Transcriptional Profiling of Nitrogen Fixation and the Role of NifA in the Diazotrophic Endophyte Azoarcus sp. Strain BH72

Background The model endophyte Azoarcus sp. strain BH72 is known to contribute fixed nitrogen to its host Kallar grass and also expresses nitrogenase genes endophytically in rice seedlings. Availability of nitrogen is a signal regulating the transcription of nitrogenase genes. Therefore, we analysed global transcription in response to differences in the nitrogen source. Methodology/Principal Findings A DNA microarray, comprising 70-mer oligonucleotides representing 3989 open reading frames of the genome of strain BH72, was used for transcriptome studies. Transcription profiles of cells grown microaerobically on N2 versus ammonium were compared. Expression of 7.2% of the genes was significantly up-regulated, and 5.8% down-regulated upon N2 fixation, respectively. A parallel genome-wide prediction of σ54-type promoter elements mapped to the upstream region of 38 sequences of which 36 were modulated under the N2 response. In addition to modulation of genes related to N2 fixation, the expressions of gene clusters that might be related to plant-microbe interaction and of several transcription factors were significantly enhanced. While comparing under N2-fixation conditions the transcriptome of wild type with a nifLA− insertion mutant, NifA being the essential transcriptional activator for nif genes, 24.5% of the genome was found to be affected in expression. A genome-wide prediction of 29 NifA binding sequences matched to 25 of the target genes whose expression was differential during microarray analysis, some of which were putatively negatively regulated by NifA. For selected genes, differential expression was corroborated by real time RT-PCR studies. Conclusion/Significance Our data suggest that life under conditions of nitrogen fixation is an important part of the lifestyle of strain BH72 in roots, as a wide range of genes far beyond the nif regulon is modulated. Moreover, the NifA regulon in strain BH72 appears to encompass a wider range of cellular functions beyond the regulation of nif genes.


Introduction
Biological nitrogen fixation, involving the enzymatic conversion of atmospheric nitrogen to ammonia by nitrogenase, is an important process to maintain soil fertility and life on earth, counterbalancing loss of nitrogen e.g. by denitrification. It is catalyzed by the two-component nitrogenase complex restricted to prokaryotes, the reaction being oxygen-sensitive demanding high amounts of energy and reductants. Nitrogen-fixing microorganisms encompass a broad habitat range from free living forms in soils and water to endophytic association with grasses or in rootnodule symbioses with legumes. Consequently, they have evolved sophisticated regulatory networks that respond to multiple environmental cues [1]. Regulation of nif gene expression has been most extensively studied in diazotrophic Proteobacteria [2]. Albeit the genes necessary for nitrogen fixation in many diazotrophs have common structures and functions, the mechanisms by which cellular nitrogen levels are sensed and nitrogen signals are transmitted can vary considerably among different nitrogen-fixing bacteria [3]. Oxygen and fixed nitrogen, such as ammonium, are the important environmental signals that regulate nitrogen fixation.
Azoarcus sp. strain BH72 is a diazotrophic model endophyte of grasses [4], belonging to Betaproteobacteria [5]. N 2 fixation and nifHDK transcription occur only under microaerobic and nitrogenlimiting conditions in this strictly respiratory bacterium [6,7]. The structural genes of nitrogenase nifHDK were even found to be expressed and translated in the aerenchyma of rice seedlings [6,8], and fixed nitrogen can be contributed in an unculturable state to the host plant Kallar grass [9].
Azoarcus sp. strain BH72 is apparently highly adapted to environments poor in available nitrogen sources, which correlates with its role as an N 2 -fixing endophyte. (i) A low-affinity glutamate dehydrogenase (GDH) for ammonium assimilation is lacking, a feature highly unusual in free-living bacteria; only the high-affinity ATP-consuming assimilation system (GS[2x]-GOGAT) is present [10]. (ii) Four genes encoding putative high-affinity ammonia transporters exist (amtB/Y/D/E), one of them with an additional regulatory domain [11]. (iii) Single copies of structural genes for the molybdenum-dependent nitrogenase complex and all genes required for cofactor synthesis and maturation of the nitrogenase are present in strain BH72 [10]. (iv) A battery of electron transporters which might contribute to nitrogenase activity is encoded in the genome, including two flavodoxins (nifF1, nifF2), 12 genes for ferredoxin-like proteins, and two gene clusters for putative electron transport systems to ferredoxins (rnf1, rnf2) [11]. Thus, nitrogen fixation appears to be an important part of the lifestyle of this bacterium, however genome-wide expression profiles are not known under these conditions.
In Azoarcus sp. strain BH72 and other nitrogen-fixing symbiotic bacteria, RpoN and NifA are master regulators of nitrogen fixation genes. The alternative sigma factor 54 (RpoN) is employed by many bacteria to transcribe genes involved in a wide variety of cellular functions such as nitrogen utilization [12], virulence, stress responses [13] and flagellum biosynthesis [14]. RpoN binds to -24/212-type promoters with consensus sequence TGGCACG-N4-TTGC [15,16]. However, for initiation of transcription it requires additionally an Enhancer Binding Protein (EBP), such as the transcriptional regulator NifA [1,17]. Also in Azoarcus sp. strain BH72, NifA plays a central role as transcriptional activator for nitrogenase gene (nifHDK) expression [18]. As otherwise only reported for Gammaproteobacteria [19], in strain BH72 an additional regulatory protein NifL is involved as an anti-activator of NifA that regulates its activity in response to oxygen and combined nitrogen [18]. However, the entire regulon of the NifLA system has not yet been determined.
In order to characterize the nitrogen response of Azoarcus sp. strain BH72, we used a genome-wide microarray [20] to analyze the transcriptome. Besides for typically expected nif genes, modulation of expression was observed under nitrogen fixation versus N-replete conditions for several non-nif genes as well as genes encoding for hypothetical protein(s). In a parallel genome wide in silico approach, several corresponding RpoN and NifA binding sites were predicted. Transcriptome profiling in a nifLA insertional mutant revealed a relatively large regulon indirectly or directly affected by NifLA.

