Strains and plasmids used are listed in Supplementary Table 1. P. fluorescens F113 and derivatives were grown in SA medium (Scher and Baker, 1982), with shaking overnight at 28°C. Purified agar (1.5%; Pronadisa, Spain) was added for solid medium. Pseudomonas strains were grown in LB medium for RNA and protein extraction assays. When it was necessary, antibiotics were used at the following concentrations (μg ml^-1): rifampicin 100; kanamycin 50; gentamicin 3; spectinomycin 100; and tetracycline 70. E. coli strains were grown in LB medium with shaking overnight at 37°C. In this case, antibiotics were used at the following concentrations (μg ml^-1): kanamycin 25; gentamicin 10; ampicillin 100; spectinomycin 25; chloramphenicol 30; and tetracycline 10.

Swimming was tested on SA medium with 0.3% purified agar. Bacteria were inoculated with a sterile toothpick in plates of 50 mm in diameter containing 12 ml of SA medium and incubated at 28°C. Swimming haloes were measured after 24 h. Every assay was performed three times with four replicates each time. Average swimming halo diameters (mm) and standard deviations were calculated and all dates were compared using statistic application software SPSS. Statistical significance was calculated by the Bonferroni test of variance analysis (ANOVA; P < 0.05).

Mutants generation was performed by directed mutagenesis. Primers were designed (Supplementary Table 2) to amplify small fragments of flhDC, fliC2, cyaA, and vfr genes [300–500 nucleotides (nts); flhDC-F/R; fliC2-F/R; cyaA-F/R; vfr-F/R]. First, these internal fragments were cloned into cloning pGEM^®T-easy vector and then, into the suicide vector pK18mobsacB (Schäfer et al., 1994) or pG18mob2 (Kirchner and Tauch, 2003; Supplementary Table 1). The constructs were introduced into F113 or F113-derived strains by triparental mating (in LB medium at 28°C during 16 h) using helper plasmid pRK600 (Finan et al., 1986; Supplementary Table 1) and selected for homologous recombination. The same method was used in the construction of double and triple mutants. All mutants were checked by Southern blotting and PCR. Primers were designed to amplify flhDC (flhDC-F2/R2; Supplementary Table 2). Overexpression of flhDC gene was achieved by cloning them under the control of the IPTG-inducible promoter present in the pVLT31 plasmid (de Lorenzo et al., 1993).

Pseudomonas fluorescens F113 and derivatives were grown in LB medium until optical density (OD₆₀₀) of 0.8. Formvar-coated grids were placed on the top of a drop of bacterial cells for 30 s to allow bacterial adhesion. Grids were stained for 1 min 15 s with a 1% solution of potassium phosphotungstate and washed with a drop of sterile water (thrice for 30 s). In order to visualize the preparations, the microscope JEOL JEM 1010 (100 kV) was used (Laboratorio de Microscopía de Transmisión del Servicio Interdepartamental de Investigación, UAM) and the images were taken by the BioScan de Gatan camera. The images were analyzed through DigitalMicrograph 3.1 analyze system.

Standard methods (Sambrook et al., 1989) were used for DNA extraction, gene cloning, plasmid preparations, and agarose gel electrophoresis. Southern blots were performed with a non-radioactive detection kit, and a chemiluminescence method was used to detect hybridization signals according to the instructions of the manufacturer (Roche Boehringer Mannheim). PCR reactions were performed using the Tth enzyme (Biotools) under standard conditions. DNA sequencing was done by chain-termination method using DyeDeoxy terminator cycle sequencing kit protocol as described by the manufacturer (Applied Biosystems).

Total RNA was extracted using Trizol^® according to manufacturer’s specifications (Invitrogen) from F113 wild-type and derivatives grown at 0.8 OD₆₀₀ in LB medium. Genomic DNA was removed by RQ1 RNase-Free DNase treatment (Promega) for 30 min at 37°C. Later, RNA was purified using Trizol^®. The concentration of RNA was spectrophotometrically determined in a Nanodrop^® and integrity was verified in denaturing agarose gels. RT-PCRs were carried out using Illustra Ready-To-Go^TM RT-PCR Beads kit from Amersham GE Healthcare. qRT-PCRs were performed in two steps: a first step of cDNA synthesis using the SuperScript^®III First-Strand Synthesis System from Invitrogen and a second step of qPCR using the Power SYBR^®Green PCR Master Mix from Applied Biosystems. Relative expression was estimated as 2^-ΔΔCt using 16S DNA as housekeeping and wild-type ΔCt values as calibrator, as described by Livak and Schmittgen (2001). Every assay was performed three times with three replicates each time.

