Discovery of a Diverse Set of Bacteria That Build Their Cell Walls without the Canonical Peptidoglycan Polymerase aPBP

ABSTRACT Peptidoglycan (PG) is a highly cross-linked peptide-glycan mesh that confers structural rigidity and shape to most bacterial cells. Polymerization of new PG is usually achieved by the concerted activity of two membrane-bound machineries, class-A penicillin binding proteins (aPBPs) and class-B penicillin binding proteins (bPBPs) in complex with shape, elongation, division, and sporulation (SEDS) proteins. Here, we have identified four phylogenetically distinct groups of bacteria that lack any identifiable aPBPs. We performed experiments on a panel of species within one of these groups, the Rickettsiales, and found that bacteria lacking aPBPs build a PG-like cell wall with minimal abundance and rigidity relative to cell walls of aPBP-containing bacteria. This reduced cell wall may have evolved to minimize the activation of host responses to pathogens and endosymbionts while retaining the minimal PG-biosynthesis machinery required for cell elongation and division. We term these “peptidoglycan-intermediate” bacteria, a cohort of host-associated species that includes some human pathogens.

Polymerization of the PG precursor, lipid II, into growing PG strands has two enzymatic requirements: glycosyltransferase (GTase) activity to form b1-4-glycosidic bonds between disaccharide residues, and transpeptidase (TPase) activity to cross-link the peptide side chains on adjacent strands. Class A penicillin binding proteins (aPBPs) are large transmembrane proteins that possess both activities (3,4). In contrast, class B PBPs (bPBPs) are transmembrane proteins that only possess TPase activity. It was recently shown that shape, elongation, division, and sporulation (SEDS) proteins, which form a complex with bPBPs, possess GTase activity in some organisms (5)(6)(7)(8)(9). Thus, it is possible that bPBP/SEDS and aPBPs form two complementary cell wall synthetic motors that both possess full TPase and GTase activity.The aPBPs are widespread and essential in most bacterial species and were long thought to be the primary drivers of PG polymerization; however, several lines of recent evidence suggest that most nascent PG growth is, in fact, driven by the newly described bPBP/SEDS activity, with aPBPs playing a supportive role in repairing cell wall defects. First, the subcellular localization of aPBP within the cytoplasmic membrane is distinct from foci formed by bPBP/SEDS pairs in the Gram-positive bacterium Bacillus subtilis and the Gram-negative bacterium Escherichia coli, indicating distinct activities (10). Second, it has been shown that all four aPBP genes can be deleted in B. subtilis to generate slow-growing but viable bacteria (11). Third, decreased expression of aPBPs in E. coli results in cells that remain rod shaped, but with decreased levels of PG cross-linking and less ability to sustain cell wall damage (12). These data suggest distinct, though partly complementary, roles for aPBPs and bPBPs in PG polymerization.
Here, we identify a group of bacteria that lack any identifiable aPBP genes but can synthesize a PG-like structure that is required for viability, demonstrating that aPBPs are dispensable for PG polymerization. Compared with closely related species that have aPBPs, the PG in these "PG-intermediate" (PGi) organisms is less abundant, and the cells do not have a uniform rod shape. These PGi organisms all exhibit an obligate intracellular or endosymbiotic lifestyle, and the osmotic protection and avoidance of innate immune receptors may have driven the selection for a minimal PG cell wall.

RESULTS
The class A penicillin binding protein gene has been lost at least four times during evolution. We previously used a comparative genomics approach to characterize the distribution of cell wall biosynthesis genes in obligate intracellular bacteria, which identified species that possessed almost all genes in the classical PG biosynthesis pathway but specifically lacked any identifiable aPBPs (13). Meeske et al. also performed a phylogenetic analysis showing that SEDS/bPBPs are more widely conserved than aPBPs and that some organisms lack aPBPs but retain a SEDS/bPBP synthase (6). Here, we extended this analysis to include 119 different strains across 9 major bacterial groups, including the identifiable obligate intracellular bacteria, facultative intracellular bacteria, and endosymbionts with complete genomes available in KEGG, as well as related freeliving bacteria ( Fig. 1A and Fig. S1). We identified clusters of bacteria that lack aPBPs in four unrelated groups, the Rickettsiales, the Actinobacteria, the Gammaproteobacteria, and the Planctomycetes Verrucomicrobia Chlamydiae (PVC) superphylum. These organisms were classified as PGi, with the criterion that they possess most lipid II biosynthesis genes, at least one SEDS and bPBP gene, but no identifiable aPBP gene. We also identified bacteria in five diverse groups that had lost the majority of their PG biosynthesis genes and would be predicted to lack PG synthesis capability (PG-negative [PGn]). In contrast, we term species that encode aPBPs, bPBPs/SEDS, and other genes in the PG biosynthesis "PG-classical" (PGc).