Results and Discussion
General Nitrogen Response of Strain Azoarcus sp. Strain BH72 In the order to analyze the nitrogen response of Azoarcus sp. strain BH72, wild type cells were grown under microaerobic conditions either on N 2 or on combined nitrogen (ammonium chloride). The global gene expression profile was compared by oligonucleotide-based transcriptome microarray studies. A total of 524 (13.1%) genes exhibited more than a 1.8 fold change in expression under nitrogen fixation in comparison to N-replete conditions (P,0.05). This cut-off was used as it allowed better coverage of cotranscribed genes, some of which were only moderately regulated. Among these genes, the expression of 290 (7.2%) genes was enhanced and the expression of 234 (5.8%) genes was repressed (see Table S1, Figure 1) under N 2 fixation. The genome of strain BH72 harbors 55 genes (azo0510 to azo0564) located in a nif cluster encoding for proteins directly or indirectly involved in nitrogen fixation according to their annotation [10], accounting only about 1.3% of the protein encoding genes ( Figure 2). Therefore, the N 2 response in strain BH72 elicited the expression of a 10-fold higher number of genes even outside the nif cluster. Selected results obtained by the microarray approach were further examined by real-time RT-PCR. Similarly, 549 out of 4146 predicted genes had enhanced expression under nitrogen fixation in Pseudomonas stutzeri A1501, accounting about 13% [21]. For Azotobacter vinelandii, transcript levels for even 30% of the genes changed more than two-fold during diazotrophic growth compared to the N-replete control [22].
Genes modulated under N 2 fixation were evenly distributed across most general COG functional categories (Clusters of Orthologus Group(s)), however the group belonging to ''not in  Table S1). Accordingly, 28% of the up-regulated genes and 12% of downregulated genes under N 2 fixation belonged to this group. Even several gene clusters with unknown functions exhibited enhanced expression under N 2 fixation, such as cluster azo1297-azo1310 or cluster azo3470-azo3474 ( Figure 3, Table S1). Interestingly, recently it has been shown that some of them encode components of a type VI secretion system (see below), [23].
The group of metabolism-related genes represented an also frequently modulated COG category. However, the majority of the members were down-regulated under N 2 fixation, particularly those belonging to carbohydrate, coenzyme, lipid and secondary metabolite metabolism subgroups, respectively ( Figure 1). COG U represented by proteins involved in secretion was also strongly overrepresented under N 2 fixation. Interestingly, genes from the subcategory L representing replication, recombination and repair proteins were mainly up-regulated under N 2 fixation, while members representing transcription (K) as well as translation, ribosomal structure and biogenesis subfamilies (J) were mainly found to be down-regulated ( Figure 1).

Differential Expression of Genes Related to Nitrogen Fixation
Several gene clusters were found to be significantly up-regulated under N 2 fixation, indicating functional relationships. They can be divided into two broad categories. One of them included genes which are directly or indirectly related to nitrogen fixation represented by gene clusters involved directly in nitrogenase synthesis, maturation and function. As expected, up-regulation of most genes of the nif cluster (azo0512-azo0562) was detected, with genes encoding proteins involved in nitrogenase maturation, nitrogenase enzyme and electron transporters ( Figure 2; Table  S1). Although nifH up-regulation was not found to be statistically significant due to spot inhomogeneity, it was shown previously that the expression of nifH, fdxN, and nifLA were elevated under N 2 fixation [11,24]. Interestingly, specifically the components of rnf1complex encoded within the nif cluster and not genes of the rnf2 complex exhibited enhanced expression under N 2 fixation. The Rnf1 complex has been reported to couple the energy of ion transport to reduce ferredoxin, and in strain BH72 it appears to play a role in electron transfer to nitrogenase and in regulation of the ''switch-off'' of nitrogenase in response to ammonia [11]. In addition, nif genes outside the nif cluster and in distant locations of the chromosome were differentially regulated too; nifF1 (azo0014) and nifZ (azo3367) along with fer22 (azo3368) (related to nitrogenase maturation) were also induced 2.7, 2.6, 11.5 fold, respectively ( Figure 3, Table S1). Likewise, genes encoding for proteins related to efficient N 2 fixation electron transfer, ATP synthesis and molybdopterin biosynthesis, an important component of FeMo-co, exhibited enhanced expression under nitrogen fixation, respectively (Table S1). Accordingly, genes for molybdenum transporters like modA1 and modE were up-regulated. Also several genes coding for electron transfer flavoproteins like etf1, etfA2, etfB2, etfA3, etfB3, flavodoxin isiB, probable cytochrome cc42 and even nifY2 (nitrogenase maturation protein) were upregulated. Recently it has been reported that NifB and NifEN protein levels are regulated by protease ClpXP under N 2 fixation conditions in A. vinelandii [25]. This ClpX protein of A. vinelandii has highest similarity (76% identical, 84% similar) to the respective ClpX copy (azo2070) of Azoarcus sp. strain BH72. Concordantly, clpX and clpP expressions were enhanced by 2 fold in strain BH72 under N 2 fixation, as well ( Figure 3).

Nitrogen Response of Gene Clusters Not Related to Nitrogen Fixation
Several of the gene clusters differentially regulated in response to nitrogen fixation conditions encoded proteins that were not obviously related to nitrogen metabolism but with completely different functions. This may in part reflect the difference in growth conditions and generation times, although they do not vastly differ (2.1060.09 on N 2 , 1.6660.13 on NH 4 + ) [26]. Transport and secretion: Several secretion clusters were found to be induced under N 2 fixation in strain BH72. In agreement to the up-regulation of genes in COG group U (Figure 1), a gene cluster encoding for putative components of a type II secretion system (azo0801-azo0805) components, was strongly up-regulated under N 2 fixation ( Figure 3; Table S1). For confirmation of differential gene expression in unexpected cases such as this (see also below), an independent method was used. Quantitative RT-PCR analysis validated the induced expression of azo0805 by even 21.465.7 fold under N 2 response. Proteins transported by the type II secretion pathway have to first translocate across the cytoplasmic membrane via the Sec system and then fold into a translocation-competent conformation in the periplasm. Secreted proteins may include proteases, cellulases, pectinases, phospholipases, lipases, and toxins [27]. This appeared to be a relatively specific induction, as a second type II secretion gsp gene cluster (azo2097-azo2084) was not affected. Another type of protein secretion system was induced under N 2 fixation. This cluster was characterized to encode components of type VI secretion system (azo1299 to azo1307) [23] (Table S1). Type VI secretion system gene clusters contain from 15 to more than 20 genes, two of which encode Hcp and VgrG, that are nearly universally secreted components of the system [28,29]. As in several symbiotic or pathogenic interactions with eukaryotes [30], the type VI protein secretion appears to play a role also in Azoarcus-rice interactions [23].
