Logo of narLink to Publisher's site
Nucleic Acids Res. 2007 Nov; 35(20): 6762–6777.
Published online 2007 Oct 5. doi:  10.1093/nar/gkm631
PMCID: PMC2189734

Discovery of Fur binding site clusters in Escherichia coli by information theory models


Fur is a DNA binding protein that represses bacterial iron uptake systems. Eleven footprinted Escherichia coli Fur binding sites were used to create an initial information theory model of Fur binding, which was then refined by adding 13 experimentally confirmed sites. When the refined model was scanned across all available footprinted sequences, sequence walkers, which are visual depictions of predicted binding sites, frequently appeared in clusters that fit the footprints (83% coverage). This indicated that the model can accurately predict Fur binding. Within the clusters, individual walkers were separated from their neighbors by exactly 3 or 6 bases, consistent with models in which Fur dimers bind on different faces of the DNA helix. When the E. coli genome was scanned, we found 363 unique clusters, which includes all known Fur-repressed genes that are involved in iron metabolism. In contrast, only a few of the known Fur-activated genes have predicted Fur binding sites at their promoters. These observations suggest that Fur is either a direct repressor or an indirect activator. The Pseudomonas aeruginosa and Bacillus subtilis Fur models are highly similar to the E. coli Fur model, suggesting that the Fur–DNA recognition mechanism may be conserved for even distantly related bacteria.


The protein Fur is the 16.8 kDa product of the fur (ferric uptake regulation) gene in Escherichia coli (1), so named because it was first observed to repress the transcription of genes that code for components of ferric (Fe+3) uptake systems found in the cell membrane. Since then, Fur also has been found to regulate other genes that are not directly related to iron transport, such as those encoding hemolysin, Shiga-like toxin and manganese superoxide dismutase (2–5).

Fur binds to DNA and represses transcription in the presence of divalent metal ions. The ion is thought to be Fe+2 in vivo (6), however, DNase I footprinting experiments have shown that Fur also binds to DNA in the presence of Mn+2, Co+2, Cu+2, Cd+2, and Zn+2 (7). Recent studies have suggested that purified Fur contains at least one Zn+2 ion as a structural stabilizer (8). Fur has been observed to bind to DNA as a dimer and in higher order polymers (7,9), and electron microscopy has shown polymerization of Fur on DNA under high concentrations of protein and metal ions (2).

Numerous strategies have been employed to find new Fur binding sites. Various consensus sequences have been derived from both footprinted and non-footprinted Fur binding sites (3,7,10) and these have been compared to sequences in the promoter region of suspected iron-regulated genes. Putative Fur targets were then investigated further through genetic and biochemical experiments. Stojiljkovic et al. created a successful ‘Fur titration assay’ to locate new Fur binding sites using an fhuF:lacZ fusion and Fur consensus sequence-containing plasmid titrant on MacConkey plates (1). Several new iron-regulated genes in E. coli were discovered using this consensus sequence-based technique. In addition to the above, studies have also been carried out using E. coli Fur for DNase I footprinting with non-E. coli DNA (11,12). Recently, transcriptional profiles of E. coli genes have been used to determine those that are regulated by iron and Fur by evaluating mRNA levels in the absence of iron or Fur protein (13).

Another method for finding Fur-regulated genes is to use molecular information theory to locate new binding sites. Using this approach, classical information theory (14,15) is applied to molecular biology (16). First, a set of binding sites is aligned by maximizing the information content (17), and then the average pattern at the sites is represented by a computer graphic called a sequence logo (18). Next, the conservation of bases in the aligned set is used to create a weight matrix model that assigns a weight in bits to each base at each position according to its frequency in the data set (19). This can be displayed using the sequence walker graphic (20).

In addition to displaying details of binding sites, sequence logos can be used to understand the mechanism of binding. In instances where factors bind in overlapping clusters, it is difficult to assign the relative contribution of a base in an overlapping region to the appropriate binder or to determine the range of the binding site. Here, we tested several Fur binding site models that were obtained by multiply aligning Fur binding sequences using different window sizes, and identified the model that best represents binding by a single Fur dimer.

Information theory has previously been used to build two models to evaluate and predict Fur binding sites (13,21). Both models used ad hoc variations of information theory to assign scores to the predicted binding sites, rather than classical information content in bits. In one case the model was built using some sites that had not been footprinted by Fur and were probably not aligned to maximize the information content (21).

The most rigorous approach to model building is to use a data set comprised of only footprinted binding sites from one species. By restricting the data set to experimentally proven sites, one is certain that the model will reflect the binding characteristics of the protein; the use of a single species ensures that the protein and DNA binding sequences evolved together and therefore correspond to one another (22). Many biases from previous models are thereby avoided, but not all (23). The resulting experimentally supported model is then scanned across the entire genome of the species, looking for sequences that contain a positive amount of information as evaluated by the weight matrix (19). Sequence walkers, which are graphical representations of individual binding sites, then display probable binding sites on the genome based on the model of proven sites (20). This method was successfully used to discover that the OxyR transcription factor controls the expression of the fur gene (24), to identify additional sites for proteins such as Fis, SoxS, OxyR and PXR/RXR (23,25–27), to characterize splice junctions (28), and to discover T7 islands, a unique class of mobile genetic elements (29,30). In this study, information theory has allowed us to identify new Fur binding sites, 13 of which we confirmed experimentally.



Programs used in this study are available at http://www.ccrnp.ncifcrf.gov/∼toms/. A web-based tool for searching for Fur sites is available at http://www.ccrnp.ncifcrf.gov/∼toms/delilaserver.html.

Creating the Fur model

Eleven footprinted E. coli Fur binding sequences were extracted from the E. coli K-12 genome (NC_000913) by the delila program (31). These sites are from the promoters of the genes cir, fecA, fecI, fur, sodA, tonB and iucA, along with two bidirectional promoter regions for the genes fepA-fes and fepB-entC (8,32–39). The promoter region at fepB-entC has two distinctly protected regions; each region was included in the data set as individual sequences (fepB and entC). The promoter of iucA has an exceptionally long secondary footprint and so two regions were used from the E. coli plasmid ColV-K30 (M10930, from 347 to 370, and from 365 to 393) (7). The complement of each footprinted sequence was also included, since Fur binds as a dimer (6,7,9,40). The program malign was used to obtain an alignment of the sequences that maximizes the information content within defined windows of the alignment (17).

As the initial model described above contains only 11 sites, we then refined the model by including 13 more sites that were identified in the genome by a search and confirmed by gel shift experiments in this study (see subsequently). The validity of the refined model was tested by scanning it across all available E. coli Fur footprinted regions, using the programs scan and lister to create sequence walkers (19,20). The model is verified if the walkers correspond to DNase I footprint regions. For further verification, the model was also scanned across synthetic oligonucleotides that contain GATAAT repeats and which had been previously footprinted (5).

To compare Fur models of different bacteria, we also built Pseudomonas aeruginosa and Bacillus subtilis models by using footprinted Fur binding sequences from these two species (41,42).

Scanning for Fur binding sites

Fur in high concentration can bind weak sites, while in low concentration it binds only strong sites. The affinity of Fur to its binding site also varies significantly with the availability of metal. Furthermore, as demonstrated by footprint data, Fur binding can extend to flanking regions that bear little resemblance to the strong Fur binding sites. This suggests that it may not be practical to use a fixed cutoff for a Fur binding site model under different conditions. In our study, we used two cutoff values for different scans.

For the whole E. coli genome scan, we used the lowest information in the set of sites used to build the model as the cutoff. This should minimize false positives in our predictions, as the model was built from experimentally proven sites. Groups of sites that were within 100 bases of each other were identified using the localbest program, and the strongest one was selected to represent the region. This ensured that each region in the data set was unique. All sites with Ri values greater than 17.0 bits were extracted (Table 4). This value was chosen simply to allow for a manageable set of regions for further analysis.

Table 4.
Strong Fur clusters in the E. coli genome predicted by the 24-site M9 model

As Fur can extend binding to much weaker regions in the footprints, we should use a lower cutoff for scanning footprinted sequences. The second law of thermodynamics sets a theoretical lower bound for the information content of a binding site (Ri) at zero bits (19). Therefore we used a zero-bit cutoff when scanning footprinted sequences.

