Specificity and genetic polymorphism in the Vfm quorum sensing system of plant pathogenic bacteria of the genus Dickeya

Summary The Vfm quorum sensing (QS) system is preponderant for the virulence of different species of the bacterial genus Dickeya. The vfm gene cluster encodes 26 genes involved in the production, sensing or transduction of the QS signal. To date, the Vfm QS signal has escaped detection by analytical chemistry methods. However, we report here a strain‐specific polymorphism in the biosynthesis genes vfmO and vfmP, which is predicted to be related to the production of different analogues of the QS signal. Consequently, the Vfm communication could be impossible between strains possessing different variants of the genes vfmO/P. We constructed three Vfm QS biosensor strains possessing different vfmO/P variants and compared these biosensors for their responses to samples prepared from 34 Dickeya strains possessing different vfmO/P variants. A pattern of specificity was demonstrated, providing evidence that the polymorphism in the genes vfmO/P determines the biosynthesis of different analogues of the QS signal. Unexpectedly, this vfmO/P‐dependent pattern of specificity is linked to a polymorphism in the ABC transporter gene vfmG, suggesting an adaptation of the putative permease VfmG to specifically bind different analogues of the QS signal. Accordingly, we discuss the possible involvement of VfmG as co‐sensor of the Vfm two‐component regulatory system.

Quorum sensing (QS) is a cell-to-cell communication process that enables bacteria to collectively, and in a synchronized manner, modify their behaviour in response to changes in the cell density. Among the functions under QS control stand the formation of biofilms, the acquisition of nutrients, the conjugative transfer of plasmids, the production of antibiotics, and, more generally, the production of virulence factors (for reviews, see Grandclément et al., 2016;Hawver et al., 2016;Papenfort and Bassler, 2016). A successful bacterial infection is based on finely tuned regulatory mechanisms, including QS systems. Two QS mechanisms have been described in Dickeya. The first system is typically found in Gramnegative bacteria and involves the production and the cytoplasmic recognition of N-acyl-homoserine lactones (Nasser et al., 1998). The second QS system is original, depending on a signal encoded by a locus called vfm for virulence factor modulating (Nasser et al., 2013). All 126 sequenced genomes of Dickeya strains available until April 2021 do contain the vfm locus but, so far, the Vfm QS system has been studied only in three species: it was shown to be preponderant for the virulence of the D. dadantii strain 3937 (Nasser et al., 2013) and for various strains of the agronomically important potato pathogen D. solani (Potrykus et al., 2018). In strain EC1, a rice pathogen recently reclassified as D. oryzae , the Vfm QS system also modulates multiple virulence traits (Lv et al., 2019).
Initially, the vfm locus was discovered and genetically characterized in the model strain D. dadantii 3937 (Nasser et al., 2013). It encodes 26 genes annotated as vfmA to vfmZ and is expected to be required for the biosynthesis, sensing or transduction of the Vfm QS signal (Fig. 1). It does not share any gene homology with other known and studied QS loci, indicating that both the structure of the Vfm QS signal and its corresponding signalling cascade exhibit unique features. Despite several attempts, the Vfm QS signal has never been isolated and has escaped detection by analytical chemistry methods (Nasser et al., 2013;Lv et al., 2019). The products of the genes vfmM, vfmO and vfmP are annotated as amino acid-activating adenylation domain proteins, indicating that the Vfm signal is a non-ribosomally synthesized peptide. Short peptides used as QS signals are common among Gram-positive bacteria, but these autoinducing peptides (AIP) correspond to ribosomally synthesized peptides. Each AIP is synthesized as a longer peptide precursor which is subsequently exported and modified (Otto et al., 1998;Cook and Federle, 2014).
Based on functional studies in D. dadantii 3937 and D. oryzae EC1, the vfm locus was shown to activate virulence through a signalling cascade involving the twocomponent system VfmI/VfmH (Nasser et al., 2013;Lv et al., 2019). When the Vfm QS signal accumulates in the extracellular medium up to a certain threshold concentration, this two-component system initiates the transcriptional expression of the vfm genes, rapidly provoking the activation of the expression of virulence genes (Nasser et al., 2013;Lv et al., 2019). Interestingly, the AIP QS system of several Gram-positive bacteria also employs a two-component regulatory system for the signal transduction and activation of the virulence genes (Hoch and Silhavy, 1995;Hvarstein, 2003). The most studied example is the Agr QS system of the genus Staphylococcus. A remarkable feature of the agr locus is the genetic link between a polymorphism observed within the ligand-binding domain of the histidine kinase and a polymorphism observed within the sequence of the mature AIP signal (Dufour et al., 2002). In Staphylococcus aureus, these genetically linked allelic variations were shown to define four Agr QS specificity groups, each producing a unique AIP analogue associated with a unique sequence of the ligand-binding domain of the histidine kinase. Among the four AIP analogue-histidine kinase pairs, two are more efficient for the transduction of the QS signal, illustrating the fact that the polymorphism observed in the agr locus of S. aureus affects the bacterial pathogenicity by impacting the efficiency of the Agr QS system (Geisinger et al., 2012).
A previous comparative genomic analysis of 47 strains of the genus Dickeya was performed to understand the phylogenetic relationship between the different species (Duprey et al., 2019). In this work, several phylogenetic trees were constructed both on core proteins and on proteins involved in the bacterial virulence. While the phylogenetic trees based on core proteins largely correlate with the species classification, we noticed that this correlation is much less clear for the phylogenetic trees of some Vfm proteins (Duprey et al., 2019). This result suggests that horizontal gene transfer occurred among the species of Dickeya at the level of some vfm genes. For example, the proteins VfmG and VfmP of the D. dadantii strain NCPPB 3537 belong to phylogenetic groups distant from those containing the proteins VfmG and VfmP of other D. dadantii strains. In all cases, however, the phylogenetic trees of VfmG and VfmP correlate with each other, suggesting a link between the polymorphism in the gene vfmG and the polymorphism in the gene vfmP. While the gene vfmG encodes the permease component of an ABC transporter, the gene vfmP encodes an amino acid-activating adenylation domain protein predicted to govern the specific selection of one amino acid during the non-ribosomal assembly of the peptidic part of the Vfm QS signal.
