Bacterial size matters: Multiple mechanisms controlling septum cleavage and diplococcus formation are critical for the virulence of the opportunistic pathogen Enterococcus faecalis

Enterococcus faecalis is an opportunistic pathogen frequently isolated in clinical settings. This organism is intrinsically resistant to several clinically relevant antibiotics and can transfer resistance to other pathogens. Although E. faecalis has emerged as a major nosocomial pathogen, the mechanisms underlying the virulence of this organism remain elusive. We studied the regulation of daughter cell separation during growth and explored the impact of this process on pathogenesis. We demonstrate that the activity of the AtlA peptidoglycan hydrolase, an enzyme dedicated to septum cleavage, is controlled by several mechanisms, including glycosylation and recognition of the peptidoglycan substrate. We show that the long cell chains of E. faecalis mutants are more susceptible to phagocytosis and are no longer able to cause lethality in the zebrafish model of infection. Altogether, this work indicates that control of cell separation during division underpins the pathogenesis of E. faecalis infections and represents a novel enterococcal virulence factor. We propose that inhibition of septum cleavage during division represents an attractive therapeutic strategy to control infections.


Introduction
Enterococci are Gram-positive commensal bacteria colonizing the gastrointestinal tract of humans. They are opportunistic nosocomial pathogens that can cause a wide range of lifethreatening infections in immunocompromised patients or following antibiotic-induced dysbiosis [1]. The emergence of enterococci as nosocomial pathogens relies on the capacity of these bacteria to colonize the host and to grow in a wide range of harsh conditions (e.g., in the presence of bile salts or in iron-depleted environments) [2]. Enterococci are intrinsically resistant to multiple antibiotics, such as cephalosporins and several aminoglycosides, and can also acquire resistance to glycopeptides. Vancomycin-Resistant Enterococci (VRE) represent a major problem in clinical settings as they can potentially transfer resistance genes to other pathogens such as Staphylococcus aureus [3,4].
Two enterococcus species, Enterococcus faecium and Enterococcus faecalis are the most clinically relevant [5]. E. faecium infections are caused by a particular subset of clones specifically found in hospital settings that share several acquired mobile genetic elements [6]. In contrast, the E. faecalis strains responsible for hospital-acquired infections are also found in healthy individuals and genes associated with virulence are not only exclusively present in clinical isolates [7]. How this organism can cause infections therefore remains poorly understood.
In the present work, we study the regulation of daughter cell separation during cell division and explore the impact of this process on pathogenesis. We previously revealed that in E. faecalis, one peptidoglycan (PG) hydrolase with N-acetylglucosaminidase activity (named AtlA) is dedicated to septum cleavage to allow separation of daughter cells at the end of division [8]. Using a combination of in vitro experiments and sets of isogenic strains, we describe multiple mechanisms controlling the activity of AtlA. We show that control of septum cleavage during growth underpins the formation of diplococci and short chains, a property critical to cause lethality in the zebrafish model of infection.

Results
High expression levels of septum hydrolytic enzymes are not sufficient for cell separation proteins expressed under their own promoters (PatlA::atlA-his and PatlB::atlB-his). Culture samples were harvested at the end of exponential growth (OD 600 = 1) and His-tagged AtlA and AtlB were detected by western blotting using an anti-histidine serum (Fig 1A). Unlike Histagged AtlA, AtlB was barely detectable (Fig 1A, lanes 2-5 and lanes 6-9; see arrowhead on Fig  1A). Next, we tested whether increasing the level of expression of AtlB could enable this peptidoglycan hydrolase to cleave the septum efficiently. We built a strain producing an AtlB-His tagged protein expressed under the atlA promoter (PatlA::atlB-his), thereby replacing the AtlA open reading frame by AtlB. As expected, expression of atlB-his under the control of the atlA promoter increased the production of AtlB-His to levels similar to AtlA-His (Fig 1A, compare lanes 2-5 with lanes [10][11][12][13]. The impact of AtlB production on septum cleavage was analyzed by flow cytometry to measure bacterial chain lengths, as previously described [8]. Increasing the production of AtlB to levels similar to those of AtlA was not sufficient to shorten bacterial cell chains (Fig 1B). The cell chain length of the PatlA::atlB-his strain was not significantly different from that of the ΔatlA strains (P>0.05; n = 3). This result indicated that the relatively low production level of AtlB does not account for the low septum cleavage activity of this enzyme. This prompted us to explore the enzymatic properties of AtlA and their impact on cell separation.