Fur has been suggested to repress genes by binding to their promoter elements (−35 and −10 regions), thereby blocking the access of polymerase to the promoters (33,43–46). To determine how many of the predicted Fur binding clusters overlap with promoters, we used an information theory-based flexible σ70 binding model to scan the Fur clusters. The σ70 model was built from 401 experimentally proven E. coli promoters by uniformly taking into account the information present in the −10, the −35, and the uncertainty of the spacing between them (47); the same method has been successfully used to model prokaryotic ribosome binding sites (48). The information content (average conservation) of the promoter model is low (An external file that holds a picture, illustration, etc.
Object name is gkm631i1.jpg bits, SD=2.8 bits for the Ri distribution), and the individual information for a site in the model ranges from 0.3 to 12.6 bits. The probability that a site is less than 4 bits is only 18.6%, so we used 4 bits as a cutoff to predict reasonably strong non-activated promoters (47).

Gel mobility shift assays

Two sets of oligonucleotides containing predicted Fur binding sites in E. coli were designed and synthesized (Oligos Etc). All oligonucleotides were self-complementary, had a 5′-GCTA-3′ overhang on the 5′ end, and contained a hairpin loop. Oligos exbB, exbD and fhuF contained a hairpin of the sequence 5′-C GCGAAGC G-3′, while the other 12 oligos [yoeA, fepD-entS (formerly ybdA), gpmA, ryhB, fhuA, nohA, oppA, gspC, garP (formerly yhaU), yahA, fadD and ygaC] contained a hairpin of the sequence 5′-ACGATCGC GCGAAGC GCGATCGT-3′ in the center. Such loops form a structure that is stabilized by base pairing between G3 and A5 of the central seven bases of each loop (49), and they are convenient for use in DNA mobility shift assays because of the exact equimolar concentrations of the complementary strands and their high stability (25).

Three oligos containing the promoter regions for exbB, ygaC and the upstream region of exbD were created to test previously published consensus sequence predictions (50,51). Eight oligos contained potential Fur-controlled promoters identified using both an 11-site Fur model and a σ70 model as described above [yoeA, fepD-entS (formerly ybdA), gpmA, ryhB, fhuA, nohA, oppA and fhuF]. We were also interested in whether Fur binds intragenically, so four oligos were synthesized that contained strong predicted sites found within gene coding regions (gspC, garP, yahA and fadD). These sequences can be found in Supplementary Table S1 and Figure S1.

Gel mobility shift assays were performed using the 15 oligos (Figure 3). The oligonucleotides were labeled by a fill-in reaction with Taq DNA polymerase. The 20 μl of labeling mixture [20 pmol oligo, 2 mM MgCl2, 1 × PCR buffer, 5 units Taq polymerase (GibcoBRL®), and 10 μM tetramethylrhodamine-6-dUTP (NEN)] was incubated at 72C for 1 h. 80 μl of ddH2O was added to the mixture, followed by two phenol extractions, a phenol/chloroform extraction and a chloroform extraction. The labeled oligonucleotides were diluted 1:5 in TE (10 mM Tris–Cl, pH 7.5, 1 mM EDTA), boiled for 10 min and then placed on ice to prevent dimerization and trimerization.

Figure 3.
Gel mobility shift assays to test predicted Fur binding sites. Oligonucleotides containing predicted Fur binding sites were incubated without (−) or with (+) 150 nM E. coli Fur protein and gel electrophoresed to test the 11-site M9 model (Figure ...

The labeled oligonucleotides (5 μl each, 0.2 pmol) were then incubated in Fur binding buffer (10 mM Bis–Tris–HCl, pH 7.5, 5 μg/ml sonicated salmon sperm DNA, 5% glycerol, 100 μM MnCl2, 100 μg/ml BSA, 1 mM MgCl2, 40 mM KCl) with 150 nM Fur protein at 37C for 13 min (7). Fur was a gift from T. O'H;alloran and C. Outten. Samples were electrophoresed on a 5% polyacrylamide gel in Fur electrophoresis buffer (0.1 M Bis–Tris–HCl, pH 7.5, 10 mM MnCl2) at 120 V for ∼1 h.

Bands were visualized with an FMBIO II fluorescent scanner (Hitachi), with an excitation wavelength of 532 nm and a 585 nm filter for detection of tetramethylrhodamine.


To generate the fhuF promoter construct (pGSO129) used to test for Fur binding, a 240-bp fragment amplified by PCR from genomic DNA (using the primers 5′-GCG GCT GGA GAT GAA TTC GCC AGA TG and 5′-GCC CTG CAA TCA GGG ATC CCG GCA GC) was cloned into the BamHI and EcoRI sites of pUC18 (27). Purified Fur protein, generously provided by C. Outten and T. O'H;alloran, was incubated with a Mn2+-containing buffer according to de Lorenzo et al. (7). DNase I footprinting then was carried out as described previously (52). The 240-bp BamHI–EcoRI fragment of pGSO129 was labeled with [γ-32P]ATP at either the BamHI site (top strand) or the EcoRI site (bottom strand). The labeled fragments were incubated with 0, 50, 100, 200 and 400 nM purified Fur protein at room temperature for 5 min. The samples then were subjected to limited DNase I digestion, purified and separated on 8% polyacrylamide sequencing gels.


E. coli Fur binding model

Eleven footprinted Fur binding sites (see Table 3 and Supplementary Figure S1) (8,32–39) from E. coli K-12 (53) were used to create an initial information theory model. Because Fur binds as a dimer (9,54), the model included both the footprinted Fur sequences and their complements. The sequences were aligned using the program malign, which maximizes the information content within a region of the alignment (a ‘window’) by shuffling the sequences back and forth (17). Multiple alignment with a window size of (−12, +12) gave a sequence logo that shows base conservation in the range of (−12, +12) (Figure 1A). For convenience we named this logo M12.

Figure 1.
Different alignments of footprinted E. coli Fur binding sites. Sequence alignments were done using the program malign with different window sizes [from (−15, +15) to (−5, +5)] (17) on 11 footprinted E. coli Fur binding sequences and their ...
Table 3.
Scan of footprinted Fur binding sites in E. coli

The M12 logo appears to consist of two overlapping direct repeats, one is from −12 to +6 and the other from −6 to +12. Because the multiple alignment process can incorporate one or the other of these repeats, this suggests that different alignments may be obtained when other windows are used to align the sequences. Because two Fur dimers can bind overlapping sites with 6-base separation (40,55), we then multiply aligned the sequences by using window sizes ranging from (−15, +15) to (−5, +5). These sizes should cover the sites recognized by one or two Fur dimers (9). A total of seven different alignments were obtained; the corresponding sequence logos are shown in Supplementary Figure S2. According to the patterns shown in the logos, the seven alignments that were obtained can be further consolidated into and assigned to three basic, alignment classes; these alignment classes are represented by logos M12, M9 and M6 (Figure 1). Each of the three alignments itself is an inverted repeat; the M12 logo also appears to consist of two overlapping M9 sites (from −12 to +6 and from −6 to +12), and the M9 logo looks like two overlapping M6 sites (from −9 to +3 and from −3 to +9) (Figure 1).

To determine which of the three alignment classes best represents a single-dimer binding model, some published experimental data was analyzed. Lavrrar and McIntosh (55) used the Ferguson method (56) to determine how many E. coli Fur dimers can bind on five synthetic oligos and two natural sites, fepB and entS. We applied the three models to their sequences and used sequence walkers to show the predicted sites of each model. A full list of the predictions (with walkers >0 bits) is given in Supplementary Figure S3. The results show that the models frequently found overlapping sites that are separated by 3 or 6 bases. The sites that are 6 bases apart can both be strong, while 3-base separated sites have one strong site and one weak site (Supplementary Figure S3). Furthermore, it has been shown that Fur dimers can simultaneously bind two overlapping sites with 6-bp spacing (40), but there is no experimental data showing that 3-base separated overlapping sites can be bound at the same time. Therefore to analyze the results, we only counted major walkers (strong walkers that are separated by 6 bases) for each model in each oligo. A summary of the results (with major walkers) is available in Table 1; two examples of the prediction (with major walkers) are shown in Figure 2A and B. The results show that the M9 model made correct predictions for five of the seven oligos, while the M12 and M6 models both made only two correct predictions. Furthermore, the M12 model always predicted one less site than the M9 model. These results suggest that the M9 model may be a single-dimer binding model, the M12 model may represent a two-dimer binding model, while the M6 alignment may be just a result of compression of the binding sequences into a small alignment window.

