Metabolomics and Transcriptomics Identify Multiple Downstream Targets of Paraburkholderia phymatum σ54 During Symbiosis with Phaseolus vulgaris

RpoN (or σ54) is the key sigma factor for the regulation of transcription of nitrogen fixation genes in diazotrophic bacteria, which include α- and β-rhizobia. Our previous studies showed that an rpoN mutant of the β-rhizobial strain Paraburkholderia phymatum STM815T formed root nodules on Phaseolus vulgaris cv. Negro jamapa, which were unable to reduce atmospheric nitrogen into ammonia. In an effort to further characterize the RpoN regulon of P. phymatum, transcriptomics was combined with a powerful metabolomics approach. The metabolome of P. vulgaris root nodules infected by a P. phymatum rpoN Fix− mutant revealed statistically significant metabolic changes compared to wild-type Fix+ nodules, including reduced amounts of chorismate and elevated levels of flavonoids. A transcriptome analysis on Fix− and Fix+ nodules—combined with a search for RpoN binding sequences in promoter regions of regulated genes—confirmed the expected control of σ54 on nitrogen fixation genes in nodules. The transcriptomic data also allowed us to identify additional target genes, whose differential expression was able to explain the observed metabolite changes in numerous cases. Moreover, the genes encoding the two-component regulatory system NtrBC were downregulated in root nodules induced by the rpoN mutant, and contained a putative RpoN binding motif in their promoter region, suggesting direct regulation. The construction and characterization of an ntrB mutant strain revealed impaired nitrogen assimilation in free-living conditions, as well as a noticeable symbiotic phenotype, as fewer but heavier nodules were formed on P. vulgaris roots.


Introduction
Paraburkholderia phymatum STM815 T is a nitrogen-fixing soil bacterium that belongs to the β-proteobacteria (β-rhizobia); it is able to induce nodules on the roots of Mimosoid and Papilionoid legumes [1]. While it had first been reported as a symbiont of the Mimosa spp. native to South America [2,3], it was later demonstrated to be a highly promiscuous nodulator of several Mimosoid genera [4,5], as well as some Papilionoid legumes [6,7].
Nitrogen is an essential component of all amino acids and nucleic acids. It is therefore considered to be a key limiting factor for plant growth and development. The ability of legumes to form a symbiotic association with nitrogen-fixing rhizobia gives the legumes a clear advantage over other