Some analogies with the genetically linked allelic variations observed within the agr locus, which define four QS specificity groups in S. aureus, raise several questions about the functioning of the Dickeya Vfm QS system. Is the polymorphism in the biosynthesis gene vfmP related to the production of different analogues of the QS signal? What is the functional link between the biosynthesis gene vfmP and the ABC permease gene vfmG that could explain the correlated polymorphism observed in these two genes? Is there polymorphism in other vfm genes, in addition to vfmG and vfmP? Does the polymorphism in the vfm genes modify the specificity of the Vfm QS system? In order to address these questions, we further analyzed the polymorphism in the vfm genes and we propose hypotheses regarding the impact of the polymorphism in the vfm genes on the specificity of the Vfm QS system. Finally, we obtained experimental proof using three newly constructed Vfm biosensor strains and a large set of Dickeya wild-type strains that allowed us to explore and validate these hypotheses.

Results
Identification of polymorphism in VfmO and VfmP by prediction of the substrate specificity of the Vfm amino acid--activating adenylation domain proteins The genes vfmM, vfmO and vfmP encode the three amino acid--activating adenylation domain proteins VfmM, VfmO and VfmP respectively. Such stand-alone adenylation domain proteins are specific to the type II non-ribosomal peptide synthesis (NRPS) (for review , Jaremko et al., 2020). In bacteria, type II NRPS has been less studied than type I NRPS which involves large modular and multidomain synthetases (for review, Süssmuth and Mainz, 2017). Nevertheless, we used the well-documented data available on the bacterial type I NRPS to predict the nature of the amino acids activated by the proteins VfmM, VfmO and VfmP, respectively. The models of prediction of the bacterial type I NRPS are based on a substrate specificity-conferring signature corresponding to 10 residues present at identical positions in the amino acid sequence of all type I NRPS adenylation domains (Stachelhaus et al., 1999;Finking and Marahiel, 2004;Süssmuth and Mainz, 2017). The 10 residues of these signatures are identified on the basis of the alignment of the amino acid sequence of the protein to be characterized versus the amino acid sequence of experimentally characterized adenylation domains. We used these models to analyze the potential polymorphism of the proteins VfmM, VfmO and VfmP among the 126 Dickeya strains whose genome sequences are available in public databases until April 2021 (Table 1). The protein VfmM is predicted to incorporate the same amino acid in the 126 sequenced Dickeya strains, as the proteins VfmM of all these strains exhibit the same type I NRPS-like substrate specificity-conferring signature (data not shown). In contrast, different forms of the proteins VfmO and VfmP, resulting from genetic polymorphism, are predicted to incorporate different amino acids Fig. 1. Functional annotation of the vfm gene cluster of Dickeya dadantii 3937 (taken from the GenBank accession NC_014500). In the upper part, the 26 genes of the gene cluster are presented by arrows. Green arrows correspond to genes annotated as biosynthesis genes, red arrows are for genes annotated as regulatory or transporter genes and black arrows for genes encoding hypothetical proteins. The length of arrows is proportional to the gene length according to the scale indicated at the right. The predicted function of each gene is shown below, with the same colour code. Abbreviations: ABC: ATP binding cassette, ACP: acyl carrier protein, MATE: multidrug and toxic compound extrusion. The corresponding GenBank tags are indicated for each vfm genes.
according to the strains. We identified three and four different type I NRPS-like substrate specificity-conferring signatures for the proteins VfmO and VfmP respectively ( Table 1). The position in the sequences of the proteins VfmO and VfmP of the residues identified as constituting these signatures is provided in Fig. S1. To further analyze these signatures, we used the same alignment to also determine the type I NRPS-like substrate specificityconferring signature of the type II NRPS stand-alone adenylation domain protein DltA, which is required for the biosynthesis of cell wall in Bacillus subtilis and which was experimentally shown to be specific for D-alanine (Kittilä et al., 2016). Interestingly, this DltA signature (DLMTMCTVAK) shares six or seven residues with the three signatures identified for the protein VfmO (DAY-VICTVAK, DVYVICTVAKor DVFVMCTVAK; Table 1). However, since none of the VfmO signatures matches exactly to the DltA signature, it is not possible to predict that one of them is specific for D-alanine.
Since none of the three VfmO signatures and none of the four VfmP signatures match exactly to any known type I NRPS-like specificity-conferring signature or to the DltA signature, predictions regarding the nature of the activated amino acid substrate have not been possible.
Nevertheless, the variations in the signatures of VfmO and VfmP suggest the production of different analogues of the Vfm QS signal, each analogue differing at the level of the amino acids activated by the enzymes VfmO or VfmP. Accordingly, different combinations of the variants of the genes vfmO and vfmP are predicted to be related to the production of different analogues of the Vfm QS signal. Among the 126 sequenced Dickeya strains, the three variants of the gene vfmO associated with different NRPS signatures have been referred to as vfmO1 to vfmO3, and the four variants of the gene vfmP associated with different NRPS signatures have been referred to as vfmP1 to vfmP4, respectively (Table 1, Fig. S1).