Figure 2.
Sequence walkers to test E. coli (A and B) and B. subtilis (C and D) Fur models. (A and B) The three E. coli Fur models (M12, M9 and M6) were tested with two E. coli Fur binding sites, fepB and entS (55). Sequence walkers of each model (green, M12; red, ...
Table 1.
Information theory test of 11-site E. coli Fur models M12, M9 and M6

Confirmation of 13 predicted sites and model refinement

Since our E. coli Fur model had only 11 sites, we wanted to refine the model by including more sites. A genome scan with the 11-site M9 model revealed 389 sites above 11 bits (the lowest information of a site in the model is 11.1 bits). Within these, 13 sites (from 12 to 26 bits) were selected for experimental confirmation (Figure 3, Supplementary Table S1 and Figure S1). These include eight sites located in promoter regions [yoeA, fepD-entS (43), gpmA (= pgm) (51), ryhB (= yhhX-yhhY) (51,57), fhuA (13), nohA (51), oppA (13) and fhuF (51)], and four sites inside genes (gspC, garP, yahA and fadD). We also included two sites that were predicted by consensus models to be in ygaC and exbD (50,51), since our model only detected a weak site (1.4 bits) in ygaC, and a negative site (−5.0 bits) in exbD (Figure 3). Gel mobility shift assays showed that all 13 sites predicted by the M9 model shifted when incubated with Fur protein, while the exbD site did not shift, and the ygaC site only showed extremely weak binding (Figure 3).

McHugh et al. (13) found that both exbB and exbD were regulated by Fur, but their model did not detect a binding site in the exbB promoter while our information theory model successfully predicted that Fur binding site, and our gel shift results confirmed this (Figure 3). The exbD gene is located only seven bases downstream of the exbB gene. Both genes have the same orientation, so they may belong to a single operon cotranscribed from the exbB promoter and coregulated by Fur.

To strengthen our model, we added the 13 confirmed sites to the 11 previously footprinted sites (Figure 4A). As with the 11 sites, multiple alignment of the 24 sites using different window sizes also gave three classes of alignments, M12, M9 and M6 (Supplementary Figure S4). When these models were tested against the same set of sequences as that used to test the 11-site models (Table 1, Supplementary Figure S3), similar, but cleaner results were obtained, i.e. the weak walkers that were not counted previously become weaker (some are even below 0 bits), but the strong walkers still have similar Ri values (Supplementary Figure S5). This confirms that our counting of major walkers predicted by the initial 11-site models was reasonable.

Figure 4.
Comparison of M9 Fur models of three bacterial species. Experimentally proven Fur binding sites from E. coli (A), P. aeruginosa (B) and B. subtilis (C) were used to build the models for these three bacteria. The marked region from −7 to +7 in ...

P. aeruginosa and B. subtilis Fur models

To compare E. coli Fur models with other bacterial Fur models, we also built P. aeruginosa and B. subtilis Fur binding site models. As with E. coli Fur binding sites, we applied the same method to align footprinted Fur binding sites from P. aeruginosa (41) and B. subtilis (42). For both bacteria we obtained two classes of alignments (Supplementary Figure S6), which correspond to the E. coli M9 and M6 logos (Figure 1B and C), respectively. No M12 alignments were obtained for these two bacteria, probably because some sites used to make the models are single-dimer binding sites (see Discussion Section). Both P. aeruginosa and B. subtilis M9 models are highly similar to the E. coli M9 model (Figure 4). The P. aeruginosa M6 logo looks almost identical to the E. coli M6 logo, while the B. subtilis M6 logo is less similar to the E. coli M6 logo (Figure 1C, Supplementary Figure S6B and D).

As with Lavrrar and McIntosh's; work on E. coli Fur binding sites (55), Baichoo and Helmann (40) performed similar experiments on eight synthetic oligos and two natural sites (feuA and dhbA) with B. subtilis Fur, and determined how many Fur dimers can bind each of these oligos. We tested the B. subtilis M9 and M6 models with these 10 sequences (Table 2, Figure 2C and D, and Supplementary Figure S7). The results show that the M9 model made correct predictions for all 10 sequences, while the M6 model was correct for only four oligos. For most of the oligos, the M6 model predicted one more site than the M9 model did. Similar to our observations in E. coli, these results strongly suggest that the M9 model is a single-dimer binding model, while the M6 alignment is due to compression of the binding sequences.

Table 2.
Information theory test of B. subtilis Fur models M9 and M6

Relative to E. coli Fur (NP_415209), the P. aeruginosa Fur (NP_253452) is moderately diverged (58% identical in protein sequence), and the B. subtilis Fur (NP_390233) is distantly diverged (33% identity). However, a comparison of the three M9 models from these three species shows that these models are highly similar to each other (Figure 4), suggesting that the Fur–DNA recognition mechanism is conserved in even distantly related species.

Scans of footprinted regions

Gel shift experiments cannot give precise Fur binding regions. Originally using the 11-site M9 model and later using the 24-site M9 E. coli Fur model, we predicted Fur sites in the fhuF promoter region and then, to validate our model, we performed DNase I footprinting on that region. To further confirm the 24-site M9 model, we also scanned it across all other available footprinted E. coli Fur binding regions, and correlated our predictions with the published footprints. Fur has been documented to exhibit ‘secondary footprinting’, protecting extended regions of DNA under higher protein concentrations (Supplementary Figure S8) (33,36,46). To reveal the extended regions we used a cutoff at zero bits, since that is the theoretical lowest information for binding (19).

The E. coli 24-site M9 model predicts two separated Fur binding clusters in the fhuF promoter region; one cluster contains three strong sites of 16.1, 20.8 and 22.3 bits, and the other contains two weaker sites of 14.6 and 9.3 bits (Figure 5A). The fhuF oligo used in the gel shift experiments in Figure 3 only contains the strong Fur cluster. To determine if both predicted clusters are protected by Fur, we performed DNase I footprinting on a larger fhuF promoter region (see Materials and Methods Section) that contains both predicted Fur binding clusters. The results show that there are indeed two Fur-protected regions in the fhuF promoter, one has high affinity with Fur (protected by 50 nM of Fur), and the other has a lower affinity (weakly protected by 50 nM of Fur) (Figure 5B). The predicted strong Fur binding cluster covers 91% of the high-affinity footprinted region, and the weak Fur cluster covers 66% of the low-affinity region (Table 3, Figure 5A).

Figure 5.
Fur model scan and DNase I footprints of the fhuF promoter region. (A) The promoter region was scanned with the 24-site M9 model, and two clusters of sequence walkers (Ri > 0 bits) were detected, with each closely corresponding to one of the two ...

There are 11 other footprinted E. coli Fur binding regions, including the 10 regions used to build the initial 11-site models, and the fepD-entS promoter region (58). When these footprints were scanned, similar coverage was obtained (Table 3, Supplementary Figure S8). All 11 footprinted sequences displayed clusters of multiple overlapping Fur walkers; in these clusters, the majority of the walkers were separated by six bases; some were separated by three bases (Table 3, Supplementary Figure S8). For 7 of the 11 regions (fepA-fes, fepB, entC, fur, tonB, cir and iucA), sequence walkers cover ∼83% (72–93%) of the footprints. No sequence walkers cover the secondary footprints in the fecA, fepD-entS and sodA regions, resulting in low coverage of these three footprints (66%, 54% and 56%, respectively). The secondary sodA footprint was only protected with a high concentration of purified Fur (200 and 500 nM); it was not protected by crude extracts containing overproduced Fur (36). In the other region, fecI, several low-information content walkers (0.2 and 0.8 bits) extend past the range of the footprint, resulting in 144% coverage of the footprint.

