High-Resolution Transcriptomic Analyses of Sinorhizobium sp. NGR234 Bacteroids in Determinate Nodules of Vigna unguiculata and Indeterminate Nodules of Leucaena leucocephala

The rhizobium-legume symbiosis is a model system for studying mutualistic interactions between bacteria and eukaryotes. Sinorhizobium sp. NGR234 is distinguished by its ability to form either indeterminate nodules or determinate nodules with diverse legumes. Here, we presented a high-resolution RNA-seq transcriptomic analysis of NGR234 bacteroids in indeterminate nodules of Leucaena leucocephala and determinate nodules of Vigna unguiculata. In contrast to exponentially growing free-living bacteria, non-growing bacteroids from both legumes recruited several common cellular functions such as cbb3 oxidase, thiamine biosynthesis, nitrate reduction pathway (NO-producing), succinate metabolism, PHB (poly-3-hydroxybutyrate) biosynthesis and phosphate/phosphonate transporters. However, different transcription profiles between bacteroids from two legumes were also uncovered for genes involved in the biosynthesis of exopolysaccharides, lipopolysaccharides, T3SS (type three secretion system) and effector proteins, cytochrome bd ubiquinol oxidase, PQQ (pyrroloquinoline quinone), cytochrome c550, pseudoazurin, biotin, phasins and glycolate oxidase, and in the metabolism of glutamate and phenylalanine. Noteworthy were the distinct expression patterns of genes encoding phasins, which are thought to be involved in regulating the surface/volume ratio of PHB granules. These patterns are in good agreement with the observed granule size difference between bacteroids from L. leucocephala and V. unguiculata.


Introduction
The rhizobium-legume symbiosis has been a model system for studying the mutualistic interactions for many years [1]. This system is characterized by its ability to form the symbiotic nodules, in which rhizobia differentiate into bacteroids to fix atmospheric nitrogen to the benefit of the legume while the host provides carbon sources to these microsymbionts. Symbiotic nodules could be simply categorized into either determinate or indeterminate nodules based on the activity of the nodule meristem. In the determinate nodules (such as those of Lotus japonicus, Glycine max and Phaseolus vulgaris), the meristem functions until the formation of the nodule primordium and produces synchronously developed infected cells, whereas the meristem persists in the indeterminate nodules (such as Medicago sativa, Pisum sativum and Vicia sativa) which are composed of different zones showing a clear developmental gradient: the apical meristem, the invasion zone, the interzone, the nitrogen-fixing zone, and the senescence zone [2,3,4]. Enlarged and nonreproductive bacteroids were thought to be a characteristic feature of indeterminate nodules of Inverse-Repeat Legume Clade (IRLC) including M. sativa, P. sativum and V. sativa, in contrast, morphologically unchanged and reproductive bacteroids were commonly found in determinate nodules of Milletioids plants including L. japonicus, G. max and P. vulgaris [5]. However, the latest study of the distribution of these two kinds of bacteroids in 40 legume species in the subfamily Papilionoideae [6] suggested that there was no clear correlation between nodule types and morphologies of bacteroids. Moreover, morphologically unchanged bacteroids might be the ancestral form to enlarged bacteroids [6].
High throughput transcriptomics and proteomics studies have revealed global gene expression profiles of enlarged bacteroids in indeterminate nodules [7,8,9,10,11] or morphologically unchanged bacteroids in determinate nodules [12,13,14,15,16]. Few of them compared host-specific bacteroid transcripts and/or proteins [11,16]. Recently, exciting progresses have been made in identifying the determinants of the interactions between host cells and enlarged bacteroids in indeterminate nodules of Medicago and Pisum [17,18,19,20,21]. But little information was known for the persistence mechanisms of both types of bacteroids in determinate nodules and morphologically unchanged bacteroids in indeterminate nodules, partially due to different strategies used by diverse legume-rhizobium systems [2,6,22,23]. Sinorhizobium sp. NGR234 is well known for its ability to nodulate legume hosts from as many as 112 genera forming either determinate or indeterminate nodules [24]. Thus NGR234 could serve as an excellent model for investigating the adaptation mechanisms of rhizobia to diverse conditions within different types of nodules. In this study, RNAseq was used to investigate the transcriptomic differences between free-living NGR234 and NGR234 bacteroids in either determinate nodules of V. unguiculata or indeterminate nodules of L. leucocephala. Both common and distinct transcription patterns of NGR234 bacteroids in these two legumes were analyzed.

