The strains and plasmids used in this work are listed in Supplementary Table 5. Escherichia coli was grown on Luria–Bertani (LB) (Miller, 1972) medium at 37°C. Rhizobium and Ensifer strains were grown on Tryptone–Yeast extract (TY) (Beringer, 1974),Yeast extract–Mannitol (YEM) (Vincent, 1970), or Vincent Minimal Medium (VMM) (Vincent, 1970) at 28°C. For solid media 15 g of agar per liter of medium were added. When needed, Congo Red was added at 0.25% (w/v) and Calcofluor was added at 0.02% (w/v). The final concentration of antibiotics used was (in μg ml^–1): gentamicin (Gm) 10, kanamycin (Km) 25, and tetracycline (Tc) 10 for E. coli. For Rhizobium and Ensifer: streptomycin (Sm) 400, nalidixic acid (Nal) 20, neomycin (Nm) 60, rifampicin (Rif) 100, spectinomycin (Sp) 100, Gm 30, and Tc 5.

Ensifer meliloti and R. favelukesii were cultivated in YEM medium. Cultures were centrifuged 45 min at 10,000 × g. The EPS in the supernatant was precipitated with three volumes of ethanol overnight at −20°C and stored until processed. Then, the samples were centrifuged for 45 min at 10,000 × g and the pellets were resuspended in water. The quantification was performed by the anthrone method (Loewus, 1952). Amounts of EPS were normalized per mg of total proteins, which was determined by Bradford assay with Coomassie brilliant blue (Bradford, 1976).

Bacteria were grown in shaker flasks containing 200 ml of VMM at 30°C for 7 days. The precipitation and purification of EPS were done according to Müller et al. (1988). For monosaccharide analysis, the EPS was hydrolysed in 2 M trifluoroacetic acid for 2 h at 120°C in sealed glass vials. The hydrolysate was subsequently dried using a rotary evaporator. Remaining amounts of trifluoracetic acid were removed by the addition of isopropanol and dried again using the rotary evaporator. This treatment was carried out twice. Samples were dissolved in water and the sugars were analyzed by high-performance anion-exchange chromatography with pulsed amperometric detection on a Carbo Pac PA1 column (250 by 4 mm; Dionex, Sunnyvale, Calif.) and isocratic elution (1 ml/min) with 16 mM NaOH.

The EPS was analyzed by proton nuclear magnetic resonance (¹H-NMR) spectroscopy. For ¹H-NMR analysis, purified EPS was dissolved in deuterium oxide (99.7%), lyophilized, redissolved, and again lyophilized. Finally, 10 mg/ml of EPS was dissolved in deuterium oxide (99.99%), and sonicated for 5 min at room temperature. Spectra were recorded at 600 MHz and 80°C (Bruker Avance III–600).

Bacterial matings were performed as described by Simon et al. (1983). The visualization of plasmids in the transconjugants (plasmid profiles) was evaluated on Eckhardt-gels (Eckhardt, 1978) as modified by Hynes and McGregor (1990).

Total DNA and plasmid preparations, restriction-enzyme analysis, cloning procedures, and E. coli transformation were performed according to previously established techniques (Sambrook et al., 1989). PCR amplification was carried out with recombinant Taq DNA polymerase or Pfu DNA polymerase as specified by the manufacturers. Primers are listed in Supplementary Table 6.

For comparative genomics studies, the genes were searched for by means of BLASTp on the NCBI webpage. To examine genetic neighborhood, https://img.jgi.doe.gov/ was also used.

For the construction of the phylogenetic trees, the proteins were aligned with the module of Clustal implemented in MEGAX (Kumar et al., 2018). Prottest2.4 was used to determine the models of protein evolution (Abascal et al., 2005). The best model was LG+I+G for ExoB, LG+G+F for ExoH, LG+G for ExoV and LG+G for ExoY. Maximum likelihood (ML) trees were inferred under the selected model using PhyML v3.1 (Guindon and Gascuel, 2003). The robustness of the ML topologies was evaluated using a Shimodaira-Hasegawa-like test for branches implemented in PhyML v3.1 (values display in the nodes are multiplied by 100). We employed the best of NNIs and SPRs algorithms to search the tree topology and 100 random trees as initial trees. The accession numbers for the proteins selected for the phylogeny are display in each tree.

