Burkholderia Species Are the Most Common and Preferred Nodulating Symbionts of the Piptadenia Group (Tribe Mimoseae)

Burkholderia legume symbionts (also called α-rhizobia) are ancient in origin and are the main nitrogen-fixing symbionts of species belonging to the large genus Mimosa in Brazil. We investigated the extent of the affinity between Burkholderia and species in the tribe Mimoseae by studying symbionts of the genera Piptadenia (P.), Parapiptadenia (Pp.), Pseudopiptadenia (Ps.), Pityrocarpa (Py.), Anadenanthera (A.) and Microlobius (Mi.), all of which are native to Brazil and are phylogenetically close to Mimosa, and which together with Mimosa comprise the “Piptadenia group”. We characterized 196 strains sampled from 18 species from 17 locations in Brazil using two neutral markers and two symbiotic genes in order to assess their species affiliations and the evolution of their symbiosis genes. We found that Burkholderia are common and highly diversified symbionts of species in the Piptadenia group, comprising nine Burkholderia species, of which three are new ones and one was never reported as symbiotic (B. phenoliruptrix). However, α-rhizobia were also detected and were occasionally dominant on a few species. A strong sampling site effect on the rhizobial nature of symbionts was detected, with the symbiont pattern of the same legume species changing drastically from location to location, even switching from β to α-rhizobia. Coinoculation assays showed a strong affinity of all the Piptadenia group species towards Burkholderia genotypes, with the exception of Mi. foetidus. Phylogenetic analyses of neutral and symbiotic markers showed that symbiosis genes in Burkholderia from the Piptadenia group have evolved mainly through vertical transfer, but also by horizontal transfer in two species.


Introduction
Legumes have developed a symbiosis with a polyphyletic group of bacteria commonly called rhizobia. This symbiosis leads to the formation of a specialized organ, the root nodule, within which rhizobia differentiates into bacteroids. Bacteroids fix atmospheric nitrogen, and feed the plant with combined nitrogen in exchange for carbon compounds derived from photosynthesis by the legume host. Over the last several decades, numerous diversity studies have focused on rhizobia, but their diversity and the number of investigated legumes hosts remain far from being complete due to the large number of legume species (.18000) [1].
The relationship between b-rhizobia and Mimosa spp. was investigated in more depth in two studies reported by [11] and [25]. These authors analyzed the diversity of symbionts from nodules of c. 70 diverse Mimosa species growing in the two major biomes of Brazil (Cerrado and Caatinga) in which the genus Mimosa has evolved and diversified into more than 200 native and endemic species [26]. They concluded on the generic character of nodulation in the genus Mimosa that the preferred symbionts of this genus in Brazil are Burkholderia. Identical tree topologies between neutral and symbiotic markers were also observed on Brazilian and French Guianan Mimosa-nodulating burkholderia [11], [12], indicating a monophyletic origin and single acquisition of symbiotic genes by a Burkholderia ancestor, followed by vertical transfer of nodulation genes during species diversification [11]. On the other hand, based on phylogenies of nodulation genes, Cupriavidus taiwanensis was found to be a more recent symbiont of Mimosa spp., which acquired nodulation genes from a Burkholderia ancestor [8], [12], [27]. Although most species of Mimosa are mainly nodulated by Burkholderia spp., and a few by Cupriavidus spp., some of them can also form effective symbioses with arhizobia [8], [11], [28], [29]. Later studies have underlined the existence of genetic and environmental factors that could affect the preference of legumes species for nodulation with either aor brhizobia; these include soil pH or the presence of combined nitrogen [11], [12], [16], [24], [25].
The Mimosa genus is closely related to a number of genera including Piptadenia (P.), Parapiptadenia (Pp.), Pityrocarpa (Py.), Anadenanthera (A.), Stryphnodendron (St.), and Microlobius (Mi.) (syn. Goldmania) within the tribe Mimoseae. Jobson & Luckow [30] have investigated the phylogeny of these different genera and subdivided the Piptadenia genus into three clades: the Piptadenia genus sensu stricto (or Eupiptadenia clade) that is a sister clade to Mimosa, and includes, for example P. flava (type species), P. floribunda, P. gonoacantha, P. paniculata; the Pityrocarpa clade (Py. leucoxylon, Py. monoliformis, Py. obliqua) that is closer to the genera Stryphnodendron and Parapiptadenia than to the Eupiptadenia clade; and finally P. viridiflora that is outgrouped from the previous genera and represent a particular case deserving a new generic name [30]. Interestingly, all these genera contain woody species native to South America, particularly to Brazil [31], [30]. Some of them are currently exploited by locals owing to their economical values, such as A. peregrina [32], [33], Pp. rigida [34], [35] and P. gonoacantha [36]. Although their ability to establish associations with rhizobia is documented [37], [38], [39], [40], [41], [42], [43], information about the rhizobial diversity and symbiotic efficiency on these plant species is scarce. A recent study of symbionts of Parapiptadenia rigida in Uruguay demonstrated that this species is nodulated by rhizobial strains belonging to the genera Burkholderia, Cupriavidus and Rhizobium, among which the Burkholderia genotypes were the predominant group [44]. Symbiosis with b-rhizobia in the tribe Mimoseae thus appears to extend outside Mimosa and be more common than previously expected.
In this study our objectives were (i) to investigate the extent of Burkholderia affinity within the tribe Mimoseae by focusing on native species in the Piptadenia genus and in related genera in the Piptadenia Group described by Jobson & Luckow [30]; (ii) to characterize the diversity of rhizobia and their symbiotic genes in this group of legumes; (iii) to examine further symbiotic specificity within the Piptadenia Group.
To achieve these objectives, we isolated rhizobia from diverse species in the Piptadenia Group in their native areas in Brazil, characterized their taxonomic and symbiotic diversity, and assessed their host specificity via coinoculation assays. We found a large diversity of Burkholderia species, but also a-rhizobia, and with a few exceptions an affinity of most plant species towards Burkholderia rather than to a-rhizobia. We discuss the evolutionary patterns of both taxonomic and symbiotic markers and the putative co-adaptation between Burkholderia and Piptadenia Group species.

