Reactive Oxygen Species-Inducible ECF σ Factors of Bradyrhizobium japonicum

Extracytoplasmic function (ECF) σ factors control the transcription of genes involved in different cellular functions, such as stress responses, metal homeostasis, virulence-related traits, and cell envelope structure. The genome of Bradyrhizobium japonicum, the nitrogen-fixing soybean endosymbiont, encodes 17 putative ECF σ factors belonging to nine different ECF σ factor families. The genes for two of them, ecfQ (bll1028) and ecfF (blr3038), are highly induced in response to the reactive oxygen species hydrogen peroxide (H2O2) and singlet oxygen (1O2). The ecfF gene is followed by the predicted anti-σ factor gene osrA (blr3039). Mutants lacking EcfQ, EcfF plus OsrA, OsrA alone, or both σ factors plus OsrA were phenotypically characterized. While the symbiotic properties of all mutants were indistinguishable from the wild type, they showed increased sensitivity to singlet oxygen under free-living conditions. Possible target genes of EcfQ and EcfF were determined by microarray analyses, and candidate genes were compared with the H2O2-responsive regulon. These experiments disclosed that the two σ factors control rather small and, for the most part, distinct sets of genes, with about half of the genes representing 13% of the members of H2O2-responsive regulon. To get more insight into transcriptional regulation of both σ factors, the 5′ ends of ecfQ and ecfF mRNA were determined. The presence of conserved sequence motifs in the promoter region of ecfQ and genes encoding EcfQ-like σ factors in related α-proteobacteria suggests regulation via a yet unknown transcription factor. By contrast, we have evidence that ecfF is autoregulated by transcription from an EcfF-dependent consensus promoter, and its product is negatively regulated via protein-protein interaction with OsrA. Conserved cysteine residues 129 and 179 of OsrA are required for normal function of OsrA. Cysteine 179 is essential for release of EcfF from an EcfF-OsrA complex upon H2O2 stress while cysteine 129 is possibly needed for EcfF-OsrA interaction.


Introduction
Extracytoplasmic function (ECF) s factors are alternative bacterial RNA polymerase s factors that play key roles in the response and adaptation of bacteria to different stresses and environments (for reviews and a comprehensive classification, see [1,2]). ECF s factors are members of the s 70 family, which is divided into four groups. Primary s factors of group 1, to which the housekeeping s factors belong, contain four conserved domains 1 to 4 and some of them also comprise an additional non-conserved region. They usually recognize promoters with the sequence TTGaca (235) and TAtaaT (210) [3]. In contrast, ECF s factors belong to group 4 of the s 70 family and contain only the conserved domains 2 and 4. Many of them are thought to respond to environmental signals, they are often associated with an anti-s factor, and usually auto-regulate their own expression [1,2,4]. Among the environmental cues are reactive oxygen species (ROS) which almost all bacteria encounter and against which even anaerobes have evolved defense mechanisms [5][6][7]. In aerobic organisms, ROS are generated also endogenously, e.g. by incomplete reduction of oxygen during respiration. The term ROS is generic, embracing not only free radicals such as superoxide anion (O 2 2 ) and hydroxyl radicals (OH N ) but also hydrogen peroxide (H 2 O 2 ) and singlet oxygen ( 1 O 2 ) (for reviews, see [8,9]). Generation of ROS occurs via different routes (for reviews, see [10][11][12]). Briefly, the best studied enzymatic generation of superoxide, and consequently hydrogen peroxide, originates from NADPH oxidases that catalyze the production of superoxide by the one-electron reduction of molecular oxygen using NADPH as an electron donor [11,13]. The main source of singlet oxygen is the photosynthetic apparatus where it is generated in photosystem II as a side product by energy transfer from excited triplet-state chlorophyll pigments to O 2 [14]. Energy can also be transferred to molecular oxygen by excited photosensitizers such as phytoalexins which are produced by plants in response to pathogens [15]. Apart from plant-derived sources, singlet oxygen is also produced in natural waters by the exposure of chromophoric dissolved organic matter to light [16].
Several ECF s factors have been described to play a role in the response of bacteria to oxidative stress. Examples are Streptomyces coelicolor SigR which responds to disulfide stress produced by superoxide and diamide [17], Caulobacter crescentus SigT which is necessary for survival under osmotic and oxidative stress [18], and SigF of the same organism mediating the response to oxidative stress in stationary phase [19]. In the photosynthetic bacterium Rhodobacter sphaeroides, transcription of rpoE is increased upon singlet oxygen stress ( [20]; for review, see [21]) while RpoE activity is controlled by the anti-s factor ChrR [22]. Orthologs of the RpoE-ChrR system are present in various bacterial species [23] including C. crescentus [24] and Myxococcus xanthus [25].
Rhizobia, soil bacteria that fix nitrogen in symbiosis with leguminous plants, are exposed to a wide range of environmental stimuli, including ROS, both in their free-living state in the soil and in the interaction with host plants, i.e., during infection and establishment of symbiosis, during nitrogen fixation in root nodules, and during senescence of these nodules ( [26,27]; for reviews, see [28,29]). Accordingly, rhizobia use a set of transcription regulators to reprogram gene expression in order to cope with these stresses. Notably, during symbiosis ROS act as signaling molecules and are needed for an efficient Rhizobium-legume interaction [29].
The soybean endosymbiont Bradyrhizobium japonicum encodes a total of 23 predicted s factor-coding genes in its genome [30,31]. Whereas two of them are s 54 -type factors, 21 belong to the s 70 family. The latter category includes the housekeeping s factor SigA (group 1), three RpoH s factors (group 3), and 17 ECF s factors (group 4) whose relationship is depicted in the phylogenetic tree shown in Figure 1. With SigA as an outgroup, the tree subdivides the ECF s factors into two groups of 12 and 5 members documenting substantial diversity among them. In the larger group, three pairs of similar s factors are found: EcfS (Blr4928) and Blr3038, Bll1028 and Blr3042, and Bll6484 and Bll2628, which show 35%, 55%, and 45% amino acid sequence identity, respectively. Up to now, only two members of the B. japonicum ECF s family have been functionally studied in more detail. Most recently, it was described that EcfS (Blr4928) plays a critical role in the establishment of a functional symbiosis with soybean [32].
Previously, we have shown that EcfG (Blr7797) is involved in tolerance to heat and desiccation as well as in the symbiotic interaction with soybean and mungbean [31]. Other functionally studied EcfG orthologs in rhizobia include RpoE2 of Sinorhizobium meliloti [33] and RpoE4 of Rhizobium etli [34] which control regulons typical of general stress response as does s T of C. crescentus [35]. Similarly, ECF s factor RpoE of the plant-associated bacterium Azospirillum brasilense is involved in tolerance to singlet oxygen and other abiotic stresses [36]. Yet another rhizobial ECF s factor, RpoI of Rhizobium leguminosarum, is required for synthesis of the siderophore vicibactin and iron uptake [37,38].
The transcriptome analysis of H 2 O 2 -stressed B. japonicum cells, which is presented here, revealed that the expression of two predicted ECF s factors is induced by H 2 O 2 and also in response to treatment with other ROS: Bll1028 (hereafter named EcfQ in accordance with its ECF s factor function and annotation as CarQ in Rhizobase (http://genome-legacy.kazusa.or.jp/ rhizobase/Bradyrhizobium) and Blr3038 (hereafter termed EcfF according to the SigF prototype ECF s factor of this class [2]). We have determined the regulons of both s factors, and demonstrated that mutant strains lacking either one or both ECF s factor(s) show increased sensitivity to singlet oxygen. Furthermore, we have analyzed the distinct regulatory mechanisms controlling synthesis and activity of EcfQ and EcfF. While expression of both genes is controlled at the transcriptional level, activity of EcfF is additionally regulated by protein-protein interaction with its cognate anti-s factor Blr3039 (hereafter termed OsrA for oxidative stress-response anti-s factor). Conserved cysteine residues of OsrA are involved in H 2 O 2 responsiveness and inhibition of EcfF activity under non-stressed conditions.

