Interactions among Redox Regulators and the CtrA Phosphorelay in Dinoroseobacter shibae and Rhodobacter capsulatus.

Bacteria employ regulatory networks to detect environmental signals and respond appropriately, often by adjusting gene expression. Some regulatory networks influence many genes, and many genes are affected by multiple regulatory networks. Here, we investigate the extent to which regulatory systems controlling aerobic–anaerobic energetics overlap with the CtrA phosphorelay, an important system that controls a variety of behavioral processes, in two metabolically versatile alphaproteobacteria, Dinoroseobacter shibae and Rhodobacter capsulatus. We analyzed ten available transcriptomic datasets from relevant regulator deletion strains and environmental changes. We found that in D. shibae, the CtrA phosphorelay represses three of the four aerobic–anaerobic Crp/Fnr superfamily regulator-encoding genes (fnrL, dnrD, and especially dnrF). At the same time, all four Crp/Fnr regulators repress all three phosphorelay genes. Loss of dnrD or dnrF resulted in activation of the entire examined CtrA regulon, regardless of oxygen tension. In R. capsulatus FnrL, in silico and ChIP-seq data also suggested regulation of the CtrA regulon, but it was only with loss of the redox regulator RegA where an actual transcriptional effect on the CtrA regulon was observed. For the first time, we show that there are complex interactions between redox regulators and the CtrA phosphorelays in these bacteria and we present several models for how these interactions might occur.


Introduction
Bacteria sense and process environmental signals in order to adapt to changes in their surroundings. These signals are relayed through regulatory networks that adjust the cells' behavior, often through changes in gene expression. The alphaproteobacterium Dinoroseobacter shibae is a member of the marine roseobacter group and an aerobic anoxygenic photoheterotrophic bacterium, capable of both aerobic and anaerobic respiration [1]. It can be free-living or an algal symbiont [1] and is a metabolically versatile bacterium able to adapt to changes in its highly dynamic environment. For example, at the end of an algal bloom when the oxygen concentration drops, an alternative terminal electron acceptor such as nitrate can be used for respiration [1,2].
The response to the change from aerobic to anaerobic conditions is controlled by four Crp/Fnr transcriptional regulators in D. shibae [3]. Crp/Fnr regulators are widely distributed among bacteria and form a superfamily consisting of 14 phylogenetic subgroups [4]. The versatility of this family is reflected by both the wide range of signals that are sensed, such as temperature [5], oxygen [6], and nitric oxide (NO) [7], and the range of metabolic processes regulated upon activation, which include respiration-related processes and especially the transition between aerobic and anaerobic lifestyles [3,8].

Datasets Analyzed in this Study
Ten published and accessible microarray and RNA-seq transcriptomic datasets for chosen gene knockout strains and experiments monitoring responses to changes in environmental conditions were obtained from the NCBI GenBank database (Table 1). Table 1. Description of the transcriptomic datasets analyzed in this study.

Processing and Analysis of Datasets
This study includes four different types of transcriptomic data ( Table 1) that could not be processed and analyzed as one dataset. We therefore used the changes in transcript levels (log 2 fold change) compared to the controls used in the respective studies (e.g., wild type or time point before changes in the environmental conditions) for each dataset. RNA-seq data from D. shibae (reads per gene) and R. capsulatus (log 2 fold change) were obtained from the respective publications (Table 1).
Agilent microarray datasets were processed using the LIMMA package in R [35]. Background correction was performed with the "normexp" method and an offset of 10. Two-color microarrays were normalized with the "loess" method before quantile normalization. Signals/intensities from spots were averaged.
Affymetrix microarray datasets were processed using the R packages LIMMA, makecdfenv, and affy [35][36][37]. The CDF environment for GSE18149 was generated using the corresponding CDF file downloaded from GEO (accession GPL9198). Data were normalized with the rma function. A linear fit model was generated for comparison.
In order to analyze the CckA and ChpT regulons, thresholds were set that allowed definition of regulated and non-regulated genes. These thresholds were applied to the log 2 fold change in transcript level values in the cckA and chpT deletion mutants. A gene was not considered regulated when its log 2 fold change was between 1 and −1 while a log 2 fold change value above 1 or below −1 indicated an affected gene. The analyzed genes were grouped based on published information about their functional categories as described (Supplementary Table S1).