Metabolomic Analysis of P. vulgaris Root Nodules Infected with P. phymatum STM815 T Wild-Type and with a rpoN Mutant
We have previously shown that the alternative sigma factor RpoN (σ 54 ) of P. phymatum STM815 T is required for the development of an effective symbiotic interaction between this β-rhizobial strain and the legume P. vulgaris [23]. Although the rpoN mutant formed a similar number and size of nodules on P. vulgaris compared to the wild type, bacteroids were unable to convert atmospheric dinitrogen to ammonium. Furthermore, the nitrogen content of the shoot was significantly reduced compared to plants infected by the wild type. To gain a better understanding of the molecules and metabolic pathways relevant during the P. vulgaris-P. phymatum symbiosis, we carried out two experiments: we first compared the metabolite levels measured in P. vulgaris nodules infected with P. phymatum wild-type with levels in nodules induced by a rpoN mutant strain 21 days post-infection (dpi). In addition, in order to establish a metabolic baseline, we compared the metabolite levels in Fix + nodules with those from uninfected root material. Three independent biological replicates were prepared, and aliquots from each replicate were injected twice (as technical replicates) in a 6550 iFunnel Q-TOF mass spectrometer (see Section 3. Materials and Methods) [33]. After discarding unassigned metabolites, low-abundant signals and adducts, a total of 409 ions, corresponding to known deprotonated metabolites from bacterial or plant origin, were identified. A principal component analysis (PCA) separated the different biological samples, according to the presence or absence of the symbiont (in the first dimension) and to the genetic background of the two strains (wt versus rpoN mutant, in the second dimension) (Figure 1).
To discover general RpoN-dependent metabolic trends, a metabolite set enrichment analysis for both significantly decreased and increased metabolites was performed, which identified several affected metabolic pathways (Table 1).  A comparison of ion intensities revealed that 147 (36%) of all detected and assigned ions were differentially accumulated in nodules induced by the wild-type strain (wt nod) versus those by the rpoN mutant (rpoN mt nod) (increase/decrease of more than 1.4-fold change is equivalent to a log 2 (FC) of 0.5, and q-value < 0.01; see Section 3. Materials and Methods,) ( Table 2). Among these, 70 metabolite ions showed a significant decrease in nodules infected by the rpoN mutant, while 77 had significantly increased intensities. One example of an over-represented category is the flavonoid/isoflavonoid biosynthesis pathway, which covered 13 of the 77 compounds that were more abundant in Fix − nodules (Table 2). Flavonoids are phenolic compounds, which are secreted by plants in order to attract rhizobia to the root. In response to flavonoids, rhizobia activate nod (nodulation) gene expression, leading to the synthesis of lipochitooligosaccharides, also known as Nod factors. Besides this well-studied role, flavonoids have multiple functions, including a role as an antioxidant [34] and as a defense mechanism against pathogens [35]. The rpoN Fix − mutant may therefore be recognized as a cheater by P. vulgaris, which accumulates flavonoids such as naringenin to defend itself. Naringenin has been detected in root exudates of a 9-day-old P. vulgaris seedling, and has been shown to induce expression of nod genes in Rhizobium leguminosarum [36]. A similar stress situation was observed in soybean nodules induced by the Bradyrhizobium diazoefficiens Fix − regulatory mutant nifA, where a phytoalexin was accumulating as a defense reaction [16,37]. Other examples of pathways over-represented among metabolites accumulating in Fix − nodules included the citrate cycle (TCA), as well as the connected glyoxylate and dicarboxylate pathways. In fact, isocitrate, aconitate, oxalate, and tartaric acid accumulated inside Fix − nodules ( Table 2). The amino acid lysine and the metabolites involved in its biosynthesis (diaminopimelate) and degradation (pipecolate) were found in the list of compounds with reduced amounts in Fix − nodules (Tables 1  and 2). Several acyl donors involved in fatty acid elongation (octanoyl-CoA, tetradecanoyl-CoA, (S)-3-hydroxytetradecanoyl-CoA and 3-hydroxy-5-methylhex-4-enoyl-CoA) also showed decreased amounts in Fix − nodules. The ion corresponding to chorismate was the metabolite with the highest decrease in Fix − nodules, compared to wild-type nodules. Chorismate is a precursor for the aromatic amino acids (phenylalanine, tryptophan, and tyrosine), indole derivatives, salicylic acid, alkaloids, vitamin K, and folate. Interestingly, phenylalanine is transformed into 4-coumaroyl-CoA, which is a precursor for flavonoid biosynthesis [38]. Therefore, the decrease in chorismate levels is in line with the accumulation of flavonoids/isoflavonoids in Fix − nodules. While the glutamine level was much lower in nodules occupied by the rpoN mutant than in wild-type nodules, glutamate was more abundant, suggesting that RpoN controls nitrogen assimilation inside the nodule.    We also compared the abundance of metabolites detected in wild type-infected P. vulgaris nodules with those from the roots of uninfected plants (baseline): 169 compounds showed increased abundance in nodules, while 62 accumulated in the roots (Table S1). Interestingly, out of the 70 metabolites with significantly decreased abundance in nodules infected by the rpoN mutant (s. above), 63 (90%) were also found among the 169 metabolites accumulating in the nodules compared to the roots. This suggested an important role of P. phymatum RpoN in P. vulgaris nodule physiology ( Figure S1). For instance, chorismate was the metabolite decreasing the most in Fix − nodules, compared to Fix + nodules, and the compound that accumulated the most in nodules compared to roots. Examples of compounds following the same pattern of a decrease in Fix − nodules compared to Fix + nodules were glutamine, arginine, alanine, and lysine. In contrast, only 17 out of the 77 metabolites accumulated in nodules occupied by the rpoN mutant displayed increased amounts in the roots, compared to wild-type nodules ( Figure S2). This result suggested that P. vulgaris reacted to the presence of the P. phymatum Fix − mutant with a specific reaction, including the accumulation of flavonoids/isoflavonoids, nucleotides, and nucleotide sugars.
Very little is known about the nutrients (carbon and nitrogen sources) consumed by the β-rhizobial strain P. phymatum during symbiosis with legumes. To mimic the root nodule environment, we performed a Biolog experiment by growing P. phymatum wild-type cells under micro-oxic conditions, in 96 well plates with different carbon (C) or nitrogen (N) sources. A list of the compounds used by P. phymatum under micro-oxic conditions is provided in Table S2. Interestingly, several compounds that were more abundant in nodules compared to roots (Table S1), such as glutamine and arginine, were utilized well by P. phymatum as C and N sources (Table S2).

