Bradyrhizobium diazoefficiens USDA110 Nodulation of Aeschynomene afraspera Is Associated with Atypical Terminal Bacteroid Differentiation and Suboptimal Symbiotic Efficiency

Legume-rhizobium symbiosis is a major ecological process in the nitrogen cycle, responsible for the main input of fixed nitrogen into the biosphere. The efficiency of this symbiosis relies on the coevolution of the partners.

ABSTRACT Legume plants can form root organs called nodules where they house intracellular symbiotic rhizobium bacteria. Within nodule cells, rhizobia differentiate into bacteroids, which fix nitrogen for the benefit of the plant. Depending on the combination of host plants and rhizobial strains, the output of rhizobium-legume interactions varies from nonfixing associations to symbioses that are highly beneficial for the plant. Bradyrhizobium diazoefficiens USDA110 was isolated as a soybean symbiont, but it can also establish a functional symbiotic interaction with Aeschynomene afraspera. In contrast to soybean, A. afraspera triggers terminal bacteroid differentiation, a process involving bacterial cell elongation, polyploidy, and increased membrane permeability, leading to a loss of bacterial viability while plants increase their symbiotic benefit. A combination of plant metabolomics, bacterial proteomics, and transcriptomics along with cytological analyses were used to study the physiology of USDA110 bacteroids in these two host plants. We show that USDA110 establishes a poorly efficient symbiosis with A. afraspera despite the full activation of the bacterial symbiotic program. We found molecular signatures of high levels of stress in A. afraspera bacteroids, whereas those of terminal bacteroid differentiation were only partially activated. Finally, we show that in A. afraspera, USDA110 bacteroids undergo atypical terminal differentiation hallmarked by the disconnection of the canonical features of this process. This study pinpoints how a rhizobium strain can adapt its physiology to a new host and cope with terminal differentiation when it did not coevolve with such a host. IMPORTANCE Legume-rhizobium symbiosis is a major ecological process in the nitrogen cycle, responsible for the main input of fixed nitrogen into the biosphere. The efficiency of this symbiosis relies on the coevolution of the partners. Some, but not all, legume plants optimize their return on investment in the symbiosis by imposing on their microsymbionts a terminal differentiation program that increases their symbiotic efficiency but imposes a high level of stress and drastically reduces their viability. We combined multi-omics with physiological analyses to show that the symbiotic couple formed by Bradyrhizobium diazoefficiens USDA110 and Aeschynomene afraspera, in which the host and symbiont did not evolve together, is functional but displays a low symbiotic efficiency associated with a disconnection of terminal bacteroid differentiation features. a phylogenetically distant host belonging to the dalbergioid clade that naturally interacts with photosynthetic rhizobia such as Bradyrhizobium sp. strain ORS285 (Fig. 1A to C) (14)(15)(16)(17)(18). To evaluate the efficiency of this symbiosis, the nitrogenase activity of USDA110-and ORS285-infected plants and their nitrogen content were determined. Although nitrogenase activity was detected in both types of nodules, it was significantly lower in USDA110-nodulated plants (Fig. 1D). A similar trend was observed for mass gain per nodule mass, although this difference is not significant (Fig. 1E). Nitrogen and carbon contents also seemed to be reduced in USDA110-nodulated plants compared to ORS285-nodulated plants, reaching levels similar to those of noninoculated plants (see Fig. S1 in the supplemental material). Accordingly, ORS285-nodulated A. afraspera plants displayed darker green leaves than those interacting with USDA110. Moreover, the plants inoculated with ORS285 are clearly much larger than the ones inoculated with USDA110 at later time points (21 days postinoculation [dpi]) (Fig. 1C), validating our conclusions drawn from the physiological analysis at 14 dpi that USDA110 is a poor symbiotic partner for A. afraspera, although the symbiosis is functional.
Moreover, the shoot/root mass ratio, a metric that reflects the nutritional status of the plant, was reduced in USDA110-nodulated A. afraspera plants compared to ORS285nodulated plants, indicating that the plant nutritional needs were not fulfilled ( Fig. S2) (19). To characterize further this suboptimal symbiosis, we analyzed the whole-nodule metabolome. Soybean nodules infected with USDA110 were used as a reference (Fig. S3). Allantoin, which is known to be the major nitrogen form exported from soybean nodules, was specifically detected in them (Fig. 1F) (20). On the contrary, asparagine and glutamine are the principal exported nitrogen compounds in A. afraspera nodules, and their amount was smaller in USDA110-infected nodules than in ORS285-infected nodules, indicating reduced nitrogen fixation by the bacteroids (Fig. 1F) (18).
In addition, we found specifically in USDA110-infected A. afraspera nodules the accumulation of sucrose, phosphoric acid, and ascorbate and, oppositely, a strong reduction in the trehalose content ( Fig. 1F; Fig. S3). Sucrose derived from phloem sap is metabolized in nodules to fuel the bacteroids with carbon substrates, usually dicarboxylates. The accumulation of sucrose in nodules indicates symbiotic dysfunction. Also, the accumulation of phosphoric acid in nodules suggests that nitrogen fixation is not reaching its optimal rate (18). Ascorbate has been shown to increase nitrogen fixation activity by modulating the redox status of leghemoglobin (21,22). Thus, its accumulation in nodules with a reduced nitrogen fixation capacity could be a stress response to rescue nitrogen fixation in nodules that do not fix nitrogen efficiently. A trehalose biosynthesis gene is upregulated in ORS285 bacteroids in A. afraspera, suggesting that TBD is accompanied by the synthesis of this osmoprotectant disaccharide (17). The lower synthesis in USDA110 bacteroids suggests an altered TBD. Together, these data indicate metabolic disorder in the USDA110-infected nodules, in agreement with USDA110 being a suboptimal symbiont of A. afraspera.
