A deletion mutant of Fur was constructed from wild-type S. oneidensis strain MR-1. The fur mutant formed smaller colonies than wild-type strain after two days' incubation at 30°C (Fig. 1), suggestive of a growth deficiency caused by Fur inactivation. Moreover, colonies of the fur mutant appeared paler in color than MR-1 colonies. Wild-type S. oneidensis colonies have characteristic reddish or pink pigments, which is attributed to high heme content in the cells 17. The pale color of colonies of the fur mutant suggested that there was less heme, presumably due to the low level of intracellular iron as a heme co-factor. The low level of intracellular iron has been documented in an E. coli fur mutant 26.

To test the role of Fur in the response of S. oneidensis to iron, we grew the fur mutant and the wild-type strain in LB medium with different concentrations of the iron chelator 2, 2'-dipyridyl to create iron depletion conditions. LB medium contains ~17 uM iron, hence it is considered as an iron-rich medium 26. In LB, the fur mutant displayed longer lag phase, slower growth rate at logarithmic phase and lower cell density at stationary phase than MR-1 (Fig. 2). When iron was depleted from the growth medium by addition of iron chelator, both strains clearly displayed growth inhibition. Interestingly, when both strains were grown with 160 uM iron chelator, no significant difference was detected between the mutant and MR-1. With 240 uM iron chelator, the fur mutant displayed a much shorter lag phase than MR-1 and reached a higher cell density at stationary phase, suggesting that the fur mutant better tolerated the iron depletion stress.

Fur is a pleiotropic transcriptional factor controlling the expression of both iron-regulated and non-iron-regulated genes 9. Fur has been implicated in acid tolerance response, since E. coli and Salmonella typhimurium harboring fur mutations are acid-sensitive 2728. Therefore, the acid tolerance of S. oneidensisfur mutant was examined. Wild-type and the fur mutant were grown in LB medium buffered at pH5.5 and 7 (Fig. 3). At pH of 5.5, the fur mutant but not wild-type strain had severe growth defect, demonstrating that the fur mutant was more sensitive to acidic condition.

For microarray experiments, mid-log phase grown fur mutant was treated with iron chelator for one hour and then ferrous sulfate. The concentration of 160 uM iron chelator was used for iron depletion during the microarray experiments because growth was clearly inhibited, but not completely abolished at this concentration (Fig. 2). Samples were collected at multiple time points during iron depletion and repletion and used for global transcriptomic analyses. To allow for comparison of any pairs of samples, S. oneidensis genomic DNA was used as a common reference in each microarray experiment (see Methods and Materials for details). This strategy has been successfully employed in several studies 29303132. In addition, since S. oneidensisfur mutant in previous studies was generated from a rifampin-resistant strain (DSP10) instead of wild-type S. oneidensis MR-1, we decided to repeat the microarray comparison between the fur mutant and wild-type under aerobic condition, using the fur mutant derived from MR-1. Therefore, MR-1 was grown in LB to mid-logarithmic phase, RNA was prepared and used for microarray experiments. This allowed for comparison of gene expression profiles of the fur mutant and MR-1 (see below). To evaluate the reliability of microarray data, quantitative RT-PCR was performed on a few selected genes (Additional file Additional file 1). A high Pearson correlation coefficient of 0.92 was observed between RT-PCR and microarray results.

Global gene expression profiles of the fur mutant and MR-1 grown in LB medium were compared. A total number of 118 and 56 genes were significantly up- and down- regulated, respectively (Fig. 4). As commonly seen in microarray datasets, a large portion of up- or down-regulated genes corresponded to genes with unknown function, indicating a much broader modulon than deduced by gene annotation alone. Among the genes with defined functions (Table 1), two groups of genes were most noticeable: genes encoding iron acquisition proteins and ribosome proteins. The high level induction of iron acquisition genes agreed with the expected role of Fur in regulating iron uptake and transport. It provided an explanation for the better growth of the fur mutant under iron-depleted condition (Fig. 2), since the up-regulation of iron acquisition genes imposed an advantage for rapid adaptation to the iron depletion. The induction of genes encoding ribosomal proteins was rather unexpected, as the fur mutant grown in LB medium displayed growth defects. It was possible that these genes were induced to accommodate the requirement to synthesize iron acquisition proteins induced by iron depletion. Alternatively, their induction could be due to cellular response to overcome the growth deficiency in LB. Notably, a conserved Fur binding motif could be identified in the promoter region of almost all of the iron acquisition genes in Table 1, while no ribosome subunit gene except SO0232 had a Fur binding motif in its promoter (data not shown). Therefore, it was likely that genes encoding ribosome proteins were not the direct targets of Fur. Lastly, several acid resistance systems have been characterized in γ-proteobacteria 3334. However, expression of components of these systems (e.g. Crp, RpoS, glutamate-, arginine- and lysine-decarboxylases) was not significantly altered in the fur mutant. Since the acid tolerance of the fur mutant was reduced, it was possible that these genes were regulated at post-transcriptional level, or the core components of these systems were not yet annotated in S. oneidensis, or S. oneidensis could employ a novel and yet unidentified mechanism for acid tolerance response.