Growth Conditions for Bacterial Strain and Plants
The broad-host range strain Sinorhizobium sp. NGR234 [24] was grown in liquid TY (tryptone yeast extract) medium [25] at 28uC. Bacterial culture with OD 600 = 0.6 (the optical density at 600 nm) was used for inoculating legume hosts V. unguiculata and L. leucocephala forming determinate and indeterminate nodules respectively with NGR234. Seeds of V. unguiculata were surfacesterilized by successive treatments with 95% ethanol for 30 sec and 0.2% (w/v) HgCl 2 for 5 min, and were then washed for 10 times by using autoclaved deionized water. Seeds of L. leucocephala were first treated with sulfuric acid for 30 min, followed by washing six times in sterilized water. They were then surface-sterilized as described for seeds of V. unguiculata. All the surface-sterilized seeds were germinated on 0.6% agar-water plates in the dark at 28uC for 24-48 h. Germinated seeds were planted in vermiculite moisturized with low-N nutrient solution in Leonard jars [26] and were inoculated with 1 ml of bacterial culture with OD 600 = 0.6 per plant. All the plants were grown at 24uC in a plant growth room with a daylight illumination period of 12 h. Nodules were harvested 21 DPI (days post inoculation) for V. unguiculata and 35 DPI for L. leucocephala when the acetylene reductase activity of nodules reached the peak level. Three biological replications were done. These nodules were either frozen in liquid nitrogen and stored at -80uC until RNA extraction, or used for sample preparations for electron and light microscopy.

Light and Electron Microscopy
Nodules were fixed in 2.5% glutaraldehyde in 0.05 M cacodylate buffer [27]. For light microscopy, fixed nodules were washed, dehydrated, embedded in Technovit 7100 (Kulzer Histo-Technik), according to the manufacturer's instructions. Sections of 2 mm were cut on a Leica Ultracut C6i and stained with 1% toluidine blue for 40 sec. For electron microscopy, fixed nodules were washed with 0.1 M phosphate buffer and postfixed in 0.1 M phosphate buffer containing 1% (wt/vol) OsO 4 . The samples were then washed with 0.1 M phosphate buffer and dehydrated with increasing volumes of acetone (30%, 50%, 70%, 90%, and 100%). The samples were embedded in the SPURR epoxy. A Leica Ultracut C6i was used to obtain ultrathin sections (80 nm thick) of these nodule samples. The resulting sections were stained with uranyl acetate and lead citrate and finally observed in JEM-1230 transmission electron microscope.

Isolation and Purification of Bacteroids
A slightly modified approach described by Day et al. [28] was used for bacteroid isolation from frozen nodules of V. unguiculata and L. leucocephala. Briefly, 2-5 g nodules were extensively ground by using a pre-chilled pestle in 15 ml of extraction buffer (10 mM DTT, 300 mM sucrose, 10 mM phosphate buffer pH 7.0, 2 mM MgCl 2 and 0.33 g PVP). In order to remove large particles of plant cell debris, the mixture was centrifuged at 4006g for 10min, 4uC. The supernatant was centrifuged at 120006g for 10 min, 4uC. The resulting pellet containing bacteroids and small plant cell debris was resuspended in 10 ml extraction buffer. Then 10 ml 30% percoll, 5 ml 60% percoll and 5 ml 80% percoll were added into the same tube. After the centrifugation at 40006g for 15 min, 4uC, the layer between 60%-80% percoll containing bacteroids was diluted into 20 ml using the extraction buffer. The resulting solution was transferred to the tube containing 5 mL 60% percoll and 5 mL 80% percoll and subject to another round of centrifugation at 40006g for 15 min, 4uC. The cushion above 80% percoll containing bacteroids was resuspended in 20 mL 0.8% NaCl and centrifuged at 120006g for 5 min. The pellet was collected for RNA extraction of bacteroids.

RNA Extraction
Bacteroids were ground by using a pestle in liquid nitrogen. They were then subjected to RNA extraction using QIAGEN RNeasy mini kit according to the manufacturer's instructions. For free-living bacteria, RNA from a bacterial culture with OD 600 = 0.5 was extracted by using the same QIAGEN RNeasy mini kit. RNA quality was assessed by using an Agilent 2100 Bioanalyzer. RNA integrity number (RIN, average 6 SE) was 8.8060. 10, 9.8760.13 and 9.9360.07 for RNA sample of L. leucocephala bacteroids, V. unguiculata bacteroids and the free culture of NGR234, respectively, indicating the good quality of RNA samples in this study.

