A Reassessment of the FNR Regulon and Transcriptomic Analysis of the Effects of Nitrate, Nitrite, NarXL, and NarQP as Escherichia coli K12 Adapts from Aerobic to Anaerobic Growth*

The transcription factor FNR, the regulator of fumarate and nitrate reduction, regulates major changes as Escherichia coli adapts from aerobic to anaerobic growth. In an anaerobic glycerol/trimethylamine N-oxide/fumarate medium, the fnr mutant grew as well as the parental strain, E. coli K12 MG1655, enabling us to reveal the response to oxygen, nitrate, and nitrite in the absence of glucose repression or artifacts because of variations in growth rate. Hence, many of the discrepancies between previous microarray studies of the E. coli FNR regulon were resolved. The current microarray data confirmed 31 of the previously characterized FNR-regulated operons. Forty four operons not previously known to be included in the FNR regulon were activated by FNR, and a further 28 operons appeared to be repressed. For each of these operons, a match to the consensus FNR-binding site sequence was identified. The FNR regulon therefore minimally includes at least 103, and possibly as many as 115, operons. Comparison of transcripts in the parental strain and a narXL deletion mutant revealed that transcription of 51 operons is activated, directly or indirectly, by NarL, and a further 41 operons are repressed. The narP gene was also deleted from the narXL mutant to reveal the extent of regulation by phosphorylated NarP. Fourteen promoters were more active in the narP+ strain than in the mutant, and a further 37 were strongly repressed. This is the first report that NarP might function as a global repressor as well as a transcription activator. The data also revealed possible new defense mechanisms against reactive nitrogen species.

In several recent studies, genome-wide transcription data have been analyzed to reveal the extent of the biochemical changes as Escherichia coli K12 adapts from aerobic to anaerobic growth. Salmon et al. (1) compared RNA isolated from cultures of strain MC4100 that had been grown aerobically or anaerobically and also from an anaerobic culture of an fnr mutant that lacks FNR, 2 the regulator of fumarate and nitrate reduction, which is a global regulator of many oxygen-regulated genes.
In similar experiments, Kang et al. (2) grew both strain MG1655, for which the complete genome sequence is available, and an isogenic fnr mutant aerobically and anaerobically in a minimal salts medium, and compared their transcriptome data with those from the previous study. In both groups of experiments, glucose was used as the carbon source for growth, and in both studies rigorous statistical methods and cluster analysis were used to analyze the data. In the former study, expression levels of 1,445 genes changed in response to the availability of oxygen.
Although the corresponding figure in the study of Kang et al. (2) was 962, only 334 genes were common to both data sets, and of those, 123 appeared to be regulated in opposite directions. Thus only 211 genes showed similar responses, 10% of the 2073 genes for which changes were observed.
Both of the previous studies were valuable in revealing the far greater extent of changes in response to FNR and oxygen than had been revealed by previous operon-by-operon approaches or from proteomic studies (3,4). They also revealed previously unsuspected possible links between different global regulons, in addition to the already established dependence of the ArcB-ArcA regulon on FNR (5). However, neither study was designed to define the number of operons that are regulated directly rather than indirectly by FNR, nor were different anaerobic growth conditions used to identify operons that are either repressed during growth with glucose (6) or are co-dependent upon nitrate-activated NarXL or NarQP for expression (7)(8)(9)(10). Consequently, changes in expression levels of operons known from a range of genetic, biochemical, and in vitro data to be dependent upon FNR activation were not detected in the previous studies, and the nirB promoter in which there is a consensus binding site for FNR activation even appeared to be repressed by FNR (1).
We now report results from a more extensive study designed to reveal the extent of the FNR regulon in E. coli strain MG1655. Particular attention has been paid to defining growth conditions in which most (but clearly not all) FNR-dependent promoters are active. A specific challenge was to find growth conditions in which the anaerobic growth rates of the fnr mutant are comparable with those of the parental strain, but glucose repression was avoided. The effects of nitrate and nitrite on the E. coli transcriptome during anaerobic growth have also been investigated, resulting in a new list of operons that are apparently regulated by both FNR and NarL. Proposals based solely upon microarray and bioinformatic information were then confirmed by independent experiments, including analysis of the transcriptomes of narXL and narXL-narP deletion strains during anaerobic growth in the presence of nitrate.

