp in length. Visual inspection of these gene regions identified sparse insertion density within the local genomic area leading to a likely false positive for conditional essentiality. The ΔbamB dataset is also significantly enriched for three other KEGG pathways: “Homologous recombination”, “Mismatch repair” and “DNA replication” (Fig 2B and Table S2). Together, the KEGG pathway enrichment analysis identifies the importance of LPS and ECA biosynthesis in these mutants as well as a potential link between OMP biogenesis and DNA replication. This directed our further investigations.

Genes required for heptose incorporation into LPS are synthetically lethal with bamB, surA and degP

We identified that fewer mutants were recovered with transposon insertions in genes involved in LPS assembly in the ΔbamB, ΔsurA and ΔdegP backgrounds than in the parent. In addition, genes involved in “Biosynthesis of nucleotide sugars” were significantly enriched, with genes identified being involved in the synthesis of heptose, a core component of LPS (Fig 2B and Table S2). Despite this category being enriched in the ΔbamE dataset, we found this was due to one gene in particular, glmS, and that the result was actually specific to the ΔbamB, ΔsurA and ΔdegP TraDIS libraries (Fig 3A and Fig S6). Synthesis of heptose consists of five main steps before it is incorporated into the LPS inner core (Fig 3B). In the ΔsurA and ΔdegP TraDIS libraries, all four genes involved in synthesis of heptose were essential: gmhA, gmhB, hldE and waaD. In the ΔbamB background all were essential except for gmhB, which is potentially because a ΔgmhB mutant is not completely devoid of heptose [37]. These data suggest that heptose production was functionally more important in ΔdegP and ΔsurA than in the ΔbamB mutant. The incorporation of heptose into the LPS structure was also functionally important. In ΔdegP and ΔsurA mutants, the genes waaC and waaF were conditionally essential, which are responsible for transfer of the first and second heptose onto the LPS inner core, respectively (Fig 3A and 3B) [38, 39]. In contrast, in the ΔbamB mutant, waaC was conditionally essential whereas waaF was not. This implies that while ΔdegP and ΔsurA mutants are not able to tolerate having LPS with only one heptose residue, the ΔbamB mutant is.

Figure 3

LPS truncation increases membrane fluidity and decreases Bam complex activity

Increased outer membrane fluidity decreases Bam activity and modifications to LPS structure can lead to changes in membrane fluidity [40]. Thus, we hypothesised that changes in LPS core structure would affect membrane fluidity to differing degrees, which could affect Bam complex efficiency. To test this hypothesis we assessed membrane fluidity and Bam complex activity in strains lacking the genes encoding the heptose biosynthesis pathway (ΔgmhA, ΔgmhB, ΔhldE, and ΔwaaD) and LPS core synthesis (ΔwaaC, ΔwaaF, ΔwaaP, ΔwaaG and ΔwaaY) (Fig 3). There were no differences in transposon insertion index for the gene waaY, which encodes the enzyme responsible for phosphorylation of the second heptose in the LPS inner core. Therefore, ΔwaaY should act as a control in this experiment [41]. In addition, ΔwaaP and ΔwaaG, which are responsible for phosphate addition to the first heptose, and incorporation of the first glucose, respectively [42], were included as a small decrease in the insertion indexes for these genes were observed in the ΔsurA and ΔdegP TraDIS datasets (Fig S7).

Fluidity of the membrane for each mutant was measured using the lipophilic pyrene probe, pyrene decanoic acid. The pyrene monomer can undergo excimer formation and demonstrates a shift in fluorescence, a process which is dependent on the ease of mobility within the membrane [43, 44]. Formation of the excimer was measured and compared to that of the parent strain (Fig 3C). The largest increase in membrane fluidity occurred in knockouts of genes required for heptose biosynthesis: ΔgmhA, ΔgmhB, ΔhldE, and ΔwaaD. Of this group, the ΔgmhB mutant had the smallest increase in membrane fluidity and this gene was also not essential in the ΔbamB background (Fig 3A and 3C). Of the genes required for LPS core biosynthesis, membrane fluidity was higher in ΔwaaC, ΔwaaF and ΔwaaG than in the parent strain with the biggest change being in ΔwaaC and the smallest in ΔwaaG. However, there was no significant increase in fluidity in the ΔwaaP mutant. Membrane fluidity of the ΔwaaF, ΔwaaP and ΔwaaG mutants was intermediate between that of the heptoseless ΔwaaC mutant and the parent, with the severity of the effect correlating to the severity of LPS core truncation. No significant change in membrane fluidity was measured in the ΔwaaY mutant control (Fig 3C).

Next, we investigated whether impairment of heptose synthesis affected activity of the Bam complex, which was monitored by an in vivo OmpT fluorescence assay. The outer membrane protease OmpT requires the Bam complex for folding and insertion into the membrane and is able to cleave a fluorogenic peptide, which is monitored by increased fluorescence over time [10]. Except for ΔwaaY, all mutants demonstrated a minimum 40% decrease in OmpT activity compared to the parent strain. The heptoseless ΔwaaC mutant exhibited the greatest decrease in OmpT activity. The effect of ΔwaaF, ΔwaaP and ΔwaaG mutations on OmpT activity levels was comparable to each other and intermediate between that of ΔwaaC and the parent strain (Fig 3D). In summary, mutations that lead to heptoseless LPS had the biggest increase in membrane fluidity and the lowest levels of OmpT activity with a graded response based on the severity of LPS core truncations. This suggests there is a correlation between LPS core structure, membrane fluidity and Bam complex activity that leads to a variety of Bam complex activity levels depending on the form of LPS in the outer membrane.