N-terminal proteolytic cleavage of recombinant AtlA stimulates septum cleavage
Previous studies indicated that truncation of the AtlA N-terminal domain (residues 54 to 172) only had a marginal impact on the activity of the recombinant enzyme tested in vitro against whole PG molecules (sacculi) as a substrate [10]. We sought to re-investigate the contribution of the N-terminal domain to AtlA activity using an in vitro assay (as described in [8]) to specifically measure septum cleavage. A recombinant AtlA protein (residues 54 to 737) harboring an N-terminal 6-Histidine tag and a Tobacco Etch Virus (TEV) protease site at the end of the Nterminal domain was expressed in Escherichia coli and purified (Fig 2A, lane 1). Following cleavage with TEV protease, truncated AtlA was recovered by metal affinity chromatography (Fig 2A, lanes 2 to 4). In agreement with previous results, the specific activity of the truncated AtlA (AtlA ΔN ; 488.2 ± 155.6 ΔOD 600 /nmole/min) was slightly higher than that of AtlA (252.5 ± 27.4 ΔOD 600 /nmole/min) when assayed against whole sacculi ( ÃÃ P = 0.0018; n = 9; Fig 2B). However, using the specific septum cleavage assay, AtlA ΔN was more than 10-fold active than the full-length AtlA enzyme. Whilst 1.6 ± 1.2 pmoles of AtlA ΔN were sufficient to disperse 50% of the cell chains, 18.2 ± 2.0 pmoles of AtlA were required for a similar septum cleavage activity ( ÃÃÃ P = 0.0008; n = 3; Fig 2C). . The amino acid sequences between N-terminal (orange) and catalytic (grey) domains in AtlA and AtlA TEV are described. The TEV cleavage site is boxed. B. Comparison of specific enzymatic activities of AtlA and AtlA ΔN in vitro using whole PG sacculi as a substrate; **P = 0.0018; n = 9. C. Flow cytometry analysis of septum cleavage activity of recombinant full-length AtlA and the N-terminally truncated variant (AtlA ΔN ). Activity is expressed as a percentage of the median forward scattered (FSC) light value corresponding to cell chains formed by the atlA mutant (ΔatlA) used as a substrate; ***P = 0.0008; n = 3.