Overlap of the Crp/Fnr and CtrA Regulons in Dinorosebacter shibae
The possible interaction between the Crp/Fnr regulator and CtrA phosphorelay networks was first assessed using transcriptomic datasets for regulator deletion mutants. The changes of transcript levels of known Crp/Fnr-and CtrA-controlled traits revealed a strong overlap of both regulons, with the regulator-encoding genes themselves affected by losses of the other regulators ( Figure 1). Under both aerobic and anaerobic conditions, the loss of dnrD or dnrF resulted in increased transcript levels of the CtrA phosphorelay, QS, flagellar motility, tad pilus, competence and recombination, gene transfer agent (GTA), divL and c-di-GMP signaling genes ( Figure 1A). In all datasets, the GTA genes showed comparatively small changes in transcript levels ( Figure 1), probably as a result of a small subpopulation actually expressing these genes [13]. Only the loss of dnrF led to a change in gene expression between aerobic and anaerobic conditions, since a greater increase in the transcript levels could be observed under anaerobic conditions for most of its regulon ( Figure 1C). The loss of fnrL or dnrE resulted in increased transcript levels of ctrA, cckA, chpT, luxI 1 , luxR 1 , and luxR 2 but had little to no effect on the downstream CtrA regulon ( Figure 1B).

Overlap of the Crp/Fnr and CtrA Regulons in Dinorosebacter shibae
The possible interaction between the Crp/Fnr regulator and CtrA phosphorelay networks was first assessed using transcriptomic datasets for regulator deletion mutants. The changes of transcript levels of known Crp/Fnr-and CtrA-controlled traits revealed a strong overlap of both regulons, with the regulator-encoding genes themselves affected by losses of the other regulators ( Figure 1). Under both aerobic and anaerobic conditions, the loss of dnrD or dnrF resulted in increased transcript levels of the CtrA phosphorelay, QS, flagellar motility, tad pilus, competence and recombination, gene transfer agent (GTA), divL and c-di-GMP signaling genes ( Figure 1A). In all datasets, the GTA genes showed comparatively small changes in transcript levels ( Figure 1), probably as a result of a small subpopulation actually expressing these genes [13]. Only the loss of dnrF led to a change in gene expression between aerobic and anaerobic conditions, since a greater increase in the transcript levels could be observed under anaerobic conditions for most of its regulon ( Figure 1C). The loss of fnrL or dnrE resulted in increased transcript levels of ctrA, cckA, chpT, luxI1, luxR1, and luxR2 but had little to no effect on the downstream CtrA regulon ( Figure 1B).  Transcriptomic data for genes in selected functional groups in different knockout strains. The four Crp/Fnr regulator knockouts were grown under aerobic (ae) or anaerobic (an) conditions. The log 2 fold changes compared to the respective wild type (WT) (A,B) or against themselves grown at different conditions, are shown (C). The CtrA phosphorelay and quorum sensing system knockouts were grown aerobically to the stationary phase and compared to the WT (D). The functional group assignments on the right are based on published information as described in Supplementary Table S1. Note: the ∆luxI 1 strain retains a portion of the gene that can therefore result in mapped reads. Almost all examined genes showed an opposite pattern in the CtrA phosphorelay and luxI 1 mutants ( Figure 1D) compared to dnrD and dnrF ( Figure 1A). Most of the genes showed decreased transcript levels in strains lacking any of the CtrA phosphorelay genes, with the exceptions of the Crp/Fnr regulators where the largest increase was found for dnrF ( Figure 1D). Loss of luxI 1 resulted in increased transcript levels for fnrL, dnrD, and dnrF, but no changes were observed for dnrE ( Figure 1D).