Generation of cyclic ECA is essential in the absence of the chaperone SurA

Genes involved in ‘Biosynthesis of nucleotide sugars’ were significantly enriched as conditionally essential in the ΔsurA mutant. The genes identified are involved in biosynthesis of ECA (Fig 2B), which is a highly conserved carbohydrate-derived molecule present on the external leaflet of the outer membrane and in the periplasm of Enterobacteriaceae [45, 46]. The ECA biosynthesis pathway genes were all conditionally essential with the exception of wxyE, which is required for survival of both the parent and the ΔsurA mutant (Fig 4A and 4B). In addition, fewer mutants were recovered with transposon insertions in the genes rffH and rffG than in the parent TraDIS library (Fig 4A). The gene products RffH and RffG catalyse the same enzymatic reaction and are homologous in sequence to the genes RfbA and RfbB, respectively. However, they form part of different operons and function in separate pathways despite rffG being able to complement an RfbB defective strain [47, 48]. This likely explains why rffH and rffG are not entirely essential in the ΔsurA background.

Figure 4

ECA exists in three forms that all share the same biosynthetic pathway: ECA that is covalently linked to LPS (ECA_LPS), covalently linked to diacylglycerol-phosphate (ECA_PG), or a cyclic form (ECA_CYC) that is localised to the periplasm as opposed to being surface exposed [45, 49]. We sought to determine which form of ECA is conditionally essential in ΔsurA. Synthesis of ECA_LPS is facilitated by the O-antigen ligase WaaL, which is responsible for attaching the ECA molecule onto the LPS core [50, 51]. However, in the ΔsurA TraDIS library, the gene waaL is non-essential (Fig 4C). This suggests that ECA_LPS is not essential in the ΔsurA mutant. The formation of ECA_PG is completed by attachment of linear ECA chains to diacylglycerol by a phosphodiester bond through an unknown mechanism [50, 52]. To determine the essentiality of ECA_PG we targeted synthesis of the donor molecule for ECA_PG, the phospholipid phosphatidylglycerol [53]. The gene pgsA encodes phosphatidylglycerophosphate synthase, which catalyses the first committed step in biosynthesis of phosphatidylglycerol. However, loss of pgsA is lethal due to mislocalisation of Braun’s lipoprotein, Lpp, to the inner membrane and activation of the Rcs stress system. These issues can be resolved by making an ΔlppΔrcsF mutant before constructing the ΔpgsA mutation [54–56]. The ΔpgsAΔsurAΔlppΔrcsF quadruple mutant was viably constructed and the absence of phosphatidylglycerol was confirmed by phospholipid extraction and thin layer chromatography (Fig 4D). This confirmed that the phospholipid donor for synthesis of ECA_PG is not essential in the ΔsurA mutant. Lastly, the gene wzzE is not required for the production of ECA_LPS or ECA_PG, but is required for production of ECA_CYC [50, 57, 58]. In the ΔsurA TraDIS library, the gene wzzE contained a conditionally essential region indicating that ECA_CYC is likely to be the form of ECA that is conditionally essential in a ΔsurA mutant (Fig 4A) [57].

meso-DAP is essential in the absence of BamB, BamC or BamE

The TraDIS data were searched for conditionally essential genes involved in cell envelope biogenesis pathways other than LPS and ECA biosynthesis. The gene dapF was identified as conditionally essential in ΔbamB and ΔbamC with visual inspection indicating decreased insertions in dapF in ΔbamE (Fig 5A). This was particularly surprising considering the lack of strong phenotypes in the ΔbamC background (Fig 1). DapF converts LL-diaminopimelate (LL-DAP) to meso-diaminopimelate (meso-DAP), which is then either decarboxylated to produce L-lysine by the enzyme LysA or is used in biosynthesis of peptidoglycan [59, 60]. The gene lysA was non-essential in all strains tested, suggesting the essentiality of dapF in these mutants is due to the requirement for meso-DAP in peptidoglycan (Fig 5A and 5B). To validate the genetic interaction between ΔdapF and components of the Bam complex, the dapF gene was knocked out in the BW25113 parent strain, ΔbamB, ΔbamC or ΔbamE mutants. The double knockouts were selected in the presence of externally supplied meso-DAP, which alleviated the loss of dapF. Cells were then grown in liquid media in the presence of meso-DAP before being serially diluted in un-supplemented media and assayed for survival in the presence or absence of meso-DAP by efficiency of plating assay. All strains grew equally as well as the parent on LB agar supplemented with 1 mM meso-DAP, except for ΔbamBΔdapF which showed a minor decrease in CFU and colony size (Fig 5C). However, in the absence of meso-DAP the single ΔdapF mutant demonstrated a decrease in survival and the double mutants were not viable (Fig 5C). Lastly, considering that loss of DapF will lead to a weaker peptidoglycan layer due to decreased crosslinks, we sought to negate the synthetic-lethal phenotype by growth in the presence of sucrose as an osmoprotectant [61, 62]. Unexpectedly, while the presence of 10% sucrose enabled survival of the single dapF mutant to levels comparable to the parent, this was insufficient to restore growth of the double mutant (Fig 5C). This data demonstrates that peptidoglycan structure is of increased importance in the absence of full Bam complex activity and that this may not simply be due to structural support for the envelope.