In a previously published study, Escolar et al. synthesized oligonucleotides that contained repeats of the sequence GATAAT and determined Fur binding by footprint experiments (5). No footprint was seen with one insert; correspondingly, no sequence walkers were observed in that sequence (Supplementary Figure S9). For the oligos with three, four and five inserts, sequence walkers cover 83–90% of the footprints. The 2-insert sequence had a weak interaction with Fur at high protein concentrations and two overlapping walkers of 3.0 and 4.2 bits appeared, which cover the repeats (5). These results demonstrate that the Fur model accurately predicts Fur binding on both natural and synthetic sequences.

Whole genome scan

The lowest information of any site in the 24-site M9 model is 10.1 bits (Supplementary Figure S1). Since all sites in the model are proven binding sites, we used 10 bits as the cutoff for a whole genome scan. This presumably minimizes false positives in our predictions. When the model was scanned across the E. coli K-12 genome (NC_000913), a total of 412 sites were found above 10 bits. These sites belong to 363 unique clusters. A list of the clusters is available at http://www.ccrnp.ncifcrf.gov/~toms/papers/fur/.

The 363 regions were extracted from −50 to +50 around the strongest site in each region, and scanned with the 24-site M9 model using a cutoff of 0 bits. We then determined the exact region of each cluster that is covered by walkers greater than 0 bits. The results show that these clusters contain 1–14 walkers, covering a region from 19–142 bases (35.0 bases on average). This is comparable to the known footprints (30–103 bases) (Table 3, Supplementary Figure S8). Of these clusters, 155 are in intergenic regions, 143 are entirely inside gene coding regions and 65 overlap the boundary between coding and intergenic regions.

To determine how many of the Fur clusters overlap with reasonably strong promoters, we scanned these 363 Fur clusters with a σ70 promoter model (47) using a 4-bit cutoff. The results showed that 303 clusters overlap with one or more strong promoters (Ri > 4 bits). The other 60 clusters do not overlap with a strong promoter. When we scanned the flanking regions of these 60 clusters (−50 to 0 of the left edge, and 0 to +50 of the right edge of each cluster), only 23 clusters were found to have strong promoters (Ri > 4 bits) in the flanking regions. We examined the locations of Fur sites relative to known promoter elements (transcriptional start site, −10 and −35) and found that they are more tightly distributed around the promoter elements than observed with the DNA binding proteins Fis, H-NS and IHF (47) (data not shown). These results suggest that Fur may mainly act as a direct repressor of genes by binding to and blocking the promoters, and that direct activation by Fur through binding to the region upstream of a promoter, as suggested in Neisseria meningitidis by Delany et al. (59), may not be a common mechanism of Fur regulation in E. coli.

Out of the 363 clusters, 42 were found containing at least one predicted Fur binding site above 17.0 bits (our arbitrary cutoff used to locate significant regions) (Table 4). Of the 42 clusters, 39 were found to overlap with one or more strong σ70 promoters (>4 bits). The other three clusters, fepD-entS, yddA and fecA, do not overlap. Scanning the flanking regions of these three clusters with the σ70 promoter model revealed that these Fur clusters are located between the respective promoters and translational starts. There may be only weak repression of these genes by Fur, as none of them were significantly repressed by Fur according to microarray analyses (13,60).

Scans of other proposed Fur-regulated genes

Many genes have been proposed to be Fur-regulated by comparing conventional consensus sequences to promoter regions and also by homology to systems in other organisms. Kammler et al. annotated two Fur binding sites in the promoter region of feoA in E. coli by comparison to a consensus sequence (10). The 24-site M9 model predicts two clusters of sites in the same region with each cluster covering one of the marked Fur boxes. One cluster has one major site of 10.5 bits (at 3538006), the other cluster has two major overlapping sites of 10.9 (at 3538059) and 13.4 bits (at 3538065), respectively. The same authors have confirmed that Fur does bind to this region in vivo, but the exact Fur binding regions have not been determined by footprint experiments.

Vassinova and Kozyrev used an in vivo selection to locate Fur sites on Sau3A fragments from the E. coli genome (51). The five regions from Figure 2 of their paper were analyzed using sequence walkers. Using an unidentified consensus sequence, yhhX (from 3578828 to 3577791) was predicted by Vassinova and Kozyrev to have two Fur binding sites. In the same region we found one strong site of 25.2 bits (at 3579054) (Table 4), overlapping two weaker sites (8.2 and 8.3 bits). This cluster of Fur sites is actually located right in front of an sRNA gene, ryhB (from 3579039 to 3578946) (Table 4) (61,62), which has been shown to be repressed by Fur (57). The promoter region of gpmA (from 786818 to 786066, named pgm in reference 51), was predicted by consensus to contain two sites. One of the highest sites (27.4 bits) was found in this region (at 786853), overlapping two weaker sites of 7.5 and 5.6 bits. The consensus sequence-predicted Fur site by Vassinova and Kozyrev (51) is located about 600 bases upstream of the ygaC gene. In the same region, our 11-site M9 model only found a 1.4-bit site (Figure 3), and the 24-site M9 model found a negative site of −3.4 bits (Supplementary Figure S1). Instead, in the nearby regions the 24-site M9 model found a 7.4-bit site (11 bases upstream of ygaC) and a 10.1-bit site (728 bases upstream of ygaC). For nohA (from 1634391 to 1633822), two overlapping sites of 22.3 and 12.6 bits were found at 1634624 and 1634630 (Table 4). The fifth region was fhuF (from 4603686 to 4602898). Figure 5 shows the DNase I footprints and predicted walkers in this region. Only the high-affinity region (fhuF1) was identified by the consensus sequence. Thus four of the five Sau3A fragments (yhhX, gpmA, nohA and fhuF) selected to have Fur binding sites in vivo were identified in our genome scan (Table 4), and our gel shift results confirmed this (Figure 3); the remaining site, ygaC, had a weak walker of 1.4 bits (by the 11-site M9 model) and showed extremely weak binding (Figure 3).

Eick-Helmerich and Braun matched a Fur consensus sequence to the promoters of exbB and exbD (50). One strong walker of 19.1 bits was found in the exbB promoter region, overlapping the ‘Fur box’. The region upstream of exbD showed no walkers with Ri greater than 0 bits, even though it had also been predicted to contain a Fur binding site using the same consensus sequence as used in the exbB promoter. The highest walker that we could detect in the region upstream of the exbD gene is −7.1 bits (by the 24-site M9 model, Figure 3 and Supplementary Figure S1), significantly lower than the theoretical lower bound of a binding site (0 bits) (19). This indicates that Fur should not bind this region, and our gel shift results confirmed this (Figure 3).

A microarray analysis by McHugh et al. found 101 genes to be regulated by Fur, 53 of which were repressed and 48 activated (13). The 53 Fur-repressed genes belong to 32 transcription units (individual genes or operons), 18 of which are involved in iron metabolism, 14 in energy metabolism and other functions. Using the 24-site M9 model, we predicted strong Fur binding clusters (higher than 13 bits) for all 18 iron metabolism-related transcription units. Of the other 14 transcription units, only two have a Fur binding site higher than 10 bits (nrdHIEF, 10.1 bits; fimE, 10.8 bits). The 48 Fur-activated genes belong to 34 transcription units, only three of these (garPLRK, ynaE and ydfK) have a predicted Fur binding cluster. The garPLRK operon has a weak Fur cluster that contains one major site of 10.4 bits. The ynaE and ydfK genes are almost identical in both promoter and coding regions, and the two Fur clusters (17.1 bits, centered at +8 from the translational start) are the same (Table 4). These results strongly suggest that in E. coli Fur is a direct repressor of iron metabolism genes and an indirect repressor or activator of other genes.

McHugh et al. (13) used an altered information theory approach for modeling Fur binding (63) to identify which genes are under direct Fur control. Our model identified sites in all genes that their model did, as well as four others, fhuE (12.4 bits, at 1160880), exbB (19.1 bits, at 3150076), fimE (10.8 bits, at 4539697) and ydfK (17.1 bits, at 1631071).