Bacterial Isolates Sampling and Maintenance
Collection of material was authorized by IBAMA Nu. 058/ 2006. The bacterial collection was built through several sampling campaigns performed between 1984 and 2010, from 9 states in Brazil (Rio de Janeiro, Sao Paulo, Minas Gerais, Bahia, Mato Grosso do Sul, Paranà, Pernambuco, Espirito Santo and Pará). Sample locations were concentrated on the main centers of diversification of the Piptadenia Group (Mata Atlantica) and are presented in Figure S1. Details on sampling (gps coordinates, year, season, soil type and pH) are presented in Table S1. STM strains were sampled from trees from various places of Rio de Janeiro State in April 2010; SFM, BR and CVRD strains were sampled from plants growing in plant nursery beds in various states of Brazil (Table S2). Nodules sampled from the field as described in [37] were dried on silicagel until symbiont isolation. Nodules were then rehydrated for 30 min in sterile distilled water, surface sterilized by immersion 30 seconds in 70% ethanol followed by 1 to 3 min in 3% hydrogen peroxide, and washed three times in sterile distilled water. Nodules were then individually crushed and streaked on yeast mannitol agar (YMA, [45]) plates containing bromophenol blue. All YMA plates were incubated at 28uC. Single colonies were picked and checked for purity by repeated streaking and microscope examination. All pure isolates were stored at 280uC in YM broth plus 20% (w/v) glycerol. For Stryphnodendron species, no nodules could be found on two locations, so a plant trapping approach was developed by growing seedlings on a top soil harvested under each tree, and nodules were harvested and processed as described above. As these rhizobia were trapped, they were not treated as natural symbionts, and were only included on Figure 1C to assess rhizobial patterns among the Piptadenia group. Strains from previous studies on P. flava [42], Pp. rigida [44], [46], as well as P. gonoacantha and Pp. pterosperma [46] were also included in this study. A total of 196 strains were included in the study, of which 63 ''representative'' strains are listed in Table 1 and were chosen as one strain per host plant per geographical origin per unique 16 S rDNA ribotype (as described in molecular method section).