Figure 5

Regulation of DNA replication initiation is disrupted in the absence of BamB

Our TraDIS experiment identified that in the ΔbamB dataset three KEGG pathways were significantly enriched: “Homologous recombination”, “Mismatch repair” and “DNA replication”. Bacterial DNA replication is mainly regulated at the initiation step by DnaA (Fig 2A and 6A). The ATP-bound form of this protein binds to DnaA boxes present at the chromosomal origin of replication (oriC), with both high and low affinity, and directs replisome assembly [63]. Briefly, DnaA-ATP forms an oligomeric complex that facilitates unwinding of the DNA strands, which in turn enables assembly of the replisome: the complex responsible for replication of the bacterial chromosome. In E. coli, DnaA binding to its cognate DnaA boxes is modulated by regulatory proteins. Negative regulation is achieved by two proteins. The protein SeqA sequesters newly replicated origins, preventing premature reinitiation of replication [64–66]. The Hda protein works together with the β-sliding clamp to stimulate ATP hydrolysis of DnaA-ATP, which in the ADP-bound form is unable to promote initiation of DNA replication [64–68]. Positive regulation is facilitated by the DnaA initiator-associating factor, DiaA, which directly interacts with the N-terminal domain of DnaA to stimulate initiation at oriC [69, 70].

Figure 6

Analysis of the TraDIS data identified the genes holC, holD, matP, seqA and hda as conditionally essential in the ΔbamB mutant (Table S1). Within this set of genes, we chose to focus on those with the clearest conditional essentiality from the TraDIS data, hda and seqA. Insertions within the negative regulator encoding genes, seqA and hda, led to decreased survival of ΔbamB. In contrast, increased fitness of ΔbamB was observed for cells containing disruptions of the positive regulator gene diaA (Fig 6B). We confirmed the findings regarding the negative regulators by using CRISPRi. The bamB::aph allele was transferred into the dCas9 encoding strain E. coli LC-E18 and the kanamycin resistance marker was removed. The pSGRNA plasmid was used to express guide RNA targeting the seqA or hda genes in either the parent or ΔbamB. Expression of guide RNA targeting the seqA and hda genes led to significantly decreased survival of ΔbamB when compared to the parent strain, therefore confirming the observations made from TraDIS (Fig 6C). The positive genetic interaction between bamB and diaA was confirmed by measuring fitness of the ΔbamBΔdiaA mutant compared to both single deletion mutants. Fitness of the double mutant was greater than the expected fitness for this strain (Fig 6D).

Considering that loss of negative regulators of DNA replication initiation led to decreased fitness of ΔbamB, and loss of the positive regulator increases fitness, we hypothesised that the ΔbamB mutant may already have defective replication control. Therefore, we compared chromosomal content of the wild-type and ΔbamB strain by using an established replication run-out assay [71]. Both strains were grown to early exponential phase under conditions supporting fast cell growth (LB + 0.2% glucose), when cells contain multiple replicating chromosomes. We reasoned that dysfunction of replication control is usually exacerbated when cells grow with overlapping replication rounds. Next, cells were collected and treated with rifampicin to inhibit further rounds of DNA replication initiation, and with cephalexin to prevent cell division. Fixed cells were stained for DNA content with Sytox Green and analysed by flow cytometry. The comparison of chromosomal copy number revealed a higher proportion of cells with 16 chromosome copies for the ΔbamB mutant, which implies that initiation of replication occurs earlier during the cell cycle in this strain relative to the parent. Strikingly, a population of ΔbamB cells also displayed a number of chromosomes deviating from the normally observed 2ⁿ, as manifested by an additional peak between the 8 and 16 chromosome peaks. This is a hallmark of asynchronous DNA replication events, when not all replication origins in the cell fire simultaneously (Fig 6E, S8-10). Analysis of the other Bam-accessory lipoprotein gene mutants and Δskp by replication run-out assay revealed no significant changes from the parent pattern (Fig S11). Therefore, our results suggest that E. coli over-initiates DNA replication in the absence of BamB. This is consistent with the observations that ΔbamB cells are more fit if they lose the positive regulator of DNA replication, DiaA, but have decreased fitness in the absence of the negative regulators SeqA and Hda.

OMP trafficking protein conservation varies throughout Gram-negative bacteria

The data presented here demonstrate that in E. coli, the Bam-associated proteins likely have specialised functions that are potentially coordinated with LPS structure, the presence and form of ECA, peptidoglycan biogenesis and DNA replication control. Considering these results, we revisited the evolutionary conservation of Bam complex subunits and the periplasmic chaperone pathway proteins [72]. Reference genomes for a diverse range of Gram-negative bacterial representative strains were used to search for sequences coding the Bam complex subunits (as observed in E. coli) or the chaperone pathway proteins SurA, Skp and DegP. A combination of Prokka annotation, hmmsearch, SignalP 6.0 [73] and manual curation were used to generate a neighbour joining tree with a heat map of gene counts in each organism (Fig 7). While we found broad conservation of all query proteins within the gamma- and beta-proteobacteria, there are some notable exceptions. The BamB and BamC lipoproteins are not conserved within two species of the aphid endosymbiont Buchnera aphidicola and the data confirm previous unpublished observations that a strain of B. aphidicola, (subspecies Baizongia pistaciae) appears to have entirely lost genes encoding the Bam complex [74]. Likely this is due to the strain containing only a few OMPs, therefore no longer requiring a dedicated OMP assembly machinery.