RpoN Regulon in Symbiosis, as Determined by Transcriptomics
To further dissect the role of RpoN during symbiosis, and to complement the metabolomics data with an additional data type, we investigated its regulon by differential transcript profiling analysis. Therefore, the P. vulgaris root nodules infected with a P. phymatum wild-type or an rpoN mutant strain were processed at 21 dpi and analyzed by transcriptomics. Two independent biological replicates per strain were processed and sequenced. Subsequently, the unique reads mapping to the P. phymatum genome were used for a differential gene expression analysis using the DESeq package [39]. The analysis revealed that among the top 500 differentially expressed genes (DESeq analysis p-value ≤ 0.02, with log 2 (FC) ≥ 1.35 and ≤ −1.25), 322 genes were positively regulated by RpoN. Of these regulated P. phymatum genes, 39% were located on chromosome 1 (3.48 Mb), 22% on chromosome 2 (2.7 Mb), 14% on plasmid 1 (1.9 Mb), and 25% on the symbiotic plasmid (0.59 Mb) ( Figure S3). This indicated a significant enrichment of positively-regulated genes on the symbiotic plasmid, which only accounts for 6% of all P. phymatum genes ( Figure S3). To categorize these differences into modules of biological relevance, the top 500 differentially regulated genes were assigned to functional categories ( Figure 2).

Figure 2.
Functional categories of the top 500 differentially expressed genes in P. phymatum nodules infected by the wild type versus an rpoN mutant during symbiosis, with P. vulgaris (genes up-regulated in the rpoN mt are in black, those that were down-regulated are in grey) according to classification by eggNOG [40]. Percentages were calculated by dividing the number of significantly up-regulated (322) or down-regulated (178) genes in each category by the total number of retained genes in the same category. The asterisks (*) indicate statistical significance (Fischer test, p-value < 0.01).
Among the top 500 regulated genes, 178 were up-regulated in nodules induced by the rpoN mutant. The eggNOG [40] category "cell motility" was the only category found to be significantly over-represented among these genes ( Figure 2); the category includes flagella structural genes (Bphy_2931, fliF; Bphy_2937, fliL; Bphy_2949-51, fliCD) and a regulatory gene (Bphy_2962, flhF). It is known that most bacteria are able to directionally move in response to different external stimuli, e.g., to swim away from harmful compounds [41]. The up-regulation of flagellar genes in this case might be due to the hostile environment in which the Fix − bacteroids are embedded. Indeed, metabolomics analysis revealed an accumulation of flavonoids in nodules induced by the rpoN mutant strain, which were shown to be potentially noxious for other bacteria [42,43]. In line with the suggestion that rpoN mutant bacteroids undergo stressful conditions, several resistance-nodulation-division (RND) efflux transporter genes (Bphy_1582, Bphy_3489, Bphy_3490, Bphy_3492, Bphy_4022, and Bphy_4998) [44], as well as several genes involved in the response to oxidative stress (the alkyl hydroperoxide reductases Bphy_1001 and Bphy_3656) were up-regulated. In contrast, among the 322 genes positively controlled by RpoN, the three categories "cell wall/membrane/envelope biogenesis", "energy production and conversion", and "inorganic ion transport and metabolism" were over-represented ( Figure 2). Within the category "cell wall/membrane/envelope biogenesis", a cluster involved in the formation of peptidoglycan (PG) (Bphy_2672-78) and the D-alanine:D-alanine ligase encoding gene Bphy_2671 (ddl) were identified. Moreover, two potential gene clusters (Bphy_1676-95 and Bphy_2458-76) coding for exopolysaccharides (EPS) and lipopolysaccharides (LPS), respectively, as well as genes coding for several glycosyltransferases (Bphy_1681, Bphy_2460, Bphy_2464, Bphy_2468-69, Bphy_3557, and Bphy_7707) were found. The two categories "energy production and conversion" and "ion transport and metabolism" had already previously been shown to be over-represented during symbiosis with P. vulgaris, when free-living cells were used as baseline [23]. In fact, the functional category "energy production and conversion" includes genes with important symbiotic functions, such as nitrogen fixation (nif cluster), hydrogenase (hyd cluster), and a cytochrome o ubiquinol oxidase operon (cyoABCD). Other genes encoding an ATP synthase (Bphy_3029, Bphy_3031-32) and a pyruvate dehydrogenase (Bphy_3758-60) were also found in the same category, and showed induced and RpoN-dependent expression during symbiosis. Among RpoN-activated genes, the over-represented category "inorganic ion transport and metabolism" included amtB (Bphy_0257) encoding an ammonium transporter, genes coding for different ATP-binding cassette (ABC)-transporters for sulfate (Bphy_1627, Bphy_1629 and Bphy_7234-36), nitrate/sulphonate/bicarbonate (Bphy_3603), aliphatic sulphonate (Bphy_5227 and Bphy_5229), taurine (Bphy_6080), and urea (Bphy_2251-52 and Bphy_2255). Moreover, several regulatory genes were found among the top 500 genes showing reduced expression in Fix − nodules (Table S3): ntrB and ntrC, which are part of a two-component regulatory system (2CRS), known to be relevant for nitrogen metabolism in several bacteria (s. below) [45]; Bphy_1669 coding for a cyclic AMP receptor protein (Crp)-fumarate and nitrate reduction regulator (FNR) family transcriptional regulator; six LysR-type transcriptional regulators; and seven regulators of the AraC family.
A large overlap of 52% (168 genes) was identified among the 322 genes positively regulated by RpoN during symbiosis and the genes with increased expression during symbiosis with P. vulgaris, compared to free-living conditions reported earlier [23] ( Figure S4 and Table S4). On top of the nif, hyd, and cyo genes, several genes from the functional category "inorganic ion transport" described above were found to be co-regulated. Among them, we found several ATP-binding cassette (ABC)-transporter genes for sulfate (Bphy_1627 and Bphy_1629), nitrate/sulphonate/bicarbonate (Bphy_3603), aliphatic sulphonate (Bphy_5227 and Bphy_5229), taurine (Bphy_6080), and urea (Bphy_2251-52). Furthermore, the expression of two transcriptional regulatory genes, ntrC and Bphy_1669, was induced in nodules (compared to roots) and RpoN-dependent.
Q-PCR analysis was performed on selected genes showing increased, decreased, and unchanged expression in Fix − nodules compared to Fix + . In all cases, the trend of the RNA-Seq results could be confirmed (Table S5).
In summary, the symbiotic RpoN regulon of P. phymatum showed a prominent overlap with RpoN regulons, determined previously using microarrays in nodules formed by two α-rhizobial strains (Bradyrhizobium diazoefficiens and Rhizobium etli) [11,29,46]. Most notable were the activation of nitrogenase and hydrogenase gene expression, the activation of genes involved in respiration, and the transport of ammonium.