The Role of ChpT in Signal Integration
In D. shibae, deletion of neither ctrA nor cckA had an influence on expression of chpT, whereas the loss of either ctrA or chpT resulted in decreased expression of cckA ( Figure 1D) [11]. However, all three CtrA phosphorelay component genes showed reduced transcript levels in the absence of the AHL synthase luxI 1 ( Figure 1D) [13], whereas loss of the Crp/Fnr regulators resulted in increased transcript levels of these genes ( Figure 1A,B). Therefore, in contrast to ctrA and ccka, chpT is not regulated by the CtrA phosphorelay itself, but by other factors that can thereby control the phosphorylation state of CtrA. These findings also suggest that chpT transcription is regulated oppositely by QS and the Crp/Fnr regulators. This is supported by binding site predictions for FnrL [3] that suggest it binds at the promoter of chpT and clpX, which encodes a protease known to cleave CtrA [3,38,39]. Deletion of fnrL strongly increased the expression of chpT but only resulted in minimal changes for clpX ( Figure 1B). Binding site prediction for the Dnr regulators did not find any binding sites near clpX or the CtrA phosphorelay genes [3].
It was previously found that more genes were affected by the loss of chpT than cckA [11], suggesting ChpT regulates some genes independent of CckA and that a different kinase might regulate its activity and thereby affect downstream gene expression. Among the genes affected by the loss of chpT but not cckA, dnrF was the most upregulated gene during exponential growth while lexA and recA were among those most downregulated genes in both exponential and stationary phases ( Figure 2). Although there was a small increase in transcript levels of dnrF in the cckA deletion strain during exponential growth, it did not pass the threshold we defined (see Materials and Methods). These findings suggest a link between dnrF and chpT.
Additional discrepancies between CckA and ChpT are apparent from their opposing effects on the nap gene cluster during exponential growth (Figure 2A), although this is not maintained in stationary phase ( Figure 2B). In exponential phase, loss of cckA led to decreased transcript levels of the nap gene cluster, while the loss of ctrA and chpT led to increased levels ( Figure 2A). This cluster is the only denitrification cluster activated by FnrL but repressed by the three Dnr regulators [3]. Interestingly, transcript levels of all four denitrification gene clusters were increased in the AHL synthase knockout ∆luxI 2 but were unaffected in ∆luxI 1 ( Figure 2C).

Time-Resolved Evaluation of Environmental Changes and the Regulation of c-di-GMP Signaling Genes
Interactions between the networks in D. shibae were further analyzed using time-resolved transcriptomic datasets. These were collected following the switch from aerobic to anaerobic conditions in wild type cells ( Figures 3A and 4A) [31], following the external addition of AHL autoinducer to the AHL synthase mutant ∆luxI 1 (Figures 3B and 4B) [13], and through the culture growth phases for ∆luxI 1 in the absence of AHL ( Figure 4C) [14]. Additional discrepancies between CckA and ChpT are apparent from their opposing effects on the nap gene cluster during exponential growth (Figure 2A), although this is not maintained in stationary phase ( Figure 2B). In exponential phase, loss of cckA led to decreased transcript levels of the nap gene cluster, while the loss of ctrA and chpT led to increased levels ( Figure 2A). This cluster is the only denitrification cluster activated by FnrL but repressed by the three Dnr regulators [3]. Interestingly, transcript levels of all four denitrification gene clusters were increased in the AHL synthase knockout ΔluxI2 but were unaffected in ΔluxI1 ( Figure 2C).