This result showed that aPBPs are dispensable for growth in host-associated bacteria. We extended this analysis to all 4,985 closed bacterial genomes available on KEGG without distinguishing organisms by lifestyle ( Fig. 1B and Table S1). We found that the majority (3,935) encoded aPBPs, bPBPs, and SEDS (PGc), a smaller fraction (988) had specifically lost aPBPs (PGi), and a small number (20) had lost all three classes of genes (PGn). In contrast, we only identified 20 instances of strains that had retained an aPBP but lost SEDS and/or bPBP. This analysis builds on previous reports of organisms FIG 1 The class A PBP gene has been lost at least four times during evolution. (A) Analysis of key genes in the peptidoglycan biosynthesis pathway showing their presence or absence in obligate intracellular, facultative intracellular, host-associated, and free-living bacteria across 9 major bacterial groups. The predicted peptidoglycan (PG) status is shown as well as the primary site of replication of the bacteria. Genes involved in lipid II biosynthesis, SEDS (Continued on next page) Peptidoglycan Intermediate Bacteria lacking aPBPs (6,13) and demonstrates that while aPBPs are dispensable for growth, SEDS/bPBPs have been almost universally retained.
The Rickettsiales are an order of obligate intracellular bacteria that are dominated in nature by arthropod endosymbionts (Wolbachia strains and other species/strains) but also include a number of important arthropod-transmitted animal pathogens (species of Neorickettsia, Anaplasma, Ehrlichia, Orientia, Rickettsia). The pathogens Ehrlichia spp., Anaplasma phagocytophilum, Neorickettsia sennetsu, and Neorickettsia risticii were all classified as PGn, while the closely related pathogens Anaplasma marginale, Anaplasma centrale, Neorickettsia helminthoeca, Orientia tsutsugamushi, and Wolbachia strains were all classified as PGi. Remarkably, despite significant genome reduction, all 44 Rickettsia species/strains contain a full complement of PG synthesis genes and were classified as PGc.
The Gammaproteobacteria include obligate intracellular, bacteriocyte-associated intracellular, facultative intracellular, host-associated extracellular, and free-living extracellular species. Out of 30 Gammaproteobacteria strains, we identified 7 PGn/unclassified, 6 PGi and 17 PGc. We found three instances where closely related strains had distinct classifications: (i) Buchnera aphidicola, where five strains were predicted to be PGn (or unclassified) and two other strains of B. aphidicola (APS, Bp) were predicted to be PGc. B. aphidicola strain JF98 was the only strain in our data set that retained aPBP but lost bPBP/SEDS; however, this also lacks murB, murE, and murF and is likely in the process of pathway degradation. (ii) Baumannia cicadellinicola, where one strain ("Candidatus Baumannia cicadellinicola" BGSS) was classified as PGc while another strain (B. cicadellinicola Hc) lacked aPBP and was classified as PGi. (iii) Coxielliae, where Coxiella burnettii was classified as PGc and a Coxiella endosymbiont of Amblyomma americanum was classified as PGi.
The PVC group includes obligate intracellular pathogens (species of Chlamydia, Waddlia), amoeba endosymbionts (Simkania spp.), and extracellular bacterial species (Akkermansia muciniphila, Planctomyces spp.). We analyzed seven distinct species, and all except Akkermansia were classified as PGi. Akkermansia, which is a host-associated extracellular bacterium, retained aPBP and was classified as PGc.
The Actinobacteria are a group of Gram-positive bacteria that includes Mycobacterium species. We analyzed three species and found that Tropheryma whipplei, which is a human pathogen generally thought to have an obligate intracellular lifestyle, was classified as PGi, while the other species were classified as PGc.