EXPERIMENTAL PROCEDURES
Strains and Mutant Construction-E. coli K12 strain MG1655 (CGSC7740), for which the chromosomal DNA sequence is known, was used throughout this study (11). A ⌬fnr derivative was constructed by the gene-replacement method using primers K12 FNR1 and K12 FNR2 to amplify the chloramphenicol acetyltransferase gene cassette from plasmid pKD3 (12). The nucleotides underlined for each primer ( Table  1) represent sequences that are homologous to genomic K12 DNA flanking fnr, and were used for the pKD46 Red-mediated recombination of the chloramphenicol resistance gene cassette into the E. coli MG1655 chromosome (12). After electroporation, MG1655 transformants were grown on Luria Bertani agar (13) supplemented with 50 g/ml chloramphenicol. Primers FNR PRIMER A and FNR PRIMER B were used to screen chloramphenicol-resistant colonies for replacement of fnr by the chloramphenicol resistance cassette. PCR products were sequenced using a BigDye TM version 3 sequencing kit (Applied Biosystems, Warrington, UK) and analyzed on an ABI 3700 DNA sequencer (Applied Biosystems). The fnr mutant was designated strain JCB1001.
The narXL deletion was constructed in strain JCB4021 (14) by the same method using primers NarX-P1 and NarL-P2 and with plasmid pKD3 DNA as PCR template. The deletion was transferred by bacteriophage P1-mediated transduction into strain MG1655 with selection for chloramphenicol resistance. Purified transductants were checked by PCR with primers XL1 and XL2 and phenotypically for the correct retention of adjacent narK and narGHJI genes. The resulting strain was designated JCB1002. A further derivative in which the narP gene was inactivated was constructed by using bacteriophage P1 propagated in strain JCB3875 (8) to transduce strain JCB1002 to chloramphenicol resistance. To confirm that the narP allele had replaced the narP ϩ gene, purified transductants and strain JCB1002 were transformed with a napF::lacZ transcriptional fusion plasmid. Transformants were grown anaerobically in minimal medium supplemented with 0.4% glucose in the presence or absence of 20 mM nitrate, and the resulting ␤-galactosidase activity was assayed. The purified narXL narP mutant in which transcription from pnapF was no longer activated in the presence of nitrate was designated strain JCB1003. The deletion mutants were finally verified as containing unique chromosomal deletions using genomic DNA microarray hybridizations. Strain JCB1011, a gift from Dr. Hirofumi Aiba, is a derivative of E. coli MG1655 carrying a chromosomally encoded C-terminal 3xFLAG-tagged fnr gene.
Growth Conditions-Stock cultures were stored as liquid cultures in 50% glycerol at Ϫ20°C. A standardized procedure was used to prepare inocula for cultures used to grow the various strains in different media. Inocula were streaked onto a fresh nutrient agar (Oxoid) plate. After 16 h at 37°C, a 0.5-cm streak of colonies showing confluent growth was transferred into a 250-ml conical flask containing 30 ml of minimal salts medium (15) supplemented with 0.4% glycerol, 40 mM sodium fumarate, 20 mM trimethylamine N-oxide (TMAO), and 10% (v/v) Luria-Bertani broth (LB). The culture was aerated rapidly at 37°C until the absorbance at 650 nm had increased to 0.25 (100 g of bacterial dry mass per ml). For a few experiments, 70 ml of the same medium in a 500-ml conical flask was used.
For anaerobic cultures, 20 ml of the aerobic inoculum was transferred to 80 ml of the same medium in a 100-ml conical flask, and the culture was incubated without agitation at 37°C until the optical density had increased to 0.5-0.6 (corresponding to a biomass density of 0.20 -0.24 g/liter). TMAO or fumarate was omitted in some control experiments, and 5 mM sodium nitrite or 20 mM potassium nitrate was added, as required. Both nitrite and nitrate were still present in the appropriate cultures when samples were harvested.