AtlA N-terminal domain glycosylation inhibits septum cleavage
The N-terminal domain of AtlA contains a high proportion of threonine and serine residues (28% and 12%, respectively; S1D Fig). This property prompted us to test whether this domain can be O-glycosylated. To purify AtlA produced by E. faecalis, a recombinant strain expressing a C-terminally 6His-tagged AtlA protein and a TEV site between the N-terminal and catalytic domains was constructed by allele exchange (Fig 4A). Cell surface proteins were extracted with 8M urea and His-tagged AtlA protein was purified by metal affinity chromatography. Two major bands of 75 kDa and 62 kDa matching the expected molecular weights of the full-length and N-terminally truncated AtlA, respectively, were detected ( Fig 4B, lane  1). A carbohydrate moiety was detected on the full-length AtlA protein, but absent on the truncated AtlA ( Fig 4B, lane 2), suggesting that glycosylation occurred at the N-terminal domain of AtlA. To confirm this hypothesis, exponentially growing cells were harvested and incubated in the presence of TEV protease. This treatment released a glycosylated polypeptide matching the apparent molecular weight of the N-terminal domain (Fig 4C, lanes 2 and  3 and Fig 2A, lane 3). No glycosylated polypeptide was detected when the protease was omitted, therefore indicating that the N-terminal domain of AtlA is glycosylated. In Gram-positive bacteria, two glycosyl transferases named GtfA and GtfB have been shown to be essential for surface protein glycosylation [11][12][13]. We used allele exchange to inactivate two putative glycosyl transferase homologs (gtfA and gtfB; EF2891 and EF2892 in E. faecalis V583) sharing the same glycosyl transferase domain (PFAM PF00534). Following incubation of cells harboring an in-frame deletion of the gtfAB locus in the presence of TEV protease, no glycosylated peptide could be detected (Fig 4C, lanes 4-6). Altogether, these results show that the Nterminal domain of AtlA is glycosylated and that this posttranslational modification requires the glycosyltransferases gtfAB. Next, we explored the impact of AtlA glycosylation on septum cleavage during growth by measuring the bacterial chain length of gtfAB mutants by flow cytometry (Fig 4D) and light microscopy (S2 Fig). We compared the cell chain lengths of E. faecalis JH2-2 (WT) forming mostly diplococci and short chains (2-4 cells) and atlA 1-4 , producing AtlA 1-4 lacking two LysM modules (6-12 cells) with the chain length of their ΔgtfAB counterparts. Pairwise comparisons of cell chain length by flow cytometry revealed that ΔgtfAB mutants formed shorter chains than parental strains, thus indicating that the lack of glycosylation enhanced AtlA septum cleavage activity. The cell chains of the parental strain (WT; FSC = 46.09 ± 0.43) were longer than those from the ΔgtfAB mutant (FSC = 42.04 ± 0.66; ÃÃ P = 0.0017; n = 3). A more pronounced difference was measured Swapping AtlA N-acetylglucosaminidase domain for other hydrolase domains cleaving distinct peptidoglycan bonds does not abolish septum cleavage in E. faecalis Substrate recognition by the catalytic domain of peptidoglycan hydrolases is an important factor modulating enzymatic activity [14]. The fact that AtlA is dedicated to septum cleavage can therefore be underpinned by the recognition and cleavage of a specific peptidoglycan structure present at the septum. We hypothesized that if such is the case, the N-acetylglucosaminidase activity of AtlA should be essential for septum cleavage. To explore this possibility, we constructed a gene replacement vector encoding an allele of atlA with a catalytic domain flanked by two restriction sites (NcoI and BglII). These two sites were used to swap the Nacetylglucosaminidase domain of AtlA for the catalytic domains of E. faecalis N-acetylmuramidase AtlB [8], Staphylococcus aureus N-acetylmuramoyl-L-Alanine amidase Atl [15] or Streptococcus thermophilus D,L endopeptidase Cse [16] (Fig 5A and S3 Fig). AtlA alleles encoding variants with distinct peptidoglycan cleavage specificities ( Fig 5B) were introduced on the chromosome by gene replacement and expressed as a single copy under the atlA promoter. Western blot analyses indicated that all AtlA variants were expressed at similar levels, except AtlA Cse , which was subject to proteolysis (S3C Fig) The septum cleavage activity in each strain was analyzed using flow cytometry. Cell chain length of the ΔatlA deletion mutant was used as a reference to define the forward scattered light value corresponding to maximal (100%) chain length. Introduction of the NcoI and BamHI restriction sites had a limited impact on the size of the cell chains (11.68 ± 1.35% of the ΔatlA mutant chains versus 10.59 ± 1.36% of the ΔatlA mutant for the parental JH2-2 strain). All strains expressing AtlA variants with altered enzymatic specificity formed shorter chains than the ΔatlA mutant in exponential phase ( Fig 5C). This result indicated that the peptidoglycan cleavage specificity of AtlA is not an essential property of the enzyme for septum cleavage. The forward scattered light measurements corresponding to the strains expressing AtlA variants compared to the ΔatlA mutant ranged from 28.71 ± 5.38% of the ΔatlA mutant for the strain expressing AtlB with a muramidase activity to 53.63 ± 1.3% of the ΔatlA mutant for the strain expressing AtlA with an endopeptidase activity. The relatively higher forward scattered light values associated with this strain could be due to the proteolysis of the chimeric protein (S3C Fig). The chain lengths of strains expressing chimeric proteins were all significantly shorter than those of the ΔatlA mutant ( ÃÃ P<0.01; n = 3).
with the TEV protease were analysed in parallel. In both cases, a glycosylated polypeptide with the expected molecular weight for the N-terminal domain (see Fig 2A) was detected while no signal was observed in the negative control. When a similar experiment was repeated with protein extracts from a ΔgtfAB mutant, no glycosylated protein was detected, indicating that this operon is involved in the post translational modification of AtlA. D. Comparison of median forward scattered (FSC) light values corresponding to the cell chain lengths of WT, ΔgtfAB, atlA 1-4 , atlA 1-4 ΔgtfAB, ΔatlA and ΔatlAΔgtfAB strains. https://doi.org/10.1371/journal.ppat.1006526.g004 Control of cell chains length is critical for E. faecalis virulence Control of cell chains length is critical for E. faecalis virulence that LysM A is critical for optimal septum cleavage. Swapping LysM B for LysM A in AtlB led to a septum cleavage activity comparable to that of AtlA ( Fig 6C). The contribution of individual LysM repeats to the septum cleavage activity was investigated. A set of strains expressing AtlA with 1 to 5 LysM repeats (henceforth referred to as atlA 1 to atlA 1-5 strains) was constructed by allele exchange (Fig 6D). Western blot and zymogram analyses (S4A and S4B Fig) showed that all strains produced AtlA proteins with the expected size and in similar amounts, except for the atlA 1 strain in which lower amounts were detected. AtlA activity decreased as LysM repeats were truncated. Next, we measured the impact of LysM truncations on the septum cleavage activity by measuring the cell chain length. Flow cytometry analyses revealed that sequential truncation of LysM modules led to a stepwise increase of cell chain length (Fig 6E and S4C Fig). This result indicated that optimal septum cleavage requires the full-length LysM domain, each module providing an additive contribution to AtA enzymatic activity.