Overview of the USDA110 bacteroid proteomes and transcriptomes. In order to better understand the poor interaction between USDA110 and A. afraspera, bacteroid physiology was analyzed through transcriptome and proteome analyses. Efficient soybean bacteroids and free-living USDA110 cells cultivated in rich medium (exponential growth phase under aerobic conditions) were used as references ( Fig. 2A).
Prior to quantification of transcript abundances or identification and quantification of protein accumulation, the transcriptome data set was used to reannotate the USDA110 genome with the EugenePP pipeline (23). This allowed the definition of 876 new coding DNA sequences (CDSs), ranging from 92 to 1,091 bp (median size, 215 bp or 71.6 amino acids [aa]), with 11.5% of them having a predicted function or at least a match using InterProScan (IPR). This extends the total number of CDSs in the USDA110 genome to 9,171. Moreover, we also identified 246 noncoding RNAs (ncRNAs), ranging from 49 to 765 bp (median, 76 bp), which were not annotated previously. Proteomic evidence could be found for 28 new CDSs (3.2% of the new CDSs; median size, 97.6 aa). The complete reannotation of the genome is described in Table S1.
In the proteome data set, 1,808 USDA110 proteins were identified. Principal-component analysis (PCA) of all the replicate samples and sample types revealed their partitioning according to the tested conditions. The first axis of the PCA (40.2% of the observed variance) separated bacteroid profiles from the exponential-phase culture, whereas the second axis separated G. max bacteroids from A. afraspera bacteroids (14.9% of the observed variance) (Fig. 2B). The samples of the transcriptome data sets were similarly distributed on the PCA plot, with the first axis explaining 42.6% of the observed variance and the second axis explaining 23.5% of the observed variance (Fig. 2B).
Although differences were less pronounced in the proteome data set than in the transcriptome data set, Clusters of Orthologous Genes (COG) analysis showed similar profiles across functional categories, except for membrane proteins, which were less well identified by proteomics than by transcriptomics (Fig. S4). In the transcriptomic data set, 3,150 genes were differentially expressed under at least one condition (differentially expressed genes [DEGs]). Among the 1,808 proteins identified, 815 showed differential accumulation (differentially accumulated proteins [DAPs]), and 438 of the cognate genes were also differentially expressed in the transcriptome data sets, whereas 175 DEGs were not DAPs (Fig. 2C).
We analyzed the Pearson correlations between transcriptomic and proteomic profiles and found that ;66% of the bacterial functions that showed significant differences by both approaches displayed a high correlation coefficient (r . 0.9), whereas ,1% of the functions showed a strong negative correlation (r of less than 20.9) (Fig. 2D). This observation suggests that the transcriptome (which provides a more exhaustive view than the proteome) and the proteome provide a complementary picture of bacterial physiology, and they tend to show congruent information if we restrict our analysis to the genes with differential accumulation/expression (Fig. 2E). However, there were still ;66% of the DEGs, which were detected by the proteomic analysis, that were not DAPs. Our description of the bacterial functions is primarily based on the functions that were both DEGs and DAPs, as there is stronger evidence of their modulation under the tested conditions. The transcriptome alone is used only when proteomics is not informative, for example, to analyze regulons and stimulons that have been described previously in USDA110.
Symbiotic functions common to both types of USDA110 bacteroids. Among the 815 DAPs, 705 and 699 proteins were significantly differentially accumulated in G. max and A. afraspera, respectively, compared to the bacterial culture control. Strikingly, 646 proteins were commonly differentially accumulated in both plant nodules (Table S1).
In the transcriptomic data set, 1,999 DEGs, representing ;21% of the genome, were identified between the bacterial culture and the bacteroids, regardless of the host. Among them, 1,076 genes displayed higher expression levels in nodules (including 7 newly annotated ncRNAs and 1 newly annotated CDS among the 20 differentially expressed genes with the highest fold changes), and 923 genes were repressed in planta (including 2 newly annotated ncRNAs and 2 newly annotated CDSs among the 20 DEGs with the highest fold changes) (Table S1).
Restricting the analysis to the bacterial functions that were both differentially expressed (DEGs) and differentially accumulated (DAPs) in planta in both hosts compared to the bacterial culture led to the identification of 222 genes/proteins, with 150 being upregulated and 72 being repressed in planta, respectively (Fig. 3A). Notably, six newly annotated genes were in this gene list, including one putative regulator (Bd110_01119) that was induced during symbiosis. Among the functions common to DEGs and DAPs in planta, only four functions showed opposite trends by proteomics and transcriptomics.