Together, this analysis led us to the hypothesis that the PGi bacteria produce a PG wall that is different from that produced by PGc bacteria and that this would confer an advantage within the obligate intracellular/endosymbiotic life cycle. In order to test this hypothesis, we selected a group of six related organisms from within one of these groups, the Rickettsiales, and set out to determine whether they synthesized PG walls with different characteristics. For this analysis we selected representative PGc (Rickettsia canadensis), PGi (Orientia tsutsugamushi, Wolbachia endosymbiont of melanogaster [Wm], and Anaplasma marginale), and PGn (Anaplasma phagocytophilum, and Ehrlichia chaffeensis) species.
PG-intermediate bacterial species build a PG-like structure. We used a NOD1 reporter assay to assess whether the PGc and PGi species generated detectable PG-like structures. NOD1 is a mammalian innate immune receptor that detects PG fragments containing the PG-specific amino acid meso-DAP that is commonly found in Gram-negative  Table S1. bacterial PG (14). The NOD1 reporter assay measures the response of human embryonic kidney (Hek1) reporter cells to meso-DAP-containing PG fragments (15). We purified 6 Rickettsiales species from host cells and added them to NOD1 reporter cells at a constant bacterial concentration ( Fig. 2A and B). While PGn species did not stimulate a NOD1 response, both PGc and PGi induced NOD1 activation. This demonstrated the presence of meso-DAP-containing PG in R. canadensis, A. marginale, O. tsutsugamushi, and Wolbachia pipientis (Wp). The magnitude of this activation was higher in the PGc species R. canadensis than in PGi organisms, suggesting a higher abundance of PG in this species.
We analyzed PG levels at different stages of the bacterial life cycle. The Anaplasmataceae A. marginale, A. phagocytophilum, and E. chaffeensis undergo a biphasic life cycle in which (B) Quantification of relative PG levels using a NOD1 reporter cell line. Extracellular (e) and intracellular (i) bacteria were isolated from host cells and quantified by qPCR, and 1 Â 10E4 bacteria were added to HEK-Blue NOD1 reporter cells. Activation of NOD1 in response to peptidoglycan was measured by secreted alkaline phosphatase activity. Positive control (pos) was 10 mg/ml iE-DAP. Individual values and median are shown, and statistical analyses show the difference between two groups as measured by an unpaired t test using the software GraphPad Prism. Gray asterisks show the results of statistical comparisons between each group and the positive control. Red asterisks show the difference between the i and e population of a particular bacterial species. (C) Sensitivity of Rickettsiales to cell wall-targeting drugs. Bacteria were grown in the presence of drugs at the following concentrations: chloramphenicol, 100 mg/ml; penicillin G, 150 mg/ml; Dcycloserine 250 mg/ml; and phosphomycin, 40 mg/ml. The bacterial copy number after growth was compared with that in the absence of drugs. A bacterial species was scored as susceptible to that drug if the bacterial copy number was reduced in a statistically significant manner as measured by an unpaired t test. Gray cell, bacteria susceptible to drug; white cell, bacteria resistant to drug. Raw data are given in Fig. S1. (D) Quantification of bacterial regrowth after 10 min of incubation in sucrose phosphate buffer (SPG) or water. Bacteria were isolated from host cells, exposed to osmotic shock (pure water) or an osmotically protective buffer (SPG) for 10 min, and then grown in host cells for 7 days. Bacterial growth was measured by qPCR. Individual values and median are shown, and an unpaired t test was used to compare the water-treated and untreated groups within each bacterial species using the software GraphPad Prism. (E) Visualization of a PGlike structure in R. canadensis and PGi species. Bacteria were grown in the presence of a D-alanine analog containing an alkyne group (EDA). After fixation, incorporation of probes into nascent PG was detected by labeling with azide-alexa488 using a copper catalyzed click reaction. Bacteria were counterstained using bacterium-specific antibodies. PGc (R. Peptidoglycan Intermediate Bacteria ® they differentiate between a replicative intracellular form and an infectious extracellular form (16). We recently showed that O. tsutsugamushi also differentiates into distinct intracellular and extracellular populations (manuscript in preparation). In contrast, differentiation has not been described in R. canadensis and Wp. We isolated bacteria from intracellular and extracellular populations in A. marginale, O. tsutsugamushi, A. phagocytophilum, and E. chaffeensis and found that in the PGi species A. marginale and O. tsutsugamushi, the PG levels were significantly higher in the intracellular population (Fig. 2B), at which time the bacteria are actively undergoing growth and division. PG could not be detected in either intracellular or extracellular populations of A. phagocytophilum or E. chaffeensis.