The indirect activation by Fur requires an intermediate regulator, which could directly target Fur-activated genes. A small regulatory RNA gene, ryhB is one such important player. Earlier study revealed several genes that are directly repressed by ryhB (57). Recently, a microarray analysis of ryhB control by Masse et al. (60) expanded this direct target set to 18 operons. They also identified an additional 10 indirectly down-regulated operons and 10 up-regulated operons. The 10 ryhB indirectly repressed operons are directly repressed by Fur, and for all of these operons we found strong Fur binding clusters (higher than 13 bits) in the corresponding promoter regions. In contrast, only a few of the other 28 ryhB-controlled operons (18 directly repressed operons and 10 up-regulated operons) have a predicted Fur binding cluster above 10 bits. These include acnA (17.6 bits, at 1333484), ydhD (10.9 bits, at 1732367) and oppA (21.5 bits, at 1298973). The acnA and oppA Fur clusters are both strong; the former cluster has three overlapping sites of 17.6, 9.8 and 14.2 bits, and the latter has two overlapping sites of 21.5 and 9.6 bits (Table 4). However, microarray analysis did not find these two genes to be repressed by Fur (13,60), suggesting that the Fur binding clusters may not be the primary control elements of these genes.


Fur binding models

In this study, we used experimentally proven sites to create information theory models of Fur binding. Our models, especially the 24-site M9 model, approximate the binding characteristics of Fur more fully than previous models that depended on conventional consensus sequences, data from multiple species and sequences that were not footprinted (1,32,51). The rigorous approach used here revealed new binding sites, disproved two sites predicted by a consensus sequence, and clarified the manner in which the protein binds.

To produce these models, binding sequences were multiply aligned to maximize the information content in different window sizes. Because multiple Fur sites overlap, our analysis of proven Fur binding sequences from three bacterial species gave two or three different alignments, depending on the species (Figure 1, Supplementary Figure S6). By testing these models against published data, we showed clearly that the M9 model represents single-dimer binding sites, the M12 model is for two-dimer binding, while the compact M6 alignment was caused by compression of the binding sequences into a small window (Tables 1 and and2,2, Figures 1 and and2,2, Supplementary Figures S3, S5 and S7).

As evaluated by their respective M9 models, our initial set of 11 footprinted E. coli Fur binding sites all contain overlapping sites of 6-base spacing (Supplementary Figure S8), while only 7 of the 20 P. aeruginosa Fur binding sites, and 6 of the 22 B. subtilis Fur binding sites are single-dimer binding sites (data not shown). This scarcity of multiple sites explains why an M12 alignment can be obtained for E. coli, but not for P. aeruginosa or B. subtilis. When we performed multiple alignment by only using overlapping sites of 6-bp spacing, we obtained M12 alignments for both P. aeruginosa and B. subtilis (data not shown). These observations also support the use of M9 as a single-dimer model.

Our multiple alignment method should apply to any DNA binding sites that contain internal direct repeats, as repetition within binding sites may cause alternative alignments of the sites. In the RegulonDB (http://regulondb.ccg.unam.mx) (64), a Fur binding model was created by multiply aligning 47 Fur sites, although it is not clear what these 47 sites are and how the alignment was made. Based on their alignment (http://regulondb.ccg.unam.mx/html/matrix_Alignment.jsp), we made a sequence logo (data not shown), which we found to be similar to our M6 logos. Although the M6 models appear to be able to predict some sites (e.g. Tables 1 and and2,2, Supplementary Figures S3, S5 and S7), the models only represent the internal part of overlapping Fur binding sites, i.e. from −6 to +6 of the M12 model (Figure 1A).

A 1.8 Å crystal structure for P. aeruginosa Fur was obtained by Pohl et al. and based on this, the authors proposed a single-dimer binding model (9). The P. aeruginosa crystal structure contains a putative DNA binding α-helix H4 and a loop between helices H1 and H2, both of which were proposed to be involved in binding DNA through contacting bases in the major groove. Thus, a single dimer would protect two consecutive major grooves on the same face of DNA, and the minimal recognition unit of Fur should be close to 20 bp (9). Our M9 binding site models (19 bp, a 9-1-9 inverted repeat) fit with this single-dimer binding model closely. Pohl et al. also proposed a two-dimer binding model, in which two Fur dimers bind two overlapping sites that are 5-bp apart (9). However, in our whole genome scans, and footprint scans as well, we did not observe any case of 5-bp separated overlapping sites above 5 bits. Instead, we mostly found 3- or 6-bp separated overlapping sites. Above five bits, we found 95 cases of 3-bp spacing and 92 cases of 6-bp spacing.

Several other consensus-based Fur binding site models have been proposed to interpret a 19-bp consensus ‘Fur box’ (Figure 6) (65). Within these, two earlier models, the classical model and hexamer model (5,7,37,54), interpreted the ‘Fur box’ as a single recognition unit of a Fur dimer. Later, Lavrrar and McIntosh suggested that a 13-bp inverted repeat (6-1-6) is the minimal unit recognized by a single Fur dimer, and that two overlapping ‘6-1-6’ motifs correspond to the Fur box, which is required for high-affinity binding of Fur (55,58). Baichoo and Helmann reinterpreted the Fur box as two overlapping 7-1-7 inverted repeats with a 6-bp spacing, and suggested that a 7-1-7 site, but not 6-1-6, represents the minimal recognition unit of Fur (40). Baichoo and Helmann also demonstrated that high-affinity binding by Fur can happen on a single-dimer binding site, i.e. an inverted 7-1-7 site.

Figure 6.
Different Fur binding models to interpret the Fur box.

The 7-1-7 model basically agrees with our M9 models (Figure 6). The main difference is that the 7-1-7 model did not count the weakly conserved bases at positions −9, −8, +8 and +9 in the B. subtilis Fur binding site alignment (Figure 4C). As the E. coli and P. aeruginosa M9 models have highly conserved bases at these four positions (Figure 4A and B), and also the Fur binding mechanisms for these three species are similar (9), we suggest that a minimal Fur binding unit of 19 bp (9-1-9), as represented by our M9 models, is more reasonable.

Relative to our M9 models, the Fur box only represents the internal part of two overlapping sites that are separated by 6 bases, i.e. from −9 to +9 of the M12 logo (Figures 1A and 6), thus any predictions based on the Fur box may not be accurate. Two sites (exbD and ygaC) that were predicted by matching to the Fur box were shown to be non-binding (Figure 3).

Many difficulties in understanding Fur binding sites can be attributed to the choice of consensus sequences as a model (66). In contrast to the individual information weight matrix model, the consensus sequence method ignores the varying importance of bases by treating mismatches equivalently. In addition, the consensus method does not have a criterion for an acceptable number of mismatches.

Sequence walkers predicted by the 24-site M9 model cover ∼83% of the footprints (Table 3, Figure 5, Supplementary Figure S8). Similar results were obtained for both the P. aeruginosa and B. subtilis M9 models (data not shown). These results strongly suggest that our M9 models can accurately predict Fur binding.

Fur Binding Site Clusters

The whole genome scan with the 24-site M9 model revealed 363 Fur binding clusters in E. coli, most of which contain multiple walkers (up to 14), covering regions up to about 140 bases. This is comparable to the size of known footprints, which ranges from 30 to 103 bases (Table 3, Supplementary Figure S8). Most of the clusters (303 out of 363, 83%) were found to overlap with one or more strong σ70 promoters. Indeed, the four Fur clusters in the middle of coding regions gspC, garP, yahA and fadD, that we tested by gel mobility shift assays (Figure 3), all have one or more potential σ70 sites (Ri > 4 bits) exactly overlapping them (see Figures S10–S13). The function of these control elements is unknown.

We correlated our whole genome scan with published microarray data about Fur in E. coli (13,60). The results showed that all operons that are involved in iron metabolism have a predicted strong Fur binding cluster (higher than 13 bits), suggesting direct repression of these genes by Fur. In contrast, only a few of other Fur-activated or repressed operons have a predicted Fur cluster higher than 10 bits, suggesting indirect control of these genes by Fur. It has been shown that RNA gene ryhB directly targets and represses a set of other genes and ryhB itself is directly repressed by Fur, thus ryhB plays an important role in indirect activation of genes by Fur in E. coli (13,57,60).

It has been suggested that in N. meningitidis Fur can also directly activate genes by binding to the region upstream of the promoters (59). However, our whole genome analysis did not support this activation mechanism in E. coli. The E. coli ryhB gene is located between genes yhhX and yhhY. No corresponding homologs of the genes ryhB, yhhX or yhhY can be found in N. meningitidis. This suggests that Fur activation mechanisms may be different in different species.