Construction of E. coli Whole-genome Array-Highly specific 70-mer oligonucleotides from the Operon E. coli Array Ready Oligonucleotide Set TM version 1.0 were printed with Lucidea TM Universal Scorecard TM (Amersham Biosciences) controls, as described previously (16), onto Corning UltraGAPS TM slides (Corning Life Sciences, Netherlands) using Corning Pronto! spotting solution. Slides were desiccated under vacuum for 48 h and then UV light cross-linked with a dose of 600 mJ using a Stratalinker (Stratagene limited, UK). Slides were stored under desiccating conditions in the dark until use. RNA and DNA Isolation-Pool RNA was isolated from four independently grown cultures of E. coli K12 MG1655. Except for the control experiments for which the pool of RNA was prepared from aerobically grown bacteria, the pool of control RNA was isolated from independent cultures of bacteria grown anaerobically in the same medium in the absence of nitrate or nitrite but in the presence of TMAO. Test RNA was isolated from three independently grown cultures for each growth condition. Samples were mixed with 2 volumes of RNA Protect (Qiagen) to stabilize the total RNA. An RNeasy kit (Qiagen) was used to extract the total RNA according to manufacturer's instructions. Contaminating DNA was removed by using on-column DNase I digestion (Qiagen). The quality and quantity of the RNA preparations were determined with an Agilent 2100 Bioanalyzer by using the RNA 6000 nanoassay Labchip (Agilent, Stockport, UK).
DNA was isolated from 15 ml of culture grown to OD 0.7-0.8 using Genomic Tip 500/G, genomic DNA buffer set, and proteinase K stock solution (Qiagen) following the manufacturer's instructions. The quality and quantity of DNA were determined by gel electrophoresis (0.8% agarose gel) with the aid of Bioline Hyperladder I.
Reverse Transcription and Labeling-Total RNA (20 g) was reversetranscribed to Cy3-and Cy5-labeled cDNA using the CyScribe postlabeling kit (Amersham Biosciences/GE Healthcare) according to the manufacturer's instructions. Pool samples were labeled with Cy3 N-hydroxysuccinimide ester dye, and samples from the various other growth conditions and genetic backgrounds were labeled with Cy5 N-hydroxysuccinimide ester (Amersham Biosciences). Lucidea Universal Scorecard spike mRNA reference controls (Amersham Biosciences) were added to the reverse transcriptase reactions for both samples.
DNA (5 g) isolated from each deletion mutant and the wild type strain was labeled with FluoroLink TM Cy5-dCTP and FluoroLink TM Cy3-dCTP, respectively, using random hexamer primers at final concentrations of 60 ng/l (Invitrogen), 50 units of Klenow fragment of DNA polymerase I (Biolab UK), dNTPs at 0.1 mM dA/G/TTP, 0.04 mM dCTP, and 1ϫ EcoPol buffer in a 50-l reaction. Samples were incubated at 37°C for at least 90 min. Labeled genomic DNA was purified using QIAquick PCR purification columns (Qiagen) following the manufacturer's protocol. DNA was eluted in 2ϫ 35-l sterile high pressure liquid chromatography grade water.
Hybridization-Prior to hybridization, the slides were treated with the background-reducing Pronto! Pre-Soak System (Corning Life Sciences). The slides were incubated for 20 min in prewarmed Universal Pre-Soak solution at 42°C and washed twice in 0.1ϫ SSC, 0.1% SDS for 30 s at room temperature. Slides were immediately transferred into prewarmed prehybridization solution (5ϫ SSC, 0.1% SDS, and 0.1% bovine serum albumin) and incubated for 2-4 h at 42°C. The microarray slides were finally washed at room temperature once in 0.1ϫ SSC, 0.1% SDS for 1 min and twice in 0.1ϫ SSC for 30 s, briefly dipped in water and then ethanol, and dried by centrifugation for 5 min at 800 ϫ g.
For each experiment, equal quantities (80 pmol) of each Cy5-and Cy3-labeled cDNA were added to a final volume of 80 l of hybridization solution containing 25% formamide, 10 mg of bovine serum albumin (fraction V) per ml, 5ϫ SSC (1ϫ SSC is 0.15 M NaCl, 0.015 M sodium citrate), 0.1% SDS, 8 g of poly(A), and 1ϫ Denhardt's solution. The cDNA probes were denatured at 95°C for 3 min and hybridized for 16 h at 42°C. After hybridization was complete, the slides were washed in 2ϫ SSC, 0.1% SDS at 42°C for 2 min in 0.1ϫ SSC, 0.1% SDS for 2 min at room temperature, and finally twice in 0.1ϫ SSC for 2 min at room temperature. The microarray slides were dried by centrifugation for 5 min at 800 ϫ g and were then scanned at 532 and 630 nm by using a Genepix 4000A scanner (Axon Instruments, Union City, CA).