Plant Nodulation and Specificity Tests
Seeds of P. gonoacantha, Py. monoliformis, Mi. foetidus and A. colubrina were obtained from Instituto Brasileiro de Florestas (Londrina, Paraná Brasil) or Embrapa Agrobiologia (Seropédica; Brazil). M. pudica and siratro seeds were obtained from B&T World Seeds (Paguignan, France) and from University Cheikh Anta Diop (Dakar, Senegal), respectively. P. gonoacantha seeds were scarified and surface sterilized using 96% H 2 SO 4 and 3% calcium hypochlorite (2 min and 3 min, respectively), while A. colubrina, A. peregrina, and Pp. rigida seeds were immersed for 3 min in 3% calcium hypochlorite; and Mi. foetidus seeds were sterilized with 96% H 2 SO 4 during 30 min. Py. monoliformis seeds were germinated by immersion in concentrated H 2 SO 4 for 5 min, then washed with sterile dH 2 O and afterwards were soaked in 3% calcium hypochloride for 3 min. All species seeds were then washed four times with sterile dH 2 O, before being germinated on 0.8% water agar plates at 28uC in the dark. M. pudica and siratro seeds were sterilized and germinated as described in [12].
For nodulation tests, all species were grown in Gibson tubes containing Jensen medium [45] filled with dH 2 O, except Piptadenia species for which tubes were filled with sterile attapulgite (OIL DRI US Special, IIIR, Damolin) and supplemented with sterile dH 2 O. All plants were then grown in a chamber at 26uC (relative humidity 40%) with a 16 h light/8 h night cycle. Inoculation was performed by adding 1 ml of exponential bacterial culture grown in broth YM medium. Cross-contamination was investigated by Figure 1. Comparison between Piptadenia's group plant phylogeny and the occurrence of alpha and beta-rhizobia as nodule symbionts in the field or during coinoculation experiments. The plant phylogeny (A) is based on a trnL-F/trnK-matK combined dataset, and was built by parsimony with TNT1.1 (default parameters, on www.phylogeny.fr) using the Jobson & Luckow [30] dataset (downloaded from Treebase, study number S1763, and amended with the P. trisperma from this study). The % of a and b-rhizobia per legume host (in bold) from field sampling (B) or from the coinoculation experiment (C) are represented as white (Burkholderia) and black (a-rhizobia) squares, with the number of strains sampled within each square. *: statistics of symbionts from [44] and [11]. The grey colored square for Stryphnodendron indicates that % of a-rhizobia originates from a trapping experiment on soil (see Material & Methods section). doi:10.1371/journal.pone.0063478.g001   For plant-rhizobium coinoculation experiments, plants were grown in Gibson tubes filled with attapulgite, moistened with sterile water and inoculated with 1 ml of an equal mixture of 12 representative strains calibrated at 10 e 6 cells using KOVA slides microscope counting. Plants were incubated in the same conditions than as described above, watered with dH 2 O and harvested 4 weeks post inoculation. Eight nodules were harvested per nodulated plant (12 replicates), surface sterilized 3 min with 3% calcium hypochlorite, placed in a microplate, crushed using a replicator, grown on YMA microplates, and genotyped on recA gene marker as described in the molecular methods section.

Molecular Methods
Extraction of genomic DNA of bacterial strains was performed using the proteinase K lysis procedure described previously [47]. PCR amplifications were performed with GO-Taq Polymerase (Promega) following the manufacturer instructions. All PCR templates were generated with specific primer sets listed in Table  S2. 16 S rDNA amplification was carried out as previously described [12]. A 800-bp recA fragment was amplified by PCR and sequenced using the primers recABurk1F and recABurk1R for brhizobia, as described in [11], while two specific couples of primers were used for Bradyrhizobium (TSrecAf and TSrecAr) and Rhizobium (recAf and recAr) using PCR conditions described in [48] and [49], respectively. A 440-bp fragment of nifH was amplified and sequenced on Burkholderia strains as described in [8], and on Rhizobium and Bradyrhizobium strains as described in [50]. A 600-bp fragment of nodC was amplified and sequenced for Burkholderia strains using primers described in [11] except for BSP1, BD, BPL and BSa strains for which a specific primer set (nodCPipF and nodCPipR) was designed (500 bp fragment). For a-rhizobia, 600 bp of nodC for Bradyrhizobium and 800 bp for Rhizobium strains were amplified with primer sets nodCfor540-nodCrev1560 and nodCF-nodCI respectively [51], [50]. DNA sequencing was subcontracted to Genoscreen Inc., using ABI3730 sequencers, and the same primers as used for PCR, except for partial 16 S rRNA sequencing for which an internal primer was used (16 S-1080 r).
For the identification of the rhizobial nodule occupant from the plant coinoculation experiment, DNA from surface-sterilized crushed nodule were heat-shocked in 20 ml using a PCR machine (1 cycle of 2 min at 96uC, followed by 7 cycles of 10 seconds at 96uC then 4uC, with a final cycle of 2 min at 4uC), and 1 ml was used to amplify a recA gene fragment as described above. PCR templates were sequenced as previously described.
Plant phylogenetic markers trnL-F and trnk-matK of Stryphnodendron sp. and P. trisperma were PCR-amplified and sequenced as described by [30], from leaves collected from the field.