Library Construction and Strand-specific RNA Sequencing
Total RNA was sent to BGI-Shenzhen for further treatments, library construction and strand-specific RNA sequencing. Briefly, total RNA was treated with RNase-free DNase I for 30 min at 37uC to remove residual DNA. Total RNA was then treated with Ribo-Zero rRNA Removal Kit (Gram-Negative Bacteria) according to the manufacturer's instructions to remove the ribosomal RNA before preparing RNA libraries for deep sequencing. 5 mg of total RNA was used as the starting material for treatment. The mRNA-enriched RNA was chemically fragmented to 150,200 bp using divalent cations under elevated temperature. The cleaved RNA fragments were copied into first strand cDNA using reverse transcriptase and random primers. Non-incorporated nucleotides were removed and dTTP was substituted by dUTP during the synthesis of the second strand [29]. These cDNA then went through an end repair process, the addition of a single ''A'' base, and then ligation of adapters. The ligation products were then purified and subsequently digested with N-glycosylase (UNG; Applied Biosystems) to remove the second-strand cDNA. The products were then enriched by 15 cycles of PCR cycles with phusion polymerase to create the final cDNA library. Libraries were sequenced on an Illumina Hiseq 2000 platform.

Sequence Analyses
Clean reads were mapped to the reference genome of Sinorhizobium sp. NGR234 [30] using SOAP2 [31]. Mismatches no more than 5 bases were allowed in the alignment. To eliminate the influence of different gene length and sequencing discrepancy on the calculation of gene expression, the RPKM (reads per kilobase per million mapped reads) method was used to calculate gene expression level [32]. Genes with the ratio of RPKMs of the two samples above 2, Benjamini FDR (False Discovery Rate) #0.001 and the coverage value larger than 80% in the transcriptionally up-regulated condition were chosen as the differentially expressed genes (DEGs) between two samples. IGV [33] was used to visualize the expression patterns across the genome. KEGG pathway annotations for Sinorhizobium sp. NGR234 were retrieved from the KEGG database [34] and used in the pathway enrichment analysis of DEGs by using Gitools [35]. The Benjamini FDR corrected P value ,0.05 (two-tailed) for Fisher exact test was used to define the enriched pathway.

qRT-PCR
To validate the results of RNA-seq, quantitative reverse transcription PCR (qRT-PCR) experiments were performed in triplicate for 13 genes having different expression profiles. Three biological replicates were analyzed. Single strand cDNA was synthesized by using the GoScript TM Reverse Transcription System kit (Promega). Quantitative PCR was performed by using 25 mL of Light Cycler 480 SYBR Green I Master (Roche) and a Light Cycler 480 real-time PCR system (Roche). The PCR procedures were as follows: 95uC for 2 min; 40 cycles of 95uC for 15 sec, 60uC for 1 min. PCR results were analyzed by relative quantification methods using the 16S rRNA gene (NGR_c26520) as the reference gene.

Results and Discussion
Distinct Characteristics of Nodules from V. unguiculata and L. leucocephala Sinorhizobium sp. NGR234 formed spherical nodules on V. unguiculata and elongated nodules on L. leucocephala ( Figure 1A and 1C). These elongated indeterminate nodules contain a typical meristem zone on the distal part of nodules ( Figure 1D). However, bacteroids in determinate or indeterminate nodules of these two legume hosts show several distinct characteristics compared to those in the typical determinate (L. japonicus, G. max etc.) or indeterminate (M. truncatula, P. sativa etc.) nodules [2]. Bacteroids in nodules of V. unguiculata ( Figure 1E) and L. leucocephala ( Figure 1F) are ,2.5 mm in length, suggesting that they all belong to the morphologically unchanged bacteroids [6]. Moreover, there is just one bacteroid surrounded by each peribacteroid membrane in nodules from these two legume hosts ( Figure 1E and 1F). Notably, poly-3-hydroxybutyrate (PHB) granules in bacteroids of L. leucocephala nodules ( Figure 1F) were larger than those in bacteroids of V. unguiculata ( Figure 1E).