In Silico Identification of RpoN Binding Sites
A genome-wide in-silico search for RpoN DNA binding motifs (GGCACG-N4-TTGC) in all P. phymatum promoter sequences was conducted, relying on a previously described position-specific frequency matrix (PSFM) of RpoN binding sites in the closely related strain B. cenocepacia H111 [27]. A total of 409 putative RpoN binding boxes were detected (see Section 3). Interestingly, 86 out of the 322 genes down-regulated in nodules infected by the rpoN mutant showed a potential RpoN binding site in their promoter region, and might therefore be directly controlled by this sigma factor. Several of the genes that showed RpoN-dependent expression in root nodules also contained a highly significant (p-value ≤ 6.58 × 10 −5 ) RpoN binding sequence in their promoter region: the promoter of Bphy_0256, coding for nitrogen regulatory protein P-II (glnB1), which is localized upstream of the ammonium transporter gene (amtB) ( Figure 3A, Table 3); the promoter of the glutamine synthetase encoding gene (Bphy_1481, glnA); followed by two genes of the 2CRS NtrBC ( Figure 3B), which were also present among the 322 top down-regulated genes. Moreover, a highly significant RpoN-binding sequence was localized in front of Bphy_2251, which encodes an urea ABC transporter ( Figure 3D), and upstream of an operon coding for a hydantoinase (Bphy_6150-52, Figure 3C), which usually hydrolyzes cyclic amide bonds [47]. In a previous study on B. diazoefficiens symbiotic gene expression, a gene coding for a hydantoin utilization protein was also found to be a direct RpoN target [11]. Other genes harboring a putative RpoN binding motif in the promoter region that were down-regulated in Fix − nodules are listed in Table 3. These included a gene in the hydrogenase cluster (Bphy_7265, hydA) and genes in the nif cluster (Bphy_7729, nifE; Bphy_7741, nifV; Bphy_7742, nifB; Bphy_7753, nifH 1 and Bphy_7808, nifH 2 ). The potential RpoN binding motif in P. phymatum is illustrated in Figure S5.  (Table S3) are colored in grey, and their log 2 fold expression changes are shown below. Black arrows indicate the position of the RpoN box; the distance (in nucleotides) from the middle of the box to the translation start site is indicated below the arrow.