Medicago sativa (Alfalfa Sardi ten) seeds where surface-sterilized for 5 min in a 70% (v/v) ethanol solution, then washed with sterile distilled water. Subsequently, seeds were immersed in a solution of sodium hypochlorite (11 g l^–1) for 15 min and washed six times with sterile distilled water. Seeds where placed on water agar (0.8% w/v) overnight. Afterward, seeds were placed on ethylene-oxide-sterilized plastic growth pouches containing 10 ml of Rolfe medium (Rolfe et al., 1980), with modifications (Barsch et al., 2006). Germination occurred in the pouches, and the roots developed through the hole in the paper. Seeds that did not germinate or did not grown through the hole were discarded. Three days post-germination, plants were inoculated with 10⁶ colony-forming units of each strain used in this work (at least 25 plants per strain). Plants were cultured in a growth chamber at 22°C with a 16-h photoperiod, watered with modified Rolfe medium and water. The CFU that each inocula contained were estimated by plate counts. Plant assays were repeated at least twice.

Plants were harvested 4 weeks after inoculation. Nodules were counted, weighted and conserved at 4°C. The shoot dry matter weight was measured. For nodule occupancy, nodules were sterilized and treated as previously described (Torres Tejerizo et al., 2010; Montiel et al., 2016). Briefly, the nodules that previously were weighted and counted, were surface-sterilized with H₂O₂ (30 volume) for 10 min, washed with sterile distilled water and then crushed in 200 ul of sterile isotonic solution followed by plating serial dilutions on TY with the corresponding antibiotics. 2–3 days after, CFU were counted.

For microscopy, nodules were embedded in 6% (w/v) agarose. Nodule sections of 60 μm were obtained using a Leica VT1000 S Vibratome. Fresh sections were then stained with Live/Dead BacLight bacterial Viability Kit (Thermo Fisher). The staining solutions were then removed and nodule sections were resuspended in PBS buffer. Images were acquired with a Leica TCS SP5 confocal microscope.

During the symbiotic interaction between E. meliloti and alfalfa, different polysaccharides play relevant roles. Among them, EPS I (succinoglycan) is critical (Niehaus and Becker, 1998), but in some strains, the lack of EPS I may be overcome by EPS II (galactoglucan) or KPS (capsular polysaccharide, also referred as K antigen) (Pellock et al., 2000). As R. favelukesii LPU83 is able to infect alfalfa with some particular features (as mentioned in the introduction), the orthologs to the genes known to be involved in polysaccharides synthesis were searched for using BLAST. In addition to establishing a minimum of 35% identity as a cut-off to define orthologs (Rost, 1999), query coverage (>70%) and synteny were also considered in the analyses.

The presence of two main clusters of exo genes in LPU83, one in the chromosome and another in plasmid pLPU83a point out that an event of horizontal gene transfer (HGT) could have occurred during the evolution of LPU83, where the strain acquired the exo genes from other bacteria. It is worth mentioning that previous work from our group has shown that pLPU83a is able to perform conjugative transfer (Torres Tejerizo et al., 2010), and the regulation of this phenomena depends on their genomic background and environmental conditions (Torres Tejerizo et al., 2014). Moreover, the plasmid gene cluster could be complementing the chromosomal genes to get a fully functional EPS I production.

