The bacteria used in this study are indicated in Supplementary Table S1. The Bradyrhizobium strain ORS278 and its derivatives mutants were grown in yeast extract mannitol (YM) medium at 34°C or in minimal medium buffered nodulation medium (BNM) (Podlešáková et al., 2013). Escherichia coli strains were grown in Luria-Bertani medium (LB) at 37°C. When required, the media were supplemented with kanamycin (100 μg ml^-1) or spectinomycin (20 μg ml^-1) or a mixture of the two (100 and 20 μg ml^-1).

Standard molecular biology techniques were used for all cloning work. All the primers and plasmids used for cloning of DNA fragments are listed respectively in Supplementary Tables S1, S2. For the construction of Bradyrhizobium strain ORS278 insertional mutants, 300–400 base pair (bp) internal fragments were amplified by PCR and cloned into the plasmid pVO155-npt2-GFP (Okazaki et al., 2016). The constructions were first transferred into the E. coli S17-1 strain and then into ORS278 by mating as previously described (Giraud et al., 2010). For the construction of mutants by deletion mutagenesis, flanking regions of the genes were first amplified by PCR and then merged by overlap extension PCR. The overlap PCR fragment obtained was then digested by restriction enzymes corresponding to the restrictions sites of the primers (Supplementary Table S1) and was subsequently cloned in the sacB suicide pNPTS129 plasmid by ligation (Tsai and Alley, 2000). The spectinomycin resistance gene cassette from the pHRP315 vector (Parales and Harwood, 1993) was liberated by BamHI digestion and was introduced between the upstream and downstream regions previously cloned in the pNPTS129 plasmid. The resulting plasmid was then transferred into ORS278 by biparental conjugation, as previously described (Giraud et al., 2010). Antibiotic selection was used to select single recombinant which were verified by PCR. Double recombinant clones were then obtained by growth on sucrose in the presence of spectinomycin but without the addition of kanamycin. Candidate clones were checked for the loss of kanamycin resistance from the pNPTS129 plasmid, and the deletion of the gene was verified by PCR.

Lipopolysaccharide was isolated and purified from the ORS278 WT strain and the lpxXl, BRADO4680 and BRADO0045 mutants, as previously described, and the lipid A isolated through a mild acid hydrolysis (Silipo et al., 2014). Like for MS data, the lipid A was analyzed using a 4800 MALDI MS/MS (ABSciex) mass spectrometer equipped with a Nd:YAG laser at a wavelength (λ) of 355 nm with a 500-ps pulse and a 200 Hz firing rate. External calibrations were performed using an ABSciex calibration mixture, allowing mass accuracies close to 55 ppm. An aliquot of bacterial pellets was also subjected to a micro-extraction procedure to obtain the lipid A moiety, as described by El Hamidi et al. (2005), and analyzed in an ABSCIEX TOF/TOF^TM 5800 Applied Biosystems mass spectrometer equipped with an Nd:YLF laser with a λ of 345 nm, a <500-ps pulse length and a repetition rate of up to 1000 Hz. All mass spectra were acquired either in negative (not shown) and positive polarity. In the MS experiments, each spectrum resulted from the accumulation of 1,500 laser shots, whereas 5,000–7,000 shots were summed for the MS/MS data acquisitions (Spina et al., 2000; Sturiale et al., 2011). Samples were dissolved in CHCl₃/CH₃OH (50:50, v/v) at a concentration of 1 mg ml^-1. Matrix solution was prepared by dissolving 2,4,6 trihydroxyacetophenone (THAP) in CH₃OH/0.1% trifluoroacetic acid/CH₃CN (7:2:1, by volume) at a concentration of 75 mg ml^-1. One microliter of the sample/matrix solution (1:1, v/v) was deposited on the well plate and allowed to dry at room temperature (Leone et al., 2007).