Time-Resolved Evaluation of Environmental Changes and the Regulation of c-di-GMP Signaling Genes
Interactions between the networks in D. shibae were further analyzed using time-resolved transcriptomic datasets. These were collected following the switch from aerobic to anaerobic conditions in wild type cells ( Figures 3A and 4A) [31], following the external addition of AHL autoinducer to the AHL synthase mutant ΔluxI1 ( Figures 3B and 4B) [13], and through the culture growth phases for ΔluxI1 in the absence of AHL ( Figure 4C) [14]. Samples for the ctrA, cckA, and chpT knockout mutants were analyzed at mid-exponential (OD 0.4) (A) and stationary (six hours after onset of stationary phase) (B) phases of growth. The ∆luxI 1 data were obtained during stationary phase, six hours after the onset of stationary phase, and the ∆luxI 2 data were obtained during the mid-exponential growth phase (C).
Upon the shift to anaerobic conditions, all three dnr genes showed an immediate increase in transcript levels for 30 min and then stayed constant, whereas those of fnrL decreased ( Figure 4A). These changes corresponded with increased transcript levels of the denitrification gene clusters, with the nap cluster showing a slightly different pattern than the nir and nos clusters ( Figure 3A). Slight increases were observed for the c-di-GMP signaling, flagellar, tad pilus, and QS genes ( Figure 3A). Four of the five c-di-GMP signaling genes showed increased transcript levels following the transfer to an anaerobic environment, whereas dgc2 showed a slight decrease ( Figure 4A).
The addition of AHL to the ∆luxI 1 strain led to increased transcript levels for all CtrA-and QS-controlled genes ( Figure 3B). This included the CtrA phosphorelay and c-di-GMP signaling genes, with dgc2 showing the largest increase ( Figure 4B). No effect was visible for the Crp/Fnr regulator-encoding genes ( Figure 4B) and only a minor increase of the nap gene cluster was observed among the denitrification genes ( Figure 3B).
Due to the increased transcript levels observed for CtrA regulon genes in the dnrD and dnrF deletion strains, it was expected that the same genes would also be decreased under anaerobic conditions. Instead, it turned out that the change from aerobic to anaerobic conditions ( Figure 3) resulted in increased transcript levels for these genes. However, this increase was small, and effects were not observed for some genes that appeared to be controlled by the individual regulators based on the knockout transcriptomic data (Figure 1). This included the regulation of the CtrA phosphorelay genes by the Crp/Fnr regulators. Vice versa, loss of the CtrA phosphorelay genes indicated their repression of Crp/Fnr regulator gene expression ( Figure 1D), but the contrary was observed in the respective physiological datasets where the Crp/Fnr regulators seem to be upregulated ( Figure 3B). Notably however, in both physiological datasets, dgc2 stands out as distinctly affected compared to other c-di-GMP signaling genes ( Figure 4A,B). Also, in the non-induced ∆luxI 1 culture, no influence of the QS null mutant on the Crp/Fnr regulators was observed, but the CtrA phosphorelay and c-di-GMP signaling genes were down-regulated ( Figure 4C). Upon the shift to anaerobic conditions, all three dnr genes showed an immediate increase in transcript levels for 30 min and then stayed constant, whereas those of fnrL decreased ( Figure 4A). These changes corresponded with increased transcript levels of the denitrification gene clusters, with the nap cluster showing a slightly different pattern than the nir and nos clusters ( Figure 3A). Slight increases were observed for the c-di-GMP signaling, flagellar, tad pilus, and QS genes ( Figure 3A). Four of the five c-di-GMP signaling genes showed increased transcript levels following the transfer to an anaerobic environment, whereas dgc2 showed a slight decrease ( Figure 4A). Interestingly, in contrast to fnrL, dgc2, and chpT, the other Crp/Fnr regulators, c-di-GMP signaling, and CtrA phosphorelay genes all decreased at the onset of the stationary phase ( Figure 4C). Moreover, analysis of the Crp/Fnr knockout data showed that the loss of dnrF or dnrD resulted in increased transcript levels of four of the c-di-GMP signaling genes under anaerobic growth conditions, with only dgc2 being unaffected ( Figure S1A). Loss of luxI 1 and the CtrA phosphorelay genes resulted in decreased transcripts for all five genes ( Figure S1B,C), although the effects on dgc2 were smaller than for the other genes in the stationary phase ( Figure S1C).  The addition of AHL to the ΔluxI1 strain led to increased transcript levels for all CtrA-and QScontrolled genes ( Figure 3B). This included the CtrA phosphorelay and c-di-GMP signaling genes, with dgc2 showing the largest increase ( Figure 4B). No effect was visible for the Crp/Fnr regulatorencoding genes ( Figure 4B) and only a minor increase of the nap gene cluster was observed among the denitrification genes ( Figure 3B).
Due to the increased transcript levels observed for CtrA regulon genes in the dnrD and dnrF deletion strains, it was expected that the same genes would also be decreased under anaerobic conditions. Instead, it turned out that the change from aerobic to anaerobic conditions ( Figure 3) resulted in increased transcript levels for these genes. However, this increase was small, and effects were not observed for some genes that appeared to be controlled by the individual regulators based on the knockout transcriptomic data (Figure 1). This included the regulation of the CtrA phosphorelay genes by the Crp/Fnr regulators. Vice versa, loss of the CtrA phosphorelay genes indicated their repression of Crp/Fnr regulator gene expression ( Figure 1D), but the contrary was observed in the respective physiological datasets where the Crp/Fnr regulators seem to be upregulated ( Figure 3B). Notably however, in both physiological datasets, dgc2 stands out as distinctly affected compared to other c-di-GMP signaling genes ( Figure 4A,B). Also, in the noninduced ΔluxI1 culture, no influence of the QS null mutant on the Crp/Fnr regulators was observed, but the CtrA phosphorelay and c-di-GMP signaling genes were down-regulated ( Figure 4C).
Interestingly, in contrast to fnrL, dgc2, and chpT, the other Crp/Fnr regulators, c-di-GMP signaling, and CtrA phosphorelay genes all decreased at the onset of the stationary phase ( Figure  4C). Moreover, analysis of the Crp/Fnr knockout data showed that the loss of dnrF or dnrD resulted in increased transcript levels of four of the c-di-GMP signaling genes under anaerobic growth conditions, with only dgc2 being unaffected ( Figure S1A). Loss of luxI1 and the CtrA phosphorelay genes resulted in decreased transcripts for all five genes ( Figure S1B,C), although the effects on dgc2 were smaller than for the other genes in the stationary phase ( Figure S1C).