Integration of Metabolomics and Transcriptome Data
Several changes in P. phymatum transcript expression were supported by respective metabolic alterations. For example, the up-regulation of the citrate synthase gene gltA (Bphy_5205) was in agreement with the accumulation of aconitate and isocitrate in nodules induced by the rpoN mutant. The decrease in oxo-glutarate, and the increase in glyoxylate in nodules containing the rpoN mutant, suggested that the glyoxylate cycle is more active in nodules induced by this Fix − mutant. Indeed, the isocitrate lyase encoding gene (Bphy_1368), which catalyzes the cleavage of isocitrate to succinate and glyoxylate, showed increased expression in nodules infected by the Fix − mutant.
The down-regulation of the D-alanine:D-alanine ligase encoding gene ddl (Bphy_2671) and of the downstream mur cluster coding for the cell wall component murein or peptidoglycan (PG) was in line with a decrease in the level of the metabolite D-alanyl:D-alanine in nodules induced by the rpoN mutant. The D-alanine:D-alanine ligase catalyzes the dimerization of two D-alanine molecules to generate the dipeptide D-alanyl:D-alanine, which is involved in the biosynthesis of PG. The level of this dipeptide was also increased in nodules compared to roots, and, accordingly, we have previously shown that Bphy_2671 had increased expression in root nodules compared to free-living conditions [23]. Interestingly, the results from Wu and colleagues, who showed that certain flavonoids, such as apigenin-which is accumulating in Fix − nodules-target and inhibit the activity of the D-alanine:D-alanine ligase by competing with ATP [48], support our observations. We can therefore speculate that the accumulation of flavonoids in nodules induced by the rpoN mutant may lead to defective PG biosynthesis in Fix − bacteroids.
The reduced amount of pentose in Fix − nodules is in line with an increased expression of the ribose-phosphate pyrophosphokinase encoding gene Bphy_0315, as well as with an increased level of several intermediates in the purine metabolism (e.g. inosine and ITP). The degradation of purines leads to the formation of ureides, which are key nitrogen compounds for the growth and yield of legumes forming determinate nodules [49]. The decrease in the level of the ureide allantoic acid in Fix − nodules is supported by an up-regulation of Bphy_2595 coding for an allantoicase, and by an increased level of glyoxylate. Allantoin and allantoic acid were both among the compounds accumulating in P. phymatum-P. vulgaris nodules compared to uninfected roots (Table S1).
In agreement with reduced amounts of the valine precursor ketovaline in Fix − nodules, we found an increased expression of the 2-isopropylmalate synthase gene, whose product converts ketovaline into isopropylmalate (leuA, Bphy_2017). Interestingly, valine was one of the nitrogen sources that was better used by P. phymatum under micro-oxic growth conditions (Table S2).
In nodules induced by the rpoN mutant, both the level of lipoamide and the expression of the Bphy_3757-61 cluster coding for a pyruvate dehydrogenase significantly decreased. Moreover, the search for RpoN binding boxes detected a potential binding site in the promoter region of Bphy_3761. In B. diazoefficiens-soybean root nodules, the expression of the pyruvate dehydrogenase complex was also RpoN-dependent, and the promoter displayed a clear RpoN binding site [11], suggesting direct control of RpoN on this key enzyme in both αand β-rhizobia.
The reduced amounts of several fatty acyl-CoA molecules in Fix-nodules is in agreement with an upregulation of the enoyl-CoA hydratase encoding gene Bphy_4071, and of the adjacent Bphy_4072 coding for a protein with a acyl-CoA dehydrogenase domain. Both enzymes are important to metabolizing fatty acids to generate both acetyl CoA and energy.
The level of glutamine, which was increased in nodules compared to roots, depends on a functional RpoN. In fact, in Fix − nodules glutamine was detected in lower amounts, and the expression of two glutamine synthetase encoding genes-Bphy_1481 (glnA, Figure 3B) and Bphy_7784-also decreased. Accordingly, an increase in the level of glutamate was observed in Fix − nodules. This last result strongly suggested that RpoN is controlling the conversion of glutamate into glutamine in P. vulgaris root nodules.

Role of ntrB during Symbiosis
The gene ntrB is part of the 2CRS NtrBC, and codes for the sensor kinase necessary for the control of nitrogen metabolism in several bacteria [50,51]. Its companion, transcriptional regulator NtrC, is located downstream, and usually acts together with the sigma factor RpoN to regulate the transcription of several genes. As mentioned above, the expression of ntrB and ntrC was significantly down-regulated in nodules infected by the rpoN mutant versus P. phymatum wild-type nodules. Furthermore, ntrB had previously been shown to be up-regulated during symbiosis with bean compared to free-living conditions [23]. The presence of a putative RpoN binding box in the promoter region ( Figure 3B) led us to investigate the role of NtrB during symbiosis, by constructing a mutant strain.
The ntrB deletion mutant (∆ntrB), as well as the nalidixic acid-resistant wild type (wt nal R ), were inoculated on seedlings of P. vulgaris. After three weeks of growth in a plant chamber, several symbiotic parameters were determined, including the number of nodules per plant, the dry weight per nodule, and the relative nitrogenase activity (Figure 4). A lower number of nodules, but with a significantly higher dry weight per nodule, were counted on plants infected by ∆ntrB, compared to the ones infected by the nalidixic acid-resistant wild type ( Figure 4A,B). The nodules induced by the ∆ntrB mutant showed a 36% reduction in the relative nitrogenase activity compared to wild-type nodules ( Figure 4C). However, this reduction was not statistically significant (when choosing a significance level α of 0.05). These results suggested that, contrary to RpoN, NtrB from the NtrBC 2CRS plays a marginal role for an efficient establishment of a symbiosis with bean.