The proteome and transcriptome data provided a coherent view of the nitrogen fixation metabolism of B. diazoefficiens under the tested conditions. Key enzymes involved in microoxic respiration and nitrogen fixation were detected among the proteins having the highest spectral numbers in the nodule samples ( Fig. 3A; Table S1), and the corresponding genes were among the most strongly expressed ones in bacteroids while being almost undetectable under free-living conditions. This includes, for instance, the nitrogenase and the nitrogenase reductase subunits, which constitute the nitrogenase enzyme complex responsible for nitrogen conversion into ammonia. They belong to a locus of 21 genes from blr1743 (nifD) to bll1778 (putative ahpC), including the genes involved in nitrogenase cofactor biosynthesis, electron transport to nitrogenase, and microaerobic respiration, that were among the most highly expressed ones in bacteroids of both host plants, at both the gene and protein expression levels. The slightly higher level of the dinitrogenase reductase NifH detected by proteomics was not supported by Western blot analysis, which showed apparently similar protein levels under both bacteroid conditions (Fig. S5). Strikingly, the two bacteroid types did not show a notable difference in the expression of these genes and proteins, suggesting that the activation of the nitrogen fixation machinery is not a limiting factor underlying the suboptimal efficiency of strain USDA110 in A. afraspera nodules.
In addition to these expected bacteroid functions, many other proteins were identified that specifically and strongly accumulated in both nodule types. This is the case for the chaperonins GroS3 and GroL3, which were strongly upregulated and reached high gene expression and protein levels in both bacteroids. The upregulation of these chaperonins is remarkable because other GroEL/GroES proteins (GroS2, GroES, GroL1, GroL6, and GroL7) were also very strongly accumulated in a constitutive manner. This indicates that bacteroids have a high demand for protein folding, possibly requiring specific GroEL isoforms, a situation reminiscent of the requirement of one out of five GroEL isoforms for symbiosis in Sinorhizobium meliloti, the symbiont of Medicago sativa (12,24). Another example of a bacteroid-specific function is the hydrogenase uptake system, whose gene expression was induced in both bacteroid types from nearly no expression in culture. The hydrogenase subunit HupL (bll6941) was found among the proteins displaying the highest spectral number in the nodule samples, suggesting important electron recycling in bacteroids of the two hosts. Another one is the 1-aminocyclopropane-1-carboxylic acid (ACC) deaminase (blr0241), which was also among the most strongly accumulated proteins in nodules and was significantly less abundant in free-living USDA110. An outer membrane protein (bll1872) belonging to the NifA regulon (25) was also strongly induced in planta, with a transcript level among the top 10 genes in A. afraspera. Additionally, a locus of seven genes (blr7916 to blr7922 [blr7916-blr7922]) encoding an amidase enzyme and a putative peptide transporter composed of two transmembrane domain proteins, two ATPases, and two solute binding proteins was strongly upregulated in the two bacteroid types, with three proteins also being overaccumulated in planta ( Fig. 3A; Table S1).
Taken together, these data show that both bacteroid types displayed a typical nitrogen fixation-oriented metabolism, with a partial shutdown of housekeeping functions. This indicates that despite the apparent reduced symbiotic efficiency of USDA110 in A. afraspera nodules, the bacterium fully expressed its symbiotic program within this nonnative host as it does in soybean, its original host. Thus, the suboptimal functioning of the A. afraspera nodules did not seem to come from a bacterial defect to express the symbiotic program but possibly came from an unfavorable host microenvironment or a lack of metabolic integration of these maladapted partners.
Host-specific functions. Comparison of the A. afraspera and G. max bacteroids also revealed significant differences in the proteomes and transcriptomes. At the transcriptomic level, 935 DEGs could be identified between the two bacteroid types (509 A. afraspera . G. max and 426 G. max . A. afraspera). One notable feature of the transcriptome is the identification of 4 newly annotated ncRNAs and 1 new CDS among the 20 most induced DEGs in A. afraspera nodules and the presence of 5 newly annotated CDSs among the 20 most induced DEGs in G. max nodules (Table S1). However, when considering only the functions that displayed congruent and significant differences in terms of transcript and protein levels between plant hosts, these fell to 63 genes/proteins, with 33 being induced in A. afraspera nodules and 30 being induced in G. max nodules (Fig. 3B).
Interestingly, the phenylacetic acid degradation pathway (PaaABCDEIK [blr2891-blr2897]) was highly expressed in A. afraspera nodules (although only PaaABCD and PaaK have been detected by proteomics), as was an as-yet-uncharacterized cluster of genes putatively involved in toluene degradation (blr3675-blr3680). The chaperone GroEL2 was also specifically induced in A. afraspera. Similarly, three S1 peptidases (Dop [blr2591, blr3130, and blr7274]) were highly expressed in the nodules of the latter host together with an RND efflux pump (bll3903) and an LTXXQ motif protein (bll6433), a motif also found in the periplasmic stress response protein CpxP (26). The overaccumulation of these proteins suggests that bacteroids are facing stressful conditions during this interaction with A. afraspera. An uncharacterized ABC transporter solute binding protein (blr7922) was also overexpressed in A. afraspera.