Effects on the CtrA Regulon during Coculture of Dinoroseobacter shibae and Its Algal Host
In the two-phase interaction of D. shibae with its dinoflagellate host Prorocentrum minimum [14,40], a mutualistic growth phase (0 to 21 days of cocultivation) is followed by a pathogenic growth phase (21 to 30 days of cocultivation) that leads to death of the algae [15]. Analysis of the transcriptomic data of D. shibae during cocultivation showed an overall increase in the transcription for the CtrA regulon genes during the transition between the two phases (day 24 compared to day 18), followed by a decrease during the late-pathogenic phase, after 30 days (Supplementary Figure S2).
Of the CtrA phosphorelay genes, only chpT remained upregulated during the pathogenic interaction. Evaluation of the denitrification gene clusters showed strong variation among these genes (Supplementary Figure S2), likely arising from overall low expression levels, and this made it difficult to draw any conclusions.

RegA Activates the CtrA Regulon in Rhodobacter capsulatus
Next, we asked if the observed overlap between redox regulators and the CtrA phosphorelay system is conserved in another member of the family Rhodobacteraceae. For R. capsulatus, transcriptomic data were available for knockout mutants of ctrA, cckA, and the known redox regulator-encoding genes fnrL, regA, and crtJ. We identified three additional Crp/Fnr regulator-encoding genes in this bacterium based on blast searches (RCAP_rcp00107, RCAP_rcc01561, RCAP_rcc03255), but these genes showed no evidence of differential regulation in any of the analyzed datasets and we did not consider them further. A blast search also identified a homologue (RCAP_rcc02630) of the HNOX-encoding gene of D. shibae (Dshi_2815). This gene encodes a protein with a predicted heme nitric oxide/oxygen binding (HNOB) domain and is located adjacent to a c-di-GMP signaling gene (RCAP_rcc02629) that was recently demonstrated to affect GTA production and motility in R. capsulatus [41]. When bound to NO, the HNOX homologue in D. shibae inhibits the activity of the diguanylate cyclase Dgc1, which is encoded by the neighboring gene [23].
FnrL is the only Crp/Fnr regulator that has been studied in R. capsulatus [29]. Its loss did not result in any large changes in transcript levels for the examined traits under anaerobic phototrophic conditions ( Figure 5), and the same was observed for the loss of crtJ, which encodes a transcription factor that controls numerous photosynthesis and cytochrome genes [32] (Figure 5). RegA is the response regulator of the RegB/A two-component system that controls photosynthesis, nitrogen and carbon fixation, denitrification, and respiration genes in response to oxygen availability [26]. In contrast to fnrL and crtJ, we found that the loss of regA resulted in a strong decrease in transcript levels of the CtrA regulon genes ( Figure 5), indicating that RegA acts as a direct or indirect activator of these genes. Like the genes involved in regulation of photosynthesis and the change between aerobic/anaerobic lifestyle in D. shibae, loss of regA affected chpT the most among the CtrA phosphorelay genes in R. capsulatus. Loss of the CtrA phosphorelay genes had no influence on transcription of fnrL, regA, regB, or the other putative Crp/Fnr regulator-encoding genes (Supplementary Figure S3). A comparison of photosynthetic anaerobic growth and aerobic cultivation in R. capsulatus showed the CtrA-regulated traits have reduced transcript levels under anaerobic conditions ( Figure 5).

The Crp/Fnr and CtrA/QS Regulons Overlap in Dinoroseobacter shibae
Our analysis revealed an inverse regulatory crosstalk between the Crp/Fnr and CtrA systems in D. shibae. We found the denitrification gene clusters and Crp/Fnr regulator genes, especially dnrF, to be part of the CtrA phosphorelay and LuxI2 regulons. DnrE was affected exclusively by loss of LuxI2, whereas loss of LuxI1 only had minor effects on fnrL, dnrD, and dnrF and no effect on dnrE. In addition to their regulation by LuxI1, which signals cell density, the Crp/Fnr regulators integrate oxygen and NO levels and affect all three CtrA phosphorelay genes.
Until now, overlapping regulation by the Crp/Fnr and CtrA systems has only been noted in D.