Gene Expression Overview
RNA-Seq experiments produced 1.169 Gb sequences for the free-living bacteria, V. unguiculata and L. leucocephala bacteroids, respectively. As shown in Table 1, 97.92%, 87.23% and 56.83% of obtained reads in the three treatments were mapped to Sinorhizobium sp. NGR234 genome. These mapping results were consistent with a relatively higher level of the contamination by plant RNA in the RNA samples of L. leucocephala bacteroids than in V. unguiculata bacteroids (data not shown). However, around 80 Mb sequences from the L. leucocephala bacteroids' sample were uniquely mapped to known CDS of Sinorhizobium sp. NGR234 genome (6.9 Mb), and this was among the largest RNA-seq data relative to the genome size of bacteria [36,37,38,39].
Transcriptomes of bacteroids from nodules of V. unguiculata (21 DPI) and L. leucocephala (35 DPI) were compared to that of free-living Sinorhizobium sp. NGR234 in exponential growth stage (Table S1). As shown in Table 2 and Table S2, among the 3143 DEGs between V. unguiculata bacteroids and free-living bacteria, 90.2% DEGs of pNGR234a, 66.4% DEGs of pNGR234b and 29.1% DEGs of the chromosome were up-regulated in V. unguiculata bacteroids. Similarly, 96.6% DEGs of pNGR234a, 84.1% DEGs of pNGR234b and 35.9% DEGs of the chromosome were up-regulated in L. leucocephala bacteroids compared to those of free-living bacteria ( Table 2 and Table S2). These biased distribution patterns of DEGs in three replicons of Sinorhizobium sp. NGR234 ( Figure 2) also suggested an active role of two plasmids in symbiotic adaptations to both determinate (V. unguiculata) and indeterminate (L. leucocephala) nodules. This is consistent with the view that rhizobial extrachromosomal elements are important in niche adaptations [40]. Moreover, V. unguiculata and L. leucocephala bacteroids shared a common subset of 2072 DEGs compared to free-living bacteria (Table S2), and 99.1% of these DEGs showed the same direction of regulation in both legumes hosts. Moreover, among these 2072 DEGs, 141/146 DEGs of pNGR234a, 391/ 523 DEGs of pNGR234b and 412/1403 DEGs of the chromosome were up-regulated in both V. unguiculata and L. leucocephala bacteroids. As shown in Table S3, the enrichment analysis revealed that up-regulated DEGs in bacteroids from both hosts were particularly enriched in the KEGG pathways for microbial metabolism in diverse environments (rhi01120), ABC transporters (rhi02010), nitrogen metabolism (rhi00910), fatty acid metabolism (rhi00071) and benzoate degradation (rhi00362). In contrast, down-regulated genes were enriched in the KEGG pathways for ribosomes (rhi03010), pyrimidine metabolism (rhi00240), flagellar assembly (rhi02040) and aminoacyl-tRNA biosynthesis (rhi00970). Despite these similar expression profiles, distinct transcriptional differences were also observed between bacteroids from V. unguiculata and L. leucocephala such as phenylalanine metabolism (rhi00360) (Table S2-S3 and see discussion below).
To validate the results of RNA-seq, we performed qRT-PCR on 13 genes with different expression profiles (Table S4). These include genes up-regulated or down-regulated in both L. leucocephala and V. unguiculata bacteroids with (or without) significant differences between expression levels in two hosts, and genes reversely regulated in bacteroids from L. leucocephala and V. unguiculata. As shown in Figure 3, the RNA-seq data agree well with the qRT-PCR data and Pearson correlation coefficient value was 0.934 (P,0.0001), despite the few differences that are often observed between qRT-PCR and microarrays results [10] [11] [41]. Similar results were obtained for qRT-PCR experiments with three independent biological replications (Table S4).