Next, we determined the susceptibility of the same six species to PG-targeting drugs, in order to determine whether they made a PG-like structure that was required for their growth (Fig. 2C and Fig. S2). We used the bacterial ribosomal inhibitor chloramphenicol as a positive control, because this is known to be effective against Rickettsiales. We tested the susceptibility to D-cycloserine (targets proteins involved in the isomerization of L-Ala to D-Ala and its dimerization into D-Ala-D-Ala); penicillin G (targets TPase activity of aPBP and bPBP), and fosfomycin (targets MurA, a cytoplasmic protein required for early stages of synthesis of the PG precursor lipid II). The PGc species R. canadensis was sensitive to all cell wall-targeting drugs that we tested. The PGi organisms were all sensitive to D-cycloserine and fosfomycin, consistent with the presence of an essential PG-like structure and as shown previously for some of these (17)(18)(19). A. marginale was susceptible to penicillin G; both O. tsutsugamushi and Wp were not. The susceptibility of PGi organisms to D-cycloserine and fosfomycin demonstrates that they build a PG-like structure that is required for their growth; however, the differential susceptibility of PGi organisms to penicillin G cannot currently be explained and may reflect different PBP modifications or differences in membrane structure or permeability. PGn organisms were insensitive to all the cell wall-targeting drugs that we tested, consistent with a lack of a functional PG cell wall.
Next, we tested whether the PG-like structure of PGi species conferred osmotic protection. We isolated bacteria from their host cells, exposed them to hypoosmotic shock (water), and assessed the effect on subsequent bacterial growth (Fig. 2D). We found that the PGn species were highly sensitive to hypoosmotic shock, whereas growth of the PGi species A. marginale and Wp as well as the PGc R. canadensis was not affected. Growth of the PGi O. tsutsugamushi was affected by hypoosmotic shock, demonstrating that the amount of PG in this organism was not sufficient to confer complete protection. While the mechanistic basis of this difference is unknown, it may reflect the fact that A. marginale and Wp replicate within a membrane-bound vacuole and may have evolved additional rigidity in their membranes to survive vacuole-associated hypoosmolarity compared with the cytoplasm-residing PGi species O. tsutsugamushi.
Visualization of a PG-like structure in PG-intermediate species using a PG-specific metabolic probe. We used a clickable D-amino acid analog, ethynyl-D-alanine (EDA), to determine the spatial localization of PG in PGc and PGi Rickettsiales species (Fig. 2E). This is an orthogonal chemical probe that incorporates into the PG of growing bacterial cells and can be conjugated to a fluorophore after fixation for fluorescence microscopy analysis. We have previously used this to label the PGi O. tsutsugamushi (17). This probe labeled both PGc and PGi species in our study, indicating the presence of some form of PG sacculus. However, we found that the labeling was frequently unsuccessful despite carefully controlling for variability. This may be due to batch-dependent differences in permeability of host cells or differences in the activity of PG synthesis machinery at different stages of bacterial growth. The PGn species A. phagocytophilum and E. chaffeensis could never be labeled with EDA, consistent with our hypothesis that these lack any PG-like structure and demonstrating that the probes do not bind nonspecifically to bacterial cells under these experimental conditions. It has been shown that some PGi Chlamydiales species only synthesize PG at their septum (20). Such localization was never observed in any of the species studied here.