Nucleotide Sequence Accession Numbers
The sequences have been deposited in EMBL database under accession numbers HE983632 to HE983823 and HF536727 to HF536767, and are listed by gene for each strain in Table S3. Sequence alignments are available upon request.

Building a Collection of Rhizobia from Piptadenia Group Species
We sampled at least one representative species of the different clades and genera in the Piptadenia Group (see species in bold in Figure 1A): Piptadenia sensu stricto (the ''Eupiptadenia'' clade e.g. P. gonoacantha, P. paniculata, P. adiantoides), the Pityrocarpa clade (Py. moniliformis), the P. viridiflora clade, and the Anadenanthera, Parapiptadenia, Stryphnodendron and Pseudopiptadenia clades. The species assignment of each plant host was confirmed by the Botanical garden of Rio de Janeiro (R. Ribeiro). For five species (Stryphnodendron adstringens, P. trisperma, P. paniculata, P. gonoacantha and A. peregrina), the taxonomic position was confirmed by sequencing the trnL-F marker, and comparing it to the Jobson & Luckow dataset [30]. Among all sampled legume species, we confirmed that 17 of them are nodulated, and one, Piptadenia trisperma is a new report for nodulation (according to nodulation data in GRIN database, http://www.ars-grin.gov/,sbmljw/ cgi-bin/nodulation.pl and [58]). The rhizobial collection was composed of 196 isolates and details are presented in Table 1, where they are classified according to original host legume, £ number of isolates from the same host, location and 16 S haplotype per representative strain listed in the first column. # positive nodulation tests obtained from the representative strain.  representative strain and site of isolation. Representative strains for each location and legume host were chosen according to their 16 S rDNA sequence haplotype (named ribotype in Table 1): one strain per legume host and per sampling location with a unique 16 S rDNA sequence based on a 800 bp alignment was kept for further analyses, and designated as representative strains (for a total of 63 for strains from this work, and 79 when including reference strains from other studies).
Nodulation ability of the representative strains was confirmed either on their original host or on M. pudica and siratro when seeds of the original host could not be obtained (see Table 1, Nodulation column), except for 15 strains which were not tested. No nodulation was observed on M. pudica or siratro by five strains (STM7339, STM7342, STM7420, STM7426, STM7444), but these strains were kept as the original host could not be tested. Details about nodule size, shape and functionality for each Burkholderia species are given in Table S4 and Figure S2, together with pictures of nodules on M. pudica and P. gonoacantha. All nodulating strains appeared to fix nitrogen with M. pudica (based on red color within nodules usually linked to presence of leghaemoglobin, and plant development, see Table S4), except some B. sabiae (carrying nodC1 variant) and B. nodosa (nodC5 variant) strains that produced ineffective nodules. Three nodulation groups were observed on M. pudica as described in Table S4 (all nodules being indeterminate): small inefficient nodules (BSa, BN, Figure S2A), small round shape nodules (BSP1, BD, BPL, BSP4, Figure S2B), and elongated nodules with a peanut shape (BSa, BSP3, Figure S2C).
We also incorporated in our analyses strains BR9001 to BR9004, and BR4812 that were isolated from the Piptadenia Group [46] but which were not previously characterized at the taxonomic level, as well as strains from other studies: two from P. flava (UPRM8020 & UPRM8021) [42], two from P. viridiflora (JPY570, JPY565, Gross et al., unpublished), one each from P. stipulacea (JPY584), Pp. blanchetti (EG100) and Ps. bahiana (EG118) (Gross et al., unpublished), and three frequent genotypes of Burkholderia and Rhizobium from a study of Pp. rigida symbionts in Uruguay [44]. Based on their 16 S rDNA sequence haplotype, the bacterial collection could be grouped into three genera, with 107 strains in the genus Burkholderia, 48 in Rhizobium and 41 in Bradyrhizobium.
In order to establish if patterns between the occurrence of particular symbionts and plant phylogeny existed, we mapped the % of aand b-rhizobia (as means across all sampling locations) per legume species onto the trnL-trnF matK-trnK phylogeny ( Figure 1B). Although some legume species were poorly sampled, Burkholderia symbionts were still found along all the phylogeny of the Piptadenia Group species, and in the majority of most sampled species. A few legume species were devoid of Burkholderia symbionts, but these were mainly poorly sampled species, and so the presence of b-rhizobia in these species cannot be excluded.