Growth Arrest of Bacteroids in Nodule Environments of V. unguiculata and L. leucocephala
Consistent with the non-growing state of nitrogen-fixing bacteroids, the Fisher exact test with Benjamini FDR corrected P values (two-tailed) revealed that the down-regulated genes in both L. leucocephala and V. unguiculata bacteroids (Table S3) were significantly (P,0.05) enriched in pathways for ribosomes (rhi03010), flagellar assembly (rhi02040), aminoacyl-tRNA biosynthesis (rhi00970) and pyrimidine metabolism (rhi00240) among others. Notably, all the flagellar assembly proteins except NGR_c28210 (the second copy of MotB) were down-regulated in bacteroids. This is in contrast to the induction of flagellar proteins in Bradyrhizobium japonicum by genistein application and the requirement of flagella in biofilm formation and competitive nodulation of S. meliloti [42,43,44,45]. However, it was also reported that nonflagellated mutants of S. meliloti formed normal nitrogen-fixing nodules on alfalfa [46], suggesting that flagella are dispensable for function of bacteroids.
Iron complex coding genes (NGR_b02510-NGR_b02530; NGR_b11200-NGR_b11230) were highly expressed in the free-   (Table 3). Moreover, a number of ABC transporters were significantly repressed in bacteroids such as transporter proteins of alpha-glucoside, trehalose/maltose, D-xylose, fructose, dipeptide, lipoprotein-releasing, lipopolysaccharide and peptides/nickel (Table 3). These expression patterns revealed a drastic transcriptional change of ABC transporters between bacteroids and free-living bacteria, suggesting the importance of compound exchange between two symbiotic partners. The observation of overrepresented DEGs in ABC transporters was also documented earlier for other rhizobialegume symbioses [52].

Secretion Systems
A remarkable number of secretion systems were found in NGR234 [30]. Among the six Type I transporter genes, tolC (NGR_c13520) and prsDE (aprDE, NGR_b10690 and NGR_b10700) were down-regulated in bacteroids (2-3 folds), whereas NGR_c30050, NGR_c30060 and NGR_c30070 were up-regulated in bacteroids (5.5-16.3 folds). Although the prsD mutant of R. leguminosarum was defective in nitrogen fixation on peas [53], the role of the Type I secretion system in NGR234 remains elusive.
NGR234 has two Type III clusters, and only T3SS-I locus (pNGR234a) was reported to modulate the nodulation of many legume hosts excluding L. leucocephala and V. unguiculata [30,54]. However, it was reported that all of the T3SS-I locus genes were induced by flavonoids and that nolB, rhcJ, nolU and nolV were detected in mature nodules of V. unguiculata [55]. In this study, 13/ 20 and 19/20 T3SS-I locus genes (NGR_a00520-NGR_a00700, NGR_a00790) were up-regulated in V. unguiculata bacteroids (2-to 11-fold) and in L. leucocephala bacteroids (6-to 172-fold), respectively, compared to the free-living condition. In line with differential expression patterns of T3SS-I locus genes in two hosts, nopP (NGR_a00570), nopX (NGR_a00700) and nopL (NGR_a00770) encoding effector proteins were specifically up-regulated in L. leucocephala nodules (RPKM = 273-476 versus RPKM = 13-27 in V. unguiculata nodules). Considering the presence of bacteria from infection zone of L. leucocephala nodules, our finding is to a certain extent consistent with the view that the stimulation of T3SS coincides with development of the infection thread [56]. Although 8/22 and 22/22 T3SS-II locus genes (pNGR234b, NGR_b22800-NGR_b23010) were up-regulated in V. unguiculata bacteroids (2-356 folds) and in L. leucocephala bacteroids (2-132 folds) respectively, the deletion of seven T3SS-II locus genes (NGR_b22890-NGR_b22950) up-regulated in both V. unguiculata and L. leucocephala did not show any defects in symbiosis [30]. The Type IV cluster (NGR_b10250-NGR_b10360) was constitutively expressed, but at a very low level in both the free-living condition and nodules (RPKM = 1-40).
In the free-living culture, NGR234 synthesizes primarily rough lipopolysaccharides (LPS) and only trace amounts of smooth LPS [58]. A new smooth LPS species with a modified lipid A-Core and rhamnan O-antigen is induced by flavonoid and is present in bacteroids of V. unguiculata [58,59,60]. In line with these observations, putative O-antigen biosynthesis protein coding genes NGR_b11970 and NGR_b14100 involved in lipid A biosynthesis were up-regulated in bacteroids of both legumes (2-to 7-fold).
Noteworthy the rgpF-rmlB gene cluster (NGR_a03500-NGR_a03580), necessary for the synthesis of rhamnan O-antigen [58,59,61], was expressed higher in bacteroids of L. leucocephala than in V. unguiculata bacteroids and in the free-living condition. It has been reported that this gene cluster is up-regulated in the following signaling pathways: NodD1RSyrM2RNodD2RFixF, and NodD1RTtsIRRmlB-WbgA [62,63]. These regulation patterns led to the hypothesis that the genes absolutely required for rhamnan production are expressed after the bacteria have entered the plant but before they are released into cortical cells of the nodules [61]. Thus, the higher expression of rgpF-rmlB in L. leucocephala bacteroids might be due to the mixture of bacteroids with those bacterial cells in infection threads, a distinct characteristic for indeterminate nodules. In contrast to these up-regulated genes involved in the biosynthesis of lipid A and O-antigen in nodules from either L. leucocephala or both legumes, genes associated to the core region biosynthesis, lpsB and kdtA (NGR_c04250, NGR_c15710), the synthesis and modification of lipid A, lpxABD, acpXL (NGR_c18080, NGR_c13420, NGR_c13440, NGR_c13460, NGR_c18030), unusual sugar (NGR_c12790, NGR_c35510, NGR_c04250), and LPS ABC transporters (NGR_c00180, NGR_c09020-NGR_c09030) were down-regulated in bacteroids from both legumes (2-to 11-fold). This suggested a general down-regulation for LPS production in bacteroids. Similarly, lpxD (NGR_c13420 homolog) involved in lipid A synthesis was down-regulated in R. etli bacteroids [41]. However, mutants in either acpXL down-regulated in bacteroids or members of the rgpF-rmlB cluster could not form pink nodules with V. unguiculata but efficiently nodulate on L. leucocephala [61,64]. These findings suggested that differences in nodule structures and associated characteristics between L. leucocephala and V. unguiculata may lead to the observed differential expression patterns of EPS and LPS biosynthesis genes. However, the symbiotic significance of these expression profiles, such as the observed expression of rgpF-rmlB cluster in L. leucocephala, may depend on their interactions with other cellular functions of bacteria and host responses.