Analysis of the cell shape of PGc, PGi, and PGn Rickettsiales. The low levels of NOD1 activation in PGi species (Fig. 2A), combined with the fact that PG has been difficult to detect in both O. tsutsugamushi and Wp, suggests that the absence of aPBP leads to a reduced amount of PG in the cell wall. We reasoned that this may result in reduced structural rigidity and that this would affect bacterial shape. Infected cells were examined by transmission electron microscopy to document the shape of the six Rickettsiales species within host cells (Fig. 3). We found that PGc R. canadensis forms Peptidoglycan Intermediate Bacteria ® regular, rod-shaped cells, while PGi and PGn species adopt irregular and/or round cells. This finding is consistent with an electron microscopy analysis showing that the Gammaproteobacteria "Candidatus Baumannia cicadellinicola" strain BGSS (endosymbiont of blue-green sharpshooter), classified here as PGc, is rod shaped while the closely related Baumannia cicadellinicola strain Hc (endosymbiont of glassywinged sharpshooter) classified here as PGi, is irregular and non-rod shaped (21).

DISCUSSION
Here, we used comparative genomics to identify 16 species across 4 major bacterial groups that have lost aPBPs but retained genes for synthesis of the PG precursor lipid II as well as at least one bPBP and one SEDS gene. We characterized three of these species from within the order Rickettsiales and showed that they generate a PG-like structure that is required for bacterial growth. It has already been shown that multiple chlamydial species, which are not closely related to Rickettsiales, generate minimal PG-like structures (20,22), and we hypothesize that an absence of aPBPs in Gammaproteobacteria and Actinobacteria will also generate bacteria with similar cell wall characteristics.
The species that lack aPBP were almost all associated with an intracellular lifestyle (15 out of 16). There are two possible reasons why adaptation to an intracellular lifestyle would confer selective pressure to reduce the amount of PG in the bacterial cell wall. First, PG is a strong stimulator of the host immune system via peptidoglycan recognition proteins such as NOD1/NOD2 and PG recognition proteins in mammals and peptidoglycan recognition proteins (PGRPs) in invertebrates. Activation of these systems leads to induction of antibacterial mechanisms that intracellular bacteria aim to evade, and we show here that both PGc and PGi Rickettsiales are able to activate NOD1. There is substantial variation in the abundance and distribution of PGRPs in invertebrate hosts (23), and differences in the magnitude of PG-sensing between hosts may underpin why some host-associated bacteria were able to retain a classical cell wall (PGc) while others were under strong selective pressure to reduce or remove it (PGi/PGn). Second, the intracellular niche (whether cytosolic or vacuolar) offers an osmotically protective environment, and therefore the bacteria may not require cell walls with the same rigidity as their freeliving counterparts. Rickettsia species counter this argument yet are the only analyzed Rickettsiales species that synthesize lipopolysaccharide (24), which may require a more rigid cell wall for scaffolding this large glycoconjugate.
How do PGi bacteria build a PG wall in the absence of aPBPs, a bifunctional TPase and GTase normally considered a major driver of PG polymerization? All the PGi strains identified in our analysis retained at least one copy of the two known bPBP and SEDS genes. bPBPs have TPase activity, and SEDS from E. coli and B. subtilis are known to have GTase activity (5,6,8,9). Thus, it is plausible that bPBP/SEDS complexes are the major drivers of PG polymerization in aPBP-negative PGi organisms. This is supported by the observation that aPBPs in B. subtilis are dispensable for growth (11), resulting in a reliance on bPBP/SEDS-driven PG polymerization analogous to the PGi species in our study. Our model for PG polymerization in PGc, PGi, and PGn bacteria is shown in Fig. 4.
These results also raise the following question: what is the structure of the bacterial cell wall in PGi organisms? The reduced level of NOD1 activation shown in this study compared with E. coli and the PGc R. canadensis, the difficulty in D-alanine orthogonal probe labeling, and the historical inability to detect PG in PGi organisms (18,(25)(26)(27)(28) suggest that the PG in these bacteria is not abundant, and in organisms with biphasic lifestyles this structure is further depleted during some stages of the growth cycle. Since PG in Gram-negative bacteria is only one or a few layers thick (1), the observed low abundance is likely to result from PG consisting of shorter glycan strands and/or less peptide cross-linking. Diminished cross-linking within the PG sacculus would result in a larger mesh size and lower structural support to the underlying membrane (Fig. 4). This would explain the reduced osmotic protection of PGi Rickettsiales compared with the closely related PGc relatives, as well as the loss of regularity in cell shape that would result from a less extensive-and therefore less rigid-PG sacculus. This interpretation of the PG structure in PGi organisms is consistent with a role for aPBP in increasing the cross-linking and repairing damage (12) rather than being the major driver of PG polymerization.