Cell separation defects abolish E. faecalis virulence in the zebrafish model of infection
We investigated whether the formation of long chains in E. faecalis has an impact on virulence using the zebrafish model of infection [17]. We compared the lethality induced by the wildtype OG1RF strain to that of an in-frame atlA deletion mutant OG1RF (ΔatlA) forming long chains. In the ΔatlA mutant, each chain (equivalent to 1 CFU) can contain several viable cells. To eliminate this bias, ΔatlA chains were sonicated to mechanically separate cells (S5A and S5B Fig) [18] and establish the number of cells per CFU. This information was then used to inject the same number of cells (but different CFU numbers) for each strain. Zebrafish embryos were infected 30h post fertilization (hpf) and survival was monitored over the following 90h. One of three independent experiments is shown in Fig 7A whilst the results for three biological replicates are shown in S5C and S5D Fig. Injection of ca. 1,000 OG1RF cells killed between 60% and 77% of larvae (n = 79) depending on the experiment. In contrast, the ΔatlA mutant only killed 4% to 20% of larvae (n = 88), showing a significant reduction in virulence ( ÃÃÃ P<0.001 for the experiment shown in Fig 7A). At this stage, we envisaged that this difference could be attributed to several factors: (i) the increased chain length of the ΔatlA mutant, (ii) (an)other alteration(s) in cell surface properties (like in the Streptococcus mutans atlA mutant; [19]) or (iii) a defect in biofilm formation [20]. To specifically investigate the contribution of bacterial chain length to virulence, we subjected the ΔatlA mutant to mild sonication. This treatment is dispersing bacterial chains (  Fig 7A), killing between 46% and 59% of larvae (n = 90).
Previous work revealed that phagocyte evasion is a critical step for E. faecalis pathogenesis in the zebrafish [17]. We therefore quantified phagocytosis in zebrafish larvae infected with bacteria expressing the green fluorescent protein (GFP) (Fig 7B). Confocal microscopy images were used to measure bacterial uptake by phagocytes labelled with anti L-plastin antibodies. To specifically investigate the impact of the chain forming phenotype on phagocytosis, we compared the uptake of long and short bacterial chains formed by the ΔatlA mutant before and after sonication (ΔatlA S ). The ratio between green fluorescence area inside to outside phagocytes was significantly higher for the ΔatlA mutant ( ÃÃ P = 0.0098; n = 7). A significant difference was also measured when we compared the uptake of the ΔatlA mutant to that of wild-type ( Ã P = 0.0438; n = 8). As expected, no difference was detected between wild-type and ΔatlA S cells. Representative images used to quantify uptake are shown as an example (Fig 7B). Bacteria forming long chains were mostly found inside phagocytes (96% of total fluorescence area) as opposed to diplococci and short chains corresponding to wild-type and ΔatlA S cells (75% and 60% respectively).
Death rates were the same when immunocompromised zebrafish (n>20 per group) were infected with the wild-type OG1RF strain or with the ΔatlA mutant. In both cases, injection of 1250 cells in pu.1 morphants led to more than 90% mortality within 20h post infection (Fig 7C  and S5E Fig). This result suggested that the impaired virulence of the long chain-forming ΔatlA mutant is due to the fact that this strain is unable to evade phagocytosis during infection. To make a direct comparison of the uptake of wild-type and ΔatlA cells by phagocytes, an in vitro assay was performed using monocyte-derived macrophages obtained from peripheral blood mononuclear cells from healthy volunteers (Fig 7D). Phagocytic uptake was quantified following labeling of bacterial cells with pHrodo-S-ester, a pH-sensitive dye displaying increased fluorescence in low-pH compartments of phagosomes [21]. In each pairwise comparison, ΔatlA long chains were more efficiently phagocytosed (P = 0.0024 over 7 biological replicates) (Fig 7D). Collectively, our results indicate that formation of short chains by E. faecalis is a critical property enabling this bacterium to evade phagocytosis during pathogenesis.