Proteins were extracted from 200 ml exponential phase grown cultures (OD₆₀₀ = 0.8). In order to detach the flagella, the cultures were agitated by vortexing for 2 min and then centrifuged for 15 min at 7,500 rpm and 4°C. Extracellular proteins were extracted from the supernatant by precipitation for 10 min at RT with 10% (w/v) desoxycholic acid and then, 2 h at 4°C with 10% (w/v) trichloroacetic acid, followed by two washes with chilled acetone, and were finally resuspended in Laemmli buffer (Laemmli, 1970). Proteins were resolved by 12% SDS-PAGE and stained with Coomassie blue. The same electrophoretic conditions were used for Western blotting. Acrylamide gels were transferred onto nitrocellulose membranes for 1 h under standard conditions. The membranes were incubated with a 1:10,000 dilution of an anti-flagellin antiserum (de Weger et al., 1987) for 16 h at 4°C and then with a peroxidase-tagged secondary antibody (anti-rabbit immunoglobulin) for 1 h at RT. The enhanced chemiluminescence (ECL) method and Hyperfilm ECL (Amersham Biosciences) were used for development.

A modification of the “root-tip competitive colonization assay” (Simons et al., 1996) was used. Alfalfa seeds (Medicago sativa var. Resis) were surface disinfected in 70% ethanol for 2 min and then in diluted bleach (1:5; final sodium hypochlorite concentration, 1%) for 15 min and rinsed thoroughly with sterile distilled water. Seed vernalization was performed at 4°C for 16 h and was followed by incubation in darkness at 28°C for 1 day for germination. Germinated alfalfa seeds were sown in Leonard jar gnotobiotic systems (Villacieros et al., 2003) using ca. 3.5 l of sterile perlite as the solid substrate and 8 mM KNO₃-supplemented FP (Fahraeus, 1957) as the mineral solution (500 ml/jar; replenished every 2 days). After 2 days, alfalfa seeds were inoculated with 10⁸ cells of the appropriate strains. In competition experiments, strains were inoculated at a 1:1 ratio. In all cases 1 ml diluted culture was inoculated per plant. Plants were maintained under controlled conditions for 2 weeks: 16 h in the light at 25°C and 8 h in the dark at 18°C. Bacteria were recovered from the rhizosphere by vortexing the root tips (last centimeter of the main root) for 2 min in a tube containing FP medium and plating the appropriate dilutions on SA plates supplemented with appropriate antibiotics. Every experiment was performed three times with three replicates each time, and every replicate contained at least 20 plants.

Pseudomonas fluorescens F113 genomic sequence (NC_016830.1) from position 876600 to 961608 was used as query in Blastn v2.3.0+ (Camacho et al., 2009) to compare against complete and draft Pseudomonas genomes in NCBI database (nt and WGS on March 30, 2016). Own designed Python scripts were used to filter BLAST results. Genome annotations were obtained from NCBI database when possible; also RAST annotation pipeline (Aziz et al., 2008) was used. Synteny of the second flagellar region and its context was assessed from annotations and represented by using self-written Perl scripts. Protein homology was checked by Blastp (Camacho et al., 2009).

Phylogenomic distances between genomes harboring the flagellar island and close relative genomes was calculated with the genome-to-genome distance calculator (GGDC) 2.1 web service http://ggdc.dsmz.de (Meier-Kolthoff et al., 2013). Nucleotide sequence from each of the 45 ORFs from the flagellar island were retrieved from genomic annotation and aligned with Clustal Omega (Sievers et al., 2011). Resulting alignments were concatenated with own designed Python scripts following the order in P. fluorescens F113. The concatenated sequences were used to infer the phylogeny using maximum-likelihood method, Tamura–Nei model, and 1,000 bootstrap replicates using MEGA software (v7; Kumar et al., 2016).