The microarray analyses of the fur mutant indicated that fewer genes was significantly regulated at early time points than later in both iron depletion and repletion, and there were more down regulated genes than up regulated genes during iron depletion (Fig. 5A). Few genes encoding iron acquisition or ribosome proteins were among those up or down regulated, suggesting that their regulation was strictly Fur-dependent. Meanwhile, a group of genes related to anaerobic energy transport was repressed by iron depletion (Fig. 5B), but not identified in Fur regulon shown above (Table 1). Therefore, this process was iron responsive but Fur independent. These genes include formate dehydrogenase and multiple c-type cytochromes that are involved in anaerobic energy respiration. For example, FccA is a fumarate reductase; CymA is a key protein controlling respiration with a variety of electron acceptors, while MtrC/OmcB is an outer membrane protein required for metal reduction. A logical explanation of the repression by iron is that under the tested condition, these genes are non-essential, iron-using proteins. Upon iron depletion, they are repressed to release previously sequestrated iron to increase free intracellular iron pool. This mechanism has been well documented in E. coli435 as mediated by Fur and a small RNA RyhB. However, in S. oneidensis this biological process must be controlled by iron-responsive regulator(s) other than Fur. Another notable iron-responsive and Fur-independent process was aerobic energy transport, including TCA cycle enzymes malate dehydrogenase (Mdh), succinate dehydrogenase (SdhC), isocitrate dehydrogenase (SO1538) and citrate synthase (GltA). Other related genes include succinylarginine dihydrolase(AstB) and isovaleryl-CoA dehydrogenase (Ivd). Notably, in contrast to the genes involved in anaerobic energy transport, these genes were induced by iron depletion (Fig. 5C), suggesting that it was more energy-consuming for coping with iron starvation than the energetic need for normal growth.

Genes with the highest fold changes are not necessarily the most important genes for a given condition, since there usually is a dissociation between gene expression and physiological phenotype 36. Moreover, a large number of significantly regulated genes are linked to hypothetical proteins. To gain more insights into their functions, a gene co-expression network was constructed (Additional file Additional file 2A). Microarray data have frequently been used for gene annotation based on the observation that genes with similar expression patterns are more likely to be functionally related, that is, “guilt by association”, despite the fact that co-expression does not necessarily indicate a causal relationship at the transcriptional level. This fact is exploited in the “majority-rule” method of network annotation in which a gene is annotated based on the most commonly occurring functions of its co-expressed partners.

Functionally related genes were indeed grouped together in all clusters of the gene network that we constructed (Additional file Additional file 2A). For example, one cluster contains eleven genes encoding heat shock proteins (DnaK, DnaJ, HslV, HslU, HtpG, Lon, SecB, PrlC, ClpB, GroEL and GroES) and three unknown genes (Fig. 6A). These heat shock proteins function to accelerate turnovers of denatured or unfolded proteins and respond to multiple stress conditions 37. Their identification in this study is expected since lack of iron as protein cofactor would impair the functionality of a number of proteins. Furthermore, it can be predicted based on the cluster that the unknown genes in this cluster might function in stress response. This prediction is further supported by their operon structures. SO4161 is immediately upstream of heat shock HslU and HslV and might form an operon together. Similarly, SO2017 could form another operon with the upstream gene encoding heat shock protein HtpG. Finally, when the Gibbs Motif Sampler 38 was employed to search the promoter regions of genes in this cluster for common sequence motif(s), a conserved RpoH (σ³²) binding site was identified (Fig. 6A). In E. coli, RpoH directs the RNAP holoenzymes to specifically transcribed genes encoding heat shock proteins. Its presence in the promoter regions of SO2017 and SO4161 provides further evidence that these two genes play a role in stress response.

Another large cluster contains many c-type cytochromes (e.g. CymA, MtrA, MtrC and OmcA) and their related genes such as heme exporters and alcohol dehydrogenases (Fig. 6B). Many of them were considered as iron-responsive but Fur-independent, as discussed in the previous section. When the Gibbs Motif Sampler was applied, a palindromic sequence of “AAATGTGATCNNGNTCACA NTT” (Fig. 6B) was identified with statistical significance. This sequence is almost identical to the known binding motif of an E. coli global transcriptional factor Crp (AAATGTGATCTAGATCACA TTT) 39. In S. oneidensis, crp mutants are deficient in reducing Fe(III), Mn(IV), nitrate, fumarate and DMSO as electron acceptors 40. The identification of a highly conserved Crp-binding site at the promoters of genes in this cluster provides an explanation for the phenotypes of crp mutants. Crp might act as a master regulator of anaerobic energy transport; its inactivation leads to repression of multiple branches of anaerobic energy transport pathway. Notably, the regulation of Crp to anaerobic energy transport is iron-responsive but Fur-independent, since most genes in this cluster shows little expression difference between the fur mutant and wild-type.

