Legumes in the genus Aeschynomene form nitrogen-fixing root nodules in association with Bradyrhizobium strains. Several aquatic and subaquatic species have the additional capacity to form stem nodules, and some of them can symbiotically interact with specific strains that do not produce the common Nod factors synthesized by all other rhizobia. The question of the emergence and evolution of these nodulation characters has been the subject of recent debate.
We conducted a molecular phylogenetic analysis of 38 different Aeschynomene species. The phylogeny was reconstructed with both the chloroplast DNA trnL intron and the nuclear ribosomal DNA ITS/5.8S region. We also tested 28 Aeschynomene species for their capacity to form root and stem nodules by inoculating different rhizobial strains, including nodABC-containing strains (ORS285, USDA110) and a nodABC-lacking strain (ORS278).
Maximum likelihood analyses resolved four distinct phylogenetic groups of Aeschynomene. We found that stem nodulation may have evolved several times in the genus, and that all Aeschynomene species using a Nod-independent symbiotic process clustered in the same clade.
The phylogenetic approach suggested that Nod-independent nodulation has evolved once in this genus, and should be considered as a derived character, and this result is discussed with regard to previous experimental studies.
The Leguminosae (Fabaceae) is the third largest family of flowering plants with c. 730 genera and over 19 400 species worldwide (Mabberly, 1997; Lewis et al., 2005). Legume species are particularly diverse, both in size and in ecological habitat, and include small herbs from temperate regions as well as large tropical rainforest trees. They are agriculturally and economically important, being second only to the Poaceae (e.g. cereals). In addition to the most cultivated crops, such as soybean (Glycine max), common beans (Phaseolus vulgaris), peas (Pisum sativum), and peanuts (Arachis hypogaea), that are harvested for grain or oil, legumes are also valued for timber, fuel, forage and medicines. This economic importance of the Leguminosae is mainly the result of the ability of many of its species to form a symbiotic association with soil bacteria, commonly known as rhizobia. This symbiosis usually results in the formation of root nodules, in which the rhizobia reduce atmospheric nitrogen to ammonium. This allows the plant to grow well and produce protein-rich seeds in the absence of nitrogen fertilizer in soils. Among the three subfamilies of the Leguminosae (Caesalpinioideae, Mimosoideae, and Papilionoideae), nodulation by rhizobia is rare in caesalpinioids, more common in mimosoids, and very common in papilionoids (Sprent, 2007). In all three subfamilies, nitrogen-fixing nodules are almost exclusively located on roots. However, in a very few tropical legumes that are hydrophytic and that belong to the three papilionoid genera, Aeschynomene, Sesbania, and Discolobium, (Eaglesham & Szalay, 1983; Alazard, 1985; Eaglesham et al., 1990; Ladha et al., 1992; Loureiro et al., 1994), and to the mimosoid genus Neptunia (Schaede, 1940), nodulation by rhizobia can also occur on stems. Stem nodulation was observed for the first time on Aeschynomene aspera (Heagerup, 1928), and later well documented in Sesbania rostrata (Dreyfus & Dommergues, 1981; Goormachtig et al., 2004) and Neptunia plena (James et al., 1992). Meanwhile, stem nodulation was reported on several other species of Aeschynomene (Alazard, 1985; Becker et al., 1988), and this genus contains most of the stem-nodulated species described so far.
The stem nodulation phenotype in these various legumes is, in fact, represented by a number of different ontogenies. James et al. (1992) considered that genuine stem nodules should be vascularly connected to the stem, as observed in Aeschynomene and Discolobium (Loureiro et al., 1994, 1995; James et al., 2001). If this criterion is considered to be the main one by which the term ‘stem nodules’ is used accurately, then Neptunia, which has been shown to form ‘stem nodules’ that are connected by their vascular tissue to the bases of adventitious roots (James et al., 1992; Subba Rao et al., 1995), should thus be considered as root nodules. Similarly, Goormachtig et al. (2004) also considered that stem nodules on S. rostrata were, in fact, adventitious root nodules, especially as their development is morphologically equivalent to the development of lateral root base nodules. Environmental conditions also play a major role in stem nodulation. The stem nodules of Discolobium pulchellum and Discolobium leptophyllum are compulsorily aquatic, requiring permanent submergence in water or in flooded soil (Loureiro et al., 1994; James et al., 2001). Several Aeschynomene species (Aeschynomene elaphroxylon, Aeschynomene crassicaulis, Aeschynomene americana) also form nodules on the stem (at the base or all the way up the stem) only under waterlogged conditions (Boivin et al., 1997). However, only one of these species, Aeschynomene fluminensis, has ‘flooded’ stem nodules with a vascular system connected to the stem (Loureiro et al., 1995) rather than a connection to adventitious roots.