Role of ntrB for Nitrogen Uptake in Free-Living Conditions
To further investigate the role of P. phymatum ntrB, the ability to assimilate different nitrogen sources-such as ammonium, nitrate, and urea-was examined in free-living conditions. We found that the growth of the wild type was virtually undistinguishable to that of the nalidixic acid-resistant wild type for all of the three tested nitrogen sources ( Figure 5). In contrast, the ∆ntrB mutant had a growth defect in the presence of nitrate and urea, but not with ammonium as a nitrogen source ( Figure 5). Therefore, in line with previous studies in other bacteria [26,52], P. phymatum NtrB-similar to RpoN-is involved in the assimilation of nitrate and urea. Figure 5. Utilization of selected nitrogen sources by P. phymatum wild-type (wt), nalidixic acid-resistant wild type (wt nal R ), and by a ntrB deletion mutant strain (∆ntrB). Growth was assessed with at least two independent replicates. Error bars indicate standard deviation (SD). For each group of columns, values with the same letter are not statistically different, while those with different letters are (ANOVA, Tukey's test, p < 0.001).

Bacterial Strains, Media, and Cultivation
The bacterial strains, plasmids, and primers used in this work are listed in Table S6. Escherichia coli cells were cultivated in a modified Luria-Bertani liquid medium (LB [53]; 10 g of tryptone, 5 g of yeast extract, and 4 g NaCl per liter), whereas P. phymatum was grown aerobically in the LB medium without salt. The antibiotics chloramphenicol (20 µg/mL for E. coli and 80 µg/mL for P. phymatum), kanamycin (25 µg/mL for E. coli and 50 µg/mL for P. phymatum), and nalidixic acid (50 µg/mL for P. phymatum) were used.
The growth of two P. phymatum strains-a wild type spontaneously resistant to nalidixic acid (wt nal R ) and a deletion mutant in ntrB (∆ntrB)-were tested in a defined buffered AB minimal medium [54], with 10 mM of sodium citrate as carbon source, and further supplemented with one of the following nitrogen sources, respectively: 30 mM ammonium (Sigma-Aldrich, St. Louis, MO, USA), 30 mM nitrate (Sigma-Aldrich), and 15 mM urea (Sigma-Aldrich). For each strain, the growth from at least two independent cultures was measured. The spectrophotometer Ultrospec 2100 pro (Amersham Biosciences, Little Chalfont, UK) was used.

Plant Growth Conditions
Seeds of the common bean (Phaseolus vulgaris, cv. Negro jamapa) were surface-sterilized, as previously described [55]. Germination, inoculation, and the growth conditions for the legumes were conducted as reported previously [23]. Legumes were inoculated with wild-type P. phymatum, the nalidixic acid-resistant wild type (wt nal R ), the rpoN mutant [23], and the ntrB deletion strain (∆ntrB).

Plant Harvesting and Metabolite Extraction
For comparison of metabolite abundances, bean root nodules induced by different strains (P. phymatum wild-type and rpoN mutant) and uninfected roots were collected at 21 dpi. Three biological replicates, each in two technical replicates, were analyzed. Around 30 mg nodules or roots were used per sample. Nodules and uninfected roots were flash-frozen in liquid nitrogen immediately after harvesting. Samples were further processed to extract hydrophilic metabolites (methanol extract) as previously described [16].

Metabolite Data Analysis
The methanol extracts were injected in an Agilent 6550 QTOF instrument (Agilent Technologies, Santa Clara, CA, USA) and analyzed by non-targeted flow injection-time-of-flight mass spectrometry in negative mode ionization, as described [33]. Ions were annotated as previously described [16]. Given the existence of metabolites with identical molecular formula and weight, a larger number of candidate metabolites compared to ions was expected. In fact, a total of 409 ions with distinct m/z values were matched to 493 deprotonated molecules, which were then subjected to statistical analysis. Table S7 contains an overview of all metabolomics raw data, including ions, annotations, and intensities. A comparative statistical analysis was performed, according to Storey and colleagues [56], applying a two-tailed and heteroscedastic t-test, followed by the application of a false discovery rate (FDR) correction. A metabolite was considered to be significantly differentially abundant if the following requirements were satisfied: abs[log 2 (fold-change)] ≥ 0.5 and q-value < 0.01. Pathway enrichments were calculated using the procedure described previously, using the plant and bacterial KEGG databases [57].

