Strains that were isolated during this study are listed in Table S1. Nodule isolates and the spontaneous rifampicin-resistant derivative of reference Sinorhizobium (Ensifer) fredii strain NGR234 (Stanley et al., 1988) were grown at 27°C in/on tryptone-yeast (TY) (Beringer, 1974) or rhizobial minimal medium supplemented with 12 mM succinate (RMS) as sole carbon source (Broughton et al., 1986).

Fields and number of plants sampled in each of them are listed in Table 1. Once collected from roots, nodules were desiccated in plastic tubes containing silica gel and stored at 4°C until further analysis. Prior to isolation of nodule bacteria, plant tissues were incubated overnight at 4°C in sterile double distilled water (ddH₂O). Once nodules were rehydrated, the remaining soil traces were carefully removed. Sterilization of nodule surface was carried out using one initial 3 min incubation in 70% (v/v) EtOH followed by 3 min in 4% (w/v) sodium hypochlorite solution and thorough washing with sterile ddH₂O. Once sterilized, nodules were rolled onto TY agar plates (TYA) to check for remaining surface contaminants. Each nodule was then crushed into 50 μl of sterile ddH₂O, and an aliquot of the nodule lysate was used to inoculate a Petri dish containing TYA. All plates were incubated at 27°C and bacterial growth monitored each day. When nodule isolates started growing, a sample was used to inoculate a liquid culture which, following subsequent serial dilutions was used to isolate single colonies on RMS. Thus, all of the isolates described in this study were purified from a single colony.

To obtain a preliminary identification and sorting of nodule isolates, each strain was cultivated separately on TYA and at 27°C. When growth was sufficient, each isolate was analyzed in quadruplicate by mass spectrometry as described in Ziegler et al. (2015). Briefly, free-living bacteria were spotted onto MALDI steel target plates, overlaid with 1 μl of 25% formic acid, air-dried, and again overlaid with 1 μl of matrix solution consisting of saturated alpha-cyano-4 hydroxycinnamic acid (CHCA; Sigma-Aldrich, Buchs, Switzerland) in 33% acetonitrile (Sigma-Aldrich), 33% ethanol and 3% trifluoroacetic acid (TFA). Once dried, bacterial spots were analyzed with MALDI-TOF Mass Spectrometer Axima™ Confidence machine (Shimadzu-Bio- tech, Kyoto, Japan) using the linear positive detection mode, a laser frequency of 50 Hz and a mass range of 3–15 kDa. For each isolate, spectra consisting of 50–100 protein masses were averaged and processed using the Launch-pad™ 2.8 software (Shimadzu-Biotech). For identification purposes, mass spectra were matched against the PARPM and the “rhizobia-specific module” of the SARAMIS™ databases that were described previously (Ziegler et al., 2015). When isolates did not match any of the reference rhizobial strains, the search was extended to bacteria included in a proprietary and enlarged version of the SARAMIS™ database (Mabritec AG). For hierarchical clustering of protein masses, a binary matrix listing presence/absence of masses was generated using the SARAMIS Superspectra tool (see Table S5). Results were then imported into the PAleontological STatistics (PAST) software (Natural History Museum, Oslo University, Norway). Multivariate neighbor-joining cluster analyses with similarity-distances were calculated using correlation algorithm in PAST software (Feltens et al., 2010). Cluster analysis was saved as a Nexus file and imported into FigTree software to generate a circular tree. For sequencing analyses, genomic DNA (gDNA) of each of the selected isolates was prepared as in Chen and Kuo (1993). Amplifications of the 16S rDNA, ITS, nifH, and rpoB sequences were carried out on a T-Gradient thermocycler (Biometra, Göttingen, Germany) in 50 μl PCR reactions containing 50 ng of gDNA, 0.2 mM dNTPs, 1 unit Taq polymerase and 1 μM final concentration of each of the corresponding primers listed in Table S2. Prior to sequencing, amplified products were separated onto agarose gels and purified using the NucleoSpin gel and PCR clean-up kit (Macherey-Nagel AG, Oensingen, Switzerland). Purified amplicons were sequenced at Microsynth AG (Balgach, Switzerland) using Sanger sequencing. High quality sequences were assembled into double stranded DNA sequences that were manually curated for ambiguous nucleotide positions and primer sequences. Except for strains CI-16A, CI-34D1, and CI-34F for which only 16S rDNA sequences were obtained, the corresponding 16S and ITS sequences of selected nodule isolates were deposited in GenBank as assembled 16S-ITS-partial 23S sequences. All DNA sequences used in this study were deposited in GenBank under the accession numbers that are listed in Table S2.