USDA110 transcriptomics data in the perspective of previously described regulons and stimulons. USDA110 is one of the best-characterized rhizobial strains in terms of transcriptomic responses to various stimuli as well as the definition of regulons (27). We analyzed the behavior of these previously defined gene networks in USDA110 in our data set (Table S2). To initiate the molecular dialog that leads to nodule formation, plants secrete flavonoids like genistein in their root exudates, which are perceived by the rhizobia and trigger Nod factor production. At 14 dpi, when the nodule was formed and functioning, the genistein stimulon, which comprises the NodD1, NodVW, TtsI, and LafR regulons, was not activated anymore in bacteroids. The symbiotic regulons controlled by NifA, FixK1, FixK2, FixLJ, and sigma 54 (RpoN) were activated in planta, indicating that nitrogen fixation was taking place in both hosts. Accordingly, the nitrogen metabolism genes controlled by NtrC were activated in planta. Additionally, the PhyR/EcfG regulon involved in the general stress response (GSR) was not activated in bacteroids. However, differences between hosts were not observed for any of these regulons/stimulons. The only stimulon that showed differential expression between hosts is the one involved in aromatic compound degradation, which was highly expressed in A. afraspera nodules. A similar upregulation of the vanillate degradation pathway was observed in the transcriptome of Bradyrhizobium sp. ORS285 in A. afraspera and Aeschynomene indica nodules (17), suggesting that dalbergioid hosts display a higher aromatic compound content in nodules than G. max. In line with this hypothesis, some of the most differentially accumulated sets of proteins (A. afraspera . G. max) are involved in the degradation of phenylacetic acid (PaaABCDK and bll0339), suggesting that the bacterium converts phenylalanine (or other aromatic compounds) ultimately to fumarate through this route (Fig. 3B) (28). Similarly, enzymes of another pathway involved in phenolic compound degradation (blr3675-blr3680) were accumulated in A. afraspera nodules ( Fig. 3B; Table S1).
Expression pattern of orthologous genes between ORS285 and USDA110 in A. afraspera nodules. In a previous study (17), a transcriptome analysis was performed on Bradyrhizobium sp. ORS285 in interaction with A. afraspera and in culture. Bradyrhizobium sp. ORS285 is a strain that coevolved with A. afraspera, leading to an efficient symbiosis hallmarked by TBD, i.e., cell elongation and polyploidization of the bacteroids. In order to compare the gene expressions of these two nodule-forming rhizobia in culture and in planta, we determined the set of orthologous genes between the two strains using the Phyloprofile tool of the MicroScope-MAGE website. This analysis yielded a total of 3,725 genes (Table S3). The heat map in Fig. 4A presents the modulation of gene expression (log 2 fold change [LFC]) between A. afraspera nodules and the bacterial culture for the orthologous genes in each bacterium, regardless of their statistical significance. When taking a false discovery rate (FDR) of ,0.01 into account, we identified sets of genes that were differentially expressed in planta in either bacterium or in both bacteria (Fig. 4B).
Only 343 genes displayed differential expression (FDR of ,0.01 and jLFCj [absolute LFC] of .1.58) in planta in both bacteria compared to their respective culture controls (Fig. 4C). A majority of these genes (86.8%) exhibited congruent expression patterns. First, the nif, fix, and hup genes were commonly and highly induced in both strains during their symbiotic life with A. afraspera, a hallmark of a functional symbiosis. However, there were differences in their expression levels, with higher expression levels of the symbiotic genes in ORS285 (nifHDK represented 12.5% of all reads in A. afraspera nodules) (17) than in USDA110 (nifHDK represented only 2.5% of all reads in A. afraspera nodules), consistent with a more efficient interaction occurring between ORS285 and A. afraspera. Additionally, the Kdp high-affinity transport system, phosphate metabolism (pstCAB, phoU, phoE, and phoC), and phosphonate metabolism (phnHIJKL) were activated in planta in both bacteria ( Fig. 4B and C). The stress marker dop protease gene was also induced in both bacteria in A. afraspera nodules (Fig. 4C).
Additionally, 1,026 genes were differentially expressed solely in ORS285, and similarly, there were 604 DEGs specific to USDA110 (Fig. 4B). For example, the general secretory pathway seemed to be specifically induced in ORS285 (17). Oppositely, USDA110 displayed an induction of the rhcJQRU genes, which are involved in the injection of type 3 effector proteins that can be important for the establishment of the symbiosis, whereas they were not induced or even repressed in ORS285 (Fig. 4B). This was also the case for the nitrite reductase-encoding gene nirK (blr7089 [BRAD285_v2_0763]) (Fig. 4C). In addition, USDA110 induced the expression of an ACC deaminase (blr0241), while its ortholog was repressed in ORS285 (BRAD285_v2_3570) during symbiosis (Fig. 4C). Bacterial ACC deaminases can degrade ACC, a precursor of ethylene, and thereby modulate ethylene levels in the plant host and promote the nodulation process (29).
Bradyrhizobium diazoefficiens USDA110 bacteroids undergo bona fide TBD in Aeschynomene afraspera nodules despite very weak morphological and ploidy modifications. In a previous description of the A. afraspera-B. diazoefficiens USDA110 interaction, the typical TBD features were not observed, and the bacteroids were very similar to those in G. max, where no TBD occurs (16). At the molecular level, the accumulation of the replication initiation factor DnaA was higher in soybean than in A. afraspera (Table S1). Similarly, the MurA peptidoglycan synthesis enzyme (encoded by bll0822) that may play a role in cell elongation during TBD was detected at similar levels in both bacteroids (Table S1). Taken together, the molecular data did not clearly indicate whether USDA110 bacteroids undergo TBD in A. afraspera. Therefore, we investigated the features of the USDA110 bacteroids in A. afraspera nodules in more detail.