Nitrogen and carbon metabolism: As expected, in response to N 2 fixation the transcription of glnA (encoding glutamine synthetase) was enhanced to assimilate the fixed nitrogen, glnP and glnH (glutamine transporters) were induced, as well (Table S1). In contrast, the transcription of denitrification genes like napD1 and napE (Table S1), as well as a gene for NO 2 2 assimilation nirB (Table S1), were strongly repressed under N 2 fixation. Interestingly, expression of hoxB encoding for the small subunit of hydrogenase was enhanced by 6.57 fold (Table S1) specifically under N 2 fixation. It may be tempting to speculate that H 2 liberated as a by-product under N 2 fixation by nitrogenase can be further oxidized by an uptake hydrogenase to recycle reducing equivalents. N 2 fixation in Azoarcus sp. strain BH72 was accompanied by down-regulation of the expression of several of the Embden-Meyerhof pathway (EMP) and tricarboxylic acid cycle (TCA cycle) enzymes. Especially the transcription of the components of PDH complex as well as components of ethanol oxidation pathway producing acetyl-CoA was reduced. Expression of some enzymes of the ß oxidation of the fatty acid pathway generating acetyl-CoA as end product, were also found to be transcriptionally repressed under N 2 fixation. Perhaps occurrence of high energy charge (ATP/ADP) in N 2 fixing cells inhibits some of the enzymes and also their expression or this might reflect slower growth.
Translation and transcription: Expression of gene coding for small subunit of ribosomal protein (azo3394) as well as translation initiation factors and elongation factors (azo3419; azo3431) were down-regulated under N 2 fixation (Table S1), which might reflect the slower growth rate. This is in agreement with the generation time measured for Azoarcus sp. strain BH72 under N 2 fixation (2.10 h) as compared to that in presence of ammonia (1.12 h). The fold of repression (2.9-fold) of one of the ribosomal protein encoding genes, azo0720 under nitrogen response even corroborated with qRT-PCR approach (22.561.5 fold) under similar condition (Table 1). Consistently, expression of genes encoding DNA directed RNA polymerase subunits, rpoC, rpoB, and more general sigma factors rpoH (sigma32), and rpoD (sigma70), as well as several transcription factors like gacA and uidR, were strongly repressed under N 2 fixation. In contrast, expression of algU (rpoE) (sigma 24) and rpoN2 (sigma 54), but not rpoN1, were strongly enhanced under N 2 fixation (Table S1). Among these two copies in the genome of strain BH72, rpoN1 (azo0504) is located in proximity to the nif cluster, and rpoN2 (azo1790) widely distant from the nif cluster. A few organisms may have two copies of rpoN, such as Bradyrhizobium japonicum, Rhizobium etli, [31,32], Ralstonia solanacearum and Burkholderia fungorum, while Rhodobacter sphaeroides harbors 4 copies [33]. Generally RpoN1 is involved in the expression of the genes required for nitrogen fixation, whereas RpoN2 is required for the transcription of the class II and class III flagellar genes. The significant induction of azo1790 by 3.9 -fold (Table S1) was also verified by real time qRT-PCR (17.8362.94 fold) ( Table 1). In Rhizobium etli, rpoN1 expression was negatively autoregulated under aerobic growth and reduced during microaerobiosis and symbiosis while rpoN2 expression was specifically induced under microaerobiosis and in bacteroids [32]. R. sphaeroides harbours 4 copies of rpoN gene. rpoN1 located within nif cluster was down-regulated under shift to aerobic conditions while rpoN2 located with the flagellar genes was transiently up-regulated after shift to aerobiosis [34]. Additionally, a high number of transcription factors belonging to Fis, LysR, TetR, MerR, AraC, and LuxR families respectively were modulated in their expression: 10 were upregulated and 11 down-regulated; one of them, azo1584 encoding a hybrid sensor-response regulator, exhibited even a relatively high 6.34 fold enhanced expression on N 2 ( Figure S4). Thus, complex changes in transcriptional regulation appear to occur under conditions of nitrogen fixation. In Pseudomonas stutzeri, expression of only 6 transcription factors of the LysR, TetR or AraC family were up-regulated under N 2 fixation [21]. In Azotobacter vinelandii, an Fnr like negative transcriptional regulator of CydAB was up-regulated while IscR (Fe-S assembly Factor) was down-regulated under all conditions of N 2 fixation with nifH, vnf, and anf [22]. cluster encoding for several hypothetical proteins, the first gene in the cluster being vgrG (component of type VI secretion); cluster of genes encoding for proteases; cluster harboring genes involved in nif maturation, nifZ, ferredoxin like mobile electron carrier fer22, and component of Molybdenum transport modD; closely linked genes within the nif cluster encoding for FeMoco cofactor biosynthesis protein, hypothetical protein, sigma E factor regulatory protein and probable ferrodoxin. (B) Cluster down-regulated under N 2 fixation; genes encoding for proteins/enzymes involved in phenylacetic acid degradation pathway (paa). Fold change represented by shades of grey give the average fold of induction from 3 independent microarray experiments in each case. Black bar with round cap, symbols or Ø, represent intergenic transcription termination loop, putative RpoNbinding sites or putative NifA-binding sites, respectively, upstream of the target gene of interest as described previously in the legend of Secondary metabolism: Flagella and pili are important for microbial colonization of the host, particularly for endophytic colonization of rice roots by Azoarcus sp. strain BH72 [35,36,37]. Enhancement of expression of pilH, pilY1B and fliF under diazotrophic growth (Table S1) speaks in favor of a common mode of gene regulation. Indeed several of the pilin as well as flagellar genes are known to be under the common control of RpoN along with the nif related genes in other bacteria. One of the interesting observations under N 2 fixation was the strong downregulation of proteins involved in phenyl acetic acid (PAA) degradation pathway (paa-cluster) (Table S1, Figure 3). In Azospirillum brasilense, PAA has been reported as an auxin like molecule with anti-microbial activity [38]. It has been reported to play a role in plant growth promotion and protect the producing strain from other competing strains in natural environments. Recently, connections between virulence and gene products of the phenylacetate catabolism have been shown in Burkholderia cenocepacia [39]. The accumulation of the early products of PAA catabolism is known to have toxic effects on the host showing disease symptoms. Ring-1,2-epoxide and its phenolic breakdown product 2-hydroxyphenylacetate are obvious candidates for causing such damage [40]. PAA might play a role in rhizosphere competence of Azoarcus sp. strain BH72 outside the root, and down-regulation of the PAA degradation pathway under N 2 fixation might be beneficial for strain BH72 to establish a successful endophytic colonization as a ''disarmed plant pathogen'' inside rice root without showing signs of disease. Thus, modulation of expression of genes related to plant colonization might link the diazotrophic with the endophytic life style.

