Structural Determinants for Activity and Specificity of the Bacterial Toxin LlpA

Lectin-like bacteriotoxic proteins, identified in several plant-associated bacteria, are able to selectively kill closely related species, including several phytopathogens, such as Pseudomonas syringae and Xanthomonas species, but so far their mode of action remains unrevealed. The crystal structure of LlpABW, the prototype lectin-like bacteriocin from Pseudomonas putida, reveals an architecture of two monocot mannose-binding lectin (MMBL) domains and a C-terminal β-hairpin extension. The C-terminal MMBL domain (C-domain) adopts a fold very similar to MMBL domains from plant lectins and contains a binding site for mannose and oligomannosides. Mutational analysis indicates that an intact sugar-binding pocket in this domain is crucial for bactericidal activity. The N-terminal MMBL domain (N-domain) adopts the same fold but is structurally more divergent and lacks a functional mannose-binding site. Differential activity of engineered N/C-domain chimers derived from two LlpA homologues with different killing spectra, disclosed that the N-domain determines target specificity. Apparently this bacteriocin is assembled from two structurally similar domains that evolved separately towards dedicated functions in target recognition and bacteriotoxicity.


Introduction
In most natural settings, complex interactions occur among microorganisms, ranging from nutritional co-operation to warfare among competitors. Examples of such interplay have been reported not only between unrelated microorganisms (e.g. fungi and bacteria [1,2]), but also between distant relatives (e.g. members of different bacterial genera [3]), and even between close relatives (e.g. at inter-and intra-species levels [4,5]). A major strategy in niche colonization is the production of growth inhibitors or toxins directed at microbial competitors [6]. While a huge variety of secondary metabolites is used to target phylogenetically-distant competitors, ribosome-synthesized peptides or proteins are typically active against close relatives. These protein toxins are collectively referred to as bacteriocins, and may either be released into the environment or transferred to the host via specialized contact-dependent delivery systems [7][8][9].
Bacteriocins are structurally and mechanistically very diverse. This is reflected in the bacteriocinogenic potential of the genus Pseudomonas [10]. Their R-and F-type pyocins are multi-subunit protein complexes evolutionarily related to contractile tails of bacteriophages [11][12][13]. R-pyocins attach to specific lipopolysaccharide moieties at the cell surface of susceptible cells and insert their core structure through the cell envelope, causing depolarization of the cytoplasmic membrane [14]. The S-type pyocins of Pseudomonas aeruginosa share structural and functional features with Escherichia coli colicins [15]. Following docking onto surfaceexposed targets such as siderophore receptors [16,17], S-pyocins kill cells by nucleic acid degradation [10,17], cytoplasmic membrane damage [18], or inhibition of peptidoglycan synthesis [19,20]. Putidacin A (or LlpA BW ), first identified in Pseudomonas putida BW11M1 [21], represents a class of Pseudomonas-specific antibacterial proteins not related to any known bacteriocin. Additional llpA-like genes encoding functional bacteriocins were identified by genome mining in the biocontrol strain Pseudomonas fluorescens Pf-5 [22] and in the phytopathogen Pseudomonas syringae pv. syringae 642 [23]. Identification of this type of protein in two Xanthomonas pathovars extended its occurrence as a genus-specific killer protein [23]. The Xanthomonas LlpA precursor is proteolytically processed by removal of a characteristic Type II secretion signal peptide, whereas such N-terminal sequence is lacking in Pseudomonas homologues, indicating that secretory routes may differ among LlpA producers.
The amino acid sequence of LlpA suggests the presence of two related domains belonging to the 'monocot mannosebinding lectin' (MMBL) family [24]. The MMBL domain consists of a b-prism fold containing three potential carbohydrate-binding pockets, each generated by a QxDxNxVxY sequence (with x, any amino acid), but some sites may be inactive due to degeneracy of the signature motif [25]. This domain (Pfam domain: B_lectin -PF01453) was initially identified in lectins of monocot plants [26,27], but a more widespread occurrence of MMBL lectins has become evident and includes representatives in fungi [28,29], slime molds [30], sponges [31], and fishes [32][33][34]. The LlpA branch occupies a unique position among MMBL-domain proteins, harboring non-eukaryotic representatives and being equipped with the capacity to kill bacterial cells with bacteriocin-like specificity, a property not yet demonstrated for other family members [25]. Next to proteins with the LlpA-type tandem-MMBL organization, many other predicted MMBL proteins are encoded by bacterial genomes. Often the MMBL module is embedded in a larger protein. For one such protein, bacteriocin-like activity among Ruminococcus species, Gram-positive bacteria colonizing the rumen, was demonstrated [35].
Here we report on the crystal structure of LlpA BW as the prototype of a novel family of antibacterial proteins and explore how domain architecture and specific structural elements contribute to its activity and specificity.