Results
Transcriptional Profile of B. Japonicum in Response to H 2 O 2 -mediated Oxidative Stress In order to identify B. japonicum genes involved in oxidative stress response, global transcriptome analyses were performed with wild- Figure 1. Phylogenetic relationship of 17 predicted ECF s-factors in B. japonicum. The tree (generated by the UPGMA method [90]) is drawn to scale with respect to evolutionary distances. Bootstrap values were obtained after 1,000 repeats, and nodes with a confidence of greater than 90% ( N ) or 50% (o) are indicated. The primary s factor (SigA) sequence of B. japonicum is included as an outgroup (dashed branch). doi:10.1371/journal.pone.0043421.g001 type cells that had been treated with 2 mM H 2 O 2 for 10 min and with untreated wild-type cells (control). To mimic the symbiotic environment we used micro-oxic conditions as standard condition for all microarray analysis throughout this study. A total of 225 genes were differentially expressed in response to H 2 O 2 (144 upregulated, 81 downregulated; see Table S1), with 56% of them encoding proteins of unknown functions. Several genes known to be involved in the oxidative stress response were upregulated, such as catalase (blr0778), hydroperoxide resistance proteins (bll4012, bll0735) and methionine sulfoxide (MetSO) reductases (bll5855, blr7043). Notably, almost one third (29 genes) of the H 2 O 2regulated genes that encode proteins of known or predicted function are transcriptional regulators including five MarR-, four TetR-, and three LysR-type proteins. Furthermore, transcription of genes for three s factors was affected by H 2 O 2 exposure. While blr1883 encoding one of two s 54 -type s factors of B. japonicum was slightly down-regulated, the genes for two ECF s factors ecfQ and ecfF were strongly induced (34.8 and 14.4 fold, respectively). The latter s factors are the primary focus of this study, and some of their characteristic features are summarized in Table 1.