Figure 7

Outside of the gamma- and beta-proteobacteria, requirement for the Bam-associated proteins appears to be diverse. The lipoproteins BamB and BamC are largely absent in the alpha-proteobacteria, and the epsilon-proteobacteria only encode the core components BamA and BamD, demonstrating reductive evolution. This minimalist form of the Bam complex is sufficient for OMP insertion, which is likely due to reduced β-barrel complexity as observed in Helicobacter pylori [75] (Fig 7). While there is diversity in the presence or absence of these proteins in the organisms analysed, multiple copies of some components were also identified. We found that Pseudomonas putida encodes two copies of BamA, BamE and SurA. Outside of the gamma-proteobacteria there are numerous examples of species encoding multiple copies of BamA, particularly within the alpha-proteobacteria, such as the plant pathogen Agrobacterium tumefaciens (Fig 7). These copies may play specialised roles in folding and inserting specific cargo, or they may alleviate an increased requirement for OMP insertion in these organisms. Taken together, the diverse range of Bam complex subunit compositions within Gram-negative bacteria support the hypothesis that the Bam-associated proteins likely play specialised roles that vary in each organism.

Discussion

The Bam complex has been identified as a significant unrealised target for new antimicrobials, with several new inhibitors of this essential outer membrane biogenesis machine being identified recently [77, 78]. However, despite this focus as an important new avenue for drug discovery, and recent advances in understanding the mechanism by which the Bam complex folds and inserts OMPs into the outer membrane [79, 80], we still do not fully understand the roles of the Bam-associated proteins in E. coli. We also do not understand why the complex appears to be modular, with subunits being differentially conserved throughout Gram-negative bacteria as shown through our analysis and those done previously (Fig 7) [72]. Constituent parts of the outer membrane such as phospholipid species, type of LPS, O-antigen, ECA, lipoprotein content, and variety/flux of the unfolded OMP proteins can vary across bacterial species [81, 82]. Such variation would likely result in significant differences in the biophysical constraints of the membrane environment in which the Bam complex must operate. In addition, there is likely to be differing requirements for the chaperone pathway proteins depending on the pool of client OMPs. Considering that Skp and SurA have been shown to act on different unfolded states of OmpC, with Skp likely being more important during stress conditions, the environmental conditions experienced by each species could also dictate the requirement for each OMP chaperone [6]. Together, this strongly suggests that each accessory lipoprotein is likely to have specialised functions and that the chaperone proteins are unlikely to be truly functionally redundant. This is supported by our phenotypic profiling of knockout strains. Should each of the accessory lipoproteins merely contribute to overall activity of the Bam complex, and if there is true redundancy in the chaperone pathway, then we would expect that the knockout strains should phenocopy each other. However, we found this not to be the case (Fig 1). Our TraDIS analysis also supports this conclusion as we saw a large number of unique conditionally essential genes for each mutant (Fig 2).

Role of the outer membrane lipid environment for OMP insertion

The Bam complex must function within the constraints of the outer membrane lipid environment, therefore OMP biogenesis is likely to be affected by changes to outer membrane lipids. Indeed, we found conditional essentiality between genes involved in LPS inner core biogenesis and either bamB, surA or degP. Complete truncation of the LPS inner core by disruption of the last enzyme in the heptose biosynthesis pathway has previously been demonstrated to increase membrane fluidity and lead to decreased Bam complex activity [40]. We confirmed this observation and further demonstrate that the changes to membrane fluidity are of graded severity relating to the severity of LPS inner core truncation. This in turn leads to a graded impact on Bam complex activity (Fig 3). The ΔbamB, ΔsurA and ΔdegP knockout strains have the most severe phenotypes within the set of strains tested here (Fig 1). Also, loss of BamB, SurA or DegP each leads to OMP assembly defects, decreased numbers of folded OMPs in the outer membrane, and accumulation of unfolded OMPs in the periplasm [9, 26, 83–85]. We hypothesise that the negative effect of increased membrane fluidity on the Bam complex in combination with decreased levels of OMP assembly in the absence of BamB, SurA or DegP leads to OMP assembly levels that are too low to sustain cell viability. This is of particular significance when considering the variation in LPS structure seen between different strains, and the LPS modifications available, which could affect the level of Bam activity and influence the conservation of each Bam-associated protein [81, 82].

A role for ECA_cyc in amelioration of periplasmic protein folding stress?

The carbohydrate antigen ECA consists of three repeating sugars, is widely conserved amongst the Enterobacterales, and is produced in three forms that are either surface exposed (ECA_LPS and ECA_PG) or periplasmically localised (ECA_CYC). The invariant nature of the antigen suggests it has an important undiscovered function, however the role of ECA in outer membrane biology remains unknown [49]. We identified that ECA biogenesis becomes essential in the absence of the chaperone SurA (Fig 4A). Production of ECA requires the lipid carrier undecaprenyl phosphate (Und-P), which is limited in the cell and utilised in numerous metabolic reactions including the production of peptidoglycan, capsule, O-antigen and membrane-derived oligosaccharides. Disruption of ECA biogenesis can cause stress on these other pathways due to Und-P sequestration in ECA dead-end intermediates [49, 57, 86]. However, the first gene in the ECA pathway is conditionally essential in the absence of SurA (Fig 4A), therefore the effects are unlikely due to stress on the Und-P pool. Instead, we showed that the periplasmically localised form, ECA_cyc, is essential in the absence of SurA (Fig 4). It is possible that ECA_cyc might help stabilise the outer membrane of a ΔsurA mutant, as it has been shown to suppress the envelope permeability defect of a ΔyhdP mutant. However, the molecular details of this suppression are yet to be discovered [57]. Alternatively, ECA_cyc could be performing a chaperone function directly, however this would require in vitro assays of protein aggregation with purified products for proof.