Together, these data show that aPBP is dispensable for PG polymerization and raises questions about the structure and synthesis of PG in organisms that naturally lack this important component of the PG biosynthesis machinery.

MATERIALS AND METHODS
Comparative genomics analyses. A total of 119 bacterial taxa (Table S2) were selected for comparative analysis based on their lifestyle, phylogenomic relationship, and genome status (closed genomes only). Then, 32 genes of interest were chosen (Table S3), 10 lipid II biosynthesis genes, 2 SEDS family protein genes, 13 bPBP genes, and 7 aPBP genes. Identifiers for all taxa and genes were located in the Kyoto Encyclopedia of Genes and Genomes (KEGG) and used as input to kegghole, part of the keggerator package (24). This software queries KEGG for the presence of the input genes in all input taxa and generates a presence/absence matrix. Genomes that appeared to lack an input gene according to KEGG were manually queried using the protein sequence from a related organism and NCBI's blastp program (default settings except for an E value cutoff of 1). Genes absent from all 117 query taxa were removed. Finally, phylogenetically related taxa with identical patterns of gene presence/absence were compressed into taxon groups for simplicity.
HEK-Blue hNOD1 (InvivoGen, USA; chkb-hnod1) cells were used for NOD1 reporter assays and were grown in 25-cm 2 flasks at 37°C and 5% CO 2 with growth medium DMEM with 4.5 g/liter glucose, 10% heat-inactivated FBS, 100 U/ml penicillin, 100 mg/ml streptomycin, 100 mg/ml Normocin, and 2 mM L-glutamine. Selective antibiotics, 30 mg/ml of blasticidin and 100 mg/ml of Zeocin, were added after passage 2 to maintain the cell line. Bacterial copy number was determined by extracting DNA using alkaline lysis treatment and then performing quantitative PCR (qPCR) relative to known standards (29) ( Table 2).
Quantification of relative peptidoglycan levels using NOD1 reporter assay. HEK-Blue hNod1 cells (Invitrogen, USA; hkb-hnod1) were grown in DMEM growth medium with 4.5 g/liter glucose, 10% heatinactivated fetal bovine serum, 100 U/ml penicillin, 100 mg/ml streptomycin, 100 mg/ml Normocin, and 2 mM L-glutamine plus selective antibiotics (30 mg/ml of blasticidin and 100 mg/ml of Zeocin). Cells were seeded on clear-bottom black 96-well plates (Corning, USA; 29444-008) 2 days before infection. Bacteria were taken from prepared aliquots kept at 280°C in sucrose phosphate glutamine (SPG). All bacteria were heat-inactivated at 90°C for 30 min before being added to host cells. All infections were carried out in triplicate. After 2 days of infection, growth medium was replaced with HEK-Blue detection medium (InvivoGen, USA; hb-det2) for secreted embryonic alkaline phosphatase (SEAP) detection. The plate was further incubated at 37°C and 5% CO 2 and quantified by spectrophotometry (Synergy H1; BioTek) at 640 nm from 6 h of the addition of detection media. Results were added to Prism (GraphPad Software, San Diego, CA, USA), and an unpaired t test was performed to compare bacterial SEAP levels to positivecontrol iE-DAP (g-D-Glu-mDAP) at 10 mg/ml, as well as to perform pairwise comparisons between intracellular and extracellular bacteria in the case of Wolbachia and Orientia tsutsugamushi species.
Growth inhibition experiments. Bacteria were grown in 12-well plates in their respective cell lines in the presence of drugs at the following concentrations: 100 mg/ml chloramphenicol, 150 mg/ml penicillin G, 250 mg/ml D-cycloserine, and 40 mg/ml phosphomycin. After 5 days of growth (7 days for Wolbachia) bacterial DNA was extracted and quantified by qPCR as described above. For quantifying the effect of osmotic shock, bacteria were isolated from host cells, resuspended in pure water for 10 min, and then grown in fresh host cells. Bacterial growth after 5 days was quantified by qPCR. Results were added to Prism (GraphPad Software, San Diego, CA, USA), and an unpaired t test was performed to compare growth in the presence or absence of drugs or growth in the presence or absence of osmotic shock.