Response of EcfQ and EcfF to Different ROS
To validate microarray data obtained for ecfQ and ecfF, and to gain insight into the expression of ecfQ and ecfF upon treatment with other sources of ROS, quantitative, cDNA-based real-time PCR (qRT-PCR) analyses were performed. Besides treatment with H 2 O 2 , the following two reagents were used: paraquat (methylviologen) generating superoxide, and rose bengal in combination with light, generating singlet oxygen ( 1 O 2 ). The results shown in Table 2 document induction of ecfQ and ecfF not only in response to H 2 O 2 but also to singlet oxygen. Expression of ecfQ, but not ecfF, is also elevated under treatment with paraquat.
Phenotypic Characterization of Deletion Mutants DecfQ, D(ecfF-osrA), DosrA, and D(ecfQ, ecfF-osrA) To further elucidate the role of ECF s factors EcfQ and EcfF in oxidative stress response, mutant strains DecfQ, D(ecfF-osrA) and D(ecfQ, ecfF-osrA) were constructed ( Figure 2). In addition, strain DosrA was generated to study the predicted function of OsrA as an anti-s factor of EcfF. Finally, a deletion strain lacking Blr3042 was constructed to elucidate the function of this EcfQ paralog ( Figure 1). As the latter strain was indistinguishable from the wild type in all phenotypic tests, it will not be further discussed in this work.
Growth kinetics of DecfQ, D(ecfF-osrA), DosrA and D(ecfQ, ecfF-osrA) strains were determined under oxic, micro-oxic and anoxic conditions. Growth of the mutant strains followed a similar trend as seen with the wild type under micro-oxic conditions (data not shown). In oxic and anoxic conditions, growth of strains D(ecfF-osrA), DosrA and D(ecfQ, ecfF-osrA) but not of DecfQ was retarded compared to the wild type ( Figure 3A,B). In anoxic conditions doubling time of strain DosrA was approximately twice that of the wild type, and the final optical density reached by this mutant was lower ( Figure 3B).
All mutant strains were symbiotically proficient and indistinguishable from the wild type when tested on two different soybean varieties (Glycine max cultivar Williams 82 and cultivar ''Green Butterbean''), on mungbean (Vigna radiata) and on cowpea (Vigna unguiculata) (data not shown). Genetically linked anti-s factor a -OsrA n.a. d a According to [30]. b Based on the transcription start site mapped in this study, translation of ecfF is likely to start at a distal start codon, which leads to a shorter gene product of 170 amino acids (for details see text and Figure 6A,C). c According to [2].  Micro-oxically grown wild-type cells exposed to H 2 O 2 (experiment 1) or to paraquat (2, 3) for the indicated time were compared to untreated cells. Similarly, cells exposed to rose bengal plus light were compared to cells exposed to rose-bengal in the dark (4,5). In the control experiment (6), lightexposed cells were compared to cells grown in the dark. Further, the sensitivity of the mutants towards different ROS was tested in filter disk assays, on gradient plates and in spot tests. All four mutant strains showed increased sensitivity towards oxidative stress caused by singlet oxygen both on gradient plates (data not shown) and when spotted on PSY plates containing rose bengal ( Figure 4). In filter disk assays, the mutants showed a wildtype phenotype with respect to their sensitivity to the following compounds: (i) H 2 O 2 ; (ii) diamide, a reactive electrophilic species which affects the thiol redox balance; (iii) FeSO 4 that can generate oxidative stress via the Fenton reaction; (iv) NO-generating agents such as S-nitroso-N-acetylpenicillamine and S-nitrosoglutathione; (v) methylglyoxal, a toxic, electrophilic compound; (vi) paraquat (methylviologen) which causes the formation of superoxide (data not shown).  The Regulon of EcfQ Microarray analysis was used to identify potential target genes of EcfQ. To this end RNA was isolated from the wild type and the DecfQ strain, both grown unstressed or stressed by exposure to H 2 O 2 . Expression of nine genes differed between the wild type and the DecfQ mutant under non-stressed conditions (four up-regulated, five down-regulated; Table S2A). In H 2 O 2 -stressed cells, the number of differentially expressed genes increased to 34 with seven genes up-regulated and 27 genes down-regulated (Table S2B). The latter category might include direct targets of EcfQ given the positive regulation mode exerted by s factors. However, inspection of the DNA regions (200 bp) upstream of these genes did not reveal common motifs that might function as recognition site of EcfQ. Notably, two thirds of the differentially regulated genes encode hypothetical or functionally unknown proteins. Among genes with predicted functions are blr0337 and blr3534 which code for a subunit of putative carbon monoxide dehydrogenases and are both down-regulated in the mutant.

The Promoter Region of ecfQ and of other Genes Coding for Class 33 ECF s Factors are Conserved
When we had a closer look at the 13 a-proteobacterial genes representing the class 33 of ECF s factors to which EcfQ belongs [2] we made several observations: (i) in almost every organism of this group, except Mesorhizobium loti, there are two genes coding for this type of ECF s factor; (ii) consistently, one of them has a predicted anti-s factor gene in its proximity, but for the other, a predicted anti-s factor gene is absent (EcfQ together with five were pre-grown to early stationary phase, and aliquots of serial dilutions were spotted on plates containing 0.1 mM rose bengal (two independent dilution series per strain). The control plate shown in the upper panel was incubated in the dark while the plate shown in the lower panel was light exposed (2,000 lux) for 1 h to allow generation of singlet oxygen (for more details, see Materials and Methods). doi:10.1371/journal.pone.0043421.g004 other class 33 s factors belongs to the latter category); (iii) by aligning the upstream regions of the six genes of this second group, a striking pattern of sequence conservation was observed ( Figure 5A). To obtain information on the relative position of these elements in the promoter, the 59 end of ecfQ mRNA was determined by primer extension, using RNA isolated from the B. japonicum wild type grown under different conditions ( Figure 5B). The results of reverse transcription revealed the ecfQ transcription start point at a C located 44 nucleotides upstream of the annotated ecfQ start codon ( Figure 5B). In agreement with the microarray and qRT-PCR analyses, the amount of cDNA derived from RNA in H 2 O 2 -treated cells (lane 2) was higher than the amount derived from untreated cells (lane 1). Putative 235 and 210 promoter boxes were identified, forming the consensus GCAGAC and TAACAAT, respectively, however, the spacing between the motifs is unusually long (20 nt).
Several additional stretches of nucleotides are also conserved in the upstream region of ecfQ and the other five s factor-coding genes which belong to the same group. A stretch reading GAAAC is repeated several times in the upstream region (boxes labeled 226, 259, and 280 in Figure 5A). At box 280, the GAAAC sequence is part of the inverted repeat TGTTTC-N 17 2GAAACA ( Figure 5A). Database searches with the virtual footprint tool Prodoric to find regulators that might bind to this region revealed no obvious candidates. Also, the identified region does not resemble any described binding sites for several B. japonicum regulators such as Irr, Fur, FixK 2 , and RegR [39][40][41].