The Dickeya strains are split into five vfmO/P genetic groups predicted to produce different analogues of the Vfm QS signal Five different combinations of the variants resulting from the vfmO and vfmP polymorphism have been identified among the 126 sequenced Dickeya strains (Table 1). Thus, five Vfm genetic groups designated as groups I to V have been referred to as strains possessing the combinations vfmO1/P1, vfmO1/P2, vfmO2/P2, vfmO2/ P3 and vfmO3/P4, respectively. Among them, the groups I, III and IV include the majority of the sequenced strains (38, 61 and 25 respectively) while the groups II or V contain only three and two strains, respectively (Table 1). Interestingly, the strain appurtenance to a Vfm genetic group is not correlated with the strain classification into species (Table 1; Fig. 2). Most species include strains that are distributed into two or three different Vfm genetic groups, as is the case for D. chrysanthemi, D. dadantii, D. fangzhongdai, D. oryzae, D. parazeae, D. undicola and D. zeae ( Fig. 2; Table S1). In contrast, four species seem to be mostly or solely associated with one group. All the 33 D. dianthicola strains belong to group I and all but one of the 34 D. solani strains are members of group III (Fig. 2, Table S1). The four D. aquatica or D. lacustris strains belong to group IV, and the two D. poaceiphila strains belong to group V ( Fig. 2; Table S1).
The five Vfm genetic groups, defined by the polymorphism of the NRPS substrate specificity-conferring signatures of the proteins VfmO and VfmP, are expected to perform the biosynthesis of different analogues of the Vfm QS signal. We attempted to explore this hypothesis with experimental data.
Analysis of 52 Dickeya strains for their capacity to activate the Vfm uidA biosensor strain (group I) A Vfm biosensor strain was previously constructed for the D. dadantii strain 3937 (Nasser et al., 2013) which belongs to the Vfm genetic group I (Table 1). This strain, called biosensor I-uidA in the present study, corresponds to an insertion knock-out mutant of the gene vfmE in which the promoter of this gene is fused to the reporter gene uidA encoding β-glucuronidase (Nasser et al., 2013). In order to study the ability of the biosensor I-uidA to detect and respond to the Vfm QS signal produced by Dickeya strains belonging to different Vfm genetic groups, we tested cell-free supernatants of liquid cultures of a collection of 52 strains representative of the different Dickeya species (Table 2). Since the Vfm genetic group of strains whose genomic sequences are available was previously determined (Table S1), the group appurtenance of 31 tested strains was known. A positive response was observed for all sequenced strains belonging to groups I or II (six and one strains, respectively, Table 2). In contrast, a negative response was observed with all sequenced strains belonging to any of the other three groups (14, eight and two strains for groups III, IV and V, respectively) ( Table 2). To confirm the correlation observed between the positive response to the biosensor I-uidA and the appurtenance to groups I or II, we sequenced the genomes of the D. fangzhongdai strain NCPPB 2929 which gave a positive response and of two D. dadantii strains, NCPPB 3065 and CFBP 3694, which gave a negative response (Table 2). We also resequenced the genome of the positive strain D. chrysanthemi strain CFBP 1270 (NCPPB 516) because this strain was the only experimentally tested strain of group II. Furthermore, the genomic sequence of this strain available in GenBank contains a frameshift in the vfmO gene. Analysis of the resulting sequences confirmed that D. chrysanthemi CFBP 1270 belongs to group II and shows that there is no frameshift in the vfmO gene of this strain. Regarding the three newly sequenced strains, D. fangzhongdai NCPPB 2929 belongs to group I, and both D. dadantii strains NCPPB 3065 and CFBP 3694 belong to group IV. These additional genomic data are in full agreement with the response observed for these three strains, also showing that only strains belonging to groups I or II give a positive response to the biosensor I-uidA.    The Vfm genetic group has been determined for strains whose genome sequences are available in GenBank (Table S1) and for the three strains D. fangzhongdai NCPPB 2929 and D. dadantii NCPPB 3065 and CFBP 3964 whose genomes were sequenced in the frame of this study. n/a: Vfm genetic group not available. Comparison of the Vfm QS activity of 34 sequenced Dickeya strains using Vfm luciferase biosensor strains belonging to the three major groups I, III and IV, respectively A Vfm luciferase biosensor for group I was constructed by introducing a plasmid harbouring a luciferase gene under the control of the promoter of the gene vfmE in the D. dadantii 3937 vfmE mutant. The resulting strain was called biosensor I-luc. To construct Vfm luciferase biosensors corresponding to groups III and IV, we choose the D. solani strains IPO 2222 and RNS05-1-2A, respectively. In both strains, the vfmE coding sequence was deleted and replaced by the luciferase gene expressed under the control of the vfmE promoter. The resulting strains were called biosensors III-luc and IV-luc, respectively. The three luciferase biosensors I-luc, III-luc and IV-luc were used to test cell-free supernatants of liquid cultures of 34 sequenced strains of Dickeya previously tested with the biosensor I-uidA. Regarding the biosensor I-luc, a positive response was observed with all tested strains of groups I or II, and a negative response was observed with all tested strains belonging to any of the three other groups III, IV or V (Table 3). Regarding the biosensor IV-luc, a positive response was observed with all tested strains of group IV, and a negative response was observed with all tested strains belonging to any of the four other groups (I, II, III or V) (Table 3). Regarding the biosensor III-luc, a positive response was observed with the 14 strains of group III, and a negative response was observed with the 10 strains of group IV (Table 3). However, less reproducible positive responses of the biosensor III-luc (one to four with five repeats) were also Table 3. Response of the Vfm biosensors I-luc, III-luc and IV-luc to cell-free supernatants of cultures of Dickeya strains belonging to different Vfm genetic groups I to V.

Groups a Strains
Response of the Vfm biosensors b a The Vfm genetic group has been determined for strains whose genome sequences are available in GenBank (Table S1) and also for the D. fangzhongdai strain NCPPB 2929 and the D. dadantii strains NCPPB 3065 and CFBP 3964 whose genomes were sequenced in the frame of this study. b Number of positive tests out of the total number of tests performed with the corresponding Vfm biosensor. Responses with at least one positive test are shown with a grey background.
observed for six out of the seven strains of group I, the sole strain of group II and one of both strains of group V (Table 3). These assays have been repeated three times for the biosensors I-luc and IV-luc and five times for the biosensor III-luc. Variations among repeats were observed when using the biosensor III-luc with the cellfree supernatants of several strains belonging to the groups I, II, or V and even with one strain of group III (Table 3).