Putative phylogenetic relationships were inferred with the MEGA software version 6 (Tamura et al., 2013) using DNA sequences of selected isolates listed in Table S2. The corresponding 16S-ITS-23S, nifH, and rpoB sequences of the reference strains Bradyrhizobium genosp. CB756, B. diaozoefficiens USDA 110^T, B. elkanii USDA 76^T and USDA 3259, B. japonicum USDA 6^T, and B. pachyrhizi BR3262 were retrieved from GenBank. Reference strains were selected on the basis of taxonomic proximity to pigeonpea isolates and for best anchoring of tree branches. DNA sequences were aligned with ClustalW (for nifH and rpoB) or MUSCLE (for rRNA operon sequences) as implemented in MEGA, and manually corrected when needed. Phylogenetic trees were inferred using the neighbor-joining (NJ) method with statistical support calculated from 1000 bootstrap replicates. The best-fit nucleotide substitution model was selected according to the Bayesian information criterion (Schwarz, 1978): Tamura-Nei + Gamma (TN93+G) parameters for rpoB and 16S-ITS-23S sequences and Tamura 3 + Gamma (T92+G) parameters for nifH. In all trees, a corresponding sequence of S. fredii strain NGR234 was used as outgroup.

Seeds of C. cajan cultivars ILRI 16555 and “Light Brown” were surface-sterilized using concentrated sulphuric acid for 10 min, 0.1% (v/v) Tween 20 for 5 min and 5% hydrogen peroxide for 5 min, with intermediate washing steps using sterile ddH₂O. Seeds of Leucaena leucocephala, Macroptilium atropurpureum cv. Siratro, Tephrosia vogelii, Vigna radiata cv. King, and Vigna unguiculata cv. Red Caloona were surface sterilized as described previously (Fumeaux et al., 2011). To germinate, surface-sterilized seeds were incubated for two to three days in the dark, at 27°C and on B&D agar plates. Once germinated, seedlings were planted in Magenta jars (two plants per jar) containing vermiculite (Lewin et al., 1990), and watered using nitrogen-free B&D solution (Broughton and Dilworth, 1971). When plantlets emerged from the vermiculite, each of them was inoculated with 200 μl of a water solution containing 2 × 10⁸ freshly grown bacteria. Plants were grown in controlled conditions with a light phase of 12 h, a day temperature of 27°C, a night temperature of 20°C and 60–70% humidity. Except when stated otherwise, plants were harvested at 42 days post inoculation (dpi). Symbiotic proficiency of each inoculant was assessed using as parameters the nodule number, total nodule fresh weight, dry weight of shoots, and presence/absence of leghemoglobin inside nodules or pseudonodules. The promiscuous S. fredii strain NGR234 was used as a control inoculum since it was shown to nodulate and fix nitrogen with all plants tested in this study (Perret et al., 1999; Pueppke and Broughton, 1999). Identity of each inoculant was confirmed by PCR amplification of ITS sequences using ITS-For2 and ITS-Rev2 primers and aliquots of nodule lysates as template.

To minimize changes in cell structures, bacteria were fixed directly on an agar plate, for 1 h at room temperature using a final concentration of glutaraldehyde of 2% (v/v). After three successive washes with 1 ml of 66 mM Sörensen's sodium phosphate buffer (KH₂/Na₂HPO₄2H₂O, pH 6.8), cells were post-fixed in 2% (w/v) osmium tetroxide solution for 1 h. After another wash with Sörensen's buffer, samples were dehydrated by successive immersions of 10 min in solutions of respectively 25, 50, 75, and 95% (v/v) ethanol at 4°C. Samples were then washed three times for 10 min in 100% ethanol at room temperature, and finally incubated in isoamyl acetate for 1 h. Dehydrated samples were then mounted on aluminum mounts with silicon wafers, coated with gold and observed with a JEOL JSM 7001FA field emission scanning microscope.