We analyzed bacteroid differentiation features in USDA110 bacteroids extracted from soybean and A. afraspera nodules. The interaction between A. afraspera and Bradyrhizobium sp. ORS285 was used as a positive control for TBD features (9,30,31). TBD is characterized by cell elongation. We quantified the cell length, width, area, and shape of purified bacteroids and culture controls. Whereas ORS285 bacteroids were enlarged within A. afraspera nodules compared to their free-living counterparts, USDA110 bacteroids were similar to those of free-living bacteria in both soybean and A. afraspera ( Fig. 5A; Fig. S6). Another feature of TBD is endoreduplication. Analysis of the bacterial DNA content of ORS285 bacteroids in A. afraspera by flow cytometry shows peaks at 6C and higher (9). As expected, USDA110 bacteroids in G. max yielded only two peaks, at 1C and 2C, similarly to the cycling cells in the bacterial culture sample (Fig. 5B) (16). Strikingly, similar results were obtained for USDA110 in A. afraspera. Thus, with respect to the DNA content and cell size, the USDA110 bacteroids did not display the typical TBD features in A. afraspera nodules. Loss of membrane integrity is a third hallmark of TBD that likely strongly contributes to the loss of viability of bacteroids. Time course analyses of propidium iodide (PI) uptake by bacteroids and the corresponding culture controls were performed to assess bacteroid permeability (Fig. S7). Twenty minutes after PI application, USDA110 bacteroids from A. afraspera displayed an increased permeability that was much closer to that of ORS285 bacteroids in interaction with A. afraspera than to the low permeability of USDA110 bacteroids from G. max nodules (Fig. 5C). Also, the free-living counterparts exhibited very low permeability. Taken together, these results suggest that the envelope of USDA110 bacteroids is more permeable in the NCR-producing A. afraspera nodules, even if it does not reach the permeability level of the ORS285 strain. To analyze bacterial viability, bacteroids extracted from nodules were plated, and the CFU were determined (Fig. 5D). In G. max, USDA110 formed 1.46 Â 10 10 colonies/mg nodule (;100% survival). Oppositely, ORS285 formed only 5.42 Â 10 7 colonies/mg nodule in A. afraspera (;0.5% survival). Interestingly, USDA110 formed 1.13 Â 10 8 colonies/mg nodule in A. afraspera (;1% survival), indicating that despite the absence of cell enlargement and endoreduplication, USDA110 bacteroids lose their viability and undergo bona fide terminal differentiation in A. afraspera. Thus, in the NCR-producing plant A. afraspera, USDA110 bacteroids displayed a disconnection of the four canonical TBD features (i.e., cell size, ploidy level, membrane permeability, and cell viability).

DISCUSSION
Aeschynomene afraspera triggers atypical but terminal differentiation of USDA110 bacteroids. In a previous study, we noticed that in A. afraspera, USDA110 forms a functional symbiosis, although bacteroids do not display features that are usually associated with TBD (16). Here, we show that no endoreduplication and cell elongation of USDA110 occur in terminally differentiated bacteroids that fix nitrogen in a suboptimal way. Accordingly, the protein level of DnaA, the genome replication initiator, was higher in soybean than in A. afraspera bacteroids, and the MurA level was not different between bacteroid conditions, confirming that polyploidization and cell elongation did not occur in this host. Such unusual terminal bacteroid differentiation is reminiscent of the bacteroids in Glycyrrhiza uralensis. This plant of the IRLC expresses NCR peptides (11). However, one of its compatible symbionts, Sinorhizobium fredii strain HH103, does not undergo any loss of viability and shows no change in DNA content and no cell elongation (32), while another symbiont, Mesorhizobium tianshanense strain HAMBI 3372, showed all TBD features (33). The influence of the bacterial genotype on the terminal/nonterminal differentiation of bacteroids was also suggested in Medicago truncatula, in which the gene hrrP might confer to some Sinorhizobium strains resistance against the differentiation process triggered by some M. truncatula ecotypes (34). In these two IRLC plants (i.e., M. truncatula and G. uralensis), bacteria undergo complete TBD or no TBD at all in a strain-dependent manner, but there is no clear uncoupling of the features of TBD (cell elongation/endoreduplication/altered viability) as shown here in the case of B. diazoefficiens USDA110-A. afraspera.
The surprising differentiation of USDA110 in A. afraspera nodules raises questions about the molecular mechanisms supporting this phenomenon. We consider two possible hypotheses. First, strain USDA110 might be more sensitive to the differentiation factors of the host than strain ORS285 and be rapidly "terminally" differentiated, before the other differentiation features that are potentially important for symbiotic efficiency can take place. Alternatively, USDA110 might be resistant to the plant effectors that trigger the elongation and polyploidization features.