Comparison of the whole vfm gene cluster sequence from strains belonging to the same species but to different Vfm genetic groups To identify other potential polymorphic genes associated with variation in the specificity of the Vfm QS system, we compared the nucleotide sequence of the vfm locus of strains belonging to the same species but to different Vfm genetic groups (groups I to V as defined by variations observed in the genes vfmO and vfmP). According to these genomic comparisons, the polymorphism already observed in the biosynthesis genes vfmO and vfmP appears to be associated with the polymorphism in the ABC permease gene vfmG and the oxoacyl-ACP synthase gene vfmW.
In order to obtain further insights into the potential links between the polymorphism in the genes vfmO, vfmP, vfmG and vfmW, phylogenetic trees were constructed using the nucleotide sequences of the variable regions identified in the genes vfmO, vfmP, vfmG and vfmW (Fig. 4). For this phylogenetic analysis, we selected 18 representative strains corresponding to respectively, two triplets and five pairs of strains belonging to the same species but to different Vfm genetic groups (I, II, III or IV) and two strains of group V. In the four resulting trees, the sequences of the two strains of group V are clearly separated in a specific phylum, indicating a more distant origin. The sequences of each vfmO variant (vfmO1, vfmO2 and vfmO3) do cluster in a specific phylum, the variants vfmO1 (groups I and II) and vfmO2 (groups III and IV) being close to each other but well separated from the variants vfmO3 (group V) (Fig. 4A). Similarly, the sequences of each vfmP variant (vfmP1, vfmP2, vfmP3 and vfmP4) do cluster in a specific phylum, and the variants vfmP1 (group I) and vfmP2 (groups II and III) are close to each other but clearly separated from the two phyla formed by variants vfmP3 (group IV) and vfmP4 (group V) respectively (Fig. 4B). In the vfmG tree, the sequences of groups I, II and III are also close together but well separated from the two phyla formed by sequences of groups IV and V respectively (Fig. 4C). In the vfmW tree, the sequences of groups I, II and III cluster in a mixed phylum which is distinct from the two phyla formed by the sequences of groups IV and V, respectively (Fig. 4D).

Discussion
Biological validation of the prediction that the polymorphism in the genes vfmO and vfmP determines the production of different analogues of the Vfm QS signal Based on an in silico analysis of the polymorphism of the vfm locus, the 126 sequenced strains of Dickeya can be classified into five Vfm genetic groups, each group possessing the same combination of variants of the biosynthesis genes vfmO/vfmP (Table 1). Strains belonging to the groups I to V will be referred to as strains I to V respectively. In order to explore the hypothesis that each Vfm genetic group produces a different analogue of the Vfm QS signal, bioassays have been performed. The latter are based on the design and generation of four biosensor strains belonging to the three major Vfm genetic groups (I, III and IV) Accordingly, the designation 'analogues AI to AV' will be used to refer to putative analogues of the Vfm QS signal produced by the five Vfm genetic groups respectively.
The two biosensors of group I, I-uidA and I-luc, both gave a positive response only with samples prepared from any of the tested strains of group I or II (Tables 2 and 3). This specificity pattern of strains I and II indicates that analogues AI and AII correspond to an identical or, with regard to the structure, at least closely related molecule(s) which differ(s) significantly from the signalling molecules produced by strains III, IV and V. The samples prepared from strains III gave a positive response with the biosensor III-luc but not with the biosensors I-uidA, I-luc and IV-luc (Table 3), indicating that the signalling molecules of type AIII differ from the analogues AI/AII and AIV. The biosensor of group IV, IV-luc, gave a positive response only with samples prepared from any of the tested strains IV (Table 3). Additionally, the samples prepared from strains IV gave a positive response with the biosensor IV-luc but not with the biosensors I-uidA, I-luc and III-luc (Tables 2  and 3). This specificity pattern of strains from group IV suggests that the analogue AIV structurally differs from the analogues produced by strains I, II, III and V.
Altogether these results strongly suggest that strains I to IV produce at least three different signalling molecules of the Vfm QS signal (AI/AII, AIII and AIV), since only strains I and II seem to produce the same analogue. The isolation of the Vfm QS signals produced by strains belonging to different Vfm genetic groups, as well as the determination of their molecular mass and their chemical structure, will be required to definitively validate this hypothesis. Unfortunately, despite several attempts, the Vfm QS signal has not yet been isolated from any Dickeya strain (Nasser et al., 2013;Lv et al., 2019).
Regarding strains V, which include only two D. poaceiphila strains, the samples prepared from one strain gave a positive response with the biosensor III-luc (  Fig. 3. Characterization of variable regions in the vfm gene cluster between strains of Dickeya belonging to the same species but to different Vfm genetic groups. Top: Map of the vfm gene cluster. Genes are represented by arrows whose length is proportional to the gene length according to the scale indicated at the right. The genes vfmO, vfmP, vfmW and vfmG are indicated by the corresponding letter. White boxes below each of these four genes indicate the positions of the nucleotide sequences used to build the phylogenetic trees described in Fig. 4. A-K. Schematic representation of the alignment of nucleotide sequences of the vfm gene clusters for pairs of strains belonging to the same species but to different Vfm genetic groups (I, II, III or IV). Grey boxes correspond to the regions in which the aligned sequences of strain pairs share a nucleotide identity higher than 93%. White boxes correspond to the predicted recombined regions in which both aligned sequences share a nucleotide identity lower than 77%. Values of nucleotide identities shared by each strain pairs are indicated inside, below or above each box. strain gave no positive response with the biosensors of groups I, III or IV. Thus, a larger set of strains V and a biosensor strain corresponding to group V would be required to more precisely determine the sensing specificity of this group.
