The early phases of host-colonisation are crucial for pathogenic bacteria . However, little is known about the importance of early phases in determining host-specificity for plant pathogenic bacteria. Emphasis has been put on later phases of infection, as illustrated by the demonstration of the role of T3Es in host-specificity [1, 2, 36]. Host-recognition can be considered the first instance of host-pathogen interaction allowing bacteria to colonise. The importance of chemotaxis in plant-bacteria interactions has been clearly documented in some cases. For example the chemotactic mutant Ralstonia solanacearum is unable to colonise its host when inoculated into the soil, whereas it remains fully pathogenic when infiltrated inside plant tissues . Thus evolutionary processes of host-specificity would also be driven by selective forces at the very first steps of host-colonisation. Our data show evidence of such adaptive processes for numerous genes involved in chemotactic attraction, environment sensing, and adhesion to surfaces. These mechanisms precede the infection of a plant by xanthomonads. The mechanisms facilitating plant penetration and thus allowing infection are chemotaxis, aerotaxis, and fast multiplication inside host tissue, which relies on adhesion.
According to MK-tests results, many candidate genes undergo positive selection during bacterial colonisation on plant tissue. Of the 24 genes common among repertoires of the studied pathovars (vesicatoria, citri, and phaseoli) of X. axonopodis and X. campestris pv. campestris, nearly half are subject to positive selection. This proportion is almost equal among the two gene families (MCPs and adhesins). The McDonald-Kreitman procedure tests for adaptive divergence between two species. The test is known to be robust to non-equilibrium demography [37, 38] and to recombination . Using a small sample size would reduce the power of the test. The lack of power to detect selection would increase the risk of false negatives. Thus our results are conservative and can not be interpreted as false positive. Charlesworth and Eyre-Walker  showed that about 50% of amino-acid substitutions surveyed in the enteric bacterial genomes were subject to adaptive evolution. Thus both recognition and adhesion should be considered as selective steps for bacterial colonisation. To our knowledge this is the first report of positive selection acting on MCPs in plant-pathogenic bacterium. Our results are consistent with those of Chen et al.  and Petersen et al.  showing that positive selection acts on genes encoding surface structures of E. coli cell. These genes encode regulators of LPS O-antigen chain length, putative adhesins that affect biofilm formation, ferrichrome-iron receptors, two outer membrane porins, and more [41, 42]. In xanthomonads, the extracellular appendage of the Hrp pilus evolved under the constraint of positive selection likely to avoid recognition by plant defense surveillance systems .
Most sites that were found under selection in candidate genes by PAML analysis were located in conserved domains predicted to play a role in perception for MCPs and in adhesion for XadA1. Indeed, the Tar domain of chemoreceptors directly binds to aspartate and related amino acids . The Hep-Hag motif is found in the passenger domain of adhesins . This domain is known to contribute to the binding activity of invasins/agglutinins . This result is another argument in favour of selection pressures acting on these genes in link with the ecological behaviour of the strains. Detection of positive selection using the branch-sites model implemented in PAML has some power limitations especially if only three sequences are used. Our results, however, should be considered conservative as they reduced the number of false positives.
Many bacteria assemble multifunctional proteic structures on their surfaces that serve for adhesion. This feature might be an adaptation to different environmental conditions and, in the case of pathogenic bacteria, to different hosts or host tissues . In fact, adhesins are involved in various processes leading to host colonisation and transmission to seed by plant-pathogenic bacteria. For example, Tfp serves remarkably diverse functions, including twitching motility, cell to cell adhesion, and thus microcolony and biofilm formation . Tfp is an important virulence factor for vascular and non vascular plant pathogens [47, 48]. Moreover, Darsonval and colleagues  showed that PilA is involved not only in adhesion but also in transmission to seed, and the mutation of pilA in strain CFBP4834 of X. axonopodis pv. phaseoli GL fuscans leads to lower pathogenicity on bean (P. vulgaris). Additionally, YapH, an hemagglutinin, is required for adhesion to seed, leaves, and abiotic surfaces.
An interesting consequence of strong differential selection pressures by host is a specialisation at some early steps (i.e. chemotactic attraction) of host colonisation by xanthomonads. Character displacement at early stage of host colonisation should make infection more efficient by preventing competition for habitat between strains [49, 50]. In fact, xanthomonads are known to be phenotypically very homogeneous except in pathogenicity. Lack of selective pressures in host colonisation would lead to colonisation by a wide range of incompatible strains. Colonisation of specialized pathovars would therefore be less successful. Indeed, resource allocation would be redirected in favour of competition detrimental to pathogenicity. Finally, such host-isolation could act as an ecological isolating barrier to limit recombination between differentially adapted pathogenic strains. As in eukaryotic organisms, ecological differences in bacteria are known to promote speciation . Indeed, habitat sharing would allow recombination between strains that belong to different pathovars and that consequently produce strains that may reveal genetic incompatibilities.