In agreement with the latter possibility, the application of NCR peptides has a very limited effect on strain USDA110 compared to S. meliloti and other plant-associated bacteria (16,35). NCR insensitivity may be due to the thick hopanoid layer that is present in the outer membrane of strain USDA110, as the hopanoid biosynthesis mutant hpnH is more sensitive to NCR peptides and shows symbiotic defects in A. afraspera but not in G. max (36). Moreover, the altered peptidoglycan structure in the strain USDA110 DD-carboxypeptidase mutant resulted in an increased TBD process with endoreduplicated and elongated bacteroids in A. afraspera (16). This suggests that the envelope of strain USDA110 prevents canonical TBD from occurring. Possibly, NCR peptides are not able to reach their intracellular targets required to induce endoreduplication and cell division arrest, while their effect on cell viability through pore formation and membrane destabilization is still effective.
A survey of TBD in the legumes identified multiple occurrences of the process in several subclades of the legumes but found that the majority of legumes do not have TBD (37). The classification in this study was based on a morphological analysis of the bacteroids. Ancestral state reconstruction based on this classification suggested that the nondifferentiated bacteroids are ancestral and that TBD evolved at least five times independently in legumes (37). The discovery of bacteroids that are terminally differentiated without any obvious morphological changes opens the possibility that the occurrence of TBD might be underestimated in the legume family. Similarly, in the IRLC, the extent of morphological bacteroid differentiation was correlated with the size of the cationic NCR peptide repertoire, and in legumes with few NCR peptides, the morphological modification of bacteroids can be minor (11,33). In addition, at the molecular level, TBD was originally ascribed to the production of symbiotic antimicrobial peptides, the NCRs, by nodules (7), but more recently, other types of antimicrobial peptides such as the NCR-like, GRP, MBP1, and CAPE peptides specifically produced in nodules of different plants were proposed to contribute to bacteroid differentiation (9,(38)(39)(40). Thus, if TBD is indeed more widespread than currently estimated on the basis of morphological bacteroid features, the currently proposed evolutionary scenario of bacteroid formation might require revision.
Terminal differentiation is associated with specific stress responses. The TBD of strain USDA110 in A. afraspera is associated with a high accumulation of stress markers compared to the G. max bacteroids. These markers include four proteases (Dop, Lon-blr6174, blr3130, and blr7274) and one chaperonin (GroL4). Similar inductions of proteases and chaperonins have been reported in NCR-treated S. meliloti cultures (35), indicating that this response may be linked to the perception of A. afraspera NCR-like peptides in USDA110.
The genes encoding these stress-related proteins are not part of the well-characterized general stress response (GSR) controlled by the PhyR/EcfG signaling cascade in B. diazoefficiens USDA110 (41). On the other hand, we found that the PhyR/EcfG regulon in USDA110 is not activated in the bacteroids of both host plants (see Table S2 in the supplemental material). This observation contrasts with our previous study of the Bradyrhizobium sp. ORS285 transcriptome during symbiosis with Aeschynomene plants, which showed that the PhyR/EcfG cascade was upregulated in planta (17). Nevertheless, the expression of the Dop protease was induced in A. afraspera in both bacteria (Fig. 4C). Together, the omics data suggest that bacteroids of Bradyrhizobium spp. activate stressrelated genes in the TBD-inducing A. afraspera host but that differences exist in the activation of specific stress responses at the strain level.
Correlation between bacteroid differentiation features and symbiotic efficiency for the plant. TBD is associated with the massive production of symbiotic antimicrobial peptides such as NCR, NCR-like, and CAPE peptides in different plants (5,9,38,40). They represent ;10% of the nodule transcriptomes in M. truncatula (analysis of the data from reference 42), and their production is thus potentially a strong energetic cost for the plant, raising questions about the benefits of the TBD process. TBD appeared independently in different legume clades (9,37), suggesting that plants imposing this process obtain an advantage that might be a higher symbiotic benefit. Increased symbiotic efficiency has indeed been observed in hosts imposing TBD (17,43,44). The findings reported here, comparing bacteroids and symbiotic efficiencies in A. afraspera infected with strain ORS285 and strain USDA110, are in agreement with this hypothesis. Also, in the symbiosis of M. truncatula in interaction with different S. meliloti strains, a similar correlation was observed between the level of bacteroid differentiation and plant growth stimulation (45). However, the simultaneous analysis of the bacteroid differentiation and symbiotic performances of an extended set of Aeschynomene-Bradyrhizobium interactions has shown that, perhaps not unexpectedly, the symbiotic efficiency of the plant-bacterium couple is not correlated solely with bacteroid differentiation and that other factors can interfere with the symbiotic efficiency as well (46).
Conclusion. Bradyrhizobium diazoefficiens USDA110 is a major model of legume-rhizobium symbiosis, mainly thanks to its interaction with G. max, the most cultivated legume worldwide. Although omic studies have been conducted in this strain in symbiosis with various hosts (13,25), this is the first time that this bacterium has been studied at the molecular level in symbiosis with an NCR-producing plant that normally triggers typical terminal bacteroid differentiation in its symbionts. The symbiosis between USDA110 and A. afraspera is functional even if nitrogen fixation and plant benefits are suboptimal.