LlpA forms a rigid MMBL tandem
The crystal structure of LlpA BW from P. putida BW11M1 (LlpA BW ) shows it contains two b-prism MMBL domains, referred to as the N-domain and the C-domain following their position in the amino acid sequence ( Figure 1A,B; Figure S1). The N-domain spans residues Arg4-Pro135 while the C-domain encompasses residues Ala136-Gln253. Each domain exhibits pseudo-threefold symmetry and the corresponding subdomains will be referred to as I N , II N , III N , I C , II C and III C , respectively ( Figure 1A and Figure S1). Following these two domains, a bhairpin extension is formed by residues Pro254-His275 (the numbering used in this paper corresponds to that of the wild-type protein without His-tag [21]).
The two-domain architecture reflects the b-strand swapping that is typical in dimers of single-domain mannose-binding monocot lectins ( Figure 1A,B) [36] and which apparently is retained after the ancestral fusion or duplication of the two domains, as is also the case in certain MMBL tandems or heterodimers from monocots [37,38]. Thus, residues Asp126-Pro135 from the first MMBL sequence complement the fold of the C-domain while residues Pro245-Gln253 from the second MMBL sequence complement the fold of the N-domain. However, in LlpA BW , the relative orientation of both domains is different compared to what is observed in a canonical MMBL lectin dimer, such as snowdrop lectin [36], in the heterodimeric MMBL lectin ASA I from Allium sativum [38], or in the tandem MMBL SCAfet from Scilla campanulata [37] ( Figure 1C and Figure  S2). In contrast to these plant MMBL proteins, the resulting architecture of LlpA BW does not obey pseudo-twofold symmetry ( Figure 1C).
LlpA BW is a very rigid molecule. The two monomers present in the asymmetric unit are essentially identical with a root-meansquare deviation (RMSD) of 0.34 Å for 270 Ca atoms. This RMSD value does not change significantly when the individual domains are fitted separately (0.32 Å for 120 Ca's of the Ndomain and 0.22 Å for 115 Ca's of the C-domain), indicating that the inter-domain orientation is fixed. This stems from three sets of interactions ( Figure 2). Both domains are connected by a twostranded anti-parallel b-sheet that is involved in the b-strand swapping mentioned above and that links both domains. The Cterminal b-hairpin extension makes extensive contacts, through hydrophobic and hydrogen bonds, with both domains. Finally, the stretch Val140-Asp145 of the C-domain makes extensive contacts with stretch Val115-Asp118 and with the side chains of Ser15 and Pro32 of the N-domain.
Domains of LlpA BW are shaped by differential evolutionary pressure A superposition of the Ca-trace of the N-and C-domain of LlpA BW as well as the MMBL domain of snowdrop lectin is shown in Figure S3. Based on 79 Ca atoms that form the common bsheet core of the MMBL domains, the RMSD between the N-and C-domains of LlpA BW is 1.84 Å . While the secondary structure elements of the C-domain are restricted to the three four-stranded b-sheets of the b-prism fold, the N-domain contains three additional secondary structure elements ( Figure 1A). A three-turn a-helix (a1) is inserted in the loop between strands b9 and b10, and sheet II N contains two additional strands. Strand b69 is inserted in the loop between strands b6 and b7 and provides an anti-parallel extension to sheet II (hydrogen bonding to strand b9). Strand b19 is a short piece of b-strand that is part of the long Nterminus and forms a parallel extension on the opposite site of sheet II N (hydrogen bonding to strand b2), making this b-sheet a mixed type six-stranded one rather than the canonical fourstranded anti-parallel sheet.
Despite these additions to the b-prism fold, the common core of the N-domain more closely resembles that of the well-studied and highly conserved monocot lectins (e.g. RMSD of 1.35 Å with snowdrop lectin compared to 1.82 Å for the C-domain). This structural divergence is in contrast with the degree of conserva-