The Regulon of EcfF
To identify genes possibly controlled by s factor EcfF, microarray analyses were performed with the D(ecfF-osrA) mutant strain. In micro-oxically grown, unstressed mutant cells, expression of only three genes was slightly up-regulated apart from the obvious decrease of expression of the two deleted genes (Table  S3A). This indicated that in the wild type EcfF is mainly inactive under these growth conditions. Upon H 2 O 2 treatment, expression of 22 genes (including ecfF and osrA) differed between mutant and wild-type cells confirming that EcfF-dependent transcription is activated by H 2 O 2 exposure (Table S3B). Notably, all regulated genes had negative fold-change values, which is in line with the role of EcfF as a positive regulator and suggests that there are direct target genes in this group. Other than ecfF and osrA no genes were common to the lists of differentially expressed genes in unstressed and stressed cells.
To examine the predicted function of OsrA as an anti-s factor, we also performed microarray experiments with the DosrA mutant strain. We used unstressed, micro-oxically grown cells in these experiments because we assumed that even in unstressed cells the absence of the anti-s factor OsrA should result in up-regulation of EcfF-dependent genes if the function of OsrA in the wild type were to inhibit the activity of EcfF under these conditions. Expression of 39 B. japonicum genes (including ecfF and osrA) was altered in the DosrA strain, with 24 genes (more than 60%) encoding hypothetical or unknown proteins (Table S4). Only 3 genes had negative foldchange values (one of them being the deleted osrA gene) while the large majority of 36 genes was up-regulated, likely due to hyperactivity of EcfF caused by the absence of its anti-s factor. Elevated expression of ecfF (18.8 fold) in the DosrA background suggested that the ecfF-osrA operon is autoregulated. Interestingly, not only ecfF but also its paralog ecfS (blr4928) was more highly expressed (3.8 fold) in the mutant pointing to potential cross-talk between the two s factor2anti-s factor systems.
When the expression data generated with stressed cells of the D(ecfF-osrA) mutant was compared with that of unstressed DosrA cells, nine genes (apart from the mutated genes) showed a regulatory pattern that is expected for EcfF-OsrA being a cognate s factor2anti-s factor pair, i.e., down-regulation in the D(ecfF-osrA) mutant and up-regulation in the DosrA mutant (Table 3). Except for blr7044, all genes of this group belong to the H 2 O 2inducible genes. Notably, genes bll5855, blr7043, blr7044 all encode predicted peptide MetSO reductases.
Taking into account the predicted operon structure for bll1027-26, ecfF-osrA and blr7043-45, the genes listed in Table 3 comprise a total of seven transcription units with promoters that are primary candidates for being direct EcfF targets. When we searched in a 200-bp window upstream of the respective start codons for common putative promoter elements we could indeed identify a conserved GTAAC(g,a)-N 14-15 -(c,t)CG(t,a) motif ( Figure 6A,B). This element is remarkably similar to the tGTAACc-N 16 -CGAA promoter sequence that was proposed for group 16 of ECF s factors [2] to which EcfF belongs. The predicted EcfF target promoter preceding ecfF-osrA was confirmed by primer extension experiments ( Figure 6C). Indeed, a transcript starting at a C located 6 bp downstream of the predicted -10 box of ecfF was detected in cells exposed to H 2 O 2 but not in untreated cells. The experimentally detected transcription start site overlaps the ecfF ATG start codon annotated in Rhizobase [30] which argues for the more distal translational start codon as indicated in Figure 6C. Likewise, the annotated GTG start codon of bll5855 might be incorrect because the predicted 210 box of the respective promoter is located only 8 bp upstream of this start codon ( Figure 6A). When the EcfF consensus motif (GTAAC(g,a)-N 14-

In Vivo Interaction of EcfF and OsrA
If EcfF and OsrA functioned as a typical cognate s factor-antis factor pair they ought to interact directly at the protein level. We have used a bacterial two-hybrid system (BACTH system; [42,43]) to further evaluate this model. Plasmids pRJ9746 and pRJ9744 encoding protein fusions of EcfF and OsrA to adenylate cyclase Cya subdomains T18 and T25, respectively, were constructed ( Figure 7A). Cotransformation of E. coli BTH101 cells with these plasmids resulted in strain 1 which showed significant bgalactosidase activity ( Figure 7B). By contrast, no b-galactosidase activity above background was detected in E. coli BTH101 cells that contained either of the fusion plasmid in combination with the empty vector of the other hybrid plasmid (data not shown). This indicated that interaction of EcfF with OsrA enabled functional complementation of the T18 and T25 adenylate cyclase domains.

Conserved Cysteine 129 of OsrA Might be Required for Interaction with EcfF
Amino acid sequence alignment of OsrA with orthologous putative anti-s factors associated with group-16 s factors in other proteobacteria revealed two highly conserved cysteine residues. These residues are located at positions 129 and 179 of OsrA, and they are the only cysteines present in this protein (Figure 8, Figure  S1).
To probe the function(s) of the conserved cysteines, mutant variants of OsrA (C129S, C179S, C129S+C179S) were fused to T25 ( Figure 7A) and tested for two-hybrid interaction in combination with the T18-EcfF fusion protein. In strain 3 harboring the T25-OsrA C179S fusion, b-galactosidase activity reached about 70% of the reference strain 1 (T25-OsrA) whereas in strains 2 (T25-OsrA C129S, C179S) and 4 (T25-OsrA C129S) only background activity was detected ( Figure 7B). Assuming that the point mutations did not drastically alter protein expression levels or stability, these results imply that cysteine 129 of OsrA, but not cysteine 179, is required for interaction with EcfF.  Table 4 we conclude that (i) the complementation strategy is effective because wild-type OsrA restores the normal ecfF expression pattern (cf.