Physical and chemical properties of soil samples were determined by a commercial soil analysis laboratory (Soil-Conseil, Gland, Switzerland) using standard methods (ISO 17025). Briefly, samples were homogenized, dried for 48 h at 40°C and sieved with fraction <2 mm being used for subsequent analyses. Soil texture was determined by sedimentation once the humus was destroyed by H₂O₂. Soil type was determined using the standard USDA particle-size classification. Soluble elements (Ca, Cu, Fe, K, Mg, Mn, P, and Zn) were extracted with water and composition determined by spectrophotometry and spectrometry. pH was measured using 20 g of 2 mm sieved soil samples stirred in 50 ml ultrapure H₂O. Organic matter (%) was determined by oxidizing soil carbon with a mixture of K₂Cr₂O₇ 1N and H₂SO₄ 95–98%, total nitrogen (%) was measured using the Kjeldhal method, and available nitrogen was extracted with 0.01 M CaCl₂ solution and measured by spectrophotometry.

Two separate fields were sampled in each of the three localities that were selected for their different records of C. cajan cultivation (see Figure 1). Close to the northeast city of Bondoukou, next to the border with Ghana, cultivation of pigeonpea has been established for many years with inhabitants having given vernacular names in local dialects to this crop (e.g., “Kapkô” in Koulango). The two sampled fields of Bondoukou covered an area of 0.5 and 2 ha, and were intercropped with yam (Dioscorea sp.) and cassava (Manihot esculenta), respectively. In this area of Côte d'Ivoire, seeds of pigeonpea found on local markets had diverse sizes and colors (ranging from whitish to brown), indicating that farmers favored growth of different landraces rather than a specific cultivar. This contrasted with the C. cajan plants grown in the two fields of Kossou-Bouafla and that were exclusively offsprings of the ILRI 16555 cultivar. ILRI 16555 was introduced in 2004 via a program sponsored by the Heifer International charity for fostering the use of perennial cultures of pigeonpea as forage to cattles in cotton-based agricultural systems. Accordingly, seeds held by farmers in Kossou-Bouafla were uniformly brown. The remaining two fields of 0.5 and 1.5 ha, were sampled in the outskirts of the capital Yamoussoukro. They were used for planting uncharacterized cultivars of C. cajan. The field covering 1.5 ha was intercropped with Jatropha curcas plants, and was used for producing pigeonpea seeds for poultry feeding. In each of the six sampled fields and within 1.5 m from the trunk, root systems of three to six plants were cleared of soil. Only nodules attached to the cleared root system were collected. Once cleared of most soil particles, nodules that belonged to the same root system were stored together into a single desiccating vial and kept at 4°C until further processing. In total, 171 root nodules belonging to 28 plants were collected in the six designated fields.

A total of 85 nodule isolates were recovered and purified as single colonies. Initial identification was performed with MALDI-TOF MS as in Ziegler et al. (2015). Mass spectral analyses did not only provide a preliminary identity at the genus or species level for most of the nodule isolates, but also allowed for a cluster analysis of MS signatures. As shown in Figure 2, 62 isolates (73%) clustered into two clearly separated groups of strains (clusters I and II). The larger of these two groups included the reference B. elkanii strains USDA 76^T and USDA 3259 as well as 43 isolates of C. cajan nodules (lower left quadrant in Figure 2). The second cluster consisted of 19 closely related isolates (Figure 2, upper left quadrant) identified by MALDI-TOF MS as bradyrhizobia sp. The remaining 23 isolates were extremely diverse (Figure 2, upper-right corner), with only three strains (CI-5B, -39Bx, and CI-41A) sharing rhizobia-like features (Table S1).