The Crp/Fnr and CtrA/QS Regulons Overlap in Dinoroseobacter shibae
Our analysis revealed an inverse regulatory crosstalk between the Crp/Fnr and CtrA systems in D. shibae. We found the denitrification gene clusters and Crp/Fnr regulator genes, especially dnrF, to be part of the CtrA phosphorelay and LuxI 2 regulons. DnrE was affected exclusively by loss of LuxI 2 , whereas loss of LuxI 1 only had minor effects on fnrL, dnrD, and dnrF and no effect on dnrE. In addition to their regulation by LuxI 1 , which signals cell density, the Crp/Fnr regulators integrate oxygen and NO levels and affect all three CtrA phosphorelay genes.
Until now, overlapping regulation by the Crp/Fnr and CtrA systems has only been noted in D. shibae for flagellar genes and recA [3,12,13], and to our knowledge this level of regulatory interaction has not been reported for alphaproteobacteria. However, a comparable connection between QS and Crp/Fnr regulators has been documented for the gammaproteobacterium Pseudomonas aeruginosa where the regulons of the FnrL homolog Anr and QS synthase LasR overlap. Here, denitrification genes are induced by Anr and inhibited by LasR. Additionally, in the absence of lasR, Anr regulates production of the QS molecule 4-hydroxy-2-alkylquinoline [42]. At the protein level, nitrite reductase (NirS), a flagellar protein (FliC), and the chaperone DnaK form a complex that influences flagellar formation and motility and thus creates a link between denitrification and motility [43]. In cystic fibrosis infections, P. aeruginosa is exposed to ambient conditions with low oxygen tension. The intracellular levels of c-di-GMP increase, which leads to biofilm formation. These conditions also lead to an increase in mutations in the QS transcriptional regulator-encoding gene lasR. As lasR deletion strains grow to higher cell densities and have higher denitrification rates, it has been suspected that these mutations increase the fitness of the population during infection [44][45][46].
Combined, these observations indicate that there may be a more widely conserved interaction of Crp/Fnr regulators and QS in proteobacteria. The CtrA phosphorelay is unique to alphaproteobacteria, indicating a potential independent evolution of this regulatory crosstalk in this lineage.

Inverse Control of the CtrA Regulon by RegA and Anaerobic Photosynthetic Growth Conditions in Rhodobacter capsulatus
In R. capsulatus, the regulons of the redox-responsive two-component system RegA/B [47] and the CtrA phosphorelay overlap. Interestingly, chpT stands out because it is the only CtrA phosphorelay gene that is regulated by RegA. Similar to Dnr and Fnr in D. shibae, RegA controls the expression of photosynthesis and respiration genes [26]. ChIP-seq with RegA identified binding sites adjacent to several genes also targeted by CtrA: RCAP_rcc02857 (a c-di-GMP signaling gene involved in GTA production) and its divergently transcribed neighbor (RCAP_rcc02856), RCAP_rcc02683 (a chemotaxis gene), and dksA (a dnaK suppressor gene) [34].
As in D. shibae, transcriptomic data from a fnrL deletion strain showed no effects on the CtrA-controlled traits outside of the CtrA phosphorelay genes themselves. However, ChIP-seq and in silico binding site predictions [29] suggest FnrL binding adjacent to divL, dnaK, recA, flagellar gene clusters, the RcGTA capsid protein-encoding gene, and c-di-GMP signaling genes (including those affecting RcGTA production [41]). Similarly, ChIP-seq with CrtJ [48], a regulator controlling expression of multiple genes involved in photosynthesis, also revealed a connection to the CtrA phosphorelay. Even though the observed transcript level changes in the crtJ mutant were small, binding was found adjacent to ctrA, clpX, a luxR family gene, dnaA, spoT, ftsZ, and the first gene in the GTA structural gene cluster (RCAP_rcc01682) under aerobic and anaerobic cultivation. Binding sites adjacent to dnaK and two flagellar genes (flgB and flaA) were identified under aerobic and anaerobic conditions, respectively.
In D. shibae, deletions of the Crp/Fnr regulator-encoding genes indicated an inhibition of the CtrA regulon, but the physiological changes detected by these regulators (switch from aerobic to anaerobic conditions) showed a tendency towards activation of the CtrA regulon. The same was observed for the deletion mutants of the CtrA phosphorelay components and their regulation of the Crp/Fnr regulator genes. In R. capsulatus, we could observe a similar pattern but in reverse for regulation of the CtrA regulon by RegA. While the regA knockout indicated activation of the CtrA regulon, the switch to anaerobic photosynthetic growth conditions showed an inhibition. This is probably indicative of a more complex interaction among these regulatory systems. However, the regA deletion transcriptomic data are supported by in vivo motility tests that showed reduced swimming ability of the ∆regA strain [26].