To monitor growth in rich (YM) and minimal (BNM) media, cultures were inoculated at 10^-2 dilution using YM-grown log-phase (OD₆₀₀ = 0.5–0.7) WT or mutant strains. Growth was measured at OD₆₀₀ using a Cary 50 Scan (Varian). Different incubation temperatures were tested (28, 34, and 37°C). Growth curves were performed in duplicate.

The sensitivity to various abiotic stresses (SDS, H₂O₂, HCl, and NaCl) was assayed as previously described (Silipo et al., 2014). The experiments were conducted in triplicate for the lpxXL and BRADO4680 mutants and only once for the two BRADO0045 mutants.

The minimum inhibitory concentration (MIC) of polymyxin B was determined as previously described (Kulkarni et al., 2015) by the Etest method using the disk diffusion assay (Biomérieux, Marcy-l’Etoile, France). The experiment was performed in triplicate.

Aeschynomene indica and A. evenia seeds were surface sterilized, cultivated, and inoculated as previously described (Gully et al., 2016). For the nodulation and nitrogen fixation assay, 10 plants per condition were taken at 14 days post infection (dpi) to count the number of nodules on the roots and to analyze nitrogenase activity using an acetylene reduction assay (ARA) as previously described (Bonaldi et al., 2010).

Cytological analyses were conducted on 5–10 nodules originating from three different plants for each condition using the protocol described by Bonaldi et al. (2011). Confocal microscopy observations were carried out using a confocal laser-scanning microscope (Carl Zeiss LSM 700; Jena, Germany). Sections of ORS278 WT and BRADO4680 nodules were incubated for 20 min in live/dead staining solution [5 μM SYTO 9 and 30 μM propidium iodide (PI) in 50 mM Tris pH 7.0 buffer; Live/Dead BacLight, Invitrogen]. All the nodule sections were incubated for 15 min in 10 mM phosphate saline buffer (PBS) containing calcofluor white M2R (Sigma, Munich) at a final concentration of 0.01% (w/v) to stain the plant cell wall (Nagata and Takebe, 1970). Calcofluor was excited at 405 nm and emission signals were collected from 405 to 470 nm. For SYTO 9 or GFP and PI, an excitation wavelength of 488 and 555 nm was used with emission signal collected at 490–522 nm and 555–700 nm, respectively. Images were obtained using the ZEN 2008 software (Zeiss).

Statistical analysis was performed using XLSTAT version 2016.6 software. Differences between groups of samples were evaluated with the Tukey’s range test. Differences were considered statistically significant at a P-value < 0.01. The box plots were made in R 3.2.2 software. Results are shown as box plots. Each graph contains median, quartiles, and whiskers which show the last sample in [1st quartile -1.5^∗IQR, 3rd quartile + 1.5^∗ IQR] range. The free points on the graph represent the outliers samples.

The GenBank accession numbers of BRADO4675 to BRADO4680 are respectively CAL78405 to CAL78410 and the one of BRADO0045 is CAL74017.

A BLAST search of the ORS278 genome led to the identification of a single gene cluster containing several homologs of genes shown to be involved in VLCFA biosynthesis in the R. leguminosarum bv. viciae 3841 strain (Figure 2) (Bourassa et al., 2017). This gene cluster is composed of the CDS BRADO4675 to BRADO4679, which are homologs of acpXL, fabZXL, fabF1XL, fabF2XL, and lpxXL genes, respectively (Figure 2). Besides sharing a high level of identity (>40%) with the corresponding R. leguminosarum proteins, except for the absence of the adhA2XL gene in ORS278, the organization of the genes is perfectly conserved between the two strains (Figure 2).

We also analyzed the distribution and the organization of the VLCFA genes in other photosynthetic (ORS285, BTAi1, B. oligotrophicum S58) and non-photosynthetic (B. diazoefficiens USDA110, B. japonicum USDA124, and B. elkanii USDA76) Bradyrhizobium strains. In all cases, only one homolog region was identified in which the genes displayed similar organization to that in ORS278 notably with the absence of the adhA2XL homolog. However, the photosynthetic Bradyrhizobium strains differed from the non-photosynthetic ones by the presence, downstream of lpxXL, of a gene (BRADO4680) that codes for a protein of unknown function (Figure 2).