To verify that the initial identification and clustering of nodule bacteria was accurate, several isolates were selected on the basis of (i) their respective position in the cluster tree shown in Figure 2, (ii) the field/soil in which nodules were found, and (iii) the bacterial identification (or absence of it) obtained via MALDI-TOF MS. Accordingly, a subset of 8 strains (CI-1A, -7A, -15C, -19F, -33M, -36B, -36E, and CI-40F) was selected amongst the 43 isolates initially identified as B. elkanii (cluster I). CI-1B, -14A, -19D, -33F, and CI-41S were chosen to represent the 19 bradyrhizobia that formed the second cluster of related strains, and six (CI-5B, -16A, -34D1, -34F, -39Bx, and CI-41A) of the 23 unrelated strains were retained for 16S rDNA sequencing. Except for strains CI-5B and CI-39Bx, of which the 16S genes matched those of Rhizobium leguminosarum bv. viciae and Rhizobium sp. strain JGI 0001005-K05 respectively, 16S rDNA sequencing confirmed the identifications obtained with mass spectra analyses (see Table S1). For example, the 8 isolates selected from cluster I were found to carry 16S rRNA genes identical to those of several members of the B. elkanii super clade (Aserse et al., 2012) including B. elkanii type strain USDA 76^T, B. pachyrizi strain PAC48^T and Bradyrhizobium tropiciagri CNPSo 1112^T. Isolates CI-1B, -14A, -19D, -33F, and CI-41S of cluster II shared identical 16S ribosomal sequences that were 99.7% similar to the same type strains USDA 76^T, PAC48^T and CNPSo 1112^T, while the extremely slow-growing isolate CI-41A was found to be closely related to B. liaoningense. As to isolates CI-16A, -34D1, and CI-34F that were not identified via MALDI-TOF mass spectrometry, the respective 16S rRNA sequences best matched those of Paenibacillus sp. JDR-2 (accession number CP001656.1), Brevibacillus reuszeri DSM 9887 (LGIQ01000014.1) and Bacillus soli strain NBRC 102451 (BCVI01000121.1). These analyses further confirmed that, for those strains for which a related reference existed in the spectral databases, MALDI-TOF MS provided an accurate identification (Ziegler et al., 2015).

Although selected isolates from the same cluster shared identical 16S rDNA sequences, topology of the cladogram shown in Figure 2 suggested genetic diversity existed within each of these clusters. To examine whether selected isolates from clusters I (8 out of 43 strains) and II (5 out of 19) differed genetically, the internal transcribed spacer (ITS) separating 16S and 23S rRNA genes as well as the nifH gene that codes for the structural Fe protein of nitrogenase were amplified and sequenced. Strains CI-5B, -39Bx, and CI-41A were also included in these analyses because all three isolates possessed rhizobia-like features (see Table S1). For many years, the ITS region has been used in population genetics and molecular systematics (Gürtler and Stanisich, 1996), in particular for taxa for which 16S rRNA gene sequences lacked resolution for closely related isolates [e.g., sphingomonads (Tokajian et al., 2016)]. NifH is an essential component of the nitrogenase complex, and its gene has been widely used to survey diazotrophs in diverse ecosystems (Zehr et al., 2003).

Once deleted of the flanking 16S and 23S rRNA sequences, the amplified ITS regions of the 16 selected nodule isolates ranged in size from 760 bp (CI-41A) to 1054 bp (CI-5B) (Table S2). Comparing these amplified ITS sequences confirmed that isolates of the B. elkanii cluster I were not genetically identical: CI-15C and CI-19F shared similar ITS sequences (4 mismatches over 910 nucleotides) that were longer and differed significantly from those of the remaining six isolates (801 and 802 bp long). ITS's of CI-1A, -33M, and CI-36E were identical and formed a consensus sequence, from which the ITS's of CI-7A and CI-40F differed by only one nucleotide while that of CI-36B diverged by seven bases. By contrast, and except for the one nucleotide-shorter ITS of CI-19D, ITS's of selected isolates from cluster II were 862 bp long and differed by a maximum of only six nucleotide positions. ITS regions of CI-1B and CI-33F were identical and appeared as possibly more ancient, with that of CI-41S being the most divergent of all five selected isolates. Thus, ITS analyses confirmed that, when compared to the five isolates of cluster II that showed little diversity, strains of cluster I were more diverse and could be further subdivided into two distinct groups: a set of closely related isolates (CI-1A, -7A, -33M, -36B, -36E, and -40F) with CI-19F and CI-15C forming a more distantly related outgroup. These various subgroups of nodule isolates can be clearly observed in the unrooted-phylogeny trees of 16S-ITS and partial 23S as well as rpoB sequences shown in Figure 3 and Figure S1.