Author Summary
In their natural environments, microorganisms compete for space and nutrients, and a major strategy to assist in niche colonization is the deployment of antagonistic compounds directed at competitors, such as secondary metabolites (antibiotics) and antibacterial peptides or proteins (bacteriocins). The latter selectively kill closely related bacteria, which is also the case for members of the LlpA family. Here, we investigate the structure-function relationship for the prototype LlpA BW from a saprophytic plant-associated Pseudomonas whose genus-specific target spectrum includes several phytopathogenic pseudomonads. By determining the 3D structure of this protein, we could assign LlpA to the so-called monocot mannosebinding lectin (MMBL) family, representing its first prokaryotic member, and also add a new type of protective function, as the eukaryotic MMBL members have been linked with antiviral, antifungal, nematicidal or insecticidal activities. For the protein containing two similarly folded domains, we constructed site-specific mutants affected in carbohydrate binding and domain chimers from LlpA homologues to show that mannosespecific sugar binding mediated by one domain is required for activity and that the other domain determines target strain specificity. The strategy that evolved for these bacteriocins is reminiscent of the one used by mammalian bactericidal proteins of the RegIII family that recruited a Ctype lectin fold to kill bacteria. tion of the carbohydrate-binding motif characteristic of the monocot lectins (QxDxNxVxY) in each of the three subdomains. In the N-domains of LlpA homologues, the surface-exposed motifs III and II are not well conserved and likely lost their function during evolution. In contrast they seem to be better conserved in the C-domains ( Figure S4). Apparently, the two MMBL domains of LlpA experienced a differential evolutionary pressure resulting in different degrees of global and local (carbohydrate-binding motif) conservation, suggesting distinct functional roles for each domain.
The C-domain of LlpA BW further extends into a b-hairpin that helps to define the relative orientations of its two MMBL domains. This b-hairpin is highly bent due to a b-bulge inserted into its second b-strand ( Figure 1B). It is absent in all plant representatives including tandem MMBL proteins such as SCAfet ( Figure 1C). In bacteria it represents the most divergent part of LlpA homologues, both in primary sequence and in length ( Figure S5). Most of these C-terminal extensions terminate with a phenylalanine residue. This is reminiscent of the conserved terminal phenylalanine of outer membrane proteins from Gram-negative bacteria such as PhoE, required for their translocation to the cell envelope [39]. An equivalent extension appears to be absent in the Xanthomonas and Arthrobacter sequences ( Figure S5).

LlpA is capable of binding mannose-containing carbohydrates
Subdomains II C and III C of LlpA BW contain the typical sugarbinding signature (QxDxNxVxY) of an active MMBL mannosebinding site ( Figure S1 and S4). Soaking crystals of LlpA BW with 200 mM methyl-a-D-mannopyranoside (Me-Man) led to clear electron density of a single Me-Man in site III C of each of the two LlpA BW monomers in the asymmetric unit ( Figure S6A). This site comprises the side chains from Gln171, Asp173, Asn175 and Tyr179, which contribute to hydrogen bond interactions and the side chains of residues Val177, Asn188, Gln192 and Ala185, which contribute to van der Waals contacts with the carbohydrate ligand ( Figure 3A, Figure S7A  the pentasaccharide, the central reducing mannose is located in the shallow Me-Man binding site and the two GlcNAcb(1-2)Man moieties stretch out over the surface making only a few additional hydrogen bonds or van der Waals contacts ( Figure 3B). In the bound disaccharide, the non-reducing mannose is located in the Man-Me binding site while the reducing mannose faces the solvent and does not interact directly with the protein ( Figure 3C).
Site II C of both LlpA BW molecules in the asymmetric unit is involved in crystal packing interactions and the presence of Me-Man is therefore sterically excluded. All residues that form specific hydrogen bonds with Me-Man are retained but substitutions occur for three side chains that provide van der Waals contacts ( Figure  S4 and S8A). In contrast, site I C lost the conserved QxDxNxVxY motif ( Figure S4) and is involved in inter-domain contacts and therefore inaccessible to ligands ( Figure S8B).
The putative carbohydrate-binding sites in the N-domain of LlpA BW are less conserved. Similar to the C-domain, site I N is inaccessible and involved in inter-domain interactions ( Figure  S9A). In the II N subdomain, the canonical mannose-binding motif QxDxNxVxY is essentially absent, with only the Gln residue of the motif being conserved as Gln82 ( Figure S4). All other donors or acceptors required for hydrogen bonds with a mannose ligand are missing. In addition, the presence of Phe86 at the equivalent position of the expected Val sterically hinders the binding of mannose ( Figure S9B). The potential carbohydrate-binding site on subdomain III N is only partially conserved ( Figure S9C) and contains two relevant substitutions from the canonical signature: (1) the Tyr residue of the QxDxNxVxY motif is replaced by the shorter Gln49, thereby removing the canonical hydrogen bond between Man O4 and Tyr OH, and (2) a threonine at position 54 which may compensate the hydrogen bond lost due to the Tyr-to-Gln substitution in the canonical motif. The lack of electron density at this site in our Me-Man soak nevertheless indicates that this site does not recognize this ligand or that its affinity is so low that recognition would only be achieved in the context of a larger and as yet unidentified mannose-containing ligand. Alternatively, this putative site may possess specificity for a different monosaccharide. In order to evaluate this hypothesis, we soaked LlpA BW crystals with D-galactose, N-acetyl-D-glucosamine and L-fucose. No electron density was observed for any of these sugars, suggesting that the N-domain has a function distinct from carbohydrate recognition (data not shown).