Coordination of OMP biogenesis with peptidoglycan biosynthesis

To ensure successful cell growth and division, activity of the Bam complex must be coordinated with biogenesis of the peptidoglycan layer. Our TraDIS analysis identified that in the ΔbamB, ΔbamC and ΔbamE mutants, the gene dapF was essential (Fig 5A). The DapF enzyme is responsible for synthesis of the peptidoglycan stem-peptide component meso-DAP. The single and double mutants could be rescued by exogenous meso-DAP, however only the single dapF mutant could be rescued by the osmo-protectant sucrose, which is sufficient to allow survival during disruption of peptidoglycan biogenesis (Fig 5B) [61]. Loss of meso-DAP in the single dapF mutant leads to accumulation of LL-DAP, which is incorporated into peptidoglycan and leads to decreased crosslinking [59, 62]. In the ΔdapF mutant the peptidoglycan layer would be able to withstand less mechanical load due to decreased crosslinking. This decreased mechanical strength, in combination with the reduced capacity for load bearing by the outer membrane in the Bam complex mutants, means that the cells are unlikely to withstand osmotic stress [87]. However, while this may be true for ΔbamB and ΔbamE mutants, there are no strong phenotypic effects for ΔbamC (Fig 1) [88]. Also, the loss of BamC has only a slight effect on the membrane permeability barrier, no detectable changes to the outer membrane proteome, and no effect on OMP folding in an in vitro assay [9, 89]. Consequently, the synthetic lethality between dapF and bamC is unlikely to be due to osmotic stress tolerance, as could be the case for the other mutants.

It has recently been demonstrated that the Bam complex preferentially inserts OMPs at the cell division site and that this is coordinated by interaction with the peptidoglycan layer [90]. While all of the Bam complex proteins were shown to interact with peptidoglycan in vitro, only BamA and BamC were found to interact with peptidoglycan in whole cells when analysed by crosslinking and pull-down assays [90]. Considering the interaction of mature peptidoglycan with BamC and the altered peptidoglycan structure in the ΔdapF strain [59, 62], we hypothesise that BamC might facilitate coordination between the Bam complex and peptidoglycan biosynthesis, however this requires further investigation.

Link between OMP biogenesis and DNA replication control

Not only must envelope biogenesis processes be coordinated with each other, but synthesis of the envelope must be coordinated with cell growth and cell division. Experimental evidence suggests that both DNA replication control [91, 92] and the rate of cell surface synthesis are coupled to cell volume [93]. This provides the rationale for existence of mechanisms that modulate DNA replication rate in response to cell envelope synthesis perturbation. Here we demonstrated that in the absence of BamB, cells over-initiate DNA replication and that synchrony of replication initiation at multiple origins is compromised (Fig 6E). Also, our phenotypic screen revealed that ΔbamB has increased fitness at sub-optimal growth temperatures. This would lead to slower growth, a factor that is known to alleviate some DNA replication control defects, such as those arising from loss of SeqA [64]. In addition, we found that the ΔbamB mutant is sensitive to hydroxyurea, which specifically inhibits ribonucleotide reductase and leads to depletion of the cellular dNTPs pool, replication fork arrest and genomic instability (Fig 1) [94, 95]. Similarly to ΔbamB, mutants that over-initiate replication, such as seqA mutants, are also sensitive to hydroxyurea [96].

Led by our TraDIS results, we confirmed that seqA and hda are essential in the ΔbamB mutant (Fig 6) and that this is likely due to over-initiation of DNA replication, which can be caused by all three of these mutations [64, 66, 97]. SeqA is required to prevent premature re-initiation of replication through sequestration of newly replicated GATC sites within oriC, whereas Hda prevents re-initiation of replication through regulatory inactivation of DnaA (RIDA), in combination with the β-sliding clamp of DNA polymerase [64, 98]. Interestingly, we also identified that two components of the β-sliding clamp loader complex, HolC and HolD, become essential in the ΔbamB mutant (Table S1). These proteins contribute towards activity of the clamp loader. Therefore, the synthetic lethality could be due to decreased RIDA and over-initiation of DNA replication, as observed for the hda mutant [68, 99].

This is the first link between the Bam complex and DNA replication control, however links between the envelope and DNA replication have been observed previously. In order to initiate DNA replication, the DnaA protein must be in the ATP-bound form and regeneration of DnaA-ATP from the ADP-bound form is facilitated by interaction with phospholipids. [100, 101]. DNA replication control and envelope biogenesis are also linked by the SeqA protein, which has been associated to the purified outer membrane fraction in a cell-cycle dependent manner, however the mechanism remains unclear [102]. In addition, disrupted origin/terminus ratios in seqA mutant strains are suppressed by mutations in the waaG, waaQ and waaP genes, which are responsible for addition of the outer core of LPS, heptose III addition to the inner core and phosphorylation of heptose II of the LPS inner core, respectively [81, 103]. Interestingly, the ΔbamB mutant is also sensitive to changes in membrane fluidity brought about by LPS modifications that map to the inner core (Fig 3).

