Ras GTPase-Like Protein MglA, a Controller of Bacterial Social-Motility in Myxobacteria, Has Evolved to Control Bacterial Predation by Bdellovibrio

Bdellovibrio bacteriovorus invade Gram-negative bacteria in a predatory process requiring Type IV pili (T4P) at a single invasive pole, and also glide on surfaces to locate prey. Ras-like G-protein MglA, working with MglB and RomR in the deltaproteobacterium Myxococcus xanthus, regulates adventurous gliding and T4P-mediated social motility at both M. xanthus cell poles. Our bioinformatic analyses suggested that the GTPase activating protein (GAP)-encoding gene mglB was lost in Bdellovibrio, but critical residues for MglABd GTP-binding are conserved. Deletion of mglABd abolished prey-invasion, but not gliding, and reduced T4P formation. MglABd interacted with a previously uncharacterised tetratricopeptide repeat (TPR) domain protein Bd2492, which we show localises at the single invasive pole and is required for predation. Bd2492 and RomR also interacted with cyclic-di-GMP-binding receptor CdgA, required for rapid prey-invasion. Bd2492, RomRBd and CdgA localize to the invasive pole and may facilitate MglA-docking. Bd2492 was encoded from an operon encoding a TamAB-like secretion system. The TamA protein and RomR were found, by gene deletion tests, to be essential for viability in both predatory and non-predatory modes. Control proteins, which regulate bipolar T4P-mediated social motility in swarming groups of deltaproteobacteria, have adapted in evolution to regulate the anti-social process of unipolar prey-invasion in the “lone-hunter” Bdellovibrio. Thus GTP-binding proteins and cyclic-di-GMP inputs combine at a regulatory hub, turning on prey-invasion and allowing invasion and killing of bacterial pathogens and consequent predatory growth of Bdellovibrio.


Introduction
Bdellovibrio bacteriovorus is a small, predatory deltaproteobacterium which invades other Gram-negative bacteria wherein it replicates. Bdellovibrio can encounter their prey by fast motility, driven by rotation of a single flagellum in liquid environments [1,2], or by slow gliding motility on solid surfaces [3], but do not show social-or S-motility a process that is shown by other deltaproteobacteria (discussed below).
In Bdellovibrio invasion into the prey cell periplasm requires T4P, thus pilus-minus cells are incapable of host/prey-dependent (HD) growth and must be cultivated on artificial media as HIhost/prey-independent -cells [4,5]. In flagellate HD Bdellovibrio the T4P are at the non-flagellar pole and prey-invasion occurs only from that anterior pole. On surfaces a flagellum is not present and the Bdellovibrio glide bidirectionally. Both HD and HI Bdellovibrio can glide and invade prey on surfaces. Our study began by examining the genetics of surface motility control in Bdellovibrio. This work led us to find that proteins known for surface motility control in a second deltaproteobacterium, Myxococcus xanthus, have evolved to control predatory invasion of bacteria by Bdellovibrio.
Regulation of surface motility in the deltaproteobacterium M. xanthus (which is always non-flagellate), has been well characterised by pioneering work of the Søgaard-Andersen [6], Mignot [7], Zusman [8], Hartzell [9] and Kaiser [10] groups for its two types of bidirectional surface motility. These are social (S)-motility, swarming movement of streams of cells using retraction of T4P at alternate poles of the cells; and adventurous (A)-motility, characterised by the movement of individual cells on a surface. A-motility (or gliding), is thought to be powered by cell envelopespanning motor-protein complexes, [11,12], though the precise mechanism of movement is still being revealed [13][14][15]. In M. xanthus, T4P localize to one pole at a time. Occasionally, M. xanthus cells reverse direction; this involves a switch in the polarity of the two motility systems, including a switch in the pole at which T4P assembly occurs. Thus, M. xanthus cells can assemble T4P at both poles but at any one time, T4P are found only at one pole [16].
Recent data suggest that the four putative gliding motor-gene operons in the B. bacteriovorus HD100 genome are evolutionarily linked to those A-motility gene clusters in Myxococcus [17], with subtle distinctive absences and additions likely reflecting Bdellovibrio morphology and gliding differences.
Bdellovibrio exhibits A-motility on surfaces in a gliding process that does not use T4P [3]. In this gliding, A-motility, individual Bdellovibrio cells move bidirectionally, cells can follow each other along previous paths and reversals of individual cells and reorientations are seen. Gliding may be a particularly important mechanism by which Bdellovibrio explores biofilms and locates bacteria to prey upon [3,18]. It is critical for HD Bdellovibrio to be able to explore or leave solid surfaces by gliding (when its flagellum cannot operate). Unlike other non-predatory bacteria, Bdellovibrio HD cells cannot replicate outside prey without acquiring ''HI mutations'' to do so [19,20], thus without surface motility they could be trapped and starve. B. bacteriovorus gliding motility is slow, with cells moving, on average, 16 mm hr 21 [3] compared to the 24-36 mm hr 21 of Myxococcus [21]. Both B. bacteriovorus and M. xanthus show reversals in gliding direction. In Myxococcus, reversals during surface motility are known, from the work of the Søgaard-Andersen and Mignot labs, to be regulated by a Ras-like GTPase, MglA, which polarises the cell during gliding [6,7], and GTPaseactivating protein (GAP) protein MglB, which activates the GTPase activity of MglA to inhibit cellular reversals [6,7]. MglA is important for activation of both the A-and S-motility ''engines'' (S motility engines are T4P), at the alternating leading pole, during bidirectional movements [6,7]. In the absence of MglA, Myxococcus is both A-and S-non-motile. This means that MglA in M. xanthus, in conjunction with interacting partner RomR, regulates the localization/pole-switching activity of both T4P and gliding engine component proteins, in this bipolar bacterium. Chemotactic signals via the Frz system control cellular reversals in M. xanthus [22] via the RomR response regulator; RomR receives signals from the chemosensory Frz system and this modulates MglA activity [23,24]. Although romR is conserved in Bdellovibrio, the genes encoding the Frz apparatus are not.
Bdellovibrio gliding is controlled by the bacterial secondary messenger cyclic-di-GMP. A diguanylyl cyclase (dgcA) mutation abolishes gliding, rendering Bdellovibrio cells unable to glide out of a consumed prey cell bdelloplast on a surface, even 2 hours after making lytic pores in it [25].The c-di-GMP receptor CdgA (GVNEF -a degenerate GGDEF protein) was found to be present at the predatory pole of B. bacteriovorus and deletion of cdgA slowed prey-invasion significantly, showing a link between c-di-GMP signalling and predation [25].
Whilst the B. bacteriovorus HD100 genome encodes MglA (Bd3734; accession: NP_970444.1), it does not encode an MglB homologue [23]. This report caused us to ask how bipolar switching might be achieved during Bdellovibrio gliding on surfaces; and whether the non-equivalent poles of the monoflagellate Bdellovibrio in liquids might correlate with an alternative role for MglA Bd . Here we show that MglA Bd is required for predatory invasion, as well as being associated with changes in gliding reversal behaviour in B. bacteriovorus, but is not required for gliding motility per se. This activity of MglA Bd occurs without an MglB partner, but in a cell with a RomR Bd homologue. Both of these latter proteins are important to the control of bipolar motility in Myxobacteria. However we show that RomR Bd has an essential role for growth in Bdellovibrio. We also report a previously undescribed interacting protein partner of MglA, and show that MglA Bd and RomR Bd interact with this tetratricopeptide repeat protein (TPR) which is also required for predation. TPR is expressed from an operon that encodes a TamAB transport system and again TamA was essential for growth. Implications of this for predation and the onset of predatory growth upon prey-invasion are discussed.
Whilst MglA Mx is involved in regulation of T4P-mediated social motility in M. xanthus, we show that MglA Bd is involved in Bdellovibrio in the control of pilus extrusion for the process of T4Pmediated invasion of prey cells at the single predatory pole. We show that a complex of proteins, additional to the T4P, is required at the 'biting' pole to organise the prey-entry machinery.