Carbohydrate-binding capacity is required for LlpA toxicity
The LlpA BW motifs III N , III C and II C create potential carbohydrate binding sites that may be involved in bacteriotoxicity of the protein. We therefore examined the role of carbohydrate binding in the bactericidal function of LlpA BW . The presence of methyl-a-D-mannopyranoside up to 500 mM in the medium did not influence the activity of LlpA BW on P. syringae GR12-2R3. Glycan array profiling did not highlight any specific oligosaccharide structure that could represent a natural ligand of LlpA BW (Table S1). This could be due to the array design that is principally based on eukaryotic glycans and may therefore lack an appropriate carbohydrate for this prokaryotic toxin. Previously, it was observed that LlpA BW from concentrated culture supernatant does not agglutinate rabbit red blood cells, nor binds to a mannoseagarose affinity matrix [21].
To assess whether the mannose-recognizing QxDxNxVxY motifs in LlpA BW are nevertheless relevant for bactericidal activity, the conserved valine residue was mutated to tyrosine in subdomains III N , III C , and II C . These mutations sterically preclude mannose or any other ligand to enter the binding sites ( Figure S7C). Semi-quantitative activity assays with permeabilized E. coli cells expressing the LlpA variants in motifs III N , III C and II C were used to assess the relationship between carbohydrate binding and bactericidal activity. Modification of the III N site, for which no mannose binding was observed, does not affect the antibacterial activity against P. syringae GR12-2R3 ( Figure 4). In contrast, the altered III C pocket strongly diminishes activity, either alone or in pairwise combination with the other mutated sites (III N or II C ). A minor negative effect of the II C mutation is only apparent in a double mutant, when combined with a modified III N motif.
Purified proteins were prepared to further quantify these effects. Far UV CD spectra of these mutant forms are identical to that of native protein LlpA BW , indicating that the mutations do not affect the overall structure of the protein. Isothermal titration calorimetry (ITC) showed that LlpA BW has an affinity of 2.1 mM for the pentasaccharide Man, the highest among all the tested oligo-mannosides (See Figure 5 and Table 1 for a summary of the experimentally validated LlpA BW -carbohydrate interactions). This is in agreement with the crystal structures of the different complexes since this sugar is the one with the largest binding interface (Figure 3). Titrations of LlpA BW , of the mutants LlpA V177Y (a site III C knockout), LlpA V208Y (a site II C knockout) and of the double mutant LlpA V177Y-V208Y with a-methyl mannoside clearly pinpoint site III C as the only responsible for the sugar binding activity. Point mutations in both sites or III C (V177Y) alone, completely abrogate sugar binding. However knocking out site II C (V208Y) has little effect in binding and the affinities of LlpA V208Y for amethyl mannoside and Mana(1-3)Man are very close to the ones measured for the wild-type protein (See Table 1 and Figure 5B).
While the V208Y mutation in the II C site has no observable effect on the MIC value for P. syringae GR12-2R3, the altered III C motif engenders a 5.2-fold increase in MIC ( Figure 4). The mutant protein LlpA V177Y-V208Y suffers a further reduction in activity, yielding a 31.6-fold increased MIC compared to native LlpA BW . The biological activities of LlpA and its mutants were further assessed by live/dead staining and subsequent flow cytometry analysis ( Figure 6, Figure S10). Proportions of dead cells after 1 hour of exposure to LlpA or LlpA V208Y were comparable (10.1% and 9.7%, respectively). For LlpA V177Y this value was reduced to 6.1%, significantly lower than for LlpA. Killing activity was even further reduced for LlpA V177Y-V208Y (3.7%). These results are consistent with the MIC determination and ITC data, indicating that an active site III C is required to generate a fully active LlpA bacteriocin. The difference in bacteriotoxicity between LlpA V177Y and LlpA V177Y-V208Y suggests that site II C has a supporting role in the LlpA BW bacteriotoxicity.