We also identified that insertions in the diaA gene, which encodes a positive regulator of DNA replication initiation, can suppress ΔbamB. The diaA gene is part of a conserved four gene cluster that also encodes LpoA - a regulator of peptidoglycan biogenesis, YraN - a protein of unknown function, and DolP - a cell-division site localised outer membrane lipoprotein shown to interact with the Bam complex [104, 105]. The genes in this cluster all appear to be linked to regulating cellular growth processes such as peptidoglycan biogenesis, OMP insertion into the outer membrane and DNA replication. This is of particular interest considering our data demonstrate potential coordination of these processes, which is essential to ensure efficient growth and survival during cell division.

Together, our data demonstrate that the Bam-associated proteins have specialised roles in the cell and highlights potential future targets to understand these roles. We provide further evidence that OMP transport, folding and insertion is affected by, and potentially coordinated with, peptidoglycan and LPS structure. We also show that Bam function is correlated with DNA replication control and highlight the BamB lipoprotein, outer membrane fluidity and LPS structure as a potential route through which this could occur. This provides a strong direction for further study to understand this coordination.

Acknowledgements

This work was supported by a UKRI Future Leaders Fellowship [MR/V027204/1] and a Springboard Award [SBF005\1112] to Manuel Banzhaf. The work was also funded by the National Science Centre Poland [UMO-2014/13/B/NZ2/01139] (awarded to Monika Glinkowska), the KAUST baseline fund [BAS/1/1108-01-01] awarded to Danesh Moradigaravand who is a member of the KAUST Smart-Health initiative, and the EU ITN Train2Target [721484] that funded training of Kara Staunton.

Materials and methods

Bacterial strains and culture conditions

For TraDIS experiments and phenotypic profiling, the parent strain was E. coli K-12 BW25113. The E. coli ΔbamB, ΔbamC, ΔbamE, ΔsurA, Δskp and ΔdegP mutants were generated by transferring the relevant allele from the Keio library [106] into the clean parent strain by P1 transduction [107]. The kan^Rcassette was then removed by using the pCP20 plasmid [108] to leave a small in-frame 34 amino acid peptide consisting of residues from the FRT scar and the first amino acid and last seven of the target gene. These strains were then used for construction of transposon libraries and phenotypic profiling. The same approach was used for construction of the E. coli ΔwecA::aph, ΔwecD::aph, ΔwecF::aph, ΔompT::aph, ΔdapF::aph and ΔdiaA::aph mutants. The diaA::aph allele was also transferred into the ΔbamB derivative of E. coli BW25113 to generate the ΔbamBΔdiaA::aph double mutant. The dapF::aph allele was also transferred from the Keio library into ΔbamB, ΔbamC, ΔbamE, ΔsurA, Δskp and ΔdegP mutants in the presence of 1 mM meso-diaminopimelate in order to generate double mutants. The ΔrcsFΔlppΔpgsAΔsurA mutant was generated by transfer of alleles from the Keio library and subsequent removal of the kan^R cassette by use of the pCP20 plasmid before incorporation of the next allele by P1 transduction. The E. coli ΔgmhA::aph, ΔgmhB::aph, ΔhldE::aph, ΔwaaD::aph, ΔwaaC::aph, ΔwaaF::aph, ΔwaaP::aph, ΔwaaG::aph, ΔwaaY::aph mutants were created by using λ-Red recombination, as previously described for single-step gene inactivation [109]. All mutants were confirmed by polymerase chain reaction and strains were routinely cultured on LB agar and LB broth. The dCas9 expressing strain E. coli LC-E18 and the pSGRNA plasmid was a gift from David Bikard (Addgene plasmid # 115924) and has been described previously [110]. The E. coli LC-E18 ΔbamB derivative was constructed as described for the BW25113 ΔbamB derivative with the kan^R cassette removed in order to allow selection of the pSGRNA plasmid in this background. All strains used in this study are listed in Table S3. Bacterial cultures were grown at 37°C unless otherwise stated. Where stated, the medium was supplemented with kanamycin (50 μg/ml), carbenicillin (100 μg/ml), and meso-DAP (1 mM). For micro-dilution survival assays, bacteria were grown in 5 ml LB medium at 37°C with aeration for ∼16 hours. Cultures were normalised by optical density to OD₆₀₀ = 1.00, 10-fold serially diluted in LB, and 2 μl of each dilution was inoculated onto LB agar plates.

Phenotypic screening

The E. coli K-12 BW25113 parent strain, ΔbamB, ΔbamC, ΔbamE, ΔsurA, Δskp and ΔdegP mutants were screened against diverse stress conditions to phenotypically profile the mutants. Strains were arrayed in 384-well format and inoculated on 2% agar LB plates using a BM3-BC robot (S&P Robotic Inc.). In addition, the mutants and the wild type were subsequently inoculated on LB agar plates containing different stress conditions. The inoculated plates were incubated for 12 to 14 hours at 37°C before being imaged under controlled lighting with an 18-megapixel Canon rebelT3i (Canon) camera on the BM3-BC robot (S&P Robotic Inc.). Images of the plates were analysed using the software IRIS, which measured the size, opacity and circularity of each colony [27]. A total of 4 replica plates were generated for each stress condition. Fitness of the mutants was then scored and analysed using the ChemGAPP Small software [28]. Mean colony size for the mutant in each condition was compared to the mean colony size of that mutant in the LB agar condition, which was normalised to a fitness score of 1 [28]. Fitness scores below 1 represent decreased fitness, as a function of colony size compared to growth on LB agar, and scores above 1 indicate increased fitness in that condition. Each of the plates contained 56 replicates for the parent BW25113 strain (WT), ΔbamB, ΔbamC, ΔbamE and ΔdegP strains with 52 replicates for ΔsurA and Δskp for plate space requirements. There were 4 replicate plates for each condition, all of which were treated as individual replicates and compared to their respective wild type parent replicates on each plate before the average was taken. The probability that the two means are equal across replicates was obtained by a one-way ANOVA. Correlation of fitness ratios for each strain was assessed by calculating the Pearson correlation coefficient for averaged fitness scores across replicates using Python before being plotted as a heatmap.