Energy Metabolism
As expected, in microaerobic environment of nodules from both V. unguiculata and L. leucocephala, the nif genes directly involved in nitrogen fixation were strongly up-regulated. Symbiotic nitrogen fixation is a highly energy-demanding process [65]. In NGR234, there are two cluster of genes NGR_c22030-NGR_c22190 and NGR_c10480-NGR_c10630 encoding NADH dehydrogenase, and the former cluster was up-regulated in bacteroids while the latter was down-regulated. However, expression levels of the former genes (RPKM = 10-99) were quite low compared to the latter genes (RPKM = 150-707) in bacteroids. F-type H + -transporting ATPase coding genes (NGR_c31100-NGR_c31140 and NGR_c04470-NGR_c04500) were all down-regulated (2-to 6-fold). Both NGR_c25510-NGR_c25550 and NGR_c05230-NGR_c05300 encode cytochrome c oxidases, the former was up-regulated (11-to 41-fold, RPKM = 27-408) while the latter down-regulated (5-to 27-fold, RPKM = 15-249) in bacteroids. In NGR234, there are two clusters of genes encoding cbb3 oxidases, locus-I (NGR_c17970-NGR_c17990) and locus-II (NGR_c25780-NGR_c25810). Both loci were up-regulated in bacteroids, but showing a huge difference in expression levels: RPKM = 2453-3677 for locus-I and RPKM = 26-81 for locus-II. Interestingly, NGR_c01900 and NGR_c01910 encoding subunits of cytochrome bd ubiquinol oxidase were up-regulated in L. leucocephala (3-and 5-fold, respectively) but down-regulated in V. unguiculata (3-and 10-fold, respectively). However, it should be noted that expression levels of NGR_c01900 and NGR_c01910 in L. leucocephala were lower (RPKM = 49-51) than cbb3 oxidase locus-I (RPKM = 2453-3677). Although it was reported that Azorhizobium caulinodans uses both cytochrome cbb3 and bd as terminal oxidases for symbiotic nitrogen fixation [66], cbb3 oxidase could be the major terminal oxidase for symbiotic nitrogen fixation in bacteroids of both L. leucocephala and V. unguiculata. It has been reported that the constitutive expression of the thiamine biosynthetic pathway caused the production of cbb3 oxidase in the free-living condition and an increased capacity in nitrogen fixation during symbiosis [67,68]. In this study we presented additional evidence that thiCOGE (NGR_b02900-NGR_b02930) and thiD (NGR_b18410) were up-regulated (2-to 13-fold) in bacteroids from nodules of both L. leucocephala (RPKM = 26 -46) and V. unguiculata (RPKM = 44 -202), with a higher expression level in V. unguiculata bacteroids. This finding further supports the view that the thiamine biosynthesis genes are commonly expressed extrachromosomal genes in association with plants [40].
NGR_b03260-NGR_b03300 involved in pyrroloquinoline quinone (PQQ) biosynthesis were specifically activated in L. leucocephala bacteroids (RPKM = 169-580 versus RPKM = 4-19 in V. unguiculata bacteroids and the free-living form). PQQ has been found as a redox cofactor for membrane-bound dehydrogenases [69]. The presence of PQQ-dependent glucose dehydrogenase was also reported in rhizobia such as Rhizobium tropici and S. meliloti etc. [70]. The PQQ-linked glucose dehydrogenase has also been demonstrated as a requirement by S. meliloti for optimal nodulation efficiency and competitiveness on alfalfa roots [71]. Close to these pqq genes in NGR234 genome, NGR_b03250 encoding a periplasmic alcohol dehydrogenase with ferricytochrome c as the acceptor was specifically up-regulated in L. leucocephala bacteroids (RPKM = 1367 versus RPKM = 20 in V. unguiculata bacteroids and RPKM = 29 in the free culture). Moreover, NGR_b03210-NGR_b03240 within the same loci showed a similar expression trend as NGR_b03250-NGR_b03300, suggesting their potential role in adapting to the nodule environment of L. leucocephala. Proteins encoded by NGR_b03210-NGR_b03240 include a pseudoazurin, a signal transduction histidine kinase, a FIST containing signal transduction protein [72] and a LuxR family response transcriptional regulator. It would be interesting to study whether these signal transduction systems could regulate the expression of PQQ genes. It has also been shown that PQQ could work as an antioxidant protecting bacteria from oxidative damage, or as a nutrient to support bacterial growth [69]. However, the role of PQQ in rhizobium-legume symbiotic interactions remains largely unknown.