Disentangling Fur models

Each binding site that we have studied in detail with information theory has had unique properties and challenges for analysis. For example, ribosome binding sites were initially modeled as rigid objects (16) but when enough sites became available, it was clear that the Shine–Dalgarno sequence was being lost because it has a variable distance to the initiation codon (67) and a flexible model was required (48). The flexible model was later successfully applied to σ70 promoters (47) and that was used to study Fur sites. Likewise, information theory analysis revealed that Fis binding sites are commonly found to overlap each other (68), leading to some uncertainty as to how much nearby sites influence the sites being studied in the alignments. Overlapping Fur sites that are bound by Fur protein dimers create an even more complex situation.

Although the 24-site M9 model was quite successful, the model is comprised of information from multiple Fur sites because the Fur dimers overlap on the DNA. Since the clusters frequently display 6-base separation of sites, positions −3 to +9 of the model contain data from positions −9 to +3 of the adjacent Fur sites, e.g. see the logo of Figure 4 and the M9 walkers of Figure 2A. In addition, information from sites separated by 3 bases and the complements of all of these, also contribute to the overall information content of the model. It is not obvious how to disentangle these effects. We suggest that the M9 model can be scanned across the genome to locate strong single-dimer binding sites, which could then be experimentally confirmed and used to construct a disentangled model.

Unfortunately the distribution of the individual information in these sites would be determined by the model used to locate them and this could introduce biases. For example, the mean of the distribution can be altered just by selecting different subsets of sites from the distribution. A similar bias has occurred in binding site databases where the consensus sequence was used to locate new sites (23). Thus, although the sequence logo would no longer contain self-overlapping data, it would not necessarily represent the natural distribution of site strengths. To approach such an ultimate model may require further scanning of footprinted regions. Incorporation of the weaker sites in the footprints would, of course, re-entangle the model so such scans might be best used only to determine whether the model can detect the weaker sites. A viable solution may be to obtain all single-dimer binding sites in the genome that have more than zero bits and can be confirmed experimentally. If the resulting model matches the observed footprints, it may be considered complete. On the other hand, if the resulting model does not match the footprints, the mode of binding by Fur could be different between single dimers and clusters.


Supplementary data are available at NAR Online.

[Supplementary Data]


We thank Tom O'H;alloran and Caryn Outten for Fur protein; Pete Rogan for helping to develop the local best concept; Elaine Bucheimer for writing the localbest program and Lakshmanan Iyer, Brent Jewett, Danielle Needle, Xiao Ma, Shu Ouyang, Denise Rubens and Bruce Shapiro for their helpful comments and discussions. K.A.L. also thanks the National Cancer Institute at Frederick for sponsoring the Werner H. Kirsten Student Intern Program. This publication has been funded in part with Federal funds from the National Cancer Institute, National Institutes of Health, under contract #NO1-CO-12400. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products or organizations imply endorsement by the US Government. This research was supported in part by the Intramural Research Program of the NIH, National Cancer Institute, Center for Cancer Research. Funding to pay the Open Access publication charges for this article was provided by National Cancer Institute.

Conflict of interest statement. None declared.