The identification of only one VLCFA gene cluster among the Bradyrhizobium genomes suggests that the genes present in this region might be sufficient for the synthesis and the attachment of the various VLCFAs that are linked to the lipid A in these bacteria.

To confirm that the identified region in the ORS278 strain is involved in the synthesis of VLCFAs and in their transfer to the lipid A, we tried to mutate the different genes present in this region, including BRADO4680. Different mutagenesis strategies were tested, deletion by double crossing over in acpXL and BRADO4680 and disruption of fabZXL, fabF1XL, fabF2XL, or lpxXL by insertion of the non-replicative plasmid pVO155-npt2-GFP in the corresponding coding region. Despite several attempts, only two mutants were obtained, one in the lpxXL homolog (BRADO4679) and the second in BRADO4680. The repeated failure to select mutants in acpXL, fabZXL, fabF1XL, or fabF2XL suggests that these four genes are essential because they are required for the synthesis of VLCFAs without which the bacteria cannot survive or grow under the conditions used to select the mutants. Conversely, the selection of mutants in BRADO4680 and lpxXL could result in functional redundancy, the absence of role, or in a specific role in the synthesis or the attachment of only one type of VLCFA to the lipid A.

To investigate the impact of the BRADO4679 and BRADO4680 mutations on the synthesis of the VLCFAs, MALDI MS analyses were performed on the lipid A of the ORS278 WT and mutants strains. The mass spectrometry analysis of the WT lipid A showed a mixture of ion species which essentially varied by the acylation pattern, i.e., penta- and hexa-acylated species. All the lipid A species were stoichiometrically built up of a pentasaccharide backbone formed by a β-(1-6)-linked 2,3-diamino-2,3-di-deoxy-D-glucopyranose (DAG) carrying a GalpA residue on the vicinal DAG and an α-(1-6)-Manp disaccharide on the distal DAG residue. Indeed, the mass differences among neighboring ion species were due to acyl number and length of ester linked fatty acids (Figure 3A). The ion peak at m/z 2128.52 was established to be a penta-acylated sodiated lipid A carrying two 14:0 (3-OH), one 12:0 (3-OH), one 12:1 and one acetylated VLCFA, 26:0 (25-OAc). The most abundant sodiated hexa-acylated lipid A species (m/z 2594.81) further carried a secondary VLCFA, the 30:0 (2,29-2OH), in turn esterified, in the hepta-acylated species centered around m/z 3107.22, by the hopanoid moiety (Figures 3A,D). The ORS278 lipid A structure was therefore very similar to the one previously described for the BTAi1 strain with the presence of two VLCFAs, the longest of which harbored a hopanoid molecule (Silipo et al., 2014). The lipid A of the BRADO4680 mutant had a similar structure to that of the WT-strain (Figures 3B,D), showing that BRADO4680 is not involved in the modification of the lipid A.

In contrast, the lipid A of the lpxXl mutant showed differences in the acylation pattern. In fact, it was composed of a mixture of tetra-, penta-, and hexa-acylated species in which the 26:0 (25-OAc) VLCFA was absent (Figures 3C,E and Supplementary Figure S2), indicating that LpxXL is exclusively involved in the transfer of the C₂₆:25OH VLCFA to the lipid A of ORS278.

Very long-chain fatty acids have been shown to contribute to stress tolerance in free-living and symbiotic states, in diverse rhizobia like R. leguminosarum or S. meliloti (Becker et al., 2005; Gibson et al., 2008; Bourassa et al., 2017). To test the hypothesis that the removal of the C₂₆:25OH of ORS278 lipid A reduces the ability of the strain to resist stressful conditions, we challenged lpxXl and BRADO4680 mutants with a variety of stressors that occur during the initiation and progression of symbiosis. In addition, the growth kinetics of the lpxXl and BRADO4680 mutants at various temperatures and their resistance to membrane destabilizer were quantified to determine if these mutations have an effect on the stability of the membrane.