RNA-Sequencing and Data Processing
A modified hot acid phenol protocol [11] was used to extract the total RNA from flash-frozen root nodules induced by the wild type and the rpoN mutant (approximately 40 nodules per sample). An additional acid phenol treatment was performed for each sample. Two independent biological replicates were analyzed per condition. Next, complete removal of genomic DNA (gDNA) by DNase treatment and a quality check of the total RNA were performed [27]. The samples were subjected to plant ribosomal RNA (rRNA) removal with the Ribo-zero TM Plant-Seed/Root kit (Epicentre, Madison, DC, USA [23], prior to cDNA synthesis, with a total of 150 ng of high-quality RNA. As previously reported, the library preparation and purification was performed with the Encore Complete Prokaryotic RNA-Seq DR Muliplex System (NuGEN, San Carlos, CA, USA) [23]. The cDNA libraries were quantified using a TapeStation (Agilent Technologies) and sequenced with a HiSeq2500 instrument, single end and 125 base-pair (Illumina, San Diego, CA, USA) [23]. The sequencing reads were trimmed to 70 bp, and further processed and mapped to the P. phymatum STM815 T genome [5], utilizing the CLC Genomics Workbench v7.0 (QIAGEN CLC bio, Aarhus, Denmark), allowing up to two mismatches per read. The DESeq R-package (version 1.30.0) [39] was employed to statistically analyze the reads, uniquely mapping to the genome for differential expression [39]. From the DESeq analysis, the top 500 significantly RpoN regulated genes (ranked by ascending p-value) were selected. Table S8 contains a complete list of all P. phymatum genes and the log 2 of the fold changes (FC), comparing expression in nodules induced by an rpoN mutant versus expression in wild-type nodules. EggNOG v3.0 was used to group the differentially-expressed genes into functional categories [40].The RNA-seq raw data files of nodules induced by both the wild type and by the rpoN mutant are accessible through the GEO Series, accession number GSE111993.

q-PCR Analysis
The differential gene expression of the following P. phymatum genes-Bphy_0257 (amtB), Bphy_1479 (ntrC), Bphy_7808 (nifH 2 ), Bphy_3941 (rpoD) and Bphy_3492-was tested by quantitative reverse transcription-PCR (qRT-PCR) [27]. cDNAs were synthesized from an independent biological replicate, as described before [58]. Then qPCR was performed using Brilliant III Ultra-Fast SYBR green QPCR master mix (Agilent Technologies) and a Mx3000P instrument (Agilent Technologies,). For each cDNA sample, three dilutions-15, 7.5, and 3.75 ng/µL-were employed as templates and tested in triplicate. The sigma factor rpoD was used to normalize the fold changes in expression, which were calculated with the ∆∆C t method [59]. An annealing temperature of 58 • C was set for all the primers pairs tested.

Genome-Wide In Silico Prediction of RpoN-Binding Sequences
The R-package TFBSTools (version 1.14.2) [60] was employed to predict RpoN binding sites, within a window of 300 nucleotides (nts) upstream of the translation start site of each gene/operon in the P. phymatum genome. For the analysis, a minimum score of 85% was set as cut-off. The position-specific frequency matrix (PSFM) reported for RpoN of the closely related Burkholderia cenocepacia H111 [27] was used as input.

Construction of a Paraburkholderia Phymatum STM815 T ntrB Deletion Mutant
The DNeasy Blood and Tissue kit (Qiagen, Hilden, Germany) was utilized for the extraction of gDNA from P. phymatum. To construct a ntrB deletion mutant, two fragments flanking the gene ntrB were chosen as recombination sites for deletion. The upstream fragment 1 (666 nt) was amplified with primers Bphy1480_1 and Bphy1480_2, and the downstream fragment 2 (692 nt) was amplified with primers Bphy1480_3 and Bphy1480_4, using the gDNA of P. phymatum as a template. Both fragments were digested with XbaI, ligated, and cloned into the pGEM-T easy vector (Promega, Madison, DC, USA). A kanamycin resistance cassette was cut out from pKD4 [61] plasmid, using XbaI, and ligated in between the two fragments on the pGEM-T vector (Promega, Madison, DC, USA). Then, the whole insert was released by digestion with KpnI and sub-cloned into pSHAFT2 [62]. The resulting plasmid was mobilized into a previously-constructed, spontaneous, nalidixic acid-resistant P. phymatum strain [23]. The ntrB deletion mutant (∆ntrB) was confirmed with PCR, using primers Km_F and glnA_R.

Determination of Symbiotic Properties
Nodule number, nodule dry weight, and nitrogenase activity were calculated as described previously [23,63]. Two independent experiments, with at least nine plants per strain, were performed.

Biolog Analysis
Biolog Phenotypical MicroArray (Biolog, Hayward, CA, USA) for carbon (PM1, 2a) and nitrogen (PM3b) utilization were performed as described [64]. After two days incubation at 30 • C on R-2A agar (Sigma), P. phymatum cells were prepared for the inoculation of the Biolog PM plates. The PM plates were then incubated for three days at 30 • C under micro-oxic conditions, using the BD GasPak™ system jar (Becton Dickinson, Franklin Lakes, NJ, USA) by changing the sachet once a day. Afterwards, the OD 590 of each well was measured using a plate-reader.