Taxonomic Characterization of the Piptadenia Symbionts Collection
To further characterize the rhizobial collection at the taxonomic level and assign a putative species to each strain, a phylogenetic analysis of all representative strains based on a partition of two neutral markers (643 bp of the variable part of the 16 S rRNA gene, and a 644 bp fragment of the recA gene), was built following a Bayesian analysis (see Mat&Methods). The recA gene is a more highly resolved phylogenetic marker than the 16 S rDNA gene, and has proven to be a valuable tool for discriminating species within the Burkholderia genus [11], [12], [59]. We also included type strains of the most closely related species in the genera Burkholderia, Rhizobium and Bradyrhizobium. A phylogenetic tree of the 16 S-recA dataset is presented in Figure 2A (only for the genus Burkholderia, as a-rhizobia are presented in Figure S3); and individual gene marker phylogenies are available as Figure S4. The bacterial collection was clustered into 9 highly supported clades (posterior probabilities .0.9, bootstrap values .75%) in the genus Burkholderia (107 strains), 8 clades in Rhizobium (48 strains) and 4 in Bradyrhizobium (41 strains). In Burkholderia, most strains belonged to clades associated with the species B. nodosa (BN, 31 strains), B. phenoliruptrix (BPL, 27 strains), B. sabiae (BSa, 21 strains), and B. diazotrophica (BD, 18 strains). Other strains belonged to B. phymatum (BP, one strain), B. caribensis (BC, two strains), and three potentially undescribed species: Burkholderia sp. 1, 3 and 5 (BSP1, two strains, BSP3, four strains, BSP4, one strain). Burkholderia strains were identified among all legume species in this study, except for Mi. foetidus, Ps. psilostachya and P. adiantoides. For these three species, few rhizobial isolates were obtained (3-7), all belonging to Rhizobium and/or Bradyrhizobium genera. Each clade includes strains isolated from various hosts from the Piptadenia Group, except the BSP3 clade which was sampled only from P. gonoacantha. Some clades (BSa, BP, BC, BD, BN) contained strains isolated both from the genus Mimosa and the Piptadenia Group, while others were only found associated to the Piptadenia Group (BPL, BSP1, BSP3, BSP4) or to Mimosa species (BM, BSy, BSP2, BT). In the case of the a-rhizobia ( Figure S3), strains were clustered into 8 new clades of Rhizobium and 4 of Bradyrhizobium. Rhizobium sp. 1 (RSP1), R. tropici and RSP3 were the most represented clades with 25, 10 and 8 strains, respectively. R. tropici strains were found on the most important number of legume host (6 species). Strains of R. leucaenae and R. gallicum were also found on A. peregrina and P. flava [42], respectively. Rhizobium isolates from Pp. rigida from the Taulé et al. study [44] were grouped within the R. mesoamericanum species (UYPR7.63 & STM3625; Figure S4), a frequently sampled symbiotic species from Mimosa pudica [12], [60], [61]. In Bradyrhizobium, four clades could be distinguished and these were named Bradyrhizobium sp1 (BrSP1) to BrSP4, with no close relationship to any described species (except BrSP4 that was closely related to Br. elkanii). No obvious pattern could be deduced between host plants and a-rhizobial clades. Strains trapped from two soils by Stryphnodendron sp. were only characterized on the basis of their 16 S rDNA, and these grouped closed to Bradyrhizobium sp. 1 ( Figure S4).

Distribution of Rhizobial Species per Host Plant and Sampling Locations
A distribution of rhizobial species according to their plant host and site of sampling is presented in Figure 3. We observed a strong site sampling effect since the diversity pattern of encountered species was completely divergent for the same host sampled on different sites. This pattern heterogeneity of symbionts was even found between a and b-rhizobia, since P. gonoacantha and A. peregrina nodules hosted mainly Rhizobium strains in some sites but Burkholderia species on other sites, indicating no strict specificity of these species for a or b-rhizobia. Burkholderia nodosa was particularly frequent on the three different sampling sites of Pp. pterosperma, and was also found in P. trisperma, A. peregrina and P. gonoacantha. Burkholderia sabiae was also a promiscuous species in the Piptadenia Group (found in A. peregrina, P. viridiflora, P. trisperma, A. peregrina). On the other hand, BSP1 and BSP3 species were only found in P. gonoacantha nodules. On the plant side, some species exhibited the same pattern of symbionts diversity in the same geographical area (for example P. trisperma and Pp. pterosperma in Cabo Frio, RJ; Figure 3). However, the opposite could also be found, as P. gonoacantha and P. paniculata in Cabo Frio also exhibited divergent patterns of symbiotic association, one being nodulated by band the other by a-rhizobial species.