As shown in Figures 4A,D, the lpxXl mutant displayed similar growth to that of the WT strain at 28°C in both rich and minimal media. In contrast, at higher temperatures (34 and 37°C), the growth kinetics of the mutant were reduced compared with those of the WT strain (Figures 4B,C,E,F). These observations suggest a reduction in the stability of the membrane. In addition, disk diffusion assays showed that the lpxXl mutant was more sensitive to H₂O₂ and SDS than the WT strain (Figure 4H). The mutant was also more sensitive to osmotic stress than the WT strain, as evidenced by a reduction in its growth at a concentration of 100 mM of NaCl (Figure 4G). Because ORS278 is exposed to NCR-like peptides in Aeschynomene plants, we also tested the sensitivity of the mutants to polymyxin B which is a cationic peptide displaying similar effects than some NCRs on the alteration of the bacterial OM permeability (Mikuláss et al., 2016). Contrary to the lpxXL mutant of R. leguminosarum which is not affected in polymyxin B resistance (Bourassa et al., 2017), we observed that the lpxXl mutant of ORS278 displayed an eightfold lower MIC (8 μg/ml) for polymyxin B than the WT strain (64 μg/ml) (Figure 4I). The BRADO4680 mutant withstood all the stressors, as did the WT (Figures 4A–I). Together, these data suggest that the C₂₆:25OH VLCFA linked to the lipid A of ORS278 plays an important role in the ability of this strain to cope with various stressful conditions by increasing the stability of the OM.

To investigate if the lpxXl and BRADO4680 mutations have an effect on the symbiotic properties of ORS278, we inoculated the WT and the two mutants strains on two Aeschynomene plants, A. evenia and A. indica, that can be nodulated in a Nod-independent manner by this strain.

No effect of the BRADO4680 mutation was detected on either Aeschynomene species (Supplementary Figure S1). In contrast, a clear effect of the lpxXl mutation was observed in both Aeschynomene species. Plants inoculated with this mutant, particularly A. evenia, displayed typical symptoms of nitrogen starvation, including leaf chlorosis and reduced plant growth at 14-dpi (Figures 5A,B). These observations were correlated with reduced nitrogenase activity, as estimated by the ARA compared to the WT strain, despite the fact that the lpxXL mutant led to a higher number of nodules per plant (Figures 5C,D). In addition, some nodules induced by the lpxXL mutant in both A. evenia and A. indica plants were yellowish instead of green, as observed for the WT nodules (Figures 5E–G,L,M). The absence of chloroplasts in the epidermal tissue of the Aeschynomene nodules is typically observed in mutants with altered nitrogen fixation, such as nif or hopanoid-minus mutants (Bonaldi et al., 2010; Silipo et al., 2014). Furthermore, some A. evenia nodules elicited by the lpxXl mutant were hollow suggesting a degradation of the tissue of the infection zone (Figure 5H), which is a sign of early senescence, as already reported in inefficient nitrogen-fixing mutants (Delmotte et al., 2014; Silipo et al., 2014). In A. indica, no hollow nodules were found, but brownish compounds that autofluorescent in the red spectrum (excitation, 488 nm; emission, 600–660 nm) were observed in some lpxXl nodules (Figure 5P). This is indicative of the accumulation of polyphenol compounds generally associated with plant defense reactions, as previously described in other legumes (Bourcy et al., 2013). Finally, cytological analysis performed by confocal microscopy showed that the process of bacteroid differentiation was altered in the lpxXL mutant, in both A. evenia and A. indica plants. In fact, in the lpxXL mutant, both undifferentiated bacteria and bacteroids were observed, which were not perfectly spherical and were of abnormal size, in contrast to the WT bacteroids (Figures 5I–K,N–P). As a WT reference, we used a tagged strain containing the pVO155-npt2-GFP plasmid inserted in the BRADO5083 gene which encodes a protein of unknown function. Previous studies did not reveal any particular symbiotic defect due to the plasmid insertion.