Genome Wide Prediction of 212/224 Promoters for s 54 Factor (RpoN) Binding and of NifA Binding Sites
The RpoN regulon was analyzed in several organisms, such as E. coli [41], Pseudomonas putida [42] and several species of Rhizobiaceae [43] by use of powerful computational methods that took advantage of the high conservation of s 54 -type promoter sequences throughout diverse bacterial groups. To potentially discriminate between genes directly and indirectly regulated by RpoN and to identify other members of the RpoN regulon undetected by microarray analysis, we carried out an in silico search to locate potential RpoN-binding sites in Azoarcus sp. strain BH72 genome using the online tool for the prediction of prokaryotic promoter elements and regulons, PePPER: a web based regulon, TF (Transcription Factor), and TFBS (Transcription Factor Binding Site) mining system. The intergenic regions of the complete genome sequence of strain BH72 were scored against position-specific weight matrices (PWM) derived from selected RpoN-binding sites of bacterial species listed in the TFBS tool box of PePPER. With the default setting parameters of the online analysis tool, we were initially able to detect genome wide 173 putative RpoN-binding sites or s 54 -type promoters upstream of 162 target genes that could potentially direct the transcription of a gene in the correct orientation. The binding sequences of Azoarcus sp. strain BH72 were assembled together to generate a genome wide RpoN-binding consensus using Web logo ( Figure S1). Restricting the list to those target genes which were also modulated in their expression under N 2 fixation in microarray experiments, 38 RpoN-binding motifs were predicted to be present upstream of 36 target genes (Table S2). This probably comprised the subset of the whole RpoN regulon involved in controlling functions related to N 2 fixation accounting to 22% of all genes with predicted RpoN binding motif. Among them, genes nifB (azo0523) and narK (azo1288) were with tandem promoter sites as also detected for example for P. putida Ca-3 [44]. RpoN binding motifs were predicted upstream of nifH and nifL, respectively, which have been previously shown experimentally to be modulated under N 2 fixation [18,24]. Hence they were additionally included in the list as well, although modulation of nifH and nifL expression was not detected under current microarray experiments (Table S2). Similarly a candidate gene pilA, which is not related to N 2 fixation but known to possess a RpoN-binding motif [35], was detected by the online analysis tool, speaking in favor of its robustness. As expected, most of the listed genes belonged to N 2 fixation related functions: structural gene of nitrogenase nifH (azo0538), maturation process of nitrogenase like nifB (azo0523), nifE (azo0562) or electron transport like rnfA1 (azo0517), nifF1 (azo0014), fer22 (azo3368) or transcription factors rpoN2 (azo1790), azo2932. Others included those that are involved in nitrogen assimilation, glnA (azo0738), nitrate transport narK (azo1288) or even outer membrane efflux proteins as nodT (azo2554). Outside N 2 fixation related functions, genes encoding for hypothetical proteins with enhanced expression under N 2 fixation were also found to be preceded by RpoN-binding motifs. This included 17 different hypothetical protein encoding genes like azo2488, azo2922, azo2958, azo3080. 5 out of 17 encoded even secretory proteins (azo2346, azo3866, azo3738, azo3114, and azo3866).
Additionally, ethanol metabolism, with expression of the aldehyde dehydrogenase gene aldA (azo2939) [45] being repressed under N 2 fixation, was also found to be linked with an RpoN binding site. Additionally, binding sites for NifA, the transcriptional activator for nif genes, were predicted using the consensus TGT-N 10 -TCA in the intergenic region of the Azoarcus sp. strain BH72 (Prodoric). Sequences were searched in both the strands, relative to the context of the gene start spanning up to 500 bases. Sequences generated with a score above -6.5 with their associated target genes exhibiting NifA-regulated differential expression pattern in microarray experiments were chosen for further analysis. In this way, 29 NifA binding sequences were predicted upstream of 25 target genes (Table S3). Tandem NifA binding sequences were found to be present upstream of rnfA1, nifL, sodC and nifE, respectively (Table S3). 18 of the listed genes were positively regulated while 7 were negatively regulated in the presence of NifA. 11 out of 25 target genes were up-regulated under N 2 fixation, as well, and 8 of them also possessed an upstream RpoNbinding motif. Correspondence to expression data is discussed below.