Other species, such as Aeschynomene afraspera, Aeschynomene indica and S. rostrata, differ in their ability to readily develop stem nodules even under nonsubmerged conditions (Boivin et al., 1997). A unique feature that is found only in stem-nodulating Aeschynomene spp. is their capacity to symbiotically interact with photosynthetic bradyrhizobia (Evans et al., 1990; Giraud & Fleischman, 2004). It has been shown in Aeschynomene sensitiva that the photosynthetic activity of these bradyrhizobia facilitates ex planta survival and infectivity, and thus it could affect their biological nitrogen fixation during stem nodulation (Giraud et al., 2000).
Based on the stem and/or root nodulation ability of Bradyrhizobium isolates, three cross-inoculation groups of Aeschynomene spp. were initially defined by Alazard (1985). Group I (representative species A. americana) formed root nodules and/or adventitious root nodules on the stem, and was associated with ‘classical’ Bradyrhizobium strains in group A (Fig. 1). Species of group II (A. afraspera) formed profuse stem nodules under nonsubmerged conditions with both nonphotosynthetic (group A and B) and photosynthetic (group C) Bradyrhizobium. Group III (A. indica) formed sparse stem nodules under nonsubmerged conditions with photosynthetic strains (groups C and D). Since then, this system has become more complicated, as A. fluminensis has been shown to form stem nodules with photosynthetic strains but only in aquatic conditions (Loureiro et al., 1995). Moreover, new nonphotosynthetic isolates able to form stem nodules on group III Aeschynomene spp. have been discovered (Miché et al., 2010). Most surprisingly, sequencing of the whole genome of two group D Bradyrhizobium strains (ORS278 and BTAi1) showed that they did not contain the canonical nodABC-genes (nod) genes required for the synthesis of Nod factors (NFs), the signal molecules produced by all other rhizobia that had always been suggested as compulsory for the initiation of symbiotic nodules on legumes (Giraud et al., 2007). This result was further confirmed by a comparative genomic study of six additional strains (groups D and E) that were representative of the phylogenetic diversity of Bradyrhizobium isolated from group III plants (Fig. 1) (Mornico et al., 2012). It has finally been shown that group C strains, which form nodules on both Aeschynomene groups II and III, contain the canonical nodABC genes and produce NFs (Chaintreuil et al., 2001; Renier et al., 2011). Deletion of the nodB gene in one of these strains (ORS285) blocked nodulation of A. afraspera (group II), but did not affect nodulation of A. indica or A. sensitiva (group III), thus proving that ORS285 is able to use both Nod-dependent and Nod-independent symbiotic processes, depending on the host plant (Bonaldi et al., 2011).
The genus Aeschynomene contains 161 (http://www.theplantlist.org) to 180 species (Klitgaard & Lavin, 2005), half of them described from the New World, mainly South and Central America, and the other half have been found across the tropical regions of Africa, Southeast Asia, Australia and the Pacific Islands (Rudd, 1955; Verdcourt, 1971). The genus includes both herbaceous and shrubby species, annuals and perennials, some of them growing up to 8 m in height and with a basal stem width of 0.5 m (e.g. A. elaphroxylon). Half of the species are hydrophytes growing in marshes, temporary or permanent ponds, rice fields, waterlogged meadows, and along streams and riverbanks. The remaining species are more xeric and are found in savannas or dry forests.