The NRPS substrate specificity-conferring signatures associated with the different variants of the genes vfmO and vfmP are consistent with the production of different analogues of the Vfm QS signal In silico analyses were performed to specify the nonribosomal code of 10 amino acids corresponding to the type I NRPS substrate specificity-conferring signatures. The polymorphism in the biosynthesis genes vfmO and vfmP predict the incorporation of different amino acids in the Vfm QS signal, which gives rise to the production of different analogues of the Vfm QS signal differing by the amino acid sequence. This prediction is supported by the biological data obtained using the biosensor strains, which strongly suggest that the four genetic groups I to IV produce three distinct analogues of the Vfm QS signal. The biological assays indicate that strains I and II synthesize functionally similar signals, suggesting that both combinations vfmO1/P1 and vfmO1/P2 result in the production of the same (or of a very similar) analogue. We thus compared the non-ribosomal code of 10 amino acids corresponding to the VfmP1 and VfmP2 signatures: they differ by one amino acid located in position 6 (Table 1) Table S1), except for the strain NCPPB 2929 sequenced in the present study. The positions of these nucleotide sequences in the vfm gene cluster are indicated in Fig. 3. Nucleotide sequences have been aligned with MUSCLE (Edgar, 2004), the phylogenetic tree was reconstructed using the maximum likelihood method implemented in the PhyML program (Guindon and Gascuel, 2003) and graphical representation of the phylogenetic tree was performed with TreeDyn (Chevenet et al., 2006). Numbers at nodes correspond to bootstrap values (500 replicates of the original alignment). Bootstrap values are indicated only for the nodes that separate the different variants of the genes vfmO or vfmP or the main phyla for the genes vfmG or vfmW. The scale bar represents the average number of substitutions per site. among type I NRPS signatures that specify the same amino acid (Stachelhaus et al., 1999;Eppelmann et al., 2002). This position of the signature almost always exhibits an aliphatic residue, either Ala or Gly which, indeed, are found at this position in the signatures associated with the variants vfmP1 and vfmP2, respectively. Moreover, the sequences of the variants vfmP1 and vfmP2 are close in the vfmP phylogenetic tree (Fig. 4B), supporting the high level of similarity between these two variants. These observations suggest that the NRPS signatures associated with the variants vfmP1 and vfmP2 most probably confer the same amino acid specificity, leading to the biosynthesis of the same analogue of the Vfm signal in strains I and II.
The biological experiments indicate that strains I/II and III produce functionally different analogues, suggesting that the combinations vfmO1/P1 or vfmO1/P2 (strains I/II) are linked to the production of an analogue different of those produced by the combination vfmO2/P2 (strains III). Interestingly, the VfmO1 and VfmO2 signatures also differ by only one amino acid (Fig. 2). However, this variable amino acid in position 2 of the signature is located in the entrance of the substrate-specificity pocket and, as such, it plays a key role in the interaction between the type I NRPS adenylation domains and their substrate (Kudo et al., 2019). Accordingly, the NRPS signatures associated with the variants vfmO1 and vfmO2 are predicted to confer different substrate specificities, leading to the incorporation of different amino acids into the Vfm signal.
The NRPS signatures associated with the variants vfmP2, vfmP3 and vfmP4 differ by two to four residues (Table 1), most of them in positions where an enormous variability plays a major role in determining the substrate specificity (Stachelhaus et al., 1999;Eppelmann et al., 2002). This strongly supports the hypothesis that these three VfmP signatures confer specificity to three different amino acids, respectively. A similar observation was made for the NRPS signatures associated with the variants vfmO2 and vfmO3, differing from the others by two out of 10 residues, and which are predicted to confer specificity to two different amino acids (Table 1). These in silico data support the hypothesis that strains III, IV and V produce three different analogues of the Vfm QS signal which also differ from the analogue produced by strains I and II. This hypothesis is clearly in agreement with the experimental data which indicate that strains III and IV produce two different analogues also differing from those produced by strains I and II.
The polymorphism in the biosynthesis gene vfmW is linked to the polymorphism in the biosynthesis gene vfmP A polymorphism was also observed within the biosynthesis gene vfmW of the different strains (Fig. 3). The phylogenetic tree based on the alignment of the variable region of the gene vfmW from 18 strains indicates that vfmW sequences from strains IV and V cluster in two specific phyla compared to the vfmW sequences from strains I, II and III (Fig. 4D). Strains IV and V respectively possess the variants vfmP3 and vfmP4 which are not shared by the other Vfm genetic groups I, II and III. This observation suggests that the polymorphism of vfmW is associated with the polymorphism of vfmP. A suitable explanation would be that the amino acid incorporated by the enzyme VfmP into the Vfm QS signal could be the substrate of the designated oxoacyl-ACP synthase VfmW. Thus, the polymorphism of the gene vfmW would result from an adaptation to variations of this amino acid. As the NRPS signatures associated with the variants vfmP1 and vfmP2 are proposed to encode the same amino acid, the substrate of VfmW could be identical for strains I, II and III. This could explain why the vfmW sequences of strains I, II and III cluster together but not with the vfmW sequences of strains IV or V (Fig. 4D). Surprisingly, the vfmW sequences of the D. chrysanthemi strains II and III seem to be distant from the other vfmW sequences of the groups II and III.
Use of the biosensor strains strongly suggests that strains I, III and IV produce three different sensor/ receptor proteins of the Vfm QS signal The two biosensor strains of group I (I-uidA and I-luc) are activated only by samples prepared from strains I or II (expected to produce the same analogue AI/II) (Tables 2  and 3). The biosensor strain of group IV (IV-luc) is specifically activated by samples prepared from strains IV (Table 3). These patterns of specificity demonstrate that the sensory systems involved in the recognition of the Vfm QS signals AI/II and AIV are highly specific of their cognate signal. In contrast, the biosensor strain of group III (III-luc) gave a positive response with samples prepared from all strains III but also from several strains I, II, or V (Table 3). This low specificity indicates that the Vfm sensor/receptor system of group III is less selective than those of other groups and is compatible with several analogues of the Vfm QS signal. However, the low reproducibility observed between replicates using the biosensor III-luc with samples prepared from six strains of group I, one strain of group II and one strain of group V (Table 3) suggests that the affinity of the sensor/receptor system of group III is lower for the analogues AI/II and AV than for the analogue AIII.