Characterization of the NifLA Regulon: Transcriptional Activator for Nitrogen-fixation Related Genes
The transcriptional activator for nifHDK genes in Azoarcus sp. strain BH72 is the NifA protein, whose activity is known to be modulated by the anti-activator NifL in response to oxygen and combined nitrogen under free-living conditions [18]: the nifLA mutant BHLAO is unable to grow on N 2 . The NifA regulon has been analyzed only in root nodule endosymbionts up to now, such as Bradyrhizobium japonicum [46], Sinorhizobium meliloti [47] and Rhizobium etli [48]. In their regulatory network leading to symbiotic nitrogen fixation, low oxygen concentration is a major cue affecting NifA activity.
In order to globally assess the NifA regulon of strain BH72, we compared the transcriptome of wild type cells with that of the nifLA mutant strain BHLAO, both grown microaerobically under conditions allowing nitrogen fixation in presence of glutamate. The generation time of both strains was similar (wild type 2.7 h, BHLAO 2.6 h). The nifL::V strain (BHLAO) was previously constructed by marker-exchange mutagenesis, carrying a polar mutation in the nifL gene by insertion of a Sp/Sm-cartridge which also abolished nifA transcription [18]. BHLAO was not able to grow on N 2 in the absence of combined nitrogen under microaerobic condition as NifA was essential for diazotrophy as transcriptional activator. It was also shown previously that nifA expression was undetectable in presence of combined nitrogen in strain BHLAO. To further verify that nifA expression is completely abolished in strain BHLAO under conditions of strong induction, total RNA was isolated from bacteria grown under N 2 fixing condition in presence of 10 mM glutamate. As expected, a 0.421 kb amplification product of nifA was detected by RT-PCR only in wild type strain BH72 ( Figure S2A) and not in strain BHLAO. Neither of the negative controls, without an RT step or template, respectively, generated amplification products. ( Figure   S2A). The use of equal amounts of RNA in both extracts was confirmed by semi-quantitative RT-PCR using primers for amplification of 16S rRNA ( Figure S2B).
The number of genes modulated in expression exceeded the number of genes modulated under N 2 fixation. 996 genes differed in expression in the nifLA mutant in comparison to the wild type; 587 genes were positively regulated while 409 genes were negatively regulated (Table S1, Figure 4). The respective COG categories revealed that the putative regulon encompassed a broad range of functional categories; nearly all functional categories housed a fair number of candidates ( Figure S3). This might partially be attributed to different growth conditions: these experiments had to be carried out with glutamate as a nitrogen source that does not repress nif genes, in contrast to the experiments comparing N 2 -fixing and ammonium-grown wild type cells without glutamate addition. Therefore, we analyzed the effect of addition of glutamate on the transcriptome of wild type BH72 under conditions of nitrogen fixation. The glutamate response affected only transcription of 6% (Table S1) of the genome, mostly at moderately high levels, compared to the BNF response (13%) or the NifA response (24%). Moreover, the genes modulated under the NifA response independent of nitrogen response were mostly independent of the glutamate response (see below). Therefore, metabolic effects of glutamate might not fully explain the strong impact of absence of NifA. This indicated a more global nature of NifA regulation beyond nif-gene regulation. The NifA regulon in rhizobia was also found to be relatively broad, with 323 genes differentially regulated under microaerobic, free living conditions in B. japonicum [46], or even 601 genes in symbiotic S. meliloti cells [47].
As expected for NifA as the essential transcription activator for nif genes, the expression of nif structural genes as well as several of those involved in nitrogenase maturation were up-regulated under   (Figure 2). Also nif related genes outside the nif cluster like azo3366 and azo3367 (nifZ) were found to be positively regulated by NifA (Table  S1). Accordingly genes for electron transport complexes like Rnf1 (azo0512 to azo0517) or genes encoding for mobile electron carriers like flavodoxin nifF1 (azo0014) and ferredoxins like fer22 (azo3368) were positively regulated by NifA under conditions of nitrogen fixation ( Figure 5A, upper panel). Concordantly, RpoN and NifA binding sites were predicted in the upstream promoter region of azo3368 and nifF1, respectively. Data from microarray experiments were corroborated with expression data obtained from real time PCR analysis: qRT-PCR analysis showed that indeed azo3368 was strongly induced (90.7615.4 fold) under N 2 response and positively regulated (19.968.6 fold) by NifA ( Table 1). As has been shown for R. leguminosarum [49], hoxB (azo3807) encoding for hydrogenase was also under NifA control in strain BH72. Proteases encoded by clpX (azo2070) and clpP (azo2071) upregulated under N 2 -fixation were also found to be under NifA control, as in Azotobacter vinelandii [25]. As all genes of the operon were positively regulated by NifA and not affected by glutamate, this was not likely a metabolic effect.
Only 2 out of 10 transcriptional regulator genes, gcv (azo3618) and azo0625, up-regulated under N 2 -fixing conditions, were indeed positively regulated by NifA, albeit indirectly as no binding site was predicted ( Figure S4, Table S1). No RpoN-binding site could be detected upstream of all these 10 genes up-regulated under N 2 fixation (Table S1), thus modulation might be an indirect effect.

The NifA Regulon Beyond Nitrogen Fixation
Despite the known role of NifA as transcriptional activator of N 2 -fixation-related genes, almost half of the modulated transcripts were under negative control in strain BH72 (409 versus 587 positively regulated); however, this was mostly indirect, since for the majority, NifA binding sites could not be predicted. Similar ratios were found in B. japonicum (190/133 genes) [46] or even in a reversed ratio in S. meliloti (291/320 genes) [47]. Moreover, in Azoarcus sp. strain BH72 only 77 or 24 of the genes up-or downregulated under BNF, respectively, were under apparent positive or negative control of NifLA ( Figure 4, Table S1). For example, as in B. japonicum [50], the fatty acid biosynthesis cluster (azo1620-azo1627) in strain BH72 was strongly down-regulated under N 2 fixation but still, probably indirectly, activated by NifA ( Figure 5A, middle panel). In accordance to expression data from microarray analysis, qRT-PCR further validated the azo1625 modulation, with down-regulation under N 2 response by -3.861.4 fold and activation by NifA by 52.4616.6 fold.