Botanically, the genus Aeschynomene belongs to the tribe Aeschynomeneae, which has now been classified together with the Dalbergieae tribe in a monophyletic group referred to as the dalbergioid legumes, a large, mostly pantropical, group of papilionoids characterized by the presence of the Aeschynomeneae type of root nodule (Lavin et al., 2001; Sprent, 2001). Rudd (1955) published a revision of the American species of Aeschynomene, but no attempt has been made to include Old World species in this classification. The genus Aeschynomene was originally divided by Vogel (1838) into two sections: Aeschynomene L. sect. Aeschynomene, also referred to as Eu-Aeschynomene, which comprises c. 50 species with a pantropical distribution, and Aeschynomene sect. Ochopodium Vogel with > 100 pantropical species (Polhill, 1981; Rudd, 1981; Klitgaard & Lavin, 2005). This division, also retained by (Rudd, 1955), is well supported, based on morphological differences; that is, section Aeschynomene is characterized by medifixed stipules, whereas section Ochopodium has basifixed stipules. The section Ochopodium is more closely related to the genus Machaerium than to sect. Aeschynomene and to Dalbergia (Lavin et al., 2001; Ribeiro et al., 2007). Consequently, the genus Aeschynomene does not appear to be monophyletic, but this topic n eeds to be further developed using additional species from both the New and Old Worlds. The delimitation of section Aeschynomene is also problematic, as it is morphologically closely related to other genera, such as Soemmeringia, Cyclocarpa, Kotschya, Smithia, Geissaspis, Bryaspis and Humularia (Rudd, 1981). Such close relationships were more recently confirmed by DNA sequence analysis (Lavin et al., 2001; Ribeiro et al., 2007).
The aim of this study was first to conduct a molecular phylogenetic analysis including 38 species of Aeschynomene originating from both the New and Old World. By using two genomic and chloroplastic markers, ncDNA (ITS1-5.8S-ITS2) and the cpDNA (trnL) locus, respectively, we reconstructed the phylogenetic relationship among these Aeschynomene accessions, together with related species and genera. We also estimated the within-species diversity of four Aeschynomene species (A. americana, A. indica, A. sensitiva, and Aeschynomene villosa) in order to infer its possible influence on the phylogenetic reconstruction, and to link it with the geographic distribution of the accessions. We then tested most species for their ability to form root and stem nodules with various rhizobial strains, harboring nod genes or not, and consequently constructed a putative evolution of the various nodulation types found in the genus. Taxonomical and nodulation issues are discussed in the light of this phylogenetic and nodulation character-based evolution construction.
Materials and Methods
This study included 38 neotropical or Old World different Aeschynomene species (Table 1), and from one to 11 accessions per species. Whenever possible, we included several individuals from the same species, sampled from the widest possible geographical area. Seventy-one different accessions were included in total, both from our experiments and via sequences retrieved from Lavin et al. (2001) and Ribeiro et al. (2007). Table 1 lists all taxa included in the study, their sources and geographic origin, their nodulation characteristics (when tested), and EMBL accession numbers. Several related genera (15) were also included in the analyses, as previous studies had suggested their close relationship and/or their phylogenetic intermingling with the genus Aeschynomene. All of these sequences, except for four from Smithia abyssinica and Kotschya lutea, were retrieved from GenBank and were originally published in Lavin et al. (2001) and Ribeiro et al. (2007).
Table 1. Characteristics of the species and samples included in the study
Plant DNA extraction, amplification, and sequencing
Total genomic DNA was isolated using the modified CTAB extraction method (Doyle & Doyle, 1987) from new leaves of plants germinated from seeds and which were grown in our glasshouse. The chloroplast DNA trnL intron (Bakker et al., 2000) and the nuclear ribosomal DNA ITS/5.8S region (Baldwin et al., 1995) were chosen for phylogenetic analyses because they have been shown to be informative within and among closely related legume genera (Lavin et al., 2001; Ribeiro et al., 2007). The trnL (UAA) intron was amplified and sequenced using primers B49317 and A49855 (Taberlet et al., 1991). Primer pairs used for PCR to amplify and sequence the ITS region flanked the end of the 18S RNA gene ITS18 and the beginning of the 26S RNA gene ITS26 (Kass & Wink, 1997; Beyra-M & Lavin, 1999; Delgado-Salinas et al., 1999). The PCR products were purified using a Quiaquick PCR purification kit (Qiagen) according to the manufacturer's instructions. Sequencing was performed by Macrogen inc. (Seoul, Korea) using a ABI Prism 377 DNA sequencer.
Bacterial strains and culture growth conditions
For both root and stem inoculation, three rhizobial strains were used: the nod gene-lacking photosynthetic Bradyrhizobium strain ORS278, the nod gene-containing photosynthetic Bradyrhizobium strain ORS285, and the ‘classical’ nonphotosynthetic Bradyrhizobium diazoefficiens strain USDA110. All strains were grown aerobically in Yeast Mannitol (YM) medium (Vincent, 1970) on a gyratory shaker (170 rpm) at 37°C, under a light : dark cycle (16 h : 8 h) for photosynthetic Bradyrhizobium.