MglA is required for predatory invasion by B. bacteriovorus HD100
To investigate the role of MglA Bd , a deletion strategy was adopted screening for possible Bdellovibrio mutants in both prey/hostdependent (HD) and host-independent (HI) growth modes. All attempts to inactivate mglA in host/prey-dependent B. bacteriovorus HD100 were unsuccessful, despite screening many more cells than required to generate other Bdellovibrio deletion strains [26] (364 revertants obtained from second crossover events, but no deletion mutants, from three separate conjugations); suggesting that MglA Bd is essential for an aspect of the predatory life cycle.
Three host-independent (HI) DmglA Bd strains were obtained through sucrose-suicide counter-selection from a total of 76 screened. When challenged with prey, DmglA Bd HI B. bacteriovorus strains were unable to lyse E. coli in either a soft agar prey-lawn on the surface of YPSC plates, or in liquid culture ( Figure 1A). Introduction of wild type mglA Bd by in cis complementation method (as described previously [25]) restored predation ( Figure 1B) confirming that MglA Bd is essential for predatory growth.
The DmglA HI B. bacteriovorus strain could not reduce E. coli numbers in liquid culture, though this strain could still attach to the exterior of potential prey cells (Figure 2). A parallel assay showed that 43.5% of wild-type B. bacteriovorus HI cells attached to,

Author Summary
Bacterial cell polarity control is important for maintaining asymmetry of polar components such as flagella and pili. Bdellovibrio bacteriovorus is a predatory deltaproteobacterium which attaches to, and invades, other bacteria using Type IV pili (T4P) extruded from the specialised, invasive, non-flagellar pole of the cell. It was not known how that invasive pole is specified and regulated. Here we discover that a regulatory protein-hub, including Ras-GTPase-like protein MglA and cyclic-di-GMP receptor-protein CdgA, control prey-invasion. In the deltaproteobacterium, Myxococcus xanthus, MglA, with MglB and RomR, was found by others to regulate switching of T4P in social 'swarming' surface motility by swapping the pole at which T4P are found. In contrast, in B. bacteriovorus MglA regulates the process of prey-invasion and RomR, which is required for surface motility regulation in Myxococcus, is essential for growth and viability in Bdellovibrio. During evolution, B. bacteriovorus has lost mglB, possibly as T4P-pole-switching is not required; pili are only required at the invasive pole. A previously unidentified tetratricopeptide repeat (TPR) protein interacts with MglA and is essential for preyinvasion. This regulatory protein hub allows prey-invasion, likely integrating cyclic-di-GMP signals, pilus assembly and TamAB secretion in B. bacteriovorus.
or had entered, E. coli prey cells after 1 hour (Figure 2A) but no DmglA HI strain formed prey-bdelloplasts (Bdellovibrio cause the prey to round-up into 'bdelloplast' structures after invasion), even after 22 hours. Figure 2 also shows that both the DmglA HI and DpilA HI (Dbd1290, which is known to lack pili and is obligately host-independent [4]) could still attach to E. coli prey cells, albeit at a lower frequency. This suggests that pili are not a prerequisite for attachment, (although they are required for prey-invasion [4,5]), and suggests that the DmglA HI predatory defect is not due to the inability of the Bdellovibrio cell to attach to prey cells.  strain was assayed against predatory and non-predatory controls by the reduction of E. coli numbers over 48 hours. Three wild-type HI strains (HID13, HID26 and HID50) reduced E. coli numbers in liquid cultures by up to four logs (grey region shows known natural variation in predation rate between different wild-type HI isolates). The DmglA HI strain showed no reduction in E. coli numbers, comparable to a previously-studied, nonpredatory DpilA HI strain, and to E. coli with no added B. bacteriovorus. (B) Reintroduction of the mglA ORF in cis to the DmglA HI strain in plasmid pK18::mglA restored predatory growth. Error bars represent 1 SD from the mean (for predation-testing of Dbd2492 strain see Figure S5). doi:10.1371/journal.pgen.1004253.g001 The nature of the predatory defect of the DmglA HI strain was analysed further by microscopy, using a fluorescent E.coli S17-1::pMAL_p2-mCherry prey strain [27]. Addition of the DmglA HI strain to E.coli S17-1::pMAL_p2-mCherry and incubation for 22 hours demonstrated that although DmglA HI cells could attach to the outside of a prey cell, they could not invade to form bdelloplasts ( Figure 2B). A wild-type HI B. bacteriovorus strain (HID50) successfully invaded E. coli cells and killed them (as shown in Figure 1) and at the 22 hour stage was shown to have formed bdelloplasts from 26.3% of the remaining E. coli, compared to zero bdelloplasts for the DmglA HI strain. Thus the deletion of mglA Bd abolished a process required for prey-invasion.