Discussion
The formation of diplococci and short chains is a distinctive property of E. faecalis that was originally reported over a century ago [22]. This typical morphology results from the separation of daughter cells by the N-acetylglucosaminidase AtlA [8]. Here, we show that multiple mechanisms are in place to control septum cleavage by AtlA. We further demonstrate that the formation of diplococci and short chains is crucial for the virulence of this opportunistic pathogen. Cells with impaired cell separation are more prone to phagocytosis and can no longer cause infection. We propose that the control of cell chain length is a novel virulence factor that links cell division and pathogenesis in E. faecalis.
Using a specific assay to measure septum cleavage by flow cytometry, we showed that two post-translational modifications, both occurring on the N-terminal domain of AtlA contribute to down-regulate the activity of this enzyme. Our hypothesis is that the N-terminal domain, predicted to be disordered (http://prdos.hgc.jp), sterically hinders the catalytic activity of AtlA. By extension, glycosylation of the N-terminal domain is expected to further impair PG recognition and cleavage by the catalytic domain. Truncation of the N-terminal domain by extracellular proteases thus ensures optimal activity of the enzyme once it has reached its substrate at the cell surface. N-terminal truncation of the N-terminal domain occurs during growth and can be detected by zymogram [8,10]. It is tempting to assume that the proteolytic cleavage of the AtlA N-terminal domain is primarily mediated by the metalloprotease GelE, which has been associated with formation of short chains in E. faecalis [23]. However, the diplococcal using a dedicated Fiji plugin. The ratio of GFP fluorescence area outside to inside phagocytes was used to quantify bacterial uptake. Phagocytosis was significantly higher for long chains (ΔatlA) when compared to their sonicated counterparts (ΔatlA S ) (**P = 0.0098) or the wild-type cells (*P = 0.0438). No difference in uptake was found between short chains corresponding to the wild-type or sonicated ΔatlA mutant (NS, P>0.05). Representative images of phagocytes (magenta) following infection with ΔatlA, sonicated ΔatlA S and wild-type OG1RF cells shown. Phagocytes labeled with Lplastin appear in magenta, GFP-producing bacteria in green. Scale bar is 20μm. C. Survival of phagocyte-depleted zebrafish larvae (n>20) following injection with E. faecalis OG1RF (WT) or ΔatlA. D. Pairwise comparisons of phagocytosis indexes corresponding to E. faecalis OG1RF and ΔatlA uptake by human monocyte-derived macrophages (MDM). Statistical significance was determined by paired t-test; ΔatlA cells were more efficiently phagocytosed by MDM than WT cells (**P = 0.0024; n = 7). state of E. faecalis is not restricted to strains producing GelE, indicating that other proteases can process AtlA. An example is E. faecalis JH2-2: although this strain is deficient for the production of GelE, it forms diplococci and very short chains.
Interestingly, the genes essential for AtlA glycosylation were previously shown to catalyze the production of diglucosyl-diacylglycerol [24]. This implies that (an)other glycosyl transferase(s) mediate(s) the direct glycosylation of AtlA. Further investigations are required to identify the corresponding enzyme(s). This task appears relatively difficult given the functional redundancy of glycosyl transferases. The E. faecalis V583 genome encodes 15 putative enzymes that could be responsible for AtlA glycosylation. Our results showed that in the absence of AtlA, the gtfAB deletion has no impact on septum cleavage. Altough we cannot formally rule out that glycosylation of other surface proteins can modulate cell separation, this effect (if any) is limited. The identification and characterization of GtfAB substrates awaits further analysis. Another open question deals with the degree of AtlA glycosylation. It is possible that the extent of AtlA glycosylation varies during growth or in response to environmental cues.
Recent studies have explored the impact of PG structure on substrate recognition and cleavage by the catalytic domain of PG hydrolases. The N-acetylglucosaminidase LytB, the S. pneumoniae functional homolog of E. faecalis AtlA, requires fully acetylated GlcNAc moieties for cleavage. A substrate-assisted catalytic mechanism involving anchimeric assistance by the C2-acetamido group of the GlcNAc moiety is likely to underpin this requirement [25]. Another example is the pneumococcal autolysin LytA, in which several amino acids in the vicinity of the catalytic residues contribute to positioning of the substrate in the catalytic cleft so that the scissile bond is at an optimal distance from the catalytic residue [26]. Our work revealed that the N-acetylglucosaminidase domain of AtlA is not essential for septum cleavage. However, swapping the AtlA catalytic domain for another domain results in a lower septum cleavage activity. One possibility is that in E. faecalis, AtlA has evolved to optimally recognize and cleave the local PG structure at the septum. Recent work in E. coli and B. subtilis suggested that septal PG is enriched in "denuded" glycan strands resulting from N-acetylmuramoyl-Lalanine amidase activity [27]. Whether denuded glycan strands represent an optimal substrate for AtlA remains to be tested. It is expected that the lack of peptide stems will increase the binding activity of the LysM domain [28]. Thus, measurements of the N-acetylglucosaminidase activity of AtlA against glycan chains and chains substituted by peptide stems should be carried out with the catalytic domain in isolation to uncouple binding of the AtlA enzyme to its substrate from catalysis.
Our previous work revealed that AtlA LysM motifs can fold independently and do not interact, thus suggesting that they behave as "beads on a string" [28]. This model implied that instead of forming a quaternary structure, LysM repeats bind PG in a cooperative manner. Here, we showed that septum cleavage increases with the number of LysM motifs, with the formation of diplococci requiring the presence of all six repeats. The model strain V583 encodes twelve proteins with LysM domains; four contain two repeats, seven contain a single repeat, AtlA being the only one with six repeats. Bearing in mind that the formation of diplococci or short chains by E. faecalis is critical for virulence, our results suggest that the lifestyle of this organism as a commensal has favored the emergence of multimodular LysM domains in AtlA.
During infection, the size of bacterial cells has a major impact on recognition by the immune system. One example is the cording morphology of mycobacteria, which correspond to snake-like structures formed by the end-to-end and side-to-side aggregation of bacilli [29]. The formation of large bacterial aggregates impairs phagocytosis and is required for virulence [30]. Another well-documented strategy to escape host immunity is the formation of filaments after invasion of epithelial cells by uropathogenic E. coli (UPEC) [31].
Inhibition of septation enhances resistance to phagocytosis and increase survival rates of UPEC and other pathogens in the host [32]. The exact mechanism by which filamentation inhibits uptake and killing by phagocytes is unclear. In vitro experiments using anisotropic polystyrene particles and alveolar macrophages revealed that the point of contact between phagocytes and particles is critical for phagocytosis initiation [33]. It has therefore been proposed that the increased cell length in filamentous bacteria reduces the probability of macrophages to encountering the cell poles that stimulate the formation of the phagocytic cup [34]. This early step in the uptake by immune cells appears to be a limiting factor, the internalization speed of filaments itself being similar to that of smaller particles [34]. In contrast, minimization of bacterial cell chains has been described as a strategy to overcome host immunity in Streptococcus pneumoniae [35]. The effect of cell chain length in S. pneumoniae involves subversion of complement-mediated opsonophagocytosis. Interestingly, our assay using monocyte-derived macrophages indicated that in the absence of complement, cell chains can be recognized and readily engulfed by phagocytes. Unlike UPEC filaments, the long chains of enterococci are pretty flexible and often form turns (S5B Fig). This is expected to generate a contact point with phagocytes that will favor cytoskeleton remodeling to form the phagocytic cup [34]. Impaired septum cleavage is primarily expected to restrict the capacity of the bacteria to disseminate and multiply in the host and is not expected to have any impact on clearance by phagocytes. In the context of a systemic infection in zebrafish larvae, diplococci are circulating in the bloodstream. The formation of long chains limits the dissemination of the bacteria, hence increasing their probability of encountering immune cells. We propose that the sequestration of E. faecalis inside the phagocytes prevents cell multiplication and release of the metalloprotease GelE that is essential to cause tissue damage and host death [17].
E. faecalis is a common nosocomial pathogen associated with a wide range of infections that can be life-threatening. It was recently shown to promote the growth of other microorganisms during polymicrobial infections [36]. This study suggests that targeting the enzymatic activity of AtlA, the autolysin dedicated to septum cleavage, represents a novel therapeutic strategy to eradicate E. faecalis.