Root and stem inoculation
Aeschynomene seeds were surface-sterilized and dormancy was broken with 96% H2SO4 for 10–45 min, depending on the seed size for each species, and then rinsed six times in sterile distilled water to remove all traces of acid. To allow germination, seeds were placed in sterile water for 24 h at 30°C. They were then transferred either to Gibson tubes to test for root nodulation, or to glasshouse pots (plastic pots 8 cm in diameter) containing 500 g of sterilized sandy soil for stem nodulation. Inoculations of plants by the different rhizobial strains were performed both at the root level (via inoculation in hydroponic conditions, owing to the semiaquatic habit of the tested plants) and at the stem level (via inoculation of nonsubmerged stems by their careful ‘painting’ with the inocula), as already described (Giraud et al., 2000). Roots of plants grown in tubes were observed for root nodule formation 2–3 wk after inoculation. Stems of plants grown in the glasshouse were observed for stem nodulation 2–4 wk after inoculation.
Sequences alignment and phylogenetic analyses
Multiple alignments were performed for the trnL sequences with ClustalX, version 1.63b (Larkin et al., 2007), and alignments were manually corrected using GeneDoc (Nicholas et al., 1997). Phylogenetic reconstruction was performed using a maximum likelihood approach. The best model of molecular evolution for the trnL alignment was chosen using jmodeltest (Darriba et al., 2012). Most probable trees were obtained using Phyml (Guindon et al., 2009) by implementing previously estimated parameters. Statistical tests for branch supports were estimated using nonparametric bootstraps calculated on 100 replicates, implemented in Phyml.
Owing to the very high amount of nucleotide divergence among the ITS sequences, a confident alignment, including all taxa, could not be achieved easily with the former approach. We therefore conducted a Markov Chain Monte Carlo (MCMC) analysis using the software package Bali-Phy (Suchard & Redelings, 2006). Alignment uncertainties are taken into account by integrating the overall alignments in proportion to their posterior probabilities conjointly with phylogeny topology estimations. Because of the slowness of the analysis with the entire data set, we performed a first run with only 17 sequences of the 63 in the data set. These sequences were chosen according to a rough alignment obtained with ClustalX followed by NJ clustering. Each cluster was checked for the high (and alignable) similarity among sequences falling in it, and one sequence per clade was chosen. We implemented a Tamura-Nei model of molecular evolution (Tamura & Nei, 1993) with the RS07 (Redelings & Suchard, 2007) insertion/deletion model. Five independent runs were performed with this data subset, with 30 000 iterations each. Following the Bali-Phy user's guide, we estimated the SD across runs of the posterior probabilities for each run and averaged the values across splits. We also used the potential scale reduction factor to check that different runs had similar posterior distributions. The consensus alignment of the 17 sequences was then used in a second step as a guideline for the alignment of the other sequences. Based on this alignment, we estimated three different models of molecular evolution corresponding to the three ITS1, 5.8S and ITS2 regions that clearly displayed very different amounts of mutation accumulation, and reconstructed the phylogeny using these three models implemented in MrBayes software (Ronquist et al., 2012). Node supports were estimated with the posterior probabilities obtained in the MCMC analysis.
In both phylogenies, the two sequences from Andira galeaottiana and Vatairea sp. were chosen as outgroups, based on both traditional classification (Polhill, 1981) and the results of previous DNA phylogenies (Hu et al., 2000; Lavin et al., 2001; Ribeiro et al., 2007).
The topologies of the two phylogenetic trees being roughly similar, we concatenated the two data sets and reconstructed a combined tree. ITS sequences were missing for five species (Smithia ciliata, Kotschya aeschynomenoides, Kotschya ochreata, Aeschynomene purpusii, and Geissaspis descampsii), and these were indicated as missing data in the data matrix. We used MrBayes software (Ronquist et al., 2012) and applied different models of molecular evolution on each dataset partition. The four models (three for ITS and one for trnL) applied to the full dataset were used in an MCMC phylogenetic search. Searches were performed three times to check that the same equilibrium and final topology were achieved each time.
In order to study the evolution of the genus, we mapped onto the concatenated ITS-trnL phylogeny the two nodulation characters, that is, stem and Nod-independent nodulation abilities. We applied an unordered and equally weighted scheme, considering that transitions in either direction between the different states of character were equally likely, with no a priori assumptions concerning the ancestral state of the clade.