Altogether, these results show that the groups I/II, III and IV produce three different variants of the Vfm protein responsible for the recognition of the Vfm QS signal. The vfm gene cluster encodes the two-component regulatory system VfmH/VfmI. By analogy with classical two-component systems, the histidine kinase VfmI could have been the sensor/receptor of the Vfm QS signal. However, no vfmI polymorphism associated with those observed in the signal biosynthesis genes vfmO/P has been identified, as shown by the vfmI phylogenetic tree which correlates with trees of housekeeping genes but not with the trees of vfmO or vfmP (Fig. 4; Fig. S2). Since a polymorphism resulting from its adaptation to the different analogues is expected for the gene encoding the signal sensor/receptor, VfmI does not appear to be a suitable candidate for this function. Although the transcriptional regulator VfmH was experimentally shown to be involved in the transduction of the cellular response to the Vfm QS signal (Nasser et al., 2013), to date, the function of the histidine kinase VfmI has not been experimentally investigated.
The polymorphism of the ABC permease gene vfmG suggests an adaptation of the membrane protein VfmG to the specific binding of different analogues of the Vfm signal In contrast to the histidine kinase vfmI gene, the ABC permease gene vfmG shows a polymorphism associated with those of the genes vfmO/P, as the clustering of the vfmG sequences observed on the vfmG phylogenetic tree coincides with the Vfm genetic groups defined on the basis of the vfmO/P variants (Fig. 4C). According to these phylogenetic data, the polymorphism of the ABC permease protein VfmG could be linked to an adaptation of the protein VfmG to specifically bind the different analogues of the Vfm QS signal.
To support this hypothesis, we noticed that several ABC permeases have been described as acting as cosensors/receptors of two-component regulatory systems (Piepenbreier et al., 2017). An interesting example is the ABC permease BceB which acts as an accessory sensor/receptor in the two-component regulatory pathway controlling the antibiotic resistance system Bce of Bacillus subtilis (Koh et al., 2020). This example is of particular interest, as the histidine kinases of the Vfm and Bce systems both exhibit an atypical topographical organization of their functional domains. In contrast to classical histidine kinases, they both lack any predicted ligandbinding domain, as their two N-terminal transmembrane helices are separated by a very short loop (data not shown). The absence of a ligand-binding domain in the histidine kinase VfmI would explain why the polymorphism in the gene vfmI is not genetically linked to the polymorphism in the signal biosynthesis genes vfmO/P. Thus, VfmG is a better candidate than VfmI for direct binding to the Vfm signal. However, additional experimental data will be needed to clarify the role of VfmG as sensor/receptor of the Vfm QS signal.
The biological assays show that spontaneous null mutations in the hypothetical protein vfmX gene or in the multidrug transporter vfmC gene do not affect the functioning of the Vfm system The fully sequenced genome of the wild type D. dadantii strain DSM 18020 T (CFBP 1269 T , NCPPB 898 T ) contains a frameshift in the gene vfmX (Table S1) also observed in the draft genome available for this strain (contig AOOE01000040.1 in GenBank). Since this strain gave a positive response with the biosensors I-uidA and I-luc (Tables 2 and 3), this vfmX frameshift does not block the production or secretion of the Vfm QS signal. This is consistent with the previous observation that a vfmX engineered null mutant of D. dadantii 3937 is not affected for the Vfm QS activity ( (Nasser et al., 2013), vfmX is annotated as ID16066 in this reference). The gene vfmX, encoding a protein of unknown function, could therefore only play a minor role in the Vfm QS system. Alternatively, its inactivation may be functionally complemented by another unidentified gene present in the D. dadantii genome.
Four D. dianthicola strains, including CFBP 2982, exhibit a 1-kb deletion in the gene vfmC (Table S1). In addition, a defect in assembly of the vfmC gene is observed in five other D. dianthicola genomes (Table S1), suggesting that the vfmC gene is also inactivated in these strains. Since strain CFBP 2982 gave a positive response with both biosensors I-uidA and I-luc (Tables 2 and 3), the 1-kb deletion in the gene vfmC does not prevent the production or secretion of the Vfm QS signal. The gene vfmC, encoding a multidrug transporter, was proposed to be involved in the secretion of the Vfm QS signal synthesized by the bacteria (Nasser et al., 2013). However, the inactivation of this gene may be functionally complemented by another multidrug transporter gene present in the D. dianthicola genome.
The polymorphism in the vfm genes could have an impact on the pathogenicity and epidemiology of Dickeya strains The analogy of the Vfm system with the Agr QS specificity groups, which are associated with variations in virulence in S. aureus (Geisinger et al., 2012), raise further questions on the impact of the allelic variations within the vfm genes on the phytopathogenicity of the Dickeya strains. Although there are, to date, no comparative studies on the virulence of Dickeya strains on the basis of their appurtenance to different Vfm genetic groups, we just noticed that almost all strains of the two economically important species D. solani and D. oryzae belong to the Vfm genetic group III (Fig. 2), suggesting that the Vfm QS system could be particularly efficient in this group. In contrast, all strains of D. aquatica or D. lacustris, two species not linked to agricultural damages but found in water, belong to group IV. Strains of the poorly represented species D. poaceiphila, found to be weakly virulent under laboratory conditions (Hugouvieux-Cotte-Pattat et al., 2020a), are the sole members of group V (Fig. 2). However, this species may have a quite restricted host range and niche as D. poaceiphila isolates were found only on plants of the Poaceae family (grasses) and only in Australia (Hugouvieux-Cotte-Pattat et al., 2020a).