Although the expression of the gene cluster encoding for a T6SS was enhanced under N 2 fixation, it appeared to be independent of NifA control. In contrast, gene expression of 4 out of 22 secretory proteins, azo0483, azo1306, azo2346 and azo3872, which were upregulated under N 2 fixation, were positively affected by NifA, probably indirectly (Table S1).
The majority of the NifLA-modulated genes (474 under positive, 316 under negative control) were not significantly affected by the nitrogen stimulus ( Figure 4). This was unexpected, as NifA activity in NifL-containing Gammaproteobactera [51] and strain BH72 [18] is known to be controlled by the anti-activator NifL which inhibits the transcriptional activity of NifA by direct stochiometric interaction in response to elevated levels of fixed nitrogen and oxygen. Thus, one would expect regulation of genes in response to nitrogen and under control of NifA to be widely overlapping. Apparently, NifA plays an additional role under microaerobic conditions independent from this regulatory circuit. NifA in strain BH72 was shown to be functionally inactive in nifDHK activation in presence of combined nitrogen due to the binding of anti-activator NifL, and to be only active under N 2 fixation [18]; however, it had been observed that nifH expression was severely repressed albeit not completely abolished under N surplus [6,18], indicating slight residual activity of NifA. The availability of few free NifA (not bound to NifL) molecules might be sufficient to affect expression of downstream target genes. Moreover, nifA expression is detected even in the presence of combined nitrogen in strain BH72 and enhanced only 3 fold under N 2 fixation [18].
According to our set parameter for the fold change of induction with 1.8 or above, the NifA regulon encompassed 24.9% of the genome. Setting the fold change threshold to 3.5 to cover only strongly modulated genes, drastically reduced the number of modulated targets to 4.4%. Many of these differentially expressed genes that were not modulated by ammonia addition had these relatively high levels of modulation: 67 genes were up-regulated and 48 down-regulated, respectively. However, for none of these down-regulated promoters a NifA binding site was predicted (except for azo0499), which indicates an only indirect mechanism of regulation.
Among those 67 activated genes, 17 genes encoded for hypothetical proteins. Additionally, NifA affected the expression of sodC (azo0522) (superoxide dismutase) related to oxygen stress, and surE (azo1087) (exopolyphosphatase). In R. etli, a peroxidase expression is shown to be under RpoN and NifA control [52]. NifA binding sites could also be predicted upstream of both the genes. Some of the strongly induced gene clusters included genes for metabolic processes such as azo1003-azo1005 (transhydrogenases), azo1038-azo1043 (leucine/isoleucine biosynthesis), azo3418-azo3397 (ribosomal protein subunits) ( Figure 5A, lower panel), and were not affected in expression by glutamate. Although no obvious RpoN or NifA binding sites were predicted upstream of the gene clusters encoding for respiratory chain components, expression of nuo genes comprising the NADH:ubiquinone oxidoreductase gene cluster (complex I) as well as cco genes comprising the high affinity terminal oxidase (cbb 3 -type) gene cluster (complex IV) both were positively regulated in presence of NifA, respectively ( Figure 5A).
Negative regulation by NifA on targets with enhanced expression under nitrogen fixation seems to be quite unusual. Interestingly 9 out of 18 target genes in this category belonged to hypothetical proteins, but were mainly indirectly regulated by NifA according to the binding site prediction. For example, azo3471 encoding for a conserved hypothetical protein was 11fold induced by N 2 but about 3.1 -fold negatively regulated by NifA (Table S1). qRT-PCR analysis corroborated a similar tendency of induction of azo3471:21.269.6 fold induction under N 2 response and 3.261.9 fold negative regulation by NifA. Also here, no effect of glutamate was detected. Others from this category included genes involved in phospholipid biosynthesis (cfa1), PHB granule association (azo3815), chemotaxis (parA3), maintenance of cell wall integrity (rmlB) and tRNA binding (smpB). Enhanced expression of PHB associated phasin under N 2 was also observed in the P. stutzeri transcriptome [21]. A hybrid sensor response regulator (azo1584) and alternative sigma factor rpoN2 (azo1790) strongly expressed under this condition might also have some regulatory roles ( Figure S4).
Only 2 genes were repressed under N 2 fixation and negatively regulated by NifA (Table S1): a hypothetical membrane protein (azo0359 Figure 5B, middle panel) and assimilatory nitrite reductase nirB (azo1175). An RpoN-independent, NifA mediated regulation of respiratory nitrate reductase nirK has also been reported for B. japonicum [53]. Out of the 48 genes which were negatively affected by NifA in a nitrogen independent manner (Table S1), 25 are hypothetical or conserved hypothetical proteins (Table S1; Figure 5B, lower panel). Genes involved in amino acid uptake as well as branched chain amino acid metabolism were found to be repressed in the wild type as compared to strain BHLAO, along with urease and isocitrate lyase (key enzyme for glyoxalate pathway), indicative of down-regulation of pathways for alternative nitrogen sources and gluconeogenesis. The expression of nodD (transcriptional regulator of lysR fam) was strongly upregulated in nifLAbackground which can also indirectly regulate the transcription of target genes under such conditions (Table S1).
Beyond its role as essential transcription factor for nif genes, NifA was suggested to control other genes involved in nodule maturation and bacteroid persistence in symbiotic bacteria, as well [54]. A generalized role of NifA in the nodulation competitiveness of R. meliloti had also long been reported [55]. In Klebsiella pneumoniae, the nifA product can substitute for the glnG (ntrC) product (two component response regulator) as a nitrogen control regulator, replacing the ntrC product in the activation of its own promoter unidirectionally and, in addition, the promoters of several nitrogen assimilation genes, including the hut, put, aut, and glnA genes [56]. Even NifA can substitute NtrC binding of upstream activating sequences [56].
Considering the strongly modulated ($ 3.5 fold) 31 genes which were positively regulated by NifA and even up-regulated under N 2 fixation, 7 were predicted to bear NifA binding upstream elements (Table S3). These may represent the real targets of NifA, genes with nif associated function. Perhaps NifA has a direct action on these target genes, by upstream binding. Those with binding sites and negatively regulated by NifA were independent from the N 2 response. One of them, the putative repressor frlR, might be directly regulated by NifA and thus contribute to indirect regulation by NifA.