To understand if such event of HGT occurred, phylogenetic analyses were made employing four genes. One only present in the chromosomal gene cluster (exoY), one present only in the plasmid gene cluster (exoV), one present in both clusters (exoH) and the bonafide exoB, that was present in the chromosome but not in the exo gene cluster. For the phylogenetic inference, the translated products of the genes were used. The ExoB phylogenetic tree showed that it was related to Rhizobium tibeticum CCBAU85039 and Rhizobium grahamii CCGE 502 (Figure 2), strains with which it shares a close chromosomal relationship (Torres Tejerizo et al., 2016). In a similar way, ExoY (also harbored in the chromosome) clusters with the same strains. Meanwhile, ExoV, which is present in pLPU83a, groups with E. meliloti and Ensifer medicae strains. No orthologs to ExoV were found in R. tibeticum CCBAU85039 and R. grahamii CCGE 502. Finally, the phylogenetic tree of ExoH, locates the chromosomal-encoded copy near to R. tibeticum CCBAU85039 and R. grahamii CCGE 502 while the plasmid-encoded copy is closer to E. meliloti and E. medicae. These results support the hypothesis that the plasmid exo cluster arose from an HGT event, probably from an Ensifer-related strain. It was previously shown that the exo cluster present in the plasmid is flanked by two inverted repeats, which encode a Tn3-like transposase and recombinase (Castellani et al., 2019). These structures could be a reminiscence of an insertion/excision of foreign DNA.

Evidence of HGT events toward new-symbiotic strains has been shown previously, although the efficiency of the resulting symbioses may be low. Sullivan et al. (1995) and Sullivan and Ronson (1998) showed the lateral transfer of a symbiotic island from Mesorhizobium loti to non-symbiotic mesorhizobia allowing the evolution of the latter into symbiotic rhizobia. A similar event was observed in soils of the Brazilian Cerrados (Batista et al., 2007), where evidence of HGT from inoculated Bradyrhizobium strains to soil bacteria was detected. Recently, it was shown that the symbiotic plasmid of Rhizobium etli CFN42, or at least fragments of it, can be transferred to bean-endophytic bacteria, generating new strains with the ability to nodulate and fix nitrogen (Bañuelos-Vazquez et al., 2020). In our group, we showed that the gene cluster involved in the synthesis of the nodulation factors (NFs) of LPU83 has a structure similar to that of E. meliloti, with whom it also shares a close phylogenetic relationship (Del Papa et al., 2007; Torres Tejerizo et al., 2011). Thus, not only genes involved in NFs synthesis could have been recruited by HGT, but also a gene cluster involved in EPS biosynthesis, during the evolution of LPU83.

Several determinants, including EPS, play essential roles during the complex molecular crosstalk before and during the interaction between rhizobia and plants (Jones et al., 2007). Rhizobial EPS varies among strains and species, harboring different substitutions (Skorupska et al., 2006). Due to the relevant role of EPS during the symbiosis between rhizobia and Medicago spp. and the ability of LPU83 to nodulate Medicago spp. and other legumes (Wegener et al., 2001), the structure of the EPS produced by LPU83 was analyzed. EPS was precipitated from culture supernatants. Monosaccharide analysis of the hydrolyzed EPS of LPU83 indicated that the strain produces an EPS that consists of glucose and galactose in a ratio of 7 to 1, similar to EPS I of Eme2011 (Figure 3). The EPS of LPU83 was further analyzed by ¹H-NMR and gas chromatography/mass spectrometry (GC/MS) studies. A ¹H-NMR spectrum of EPS I isolated from Eme2011 was recorded as a control (Figure 3B). This spectrum was in accordance with data already published for this EPS I (Leigh et al., 1987; Keller et al., 1995). The proton resonances between 3.8 and 5.4 ppm were assigned to glycosyl components arising from ring protons, and the singlet resonances at 2.1 and 2.8 ppm were assigned to methyl protons of the 1-carboxyethylidene (pyruvate) and acetyl groups, respectively. The characteristic triplets at 3.1 and 3.25 ppm represent the methylene protons of the succinyl groups. The EPS isolated from LPU83 gave rise to virtually the same proton NMR spectrum, indicating that the main polysaccharide secreted by this strain is identical to succinoglycan from Eme2011.