Analysis of Microarray Data-The microarray images were analyzed using GenePix software (Axon Instruments), and spots that either failed to print or to hybridize were flagged and rejected (at most 20 K12 genes in any experiment). The data were imported into GeneSpring, version 6.1 (Silicon Genetics, Redwood City, CA). A Lowess curve (locally weighted linear regression curve) was fitted to the plot of log intensity versus log ratio, and 40% of the data were used to calculate the Lowess fit at each point. The curve was used to adjust the control value for each measurement. If the control channel signal was below a threshold value of 10, then 10 was used instead.
Spots with an intensity value lower than the GeneSpring error model cut-off value were then filtered out, which in different experiments were in the range 40 -70. A second filter was then applied that required a fold change less than 0.667 and greater than 1.334 across all conditions. The resulting list was used to extract genes that showed at least 2-fold differential expression levels between the test sample and the reference pool under the different growth conditions by using Student's t test and applying the Benjamini and Hochberg false discovery rate test with a p value cut off of 0.05. The resulting list was used with the same statistical analysis to extract genes that showed at least 2-fold differential expression levels between strains with different genetic backgrounds and/or between different growth conditions using Student's t test and applying Benjamini and Hochberg false discovery rate test with a p value cut-off value of 0.05.
Identification of FNR-and NarP Promoter-binding Sites-Potential FNR-and NarP-binding sites were located using the sequence motif search facility of Colibri (genolist.pasteur.fr/Colibri/genome.cgi) allowing two mismatches. For FNR, the consensus sequence TTGAT-NNNNATCAA was used. For NarP, the consensus sequence TACYN-MTNNAKNRGTA was used. The search was limited to 500 bp upstream of open reading frames. For FNR-binding sites, a positionweight matrix-based search developed by Robison et al. (17) was also consulted (available on-line at arep.med.harvard.edu/ecoli_matrices/). Operon structures and transcription start sites were taken from either RegulonDB (www.cifn.unam.mx/Computational_Genomics/regulondb/) or ecocyc (www.ecocyc.com). Only potential FNR-and NarPbinding sites within 200 bp upstream of the transcription start site were considered.
Chromatin Immunoprecipitation-FNR-DNA interactions were studied by chromatin immunoprecipitation (ChIP) (18). Cultures of E. coli strain JCB1011 carrying a chromosomally encoded C-terminal 3xFLAG-tagged fnr gene were grown as described for RNA extraction. Cross-linking, chromatin preparation, and immunoprecipitations were described previously (18), except that the tagged FNR was immunoprecipitated with anti-FLAG monoclonal antibodies (Sigma) for 16 h at 4°C. The concentration of potential FNR-bound promoter fragments was measured using real time quantitative PCR (QPCR) (19). Primers for each promoter were designed using PrimerExpress (Applied Biosystems) and are listed in Table 1. The yabN promoter, to which FNR does not bind, was used as a negative control to normalize the data. Promoter fragments enriched by 2-fold or more in at least two independent ChIP experiments, relative to the yabN promoter fragment, were scored positive.
Data Accession-Data from these experiments are accessible via Gene expression omnibus (GEO) accession number GSE3591.

Strategy for Experimental Design and Avoidance of Some Common
Artifacts-An fnr mutant is either unable to grow anaerobically in the presence of most terminal electron acceptors and a nonfermentable carbon source such as glycerol or lactate or grows far more slowly than the parental strain (20). Under such conditions, any differences in the transcriptomes of an fnr mutant and its parental strain would be due to both direct effects of FNR and to differences in growth rate. Presumably this is why in previous studies glucose was used as the carbon source for growth (1,2), despite the fact that glucose represses expression from some FNR-activated promoters (6,21). Replacement of glucose by a less repressing fermentable carbohydrate would decrease effects because of glucose repression but to an unknown extent. We therefore exploited the fact that fnr mutants can be grown anaerobically in the presence of the nonfermentable and nonrepressing carbon source, glycerol, in the presence of TMAO in addition to fumarate as the terminal electron acceptor. Furthermore, the presence of TMAO has a minimal effect on NarX-NarL or NarQ-NarP-dependent induction or repression. Under these conditions, the fnr mutant grew as well as the parental strain, but not if either TMAO or fumarate was omitted (Fig. 1), and indicated that use of the glycerol/TMAO/fumarate medium would enable us to reveal the response of the E. coli transcriptome to nitrate, nitrite, and the twocomponent regulator system NarX-NarL. In control experiments, RNA isolated from bacteria grown under rigorously anaerobic conditions in an anaerobic cabinet was compared with RNA from cultures grown in unshaken conical flasks, as used in this study. No differences in transcript levels were observed, justifying the simpler growth conditions used for this and many previous studies.