Discussion
Rhizobia are exposed to oxidative stress originating from ROS that are generated either intrinsically in aerobic metabolism or by legume host plants during rhizobial infection [29]. Here we have analyzed the transcriptional response of B. japonicum cells to oxidative stress. Special emphasis was given to the two ECF s factors EcfQ and EcfF because (i) their transcription was strongly induced upon exposure to ROS, and (ii) ECF s factors are typical regulators in the bacterial stress response. Hydrogen peroxide exposure of free-living cells which were grown micro-oxically to mimic symbiotic conditions resulted in altered transcription of more than 200 genes. Many of them are functionally uncharacterized, others are related to oxidative stress or encode transcriptional regulators. Among the latter category are five MarR-type and three LysR-type regulators which are involved in the oxidative stress response in various other bacteria [44][45][46][47][48].
In recent studies, effects of H 2 O 2 and paraquat exposure on transcription in oxically grown B. japonicum cells were described [49,50]. A comparison of that study with our own results revealed that genes induced by H 2 O 2 in both oxic and micro-oxic cells comprise those encoding hydroperoxide resistance proteins (bll4012, bll0735), putative epoxide hydrolase 1 (bll3418), a putative glutathione S-transferase (bll7849), and the ECF s factors mentioned above (ecfQ, ecfF). Fold-change values differed substantially between the two studies which is likely due to . Extension products obtained with the [ 32 P]-labeled primers pe-1028-1 and pe-1028-2 were separated on a 6% denaturing polyacrylamide gel (only results obtained with primer pe-1028-1 are shown). The sequencing ladder was generated with plasmid pRJ0211 and primer pe-1028-1 (for more details, see Materials and Methods and Table S6). Part of the predicted promoter region is shown on the right, with the experimentally assigned transcription start site indicated with an arrowhead. The sequence of the ecfQ promoter region is shown below the picture. The determined 59 end of the transcript is indicated by an arrowhead, and the annotated translation start site is underlined and printed in bold face. Putative 210 and 235 regions are shaded in grey. The N-terminus of the EcfQ protein sequence is indicated in one-letter code. doi:10.1371/journal.pone.0043421.g005 Table 3. B. japonicum genes whose expression differed in the D(ecfF-osrA) and the DosrA relative to the wild type a . Cells were grown micro-oxically and those of strain D(ecfF-osrA) were exposed to 2 mM H 2 O 2 for 10 min prior to harvest. Wild-type cells grown under the respective conditions served as reference in both experiments. Listed are genes with an absolute fold-change value of .3 in at least one of the mutants and .2 in the other mutant. b Nomenclature according to [30]. Putative operons are shown in italics with cotranscribed promoter-distal genes indented to the right. c Fold-change values from the comparison of micro-oxically grown wild-type cells exposed to 2 mM H 2 O 2 for 10 min with untreated cells (see Table S1). d Gene description according to [30] with modifications. doi:10.1371/journal.pone.0043421.t003 different growth conditions and different microarray platforms. Remarkably, the gene for catalase KatG (blr0778), whose role in protection from oxidative stress in B. japonicum is well documented [51], appeared to be induced by hydrogen peroxide treatment only in micro-oxically but not in oxically grown cells. We speculate that blr0778 is induced even in untreated aerobic cells due to endogenous ROS production that might be higher in oxic cells than in micro-oxic cells. ECF s factors EcfQ and EcfF were functionally characterized by phenotypic analysis of respective mutants and microarray analyses. Deletion mutants DecfQ, D(ecfF-osrA), DosrA and D(ecfQ, ecfF-osrA) were more sensitive to singlet oxygen, and thus confirmed that both s factors are indeed involved in oxidativestress tolerance. Singlet oxygen sensitivity of the mutants was moderately increased and restricted to this type of ROS. This might be due to an intrinsic tolerance of B. japonicum and/or the existence of functionally redundant, EcfQ2/EcfF-independent ROS-protective systems. This hypothesis is in line with the finding that the regulons of EcfQ and EcfF showed only limited overlap  Figure 6C (for details see text). B. WebLogo of the EcfF-target promoter base on the alignment from panel A. C. 59 end mapping of ecfF. For primer extension, total RNA of the wild-type strain grown under the following conditions was used: micro-oxic (lane 1), and micro-oxic, treated with 2 mM H 2 O 2 for 10 min (lane 2). Extension products obtained with the [ 32 P]labeled primers pe-3038-1 and 3038-RT-R were separated on a 6% denaturing polyacrylamide gel (only results obtained with primer pe-3038-1 are shown). The sequencing ladder was generated with plasmid pRJ9724 and primer pe-3038-1 (for details, see Materials and Methods and Table S6). Part of the predicted promoter region is shown on the left, with the experimentally assigned transcription start site ''+10 indicated with an arrowhead. Because the ATG translation start codon of ecfF as annotated in Rhizobase ( [30]; dashed-line rectangle) overlaps the transcription start, the more distal, alternative start codon seems more likely (solid-line rectangle). doi:10.1371/journal.pone.0043421.g006 with the large group of H 2 O 2 -responsive genes ( Figure S2), and it also could explain the symbiotic proficiency of the mutants.
In the absence of stress, both s factors are probably inactive because under these conditions regulation of only few genes was altered in the deletion mutants, possibly by indirect means as none of them was differentially expressed in stressed cells (Tables S2 and  S3). Nevertheless, growth of the mutant lacking OsrA was impaired even without externally applied oxidative stress, particularly under anoxic conditions, which indicates that hyperactivity of EcfF might be deleterious.
Transcriptional control of ecfQ and ecfF is likely to occur via different mechanisms. The presence of conserved motifs within the ecfQ promoter region points to the involvement of a yet unidentified transcriptional regulator, a model that is compatible with the absence of an anti-s factor gene associated with ecfQ. By contrast, the ecfF-osrA operon appears to be autoregulated, which is typical for cognate s and anti-s factors genes organized in an operon. The difference in the regulatory mode may also be responsible for the differential response of these s factor genes to treatment with paraquat.
EcfQ and EcfF control rather small and largely distinct groups of genes. Common to both regulons are only four clustered genes (bll0331-0333; blr0337) of which bll0333 encodes a precursor of a putative alcohol dehydrogenase and blr0337 a subunit of a predicted carbon monoxide dehydrogenase. Notably, genes blr0335 and blr0336 encoding two additional subunits of the latter enzyme are also controlled by EcfF. The regulon of EcfQ is functionally rather undefined because almost 70% of its members are hypothetical or unknown proteins. By contrast, more than 70% of the proteins encoded by the genes belonging to the EcfF regulon have a (predicted) functional annotation. Strikingly, half of them are oxidoreductases including MetSO reductases Bll5855 and Blr7043 whose genes were induced by H 2 O 2 treatment. A third MetSO reductase, Blr7044, that is induced 2.9-fold by H 2 O 2 exposure, is yet another member of the EcfF regulon as its expression was inversely affected in the D(ecfF-osrA) and DosrA strains (Table 3). Thus, at least three of five MetSO reductases encoded in the B. japonicum genome are H 2 O 2 responsive and controlled by EcfF/OsrA (Blr0834 and Bll6260 being the remaining two). Neisseria gonorrhoeae [52] and Neisseria meningitidis [53] represent two other bacterial species where genes encoding MetSO reductases are controlled by an ECF s factor-anti-s factor pair.
For many bacterial species the function of MetSO reductases as antioxidant repair enzyme is well documented [54][55][56][57][58][59]; for review, see [60]). Repair of oxidized methionines by MetSO reductases depends on protein electron donors such a thioredoxin (for review, see [61]). Based on its putative signal sequence Blr7043 is predicted to localize to the periplasm. Thus, for Blr7043 to function additional components are required which transfer electrons across the cytoplasmic membrane and deliver them to this enzyme. DsbDC of E. coli is a well characterized example which is needed for reduction or isomerization of disulfide bonds in the periplasm [62]. Based on the predicted topology and domain structure of Bll1026 (membrane-anchored periplasmic thioredoxin) and Bll1027 (membrane protein with a DsbD b core domain), we speculate that Blr7043 in B. japonicum may receive electrons via these proteins whose genes are co-regulated with blr7043 by EcfF/OsrA.
In vivo interaction of EcfF with OsrA was demonstrated with a bacterial two-hybrid system in E. coli. These experiments revealed that a conserved cysteine at position 129 of the anti-s factor OsrA is crucial for interaction with EcfF, and this result was further substantiated by the finding that replacement of this residue led to constitutive EcfF activity in B. japonicum. The predicted localization of C129 in the cytoplasmic membrane argues against this amino acid making direct contact with EcfF. However, it is possible that  Cells were grown micro-oxically without stress or were exposed to 2 mM H 2 O 2 for 10 min prior to harvest. Expression levels of ecfF in different backgrounds were determined by qRT-PCR and expressed as fold-change values 6 standard errors relative to the expression detected in the pseudo wild-type strain 92-29 under nonstress conditions. Data are based on three technical replicates of a representative experiment which was repeated in three biological replicates. For details, see Materials and Methods. b Control strain containing vector pSUP202pol4 chromosomally integrated downstream of the DosrA::aphII locus (see Figure 2B). doi:10.1371/journal.pone.0043421.t004 C129 is crucial for keeping OsrA in an interaction-competent conformation and that its replacement with serine interfered with this function. The second conserved cysteine of OsrA, C179, is probably involved in sensing and/or transducing the stress signal because inhibition of EcfF by the OsrA C179S variant was not released when B. japonicum cells were stressed with hydrogen peroxide. Taking into account our data and the predicted OsrA topology (Figure 8) we propose that oxidative stress detected via periplasmexposed C179 is signaled to the cytoplasmic portion of OsrA where it leads to the release of bound EcfF. Although cysteines are redox-active amino acids and thus well suited to monitor oxidative stress, C179 of OsrA is not necessarily the primary signal input site. Inspection of the OsrA amino acid sequence revealed a striking accumulation of eight methionine residues in two predicted periplasmic loops (Figure 8), with only two of them being conserved in the closest B. japonicum paralog TmrS (Blr4929), the anti-s factor of EcfS (Blr4928) (Figure 1, Figure S1; [32]). Given that three MetSO reductases are controlled by EcfF/OsrA it is tempting to speculate that the presence of multiple methionines in OsrA makes this regulator intrinsically responsive to molecules that elicit methionine oxidation.
Amino acids that are critical for oxidative stress signaling were identified previously in other anti-s factors which, however, are not homologs of OsrA. Specifically, conserved histidine or cysteine residues in the ChrR anti-s factor proteins of C. crescentus, R. sphaeroides, and N. meningitidis are required for proper regulation of the respective s E proteins in response to organic hydroperoxide [24,53,63]. Likewise, in M. xanthus blue-light responsiveness of the membrane-bound CarQ anti-s factor is controlled by another membrane-associated protein, CarF, whose anti-anti-s factor activity depends on several histidine residues [64,65].
Our study contributes to the characterization of an ECF s factor family found in B. japonicum. With EcfQ and EcfF studied here and the previously described s factors EcfG [31] and EcfS [32], functional information is now available for a total of four ECF s factors of this bacterium. While the general stress response regulator EcfG contributes to both free-living and symbiotic traits, functions of EcfS are largely confined to symbiosis and those of EcfQ and EcfF to the free-living state. The function of the EcfQ paralog Blr3042 remains enigmatic as it turned out to be dispensable under all tested conditions. Challenging goals for future studies include the characterization of signals, mechanisms of their transduction, and functions of target genes of EcfQ and EcfF.