The B. bacteriovorus mglA mutant is hypo-piliated
The DmglA HI strain showed a similar phenotype to that observed in a pilus-minus (DpilA) strain, which was known to be unable to invade prey cells [4]. We hypothesised that B. bacteriovorus DmglA might be defective in the synthesis or extrusion of pili, preventing prey cell invasion. This seemed plausible given that MglA regulates both the pole-switching of the A-motility and Type IV pilus-mediated S-motility systems in M. xanthus.
Transmission electron microscopy of HI Bdellovibrio cultures grown to an OD 600 of 0.2-0.3 showed that a wild type HI control had pili in 14.3% of cells, whilst DmglA HI had pili in only 2.3% of cells analysed (p = 0.02). These data suggested that MglA Bd regulates formation of pili; loss of mglA reduces the number of piliated cells. But, in contrast to the DpilA strain which completely lacks pilus fibres, the total inability of DmglA cells to invade, despite the presence of a low (but significant) frequency of piliated cells, suggests that these few pili present in the DmglA cells are not competent to facilitate invasion. This could be due to a defect in pilus retraction upon attachment to prey surfaces, or a requirement for another MglA-controlled factor to mediate invasion. Candidate MglA Bd -interacting proteins for invasive processes are discussed later.
MglA Bd controls gliding reversal frequency but not gliding per se Knowing that MglA Bd controls pilus-mediated bacterial invasion in B. bacteriovorus, but that in M. xanthus both pilus-mediated S-motility and gliding A-motility are MglA controlled, we used time-lapse microscopy to observe DmglA and wild-type B. bacteriovorus strains for gliding motility on 1% agarose/CaHEPES. Surface motility in B. bacteriovorus begins after a period of incubation on an agarose surface and allows exploration of environments for potential prey.
In contrast to recent studies in Myxococcus xanthus which showed that a DmglA Mx strain is non-motile on surfaces [7], and a mglA G21V strain displays hyper-reversals during A-motility [6], we found that Bdellovibrio DmglA cells showed sustained gliding runs on surfaces (Figure 3), indicating that MglA Bd is not absolutely required for Bdellovibrio cells to glide.
A Bdellovibrio strain with C-terminally His 8 -tagged MglA Bd , expressed from the endogenous bd3734 promoter in cis, with a plasmid promoter-driven wild-type copy of mglA Bd , could be grown predatorily, in contrast to the DmglA strain which was nonpredatory. In a previous study in M. xanthus, the presence of tagged MglA Mx protein in conjunction with wild-type MglA Mx allowed gliding to remain fully functional [7]. In contrast to the sustained gliding motility of the DmglA Bd strain ( Figure 3A), the predatory B. bacteriovorus HD100 MglA-His 8 showed increased reversals during gliding: on average 9.0 reversals hr 21 (n = 28), significantly more than wild-type HD100 cells with an average of 3.2 reversals hr 21 (n = 21) (p,0.001) ( Figure 3B,C). The same hyper-reversal phenotype was also observed in B. bacteriovorus HD100 MglA-mCherry cells (data not shown).
The P-loop region ( 19 GXXXXGKT 26 ) of MglA Mx was shown by Søgaard Andersen and co-workers to be important for GTP hydrolysis, and for MglA function [28], and substitutions in this region, such as G21V, were reported to decrease hydrolysis [6]. The P-loop region of MglA Bd contains a natural serine at residue 21; the corresponding G12S substitution in eukaryotic G protein Ras activates Ras protein [30], essentially locking the protein in a GTP-bound state, in the same way as a Ras G12V substitution. This suggests that MglA Bd exists in a permanently GTP-bound state. The G21-equivalent residue is a conserved glycine across 7 deltaproteobacterial genera ( Figure S1A) which all also have a conserved mglB gene, though in Bdellovibrio the equivalent residue is a serine.
The difference at residue 21 in the MglA Bd sequence suggested to us a reason why we did not observe conservation of the gene encoding MglB in Bdellovibrio, as the GAP activity of an MglB would likely be ineffective on a permanently GTP-bound MglA protein such as that suggested by the MglA Bd sequence with S at position 21. We thought that it might also explain the lack of a Bdellovibrio Frz system [23], which stimulates motility reversals in M. xanthus, as a mutation, causing MglA Mx G21V, bypasses the requirement for Frz for reversals in that deltaproteobacterium [6]. Thus we turned to examine the presence of mglB in the deltaproteobacterial relatives of Bdellovibrio. We also tested the conserved RomR Bd protein, while also looking for other proteins, specific to Bdellovibrio, with which MglA Bd might interact. In M. xanthus, RomR is found at both poles of the cell and interacts with both MglA and MglB to link the Frz system to regulate polarity control [23,24].