At first glance, it may have been proposed that the different substitution/modifications of the EPS could be related to the failure in the nodule development and differentiation observed in R. favelukesii (Eardly et al., 1985; Del Papa et al., 1999; Wegener et al., 2001). Our results demonstrate that despite the different organization of EPS I genes, whose localization is not restricted to one cluster and that the evolution of both clusters seems to integrate genes from different rhizobial linages, the EPS 1 produced is identical to that of E. meliloti. The question that remains is whether these clusters are essential for this EPS production and plant infection or not.

We showed that LPU83 produces a succinoglycan identical to the one produced by E. meliloti and that the genes involved in the synthesis are distributed between the chromosome of LPU83 and plasmid pLPU83a. This particular organization leads to the question whether both gene clusters are relevant for the production of EPS I. To determine this, a genetic approach was used. The chromosome cluster was deleted by double crossing-over and inserting a resistance to Tc, generating strain LPU83 Δchromo (the region is indicated in Figure 1), as indicated in section “Materials and Methods”. The cluster located in pLPU83a was deleted in a similar way, but a Sp resistance was used instead, generating strain LPU83 Δplasmid (also indicated in Figure 1). ExoB has been recognized as a UDP-glucose 4-epimerase which provides the UDP-galactose in E. meliloti (Buendia et al., 1991). Mutants in exoB of E. meliloti fail to produce proper EPS I and EPS II due to the lack of galactose (Putnoky et al., 1990; Buendia et al., 1991; Reuhs et al., 1995). Due to the relevance of exoB, a mutant was also made. In this case, by single recombination through the integration of pK18mob-exoB plasmid (LPU83-exoB^–). Afterward, the mutations were combined to generate a double deletion mutant (LPU83 Δchromo Δplasmid) and a triple mutant (LPU83-exoB^– Δchromo Δplasmid). Finally, LPU83-exoB^– was complemented in trans with a full copy of exoB (pBBR1MCS5-exoB). Wild type Eme2011 and Eme2011 Δexo [which harbors a markerless deletion of the exo gene cluster from exoP to exoZ (Schäper et al., 2019)] were used as controls. EPS production was quantified with the anthrone method (Figure 4A) and EPS I was visualized due to its fluorescence in medium containing calcofluor (Figure 4B). The calcofluor assays demonstrated that only wild type strains (Eme2011 and LPU83) or the LPU83-exoB^– complemented in trans were able to produce EPS I. In agreement with this result, anthrone assays also showed EPS production by the same strains. The levels of EPS of the exoB mutant complemented were slightly lower than the wild type (Figure 4A), but the calcofluor fluorescence phenotype was clearly observed (Figure 4B).

Schäper et al. (2016, 2017) have shown that in Eme2011 EPS production is enhanced when cyclic di-GMP levels are elevated. Also, a new polysaccharide was described: an arabinose-containing polysaccharide (APS) (Schäper et al., 2019). The ectopical overexpression of pleD (diguanylate cyclase) and cuxR (c-di-GMP responsive transcriptional activator) leads to high di-GMP levels and, consequently, to an APS overproduction, generating wrinkled red-colored macro-colonies on medium containing Congo red (CR). Moreover, Pérez-Mendoza et al. (2015) showed that when cyclic di-GMP levels increase, MLGs can be produced. MLGs also bind Calcofluor and CR. The morphology of LPU83 and its mutants was evaluated in medium with CR, to determine binding of this dye. Neither LPU83 nor the mutants showed wrinkled macro-colonies nor red staining. Mutants of LPU83 did not show Calcofluor fluorescence (Figure 4). These results indicate that under our experimental conditions, APS and MLGs are not synthetized.