A second key feature in the experimental design was the use in each large set of experiments of a common pool of reference RNA isolated from bacteria that had been grown anaerobically, and in which FNRactivated genes were expressed at a significant level. Preliminary experiments (data not shown) confirmed that there was excellent (Ͼ98%) agreement between hybridization data from single RNA preparations and the reference pool. It was therefore possible to make comparisons between all of the data from individual batches of experiments. A potential disadvantage of this approach was the risk that some promoters repressed by FNR would be expressed at such a low level that the microarray signals would be too low to yield reliable data and would be rejected by the statistical screens used. To check for this artifact, further experiments were completed in which the reference RNA was a pool of samples isolated from bacteria in the early exponential phase of aerobic growth.
In further control experiments, RNA was isolated from the same anaerobic cultures at different stages of the growth cycle, and concentrations of RNA from genes that are known to be activated by FNR were monitored. Similar levels of these RNA species were detected in bacteria harvested at culture densities in the range 0.14 -0.32 mg dry mass/ml, but levels were lower earlier in the growth cycle or as the growth rate declined when the culture approached the stationary phase. RNA samples from anaerobic cultures were always isolated at a culture density in the narrow range of 0.18 -0.22 mg/ml.
Overview of the Effects of Oxygen, Nitrate, and Nitrite on the Transcriptomes of MG1655 and an fnr Mutant-RNA was isolated from early exponential phase aerated cultures of MG1655 and an isogenic fnr mutant and from mid-exponential phase cultures of the same strains growing anaerobically in the presence or absence of nitrate or nitrite. RNA from three biologically independent experiments for each of the seven types of culture was compared with the reference pool of RNA from the parental strain that had been grown in the absence of nitrate or nitrite. Only greater than 2-fold differences in microarray signals and with a probability of a false positive identification of less than 5% were considered to be significant. Supplemental Table S1 lists raw data for 502 genes for which transcript levels were significantly affected by FNR during anaerobic growth in at least one growth condition. Transcripts that are less abundant (repressed) in the reference pool relative to the test sample include those that are induced directly or indirectly during aerobic growth, transcripts that are induced during anaerobic growth by nitrate or nitrite, or repressed by FNR. Conversely, transcripts that are more abundant in the reference pool than in the test sample include genes that, directly or indirectly, are activated by FNR or induced during anaerobic growth. For cultures supplemented with nitrate or nitrite, entries in supplemental Table S1 indicate the levels of induction or repression by nitrate or nitrite in the parental strain or in the fnr mutant relative to the pooled RNA from anaerobic cultures unsupplemented with nitrate or nitrite. Transcripts that were found to vary significantly between the fnr versus fnr ϩ strains were then assigned to operons, which in turn were sorted into those previously reported to be FNR-regulated on the basis of biochemical or genetic data ( Table 2) and others not previously known to be FNR-regulated, but for which a near-consensus FNR-binding site could be identified based upon at least one of the two criteria used (Table 3). Table 2, parts B and D, includes the microarray data for a few additional operons or promoters for which there is independent evidence in the literature that an operon or promoter is activated or repressed by FNR. Table 2 includes 29 operons for which there is previous evidence of FNR-dependent activation and 14 operons previously reported to be repressed by FNR. For each entry, the function of the gene products, the location of the FNR-binding sites, the evidence on which FNR dependence is based, key references to the literature, and other transcription factors known to regulate the operon are also listed. Note, however, that these microarray data alone did not confirm FNR activation of five of these operons (adhE, glpTQ, cydDC, hlyE, and arcA; Table 2, part B) or FNR repression of the seven operons listed in Table 2, part D.