B. bacteriovorus has lost the mglB gene
The majority of sequenced deltaproteobacteria genomes contain both mglA and mglB, and these are often co-transcribed at the same locus, including in M. xanthus where the MglB MX protein has an important role in motility [6,7,31]. Although the mglA gene product in B. bacteriovorus HD100 shares extensive sequence similarity with other MglA proteins, there is no mglB homologue in the HD100 genome, despite neighbouring genes (dnaX, recR, mglA and a DUF149-encoding gene) showing conserved synteny to other deltaproteobacteria that do have an mglB. The closely related B. bacteriovorus Tiberius [32] also lacks an mglB homologue. The predatory, invasive, marine bacterium Bacteriovorax marinus is also closely related to B. bacteriovorus, although the Bdellovibrio and Bacteriovorax genera have diverged separately from Myxobacteria. A 16S rRNA phylogenetic tree of the deltaproteobacteria shows the ancestral lineage leading to Bdellovibrio and Bacteriovorax diverged from the ancestral lineage leading to the clade including Myxococcus xanthus [33] and in that divergent Bdellovibrio branch we detect mglB loss ( Figure S1A). We found that in B. marinus, which also has an mglA gene (BMS_0054), there is an adjacent putative mglB homologue (BMS_0053), both genes lying downstream of recR ( Figure S1B).
BMS_0053 shares only limited sequence similarity with other MglB Roadblock domain proteins (BMS_0053, 168 residues, shares 22% identity and 43% similarity (NEEDLE global alignment) with M. xanthus MglB protein, 159 residues, Figure  S1C). This highly divergent MglB homologue in Bacteriovorax is likely still functional, since no frameshift or nonsense mutations have arisen in the B. marinus lineage, and protein sequence length is conserved; however, its function is unclear. We are unable to test whether mglB is under positive selection (dN/dS.1) in Bacteriovorax because synonymous substitution rates are saturated for available sequence comparisons (dS.2). The Bacteriovorax MglA homologue is much more conserved (66% identity and 83% similarity to MglA Bd ) and may function in an analogous predatory role to that of B. bacteriovorus.

MglA Bd interacts with TPR-(tetratricopeptide repeat) domain protein Bd2492 in vivo
As MglA Bd had both similarities and differences to MglA Mx , we sought to identify proteins that interact with MglA homologue Bd3734 in B. bacteriovorus as we reasoned that these proteins might have a predatory role. We used a pull-down co-purification assay with proteins from the predatory B. bacteriovorus strain producing MglA Bd with a C-terminal His 8 tag from the endogenous mglA Bd promoter, mentioned above. For the co-purification assay, a hostindependent isolate of the MglA Bd -His 8 strain was used, as previous array data showed that mglA Bd transcription is upregulated in wild type HI cells, (which remain predatory but are longer than attack phase Bdellovibrio). Whole cell lysates of this HI strain were used in the assay, in which the bait His-tagged protein MglA Bd binding to TALON-NX cobalt-charged resin allowed interacting proteins to be identified ( Figure S2) that were not present in the control without the His-tag.
MglA Bd co-purified with Bd2492 (accession: NP_969302.1) ( Figure S2) -a B. bacteriovorus protein with a hypothetical annotation, with predicted tetratricopeptide repeat (TPR) domains typically involved in protein-protein interactions. Bands were excised from the gel and analysed by LC-MS/MS. Corresponding regions of the wild-type HID13 control lane were also analysed, and neither MglA Bd nor Bd2492 were found in these regions, suggesting that MglA Bd and Bd2492 (TPR Bd ) interact in vivo.
Confirmation that MglA Bd and TPR Bd proteins interact by bacterial two-hybrid and heterologous co-expression in E. coli The mglA ORF and bd2492 ORF were cloned into pUT18C and pKT25 vectors containing T18 and T25 fragments of adenylate cyclase, respectively [34]. The bacterial two-hybrid assay for MglA and Bd2492 showed a strong signal ( Figure S3A-B) suggesting that the two B. bacteriovorus proteins interact. This interaction was supported by the observation that MglA copurifies with His 6-tagged Bd2492 in nickel-affinity chromatography of E. coli lysates heterologously expressing these two proteins from plasmid pD2492N/3734 ( Figure S4). Gel filtration and SDS-PAGE of purified MglA and Bd2492-His 6 indicated that the MglA-Bd2492 complex has an Mw of approximately 63 kDa and exists predominantly as a heterodimeric complex of 1:1 stoichiometry (data not shown).
TPR Bd is required for predatory invasion by B. bacteriovorus HD100 As the B. bacteriovorus mglA mutant was non-predatory, we tested whether bd2492 (encoding TPR Bd ) was essential for predatory growth. All attempts to inactivate bd2492TPR in host-dependent B. bacteriovorus HD100 were unsuccessful (68 revertants obtained from second crossover events, but no deletion mutants). Two hostindependent (HI) Dbd2492 strains were obtained through sucrosesuicide counter-selection from a total of 10 screened. When challenged with prey, Dbd2492TPR HI strains were unable to lyse E. coli in liquid culture ( Figure S5). As with the DmglA HI strains, the Dbd2492TPR HI strains could still attach to E. coli prey cells (attachment assay; 26.6% of E. coli cells had attached Bdellovibrio), but could not invade to form bdelloplasts (invasion assay; 0/389 E. coli cells). The B. bacteriovorus Dbd2492TPR HI strain was still able to glide on 1% agarose CaHEPES (data not shown).
B. bacteriovorus genes bd2492, bd2494 and bd2495 are co-transcribed and syntenic in other deltaproteobacteria TPR gene bd2492 is co-transcribed with bd2494 and bd2495 ( Figure S6). The same gene synteny is also found in M. xanthus (MXAN_5763-5766) and B. marinus SJ (BMS_0137-140) ( Figure 5) where the gene encoding a TPR domain protein is followed by genes encoding homologues of Bd2494 and Bd2495. In M. xanthus, the genes encoding homologues of Bd2492 and Bd2494 (MXAN_5766 and MXAN_5764) are interrupted by a gene encoding a putative Sec system ATPase, MXAN_5765.
RomR Bd in B. bacteriovorus interacts with TPR Bd by BTH and both proteins are located at the prey-invasion pole As mentioned in the introduction, RomR Mx interacts with the MglA MX signalling system to regulate surface motility in response to Frz system signals [23,24], but the Frz system is not conserved in Bdellovibrio. We assessed the interaction of the RomR Bd (Bd2761; accession: NP_969553.1) with the MglA-interacting protein TPR Bd (Bd2492) by bacterial two-hybrid ( Figure S3A). RomR Bd shares homology with the REC domain and C-terminal region of RomR Mx , whilst the remainder of the protein is less well conserved ( Figure S7). RomR Bd and TPR Bd interact in the BTH assay ( Figure S3A, C). We found that RomR Bd and MglA Bd interacted weakly, but not significantly (p = 0.18) (Figure S3B-C).
Fluorescent tagging of RomR Bd and TPR Bd with C-terminal mCherry revealed that both proteins are localised at only one pole of the cell. Co-incubation with E. coli prey cells confirmed that both RomR Bd -mCherry and TPR Bd -mCherry are found at the anterior, prey-interaction pole of B. bacteriovorus cells (Figure 6). Fluorescent tagging of MglA Bd with mCherry typically showed cells with diffuse fluorescence localization in cells directly after applying to 1% agarose/CaHEPES (i.e. not gliding) ( Figure 6); 63% of HD100 MglA-mCherry Bdellovibrio had diffuse fluorescence, the remainder showing a unipolar focus (28.4%) or bipolar foci (8.6%).