1. Stojiljkovic I, Baumler AJ, Hantke K. Fur regulon in gram-negative bacteria. Identification and characterization of new iron-regulated Escherichia coli genes by a fur titration assay [published erratum appears in J. Mol. Biol. 1994 Jul 15;240(3):271] J. Mol Biol. 1994;236:531–545. [PubMed]
2. Le Cam E, Frechon D, Barray M, Fourcade A, Delain E. Observation of binding and polymerization of Fur repressor onto operator-containing DNA with electron and atomic force microscopes. Proc. Natl Acad. Sci. USA. 1994;91:11816–11820. [PMC free article] [PubMed]
3. Calderwood SB, Mekalanos JJ. Iron regulation of Shiga-like toxin expression in Escherichia coli is mediated by the fur locus. J. Bacteriol. 1987;169:4759–4764. [PMC free article] [PubMed]
4. Niederhoffer EC, Naranjo CM, Bradley KL, Fee JA. Control of Escherichia coli superoxide dismutase (sodA and sodB) genes by the ferric uptake regulation (fur) locus. J. Bacteriol. 1990;172:1930–1938. [PMC free article] [PubMed]
5. Escolar L, Perez-Martin J, de Lorenzo V. Binding of the Fur (Ferric Uptake Regulator) repressor of Escherichia coli to arrays of the GATAAT sequence. J. Mol. Biol. 1998;283:537–547. [PubMed]
6. Bagg A, Neilands JB. Ferric uptake regulation protein acts as a repressor, employing iron (II) as a cofactor to bind the operator of an iron transport operon in Escherichia coli. Biochemistry. 1987;26:5471–5477. [PubMed]
7. de Lorenzo V, Wee S, Herrero M, Neilands JB. Operator sequences of the aerobactin operon of plasmid ColV-K30 binding the ferric uptake regulation (fur) repressor. J. Bacteriol. 1987;169:2624–2630. [PMC free article] [PubMed]
8. Althaus EW, Outten CE, Olson KE, Cao H, O'H;alloran TV. The ferric uptake regulation (Fur) repressor is a zinc metalloprotein. Biochemistry. 1999;38:6559–6569. [PubMed]
9. Pohl E, Haller JC, Mijovilovich A, Meyer-Klaucke W, Garman E, Vasil ML. Architecture of a protein central to iron homeostasis: crystal structure and spectroscopic analysis of the ferric uptake regulator. Mol. Microbiol. 2003;47:903–915. [PubMed]
10. Kammler M, Schon C, Hantke K. Characterization of the ferrous iron uptake system of Escherichia coli. J. Bacteriol. 1993;175:6212–6219. [PMC free article] [PubMed]
11. Desai PJ, Angerer A, Genco CA. Analysis of Fur binding to operator sequences within the Neisseria gonorrhoeae fbpA promoter. J. Bacteriol. 1996;178:5020–5023. [PMC free article] [PubMed]
12. Heidrich C, Hantke K, Bierbaum G, Sahl HG. Identification and analysis of a gene encoding a Fur-like protein of Staphylococcus epidermidis. FEMS Microbiol. Lett. 1996;140:253–259. [PubMed]
13. McHugh JP, Rodriguez-Quinones F, Abdul-Tehrani H, Svistunenko DA, Poole RK, Cooper CE, Andrews SC. Global iron-dependent gene regulation in Escherichia coli. A new mechanism for iron homeostasis. J. Biol Chem. 2003;278:29478–29486. [PubMed]
14. Shannon CE. A mathematical theory of communication. Bell Syst. Tech. J. 1948;27:379–423. 623–656. http://cm.bell-labs.com/cm/ms/what/shannonday/paper.html.
15. Pierce JR. New York: Dover Publications, Inc.; 1980. An Introduction to Information Theory: Symbols, Signals and Noise, 2nd edn.
16. Schneider TD, Stormo GD, Gold L, Ehrenfeucht A. Information content of binding sites on nucleotide sequences. J. Mol Biol. 1986;188:415–431. http://www.ccrnp.ncifcrf.gov/~toms/paper/schneider1986/ [PubMed]
17. Schneider TD, Mastronarde D. Fast multiple alignment of ungapped DNA sequences using information theory and a relaxation method. Discrete Appli. Math. 1996;71:259–268. http://www.ccrnp.ncifcrf.gov/~toms/paper/malign. [PMC free article] [PubMed]
18. Schneider TD, Stephens RM. Sequence logos: a new way to display consensus sequences. Nucleic Acids Res. 1990;18:6097–6100. http://www.ccrnp.ncifcrf.gov/~toms/paper/logopaper/ [PMC free article] [PubMed]
19. Schneider TD. Information content of individual genetic sequences. J. Theor Biol. 1997;189:427–441. http://www.ccrnp.ncifcrf.gov/~toms/paper/ri/ [PubMed]
20. Schneider TD. Sequence walkers: a graphical method to display how binding proteins interact with DNA or RNA sequences. Nucleic Acids Res. 1997;25:4408–4415. http://www.ccrnp.ncifcrf.gov/~toms/paper/walker/, erratum: NAR 26(4): 1135, 1998. [PMC free article] [PubMed]
21. Panina EM, Mironov AA, Gelfand MS. Comparative analysis of FUR regulons in gamma-proteobacteria. Nucleic Acids Res. 2001;29:5195–5206. [PMC free article] [PubMed]
22. Schneider TD. Evolution of biological information. Nucleic Acids Res. 2000;28:2794–2799. http://www.ccrnp.ncifcrf.gov/~toms/paper/ev/ [PMC free article] [PubMed]
23. Vyhlidal CA, Rogan PK, Leeder JS. Development and refinement of pregnane X receptor (PXR) DNA binding site model using information theory: insights into PXR-mediated gene regulation. J. Biol Chem. 2004;279:46779–46786. [PubMed]
24. Zheng M, Doan B, Schneider TD, Storz G. OxyR and SoxRS regulation of fur. J. Bacteriol. 1999;181:4639–4643. http://www.ccrnp.ncifcrf.gov/~toms/paper/oxyrfur/ [PMC free article] [PubMed]
25. Hengen PN, Bartram SL, Stewart LE, Schneider TD. Information analysis of Fis binding sites. Nucleic Acids Res. 1997;25:4994–5002. http://www.ccrnp.ncifcrf.gov/~toms/paper/fisinfo/ [PMC free article] [PubMed]
26. Wood TI, Griffith KL, Fawcett WP, Jair, K.-W., Schneider TD, Wolf RE. Interdependence of the position and orientation of SoxS binding sites in the transcriptional activation of the class I subset of Escherichia coli superoxide-inducible promoters. Mol. Microbiol. 1999;34:414–430. [PubMed]
27. Zheng M, Wang X, Doan B, Lewis KA, Schneider TD, Storz G. Computation-directed identification of OxyR-DNA binding sites in Escherichia coli. J. Bacteriol. 2001;183:4571–4579. [PMC free article] [PubMed]
28. Rogan PK, Faux BM, Schneider TD. Information analysis of human splice site mutations. Hum. Mutat. 1998;12:153–171. Erratum in: Hum Mutat 1999;13(1):82. http://www.ccrnp.ncifcrf.gov/~toms/paper/rfs/ [PubMed]
29. Chen Z, Schneider TD. Information theory based T7-like promoter models: classification of bacteriophages and differential evolution of promoters and their polymerases. Nucleic Acids Res. 2005;33:6172–6187. http://www.ccrnp.ncifcrf.gov/~toms/papers/t7like/ [PMC free article] [PubMed]
30. Chen Z, Schneider TD. Comparative analysis of tandem T7-like promoter containing regions in enterobacterial genomes reveals a novel group of genetic islands. Nucleic Acids Res. 2006;34:1133–1147. http://www.ccrnp.ncifcrf.gov/~toms/papers/t7island/ [PMC free article] [PubMed]
31. Schneider TD. Reading of DNA sequence logos: prediction of major groove binding by information theory. Meth. Enzymol. 1996;274:445–455. http://www.ccrnp.ncifcrf.gov/~toms/paper/oxyr/ [PubMed]
32. Griggs DW, Konisky J. Mechanism for iron-regulated transcription of the Escherichia coli cir gene: metal-dependent binding of Fur protein to the promoters. J. Bacteriol. 1989;171:1048–1052. [PMC free article] [PubMed]
33. Angerer A, Braun V. Iron regulates transcription of the Escherichia coli ferric citrate transport genes directly and through the transcription initiation proteins. Arch. Microbiol. 1998;169:483–490. [PubMed]
34. de Lorenzo V, Herrero M, Giovannini F, Neilands JB. Fur (ferric uptake regulation) protein and CAP (catabolite-activator protein) modulate transcription of fur gene in Escherichia coli. Eur. J. Biochem. 1988;173:537–546. [PubMed]
35. Hunt MD, Pettis GS, McIntosh MA. Promoter and operator determinants for fur-mediated iron regulation in the bidirectional fepA-fes control region of the Escherichia coli enterobactin gene system. J. Bacteriol. 1994;176:3944–3955. [PMC free article] [PubMed]
36. Tardat B, Touati D. Iron and oxygen regulation of Escherichia coli MnSOD expression: competition between the global regulators Fur and ArcA for binding to DNA. Mol. Microbiol. 1993;9:53–63. [PubMed]
37. Escolar L, Perez-Martin J, de Lorenzo V. Evidence of an unusually long operator for the fur repressor in the aerobactin promoter of Escherichia coli. J. Biol. Chem. 2000;275:24709–24714. [PubMed]
38. Young GM, Postle K. Repression of tonB transcription during anaerobic growth requires Fur binding at the promoter and a second factor binding upstream. Mol. Microbiol. 1994;11:943–954. [PubMed]
39. Brickman TJ, Ozenberger BA, McIntosh MA. Regulation of divergent transcription from the iron-responsive fepB-entC promoter-operator regions in Escherichia coli. J. Mol Biol. 1990;212:669–682. [PubMed]
40. Baichoo N, Helmann JD. Recognition of DNA by Fur: a reinterpretation of the Fur box consensus sequence. J. Bacteriol. 2002;184:5826–5832. [PMC free article] [PubMed]
41. Ochsner UA, Vasil ML. Gene repression by the ferric uptake regulator in Pseudomonas aeruginosa: cycle selection of iron-regulated genes. Proc. Natl Acad. Sci. USA. 1996;93:4409–4414. [PMC free article] [PubMed]
42. Baichoo N, Wang T, Ye R, Helmann JD. Global analysis of the Bacillus subtilis Fur regulon and the iron starvation stimulon. Mol. Microbiol. 2002;45:1613–1629. [PubMed]
43. Christoffersen CA, Brickman TJ, Hook-Barnard I, McIntosh MA. Regulatory architecture of the iron-regulated fepD-ybdA bidirectional promoter region in Escherichia coli. J. Bacteriol. 2001;183:2059–2070. [PMC free article] [PubMed]