Sampled species, their localities, voucher specimens, and GenBank data base accession numbers for trnL intron and ITS sequences are listed in Table 1.
ITS and trnL sequences
We sequenced 65 different plant accessions (see Table 1) from Aeschynomene and other genera in this study. Together with other sequences retrieved from the databank, in total we analyzed the sequences for 71 Aeschynomene accessions and 24 from other related species. In contrast to almost the whole set of analyzed accessions for which single PCR products were obtained after amplifying the ITS and trnL markers, the electrophoregrams obtained for the ITS sequences from some A. indica accessions displayed several double peaks. As these double peaks were common and at the same position as these accessions, we cloned and sequenced the PCR product for the accession USDA PI225551, and two different sequences were subsequently obtained. A similar situation was also previously observed for an accession of Aeschynomene evenia (IRFL 6945), for which two different ITS sequences were obtained (Arrighi et al., 2013). Therefore, for these two species (A. indica and A. evenia), as the two sequences obtained were very similar each time they were analyzed (six Single Nucleotide Polymorphisms over 617 bp for both), and also fell into the same clade (data not shown), we decided to consider only one copy of ITS for each species for the subsequent phylogenetic analysis.
The Aeschynomene ITS region obtained for 70 accessions ranged in length from 597 bp (Aeschynomene filosa) to 623 bp (A. americana). The trnL regions were sequenced for 63 Aeschynomene accessions, and eight other sequences were retrieved from GenBank. Sequences ranged from 436 bp (A. villosa and Aeschynomene parviflora) to 477 bp (A. afraspera and A. nilotica). Five Aeschynomene species were represented by > two accessions: A. americana (six accessions), A. indica (11), A. sensitiva (seven), Aeschynomene rudis (four) and A. villosa (four). The within-species nucleotide divergence among sequences was low, ranging from 0 to 1% for the ITS sequences. Divergence was even lower for trnL sequences, with a maximum value of 0.4% (two mutations over the 456 bp sequence length for A. indica). Phylogenetic analyses always clustered all accessions from the same species in the same clade (data not shown). Consequently, we only kept one accession per species in all subsequent analyses.
Alignments and phylogenetic analysis
The trnL aligned matrix included 39 sequences, each from a different Aeschynomene species, and 31 other species from 22 different genera. The ITS analyses resulted in a 786 bp long matrix, with 38 Aeschynomene sequences and 27 from other genera. The two phylogenies being very similar, we combined the two data sets into a single phylogenetic analysis (the two single locus trees are given and described in Supporting Information Figs S1, S2).
For this, we reconstructed a Maximum Likelihood Phylogeny (Fig. 2). Four Aeschynomene clades were found, each supported with high posterior probabilities from the MrBayes analysis. Clade 1 grouped together 10 Aeschynomene species, all from subgenus Ochopodium, in a sister clade to the two Machaerium and Dalbergia genera. Other Aeschynomene species fell into the same monophyletic branch, but were mixed with other genera. The first emerging clade 2 grouped four Aeschynomene species, clade 3 grouped 12 species, and finally clade 4 grouped together 12 Aeschynomene species plus five other genera in a clade with an unresolved basal branching. Several species from each of the two genera Smithia and Kotschya were grouped together, suggesting a true evolutionary relationship among them within each genus. A close relationship among Aeschynomene bella, Aeschynomene abyssinica and Geissaspis descampii was observed, but this was based on a single trnL sequence of G. descampii retrieved from Genbank and thus should be verified with alternative sequences and vouchers. The grouping of Bryaspis lupulina with several Aeschynomene species in an unresolved, well supported clade is also notable. It is worth noting that the two stem-nodulating Discolobium species, which were included in the phylogeny, formed another clade that clearly belongs to the Dalbergia tribe, but it is not related to the genus Aeschynomene.