Integration of Crp/Fnr Regulation into the CtrA Phosphorelay and Regulon
In D. shibae, CtrA binding site predictions and expression data for ctrA and cckA suggest that CtrA directly regulates its own expression and that of cckA, but not chpT [13]. Therefore, chpT transcription must be regulated from outside of the CtrA phosphorelay and upstream of CtrA. Both, regulatory control of chpT and signal integration upstream of CtrA is known for LuxI 1 [11]. A similar situation might be possible for Crp/Fnr signal integration due to their regulation of chpT ( Figure 6A). Since chpT is the only RegA-regulated CtrA phosphorelay gene in R. capsulatus (and it has a RegA binding site), it seems to play a central role here, too. However, there are also RegA binding sites associated with clpX and other genes of the CtrA regulon [26]. Interestingly, the Dnr/Fnr binding site in the nosR2 promoter in D. shibae has the sequence 5 -TTAAC-N4-GTCAA-3 [3], which shares a half-site binding motif with CtrA 5 -TTAAC-N5-GTTAAC-3 [11]. Previously, comparison between transcriptional regulation and the presence of full and half-site motifs revealed the potential importance of half-site motifs for transcriptional control by CtrA in R. capsulatus [34]. Thus, CtrA and Fnr regulators might interact with some of the same/overlapping sequences ( Figure 6B). A distinct role for ChpT is supported by the observation that loss of chpT or ctrA but not cckA results in decreased transcript levels of dnrF. It is possible that ChpT integrates signals from more than one kinase into its regulation of CtrA. To our knowledge, the only other instance of a histidine kinase affecting phosphorylation of CtrA via ChpT is CcsA from Sphingomonas melonis [49]. Potential homologues of CcsA are encoded in D. shibae (Dshi_1893) and R. capsulatus (RCAP_rcc02545), but effects on transcript levels of these genes were not observed in any of the analyzed datasets. This does not exclude their involvement but also does not allow us to draw further conclusions ( Figure 6C).

Crp/Fnr Regulation of the CtrA Regulon is Largely Independent of Oxygen Tension
Among the Crp/Fnr regulators, only loss of the NO-sensing DnrF resulted in higher inhibition activity of the CtrA system under anaerobic conditions. In P. aeruginosa, swimming motility is controlled anaerobically and aerobically and it was suggested that NirS promotes motility in multiple ways, at the protein level or via signaling pathways, depending on oxygen availability [50]. Regulation of QS traits by both NO and oxygen was also found in the interaction of Vibrio fischeri with the light organ of its squid host. Here, NO released by the host's immune system regulates the symbionts' settlement via biofilm production while the host's control of oxygen availability regulates bacterial bioluminescence in a circadian manner [51][52][53]. However, since the Crp/Fnr knockout and physiological change transcriptomic data have opposite effects on the CtrA phosphorelay (inhibition indicated by the knockouts and activation by the shift to anaerobic growth), it is difficult to determine the role of oxygen on the CtrA phosphorelay. In R. capsulatus, the knockout transcriptomic data were supported by in vivo experiments, so if the knockout transcriptomic data also reflect the actual CtrA regulon in D. shibae, the Crp/Fnr regulators have an inhibitory effect on the CtrA phosphorelay and its regulon. It is known that Dinoroseobacter establishes a mutualistic symbiosis with its dinoflagellate host via the CtrA phosphorelay and by means of flagella. It is possible this interaction is repressed towards the end of an algal bloom when oxygen concentrations change, resulting in downregulation of flagella (and other CtrA-regulated traits) via Crp/Fnr regulation. A distinct role for ChpT is supported by the observation that loss of chpT or ctrA but not cckA results in decreased transcript levels of dnrF. It is possible that ChpT integrates signals from more than one kinase into its regulation of CtrA. To our knowledge, the only other instance of a histidine kinase affecting phosphorylation of CtrA via ChpT is CcsA from Sphingomonas melonis [49]. Potential homologues of CcsA are encoded in D. shibae (Dshi_1893) and R. capsulatus (RCAP_rcc02545), but effects on transcript levels of these genes were not observed in any of the analyzed datasets. This does not exclude their involvement but also does not allow us to draw further conclusions ( Figure 6C).