Mutant Construction
Mutant strains 0202 (DecfQ), 9688 (D[ecfF-osrA]) and 9692 (DosrA) were constructed by marker-exchange mutagenesis. Briefly, the 59-and 39-flanking regions of the genes to be deleted were amplified by PCR using primer pairs listed in Table S6, cloned in the pGEM-T Easy vector (Promega Corp., Madison, WI, USA), verified by sequencing, and finally cloned in tandem in vector pSUP202pol4. A 1.2-kb kanamycin resistance cassette (aphII) derived from pBSL86 [70] was inserted between the up-and downstream regions to generate plasmids pRJ0202 (for deletion of ecfQ), pRJ9688 (for deletion of ecfF plus osrA), and pRJ9692 (for deletion of osrA). The resulting plasmids were transformed into E. coli S17-1 and then mobilized by conjugation into B. japonicum wild-type strain 110spc4 as previously described [71]. The correct genomic structure of the resulting deletion mutants 0202 (DecfQ), 9688 (D[ecfF-osrA]), and 9692 (DosrA) was verified by PCR. In strains 9688 and 9692, the cassette was inserted in the same orientation as the deleted gene(s) while in strain 0202 the cassette was oriented opposite to the deleted ecfQ gene (Figure 2). The deletion in strains 0202, 9688 and 9692 spans the genomic regions from position 191349763 to 191359446, 393559445 to 393569598 and 393569040 to 393569598, respectively.
Strain 15-02 (D[ecfQ, ecfF-osrA]) was constructed as follows: first, the kanamycin resistance cassette in strain 9688 was replaced by a spectinomycin/streptomycin resistance cassette (V) resulting in strain 9715. The cassette exchange was performed by conjugation into strain 9688 plasmid pRJ9715 whose insert corresponds to that of pRJ9688 with the kanamycin resistance cassette replaced by the V cassette inserted between the up-and downstream regions of the ecfF-osrA genes. Mutant strain 15-02 which is deleted for ecfQ and ecfF-osrA was obtained by using plasmid pRJ0202 to introduce the ecfQ deletion into strain 9715 via marker-exchange mutagenesis ( Figure 2). The resulting deletions in strain 15-02 span the same genomic regions as in the individual mutants described above.
The Dblr3042 strain 0203 was constructed by a markerless inframe deletion mutagenesis. This approach was chosen because tiling analysis of microarray data indicated that blr3042 is the promoter-proximal gene of a tricistronic operon consisting of blr3042, blr3043 and blr3044. Flanking regions of blr3042 were cloned into the suicide plasmid pK18mobsacB to yield plasmid pRJ0203. Plasmid pRJ0203 was transferred by conjugation from E. coli S17-1 to B. japonicum 110spc4. Kanamycin resistant exconjugants were selected and grown in the presence of 5% sucrose to force loss of the vector-encoded sacB gene. Resulting colonies were checked for kanamycin sensitivity, and the desired deletion was confirmed by PCR. In the resulting strain 0203 the genomic region from position 393599337 to 393599926 is deleted.
Strain 92-30 served a control and contains vector pSUP202pol4 chromosomally inserted between osrA and bll3040. For its construction, a 451-bp fragment containing the 39 end of bll3040 (genome coordinates 393569640 to 393579075) was PCR amplified using the primer pair listed in Table S6, cloned in the pGEM-T Easy vector, verified by sequencing and re-cloned in pSUP202pol4 resulting in plasmid pRJ9729. Plasmids pRJ9729, pRJ9736, pRJ9737, pRJ9738 and pRJ9730 were transformed into E. coli S17-1 and then mobilized by conjugation into B. japonicum strain 9692 as previously described [71] resulting in mutant strains 92-29, 92-36, 92-37, 92-38 and 92-30, respectively. The correct genomic structure (Figure 2) of the resulting strains was verified by PCR.