RomR Bd and TPR Bd interact with invasion-pole protein CdgA, a degenerate GVNEF c-di-GMP binding protein which is required for rapid prey-invasion by B. bacteriovorus
We found earlier that the Bdellovibrio DmglA strain does not show a hyper-reversal or non-motility phenotype ( Figure 3). Thus, the regulation of MglA Bd localization in the control of gliding reversals (in the absence of MglB and Frz) is likely to employ an alternative signalling system to that of M. xanthus. Previous work suggested that this could be c-di-GMP as we have shown [25] that lack of GGDEF protein Bd0367 DgcA abolished gliding exit from bdelloplasts.
We had also had previously noted a link between a c-di-GMP binding protein and prey-invasion in Bdellovibrio [25]. Degenerate GGDEF (GVNEF) protein CdgA, Bd3125 (accession: NP_969891.1), is located at the prey invading pole of B. bacteriovorus and lack of this polar protein causes a very significant slowing of prey-invasion with bdelloplast formation taking 40-90 minutes compared to 30-40 minutes for wild type [25]. We concluded in that paper that ''CdgA organises processes at the Bdellovibrio ''nose'' that are crucial to rapid prey-invasion''. In our current study, we found that both RomR Bd and TPR Bd (though not MglA) interacted with CdgA in the bacterial two-hybrid assay ( Figure S3), supporting this idea. Whether RomR Bd has a role in the regulation of gliding motility will be the subject of a subsequent study, but our interaction data suggested a link between RomR Bd and predatory growth (as DcdgA was affected in predation [25]), so we tested for a romR Bd deletion strain.

RomR Bd is essential in B. bacteriovorus
Given the CdgA and TPR Bd interactions found at the B. bacteriovorus invasive pole, we speculated that RomR Bd would be required for prey-invasion. Attempts to delete romR Bd , both predatorily (HD) and host-independently (HI), were unsuccessful (HD 104; HI 120 revertants screened), suggesting that RomR Bd is required for both predatory and host-independent Bdellovibrio growth.
TamA Bd , encoded from the operon encoding TPR Bd is also essential in B. bacteriovorus As RomR Bd interacted, by BTH, with TPR Bd , encoded in an operon with the tamAB genes, we speculated that the TamAB complex would also be required for predatory growth. Attempts to delete tamA Bd also proved unsuccessful (HD 140; HI 97 revertants screened), suggesting that TamA Bd is also essential for both phases of Bdellovibrio growth.

Discussion
Here we report that B. bacteriovorus use homologues of adventurous/social motility-control proteins for the process of predatory invasion of other bacteria. Whilst non-invasive M. xanthus utilise the proteins MglA and MglB to control bipolar, bidirectional surface motility [6,7] in Bdellovibrio MglA Bd has evolved to function without an MglB homologue (the mglB gene is absent) to regulate prey entry at a single pole. There are three lines of evidence to suggest this: (1) The deletion of mglA Bd caused a non-prey-invasive phenotype ( Figure 1) and severely reduced pilus formation on the cell surface; (2) the natural substitution in MglA Bd of serine for glycine (Figure 4) at the position equivalent to residue 21 in MglA Mx suggests that MglA Bd exists in a permanently GTP-bound state, and is not involved in the GTPase cycle which is key to the alternate bi-polar switching of motility proteins in M. xanthus [6,7]; (3) RomR-mCherry is unipolar in B. bacteriovorus (Figure 6), in contrast to its asymmetric bipolar localization in Myxococcus, controlling MglA Mx positioning.

RomR Bd is localised only at the prey-invasion pole and has a different phenotype to RomR Mx
We had hypothesised that RomR Bd might be involved in regulating pole activity to control gliding motility. As RomR Bd was found at the predatory pole only, this suggested an alternative role. We could not detect a significant interaction between RomR Bd and MglA Bd by BTH, but we did detect a significant interaction with Bd2492 TPR protein ( Figure S3), which is also at the anterior pole (discussed later).
The RomR Bd location at the anterior pole of B. bacteriovorus puts it where prey-invading T4P are located. Lotte Søgaard-Andersen's group showed that an mglA Mx deletion mutant resulted in unipolar RomR Mx , with RomR Mx and T4P, (used in that bacterium for bipolar social motility), found at the same pole [37]. Sequenceand localization-differences between unipolar RomR Bd and MglA Bd (in the absence of an MglB) in B. bacteriovorus, versus those in M. xanthus (which has MglB), might explain why T4P are only found at the anterior Bdellovibrio pole where they control preyinvasion.
Deletion of romR Bd abolished Bdellovibrio growth in both HI and predatory conditions, but in M. xanthus romR is viable with abolition of gliding motility and reduction of T4P-dependent social motility [23,24]. Thus RomR Bd , which does show some sequence divergence from RomR Mx (Figure S7), could be reporting T4P activity and prey-invasion, at the anterior pole, back to initiate Bdellovibrio growth. It should be recalled that predatory ''attack phase'' Bdellovibrio do not replicate outside prey, but initiate replication when prey are entered [20].