Terminal bacteroid differentiation is taking place in the NCR-producing host A. afraspera, as bacterial viability is impaired in USDA110 bacteroids, whereas morphological changes and the cell cycle switch to endoreduplication are not observed. We also show by combining proteomics and transcriptomics that the bacterial symbiotic program is expressed in A. afraspera nodules in a way similar to that in G. max, although host-specific patterns were also identified. However, the bacterium is under stressful conditions in the A. afraspera host, possibly due to the production of NCR-like peptides in this plant. The integration of data sets from different bacteria in symbiosis with a single host, like ORS285 and USDA110 in symbiosis with A. afraspera, shed light on the differences in the stress responses activated in A. afraspera and confirmed that the symbiosis is functional but suboptimal in this interaction. The molecular data presented here provide a set of candidate functions that could be analyzed for their involvement in adaptation to a new host and to the TBD process.
G. max ecotype Williams 82 and A. afraspera seeds were surface sterilized, and the plants were cultivated and infected with rhizobia for nodule formation as described previously by Barrière et al. (16). Nodules were collected at 14 days postinoculation (dpi), immediately immersed in liquid nitrogen, and stored at 280°C until use. Each tested condition (in culture and in planta) was produced in biological triplicates.
Phylogeny analysis. Nucleotide sequences of matK genes were collected from the NCBI database using accession numbers described previously (50,51) and analyzed on by Phylogeny.fr (www .phylogeny.fr). They were aligned using ClustalW with manual corrections, before running a phyML (GTR-gamma model) analysis with 500 bootstraps. A Bayesian inference tree was also generated (GTR1G1I) and provided a topology similar to the one for the maximum likelihood (ML) tree (data not shown). Trees were visualized and customized using TreeDyn.
Genome annotation and RNA-seq analysis. Nodule and bacterial culture total RNA was extracted and treated as previously described (17). Oriented (strand-specific) libraries were produced using the SOLiD total RNA-seq kit (Life Technologies) and sequenced on a SOLiD 3 station, yielding ;40 million 50-bp single reads. Trimming and normalization of the reads were performed using CLC workbench software. Subsequently, the reads were used to annotate the genome using EuGenePP (23), and mapping was performed using this new genome annotation. Analysis of the transcriptome using DESeq2 and data representation were performed as previously described (17). Differentially expressed genes (DEGs) showed an absolute log 2 fold change (jLFCj) of .1.58 (i.e., fold change of .3) with a false discovery rate (FDR) of ,0.01.
Proteomic analysis. Bacteroids were extracted from 14-dpi frozen nodules (6), while bacterial culture samples were collected as described above, and the bacterial pellets were resuspended in 220°C acetone and lysed by sonication. Protein solubilization, dosage, digestion (2% [wt/wt] trypsin), and solidphase extraction (using a Phenomenex polymeric C 18 column) were performed as described previously (52). Peptides from 800 ng of proteins were analyzed by LC-MS/MS with a Q Exactive mass spectrometer (Thermo Electron) coupled to a nanoLC Ultra 2D instrument (Eksigent) using a nanoelectrospray interface (noncoated capillary probe, 10-mm internal diameter [ID]; New Objective). Peptides were loaded on a Biosphere C 18 trap column (particle size of 5 mm, pore size of 12 nm, inner/outer diameters of 360/ 100 mm, and length of 20 mm; NanoSeparations) and rinsed for 3 min at 7.5 ml/min of 2% acetonitrile (ACN)-0.1% formic acid (FA) in water. Peptides were then separated on a Biosphere C 18 column (particle size of 3 mm, pore size of 12 nm, inner/outer diameters of 360/75 mm, and length of 300 mm; NanoSeparations) with a linear gradient from 5% buffer B (0.1% FA in ACN) and 95% buffer A (0.1% FA in water) to 35% buffer B and 65% buffer A in 80 min at 300 nl/min, followed by a rinsing step at 95% buffer B and 5% buffer A for 6 min and a regeneration step with parameters from the start of the gradient for 8 min. Peptide ions were analyzed using Xcalibur 2.1 software in the data-dependent mode with the following parameters: (i) a full MS scan was acquired for the m/z 400 to 1,400 range at a resolution of 70,000 with an automatic gain control (AGC) target of 3 Â 10 6 , and (ii) an MS 2 scan was acquired at a resolution of 17,500 with an AGC target of 5 Â 10 4 , a maximum injection time of 120 ms, and an isolation window of 3 m/z. The normalized collision energy was set to 27. The MS 2 scan was performed for the eight most intense ions in the previous full MS scan with an intensity threshold of 1 Â 10 3 and a charge of between 2 and 4. Dynamic exclusion was set to 50 s. After conversion to mzXML format using msconvert (3.0.3706) (53), data were searched using X!Tandem (version 2015.04.01.1) (54) against the USDA110 reannotated protein database and a homemade database containing current contaminants. In a first pass, trypsin was set to strict mode; cysteine carbamidomethylation was set as a fixed modification; and methionine oxidation, protein N-terminal acetylation with or without protein N-terminal methionine excision, N-terminal glutamine and carbamidomethylated cysteine deamidation, and N-terminal glutamic dehydration were set as potential modifications. In a refine pass, semienzymatic peptides were allowed. Protein inference was performed using X!TandemPipeline (version 3.4.3) (55). A protein was validated with an E value of ,10 25 and 2 different peptides with an E value of ,0.05. Proteins from the contaminant database (Glycine max proteins and unpublished Aeschynomene expressed sequence tags) were removed after inference. Proteins were quantified using the spectral counting method (56). To discriminate differentially accumulated proteins (DAPs), analysis of variance (ANOVA) was performed on the spectral counts, and proteins were considered a DAP when the P value was ,0.05.