Phylogeny of Symbiosis-related Genes of Piptadenia's Rhizobia
Phylogenetic analyses were performed on fragments of the nodC (involved in the synthesis of the Nod factor core) and nifH (involved in nitrogen fixation) genes to evaluate the origin and evolution of symbiosis in Piptadenia Group microsymbionts. The nodC and nifH data sets used for phylogenies included representative strains listed in Table 1, as well as reference strains from each clade of the Bontemps et al. [11] and Mishra et al. [12] diversity studies on Mimosa species. A nodC ML phylogeny is shown in Figure 2B, while the nifH ML tree is presented as Figure S5. All nodC sequences from Burkholderia symbiotic species of the Mimoseae tribe were monophyletic compared to a-rhizobia. Burkholderia strains from the Piptadenia Group were clustered into five nodC clades (nodC1 to nodC5 on Figure 2B). Some nodC clades corresponded to already described bacterial species, such as B. nodosa (nodC5), BSP3 (nodC2), while other species had strains carrying different alleles of nodC (e.g. B. sabiae with nodC1 or nodC3 alleles; B. phenoliruptrix with nodC3 and nodC4 alleles). In the case of B. phenoliruptrix, strains carrying either nodC3 or nodC4 were both efficient symbionts of M. pudica, while B. sabiae nodC variants (nodC1, nodC3) exhibited different nodulation phenotypes on M. pudica (inefficient/efficient, Table S4). Some clades carried strains sampled from Mimosa and Piptadenia Group species (nodC3, C5), while others were specific to either the Piptadenia Group (nodC1, C2, C4) or to Mimosa spp. (nodC6, C7). The large nodC3 clade could be divided into 2 sub-clades corresponding to different species: nodC3-1 hosted B. sabiae and BSP1, while nodC3-2 contained B. phymatum, B. phenoliruptrix and B. diazotrophica strains. The large nodC3 clade hosted strains from every studied legume species, except for P. rigida that was restricted to the nodC1 clade. Overall, no clear pattern was identified concerning nodC cladogenesis and legume hosts (except the nodC2 clade that was specific to P. gonoacantha). A better correspondence was found between the nodC and 16 S-recA trees. However, although several 16 S-recA clades corresponded to nodC clades, horizontal gene transfer was also identified in species which carried different nodC alleles (e.g. B. sabiae with BSa-1 and 2, and B. phenoliruptrix with BPL-1 & 2, Figure 2B). The nifH gene ML tree ( Figure S5) reflected the same cladogenesis as the nodC tree, but with a few exceptions e.g. some strains in the nodC1 clade (UYPR1.313, UYPR3.611 and Prigida2) were clustered in a different clade in the nifH tree (nifH3).
In the case of a-rhizobia from the Piptadenia Group ( Figure  S2B), nodC sequences from Rhizobium strains (R. tropici, RSP1, RSP2, R. leucaenae) were clustered in a clade (nodC12) together with R. tropici strains from diverse Mimosa species isolated from Papua New Guinea (NGR181) and Puerto Rico (UPRM8021) [28], [42]. There was no sequence variation on nodC fragment  Table S3. doi:10.1371/journal.pone.0063478.g002 between UPRM8021 and the bean symbiont R. tropici CIAT899 T (not shown). The only exceptions were R. gallicum strains from Piptadenia flava (e.g. UPRM8060) that grouped close to the R. etli symbiovar (sv) mimoseae clade (nodC8), and to a R. mesoamericanum strain from Pp. rigida (nodC13) that grouped close to S. meliloti 1021. The nodC and/or nifH genes from some Rhizobium strains isolated from nodules of P. paniculata (STM7339, STM7342) could not be amplified: these strains did not nodulate M. pudica, siratro or P. gonoacantha, and could not be tested on their original host P. paniculata. The symbiotic nature of these two strains thus remains uncertain. Bradyrhizobium strains from P. paniculata and Mi. foetidus clustered in three different clades (nodC9 to nodC11), with the P. paniculata isolates (nodC9, C11) being separated from Mi. foetidus ones (nodC10). Notably, Bradyrhizobium isolates from P. paniculata could not be tested on their original host, but they did nodulate siratro but not M. pudica. As observed for Burkholderia symbionts, the nifH tree ( Figure S5) reflected the nodC groupings of the arhizobia.