Ethics statement
Monocyte-derived macrophages (MDM) were isolated from whole blood from healthy volunteers at the Sheffield Royal Hallamshire hospital with written informed consent prior to inclusion in the study, as approved by the South Sheffield Research Ethics Committee (07/Q2305/7) [37]. All samples were anonymised. Animal work was carried out according to guidelines and legislation set out in UK law in the Animals (Scientific Procedures) Act 1986 under Project License PPL 40/3574. Ethical approval was granted by the University of Sheffield Local Ethical Review Panel.

Bacterial strains, plasmids and growth conditions
Bacterial strains and plasmids used in this study are described in Table 1. E. coli was grown at 37˚C in Brain Heart Infusion (BHI) broth or agar 1.5% (w/v) supplemented with 200ug/ml erythromycin (for pGhost9 derivatives) and 100ug/ml ampicillin (pET derivatives). E. faecalis strains were grown in BHI broth or agar at 37˚C, unless otherwise stated. When necessary, the medium was supplemented with 30 μg/ml erythromycin.

Zebrafish strains and maintenance
London wild type (LWT) zebrafish [38] were provided by the aquarium facility at the University of Sheffield. Embryos were maintained in E3 medium at 28˚C according to standard procedures previously described [39]. Phagocyte-depleted embryos were obtained following injection of phosphorodiamidate morpholino oligomers against pu.1 as previously described [40].

Microinjections of E. faecalis in zebrafish embryos
Cells were grown to mid-exponential phase (OD 600~0 .3) and harvested by centrifugation (5,000 x g for 10 min at room temperature). Bacteria were resuspended in filtered phosphate buffer saline (150 mM Na 2 HPO 4 , 20 mM KH 2 PO 4 , 150 mM NaCl [pH 7.5], PBS) and transferred to microcapillary pipettes. Embryos at 30 hours post fertilization (hpf) were anaesthetized, dechorionated, embedded in 3% (w/v) methylcellulose and injected individually with 2nl of a cell suspension corresponding to ca. 1,000 cells as previously described [40]. The number of cells injected was checked before and after each series of injections with a given strain. Zebrafish embryos were monitored at regular intervals until 90 h post infection (hpi).

Immunostaining of phagocytes in zebrafish
Zebrafish larvae were fixed 1.5h post infection with 4% paraformaldehyde (m/v) at 4˚C overnight and washed four times in PBS supplemented with 0.4% (v/v) TritonX 100 and 1% (v/v) DMSO (PBS-TxD). Samples were blocked in 5% (v/v) sheep serum in PBS-TxD for 1h at room temperature followed by one wash in PBS-TxD. Cells were incubated at 4˚C overnight with primary antibodies (anti-L-plastin 1:400, gift from Paul Martin, University of Bristol). After four washes in blocking solution, embryos were incubated with secondary antibodies (goat anti-rabbit IgG, Alexa Fluor 647 conjugate, Life Technologies) for 2h at room temperature. Larvae were washed four times with PBS-TxD and fixed again with 4% paraformaldehyde (m/v) for 30 min at RT. Immunolabelled embryos were mounted using 0.8% (m/v) agarose in E3 medium and imaged with a confocal microscope.

Imaging of infected larvae by confocal microscopy and quantification of uptake by phagocytes
Immunolabelled embryos were immersed in 0.8% (w/v) low melting point agarose in E3 medium and mounted flat on FluoroDish™ (World Precision Instruments Inc.). Images were collected using a DMi8 confocal microscope (Leica). Image acquisition was performed with the Volocity software and the images were processed with ImageJ 1.49v software. Bacterial phagocytosis was quantified using an ImageJ custom script called Fish Analysis v5 which can be obtained from http://sites.imagej.net/Willemsejj/ or via ImageJ updater. All bacteria were identified based on their (GFP, Channel 1) fluorescence. Subsequently, the fluorescence intensity of the phagocytes (Alexa 647, Channel 2) surrounding the phagocytosed bacteria was measured. The phagocytosed bacteria had high fluorescence intensity of Channel 2 and the cut off of 2 was used to discriminate the phagocytosed from non-phagocytosed bacteria. The area of phagocytosed bacteria was compared with the area of non-phagocytosed bacteria and their ratio was calculated.