A recent study proposed that an allelic variation in the hypothetical protein VfmB is associated with variations in the bacterial aggressiveness and competitiveness in D. solani (Blin et al., 2021). Four D. solani strains possessing the allele VfmBSer were shown to more efficiently damage potato tubers than four D. solani strains possessing the allele VfmBPro. However, this VfmB allelic variation is not linked to any variation in the specificity of the Vfm QS system since the eight D. solani strains tested share exactly the same proteins VfmG, VfmO, VfmP and VfmW (amino acid identity of 100%). As most D. solani strains, these eight strains belong to the genetic group III and positively respond to the biosensor III-luc (Table 3).
Both predictive in silico analyses and experimental results demonstrated that different strains of the same Dickeya species could belong to different Vfm genetic groups (Fig. 2). Thus, signalling via the Vfm QS system is limited to strains belonging to compatible groups. This observation raises questions regarding the cohabitation inside the same host organism or the same environment of strains belonging to different Vfm genetic groups (Fig. 2): for example, the species D. fangzhongdai includes strains of groups I and III. In this species, strains of group I can act as cheaters when they live next to strains of group III because group III strains respond to the signalling molecule AI while strains of group I do not respond to the signalling molecule AIII. Regarding species containing strains of group III and IV, such as D. chrysanthemi, D. fangzhongdai, D. parazeae or D. zeae, cohabitation of strains unable to communicate via the Vfm QS system could occur since strains of group III do not respond to the analogue AIV and inversely strains IV do not respond to the analogue AIII. Similarly, the communication via Vfm is impossible between strains I and strains IV, for example in the species D. dadantii. On the contrary, the absence of vfm polymorphism was observed in a few Dickeya species. While only a limited number of strains have been sequenced for D. aquatica, D. lacustris or D. poaceiphila, D. dianthicola is a good example of species homogeneity as all 33 analyzed strains of this species belong to the Vfm genetic group I (Fig. 2, Table S1), as well as the 16 recently sequenced strains (data not shown, Curland et al., 2021).

Conclusion
This study demonstrates the presence of polymorphism in a few genes of the vfm cluster of Dickeya strains and it provides strong evidence that the strain-specific polymorphism observed in the biosynthesis genes vfmO/P results in the production of at least three different analogues of the Vfm QS signal. This genetic polymorphism has a direct consequence on the specificity of the Vfm QS system. Interestingly, strains belonging to a same Dickeya species could belong to different Vfm specificity groups, opening new questions that remain to be explored, for instance regarding the impact of the Vfm polymorphism on the pathogenicity and the epidemiology of Dickeya strains. Does the efficiency of the Vfm QS system vary according to the Vfm polymorphism? Is this strain polymorphism linked to variations in virulence? Do strains belonging to different Vfm specificity groups coexist naturally in a same niche?
The new information about the Vfm QS system evidenced by this study could be of great interest for developing quorum quenching techniques in order to disrupt communication between Dickeya cells, with the aim to reduce their virulence. The discovery of a strainspecific polymorphism in the genes vfmO/P linked to variations in the specificity of production and recognition of different Vfm analogues is an essential knowledge for the development of control methods targeting the Vfm QS system of Dickeya strains.

Bacterial strains, media and culture conditions
The bacterial strains of different Dickeya species were obtained from international culture collections. All strains were grown at 30 C in rich medium LB or in minimal medium M63 (Miller, 1992) containing glycerol at a final concentration of 2 g L À1 as a carbon source. Kanamycin (Km) was used at a final concentration of 20-50 μg ml À1 . Ampicillin was used at a final concentration of 20-50 μg ml À1 . Chloramphenicol was used at a final concentration of 12.5 μg ml À1 . For solid media, 15 g L À1 agar was usually added.
Detection of the Vfm signal using the Vfm biosensor I-uidA The biosensor I-uidA corresponds to the strain A5243 previously prepared from the model strain D. dadantii 3937 by introducing a vfmE-uidA Km R mutation (Nasser et al., 2013). The biosensor I-uidA was used to test the cell-free culture supernatants of 52 wild-type strains belonging to different Dickeya species. The strains were grown in M63 medium supplemented with glycerol. After 24 h at 30 C, cells were discarded by centrifugation at 12 000g for 4 min and the culture supernatants were filtered and stored at 4 C. These supernatants were added in a 1/10 ratio to early exponential phases of A5243 cultures in M63 medium supplemented with glycerol and kanamycin. After 24 h at 30 C, the response of the reporter system was detected by a spectrophotometric assay of the uidA product. ß-Glucuronidase (GUS) activity was measured by monitoring the hydrolysis of pnitrophenyl-ß-D-glucuronide to yield p-nitrophenol that absorbs at λ = 405 nm (Bardonnet and Blanco, 1992). Supernatants of the wild-type strain 3937 and of the vfmE mutant A5243 were used as positive and negative controls respectively. A sample was considered as positive if its GUS activity was at least five times higher than that of the negative control.

Construction of the Vfm biosensor I-luc
The 5 0 regulatory region of the vfmE gene of strain D. dadantii 3937 was amplified by PCR with the primers RRvfmEdeb XhoI (GGCTCGAGAGGTCGTTTCCTGTTC ATCTGCGTC) and RRvfmEfin BamHI (GGGGATCCG TAGGTGTTCTGCAAGCTCATG). After treatment with both restriction enzymes XhoI and BamHI, the resulting PCR fragment was cloned between the restriction site XhoI and BglII of the plasmid pUCTer-Luc-Cm (Jiang et al., 2015). The resulting plasmid was introduced by electroporation in the vfmE mutant A5243 (Nasser et al., 2013).