This comparative analysis of repertoires of MCPs, STCRS, TBDTs, and adhesins provides useful insight into bacterial behaviour. First, the number of MCPs and more generally sensors is higher in Xanthomonas strains than in E. coli and Salmonella . E. coli and Salmonella have only five MCPs whereas strains Xav 85-10, Xac 306, and Xcc ATCC339313 have about 20 MCPs; strain Xoo KACC10331 has only about 10 MCPs. The large numbers (14 in Xav 85-10, 10 in Xac 306, 8 in Xcc ATCC339313, and 7 in Xoo KACC10331) of MCPs and other sensors repeated in tandem are unusual in bacteria, suggesting a prominent role in the life style of Xanthomonas [17, 31, 32]. Second, repertoires of MCPs, STCRS, TBDTs, and adhesins differed among the majority of pathovars and genetic lineages belonging to the tested Xanthomonas spp. and displaying different host range. Repertoires of genes coding sensors and adhesins comprised core and variable gene suites. Some genes under study were not intra-specifically conserved and hence belong to the accessory genome. Repertoires of genes involved in attraction and adhesion may evolve by gene gain or loss, probably after duplication events. In the case of Drosophila, the size of repertoires of genes encoding olfactory and gustatory receptors varies through gene duplication, pseudogenization, and gene loss. The changes among species of Drosophila show that these receptors have changed during species divergence, and their evolution might reflect species' adaptation to their chemical environment . Similarly, in xanthomonads, the variable set of sensors and adhesins may be involved in the recognition of specific components allowing strain adaptation to a particular set of hosts. Moreover, we identified signals of adaptive divergence have been identified on such genes of the variable set.
This study showed that among the same genus, Xanthomonas, the majority of pathovars and genetic lineages belonging to different species (X. axonopodis, X. campestris, X. vasicola, and X. oryzae) displayed different and unique repertoires of MCPs, STCRS, TBDTs, and adhesins while they displayed different host range. Note that the distribution of sensor and adhesin genes does not necessarily correlate with strain phylogeny. Indeed, bacteria as phylogenetically distant as X. axonopodis pv. vasculorum and X. vasicola. pv. vasculorum  share a common repertoire of sensor and adhesin genes (Figure 2) and a common host: sugarcane. This case illustrates the sharing of a common ecological niche (symptomatic host) by two phylogenetically distant bacteria.
Our results suggest that adaptation to host involves pathoadaptation but also asymptomatic colonisation steps. Indeed, several pathovars and genetic lineages shared the same repertoires whereas they are known to infect different crops. This means that they could share the same asymptomatic host range but develop symptoms on a restricted number of different host plants. This is the case for the pathovar vignicola and the genetic lineage GL2 of pathovar phaseoli. X. axonopodis pv. phaseoli GL2 and pv. vignicola strains may detect similar plant-originated molecules potentially conserved among their host plants. This hypothesis is supported by the fact that X. axonopodis pv. phaseoli GL2 and vignicola both infect legumes. X. axonopodis pv. phaseoli GL2 infects Phaseolus spp. and X. vignicola infects Vigna unguiculata; these legumes belong to the Milletioid clade and are phylogenetically closely related . Interestingly, cross inoculations would provide insight on the ecological behaviours of these two pathogens (survival, colonisation and chemotaxis responses). The four genetic lineages (fuscans, GL1, GL2, GL3) of pathovar phaseoli, which all share a common host (P. vulgaris), present distinct repertoires of MCPs, STCRS, TBDTs, and adhesins. Genetic lineage fuscans, GL2, and GL3 are phylogenetically closely related and belong to rep-PCR group 9.6 whereas GL1 is distant and belongs to rep-PCR group 9.4 [2, 30, 54]. These four distinct genetic lineages have different T3E repertoires but clustered together on the dendrogram constructed on the matrix of presence/absence of T3Es genes, supporting the hypothesis of an adaptive pathological convergence on bean . Our results suggest that, upstream of the invasive pathological stage, the four genetic lineages have different ecological behaviours. Colonisation does not necessarily lead to infection and then may not be under the same adaptive processes as host infection. We can speculate that each lineage of X. axonopodis pv. phaseoli can be found on different asymptomatic hosts.
Overall, our findings indicate that plant pathogenic bacteria belonging to different pathovars have evolved different set of genes allowing them to specifically detect favourable hosts on which they can settle. These results support a recent theory termed inverse-gene-for-gene in which infectiousness is determined by pathogen recognition of hosts signals and/or receptors . This theory is an alternative to the gene-for-gene model in which the pathogen is recognized by the host., Here, in agreement with this theory we show that many pathogen genes involved in host recognition evolved under adaptive divergence. Such a selective pressure on genes encoding for recognition of specific hosts strongly accounts for coevolutionary dynamics where pathogens are always adapting their sensors in response to hosts changes in exudates and surface structures.
Plant pathogenic xanthomonads are associated with aerial parts of plants. They are not usually encountered in other environments. Plant pathogenic pseudomonads colonise non-host habitats such as snow or water . Our attempts to isolate xanthomonads from such environments were, however, unsuccessful (our unpublished data). Saprophytic survival of xanthomonads in soil is very poor. Apart from their primary host, many xanthomonads can survive for long periods in association with weeds that grow naturally in crops. It is not yet known which weeds are susceptible to colonisation by each plant pathogenic xanthomonad. We refer in our study to the main crop contaminated by each pathovar, which certainly represents the major opportunity for bacterial multiplication.