Scripts are provided as Supplementary File 1. Statistical analyses were carried out with R software (R Core Team, 2013).

Analysis of the genome of P. fluorescens F113 showed that besides the genes required for the synthesis of the flagella, it harbors an additional 41 kb genetic region containing 45 ORFs that also showed homology with flagellar genes (Redondo-Nieto et al., 2012). This region is not present in other closely related strains belonging to the same phylogenomic group of P. fluorescens F113 (Redondo-Nieto et al., 2013; Garrido-Sanz et al., 2016). Syntenic comparison of the F113 genome and one of its closest sequenced relative, Pseudomonas brassicacearum NFM421 (Figure 1), showed that these 45 genes form a genetic island inserted in the intergenic region between genes PsF113_0737 and PsF113_0783 encoding a phosphatase and a peroxidase respectively, whose orthologs in P. brassicacearum are contiguous (PSEBRa709 and PSEBRa710). Abrupt loss of sequence homology between F113 and NFM421 indicated that the left border of the genetic island corresponds to F113 nucleotide 894337 and the right border to nucleotide 939056, the nucleotide before the stop codon of PsF113_0783. In the intergenic region between PSEBRa709 and PSEBRa710 there are 42 nts, which are present in other strains belonging to the F113 phylogenomic group but are not in the F113 genome. Those nucleotides could have been lost in F113 as result of the flagellar island insertion.

The analysis of pseudomonads genomes deposited both in the nr/nt and WGS NCBI databases, showed that besides F113, this flagellar island was also present in other seven strains, Pseudomonas extremaustralis 14-3b, Pseudomonas sp. CBZ-4, Pseudomonas veronii R4, P. brassicacearum LBUM300, Pseudomonas kilonensis 1855-344, P. fluorescens et76 and Pseudomonas putida ATH-43. Strains F113, 1855-344, LBUM300, and et76 are phylogenetically related and belong to the Pseudomonas corrugata subgroup, while strains 14-3b, CBZ-4, and R4 are also phylogenetically related and belong to the P. fluorescens subgroup (Figure 2A; Garrido-Sanz et al., 2016). Outside the P. fluorescens complex of species, the island was only found in a P. putida strain (ATH-43). The genomic location of the island is identical in strains F113, LBUM300, and 1855-344 (Figure 1) indicating that the presence of the island in their genomes originated from a single insertion event. The location of the island in strain et79 is in a different genomic context, indicating an independent insertion event (Figure 1). Similarly, the location of the flagellar island in strains R4 and 14-3 is also identical, albeit different that in the P. corrugata group strains (Figure 1). The insertion in the CBZ-4 strain is in another different region (Figure 1). These results indicate that within the P. fluorescens complex, the appearance of the flagellar island occurred in four independent insertion events. These events are also supported by a phylogenetic and phylogenomic analysis of the island (Figure 2). As shown in Figure 2B the evolution of the island is in accordance with the evolution of the genomes in which they are inserted. Furthermore, genetic organization within the island is highly conserved in all the analyzed strains (Supplementary Figure 1).

Comparison of the aminoacid sequences of the ORFs in the island present in F113 confirmed that for most of them, closest relatives outside the pseudomonads were present in A. vinelandii genomes. However, a few ORFs showed higher homology with ORFs present in other Gamma Proteobacteria (Supplementary Table 3).

We have previously shown that ectopic expression of the flhCD genes in a wild-type F113 background resulted in hypermotility (Redondo-Nieto et al., 2013) indicating that the expression of the master regulatory operon is sufficient to trigger the regulatory cascade resulting in the formation of the second flagellar apparatus. However, the introduction of the plasmid was unable to restore motility to non-motile mutants such as fliC (data not shown), indicating that production of the first flagellum is necessary for the production of the second one or that a non-functional flagellum is produced under these conditions.