TraDIS library construction

The E. coli K-12 BW25113 parent strain, ΔbamB, ΔbamC, ΔbamE, ΔsurA, Δskp and ΔdegP mutants were transformed with the EZ-Tn5™ <KAN-2> Tnp Transposome (Cambio) as previously described [32]. Approximately 1 million mutant colonies were pooled, thoroughly mixed and stored in LB supplemented with 15% glycerol at -80°C. DNA was extracted from at least two samples of each transposon library to generate independent sampling replicates for library generation.

Sequencing of TraDIS libraries

Cells from the pooled mutant library were harvested and genomic DNA was extracted for library preparation and sequencing as previously described [32]. Samples were sequenced using an Illumina MiSeq with a 150 cycle v3 cartridge.

TraDIS data analysis

Raw data were processed using a series of custom scripts as previously described [32]. The data were trimmed using Fastx barcode splitter and trimmer tools (Pearson et al., 1997) and filtered based on inline indexes. The accuracy of the transposon sequences were checked in two steps: the first 22 bases, allowing for three nucleotide base mismatches and the last 10 bases of the transposon, allowing for up to one mismatch. Using Trimmomatic, sequences with less than 20 bases in length were removed (Bolger et al., 2014). TraDIS data were then analysed using BioTraDIS (https://sanger-pathogens.github.io/Bio-Tradis/) [34] and aligned to the E. coli BW25113 reference genome CP009273.1, available from NCBI (Tatusova et al., 2014). We used SMALT (https://www.sanger.ac.uk/tool/smalt/) as an aligner with zero value for mismatch threshold. We also set the parameter smalt_r, determining how to treat multi-mapping reads, to zero. This avoided repetitive elements to count as essential. We used the scripts tradis_essentiality.R and tradis_comparison.R as part of the package to produce the list of essentiality and then to compare the control and test libraries, respectively.

Membrane fluidity assay

Membrane fluidity was measured using the membrane fluidity kit (Abcam: ab189819), as previously described, but with minor modifications [104]. Bacterial strains were grown to mid-exponential phase (OD₆₀₀ = ∼0.4-0.6) in LB medium. Cells were harvested by centrifugation, washed with phosphate buffered saline (PBS) and incubated with labelling mix (10 μM pyrenedecanoic acid (PDA), 0.08% pluronic F-127, in PBS) in the dark for 20 min at 25°C with rocking. Cells were then washed twice in PBS and re-suspended in PBS prior to measuring fluorescence (excitation = 350 nm, emission = either 400 nm or 470 nm). Membrane fluidity was estimated by measuring the ratio of excimer (470 nm) to monomer (400 nm) fluorescence. The emission spectra were compared to unlabelled cells to confirm membrane incorporation and each experiment contained triplicate technical repeats and was then repeated three times. Membrane fluidity of the mutants of interest were expressed relative to the parent E. coli strain. Experiments were performed in technical triplicate and were repeated three times. Two sample t-tests were used to assess statistical significance of differences from the WT strain.

OmpT in vivo fluorescence assay

The OmpT assay for monitoring Bam activity was performed as described previously with minor modifications [10, 16, 40]. Bacterial strains were grown to mid-exponential phase (OD₆₀₀ = ∼0.4-0.6) in LB medium and were normalised to OD₆₀₀ of 0.2 in growth media. The cell suspension (5 μl) was then added to 95 μl of 25 μM fluorogenic peptide, Abz-ARRAY(NO₂)-NH_2, diluted in PBS. Fluorescence emission was immediately measured (excitation = 325 nm, emission = 430 nm) over a period of 5 h, with readings every 20 s. OmpT activity was expressed relative to the parent strain. Experiments were performed in technical triplicate and were repeated three times. Two sample t-tests were used to assess statistical significance of differences from the WT strain.

Phospholipid extraction and thin layer chromatography

Phospholipids were extracted using an adapted Bligh-Dyer method [111]. Bacterial cultures were grown overnight (∼16 hours) at 37°C with shaking and cells were harvested by centrifugation. The bacterial cell pellets were resuspended and normalised to OD₆₀₀ = 3 and mixed with methanol and chloroform (2:1). The cell suspension was incubated at 50°C for 30 min before additional chloroform and water was added to the samples to produce a final ratio 2:2:1.8 of methanol/chloroform/water. Following centrifugation, the phospholipid-containing phase was extracted and dried under nitrogen before being stored at -20°C. Samples were re-dissolved in chloroform before being separated by thin layer chromatography on silica gel membrane using a solvent system that consisted of chloroform, methanol, acetic acid (65:25:10). The plate was subsequently stained with phosphomolybdic acid (PMA) and warmed until the lipid species were visible.

Efficiency of plating assay

Efficiency of plating assays were completed as previously described [112]. Cells were grown overnight in LB (supplemented with 1 mM mesoDAP where indicated) before being normalised to OD₆₀₀ = 1.00 and serially diluted 1:10. Following dilution, 2 µl was spotted on agar plates containing supplements where indicated and incubated at 37°C for ∼16 hours before being imaged.