Evolution of symbiotic features among Aeschynomene groups
Fifty-six Aeschynomene accessions were tested for stem and root nodulation using three different strains, USDA110, ORS285 and ORS278, which permitted the distinction of the three inoculation groups defined by Alazard (1985) (cf. Fig. 1). Our tests did not reveal any within-species variation for nodulation phenotype when several accessions were used (Table 1). Among the 26 Aeschynomene species tested, 10 were nodulated only by strain USDA110, and formed only root nodules, and thus these Aeschynomene species were classified as belonging to the inoculation group I. Four species formed root nodules with USDA110, and root and stem nodules with the nod gene-containing strain ORS285, and thus belonged to the inoculation group II. Aeschynomene fluminensis could form stem nodules with ORS285, but no stem nodules in nonsubmerged conditions, which is in accordance with the previous observations of Loureiro et al. (1995), who showed that this species could form stem nodules with photosynthetic bacteria, but only under flooded conditions. Interestingly, among the Aeschynomene species, only A. crassicaulis and Aeschynomene fluitans possess the notable characteristic of having a floating stem with nodules developed on it (Fig. 2). For A. crassicaulis, we confirmed it could be nodulated only by the nonphotosynthetic strain USDA 110, and that flooding was compulsory for stem nodulation, as reported previously by Boivin et al. (1997). Finally, 12 species formed root and stem nodules with ORS285, but also with the nod gene-lacking strain ORS278, and were thus defined as belonging to inoculation group III.
We mapped the two symbiotic characters, stem nodulation and Nod-independent nodulation, onto the combined data set phylogeny. For other genera for which we did not perform any nodulation tests, and for which no published information is available with regard to their stem nodulation, we considered them as classical root-nodulating species. Similarly, none of them were considered as able to nodulate with strains lacking nod genes, even if we could not firmly exclude this option (see the 'Discussion' section). Based on an unweighted scheme, and no a priori assumptions about ancestral states, the concatenated phylogeny suggested at least three independent emergences of the stem nodulation ability, leading to A. fluminensis, clade 3, and a group of four species within clade 4 (Fig. 2).
The evolutionary pattern of the history of the Nod gene-independent character appears to be relatively simple, as all the Aeschynomene species that can form an efficient symbiosis with the nod gene-lacking strain clustered in a single clade, thus revealing a unique emergence of this ability.
All the aquatic Aeschynomene species fall into a single clade, but delineation of the genus requires revision
The genus Aeschynomene is complex, containing between 160 and 180 different species, possibly more, and is still increasing in size, as suggested by the recently described new species Aeschynomene sousae and Aeschynomene sabulicola (Queiroz & Cardoso, 2008; Delgado-Salinas & Sotuyo, 2012). As shown in previous studies (Lavin et al., 2001; Ribeiro et al., 2007), the genus Aeschynomene is polyphyletic in the two phylogenies we obtained, with species falling into two well-separated clades. The fact that the ITS and trnL sequences are retrieved from nuclear and chloroplastic genomes, respectively, reinforces the confidence we have in a true evolutionary split between the two clades. Although the aim of this study was not to redefine the borders of the genus, the phylogenies strongly suggest that the subgenus Ochopodium (clade 1 in our phylogeny) should be elevated to the genus rank, as a sister clade of Machaerium. Moreover, since A. aspera L., the type species of the genus Aeschynomene, falls into clade 4, the subgenus Ochopodium could not retain the name ‘Aeschynomene’ and would thus require renaming. All the other Aeschynomene species fall into the same main branch. The first two emerging clades (2 and 3) mostly contain American species (with the exception of Aeschynomene tambacoundensis, which is endemic to West Africa, and A. indica, which has a pantropical distribution), whereas clade 4 includes African species and one Asian species (A. aspera). America thus appears to be the center of origin of the genus, with a secondary center of diversification in Africa.
All these Aeschynomene species within this main branch (i.e. clades 2–4) share the same aquatic or semiaquatic habitat. Other related genera (Kotschya, Smithia, Geissaspis, Soemmeringia, and Bryaspis) can also be found in humid habitats, although they are not described as hydrophytes (Lewis et al., 2005). This ability to grow in or at the border of ponds (either permanent or temporary) has apparently been acquired before the diversification of all these species and genera, and might play a role in the ability of some Aeschynomene species to make stem nodules (see next paragraph).
The intermingling of the different genera makes their true delimitation unclear. Bryaspis and Soemmeringia include two and one species, respectively, whereas the others contain from three (Geissaspis) to 30 (Kotschya) different species. We cannot rule out the possibility that the molecular phylogenies, based on two sequences, might give false, or uncertain, evolutionary reconstructions. In addition, misidentification, as well as other factors that could have confused the phylogenetic pattern (i.e. lineage sorting, ancestral polymorphism) might have interfered in our results. Nevertheless, and in spite of these possibilities, whether several Aeschynomene species should be transferred to another existing genus, or conversely, whether several other genera should be included within a larger genus Aeschynomene remains an open question that should be explored further.