Absence of MglB in Bdellovibrio is consistent with unipolar RomR Bd
Our BTH interaction data were too weak to prove a significant interaction between RomR Bd and MglA Bd . This could be interpreted to mean that RomR Bd transiently docks with MglA Bd when RomR Bd is complexed at the pole, that other partner proteins are required to contribute to this interaction, or that they do not interact, in contrast to published data for MglA Mx [23,24]. Our finding that RomR Bd is unipolar fits with evidence in M. xanthus that MglB MX is required for bipolar localization of RomR Mx [23] and the apparent loss of MglB from the preyinvasive Bdellovibrio lineage in evolution. The Bdellovibrio-like invasive B. marinus has a putative mglB gene, the product of which shows only limited sequence similarity to other MglB Roadblock domain proteins ( Figure S1). This mglB Bm gene is highly divergent from mglb Mx but likely still functional. It may be undergoing selection to evolve an alternative function, while the B. marinus mglA gene is maintained for a predatory role analogous to that in B. bacteriovorus.
MglA and MglB were shown to be conserved by Keilberg and co-workers in many deltaproteobacteria but also occur in some evolutionarily distant bacteria such as the green non sulphur bacteria, Acidobacteria and Deinococcus-Thermus group, Figure  S3 in Ref [23]. The authors calculated the following: out of a total of 70 species with (at least one) predicted MglA homologue 87% = 61/70 species have MglB and an MglA. Of the 9 without MglB, 4 bacteria had MglA G21 with no MglB; 5 had MglA A/ S21 with no MglB. Of these 9 with no MglB, only B. bacteriovorus and one other species, (a soil Acidobacterium named Candidatus koribacteria versatilis), have predicted RomR homologues. Thus bacteria with RomR and MglA and B may have interacting protein complexes that move between poles; but our study on B. bacteriovorus is the first to examine the situation in a bacterium where MglA and RomR are present but MglB is not.
As mentioned above, we detected an interaction with an additional protein that could contribute to the localization of MglA Bd and RomR Bd at the single prey-invasion pole of Bdellovibrio. This was with the unipolar tetratricopeptide repeat TPR protein, Bd2492 (TPR Bd ) shown using both His-tag pulldowns and BTH for MglA and BTH for RomR. TPR Bd could sequester either MglA Bd or RomR Bd at the prey-invasive pole, regulating their freedom to interact with each other, or promoting an interaction on the TPR Bd surface. Deletion of bd2492TPR Bd abolished prey-invasion in the same manner as DmglA Bd ( Figure  S5, Figure 1).
It was not possible to monitor localization of fluorescently tagged proteins informatively in the HI derivative strains of the non-predatory DmglA Bd and Dbd2492 mutants. This is because HI derivatives have pleomorphic cell morphotypes (HI cells naturally differ greatly in length and shape) [38], and indeed some long HI cells are predatory at both poles [25].

TPR gene bd2492 is in an operon with tamAB genes
The bd2492 gene is located upstream of, and is co-expressed ( Figure 5, Figure S6) in an operon with, gene bd2494, which encodes a transmembrane protein with a C-terminal DUF490 domain, homologous to the TamB component of the TamAB autotransporter-secretion system [39]. Bd2494 might dock with TPR Bd at the prey-invasive nose. The last gene in the operon (bd2495) encodes a 7-POTRA (polypeptide-transport-associated)domain, outer membrane protein (OMP) member of the Omp85 superfamily. The Omp85 protein family includes the BamA component of the BAM complex, known to receive and assemble beta barrel proteins during outer membrane growth [40]. The family also includes the TamA component of the TamAB complex, which aids autotransporter secretion [39]; and twoprotein secretion (TPS) proteins [41].
The TamA and TamB genes are typically adjacent in proteobacteria [39], suggesting that the adjacent B. bacteriovorus bd2494-2495 genes encode a TamAB-like transporter.
Thus our finding that MglA Bd and RomR Bd interact with a TPR protein ( Figure S3), encoded from the 59 gene of a tamAB-like operon, suggests that the Bd2494-2495 TamAB-like transport activity might be required for OMP/autotransporter proteins involved in predation. This may account for our observation that some pili are present on the DmglA mutant but that despite this, it does not invade due to an effect on TamAB-dependent predatory protein transport. Similarly, the Dbd2492 mutant was also nonpredatory ( Figure S5), but attached to prey. This suggests that either: TPR Bd and MglA Bd are important in the positioning of proteins (probably Bd2494-5 TamAB Bd ) at the predatory pole of the B. bacteriovorus cell to facilitate prey entry; or that binding of RomR Bd and MglA Bd to TPR Bd affects its activity, and that of the TamAB Bd complex, regulating predatory protein secretion.
Reinforcing our observation (mentioned above) that RomR Bd is essential, we found, by attempting to delete bd2495, that TamA Bd was also essential for both HD and HI growth of Bdellovibrio. This suggests that the activity of the TamAB complex (possibly involving a TPR-mediated interaction with RomR Bd ) is required for secretion of proteins required for prey-invasion and both predatory and HI growth. Potential candidates for TamAB export are proteins involved in the synthesis/secretion/maturation of extracellular polysaccharide (EPS) or polyelectrolytes; an earlier study proposed that RomR was responsible for stimulating polyelectrolyte secretion in M. xanthus [37]. We cannot yet define whether RomR Bd activates a TamAB dependent process that is essential for predatory and HI growth, or whether it reports on the activity of a TamAB complex, via its interaction with TPR Bd , to regulate Bdellovibrio growth. This will be the subject of a further extensive genetic study.