Exopolysaccharides production depends on many genes (Niehaus and Becker, 1998). In LPU83, the genes needed for the production of EPS I are distributed among different replicons. As shown above, both clusters and the exoB gene are needed for EPS production. In addition to the lack of orthologs of the genes needed for the synthesis of EPS II, the lack of glucose in the mutants suggest that galactoglucan is not produced in LPU83. The lack of CR staining suggests that also APS is not produced under our laboratory conditions. Altogether, LPU83 seems to be producing only EPS I, requiring both exo clusters and exoB.

As mentioned earlier, KPS and EPS II may replace, at least partially, the lack of EPS I in the effective nodulation of alfalfa by E. meliloti (Pellock et al., 2000). But when only EPS I is produced, mutants in its biosynthesis fail to nodulate (Cheng and Walker, 1998; Niehaus and Becker, 1998; Mendis et al., 2016). Also in Eme2011, mutants in the genes involved in the sulfation of NFs are not able to nodulate (Roche et al., 1991; Schultze et al., 1995). LPU83 has the particular capacity to nodulate alfalfa even in absence of sulfated NFs (Torres Tejerizo et al., 2011). This peculiarity encouraged us to evaluate the nodulation of alfalfa using all the constructed strains. As control, we included Eme2011 and Eme2011 Δexo strains. Eme2011 Δexo is a deletion mutant that lacks the exo gene cluster from exoP to exoZ, but retains the exoB gene. Eme2011 Δexo lacks the same genes as LPU83 Δchromo Δplasmid. Nodules induced by Eme2011 exhibited the typical cylindrical shape of alfalfa nodules with pink coloration. In contrast, Eme2011 Δexo showed small white round nodules (Figure 5). Some of the nodules induced by wild-type LPU83 showed the typical cylindrical shape, however they were mostly white instead of pink. Similar to the nodules observed for the Eme2011 exoB mutant (Schäper et al., 2019), the nodules induced by LPU83-exoB^– and by LPU83-exoB^– (pBBR1MCS5 empty vector) were tiny and round, while the complemented mutant, LPU83-exoB^– (pBBR1MCS5-exoB), showed bigger nodules than LPU83-exoB^–. The nodules induced by LPU83 Δplasmid were mainly round, but some of them were cylindrical. In contrast, the ones induced by LPU83 Δchromo were mainly cylindrical, but some were round. Unexpectedly, most of the nodules induced by the double and triple mutant were cylindrical. Nodules from all the derivatives of LPU83 were mostly white (Figure 5).

In the double and triple mutants, many nodules appeared with brown necrotic areas, while in the wild-type LPU83, these brown necrotic areas were observed in only a few nodules. These brown necrotic areas have been previously reported and related to plant defense symptoms, in nodules induced by EPS-deficient rhizobial mutants (Niehaus et al., 1993; Parniske et al., 1994; Fraysse et al., 2003; Berrabah et al., 2015). This led to the assumption that rhizobial EPS could act as suppressor of the plant defense response (Niehaus and Becker, 1998; Jones and Walker, 2008; Jones et al., 2008). This could cause the presence of the brown necrotic areas in nodules developed by LPU83 mutants, but it cannot explain their presence in the wild-type LPU83. One possibility is that the wild-type LPU83 triggers, at low levels, a plant defense mechanism. Studies with transcriptional fusions demonstrated that exo genes of Eme1021 were highly expressed in free-living cells but their expression decreased in later stages of symbiosis (Reuber et al., 1991). These observations were confirmed with recent high-throughput transcriptome studies (Capela et al., 2006; Roux et al., 2014). The expression of these genes might be fine-tuned by nodule signals, to allow a proper modulation of the plant defense. It cannot be disregarded that, despite the identical structure of EPS I of Eme2011 and LPU83, the spatial and temporal expression pattern of the exo genes could differ, causing a different response of the plant inside the nodules induced by LPU83.