Comparison with Previous Microarray Studies of the Effects of Oxygen and FNR on the Transcription of Previously Identified FNR-dependent
Operons-The first three columns of ratios in Table 2 provide a basis for comparison with previous microarray studies of the effects of oxygen and FNR on transcription of operons known from independent studies to be FNR-dependent. On the sole basis of insignificant induction during anaerobic growth compared with aerobic growth (Table 2, part A, 1st column, which lists the ratio of aerobic to anaerobic transcript levels for the parental strain), the hcp-hcr, fliZY, napF, nirB, nikA, aspA, and nrdD operons could not be assigned to the FNR regulon, despite independent evidence to the contrary. Furthermore, in none of the current experiments could FNR dependence of the five operons listed in Table (2), who also found that 11 of the 29 operons in Table 2, part A, were significantly activated by FNR and 4 of the 7 operons in Table 2, part C, were repressed by FNR. Note, however, the total agreement between this study and that of Kang et al. (2) concerning the five operons that failed to reveal the previously reported FNR activation ( Table 2, part B), and the seven operons that were not confirmed to be FNR-repressed (Table 2, part D). Even with only this limited data set, the conclusions that could be drawn were more consistent with published data than those from the previous studies (1, 2), but the fact that many of the differences were small indicated that other factors must be considered for the full extent of the E. coli FNR regulon to be revealed from microarray data alone. Simply avoiding the use of the repressing carbon source, glucose, allowed the FNR dependence of the nrf and nap promoters to be demonstrated, even under growth conditions that were suboptimal for expression. Table 2, 4th column, show that only five of the previously identified FNR-activated operons, hcphcr, narK, narGHJI, fdnGHJI, and nirBDC, were strongly induced by nitrate, and this induction was FNR-dependent (Table 2, 5th column). Twelve previously identified FNR-dependent operons were repressed more than 2-fold by nitrate in the parental strain (Table 2, 4th column). With the sole exception of adhE, even less transcript accumulated in the fnr mutant (Table 2, compare 4th and 5th columns), again confirming the FNR dependence of transcription from these promoters. Although transcription of the remaining previously characterized FNR-dependent operons was unaffected, or only slightly affected, by the presence of nitrate during anaerobic growth, in each case less transcript accumulated in the fnr mutant compared with the parental strain.

Effects of Nitrate and Nitrite on Transcription from Previously Identified FNR-dependent Promoters-Data in
All of the seven operons that were both previously reported and confirmed in the current experiments to be repressed by FNR were induced by nitrate in the fnr mutant. Four of these operons were also nitrateinduced in the parental strain ( Table 2, part C).
The effects of growth in the presence of nitrite on known FNR-dependent promoters were particularly interesting for four reasons. First, all five operons that were strongly activated by nitrate were also activated in the presence of nitrite (Table 2, 6th column). However, consistent with an abundance of independent studies (7, 23), both the fdnG and narG promoters were more strongly activated by nitrate than by nitrite. Second, the napF operon, which is known to be transcribed from at least two promoters (24), and the nrfA promoter ( Table 2, 6th and 10th columns), encoding the periplasmic nitrate and nitrite reductases, were  FEBRUARY 24, 2006 • VOLUME 281 • NUMBER 8 activated in the presence of nitrite, but repressed by nitrate. Third, both the hcp-hcr operon and narK, like the well studied nirB promoter, were equally active during growth in the presence of nitrate or nitrite, suggesting that the products of these genes are functionally related. Finally, in contrast to the well characterized effects of nitrate, nitrite, NarL, and NarP at the nrf and nap promoters, which are activated by NarP but antagonized or repressed by NarL, the dcuB-fumB operon was strongly repressed by both nitrate and nitrite, suggesting that it might provide the first example of a promoter that is repressed by NarP as well as by NarL. For all of these promoters, induction was strongly FNR-dependent (Table 2, 5th and 7th columns). In summary, the combined data in Table 2 justified the experimental design and methods used, having for the first time revealed by genomewide transcription experiments FNR-dependent regulation of the majority of operons that had been assigned by independent studies to the FNR regulon. Table 3 lists a further 44 operons that had not previously been shown to be FNR-dependent but appeared by one