The stem nodulation character has emerged, or been maintained, several times in the aquatic Aeschynomene species
The scarcity of stem nodulation among the legume genera has been underlined previously, being reported in only four different genera (Aeschynomene, Discolobium, Neptunia, and Sesbania; Boivin et al., 1997), with ‘genuine’ stem nodules only recognized in the first two genera, as defined in the 'Introduction'. The distribution of stem-nodulating species, with or without genuine stem nodules, throughout the Leguminosae (Lavin et al., 2001; Wojciechowski et al., 2004; the present study) strongly suggests that the ability to form these structures on stems has evolved independently several times.
Contrary to the three other stem-nodulating genera, stem nodulation in the genus Aeschynomene is widespread, which leaves open the possibility of analyzing its evolution at a much narrower evolutionary level.
Excluding clade 1 (Ochopodium) from the analysis, the most parsimonious reconstruction of evolution suggests that the ‘stem nodulation ability’ has possibly emerged three times during the diversification of the genus. The true number of transition and/or reversions is obviously difficult to assess, especially as the probability of occurrence of each event is most certainly unequal. In addition, its distribution among all these species clearly shows that this character is not stable, with several emergences or losses occurring alternately within a short period (at least in evolutionary terms).
Interestingly, clades 2–4 include the hydrophytic Aeschynomene species, and as all stem-nodulated Aeschynomene are hydrophytes, it suggests an influence of waterlogged/flooded conditions on the emergence of the stem nodulation characteristic. All the other genera within clade 4 grow along riverbanks or ponds (Lewis et al., 2005), but none have been reported to form stem nodules. A hydrophytic ecology thus appears to be an essential requirement for the evolution of the ability to form stem nodules, but it is not on its own sufficient for it, and so it would appear to be also related to factors other than environmental ones, such as bacterial ecology, or to specific recognition mechanisms between the bacteria (rhizobia) and the host plants.
Stem nodulation in the genus Aeschynomene might have evolved in a two-step process, first with a genetic predisposition (at the base of the clade) to produce adventitious root initials all along the stem. Previous studies have shown that rhizobia colonize the stem via epidermal fissures (cracks) generated by the emergence of adventitious root primordia (Sprent, 1989; Boogerd & van Rossum, 1997) A second, still unknown, mutation that would have appeared several times will have led to the various clades in which the true stem-nodulating species fall. This is possibly linked to the ability of protruding root primordia to pierce the epidermal layer and thus to form at their base a large annular cavity in which the bacteria can easily multiply (Boivin et al., 1997; Giraud et al., 2000). Deciphering the genetic mutation(s) that drive stem nodulation in each species would help in confirming this two-step hypothesis.
The phylogeny of Aeschynomene suggests that Nod gene-independent nodulation is a derived character
All Aeschynomene species that are able to form an efficient symbiosis with bacteria lacking nodulation genes, and thus not producing NFs, fall into a single clade. Moreover, this clade does not include any other species that compulsorily require NFs to interact efficiently with symbiotic rhizobia. The most parsimonious and simplest view of evolution leads to a single emergence of the capacity to interact with bacteria without the production of NF. Consequently, Nod gene-independent nodulation should be viewed as a derived and more recent character compared with NF-mediated nodulation, as previously suggested by Okubo et al. (2012).
However, this evolutionary scenario might not be so simple. Recently, Madsen et al. (2010) showed that Lotus japonicus double mutants, affected in some determinants of the NF perception and signaling pathway, were occasionally able to form functional nodules when the plants were inoculated with a compatible rhizobial strain unable to produce NF. In such cases, nodules could be formed after a bacterial intercellular infection of the root (i.e. without the formation of infection threads). Based on these results, the authors suggested that direct intercellular infection may constitute an ancient invasion path, and that the most highly evolved state envisaged would be the root hair infection mode that requires NF receptors, which is in direct contradiction with our phylogenetic conclusions based upon the genus Aeschynomene.
The uncertainty as to whether NF independence is an ancestral or derived character cannot be dispelled easily. The symbiosis between actinorhizal plants and Frankia, which is considered to have emerged before the Rhizobium–legume symbiosis, does not involve the synthesis of NFs by the bacteria, but it does recruit some common determinants of the Nod-dependent signaling pathway described in model legumes (Normand et al., 2007; Gherbi et al., 2008). Following on from these hypotheses, Okubo et al. (2012) raised several possible scenarios, including both Nod gene-independent and Nod gene-dependent ancestries, and even an hypothesis in which the Aeschynomene ancestor was not nodulated, but later acquired the Nod gene-independent symbiotic pathways. It should also be stressed that, as underlined by Madsen et al. (2010), the two alternative invasion modes, Nod gene-independent or Nod gene-dependent, are not mutually exclusive.