Assessment of Symbiotic Specificity in the Piptadenia Group Using Competition Assays
In order to assess plant specificity in the Piptadenia Group, we carried out a coinoculation experiment with 12 rhizobial genotypes (6 Burkholderia, 3 Rhizobium and 3 Bradyrhizobium strains, representative of species and nodC clades) on 6 Piptadenia Group species (Mf, Pm, Ppr, Pg, Ap, Ac) for which seeds were available. The final range of nodules analyzed were 12 to 60 on 5 to 9 plant replicates. The nodule occupants were then identified as described in the Material and Methods section. Table 2 shows the percentage of nodule occupancy per bacterial genotype for each plant host. The percentage for a or b-rhizobia was also included in Figure 1C for comparison with wild nodule sampling % and plant phylogeny. Overall, all plant species showed a preference for Burkholderia genotypes, except for Mi. foetidus which was nodulated preferentially by bradyrhizobia. One Burkholderia genotype, STM7353 (BN, nodC5) was the most prevalent (in 38 to 85% of nodules) over all other genotypes on all plant hosts (except for Mf, 9% of nodules), followed by STM7317 (BPL, nodC4) that occupied 30 to 33% of plant hosts (except on Mf, 9%, and Pg, 3%). Interestingly, the two B. sabiae genotypes tested (each carrying either the nodC1 or nodC3 allele) did not show the same nodulation pattern: STM7373 (nodC1) was present at around 20% on Mf and Ac while STM7315 (nodC3) was absent, and conversely STM7315 was present at 20% on Pm while STM7373 was absent. Such results indicate potential host plant specificity towards several nodC alleles, although competitiveness for nodulation also interferes with these results. Genotypes from the Rhizobium and Bradyrhizobium genera were poorly competitive, and only STM7379 (RT, nodC2) and STM7378 (BrSP4) were identified in one host, Mi. foetidus (these two strains originating from nodules of this species). It is interesting to note that if Mi. foetidus favors a-rhizobia symbionts, this plant species selected its native genotypes rather than a-rhizobia from P. paniculata, thus underlining specific affinities between this host and its symbionts. Another case of potential host plant specificity is that of STM7296 (BSP3, nodC2), as this poorly competitive strain was only detected on its original plant host, P. gonoacantha.

Burkholderia Species are Common and Highly Diverse Symbionts of the Piptadenia Group
Burkholderia symbiont diversity has been intensively studied on collections isolated from nodules of different species of Mimosa, most of them originating from Brazil [11], [14]. In the present study we extended the nodulation by Burkholderia symbionts to other species in the Piptadenia Group, including six genera that are sister clades to Mimosa in the tribe Mimoseae. The Piptadenia Group contains about 70 species (if excluding the genus Mimosa, that is also part of this group), which are common in tropical and subtropical America, but are mainly native to Brazil [30]. The capacity of nodulation among these clades has been investigated in previous works (see Introduction), but one of them, P. trisperma, is a new report of nodulation from this study. A large diversity of Burkholderia was found, with nine species, five being already described as symbionts of Mimosa species (B. sabiae, B. phymatum, B. caribensis, B. diazotrophica, B. nodosa; [8], [18], [19], [17], [21]), one was not previously reported as containing symbiotic strains (B. phenoliruptrix, [10], [62]), and BSP1, BSP3 and BSP4 are three putative new species, whose taxonomic status is currently being investigated (Bournaud et al., unpublished). This diversity of symbiotic Burkholderia species mirrored the one discovered on 47 species of Mimosa by Bontemps et al [11] in Brazil, which also had seven clades, of which only three overlap with this study and are symbionts of both  [11], [12] but was not detected in Piptadenia group nodules. However, these patterns are only based on our current knowledge, and given the high sampling site effect observed on the diversity of symbionts in this group, it is likely that each bacterial species could also be found in both groups of legume hosts. The legume ability to form a symbiosis with Burkholderia is not antagonistic with the presence of a-rhizobia, since the two kinds of symbionts were found on six plant species (Table 1, Figure 1B), and some species had only a-rhizobia strains sampled from them. No clear conclusion about affinities of plant species towards a or brhizobia could be drawn for several legumes for which the nodule sampling was limited (i.e. Py. monoliformis, P. flava, P. stipulacea, Pseudopiptadenia spp), or for which the site sampling effect was too strong (Fig. 3, P. gonoacantha, A. peregrina). In order to investigate this affinity, we conducted coinoculation tests with 12 genotypes of a and b-rhizobia, and could conclude that Py. monoliformis, Pp. rigida, P. gonoacantha, A. peregrina and A. colubrina had a strong preference towards Burkholderia genotypes, while Mi. foetidus had an affinity for two genotypes of Rhizobium and Bradyrhizobium, as was also found in the wild sampling study. Another species, P. paniculata, remains ambiguous in terms of symbiont preference. Unfortunately we could not investigate this topic due to the unavailability of seeds. In the case of Stryphnodendron species, trapping assays on soils recovered from the rhizosphere of two trees (identified as St. adstringens) trapped only Bradyrhizobium species. The species St. pulcherrimum species is nodulated by strains close to R. tropici but also by B. nodosa in French Guiana (Christine Le Roux, unpublished data), indicating that this genus is not only nodulated by a-rhizobia. The Stryphnodendron genus thus warrants a more focused study in order to characterize its symbionts.