Fresh nodules were counted, weighted and the number of CFU per nodule was evaluated (Figure 5). As expected, the comparison between Eme2011 and Eme2011 Δexo showed statistical significant differences in fresh nodule numbers and weight. Among the LPU83 derivatives, wild-type LPU83, double and triple mutants and LPU83 Δplasmid showed similar fresh weight, while the LPU83-exoB^– showed lower fresh weight. LPU83-exoB^– and LPU83-exoB^– (pBBR1MCS5 empty vector) showed the highest numbers of nodules per plant, but harbored less bacteria inside. The complemented mutant, LPU83-exoB^– (pBBR1MCS5-exoB), generates similar numbers of nodules as LPU83, but less than the LPU83-exoB^– mutant. No CFU were detected in Eme2011 Δexo nodules, indicating the lack of a proper infection. The absence of bacteria in mutants in EPS I production of Eme2011 was previously shown (Cheng and Walker, 1998; Niehaus and Becker, 1998). Nodules occupied by LPU83 showed the highest number of CFU per nodule. Unexpectedly, bacteria could be recovered from the nodules induced by all LPU83 mutants. LPU83 Δplasmid and the double and triple mutants showed CFU per nodule values similar to those of Eme2011. The estimation of nitrogen-fixing ability was evaluated by comparisons of the shoot-dry weights. As expected, Eme2011 showed nitrogen fixation, while Eme2011 Δexo did not. LPU83 and its mutants showed a poor nitrogen fixation and the constructed mutations did not change the behavior of the wild-type strain (Figure 5).

As the recovery of bacteria from the nodules induced by all LPU83 mutants was unexpected, nodule histology was evaluated by confocal microscopy. Nodules were cut and stained with Live-Dead BacLight. As a control, Eme2011 showed the typical tissue organization, where zones delimitation and living bacteria stained by SYTO9 were clearly visible (green fluorescence) and distributed in the nodule after the meristem (zone I) (Figures 6A,G). Nodules infected with LPU83 showed a similar pattern, but red-stained bacteria (indicative of dead bacteria) were observed. Moreover, less plant cells seem to be occupied by LPU83 than by Eme2011 (Figures 6B,H). The triple mutant, LPU83 Δplasmid and LPU83 Δchromo showed a similar pattern to the one observed in LPU83, but more cells seem to be alive. Also, more plant cells appear to be occupied by the mentioned mutants (Figures 6D–F,J–L). The LPU83-exoB^– strain showed smaller nodules, with an unclear definition of the zones and a large portion of the nodule was empty. Nevertheless, bacteria were observed inside these nodules but many of them were red-stained bacteria (Figures 6C,I). The size and shape of the nodules, the CFU per nodule obtained and the observed histology in nodules from LPU83-exoB^–, show that the alteration in exoB severely impairs the nodulation of LPU83 but, contrary to E. meliloti exoB mutant (Schäper et al., 2019), LPU83-exoB^– is still able to infect. Remarkably, LPU83-exoB^– Δchromo Δplasmid, which lacks all exo genes, presents number of nodules, size, occupation and histology more similar to LPU83 than to the mutant LPU83-exoB^–.

Overall, our results indicate that R. favelukesii LPU83 shows a lack of differentiation inside the nodules, which could explain the low nitrogen fixing rate. During the last years, nodule-specific cysteine-rich (NCRs) peptides have been intensively studied due to their role in the differentiation of the rhizobia inside the nodules (Van de Velde et al., 2010). Despite that NCRs show toxicity for the rhizobia in vitro, they are needed for the proper differentiation of the bacteria into bacteroids (Kondorosi et al., 2013; Pan and Wang, 2017). Furthermore, the membrane, the EPS and the LPS have been suggested as key components in the resistance to the NCRs (Montiel et al., 2017; Lima et al., 2020). It could be hypothesized that some other difference in the LPU83 cell envelope does not allow a correct perception of NCRs or that the expression of NCRs differs between the infection of E. meliloti and LPU83, resulting in non-proper symbiotic nodules in the case of infection by LPU83.