Role of Nitric Oxide
Recently, nitric oxide (NO) has been detected at different steps of the symbiosis between legumes and rhizobia [73,74,75]. Modulation of NO levels was demonstrated to be involved in the establishment and persistence of the symbiosis [73,76,77]. In this study, the up-regulation of NapABC (nitrate reductase, NGR_c10020-NGR_c10040, 10-to 15-fold) and NirK (nitrite reductase, NGR_c09950, 37-to 56-fold), and the down-regulation of NorC (nitric oxide reductase, NGR_c09850, 2-to 3-fold) suggested that bacteroids contributed to the NO pool within nitrogen-fixing nodules from both L. leucocephala and V. unguiculata. In line with this finding, in M. truncatula nodules formed by napA or nirK mutant of S. meliloti, the production of NO was decreased by about 35% compared with that of the wild-type control [78]. It was reported that cytochrome c550 is required for the succinatedependent nitrite reduction and might be involved in electron transfer to the copper-containing nitrite reductase of B. japonicum [79]. In this study, NGR_b03130 encoding cytochrome c550 was specifically up-regulated in L. leucocephala nodules (RPKM = 186 versus RPKM = 5-6 in V. unguiculata nodules and in the free-living condition). In line with this, NGR_b03210, encoding a pseudoazurin which was demonstrated to be an electron donor to the copper-containing nitrite reductase in other denitrifying bacteria [80], was also strongly up-regulated in L. leucocephala nodules (RPKM = 449) compared to V. unguiculata nodules (RPKM = 20) and the free-living condition (RPKM = 11). Although cytochrome c550 is not required by B. japonicum for nitrogen-fixation in determinate nodules of G. max [81], potential roles of cytochrome c550 and/or pseudoazurin in indeterminate nodules such as L. leucocephala nodules are still unknown.
A considerable number of DEGs encoding enzymes that participate in glyoxylate and dicarboxylate metabolism were also found (Table S3), suggesting a potential role of C2 metabolism in symbiosis. Recently, it was reported that this glyoxylate cycle of R. leguminosarum was more strongly induced in the rhizosphere of pea, the compatible legume host of this bacterium, than in that of alfalfa and sugar beet [82]. Moreover, the coincident expressions of glyoxylate cycle genes and a subset of nif genes in the chemoautotrophic culture of B. japonicum [84] imply potential coordination in transcriptions of these genes. However, S. meliloti mutants of two principle genes, aceA (encoding isocitrate lyase) and glcB (encoding malate synthase), in the glyoxylate cycle were not impaired in nodulation and nitrogen fixation on alfalfa [85]. NGR_c03920-NGR_c03950 expressed higher in V. unguiculata bacteroids (RPKM = 254-401) than in L. leucocephala bacteroids (RPKM = 55-105) and the free-living condition (RPKM = 24-40). They encode glycolate oxidase which catalyses the oxidation from glycolate and O 2 to glyoxylate and H 2 O 2 . The glcD (NGR_c03920 homolog) mutant of A. caulinodans formed Fixnodules on S. rostrata [86]. On the other hand, it has been reported, in S. meliloti-Medicago symbiosis, an optimal level of H 2 O 2 is required in the normal progression of infection threads and the efficient release of bacteria into nodule cells [87,88]. In this study, NGR_b11000 encoding catalase C (peroxidase) was up-regulated in both V. unguiculata bacteroids (RPKM = 19) and L. leucocephala bacteroids (RPKM = 48) compared to the free-living form (RPKM = 4). However, the symbiotic roles of glycolate oxidase and products from its activation in bacteroids remain elusive. Despite the unclear picture of the symbiotic role of glyoxylate cycle, there were large amounts of acetate and fatty acids in the nodules [89]. On the other hand, as shown in Table 3S, the down-regulation of fatty acid synthesis genes and the up-regulation of fatty acid metabolism genes in bacteroids might be related to the presence of fatty acids in nodules.