Genetic interaction analysis

For genetic interaction analysis, overnight cultures were diluted 1/100 and then grown to OD₆₀₀ = 0.8-1. Cultures were then spread on LB agar plates using sterile glass beads to create source plates, before being incubated for 12 to 14 hours at 37°C. Each strain was arrayed on LB agar plates, from the source plates, to form 384 colonies of 96 replicates for each strain using a BM3-BC robot (S&P Robotic Inc.). Plates were incubated at 37°C for 12 to 14 hours. The plate was then imaged under controlled lighting, with an 18-megapixel Canon rebelT3i (Canon) camera on the BM3-BC robot (S&P Robotic Inc.). Images were then analysed using the software called IRIS to measure the size, opacity and circularity of each colony on the plates [27]. Fitness of the mutants was then analysed and compared using the software ChemGAPP GI [28].

Flow cytometry analysis

Chromosome number measurements were performed as described previously, but with several modifications [71]. Briefly, cells were grown with aeration at 37°C until OD₆₀₀=0.15 in LB medium supplemented with 0.2% glucose. Samples were collected, treated with 150 μg/ml rifampicin, and 10 μg/ml cephalexin and incubated for 4 h at 37°C with mixing. Incubation with antibiotics results in cells containing an integral number of chromosomes, corresponding to the number of replication origins at the time of drug treatment. Subsequently, cells were harvested, washed with TBS (20 mM Tris-HCl pH 7.5, 130 mM NaCl) and fixed with cold 70% ethanol overnight. Additional samples were collected at OD₆₀₀=0.15 without antibiotic treatment and fixed as above.

Prior to flow cytometry analysis, cells were resuspended in 50 mM sodium citrate followed by RNA digestion with RNase A for 4 h. Chromosomal DNA was stained with 2 mM Sytox Green (Invitrogen) and DNA content per cell was measured with BD FACS Calibur at 488 nm Argon Ion laser. MG1655 (WT) strain grown in AB medium containing one of the following carbon sources: 0.4% sodium acetate, 0.2% glucose, 0.2% glucose + 0.5% casamino acids or in LB medium with 0.2% glucose, treated with antibiotics, fixed and stained as above was used as a standard for each flow cytometry measurement, indicating cells containing 1/2 , 2/4, 4/8 or 8/16 chromosomes, respectively. Flow cytometry data was analyzed using FlowJo ver. 10.8.0.

Conservation analysis

Reference genomes were downloaded for each species from NCBI in fasta format and annotated using Prokka [113]. The 16s rRNA sequence for each species was then extracted from Prokka ffn files in fasta format. A Kmer based neighbour joining tree was produced via extracting a one-step sliding window of 8 base kmers from each 16s rRNA sequence. Following this, the number of unique kmers was counted for each species in order to produce a Kmer profile. The Jaccard similarity between each species Kmer profiles was then used to produce a distance matrix. This distance matrix was input into the R module ape::nj to produce the neighbour joining tree. The neighbour joining tree was reformatted as a nexus file and visualised within Figtree (http://tree.bio.ed.ac.uk/software/figtree/), where the tree was rooted at the midpoint and branches were transformed to become proportional.

To count the instances of the proteins BamA, BamB, BamC, BamD, BamE, DegP, SurA, and Skp, hmmsearch was used to extract proteins with the relevant domains from the Prokka annotated faa files. The Pfam domains used were PF01103, PF13360, PF06804, PF13525, PF04355, PF09312 and PF03938 for BamA, BamB, BamC, BamD, BamE, SurA, and Skp, respectively. For DegP, both PF00595 and PF13180 were used and the intersect taken. An e-value of 0.001 was used as a threshold for significant hits by hmmsearch. For BamA, hits annotated as the query protein were inputted into hmmscan and hits with at least one POTRA domain and an Omp85 domain were counted as true BamA hits. Hits containing a Tam-POTRA domain were excluded. Where a hit had only an Omp85 gene, the adjacent gene was checked for the POTRA domains. For BamB, any hmmsearch hits annotated as BamB or hypothetical proteins by Prokka were input into SignalP 6.0 [73] and hmmscan. Proteins assigned a LIPO (Sec/SPII) signal peptide within SignalP 6.0, with only one PQQ_2 domain within hmmscan were selected as true BamB hits. For BamC, all hits from hmmsearch were inputted into hmmscan to confirm the presence of the Lipoprotein-18 domain. Hits were also inputted into Phmmer and proteins with significant homology to BamC were counted as true hits. For BamD, hypothetical proteins and BamD hits were inputted into hmmscan and Phmmer and those with homology to BamD and a non-outcompeted YfiO domain counted as true hits. For BamE, hypothetical proteins and BamE hits were inputted into hmmscan and those with a singular SmpA_OmlA domain were counted as true hits. For DegP hypothetical proteins and DegP hits were inputted into hmmscan. If a hit had a PDZ type domain, Peptidase_M50 domain or Trypsin_2 domain then it was visualised on Pfam and manually assigned as DegP based on structure. If a DegP hit had any other type of domain, it was discounted. For SurA, all hits were run through hmmscan and Phmmer. Those with a SurA_N domain, Rotamase domain and a Rotamase_3 domain and homologous to SurA within Phmmer were counted. For Skp, all hits were run through Phmmer and only hits with only an OmpH domain and with homology to Skp counted as true hits. Finally, the neighbour-joining tree was edited in iTOL [76] to produce the heatmap of gene counts.