All domains are necessary for LlpA BW functionality
The site-directed mutagenesis approach revealed an important role for the C-domain's carbohydrate-binding capacity in LlpA BW toxicity. Considering the increased binding motif degeneration in the N-domain and the fact that a Ruminococcus bacteriocin composed of only a single MMBL domain fused to an unknown domain has been identified [35], the N-domain may fulfill a distinct function, different from that of the C-domain. In order to scrutinize the contribution of individual domains to overall activity, six domain deletion constructs of llpA BW were engineered to potentially encode proteins lacking the first or second MMBL domain, a gene product devoid of the C-terminal hairpin, or a protein retaining only an individual domain (N-domain, Cdomain, or hairpin) ( Figure S11). To take the domain swapping into account, the constructs containing only a single MMBL domain were designed with a fusion of the swapped C-terminal bstrands to the corresponding domain via a short linker.
None of these deletion constructs resulted in the production of an active protein, indicating that none of the domains are dispensable. Removal of the terminal phenylalanine residue still allows expression of a functional bacteriocin in E. coli (Figure 7), but a further C-terminal truncation (deletion of Trp-His-Phe tail) resulted in a negative bacteriocin assay (data not shown). From these data we conclude that both MMBL domains as well as the C-terminal hairpin extension are required for activity of LlpA. Whether the role of the C-terminal hairpin is any other than simply stabilization of the C-domain cannot be concluded.

Target specificity of LlpA is hosted by the N-domain
In order to investigate the role of the different domains in target specificity, we created hybrid LlpA proteins using the domains of LlpA BW from P. putida BW11M1 and LlpA1 from P. fluorescens Pf-5. These two LlpA proteins share 45% sequence identity and differ in their target spectra. Strains P. syringae GR12-2R3 and P. fluorescens LMG1794 were identified as specific indicators for LlpA BW [21] and LlpA1 [22], respectively. Six constructs carrying llpA BW / llpA1chimeric genes were made with domain exchanges involving the N-domain, C-domain, and hairpin region (Figure 7 and Figure  S11). For four of these constructs activity against one of both indicators was detected. Only constructs retaining the original Ndomain give rise to inhibition of the cognate indicator strain. The C-domain or the hairpin of LlpA BW could be replaced with the corresponding LlpA1 domains without changing target specificity.
Conversely, the original specificity of LlpA1 is retained upon replacement of its C-domain by the LlpA BW equivalent.