The question of whether NF independence is ancestral, derived, or mixed ability remains open, and illustrates how experimental vs phylogenetic approaches might give conflicting results. This question is closely akin to that posed by Masson-Boivin et al. (2009) who, among several outstanding questions to be explored, asked if ‘symbiosis evolved from primitive (e.g. NF-independent, crack entry) to sophisticated (NF, infection thread) strategies’, although it should also be recognized that the NF-independent symbiosis should not necessarily always be viewed as being more primitive than the NF-dependent one, as, in evolution, simplest does not necessarily mean less evolved. Deciphering the details of the NF-independent strategy will be the next step for elucidating which of these alternative evolutionary possibilities is the correct one. As suggested by Arrighi et al. (2012), it might be achieved through the use of A. evenia as a model legume, as it displays all the characteristics required for genetic and molecular analyses (i.e. it is a short-perennial and autogamous diploid species with a relatively small genome).
Nod-independent nodulation occurrence and evolution
We cannot fully reject the possibility that the Nod gene-independent nodulation process was retained in genera and clades other than Aeschynomene, especially those that are infected following a crack entry process, except that it has never been demonstrated or proven before. The intercellular infection process observed in Aeschynomene species is supposedly found in > 25% of legumes (Sprent, 2007). We may then consider that within these thousands of species, among which a majority have never been studied in terms of their nodulating symbiosis, many of them might be able to use a Nod gene-independent interaction mechanism with their symbiotic rhizobial partners. At present, no specific genetic markers exist, in either the plant or the bacterial symbionts, to easily detect such ability, or to study its frequency and distribution along a wide taxonomic and phylogenetic sampling. The only alternative currently available is to search for the presence, or not, of nod genes within bacterial symbionts isolated from nodules, which is not trivial, except through full genome sequencing. Moreover, one plant species might be able to interact with Nod gene-producing bacteria, but still be able to interact in a Nod gene-independent manner, such as observed for the group III Aeschynomene species, thus making it even more complex to analyze the system.
On the other hand, it is worth noting that all rhizobial strains sampled so far, either with or without nod genes, which can make efficient nodules with group III Aeschynomene species, fall into a single clade (Fig. 1). This pattern suggests that the ability to form nodules without NF, rather than being driven solely by the plant, has also been dependent on a specific single bacterial evolution/mutation. Aeschynomene species in group III only form an efficient symbiosis with group D and E bacterial isolates, and reciprocally, nod gene-lacking strains are strictly specific to group III Aeschynomene species. This reciprocal specificity is reminiscent of the gene-for-gene interaction in phytopathology (Flor, 1942), and opens up opportunities for a coevolution process. The current associations might thus have resulted from both the bacterial acquisition of a Nod gene-independent specific recognition mechanism (e.g. one linked to an unknown receptor), and a specific mutation in the plant ancestor, that together would have mediated the high specificity observed in this interaction. The scarcity of Nod gene-independent symbiotic interactions among legumes could then be explained by the requirement for such a joint evolution of the two symbiotic partners.
Obviously several questions remain, such as the nature of the evolutionary pressures that would have driven the concordant evolution between the two partners during the emergence of Nod gene-independent nodulation, as well as the role of photosynthesis in the diversification of the symbiotic bacteria. All these questions, plus the genetic and physiological investigations underlying them, will pave the way for many fascinating new studies.
We are very grateful to M. Boursot for help with the glasshouse experiments, and P. Tisseyre for management of strain collection. We would like to thank F. Crozier and S. Gonzalez from ‘Herbier de Guyane’ (Cayenne, IRD) for their help in sampling Aeschynomene plants in French Guiana and J. L. Contreras (Universidad Autónoma de Puebla, México) for help with sampling in Mexico. We also thank M. Zabaleta (University of Montevideo, Uruguay), L. G. Santos (CIAT, Colombia), A. Jorge (ILRI, Ethiopia) and S. Norton (AusPGRIS, Australia) for providing various seeds. We are extremely grateful to E. James for a remarkable re-reading and correction of the manuscript. We finally thank three anonymous reviewers for very helpful and interesting comments on an earlier version of this manuscript.