Crp/Fnr Regulation of the CtrA Regulon is Largely Independent of Oxygen Tension
Among the Crp/Fnr regulators, only loss of the NO-sensing DnrF resulted in higher inhibition activity of the CtrA system under anaerobic conditions. In P. aeruginosa, swimming motility is controlled anaerobically and aerobically and it was suggested that NirS promotes motility in multiple ways, at the protein level or via signaling pathways, depending on oxygen availability [50]. Regulation of QS traits by both NO and oxygen was also found in the interaction of Vibrio fischeri with the light organ of its squid host. Here, NO released by the host's immune system regulates the symbionts' settlement via biofilm production while the host's control of oxygen availability regulates bacterial bioluminescence in a circadian manner [51][52][53]. However, since the Crp/Fnr knockout and physiological change transcriptomic data have opposite effects on the CtrA phosphorelay (inhibition indicated by the knockouts and activation by the shift to anaerobic growth), it is difficult to determine the role of oxygen on the CtrA phosphorelay. In R. capsulatus, the knockout transcriptomic data were supported by in vivo experiments, so if the knockout transcriptomic data also reflect the actual CtrA regulon in D. shibae, the Crp/Fnr regulators have an inhibitory effect on the CtrA phosphorelay and its regulon. It is known that Dinoroseobacter establishes a mutualistic symbiosis with its dinoflagellate host via the CtrA phosphorelay and by means of flagella. It is possible this interaction is repressed towards the end of an algal bloom when oxygen concentrations change, resulting in downregulation of flagella (and other CtrA-regulated traits) via Crp/Fnr regulation.

The Role of c-di-GMP
Multiple eukaryotic hosts are known to use NO for communication with microbial symbionts. In some of the characterized systems, NO is sensed by HNOX proteins, which then control c-di-GMP signaling proteins or histidine kinases encoded by genes adjacent to the HNOX-encoding gene. For example, in Vibrio harveyi, the HNOX-neighboring histidine kinase phosphorylates the QS transcription regulator LuxU [20], and in D. shibae, HNOX inhibits the c-di-GMP signaling enzyme Dgc1 [23]. However, D. shibae also has a second c-di-GMP synthesizing enzyme, Dgc2. During adaptation to anaerobic cultivation and at the onset of stationary phase, dgc2 transcriptional patterns were similar to chpT and fnrL. The transcript levels of these three genes plateaued, whereas those of the other c-di-GMP signaling, CtrA phosphorelay and Crp/FnrL genes decreased. A unique regulation of dgc2 was also observed in the dnrF, dnrD, cckA, and chpT knockout strains. Thus, both networks (Crp/Fnr and CtrA phosphorelay) regulate dgc2 and affect its expression in a similar manner as a response to the onset of stationary phase.
The role of dgc2 in the CtrA phosphorelay and FnrL networks and how it might connect both remain to be clarified. For example, it is possible that phosphorylation of the receiver domain of Dgc2 regulates its c-di-GMP synthase activity. As a result, regulation by the Crp/Fnr or CtrA phosphorelay systems could have different effects on the shared traits ( Figure 6D).

Conclusions
In this study we show that regulation of the CtrA regulon, including traits related to phenotypic heterogeneity, is additionally controlled by the aerobic-anerobic regulators Crp/Fnr in D. shibae and by FnrL/RegA in R. capsulatus. This finding is especially important for the understanding of the metabolically flexible lifestyles of these bacteria. The analysis of the available transcriptomic datasets revealed multiple possible integration sites of the Crp/Fnr signal into the CtrA phosphorelay, but a final explanation is still elusive based on these data. Nevertheless, this investigation provides the first insights into the integration of a second environmental signal into the CtrA phosphorelay and demonstrates a strong transcriptional connection between QS, CtrA-regulated traits, and Crp/Fnr regulators in alphaproteobacteria, which has an interesting parallel with QS and Crp/Fnr regulators in a second class of proteobacteria. To our knowledge, D. shibae and R. capsulatus are the first two organisms where both Dnr and HNOX NO-sensor proteins have been studied. Further investigation is necessary to clarify the interaction between the CtrA phosphorelay and the Crp/Fnr regulators. For example, it would be helpful to confirm if an additional kinase is indeed regulating ChpT in these bacteria.
Supplementary Materials: The following are available online at http://www.mdpi.com/2076-2607/8/4/562/s1: Figure S1. Transcript level changes of D. shibae c-di-GMP signaling genes; Figure S2. Comparison of changes in transcript levels during different stages of the "Jekyll and Hyde" interaction between Dinoroseobacter shibae and the dinoflagellate Prorocentrum minimum; Figure S3. Transcript level changes of various FnrL-and RegA-related genes in CtrA phosphorelay mutants during exponential and stationary phases of growth in Rhodobacter capsulatus; Table S1. Assignment of genes into functional categories.