DNA Work
Recombinant DNA work was performed according to standard protocols [72]. B. japonicum chromosomal DNA was isolated as described [71].

RNA Extraction and CDNA Synthesis
Harvest and storage of cells, RNA extraction and cDNA synthesis were done as previously described [68].

Quantitative Real-time PCR
Expression of genes ecfQ and ecfF was analyzed by reverse transcription-based quantitative real-time PCR as previously described [41]. RNA was isolated from micro-oxically grown mid-log phase wild-type cells that were either untreated or exposed prior to harvest to one of the following treatments: 2 mM H 2 O 2 for 10 min; 0.2 mM paraquat for 5 or 10 min; 0.5 mM rose bengal plus light exposure (20,000 lux) for 10 or 180 min; exposure to light for 60 min (control). Expression of the ecfF gene was analyzed in strains 92-29, 92-36, 92-37, 92-28 and 92-30 grown microoxically to mid-log phase, either untreated or exposed to 2 mM H 2 O 2 for 10 min prior harvesting. cDNA (0.2 to 20 ng) in combination with 2.5 mM of primers pairs 1028-RT-F and 1028-RT-R or 3038-RT-F/3038-RT-R (Table S6) were used for monitoring expression of ecfQ and ecfF, respectively. The primary s factor gene sigA was used as a reference for normalization (primers SigA-1155R and SigA-1069F; [41]). Data were evaluated by the method of Pfaffl [75].

Primer Extension
The transcription start site of ecfQ and ecfF were determined as previously described [76,77]. RNA was extracted from microoxically grown wild-type cells either non-stressed or treated with 2 mM H 2 O 2 for 10 min. To determine transcription start site of ecfQ cDNAs were synthesized with primers pe-1028-1 or pe-1028-2 (Table S6). The same primers were used to obtain sequencing ladders from plasmid pRJ0211 (Table S5), containing the promoter and part of the ecfQ coding region. Likewise, primers pe-3038-1 and 3038-RT-R and plasmid pRJ9724 were used for determination of the transcription start site of ecfF-osrA.