Macrophage isolation and culture
Monocyte-derived macrophages (MDM) were isolated from whole blood from healthy volunteers. Peripheral blood mononuclear cells were isolated by Ficoll Plaque (GE Healthcare) density centrifugation. To differentiate PBMC into monocyte-MDM 2×10 6 PBMC/ml were plated in RPMI 1640 media (Lonza) with 2 mmol/l L-glutamine (Gibco BRL) containing 10% human AB serum (First Link (UK) LTD) in 24-well plates (Costar). After 24h, non-adherent cells were removed, and adherent cells were cultured in RPMI with 10% heat-treated fetal bovine serum (FBS; Lonza) in 5% CO 2 at 37˚C to give a final concentration of approximately 2×10 5 MDM/ml at day 14 [37].
In vitro phagocytosis assay E. faecalis strains were grown to OD 600 = 0.6 and stored in frozen aliquots at -80˚C. Viable counts were determined upon thawing and used to calculate volumes necessary to give desired multiplicity of infection (MOI). Bacteria were labelled with pHrodo dye (pHrodo Red, succinimidyl ester, Invitrogen) as previously described [41]. Briefly, bacteria were washed in phosphate-buffered saline before being incubated with 10.2 μM pHrodo in 0.1M sodium bicarbonate pH8.3 for 30min at 37˚C protected from light. Excess dye was washed off before MDM were challenged with pHrodo labelled bacteria at MOI = 100 for 4 hours at 37˚C. Cells were then fixed in 2% paraformaldehyde and fluorescence (Ex/Em 560/585nm) measured on a Varioskan Flash multimode reader (Thermo Scientific). Relative fluorescence values (RFU) of cell only wells were subtracted from readings to control for autofluorescence.

Construction of pGhost derivatives for allele replacement
A similar strategy was followed to construct all plasmids for allele replacement except pGABhis, for which the whole insert was synthesized and cloned into pGhost9 using XhoI and EcoRI restriction sites. For pGAAhis, pGBBhis, pGDN, pGtfAB, pGAtlA1, pGAtlA1-2, pGAtlA1-3, pGAtlA1-4, pGAtlA1-5, two homology regions were amplified and fused by overlap extension using PCR [42]. A 5' homology region (referred to as H1) was amplified with oligonucleotides H11 (sense) and H12 (antisense). It was fused to a 3' homology region (referred to as H2) amplified with oligonucleotides H21 (sense) and H22 (antisense). The assembled PCR fragment flanked by two restriction sites was digested and cloned into pGhost9 similarly digested. Oligonucleotide sequences and restriction sites used for cloning are described in S1 Table. For pGAtlA Ã , three homology regions (H1, H2, H3) were fused by overlap extension. The resulting plasmid contains a catalytic domain flanked by NcoI and BamHI sites. The NcoI-BamHI fragment encoding the N-acetylglucosaminidase activity of AtlA was swapped for NcoI-BamHI fragments encoding catalytic domains with distinct catalytic activities to produce pGAtlA-Cse, pGAtlA-AtlB and pGAtlA-Ami generated by PCR (see Supplemental Experimental Procedures for oligonucleotide sequences). The sequences of the chimeric proteins encoded by these plasmids are described in S3 Fig.

Construction of E. faecalis mutants
Isogenic derivatives of E. faecalis JH2-2 were constructed by allele exchange using the procedure previously described [8]. Briefly, pGhost derivatives were electroporated into JH2-2 and transformants were selected at a permissive temperature (28˚C) on BHI plates with erythromycin. To induce single crossover recombination, transformants were grown at a non-permissive temperature (42˚C) in the presence of erythromycin. The second recombination event leading to plasmid excision was obtained after 5 serial subcultures at 28˚C without erythromycin. The last overnight subculture was plated at 42˚C without erythromycin. A clone harboring a double crossover mutation was identified by PCR and Southern blot hybridization.
Construction of pET derivatives for protein expression in E. coli pET2818 was used as an expression vector to produce C-terminally His-tagged recombinant proteins. To construct pET-AtlA TEV , a cleavage site recognized by the Tobacco Etched Virus (TEV) protease was introduced by PCR by fusing two amplified products (named H1 and H2). For pET-AtlAB and pET-AtlBA, a DNA fragment encoding the N-terminal domain of AtlA or AtlB (referred to as H1) was fused to a DNA fragment encoding the LysM domain of AtlA or AtlB (referred to as H2). The resulting fragments were cut by NcoI and BamHI and cloned into pET2818 that had been similarly digested. Specific oligonucleotides used for each construct are described in S1 Production and purification of his-tagged recombinant proteins produced in E. coli E. coli BL21(DE3) cells harboring pET-derivatives were grown to an optical density at 600 nm (OD 600 ) of 0.7 and production of recombinant proteins was induced by addition of 1 mM isopropyl-β-D-thiogalactopyranoside. The cells were harvested 4h after induction, resuspended in buffer A (50 mM Tris-HCl [pH 8.0] containing 300 mM NaCl) and sonicated (5 times 30 sec at 20% output using a Branson Sonifier 450). Soluble proteins were recovered after centrifugation (45,000 x g, 20 min at 4˚C), loaded onto Ni 2+ -nitrilotriacetate agarose resin (Qiagen GmbH, Hilden, Germany), washed with 10 mM imidazole in buffer A and eluted with 300 mM imidazole in buffer A. Recombinant his-tagged proteins were further purified by size exclusion chromatography on a Superdex75 HR 26/60 column (Amersham biosciences, Uppsala, Sweden) equilibrated with PBS. The fractions were analyzed by SDS-PAGE and pooled. Protein concentration was estimated by measuring the absorbance, using a theoretical extinction coefficient at 280 nm (http://www.expasy.org). Proteins were kept frozen at -80˚C in PBS supplemented with 25% glycerol. AtlB stocks were available from previous studies [8].