Discussion
Structure elucidation of LlpA BW from P. putida BW11M1 unequivocally assigns this bacteriocin to the MMBL lectin family, in which it constitutes the first prokaryotic member, representative for a group of bacterial proteins composed of two MMBL domains [21][22][23]. Systematic inactivation of the three potential carbohydrate-binding sites present in the N-domain (III N ) and in the C-domain (III C and II C ) of LlpA BW , revealed that a non-occluded III C pocket is required to obtain a fully active LlpA BW molecule. A negative co-operative effect on activity resulted when the II C site was additionally modified. Although mannose-containing carbohydrates can bind to the III C pocket of LlpA BW , it remains unclear  if these are part of or mimic the natural ligand required for biological activity since bacteriocin activity is not impaired in the presence of excess mannose. A mutated III N site did not provoke a negative effect on antibacterial activity. However, the N-domain appears to play a major role in target selection. This was demonstrated by assessing the differential activity of domain chimers against two target strains, diagnostic for the LlpA BW -and LlpA1specific killing.
The b-hairpin does not appear to be a specificity determinant, although it constitutes the most variable region among LlpA-like bacteriocins. Possibly, it is required for thermodynamic stability since it needs to be intact in LlpA BW . An equivalent C-terminal stretch is absent from the Xanthomonas citri LlpA-like bacteriocin [23]. From our results relying on heterologous expression in E. coli and a bacteriocin assay with permeabilized cells, it cannot be excluded that this structural element may play a role in the way an LlpA protein is exported by its native producer cells.  In general, a defensive role has been proposed for the (oligo)mannose-binding MMBL lectins based on insecticidal, nematicidal, antifungal, or even antiviral activities demonstrated for several of these proteins that are abundantly found in monocot plants [40][41][42][43][44][45][46]. Some of these plant lectins trigger apoptosis in cancer cells [47]. Also their identification in fish mucus and epithelial cells is in line with a general protective (antimicrobial) function for MMBL domains [32]. LlpA BW as a bactericidal protein fits within this picture of MMBL domains being involved in general defense mechanisms. Since no antibacterial activity has been assigned to the eukaryotic MMBL proteins, it is challenging to identify structural features that confer the intragenus-specific bacteriocin activity of LlpA, as shown for proteins from P. putida [21], P. fluorescens [22], P. syringae [23], and Xanthomonas citri [23]. Their target spectra are narrower than reported for the mammalian antibacterial C-type lectins of the RegIII family, such as mouse RegIIIc and its human homolog HIP/PAP that bind to the surface-exposed peptidoglycan layer of Gram-positive bacteria [48], and RegIIIb that also binds to the lipid A moiety of lipopolysaccharides on the cell envelope of Gram-negative bacteria [49].
The absence of any known secretory signal sequence in LlpA BW and its homologues in other Pseudomonas species is intriguing in view of their extracellular location [21]. The translocation of the outer membrane-associated mannose/ fucose-specific lectin LecB of P. aeruginosa, that also lacks such signal sequence [50], is dependent on its glycosylation [51]. Contrary to LlpA that is exported to the culture supernatant to exert its antagonistic activity, LecB remains associated with the cell envelope through interaction with the major outer membrane protein OprF [52], in line with its role in biofilm formation.
Plasmids used for antibacterial testing and sequencing were propagated in E. coli TOP10F9 (Invitrogen). E. coli BL21(DE3) (Novagen) was used as a host for plasmids driving recombinant protein expression. Genomic DNA from P. putida BW11M1 was isolated using the Puregene Yeast/Bact. Kit B (Qiagen). Plasmid DNA was extracted using the QIAprep Spin Miniprep Kit (Qiagen). Stocks were stored at 280uC in the appropriate medium in 25% (v/v) glycerol.  Figure 4) and of LlpA1 (inferred by pairwise alignment; N-domain in orange, C-domain in purple and C-terminal extension in grey) are depicted, along with those of chimeric forms (in dashed box). The LlpA variant lacking the terminal phenylalanine residue is marked with a yellow hexagon. Inhibitory activity of the respective E. coli recombinants was tested with diagnostic indicators for LlpA BW (P. syringae GR12-2R3) and LlpA1 (P. fluorescens LMG 1794). Halo sizes are semi-quantified according to size of the growth inhibition halo (+++, native halo size of LlpA BW and LlpA1; ++, halo size reduced; C, local clearing confined to producer colony spot; 2, no halo or clearing; NT, not tested). Additional chimeric and domain deletion constructs not conferring bacteriocin activity against one of the indicator strains are specified in Figure S11. doi:10.1371/journal.ppat.1003199.g007

Recombinant DNA methods
Standard methods were used for preparation of competent E. coli cells, heat shock transformation of E. coli and DNA electrophoresis [53]. Restriction enzymes were used according to the supplier's specifications (Roche Diagnostics and BIOKÉ ). DNA ligation was performed using T4 DNA ligase (Invitrogen). Plasmid sequencing was performed by GATC Biotech (Constance, Germany). Constructs that were generated are listed in Table S2 and primers are listed in Table S3.
A 921-bp fragment containing llpA BW was amplified by PCR with Platinum Pfx DNA polymerase (Invitrogen), using a C1000 Thermal Cycler (Bio-Rad). P. putida BW11M1 genomic DNA was taken as a template, and combined with primers PGPRB-3155 and PGPRB-3156. The amplicon was purified using the QIAquick PCR Purification Kit (Qiagen), digested with KpnI and BamHI, ligated in pUC18, and transformed to E. coli TOP10F9. Transformants were verified for the presence of the insert by PCR using Taq Polymerase (BIOKÉ ) with primers PGPRB-2545 and PGPRB-2546. The cloned construct (pCMPG6129) was purified and its insert confirmed by sequencing.
DpnI-mediated site-directed mutagenesis was performed to construct valine-to-tyrosine mutant forms of llpA BW in pUC18 (pCMPG6129) and N-terminal His-tagged llpA BW in pET28a (pCMPG6056 [54]). PCR conditions were: 2 min initial denaturation, followed by 16 cycles of denaturation (1 min), annealing (1 min, primer-dependent temperature) and elongation at 68uC (1 min./kb). Final elongation was for 8 min at 68uC. After PCR, samples were immediately treated with DpnI at 37uC for 2 h and transformed into E. coli TOP10F9 and selected on the appropriate medium. Plasmid inserts of selected transformants were verified by sequence analysis. Double mutants were constructed using plasmids with a single point mutation as a template.
Domain deletants of llpA BW were constructed using pCMPG6129 as a template (llpA BW from P. putida BW11M1). Chimeric constructs were obtained using pCMPG6129 and pCMPG6053 (llpA1 from P. fluorescens Pf-5 [22]) as templates. Artificial ligation of gene fragments, generated with the PCR primers specified in Table S3, was performed by using splicing by overlap extension (SOE). The resulting recombinant amino acid sequences are listed in Table S4.