Comparing the array of 38 target genes with upstream RpoNbinding site (all modulated in their expression under N 2 fixation) with the list of 25 genes with predicted NifA binding site (expression regulated by NifA in all) generated an overlap of 8 genes, in which RpoN binding promoter sequence as well as NifA binding upstream activating sequence occurred simultaneously. As expected, 4 (rnfA1, azo0536, hesB, and nifE) out of these 8 target genes were located within the nif cluster. Two out of the remaining genes situated outside the nif-cluster, encoded for mobile electron carriers involved in N 2 fixation: nifF1 (azo0014) encoding for probable flavodoxin and fer22 (azo3368) encoding for probable ferredoxin. The other 2 encoded for hypothetical protein (azo3080) and conserved hypothetical ATPase (azo3791), respectively.

Conclusion
The N 2 response of Azoarcus sp. strain BH72 resulted in upregulation of 144 genes and down-regulation of 174 genes, independent of the NifA regulon accounting to 61% of the response. Outside the nif related genes there is a significant upregulation of genes for hypothetical proteins and non-nif related gene clusters and down regulation of several general metabolismrelated genes under N 2 fixation in strain BH72. The life style may undergo drastic changes when free-living bacteria become plant associated. Interestingly, the N 2 response itself may partially act as a stimulus to trigger such changes, even in the free living state: In addition to the up-regulation of genes involved in the nitrogen fixation machinery, several protein secretion systems, motility appendages and hypothetical proteins likely to be colonizationrelated became up-regulated. On the other hand, genes particularly related to cell structure, energy metabolism and protein synthesis were found to be down-regulated under N 2 , perhaps correlating with the slow-down of general metabolism associated with nitrogen-fixing life style at slower growth. Furthermore, down-regulation of selected aromatic aminoacid (phenylacetic acid) degradation pathways under N 2 stimulus observed in this study could be beneficial for successful establishment as disarmed pathogen in plant roots. This report provides for the first time evidence for a genome-wide regulatory activity of NifA in an endophyte. Although N 2 fixation modulated 13% of the genome, the NifA regulon was found to encompass a broader and more diverse range of targets than ever expected, accounting to 24% of the genome or still 4.4% when only the strongly modulated ($ 3.5 fold) genes were considered. Considering the large number of differentially expressed genes and the relatively few genes that may underlay direct regulation by NifA, it is tempting to speculate that the regulatory role of NifA is extended by the control of certain transcriptional factors such as sigma factors, activators or repressors. Several good candidates for further investigations are azo0625, gcvA, and sigma factors like algU and rpoN2 which are all modulated under N 2 response and apparently indirectly transcriptionally regulated by NifA.

Strains and Growth Conditions
Azoarcus sp. strain BH72 was grown in an oxygen-controlled bioreactor (Biostat B; B. Braun Melsungen AG, Melsungen, Germany) at 37uC with stirring at 600 r.p.m, under microaerobic conditions (0.6% oxygen concentration) either in nitrogen free SM-medium on N 2 [57], or in the same medium supplemented with either 10 mM ammonium chloride as N-source (N-replete) or with 10 mM glutamate (poor nitrogen source). A nifLA 2 mutant of Azoarcus sp. strain BH72 was constructed by disrupting the ORF of nifL by insertion of a Sp/Sm resistance cartridge [18]. Cells were harvested at an OD 578 of 0.6 by centrifuging 45 mL of culture per tube at 90006g for 5 min at 22uC. Pellets were immediately frozen in liquid nitrogen and stored at 280uC.
Azoarcus sp. Strain BH72 Transcript Profiling Using a Genome-wide Oligonucleotide Microarray Frozen bacterial cell pellets were re-suspended in a 1:1 mixture of pre-warmed phenol-chloroform (pH 4.7) and 50 mM sodium acetate/10 mM EDTA/1% SDS (pH 5.1). Cells were incubated for 5 min at 65uC followed by 10 min of incubation on ice. For phase separation, the mixture was centrifuged for 10 min at 12uC and 80006g, and the upper phase was mixed with the same volume of phenol-chloroform (pH 4.7). This phenol extraction was repeated three times followed by extraction with chloroformisoamyl alcohol (24:1). RNA precipitation was performed with the same volume of isopropanol for 45 min on ice. The RNA was pelleted by centrifugation (1300006g, 10 min, 4uC) and washed with 70% ethanol. RNA was dissolved in 16RNAsecure (Applied Biosystems), snap frozen in liquid nitrogen and stored at 280uC until further processing.
Contaminating DNA was removed from RNA preparations by DNase I using Qiagen columns (RNeasy Mini Kit, Qiagen, Hilden, Germany) according to manufacturer's instructions. The DNase I treated RNA was eluted twice in 30 mL of elution buffer, pooled and finally stored in 16RNA secure in 280uC.