Immunofluorescence labeling, click labeling, and confocal microscopy. All fixed cells were permeabilized in 0.5% Triton X for 30 min, 100% ethanol for 1 h on ice, and 1 mg/ml lysozyme in sterile tris-EDTA for 1 h at room temperature. Primary antibodies were added for 1 h at 37°C (TSA56, 13-6, ANAF16C1, FtsZ, dog serum, and anti-p44; see Table 3). Samples were washed 3 times with phosphatebuffered saline-bovine serum albumin (PBS-BSA), and then the appropriate secondary antibodies were diluted 1:500 and incubated for 30 min at 37°C in the dark (goat anti-rat IgG Alexafluor 555 conjugate [Thermo Fisher A-21434], goat anti-rabbit, Alex Fluor 594 [Thermo Fisher A-11012], goat anti-mouse IgG superclonal Alexa Fluor 555 [Thermo Fisher A28180], and goat anti-canine IgG Texas red [Novus Biologicals NBP173511]). The nuclear stain Hoechst was diluted to 1:1,000 and included with the secondary antibody incubation. Before addition of the mounting medium, cells were washed with 1Â PBS again.
The metabolic click-labeling is based on the Click-iT L-homopropargylglycine (HPG) Alexa Fluor protein synthesis assay kits (molecular probe by Life Technologies). To incorporate HPG at each time point, infected cells were incubated in the minimal medium without L-methionine (Dulbecco's modified Eagle's medium [DMEM], catalog [cat.] no. 21013) containing 25 mM HPG for 30 min at 37°C. Labeled bacteria were washed twice in PBS plus 1 mg/ml BSA before fixing with 1% formaldehyde or methanol (antibody 13-6 only) and subsequently permeabilized with 0.5% Triton X for 30 min, 100% ethanol for 1 h on ice, and 1 mg/ml lysozyme in sterile tris-EDTA for 1 h at room temperature. After washing with PBS plus 1 mg/ml BSA, Click-iT reaction cocktail was incubated with cells for 30 min at room temperature protected from light. The component of Click-iT reaction cocktail is based on Click-iT HPG Alexa Fluor protein synthesis assay kits [cat.] no. C10428. The azide dye (Alexa Fluor 488; Invitrogen A10266) was used at final concentration of 5 mM. After the click reaction, the cells were ready for immunofluorescent fluorescence labeling and imaging as described above.
Imaging was performed using an Observer Z1 LSM700 confocal microscope with an HBO 100 illuminating system equipped with a Â63/1.4 Plan-APOCHROMAT objective lens (Carl Zeiss, Germany) and 405-nm, 488-nm, and 555-nm laser lines. In some cases, we used a TCS SP8 confocal microscope (Leica Microsystems, Germany) equipped with a Â63/1.4 Plan-APOCHROMAT oil objective lens with a 1.4-mm working distance and 405-nm, 488-nm, 552-nm, and 638-nm laser lines. Transmission electron microscopy. Bacteria were grown in their respective cell lines, harvested by trypsinization, and pelleted at 1,000 Â g in a microcentrifuge. Culture medium supernatants were removed, and the resultant cell pellets were fixed using 2.5% formaldehyde, 2.5% glutaraldehyde, and 0.1 M sodium cacodylate buffer, pH 7.4 (EMS catalog no. 15949). After overnight fixation at 4°C, the pellets were gently rinsed and floated in 0.1 M sodium cacodylate buffer, pH 7.4 and postfixed with 1% osmium tetroxide in 0.1 M sodium cacodylate buffer, followed by en-bloc staining in 1% aqueous uranyl acetate. Pellets were dehydrated through a graded series of ethanol and propylene oxide up to 100% propylene oxide and then incubated in a 1:1 mixture of propylene oxide and EMBed 812 (Electron Microscopy Sciences; 14120). Pellets were equilibrated in 100% EMBed 812 overnight, placed into fresh EMBed812 in flat embedding molds, and cured at 60°C. Ultrathin sections (;70 nm) were cut, and grids were stained with uranyl acetate and lead citrate. Sections were imaged using a Thermo Fisher FEI Tecnai 12 transmission electron microscope, and micrographs were recorded using a Gatan OneView 16-megapixel camera.
Data availability. All data are included in the figures and supplementary information.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.