Recombinant protein expression and purification
Protein isolation and purification of N-terminal His 6 -tagged LlpA BW , LlpA V177Y , LlpA V208Y , and LlpA V177Y-V208Y from E. coli BL21(DE3), carrying expression constructs pCMPG6056, pCMPG6149, pCMPG6150 and pCMPG6151 respectively, were performed as described by Parret and collaborators [54]. The presence of His-tagged protein was observed via immunodetection by Western blot, using monoclonal anti-His 6 (IgG1 from mouse; Roche Diagnostics) as primary antibody. Fractions free of other proteins, as verified by SDS-PAGE and subsequent Coomassie Blue staining, were dialyzed against bis-TRIS propane buffer (20 mM, 200 mM NaCl, pH 7.0). Concentrations of purified proteins were determined by absorbance measurement at 280 nm using molar extinction coefficients of 62910 M 21 cm 21 for LlpA BW , 64400 M 21 cm 21 for LlpA V177Y and LlpA V208Y , and 65890 M 21 cm 21 for LlpA V177Y-V208Y . Extinction coefficients were calculated according to Pace and collaborators [55].

Antibacterial assays
Bacteriocin production by E. coli cells carrying pUC18-derived constructs was assayed as follows: 2-ml drops of an overnight stationary-phase culture were spotted onto LB agar plates and incubated for 8 h at 37uC. Next, plates were exposed to chloroform vapor (30 min), aerated and subsequently overlaid with 5 ml of soft agar (0.5%), seeded with 200 ml of a cell culture of an indicator strain (,10 8 CFU/ml), followed by overnight incubation at 30uC. Next day, plates were scored for the presence of halos in the confluently grown overlay.
Antibacterial activity assays with purified recombinant His 6tagged proteins were performed as described by Ghequire and collaborators [23]. To assess the influence of added sugar, the same assay was carried out on agar medium supplemented with Dmannose (Sigma-Aldrich) or methyl-a-D-mannopyranoside (Sigma-Aldrich) to a final concentration of 0.01 M to 0.5 M.
A Bioscreen C apparatus (Oy Growth Curves Ab Ltd, Finland) was used to determine the minimum inhibitory concentration (MIC). An overnight culture (16 h) of the indicator strain was diluted to 10 4 -10 5 CFU/ml and incubated at 30uC, with a twofold dilution series of recombinant His 6 -tagged LlpA BW or mutant LlpA BW . Bis-TRIS propane buffer was used as control. The MIC value was determined as the minimum concentration of protein at which no growth of the indicator strain (OD 600 ,0.2) occurred after 24 h. At least three independent repeats, each with three replicates, were carried out.

Glycan array
His 6 -tagged LlpA BW was lyophilized and verified for antibacterial activity. After re-dissolving in MilliQ water, recombinant LlpA BW was diluted to 200 mg/ml with binding buffer (20 mM TRIS-HCl pH 7.4, 150 mM NaCl, 2 mM CaCl 2 , 2 mM MgCl 2 , 0.05% Tween 20, 1% BSA), and used to probe the printed glycan arrays [56] following the standard procedures of Core H of the Consortium for Functional Glycomics (http://www. functionalglycomics.org/). Monoclonal anti-His 6 antibodies (Roche Diagnostics) were used as primary antibodies, and fluorescently labeled anti-mouse IgG as secondary antibodies.

Circular dichroism
CD spectra were acquired on a Jasco J-715 spectropolarimeter. Curves were averaged over 8 scans, taken at 25uC using a 1 mm cuvette. Samples were dialyzed against bis-TRIS propane buffer (20 mM, NaCl 200 mM, pH 7.0), filtered and degassed prior to data acquisition. All proteins were assayed at 0.4 mg/ml.