Taken together, these data indicate that the lpxXL mutation drastically alters the ability of the ORS278 strain to form an efficient symbiosis with Aeschynomene plants.

The fact that the lipid A of the lpxXL mutant still contains the C₃₀:29OH VLCFA suggests that another acyltransferase exists that allows the specific transfer of this VLCFA to the lipid A. Sequence genome analysis of ORS278 did not enable identification of another homolog of lpxXL, suggesting that this acyltransferase should be strongly different from BRADO4679. Conversely, a search for genes annotated as containing an acyl transferase domain retrieved more than 40 candidates. This number was too high to envisage the systematic mutagenesis of all of them. On the other hand, a Tn5 mutant in one of them (BRADO0045) was previously described for its nitrogen fixing deficiency in A. indica (Bonaldi et al., 2010) suggesting that this gene plays an important role during symbiosis. This CDS has 42% identity with the ActA protein of S. meliloti, which has been shown to play an essential role in the acid tolerance of the bacteria (Tiwari et al., 1996), and 30% identity with the Apo-lipoprotein acyltransferase Lnt characterized in E. coli, which is involved in the maturation of lipoprotein (Gupta et al., 1993). Together, these different elements prompted us to examine in more detail the role of BRADO0045 in both the physiology and symbiotic properties of ORS278 and to explore the possibility of its direct involvement in the transfer of the C₃₀:29OH VLCFA to the lipid A.

To be sure that the phenotype reported for the Tn5 mutant corresponds to the inactivation of BRADO0045, we constructed a new insertional mutant in this CDS using the non-replicative plasmid pVO155-npt2-GFP. The ability of the two BRADO0045 mutants (Tn5 and pVO155) of ORS278, called Ω0045T and Ω0045P, respectively, to cope with various stresses was then analyzed as previously. Analysis of growth kinetics at 28 and 34°C showed that the two BRADO0045 mutants underwent alteration of their growth at the higher temperature, pointing to destabilization of the membrane, as observed in the lpxXL mutant (Figures 6A,B). In addition, the two mutants were found to be more sensitive to salt, oxidative and detergent stresses and to the polymyxin B (Figures 6C–E) than the WT strain. These results show that the mutation in BRADO0045 reduces the ability of ORS278 to cope with abiotic stresses in the same way as the lpxXL mutation.

To test if the mutations in BRADO0045 have an impact on the symbiotic properties of ORS278, A. indica plants were inoculated with the WT strain and the two mutants. At 14-dpi, we observed that the plants inoculated with the two mutants were smaller than the plants inoculated by the WT-strain (Figure 6F), which is correlated with lower nitrogenase activity (Figure 6G). As previously observed in the lpxXL mutant, some nodules elicited by the two BRADO0045 mutants were yellow and hollow (Figures 6H,L) and some displayed massive plant defense reactions (Figures 6I–K). In addition, cytological analysis showed that the intracellular bacteria were not differentiated or not perfectly spherical, and were of abnormal size, the same as observed in the lpxXL mutant (Figures 6M,N). The BRADO0045 protein thus appears to be essential for the establishment of an efficient symbiosis between ORS278 and A. indica.

To determine if this phenotype was due to alteration of the lipid A structure of ORS278, MALDI MS analyses were performed on the lipid A from ORS278 WT and from Ω0045P mutant. As can be seen in Supplementary Figure S3, the two spectra were very similar, showing that the effects of the BRADO0045 mutation on the phenotypic and symbiotic properties of ORS278 were not due to modification of the lipid A.