Symbiosis Genes in Burkholderia from Mimosoid Legumes are Monophyletic and have been Transferred both Vertically and Horizontally Among Bacterial Species
The Burkholderia strains from Mimosa and from the other Piptadenia Group species are monophyletic in terms of their nodC and nifH phylogenies, indicating a common ancestor at the origin of their interaction with members of the tribe Mimoseae. The comparison of symbiotic and neutral markers revealed that vertical gene transfer is the main process for the dissemination of symbiotic genes within Burkholderia species. This pattern has been already observed for Mimosa symbionts [11], [12], and has allowed for the description of the symbiotic nature of Mimosa-nodulating burkholderias as being ''ancient'', as it has been tentatively dated at c. 50 MY ago [11]. However, we found also that two species of Burkholderia (B. phenoliruptrix and B.sabiae) harbor strains with different nodC alleles, which were either inherited by vertical transfer (nodC3/BSa-2, nodC4/BPL-2), or were acquired by horizontal gene transfer from another Burkholderia clade (nodC1/ BSa-1, nodC3/BPL-1). Horizontal gene transfer thus also played a role in the co-adaptation between Burkholderia species and their legume hosts.

Host Specificity between Burkholderia Genotypes and Piptadenia Group Species is not Strong
Although gene phylogenies have informed us about the ancient character of the symbiosis between burkholderias and Piptadenias, host specificity between the different partners is not obvious. Host plants are dispersed all over the Burkholderia phylogenies ( Figure 2A,B), with no specificity pattern, except for that between BSP3 and P. gonoacantha. We tried to assess the specificity of the interactions between legume and rhizobial partners using coinoculation experiments, and detected i) a high affinity of most tested plants towards Burkholderia genotypes, ii) promiscuous strains with high competitivity for nodulation (B. nodosa, B. phenoliruptrix), as well as some trends towards specificity (e.g. P. gonoacantha/BSP3, Mi. foetidus/Bradyrhizobium). Given the strong site sampling effect detected in the sampling analyses (Figure 3), and the absence of strict host specificity, it is realistic to hypothesize that most Piptadenia Group and Burkholderia species interact with a relaxed host specificity, and that plant selection targets nodulation genes from the Burkholderia nodC monophyletic clade or from diverse arhizobia when no symbiotic Burkholderia are present, and there are few exclusions in order to maximize their chances of finding a compatible partner. The environmental (especially soil) conditions would thus play a crucial role in the survival and biogeography of the symbionts, and could be the origin of the different diversity patterns observed in our results. Several parameters have already been found to affect Burkholderia symbionts diversity and/or competition, such as soil pH [12], [24], [16], altitude [11], or nitrogen sources [28].

Conclusion
In this study we extended the symbiotic interaction with Burkholderia from Mimosa to the wider Piptadenia Group in the tribe Mimoseae. Given the ancient and diverse character of this interaction, symbiosis with Burkholderia species might be present in other genera of this tribe, and possibly even in other mimosoid tribes, such as the Ingeae, as Barrett and Parker [63] have isolated Burkholderia strains from a member of this tribe (Abarema macradenia). The fact that B. phymatum STM815 is able to nodulate several legumes in the subfamily Mimosoideae (including species in Leucaena, Prosopis, and Acacia) give clues on the possible extent of the host range of these symbionts [7]. Symbiosis with Burkholderia could thus be much more common and ecologically significant than anticipated, encompassing several tribes in the Mimosoideae.
Given the relaxed specificity between Piptadenia Group species and their symbionts, and the diversification of Burkholderia into many species, larger samplings of each species in relation to soils and environmental parameters are required in order to get a clearer picture of the diversity and phylogeography of b-rhizobia, and how they have co-evolved with species in the Mimosoideae.