Conserved synteny of the TPR tamAB genes in Myxococcus
Although a TPR protein interaction with MglA Mx or RomR Mx has not been previously reported, the MXAN_5766 gene encoding a TPR domain protein, from a gene cluster with similar synteny to the B. bacteriovorus bd2492-2495 genes ( Figure 5), has previously been implicated in M. xanthus S-motility by transposon studies carried out by the Hartzell group [42]. The low percentages of TPR ORF similarity/identity between MXAN_5766 and Bd2492 could reflect the greatly different protein sizes and may indicate interactions with additional protein partners in M. xanthus. However, in M. xanthus, similar TPR interactions with RomR, MglA and TamAB-like proteins could play a role in bipolar motility control. Whether or not this is the case in M. xanthus, it is clear that TPR, and likely TamAB Bd , proteins play an important role in defining the single, active, predatory pole of Bdellovibrio.

A protein hub controlling predatory invasion
We propose a predatory regulatory 'hub' of proteins at the B. bacteriovorus prey-invasive pole (Figure 7), with the TamABassociated Bd2492-5 TPR Bd protein complex involved in the organisation/assembly of OMPs or autotransporters at the predatory pole. This is reflective of TamAB protein functions in other bacteria (discussed in [39]). Such protein secretion could facilitate predation directly or produce other extra-cellular compounds such as EPS or polyelectrolytes, as mentioned above, which contribute to predatory invasion. Predatory proteins could be secreted in outer membrane vesicles (OMVs); Sar and Arf GTPases (homologous to MglA) have functions in vesicle transport [43] and M. xanthus vesicles likely have an extra-cellular predatory role in the ''wolf-pack'' [44] hunting of M. xanthus [45]. Our studies show that the directed prey-invasion of Bdellovibrio requires a protein encoded by a tamAB operon, suggesting synergies in TamAB-mediated predation and cell interaction processes of B. bacteriovorus and M. xanthus which is worthy of further investigation. Regulatory protein hubs are reported to control pili and flagella in other bacteria [46].

Evolutionary comparisons in Myxococcus and Bdellovibrio
Considering evolutionary differences that led Bdellovibrio to prey-invasion via a single pole, we also suggest that the absence of mglB in B. bacteriovorus (and the high degree of divergence of this gene in B. marinus), is because MglB is no longer required for pole switching of pili: B. bacteriovorus pili are found at only one -the non-flagellar, prey-invasive pole [4]. This is also concordant with B. bacteriovorus cells being incapable of S-motility (which would require pole-switching of T4P) and instead using T4P at a single pole for prey-invasion. However, the absence of an MglB homologue does suggest that an alternative mechanism for regulating reversals during gliding motility is likely to exist.
The mechanism by which reduced incidence of pili or a change in their retraction state is caused, in the B. bacteriovorus DmglA Bd strain, remains to be determined. Capeness and coworkers have recently shown that regulation of Bdellovibrio pilus retraction status does correlate with prey-invasion [26]. Pilus retraction occurs through secretin PilQ [47], which is required for predation in B. bacteriovorus [18]. The OM-assembly of a pilus-biogenesis protein such as PilQ could be affected by the Bd2492-5 TamAB complex activity. Alternatively, OMPs required for secretion of EPS might be perturbed at the Bdellovibrio pole, preventing pilus retraction; EPS is required for pilus retraction in M. xanthus [48]. These considerations will be the subject of a subsequent study.
The MglA/RomR-TPR interactions reported in this paper may have evolved from ancient interactions common to ancestors of M. xanthus and Bdellovibrio, and are now used in B. bacteriovorus for preyinvasion control. They may also underlie the motility and ''wolfpack'' predation of Myxobacteria, but the function of the M. xanthus TPR protein homologue remains to be explored. Pioneering work by Mignot/Theodoly has shown that adhesion during gliding motility is mediated by slime deposition [14,15] on a solid surface and that gliding directionality is controlled by MglA Mx [6,7] and other interacting proteins. In nature gliding of M. xanthus may occur on top of prey bacterial biofilms and we hypothesise that the Bd2492-5 TamAB associated system may have a role in producing vesicles, not only for gliding, but to damage prey cells as part of the M. xanthus wolf-pack lytic process.

Cyclic-di-GMP signalling at the predation control hub
In M. xanthus, chemotactic phospho-transfer signalling, involving Frz proteins, governs the localization of soluble RomR Mx , MglA Mx and MglB Mx proteins to alternately activate or deactivate each cell pole for surface-motility directionality [23,24]. In B. bacteriovorus, we detected an interaction between RomR Bd and the CdgA GVNEF domain c-di-GMP binding protein ( Figure S3) which has been shown to affect prey entry [25]. There is no Frz system in Bdellovibrio [23] but our finding that CdgA binds RomR Bd ( Figure S3) suggests that this c-di-GMP signalling pathway could contribute to RomR Bd localization in the control of the prey-invasive pole. Further work is underway to define any signalling-link to RomR Bd and CdgA from our previous observations that c-di-GMP synthases control gliding motility, predation and the switch from predatory to host-independent growth [25].
The data we present here show how the ''phenotype space'' and function of B. bacteriovorus MglA has diverged from that in M. xanthus. MglA Bd functions in the control of unipolar prey-invasion: a critical process in the predatory lifecycle of B. bacteriovorus. Our present observations indicate (Figure 7) that MglA Bd , RomR Bd and the interacting TPR-domain protein TPR Bd and TamAB Bd complex act at a single pole in B. bacteriovorus to facilitate preyinvasion via a mechanism that has diverged from that which controls M. xanthus S-motility.

Bacteria, plasmids and primers
Bacterial strains and plasmids used are listed in Table S1. Primers used for gene manipulation or PCR amplification are listed in Table S2.