Metabolomic analysis. Metabolites and cofactors were extracted from lyophilized nodules and analyzed by gas chromatography-mass spectrometry (GC-MS) and LC-MS, respectively, according to methods described previously by Su et al. (57) and Guérard et al. (58).
Plant biomass and nitrogen fixation analysis. Dry masses of the shoot, root, and nodules were measured, and the shoot/root mass ratio was calculated. The mass gain per gram of dry nodule was calculated as the difference between the total mean masses of the plants of interest and those of the noninoculated plants, divided by the mean mass of nodules. Thirty plants were used under each condition. Nitrogenase activity was assessed by an acetylene reduction assay (ARA) on 10 plants under each condition as previously described (31). The elemental analysis of leaf carbon and nitrogen contents was performed as described previously (18).
Analysis of B. diazoefficiens USDA110 regulons and stimulons. Gene sets defined as regulons and stimulons were collected from the literature, and the regulons/stimulons were considered activated/ repressed when $40% of the corresponding genes were DEGs in a host plant compared to the culture conditions.
Comparison of orthologous gene expression between B. diazoefficiens USDA110 and Bradyrhizobium sp. ORS285. The list of orthologous genes between USDA110 and ORS285 was determined using the Phyloprofile tool of the MicroScope-MAGE platform (59), with an identity threshold of 60%, a maxLrap of .0, and a minLrap of .0.8. The RNA-seq data from a previous study (17) and those of this study were used to produce heat maps for the genes displaying an FDR of ,0.01 (A. afraspera versus YM medium) using R (v3.6.3) and drawn using pheatmap (v1.0.12) coupled with kohonen (v3.0.10) for gene clustering using the self-organizing maps (SOM) method. The DEGs in both organisms (A. afraspera versus YM medium) were plotted for USDA110 and ORS285.
Analysis of TBD features. Bacteroids were extracted from 14-dpi nodules and analyzed using a CytoFLEX S instrument (Beckman-Coulter) (31). For ploidy and live/dead analyses, samples were stained with propidium iodide (PI) (Thermo Fisher) (50-mg · ml 21 final concentration) and Syto9 (Thermo Fisher) (1.67 mM final concentration). PI permeability was assessed over time on live bacteria. Bradyrhizobium sp. ORS285.pMG103-nptII-GFP (30) and B. diazoefficiens USDA110 sYFP2-1 (60) strains were used to distinguish bacteroids from debris during flow cytometry analysis. For each time point, the suspension was diluted 50 times for measurement in the flow cytometer. The percentage of bacteroids permeable to PI was estimated as the ratio of PI-positive over total bacteroids (green fluorescent protein [GFP]/yellow fluorescent protein [YFP] positive). Heat-killed bacteroids were used as a positive control to identify the PI-stained bacteroid population.
For bacteroid viability assays, nodules were collected and surface sterilized (1 min of 0.4% NaClO, 1 min of 70% ethanol, and two washes in sterile water). Bacteroids were subsequently prepared as previously described (31) and serially diluted and plated (5 ml per spot) in triplicate on YM medium containing 50mg · ml 21 carbenicillin. CFU were counted 5 days after plating and divided by the total nodule mass.
Bacterial cell shape, length, and width were determined using confocal microscopy image analysis. Bacteroid extracts and stationary-phase bacterial cultures were stained with 2.5 nM Syto9 for 10 min at 37°C and mounted between a slide and a coverslip. Bacterial imaging was performed on an SP8 confocal laser scanning microscope (Leica Microsystems) equipped with hybrid detectors and a 63Â oil immersion objective (Plan Apo, 1.4 numerical aperture [NA]; Leica). Under each condition, multiple z-stacks (2.7-mm width and 0.7-mm step) were automatically acquired (excitation at 488 nm and collection of fluorescence at 520 to 580 nm).
Prior to image processing, each stack was transformed as a maximum-intensity projection using ImageJ software (https://imagej.nih.gov/ij/). Bacterial detection was performed with MicrobeJ (https:// www.microbej.com/) (61). First, bacteria were automatically detected on every image using an intensitybased thresholding method with a combination of morphological filters (area, 1 to 20 mm 2 ; length, 1 mm to 1; width, 0.5 to 1.3 mm), and every object was fitted with a "rod-shaped" bacterial model. To ensure high data quality, every image was manually checked to remove false-positive results (mainly plant residues) and include rejected objects (mainly fused bacteria). Next, the morphology measurements and figures were directly extracted from MicrobeJ. ORS285 cells in culture and in symbiosis with A. afraspera were used as references for the analysis of TBD features.
Data availability. Genome annotation of Bradyrhizobium diazoefficiens USDA110 using EuGenePP is available at http://doi.org/10.25794/reference/d56qaddg. Transcriptome data have been deposited in the NCBI Gene Expression Omnibus (GEO) and are accessible through GEO series accession number GSE163004. Proteome data are available at http://moulon.inra.fr/protic/usda110_bacteroid_differentiation (login/passwords will be provided upon request).

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.