For transcriptome microarray analysis, 20 mg of total RNA was reverse transcribed with BioScript RT in respective reaction buffer (Bioline) with random hexamers and 4:1 aminomodified-dUTP/ dTTP nucleotide mix for 90 min at 42uC, after it had been controlled for DNA contamination by PCR. For coupling of fluorescent dyes to the aminoallyl-labelled first strand cDNA, aliquots of Cy3-NHS or Cy5-NHS esters (GE Healthcare) were dissolved in the elution of first strand cDNA, mixed and incubated for 90 min at room temperature in the dark. Blocking of all remaining dyes (quenching) was achieved by adding hydroxylamine (4 M) to the sample solution, followed by incubation for 15 min at room temperature. Subsequent cleaning up of labelled cDNA from remaining dyes was performed with the CyScribe GFX Purification Kit (GE Healthcare) according to manufacturer's instructions. The Cy5-and Cy3-labelled cDNA that was used in one hybridization experiment were cleaned up together. The labelled cDNA was stored at 220uC until microarray hybridization after it had been checked for successful labeling [20]. The combined Cy3/Cy5-labelled targets were dried in a vacuum concentrator and further dissolved in DIG Easy hybridization solution (Roche, Mannheim, Germany) prior to hybridization. The transcriptome microarray spotted on epoxysilane-coated Nexterion Slide E (Schott) (CeBiTec, University Bielefeld) [20] contained 70mer oligonucleotide probes for the 3,989 predicted protein-coding genes from Azoarcus sp. strain BH72 in four replicates [20]. Before hybridization, the targets were denatured at 65uC for 10 min. The hybridization of fluorescently labelled cDNA targets to oligonucleotide microarrays was performed at 42uC for 14 to 18 hours as described previously [20]. The Cy3and Cy5-fluorescence was scanned at 532 nm and 635 nm with the GenePix Scanner 4000A (Molecular Devices, Sunnyvale, CA, USA) with a pixel size of 10 mm. The image analysis was performed with the GenePix 4.1 program. After scanning, raw data images of the slides that contain information about gene expression levels were obtained by the GenePix software. These images were analyzed by identifying each spot on the array (with the help of an array layout template) with measurements of its fluorescence intensity and the corresponding background intensity. With the file transformation tool Express Converter v.2.1 (http:// www.tm4.org/utilities.html), data files generated from GenePix (.gpr) were converted to TM4 files (.mev) that could be uploaded to the MIDAS platform. Normalization with MIDAS v2.19 (TM4 suite) was done in order to reduce variability by appropriately adjusting the data. In this study, the LOWESS (locally weighted scattered plot smoothing) normalization was performed using the block mode as set-up parameter as this mode allowed correction of systematic spatial variation between the grids of the array. The normalized data for each spot were exported to Excel for further downstream analysis. Overall, three independent experiments (biological replicates) were performed. Average expression fold were determined from normalized values of the three sets. Statistical analysis was performed by one-tailed paired t-test with Bonferroni correction for three independent experiments including dye-swap. Only genes that showed an expression of at least 1.8 fold and a P-value #0.05 were regarded as being differentially expressed. The microarray data have been deposited in the GEO database (http://www.ncbi.nlm.nih.gov/geo) under accession no. GSE49394.

RT-PCR and Semi Quantitative RT-PCR
For two step RT-PCR, initially 200 ng or 10 ng DNaseI treated total RNA was reverse transcribed at 50uC for 1 hour using gene specific reverse primers: nifA with nifArev3RT (59TCGTCCAGGTGCTCGCGGCTG 39) or 16S rDNA with R518 (59ATTACCGCGGCTGCTGG 39), respectively, using SuperScriptH III Reverse Transcriptase (RT) (life technologies) according to manufacturer's instruction. Subsequent PCR amplification of the nifA cDNA in the second step was carried out with DreamTaq DNA polymerase (Thermo SCIENTIFIC) using an aliquot from the previous reverse transcriptase reaction as template and primers nifAfor1RT (59AT-GAGCGCGGCCGGTCCGATG 39) and nifArev2RT (59CACGGTTTCGTGCCCGGCGCG 39) for 35 cycles of 1 min 95uC, 1 min 65uC, 1 min 72uC [18]. For 16S cDNA amplification, semi quantitative RT-PCR was carried out in a similar way using forward primer F341 (59CCTACGGGAGG-CAGCAG39) and the previous reverse primer R518 for PCR amplification at 1 min 95uC, 1 min 50uC, 1 min 72uC [58] with samples being taken out at different cycle numbers (13)(14)(15)(16)(17)(18)(19) from the PCR reaction. In each case the products were separated on 1.5% agarose gel.

Real-time PCR
For quantitative real-time RT-PCR of selected genes, synthesis of cDNA was achieved by using gene specific reverse primers for genes azo1790, azo3471, azo0954, azo3368, azo0804, azo0805, azo1566, azo1625, azo720, azo0585, azo1077, azo3155, azo3246 and 16S rRNA by using the Verso 2-Step QRT-PCR Kit from ABgene. cDNA synthesis on 30 ng RNA and the quantitative PCR step with 162-Step QPCR Mix (ABgene) and 0.56SYBR green I dye (Molecular Probes) was carried out as previously described [20]. The 2 2DDC T method was applied for data analyses, and 16S was used as a reference gene with the following primers: 16SRTfor

Web Based Genome Wide Predictions
Genome wide prediction of RpoN binding sites was carried out by using the online tool PePPER (Prediction of Prokaryote Promoter Elements and Regulons): a web based Regulon, Transcription Factor (TF) and Transcription Factor Binding Sites (TFBS) mining system [59]. The regulon and TFBS database used by PePPER is based on DBTBS, RegulonDB and MolgenRegDB. TFBSs prediction from the PePPER tool box was used for the prediction of (RpoN-binding) considering all the listed known RpoN binding sites of the tool box in the intergenic regions of the annotated genome of Azoarcus sp. strain BH72 based on the MolGenRegDB of PWM models.
Genome wide NifA binding site prediction was carried out by using the net online programme Prodoric Database release 8.9 [60]. The new software suite Virtual Footprint version 3.0 was used for genome wide motif predictions. For prediction of NifA binding sites, Regulon Analyses mode was used. For each IUPAC code used, sensitivity threshold was set to 0.8 and core sensitivity was set at 0.9. The maximum distance to the downstream gene was set to 0.500 kb. The search was restricted to intergenic regions, and the option to remove redundant palindromic matches was also activated. The generated promoter sequences were considered only when the SEP-score was above -6.0. Intrinsic transcriptional terminators (azo_Terminator(s) were inferred from genome annotation of strain BH72 by GenDB [61]. Figure S1 Sequence logo for genome wide RpoNbinding site of Azoarcus sp. strain BH72 created by using 'WebLogo'. The consensus motif is derived from individual motifs ranging from 46 to 49 bases predicted in the intergenic upstream region of target genes using the online webserver for prediction of prokaryotic promoter elements, PePPER. 173 putative RpoN-binding motifs upstream of 162 target genes with score .6.00 were used to create the logo with the WebLogo generator http://weblogo.berkeley.edu/. (PDF)