44. Chai S, Welch TJ, Crosa JH. Characterization of the interaction between Fur and the iron transport promoter of the virulence plasmid in Vibrio anguillarum. J. Biol. Chem. 1998;273:33841–33847. [PubMed]
45. Escolar L, Perez-Martin J, de Lorenzo V. Coordinated repression in vitro of the divergent fepA-fes promoters of Escherichia coli by the iron uptake regulation (Fur) protein. J. Bacteriol. 1998;180:2579–2582. [PMC free article] [PubMed]
46. Escolar L, de Lorenzo V, Perez-Martin J. Metalloregulation in vitro of the aerobactin promoter of Escherichia coli by the Fur (ferric uptake regulation) protein. Mol. Microbiol. 1997;26:799–808. [PubMed]
47. Shultzaberger RK, Chen Z, Lewis KA, Schneider TD. Anatomy of Escherichia coli σ70 promoters. Nucleic Acids Res. 2007;35:771–788. http://www.ccrnp.ncifcrf.gov/~toms/paper/flexprom/ [PMC free article] [PubMed]
48. Shultzaberger RK, Bucheimer RE, Rudd KE, Schneider TD. Anatomy of Escherichia coli ribosome binding sites. J. Mol Biol. 2001;313:215–228. http://www.ccrnp.ncifcrf.gov/~toms/paper/flexrbs/ [PubMed]
49. Hirao I, Kawai G, Yoshizawa S, Nishimura Y, Ishido Y, Watanabe K, Miura K. Most compact hairpin-turn structure exerted by a short DNA fragment, d(GCGAAGC) in solution: an extraordinarily stable structure resistant to nucleases and heat. Nucleic Acids Res. 1994;22:576–582. [PMC free article] [PubMed]
50. Eick-Helmerich K, Braun V. Import of biopolymers into Escherichia coli: nucleotide sequences of the exbB and exbD genes are homologous to those of the tolQ and tolR genes, respectively. J. Bacteriol. 1989;171:5117–5126. [PMC free article] [PubMed]
51. Vassinova N, Kozyrev D. A method for direct cloning of Fur-regulated genes: identification of seven new Fur-regulated loci in Escherichia coli. Microbiology. 2000;146:3171–3182. [PubMed]
52. Toledano MB, Kullik I, Trinh F, Baird PT, Schneider TD, Storz G. Redox-dependent shift of OxyR-DNA contacts along an extended DNA binding site: a mechanism for differential promoter selection. Cell. 1994;78:897–909. [PubMed]
53. Blattner FR, Plunkett,G. III, Bloch CA, Perna NT, Burland V, Riley M, Collado-Vides J, Glasner JD, Rode CK, et al. The complete genome sequence of Escherichia coli K-12. Science. 1997;277:1453–1474. [PubMed]
54. Bagg A, Neilands JB. Molecular mechanism of regulation of siderophore-mediated iron assimilation. Microbiol. Rev. 1987;51:509–518. [PMC free article] [PubMed]
55. Lavrrar JL, McIntosh MA. Architecture of a Fur binding site: a comparative analysis. J. Bacteriol. 2003;185:2194–2202. [PMC free article] [PubMed]
56. Orchard K, May GE. An EMSA-based method for determining the molecular weight of a protein – DNA complex. Nucleic Acids Res. 1993;21:3335–3336. [PMC free article] [PubMed]
57. Masse E, Gottesman S. A small RNA regulates the expression of genes involved in iron metabolism in Escherichia coli. Proc. Natl Acad. Sci USA. 2002;99:4620–4625. [PMC free article] [PubMed]
58. Lavrrar JL, Christoffersen CA, McIntosh MA. Fur–DNA interactions at the bidirectional fepDGC-entS promoter region in Escherichia coli. J. Mol. Biol. 2002;322:983–995. [PubMed]
59. Delany I, Rappuoli R, Scarlato V. Fur functions as an activator and as a repressor of putative virulence genes in Neisseria meningitidis. Mol. Microbiol. 2004;52:1081–1090. [PubMed]
60. Masse E, Vanderpool CK, Gottesman S. Effect of RyhB small RNA on global iron use in Escherichia coli. J. Bacteriol. 2005;187:6962–6971. [PMC free article] [PubMed]
61. Argaman L, Hershberg R, Vogel J, Bejerano G, Wagner EG, Margalit H, Altuvia S. Novel small RNA-encoding genes in the intergenic regions of Escherichia coli. Curr. Biol. 2001;11:941–950. [PubMed]
62. Wassarman KM, Repoila F, Rosenow C, Storz G, Gottesman S. Identification of novel small RNAs using comparative genomics and microarrays. Genes Dev. 2001;15:1637–1651. [PMC free article] [PubMed]
63. Robison K, McGuire AM, Church GM. A comprehensive library of DNA-binding site matrices for 55 proteins applied to the complete Escherichia coli K-12 genome. J. Mol. Biol. 1998;284:241–254. [PubMed]
64. Salgado H, Santos-Zavaleta A, Gama-Castro S, Millan-Zarate D, Diaz-Peredo E, Sanchez-Solano F, Perez-Rueda E, Bonavides-Martinez C, Collado-Vides J. RegulonDB (version 3.2): transcriptional regulation and operon organization in Escherichia coli K-12. Nucleic Acids Res. 2001;29:72–74. [PMC free article] [PubMed]
65. Andrews SC, Robinson AK, Rodriguez-Quinones F. Bacterial iron homeostasis. FEMS Microbiol. Rev. 2003;27:215–237. [PubMed]
66. Schneider TD. Consensus Sequence Zen. Appl. Bioinformatics. 2002;1:111–119. http://www.ccrnp.ncifcrf.gov/~toms/papers/zen/ [PMC free article] [PubMed]
67. Rudd KE, Schneider TD. Compilation of E. coli ribosome binding sites. In: Miller JH, editor. A Short Course in Bacterial Genetics: A Laboratory Manual and Handbook for Escherichia coli and Related Bacteria. New York: Cold Spring Harbor, Cold Spring Harbor Laboratory Press; 1992. pp. 17.19–17.45.
68. Hengen PN, Lyakhov IG, Stewart LE, Schneider TD. Molecular flip-flops formed by overlapping Fis sites. Nucleic Acids Res. 2003;31:6663–6673. [PMC free article] [PubMed]
69. Fee JA. Regulation of sod genes in Escherichia coli: relevance to superoxide dismutase function. Mol. Microbiol. 1991;5:2599–2610. [PubMed]
70. Igarashi K, Saisho T, Yuguchi M, Kashiwagi K. Molecular mechanism of polyamine stimulation of the synthesis of oligopeptide-binding protein. J. Biol. Chem. 1997;272:4058–4064. [PubMed]
71. Yoshida M, Meksuriyen D, Kashiwagi K, Kawai G, Igarashi K. Polyamine stimulation of the synthesis of oligopeptide-binding protein (OppA). Involvement of a structural change of the Shine-Dalgarno sequence and the initiation codon aug in oppa mRNA. J. Biol. Chem. 1999;274:22723–22728. [PubMed]
72. Litwin CM, Calderwood SB. Role of iron in regulation of virulence genes. Clin. Microbiol. Rev. 1993;6:137–149. [PMC free article] [PubMed]
73. Makui H, Roig E, Cole ST, Helmann JD, Gros P, Cellier MF. Identification of the Escherichia coli K-12 Nramp orthologue (MntH) as a selective divalent metal ion transporter. Mol. Microbiol. 2000;35:1065–1078. [PubMed]
74. Patzer SI, Hantke K. Dual repression by Fe(2+)-Fur and Mn(2+)-MntR of the mntH gene, encoding an NRAMP-like Mn(2+) transporter in Escherichia coli. J. Bacteriol. 2001;183:4806–4813. [PMC free article] [PubMed]
75. Furrer JL, Sanders DN, Hook-Barnard IG, McIntosh MA. Export of the siderophore enterobactin in Escherichia coli: involvement of a 43 kDa membrane exporter. Mol. Microbiol. 2002;44:1225–1234. [PubMed]
76. Cunningham L, Gruer MJ, Guest JR. Transcriptional regulation of the aconitase genes (acnA and acnB) of Escherichia coli. Microbiology. 1997;143 (Pt 12):3795–3805. [PubMed]
77. Fuentes AM, Diaz-Mejia JJ, Maldonado-Rodriguez R, Amabile-Cuevas CF. Differential activities of the SoxR protein of Escherichia coli: SoxS is not required for gene activation under iron deprivation. FEMS Microbiol. Lett. 2001;201:271–275. [PubMed]
78. Varghese S, Tang Y, Imlay JA. Contrasting sensitivities of Escherichia coli aconitases A and B to oxidation and iron depletion. J. Bacteriol. 2003;185:221–230. [PMC free article] [PubMed]
79. Zavitz KH, DiGate RJ, Marians KJ. The PriB and PriC replication proteins of Escherichia coli. Genes, DNA sequence, overexpression, and purification. J. Biol. Chem. 1991;266:13988–13995. [PubMed]
80. Sandler SJ. Requirements for replication restart proteins during constitutive stable DNA replication in Escherichia coli K-12. Genetics. 2005;169:1799–1806. [PMC free article] [PubMed]
81. Schneider TD. Strong minor groove base conservation in sequence logos implies DNA distortion or base flipping during replication and transcription initiation. Nucleic Acids Res. 2001;29:4881–4891. http://www.ccrnp.ncifcrf.gov/~toms/paper/baseflip/ [PMC free article] [PubMed]

Articles from Nucleic Acids Research are provided here courtesy of Oxford University Press
PubReader format: click here to try


Save items

Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...


  • Gene
    Gene records that cite the current articles. Citations in Gene are added manually by NCBI or imported from outside public resources.
  • Gene (nucleotide)
    Gene (nucleotide)
    Records in Gene identified from shared sequence and PMC links.
  • GEO Profiles
    GEO Profiles
    Gene Expression Omnibus (GEO) Profiles of molecular abundance data. The current articles are references on the Gene record associated with the GEO profile.
  • MedGen
    Related information in MedGen
  • Nucleotide
    Primary database (GenBank) nucleotide records reported in the current articles as well as Reference Sequences (RefSeqs) that include the articles as references.
  • Protein
    Protein translation features of primary database (GenBank) nucleotide records reported in the current articles as well as Reference Sequences (RefSeqs) that include the articles as references.
  • PubMed
    PubMed citations for these articles

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...