Isothermal titration calorimetry
ITC titrations were carried out on an ITC200 apparatus (MicroCal). Prior to the measurement, LlpA BW , LlpA V177Y , LlpA V208Y and LlpA V177Y-V208Y was dialyzed to bis-TRIS propane buffer. Sugars were directly dissolved into the same buffer. The samples were filtered and degassed for 10 min at 25uC before being examined in the calorimeter. The titrations were carried out at 25uC, injecting the sugars (methyl-a-Dmannoside, Mana  Table 1.

X-ray data collection and structure determination
Expression, purification and crystallization of recombinant Histagged LlpA BW have been described [54]. X-ray data for native and derivative crystals were collected on EMBL beamline BW7A of the DESY synchrotron (Hamburg, Germany). For each potential derivative, the wavelength was chosen to be at the high-energy side of the absorption edge in order to ensure a usable anomalous signal. All data were scaled and merged with the HKL package of programs. Data collection statistics are given in Table S5.
The crystal structure of free LlpA BW was solved combining single isomorphous replacement with anomalous scattering (SIRAS strategy) from a p-chloromercurybenzoate derivative. The heavy-atom substructure was determined with SHELXD [57] using a resolution cutoff of 4.0 Å . Heavy-atom refinement and phasing were performed with SHARP [58]. Phase improvement by solvent flattening was performed with SOLOMON [59]. Non-crystallographic symmetry averaging with density modification [60] further improved the electron density. A partial model (94% of the residues comprising the asymmetric unit) was automatically built with ARP/wARP [61] and the remainder was built manually over several cycles of model building with Coot [62], alternated with refinement using phenix.refine [63,64]. Phasing and refinement statistics are shown in Table S5.

Carbohydrate soaks
Crystals of LlpA BW were transferred to artificial mother liquor

Flow cytometry
Overnight cultures of P. syringae GR12-2R3 (16 h) were diluted to OD 600 0.5 and washed twice with phosphate-buffered saline (PBS). Cells were treated with LlpA, mutant proteins or buffer (bis-TRIS propane buffer, negative control), at a final concentration of 50 mg/ml for 1 h, at 20uC. Next, PBS-washed bacteria were stained using the Live/Dead BacLight bacterial viability kit (Invitrogen), incubated for 15 minutes, and analyzed on a BD Influx (BD Biosciences). Excitation of the dyes was done at 488 nm, and fluorescence measured at 530 nm for SYTO 9 and at 610 nm for propidium iodide. Results were processed with FlowJo 10.0.4 software ( Figure S10). Measurements were done independently and based on six biological repeats. Results are expressed as percentages of dead cells [dead/(live+dead) * 100)].  Figure S5 Sequence alignment of the carboxy-terminal sequences of LlpA-like proteins. The P. putida LlpA BW sequence adopting a b-hairpin fold is delineated in Figure S1. The preceding conserved tryptophan residue is located C-terminally to I C ( Figure S1). The sequence logo representation visualizes the degree of consensus for each residue. Accession numbers are listed in Figure S4 Stereoview of site II C of the C-domain (colored according to atom type) superimposed on site III C of the C-domain (dark gray). The Me-Man bound in site III C is shown in red. This site is very similar to site III C but in the crystal it is inaccessible due to crystal lattice interactions. Residue labels correspond to residues of site II C . (B) Similar view showing site I C of the C-domain (colored according to atom type) superimposed on site III C of the C-domain (dark gray). The Me-Man bound in site III C is shown in red. The stretch of Ile137-Leu139 that provides a steric conflict preventing Me-Man binding in site I C , is highlighted with carbon atoms drawn in green. Residue labels correspond to residues of site I C . (JPG) Figure S9 Sites of the LlpA BW N-terminal domain. (A) Stereoview of site I N of the N-domain (colored according to atom type) superimposed on site III C of the C-domain (dark gray). The Me-Man bound in site III C is shown in red. The stretch of Ile271-Trp274 that provides a steric conflict preventing Me-Man binding in site I N is highlighted with carbon atoms drawn in green. Residue numbering corresponds to residues of site I N . (B) Similar superposition for site II N of the N-domain. Phe86 that prevents Me-Man binding to this site through a steric conflict is highlighted in green. Other residues belonging to site II N are labeled in teal. Three residues of site III C for which site II N has no structural equivalent are labeled in black. (C) Similar superposition for site III N of the N-domain. Residues belonging to site III N are labeled in teal. One residue of site III C for which site III N has no structural equivalent, is labeled in black.