To generate a fur deletion mutant from MR-1, a pDS3.1-fur suicide plasmid described in 23 was used. This plasmid contains the 5'- and 3'-flanking regions of Fur locus, with a 241-bp internal fragment of Fur locus removed. The plasmid was transformed into E. coli WM3064 strain prior to conjugal transferring into MR-1. Correct in-frame deletion was verified by sequencing PCR products using primers outside the DNA recombination region. The sequences of the primers are: (5'-GCA AGT ACA GCC GTT ATT CAC TCC-3') and (5'-GTA TCC AAA GGA TGC AAC TGG-3').

MR-1 and the fur mutant were grown to mid-log phase and diluted 1:100 into 300 ul fresh LB or M1 liquid medium. A Type FP-1100-C Bioscreen C machine (Thermo Labsystems) was used to measure the growth every thirty minutes. All physiological studies were done in triplicates so that average and standard error could be calculated. Different concentrations of iron chelator were prepared by dissolving 2, 2'-dipyripyl in water. For iron repletion, ferrous sulfate was used at the concentration of 200 uM.

Four biological replicates of MR-1 and the fur mutant were grown in LB to mid-log phase (OD₆₀₀=0.6) and cells were collected. For the fur mutant, cells were also sampled at 1, 5, 10, 20, 40, and 60 minutes after adding 2, 2'-dipyridyl to a final concentration of 160 uM. Subsequently, ferrous sulfate was added to a final concentration of 200 uM. Cells were collected at 1, 5, 10, 20, 40, and 60 minutes thereafter. Total RNA was extracted using Trizol Reagent (Invitrogen) as described previously 23. RNA samples were treated with RNase-free DNase I (Ambion, Inc.) to digest residual chromosomal DNA and then purified with RNeasy Kit (Qiagen) prior to spectrophotometric quantification at 260 and 280 nm.

cDNA was produced in a first-strand reverse transcription (RT) reaction and labeled with Cy5 dUTP (Amersham Biosciences) by direct labeling in the presence of random hexamer primers (Invitrogen). S. oneidensis MR-1 genomic DNA (gDNA) was amplified by Klenow (Invitrogen) and Cy3 dUTP was incorporated into the product (Amersham Biosciences). Fluorescein labeled probes were purified using a PCR purification kit (Qiagen). Slides were pre-hybridized at 50°C for about one hour to remove unbound DNA probes in a solution containing 50% (V/V) formamide, 9% H₂0, 3.33% SSC (Ambion, Inc.), 0.33% sodium dodecyl sulfate (Ambion, Inc.), and 0.8 μg/μL bovine serum albuminin (BSA, New England Biolabs). Slides were hybridized at 50°C overnight with Cy5- and Cy3- labeled probes in the above solution, with 0.8 μg/μL herring sperm DNA (Invitrogen) replacing BSA to prevent random binding. Pre-hybridization and hybridization were carried out in hybridization chambers (Corning). Slides were then washed on a shaker at room temperature in the following order: 7 min. in 1x SSC, 0.2% SDS; 7 min. in 0.1x SSC, 0.2% SDS; and 40 sec. in 0.1x SSC.

A ScanArray Express Microarray Scanner (PerkinElmer) was used to scan slides. Fluorescence and background intensity were quantified using ImaGene 6 software (BioDiscovery, Inc.). All spots in which signal vs. background ratios were less than 3 were discarded.

GeneSpring 7.2 (SiliconGenetics) was used to remove outliers, perform LOWESS normalization and for analysis of statistical significance via a two-way t test for two independent conditions. To calculate the ratios over different time points, samples during iron depletion (namely, C1', C5', C10', C20', C40' and C60') were compared to samples without addition of chelator (C0'), and samples during the iron repletion (F1', F5', F10', F20', F40' and F60') were compared to C60' (1' is 1 min., 5' is 5 min., etc.). Genes with |log₃ ratio| ≥1 with p<0.05 are considered significant. To construct a gene co-expression network from temporal gene expression profiles, a Random Matrix Theory (RMT) based algorithm as described in 4142 was employed. Applying the RMT method to the microarray data revealed a Pearson correlation coefficient of 0.87 as the minimal threshold to construct gene co-expression network. Since gene co-expression networks are known to be hierarchical 43, higher thresholds have been used to recognize larger functional modules. As a result, nineteen modules were distinguished (Additional file Additional file 2A). The software Pajek 44 was used to visualize the gene co-expression network. To identify the common motif, the Gibbs Motif Sampler 38 was employed according to its manual. The promoter regions of genes in the heat shock cluster were scanned to identify RpoH-binding site. For Crp-binding site, the recursive model of the Gibbs Motif Sampler was used to scan the upstream intergenic regions.

Real-time quantitative reverse transcription-PCR (RT-PCR) was performed as described previously 23, except that iQ SYBR green supermix (Bio-Rad) was used instead of SYBR green I.

The microarray data are available at http://www.cs.clemson.edu/~luofeng/yang/fur.html.