Deletion construction
Markerless deletion strains of mglA Bd and bd2492 (encoding TPR Bd ) were generated using a modified technique of that of the Pineiro lab [49], and as described previously [25]. Construction of each mutant is described in full in Text S1.

Fluorescent protein tagging
Fluorescent protein tags were generated as described previously [25] by cloning of a whole gene fused to mCherry at the 39 end. Construction of each tag is described fully in Text S1.

Fluorescent microscopy
To observe the fluorescence of B. bacteriovorus mCherry-tagged strains during attachment to E. coli prey cells, 1 ml of a B. bacteriovorus predatory culture (containing 2.5610 8 pfu ml 21 ) was concentrated 20-fold and added to a microcentrifuge tube containing 30 ml CaHEPES and 40 ml E. coli S17-1 pZMR100 (from a culture grown for 16 hours at 37uC 200 rpm in YT broth supplemented with Km 50 ) diluted to OD 600 2.0 in CaHEPES, before incubating at 29uC for 5 minutes to allow attachment to occur. Cells were immobilised on a 1% agarose/CaHEPES pad and images were taken on using a Nikon Eclipse E600 epifluorescence microscope with a 1006 objective lens and an hcRED filter (excitation 550 to 600 nm; emission 610 to 665 nm) with a Hamamatsu Orca ER camera. Images were analysed using Simple PCI software (version 5.3.1 Hamamatsu).

Host-independent predation, invasion and attachment assays
Procedures for attachment, invasion and predation assays of HI Bdellovibrio cells on E. coli prey are described in Text S1. 3 biological replicates were performed.
Gliding motility assay B. bacteriovorus gliding motility was observed on 1% agarose/ CaHEPES by timelapse microscopy as previously described [3]. Briefly, 1 ml of an predatory culture (containing 2.56 10 8 pfu ml 21 ) was concentrated 10-fold (HI cultures were not Figure 7. Model for B. bacteriovorus predatory-pole regulation during prey-invasion and its relationship to M. xanthus bipolar motility-control proteins. During prey-invasion; TamA Bd , RomR Bd and CdgA protein interactions occur (see Figure S2, S3 S4) at the single B. bacteriovorus pole. This could control localization of the TamA Bd -like OMP at the prey-interaction pole or activate it to receive, (via its POTRA domains), and secrete predatory outer membrane or autotransporter proteins. The action of this secretion via Bd2495 TamA is essential to both predatory and HI lifestyles, and RomR Bd , (which is also essential), may regulate or report the activity of the TamAB transport system, at the single predatory pole. Additional regulation of this activity could be influenced by c-di-GMP for which CdgA, (another hub protein that binds RomR and TPR Bd2492), is a receptor in Bdellovibrio. MglA Bd interacts with Bd2492 at the predatory pole but also is found more diffusely in the cell. MglA interactions may regulate prey entry via TPR Bd2492, as deletion of MglA or TPR Bd2492 abolishes prey-invasion but not prey attachment. MglA deletion in Bdellovibrio greatly reduces the level of Type IV pilus formation at the single anterior pole. In contrast Keilberg and co-workers showed that M. xanthus RomR and MglB localise bipolarly asymmetrically, while MglA typically localises at the leading cell pole, during surface movement, to regulate both A-and S-motility (M. xanthus after [23]). doi:10.1371/journal.pgen.1004253.g007 concentrated) and 8 ml was spotted onto the agarose pad. Measurements of gliding reversals were calculated after cells had been gliding for .1 hr.

Electron microscopy
To analyse percentages of piliated cells, each HI strain was back-diluted and grown to OD 600 0.1-0.5 in PY broth at 29uC 200 rpm. Cells were then stained with 2.0% phosphotungstic acid (PTA) on carbon formvar copper grids (Agar Scientific) and analysed for the presence/absence of a pilus structure, as described previously [26].

Bacterial two-hybrid and protein co-purification
Procedures for bacterial two-hybrid and protein co-purification are described in Text S1. Figure S1 Tree showing co-evolution of G21-encoding mglA with mglB versus lone mglA S21 in deltaproteobacteria. (A) A Maximum Likelihood phylogenetic tree of deltaproteobacteria small subunit rRNA gene sequences: the majority of these bacteria encode an MglA with a G21 residue -these also encode an MglB homologue. Bdellovibrio bacteriovorus and Bacteriovorax marinus diverge separately from these mglB-encoding deltaproteobacteria, including Myxococcus xanthus. The B. marinus genome encodes MglA G21 and a degenerate MglB; the B. bacteriovorus genome encodes MglA with an S21 residue, but no MglB homologue. Tree generated using Phylogeny.fr [50] and rooted with Shewanella onidensis;  Figure S3 Bacterial two-hybrid shows MglA and RomR interact with Bd2492; RomR and Bd2492 interact with CdgA. A bacterial two-hybrid (BTH) assay between Bd2492 and MglA produces a positive signal on spot tests (A); the interaction between pUT18C-MglA and pKT25-Bd2492 was confirmed by beta-galactosidase assay (C). A positive result was also obtained for a BTH interaction between RomR homologue Bd2761 and Bd2492 on spot tests (A); the interaction between pUT18C-RomR pKT25-Bd2492 was confirmed by beta-galactosidase assay (C). Both RomR and Bd2492 were found to interact with CdgA (Bd3125) by BTH (A). The interactions between pKT25 Bd3125 and pUT18C-RomR or pUT18C-Bd2492 were confirmed by beta-galactosidase assay (C). When MglA and RomR interactions were assayed with tags at either end of the proteins, one combination (pUT18C-RomR and pKNT25-MglA), indicated by an asterisk (2 independent transformants) reproducibly produced a positive result on spot tests suggesting these two proteins interact (B). This interaction could not be confirmed as significant by beta-galactosidase assay, suggesting there is no interaction (as detected by BTH) between RomR and MglA. Positive control (+) = pUT18-zip and pKT25zip and negative control (2) = pUT18C and pKT25. Error bars represent 1 SD from the mean.