Protein preparation from E. faecalis cultures
Proteins from supernatants were prepared from exponentially growing cultures (OD 600~0 .4). Supernatants were precipitated by addition of TCA (10% v/v final). After 10 min on ice, proteins were recovered by centrifugation (15,000 x g, 10 min at room temperature), washed in 100% acetone, dried and resuspended in SDS-PAGE loading buffer (1ml/equivalent OD 600 = 50).
For the detection of His-tagged AtlA and AtlB produced under the control of the atlA promoter, proteins were prepared from cultures in late exponential phase (OD 600~1 ). One ml of culture (containing both cells and supernatant) was transferred to a tube containing 250μL of glass beads (100μm in diameter, Sigma). Cells were mechanically disrupted using a FastPrep device (six cycles of 40 sec at maximum speed with 5 min pauses between cycles). Loading buffer was added to the protein samples and equivalents of 40, 20 and 10 μl of the cultures were separated on a 10% SDS-PAGE. ΔatlA JH2-2 mutant harboring an in-frame deletion of atlA [8] Western blot detection of AtlA and AtlB Proteins were transferred to a nitrocellulose membrane. After a blocking step for 1h at room temperature in Tris buffer saline (TBS, 10mM Tris-HCl pH7.4, 150mM NaCl) supplemented with tween-20 (0.025%, v/v) and milk (2%, m/v), the membrane was incubated with rabbit polyclonal anti-AtlA antibodies raised against the catalytic domain of AtlA (1:1,000 dilution) or polyclonal anti-His antibodies (1:2,000 dilution; Ebioscience). Proteins were detected using goat polyclonal anti-rabbit antibodies conjugated to horseradish peroxidase (Sigma) at a dilution of 1:20,000 and clarity Western ECL Blotting Substrate (BioRad).

Detection of AtlA activity by zymogram
Proteins from the supernatant were separated on a 10% SDS-PAGE containing Micrococcus luteus autoclaved cells as a substrate (final OD 600 = 2). After electrophoresis the gel was rinsed in distilled water and proteins were renatured at 37˚C in a buffer containing 50mM Tris-HCl (pH7.5) and 0.1% (v/v TritonX-100).

Flow cytometry analysis of cell chain length
Cells were grown overnight without agitation at 37˚C. Cells were diluted 1:100 into fresh broth (OD 600~0 .02) and grown in standing cultures to mid-exponential phase (OD 600~0 .2 to 0.4). Bacteria were diluted 1:100 in phosphate buffer saline filtered on a 0.22μm pore size membrane (Millex-GV syringe filter unit, Millipore) to eliminate salt crystals which could interfere with measurements and analyzed by flow cytometry using Millipore Guava Easy Cyte system. Light scatter data were obtained with logarithmic amplifiers for 20,000 events. To measure the septum cleavage activity of recombinant proteins, the OG1RF ΔatlA mutant was grown to exponential phase (OD 600 = 0.2), collected by centrifugation, and bacterial chains were resuspended in filtered PBS containing various concentrations of recombinant proteins. Cell size distribution was determined by flow cytometry after 15 min of incubation at 37˚C. Relative logs of forward scattered light values (FS log) were collected for 5,000 events and expressed as a percentage of the control strain incubated in the absence of enzyme.

Light and fluorescent microscopy analysis of bacteria
Cells were grown to mid-exponential phase (final OD 600~0 .3) and fixed with 1.6% paraformaldehyde in PBS for 30min at RT. After fixation, bacteria were washed twice in distilled water and mounted onto poly-L-lysine coated slides and imaged using a DeltaVision deconvolution microscope equipped with an UplanSApo 100x oil (NA 1.4) objective and a Photometrics Coolsnap HQ CCD camera. ImageJ software was used to optimize contrast and to count the numbers of cells per chain.

Statistical analyses
All experiments reported in this study correspond to at least three biological replicates. Statistical analyses were performed using GraphPad Prism version 6.0e. Comparisons between survival curves were made using the log rank (Mantel-Cox) test. Median FSC values were compared using a two-tailed, unpaired Student's t test with Welch's correction. Comparison of OG1RF and ΔatlA derivative uptake in vitro by MDM was carried out using a paired Student's t test. Comparison of uptake by zebrafish macrophages was carried out using an unpaired non-parametric Dunn's multiple comparisons test. The number of cells per chain was compared using a non-parametric Mann-Whitney U test. Statistical significance was assumed at P values below 0.05.