In order to test the functionality of the second flagellar apparatus, we checked the expression of the fliC2 gene encoding flagellin in F113 and isogenic mutants affected in motility (Figure 3A). RT-PCR experiments showed no expression of the gene in the wild-type strain, suggesting that a second flagellar apparatus is not produced by F113 under laboratory conditions. No expression was observed in the gacS, sadB, or wspR mutants. However, a kinB mutant, that had been isolated as a hypermotile mutant in a screening for motility repressors (Navazo et al., 2009) and an algU mutant, showed significant expression of the fliC2 gene (Figure 3A). Quantitative RT-PCR was used to determine the expression of flhDC and fliC2 in both mutant backgrounds compared to F113. As shown in Figure 3B, high level of expression was observed for both genes in both mutants. flhDC expression was six times higher in the kinB mutant and eight times higher in the algU mutant related to the wild-type strain. Furthermore, fliC2 expression was 4.5 times higher in kinB mutant and 11 times higher in algU than in F113.

To test whether the hypermotility phenotype of the kinB mutant was due to the production of a second flagellar apparatus, we generated flhDC and fliC2 mutations in a kinB mutant background. As shown in Figure 3C, both mutations suppressed the hypermotility phenotype of the kinB mutant, restoring wild-type motility. These results indicate that production of the second flagellar apparatus is causing the hypermotility of the kinB mutant. Furthermore, fliC2 mutation in a F113 background produces no change in the motility phenotype, confirming that the second flagellum is not expressed in the wild-type strain under the conditions tested (Figure 3C).

It has been shown that in A. vinelandii, flhDC expression is activated through c-AMP by the Vfr protein (a CRP ortholog). In order to test whether this regulation was also functioning for the second flagellar apparatus in P. fluorescens F113, we generated vfr mutants in F113 and the kinB mutant. As shown in Figure 4, the vfr mutation had no effect in the swimming phenotype of F113. However, this mutation suppressed the hypermotile phenotype of the kinB mutant, restoring motility to wild-type levels. We also tested the effect of a mutation on the cyaA gene, encoding an adenylate cyclase, in the kinB background. This mutation also restored the wild-type motility phenotype to the kinB mutant (Figure 4). These results indicate that cyaA and vfr activate flhDC expression and therefore the production of the second flagellar apparatus and that this activation is strongly repressed by the kinB gene.

Transmission electron microscopy of negatively stained cells was used to monitor the flagellar number and location in different derivatives of P. fluorescens F113 (Figure 5). The wild-type strain (Figure 5A) presents a single polar flagellum, as has been previously described. Conversely, all the strains producing the second flagellar apparatus, such as the kinB mutant (Figure 5C) and the F113 (pflhDC; Figure 5B) presented a tuft of flagella in one of the poles of the cell. The same was true for the phenotypic variant S (Sánchez-Contreras et al., 2002; Figure 5D), a hypermotile variant isolated from the rhizosphere. A kinB mutant, harboring a mutation in the fliC2 gene (Figure 5E) showed the same flagellation that the wild-type strain.

In order to confirm that hyperflagellation resulted from production of the second flagellar apparatus and not by hyperproduction of the first flagellum, we extracted extracellular proteins from cultures of the wild-type strain, the kinB mutant and a sadB mutant, that has been previously shown to overproduce the FliC flagellin. As shown in Supplementary Figure 2, an anti-FliC antiserum that does not recognize FliC2, showed FliC overproduction only for the sadB mutant.

Since the phenotypic variant S appeared to be producing the second flagellar apparatus, we decided to test fliC2 expression in a battery of hypermotile phenotypic variants isolated from alfalfa rhizosphere (Martinez-Granero et al., 2006). All the phenotypic variants expressed the fliC2 gene (Figure 6A), suggesting that the rhizosphere environment selects for the production of the second flagellar apparatus. It is important to note that phenotypic variants arise at a much lower frequency from liquid medium cultures than in the rhizosphere (Martínez-Granero et al., 2005). We also tested the competitive colonization ability of the wild-type strain against a fliC2 mutant, unable to produce the second flagella. As shown in Figure 6B, the fliC2 mutant was displaced by the wild-type strain from the root tip. Furthermore, we analyzed the competitive colonization capacity of the kinB mutant that synthesizes both flagellum, and a kinBfliC2 double mutant, which can only produce the first flagellar apparatus. As shown in Figure 6B, the kinB mutant was able to displace the kinBfliC2 double mutant. Taken together, these results show that the production of the second flagella in the rhizosphere environment is an important trait for root competitive colonization.