As shown in Figure 1E and 1F, PHB granules were observed in bacteroids of nodules from both L. leucocephala and V. unguiculata. In line with this observation, the acetyl-CoA acetyltransferase coding gene, phbA (NGR_c32720, 3-fold) and the poly-beta-hydroxybutyrate polymerase coding gene phbC1 (NGR_c34290, 4-and 7-fold, respectively) were up-regulated in both legume hosts while phbC2 (NGR_c14000) and phbB (NGR_c32710) were constitutively expressed. Noteworthy, phbZ encoding a polyhydroxybutyrate depolymerase (NGR_b03370) was up-regulated by 19-fold and 9fold in L. leucocephala and V. unguiculata, respectively. bdhA2 encoding a 3-hydroxybutyrate dehydrogenase (NGR_c23850) was constitutively expressed. Therefore, PHB level is subject to strict modulation in bacteroids. It has been reported that PHB was not accumulated in the free-living culture of R. etli when biotin was added [90]. The transcription of bdhA in free-living S. meliloti was higher in the presence of added biotin [91]. These earlier reports suggested biotin-induced PHB degradation in the free-living condition. However, bioABDF involved in biotin synthesis were all up-regulated in L. leucocephala (3-to 9-fold) but down-regulated in V. unguiculata nodules (3-fold), indicating a potential complex regulation on PHB level in bacteroids.
Notably, PHB granules in L. leucocephala bacteroids were larger than those in V. unguiculata bacteroids ( Figure 1E and 1F). Phasins encoded by phaP genes could bind to PHB granules and promote PHB synthesis by regulating the surface/volume ratio of PHB granules or by interacting with the polyhydroxyalkanoate synthase [92]. It was also reported that phaP1 and phaP2 in S. meliloti contributed to PHB accumulation in bacteroids and symbiotic nitrogen fixation [93]. NGR234 has three phaP homologs, phaP1 (NGR_c03360), phaP2 (NGR_c13240), phaP3 (NGR_a00900). As shown in Table 4, phaP3 seemed to be specifically and strongly expressed in bacteroids of both legume hosts while phaP1 and phaP2 were up-regulated in L. leucocephala bacteroids (1.8-fold) and phaP1 was down-regulated in V. unguiculata bacteroids. Therefore, these expression patterns of phaP homologs agree well with the difference in PHB granule size observed in L. leucocephala and V. unguiculata bacteroids.

Conclusions
Although Sinorhizobium sp. NGR234 is well known for its broadhost range and the ability of forming either determinate or indeterminate nodules on corresponding legumes [24], genomelevel transcriptomic adaptions to determinate or indeterminate nodules of different legumes have not been investigated for this strain before. By using RNA-seq, we conducted a high-resolution transcriptomic analysis for NGR234 bacteroids in indeterminate nodules of L. leucocephala and in determinate nodules of V. unguiculata. Both common and distinct transcription patterns were uncovered for NGR234 bacteroids in these two non-model legumes. These shed new light on the mysterious mechanisms of rhizobial adaptations to diverse legume hosts.