Microarrays
Global transcription levels were determined as described previously using a custom-designed Affymetrix chip [68,69]. RNA template for cDNA synthesis was isolated from microoxically grown cells of the wild type and mutant strains 0202, 9688 and 9692, and also of H 2 O 2 -treated cells (2 mM, 10 min) for the wild type and two mutant strains (0202, 9688). For each strain and condition, a minimum of three biological replicates was prepared. RNA extraction, cDNA synthesis, fragmentation and labeling were done as described previously [68,78]. GeneChip data analysis was performed using GeneSpring GX 7.3.1 software (Agilent). After filtering for probe sets which were called present or marginal in at least two out of three replicas, a statistical student ttest with a P-value threshold of 0.01 was applied. Genes were considered as differentially expressed if the fold-change value was ,-3 or .+3 when comparing two strains or conditions. Data sets generated in this work are deposited in the GEO database under record number GSE39165.

Bioinformatic Analyses
Phylogenetic analysis of B. japonicum ECF s factors was conducted using MEGA version 4 [79]. For alignment of nucleotide and amino acid sequences, the T-COFFEE program was used (http://www.ebi.ac.uk/Tools/msa/tcoffee/ [80,81]). Results were visualized with GeneDoc [82] and BioEdit [83]. Database searches for regulators that might bind to the up-stream region of ecfQ were done with the virtual footprint tool Prodoric (http://www.prodoric.de; [84]). Search for consensus motifs in EcfQ-and EcfF-target promoters was performed using the BioProspector suite (http://ai.stanford.edu/xsliu/BioProspector/; [85]). DNA sequences corresponding to 200 bp located upstream of the promoter-proximal genes listed in Table 3 were used as input for the analysis. Parameters were set to search for a twoblock motif with 5 nucleotides per block and a gap of 13 to 16 nucleotides between the blocks. Identified sequence motifs were aligned and visualized using the WebLogo tool (http://weblogo. berkeley.edu/; [86]). Genome-wide searches for putative EcfF target promoters focused on 200 bp regions upstream of genes or operons and were performed with the genome-scale DNA pattern search program from the RSAT collection of sequence analysis tools (http://rsat.ulb.ac.be/; [87]) Searches for amino acid sequence similarities were performed with BlastP (http://blast. ncbi.nlm.nih.gov/Blast.cgi?PAGE = Proteins). Topology prediction for OsrA was done with TOPCONS (http://topcons.cbr.su. se/; [88]). Protein localization prediction via a signal peptide search was performed using SignalP 4.0 (http://www.cbs.dtu.dk/ services/SignalP/; [89]).

Bacterial Two-hybrid System
For analysis of EcfF-OsrA interactions the BATCH system was used (Euromedex, Souffelweyersheim, France). Translational fusions of wild-type and mutant versions of OsrA to the Cterminal end of the T25 fragment of Bordetella pertussis adenylate cyclase (Cya) were generated by cloning of PCR-generated PstI-EcoRI fragments into vector pKT25 (resulting in plasmids pRJ9744, pRJ9752, pRJ9753, and pRJ9754; Table S5). Primers are listed in Table S6. For amplification of wild-type osrA, genomic DNA of B. japonicum 110spc4 was used while mutated osrA versions were amplified with plasmids pRJ9736, pRJ9737, or pRJ9738 as templates. In parallel, a translational fusion of EcfF to the C-terminal end of the Cya T18 fragment was generated. To do so, a PstI-XbaI fragment containing the wild-type ecfF gene was amplified (primer pair listed in Table S6, genomic B. japonicum DNA as template) and cloned into vector pUT18C yielding plasmid pRJ9746. All constructed plasmids were verified by sequencing. To study interaction of EcfF with different versions of OsrA, E. coli strain BTH101 was co-transformed with pRJ9746 and one of the plasmids expressing a T25-OsrA fusion. For b-galactosidase activity assays, co-transformed clones were inoculated into 6 ml LB medium containing appropriate antibiotics and 0.5 mM IPTG (isopropyl b-D-1-thiogalactopyranonoside). Cultures were grown for 18 h at 30uC, and aliquot(s) from 50 ml to 200 ml were used to determine b-galactosidase activity as described elsewhere [43]. Figure S1 Alignment of OsrA-homologs. Numbers on the right of each line refer to the position of the last amino acid within the corresponding protein sequence. Shaded in black, dark grey, and light grey are nucleotides which are identical in all, 80%, and 60% of the sequences, respectively. Arrowheads indicate conserved cysteines. GI numbers of the proteins are as follows:  Figure S2 Venn diagram of H 2 O 2 -responsive genes in the B. japonicum wild-type strain and the regulons of ECF s factors EcfQ and EcfF. Hydrogen peroxide-responsive genes were identified by transcriptome analyses of untreated wildtype cells with cells exposed to 2 mM H 2 O 2 for 10 min. Similarly, regulons of EcfQ and EcfF were determined by comparing the transcriptome of DecfQ and D(ecfF-osrA) mutant strains, respective-ly, both treated with 2 mM H 2 O 2 for 10 min, with identically stressed wild-type cells. All strains we grown micro-oxically. Size and overlap of the regulons are drawn to scale with numbers of differentially expressed genes (3-fold change cut-off) indicated in the respective segments. Total number and numbers of down-(Q) and up-regulated genes (q) are shown next to individual regulons. (TIF)

Supporting Information
Table S1 List of 225 B. japonicum genes which are differentially expressed after treatment with 2 mM H 2 O 2 for 10 min in wild-type cells grown micro-oxically in PSY medium as compared to untreated wild-type cells. (DOCX)