Diversity of selected isolates was further confirmed by comparing nifH sequences. With a maximum of four diverging nucleotide positions, NifH coding sequences of selected cluster II isolates (CI-1B, -14A, -19D, -33F, and CI-41S) were closely related to and did best match the nifH gene of B. elkanii type strain USDA 76 (see Figure 4). In contrast, nifH sequences of cluster I strains fell into two distinct subgroups, one of which included isolates CI-15C and CI-19F. Unlike what was observed in the 16S-ITS-23S phylogenetic tree (Figure 3), nifH of CI-15C and CI-19F were closer to corresponding sequences of cluster II isolates than to those of the remaining cluster I strains (see Figure 4). In fact, the best match for nifH of CI-1A, -7A, -33M, -36B, -36E, and CI-40F was the corresponding gene of Bradyrhizobium pachyrhizi strain BR3262. As in Figure 3, CI-41A stood out from cluster I and cluster II isolates, with a best match for nifH in Bradyrhizobium genosp. CB756 (SA-4). Thus, except for the discrepant nifH of CI-15C and CI-19F, the 16S-ITS-partial 23S and nifH phylogenies were similar.

Proficiency of representative CI-1B (for cluster II), CI-36E (cluster I), CI-41A (Bradyrhizobium sp. isolate) and control NGR234 strains was compared using offsprings of self-crossed ILRI 16555 and “Light Brown” plants collected in Côte d'Ivoire. As shown in Table 2 and Figure S2, pigeonpea isolates were clearly more proficient symbionts than NGR234 on the selected C. cajan cultivars, with a shoot dry weight for plants inoculated with CI-1B, -36E, and CI-41A being at least twice as much as that of plants nodulated by NGR234. While symbiotic efficacies of CI-1B and CI-41A were found to be similar on both ILRI 16555 and “Light Brown” cultivars, CI-36E appeared slightly less efficient. Interestingly, on both ILRI 16555 and “Light Brown” cultivars NGR234 made fewer (Table 2, Figure S2) but larger nodules. For example on ILRI 16555 (test #2), NGR234 formed nodules of ca. 32 mg fresh weight each whereas plants inoculated with isolates CI-1B, -36E, and -41A carried nodules of 9.4–12.1 mg fresh weight in average. Thus, when compared to pigeonpea isolates, the lower symbiotic efficacy of NGR234 on C. cajan possibly resulted from a reduced ability to initiate nodule formation rather than a lower nitrogen fixation capacity once bacteria were established inside nodules.

A number of additional but smaller scale nodulation assays were also carried out in order to better define the host-range and nodulation properties of pigeonpea isolates (Tables S3-A and S3-B). In addition to ILRI 16555 and “Light Brown” cultivars, the selected CI-1B, -36E, and CI-41A isolates were also found to nodulate (Nod+) and fix nitrogen (Fix+) with Macroptilium atropurpureum cv. Siratro, Tephrosia vogelii, Vigna unguiculata cv. Red Caloona (Table S3-A) and Vigna radiata cv. King (Table S3-B). On Leucaena leucocephala that forms Fix+ associations with NGR234 (Lewin et al., 1990), strains CI-1B, CI-36E, and CI-41A were found to induce only few pseudonodules that appeared to be Fix-, however. By contrast, isolates CI-5B and CI-39Bx failed to nodulate any of the hosts that were so far tested: a result consistent with the absence of successful nifH amplifications for both of these isolates.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.