Conclusions
In this study, we aimed to dissect the regulon of P. phymatum RpoN, the key sigma factor of nitrogen-fixing symbiosis, by analyzing and integrating metabolite changes with changes in gene expression in P. vulgaris root nodules. The importance of RpoN for symbiosis, both at the metabolite and transcript level, was evidenced by a large overlap of nodule-specific metabolites/transcripts with RpoN-dependent metabolites/transcripts (90% and 52%, respectively). Furthermore, we found numerous examples where the different metabolite abundances correlated with P. phymatum genes expression differences (e.g., PG biosynthesis, conversion of glutamine in glutamate, glyoxylate cycle) ( Figure 6). Interestingly, we found that nodules induced by the Nod + Fix − mutant seemed to accumulate significant amounts of flavonoids, which may be used as a defense reaction by the plant, which probably recognizes this strain as a Fix − cheater strain. This hypothesis is supported by the rpoN mutant-induced up-regulation of several P. phymatum RND efflux transporter genes, as well as other genes involved in the resistance to stress inside nodules. The effects of accumulating flavonoids on P. phymatum growth could be tested by adding different concentrations of flavonoids to free-living cultures. Microscopic analysis to investigate the ultrastructure of Fix + and Fix − nodules may help to gain more information about the status of their bacteroids. This first metabolome analysis of β-rhizobial root nodules contributed new insights on the differences between the physiology of nodules induced by α-and β-rhizobia. Glutamate, for example, was among the most abundant metabolites in nodules of several α-rhizobial symbiosis model systems [16,[65][66][67][68], but did not show differential abundance in bean nodules induced by P. phymatum.
In order to better characterize the adaptation of P. phymatum to papilionoid legumes, metabolomics studies should also be performed on nodules formed by P. phymatum on its native mimosoid host plant in the near future.  Figure S1: Venn diagram showing the overlap between 169 metabolites that are more abundant in nodules infected by P. phymatum wild-type (wt) versus bean root, and the 70 metabolites that are less abundant in nodules induced by P. phymatum rpoN mutant (rpoN) versus the wild-type (wt); Figure S2: Venn diagram showing the overlap between the 62 metabolites that are less abundant in nodules infected by P. phymatum wild-type (wt) versus bean root, and the 77 metabolites that are more abundant in nodules induced by P. phymatum rpoN mutant (rpoN) versus the wild-type (wt); Figure S3: Pie charts representing the percentages of genes distributed in the P. phymatum STM815 genome (A); In (B), the localization of the 322 genes showing down-regulation in nodules induced by the rpoN mutant is shown. A statistically significant enrichment of genes located on the symbiotic plasmid is observed; Figure S4: Venn diagram showing the overlap between the 322 genes significantly and positively controlled by RpoN, and the 322 genes up-regulated during symbiosis, compared to free-living conditions (*; Lardi et al., 2017); Figure S5: Consensus sequence logo for P. phymatum STM815 RpoN-binding site. Depicted is a putative RpoN-binding site for P. phymatum, based on the 409 promoter sequences. The new consensus sequence was created using WebLogo (http://weblogo.berkeley.edu/); Table S1: List of 231 metabolites showing altered abundance in bean root nodules induced by the P. phymatum wild-type, compared to the uninfected root; Table S2: C and N source utilization of P. phymatum STM815 wild-type during micro-oxic conditions, using Biolog PM plates (1, 2a for C and 3b for N). After three days incubation at 30 • C, optical density at 590 nm was measured for each well; Table S3: List of the top 500 differentially regulated genes in nodules induced by rpoN mutant versus wild-type nodules (DESeq analysis; p-value ≤ 0.02, log2 [FC] ≥ 1.35 and ≤ −1.25); Table S4: List of the 168 genes positively regulated by RpoN during symbiosis and showing induced expression in nodules compared to free-living conditions ; Table S5: qPCR analysis of several genes. Fold changes in expression of the selected genes in nodules induced by rpoN mutant (rpoN mt nod) versus expression in wild-type nodules (wt nod) are shown; Table S6: Bacterial strains, plasmids, and oligonucleotides used in this study; Table S7: Raw metabolomics data, including injections, ions, annotations, and intensities; Table S8: List of all P. phymatum genes, log 2 of the fold changes in expression, and the p-values in bean nodules infected by rpoN mutant versus wild type as assessed by DESeq analysis.
We are thankful to Hans-Martin Fischer and his lab for letting us use the gas chromatography at the Institute of Microbiology at the ETH Zürich, to Laura Dal Ben Settele for helping in the graphics of Figure 6, and Pietro Cattaneo for advising us on statistic issues. This work was supported by the Swiss National Science Foundation (31003A_153374 to Gabriella Pessi).