Construction of the Vfm biosensors III-luc and IV-luc
The biosensors III-luc and IV-luc were derived from strains D. solani IPO 2222 and RNS05-1-2A, respectively. They correspond to an in frame deletion of the vfmE coding sequence associated with its replacement by the luciferase gene luc (ΔvfmE-luc constructions). They were obtained using a technique involving a suicide plasmid and the sacB counter-selection (Edwards et al., 1998). The vector pRE112, an R6K-based suicide plasmid carrying the genes sacB and cat (Cm R ), has previously been successfully used for allelic exchange in D. dadantii (Royet et al., 2019). The plasmid pRE112-ΔvfmE-luc was constructed by cloning simultaneously three PCR fragments, consisting of two 1-kbp DNA fragments corresponding to the upstream and downstream vfmE coding sequence of D. solani IPO2222 and the luc coding sequence, into the SacI/KpnI digested vector pRE112, using the Gibson's assembly method (Gibson et al., 2009). After CaCl 2 -mediated transformation of E. coli DH5α λpir competent cells, transformants were selected onto LB-Cm plates. Plasmids were extracted with the NucleoSpin Plasmid kit (Macherey-Nagel, Düren, Germany), checked by restriction digestion and PCR analysis. They were transferred by transformation into competent cells of E. coli MFDpir, a strain producing the RP4 conjugation machinery which allows the transfer of plasmids into several Gram-negative bacteria by conjugation (Ferrières et al., 2010). After mating of E. coli MFDpir with D. solani IPO 2222 or RNS05-1-2A, cells were spread onto LB-Cm plates to select the first recombination of the pRE112-ΔvfmE-luc plasmid into the D. solani chromosome. After a second isolation on this medium, colonies were spread onto LB plates supplemented with 5% sucrose and incubated at 20 C for 2-3 days to allow the second recombination event leaving the ΔvfmE-luc constructions derived from D. solani IPO 2222 or RNS05-1-2A, respectively. Sucrose-resistant colonies were replicated onto LB-Cm plates to check plasmid loss. The correct structure of the vfmE-luc fusion was confirmed by PCR analysis.
Detection of the Vfm signal using the biosensors I-luc, III-luc and IV-luc A biological assay allowing the detection of the Vfm QS signal was carried out in 96-well plates (Thermo Scientific Nunc, Rochester, NY, USA) displaying wells shaped with a flat transparent bottom to allow measurement of the optical density and a white opaque wall to allow measurement of bioluminescence without interference from the neighbour wells. As a standard procedure, 20 μl of the sample to be tested (a bacterial culture supernatant) was mixed in each well with 180 μl of a 5-h culture in LB medium supplemented with kanamycin and ampicillin (for the biosensor I-luc) or chloramphenicol (for the biosensors III-luc and IV-luc) and with Xenolight™ D-luciferin (Perkin-Elmer, Waltham, MA, USA) at a final concentration of 225 ng ml À1 . Optical density at 600 nm and bioluminescence were concomitantly measured every 15 min for 15 h in a Tecan Spark plate reader (Tecan, Grödig, Austria). The internal temperature was set to 28 C and an orbital agitation (150 rpm, orbital radius 4 mm) was continuously applied to the plate between each measurement point. Results are expressed as the mean of the 61 bioluminescence measurements under the codename 'mean61'. A sample was considered as positive if its 'mean61' is at least twice the 'mean61' of the negative control (a cell-free supernatant of a culture of the vfmE mutant A5243).
Verification of the species appurtenance by sequencing a gapA PCR product For strain identification, the gene gapA was amplified by PCR performed using the Illustra™ PuReTaq™ Ready-To-Go™ kit (GE Healthcare, Chicago, IL, USA) on bacterial cell lysates with the primers gapAF and gapAR (Cigna et al., 2017). Sequences of the PCR products were determined by Sanger sequencing (Biofidal, Vaulxen-Velin, France).

Genome sequencing
The genomic DNA of each strain was extracted using a NucleoSpin R bacterial DNA purification kit (Macherey-Nagel). Illumina sequencing and assemblies of the reads were performed by MicrobesNG (Birmingham, UK). The resulting draft genome sequences were used to determine the sequence of the complete vfm locus of the strains D. fangzhongdai NCPPB 2929 (GenBank accession no. MZ611617), D. chrysanthemi CFBP 1270 (GenBank accession no. MZ611618), D. dadantii NCPPB 3065 (GenBank accession no. MZ611619) and D. dadantii CFBP 3694 (GenBank accession no. MZ611620).

Comparative analysis of the genomic sequences available in GenBank
The sequence of the vfm locus of 126 genomes of Dickeya available until April 2021 was retrieved from GenBank. The type I NRPS-like signature of the proteins VfmM, VfmO and VfmP was determined with NRPSpredictor2 (Röttig et al., 2011). Amino acid identities between the different allelic forms of the proteins VfmO, VfmP, VfmG and VfmW were determined with Clustal Omega (Sievers et al., 2011). Alignment of the nucleotide sequences of the vfm loci from different pairs of strains of Dickeya was performed with Nucleotide BLAST (Basic Local Alignment Search Tool) or Clustal Omega (Sievers et al., 2011). Transmembrane helices within the proteins VfmG and VfmI were predicted with HMMTOP (Tusn ady and Simon, 2001). Conserved domains within the proteins VfmF, VfmG, VfmH and VfmI have been searched with the CDD/SPARCLE database (Lu et al., 2020). Phylogenetic trees were generated with a ready-to-use pipeline [(Dereeper et al., 2008), Phylogeny.lirmm.fr]. Sequences were aligned with MUS-CLE (Edgar, 2004), and each phylogenetic tree was reconstructed using the maximum likelihood method implemented in the PhyML program (Guindon and Gascuel, 2003). The graphical representation of the phylogenetic tree was performed with TreeDyn (Chevenet et al., 2006).