or more criteria to be activated by FNR during anaerobic growth. Thirteen of these operons were also identified in the study of Kang et al. (2), but anaerobic growth in the presence of nitrate or nitrite was essential for revealing the FNR dependence of most of the remaining operons. These were the only additional differentially regulated operons for which there is a near-consensus FNR-binding site within 300 bases of the first gene of the operon. Transcription start sites have been identified for only 10 of these 44 putative operons. Two of the most convincing examples were ydfZ encoding a hypothetical protein and the ynfEFGHI operon that encodes components of an alternative sulfoxide reductase (25). As in previous studies, many of the most striking examples of FNR-dependent activation involved genes of unknown function (2) as follows: ydfZ, ynfK, yecR, yjdAZ, yjdK, and yjjW. Three exceptions were the ferric enterobactin synthesis and transport operon, fes-entF-fepE, ompW encoding an outer membrane protein and colicin S4 receptor, and cadC encoding the transcriptional activator of the cad operon (but why was the cad operon itself apparently not FNR regulated?). Perhaps the most surprising result was that only one new operon was significantly induced by nitrate and that it was by only 2.8-fold (Table 3, 4th column). A further 11 FNR-dependent operons that were repressed by nitrate included ynfEFGHI and ydhYVWXU encoding putative oxidoreductases, as well as the fes and cadC promoters. The yccM gene was induced 2.8-fold by nitrite compared with only 1.6-fold by nitrate (compare Table 3, 4th and 6th columns), suggesting that its promoter might be activated by NarP rather than by NarL. Although the malPQ transcript was 2-fold less abundant during anaerobic growth of the fnr mutant than the parental strain, it was more than 4-fold more abundant in the mutant during aerobic growth. It is therefore unlikely that malPQ is part of the FNR regulon.

Identification of Previously Undescribed FNR-dependent Operons-
Twenty-eight operons not previously known to be regulated by FNR appeared to be repressed by FNR during anaerobic growth. Table 3 lists positions of possible FNR-binding sites in the vicinity of the predicted or mapped transcription start site for each of these operons. In many cases the effects were small; exceptions were the strongly regulated cyoAB-CDE operon that during anaerobic growth was 60-fold de-repressed in an fnr mutant relative to the parental strain (and also strongly activated by nitrate), aldA encoding an NAD ϩ -linked aldehyde dehydrogenase, and kgtP encoding 2-oxoglutarate permease. FNR repression was often most apparent during anaerobic growth in the presence of nitrite. Striking examples include two operons encoding proteins of unknown function, ygbA and yibIH, both of which are also strongly induced by nitrate.

Confirmation of the Microarray Data Using a Pool of RNA from Aerobically Grown Bacteria as the Reference-The microarray experiments
in which a pool of RNA from cultures of the parental strain grown anaerobically in the absence of nitrate or nitrite was used as the reference sample had indicated that the E. coli FNR regulon consists of ϳ104 operons, 68 that are activated and 36 that are repressed by FNR. Several approaches were used to check the reliability of the assignments. First, recognizing that genes repressed by FNR would be poorly expressed during anaerobic growth, a further 21 microarray experiments were completed using a pool of RNA from multiple aerobic cultures of the parental strain as the reference. These data (not shown, but available online) failed to reveal other operons that might be regulated directly by FNR.
The Effects of a narXL Deletion Mutation on the E. coli Transcriptome; Definition of the NarXL Regulon-Some well characterized FNRdependent promoters are also regulated by NarL in response to the presence of nitrate (23), and the regulatory effects, both positive and negative, at many other promoters were most apparent during anaerobic growth in the presence of nitrate or nitrite. To reveal the extent of direct or indirect regulation by NarX-NarL, RNA was isolated from the parental strain and its isogenic narXL mutant during anaerobic growth in the presence of nitrate. Levels of each transcript were again compared with those in the RNA pool from the parental strain grown in the absence of nitrate. The data were analyzed and filtered; data were retained only from genes with at least a 2-fold difference in expression level between either the parental strain and the narXL mutant or the parental strain grown in the presence and absence of nitrate. As phosphorylated NarL binds to a degenerate heptamer sequence, there are more than 14,000 potential NarL-binding sites on the E. coli chromosome. It was therefore inappropriate to attempt to correlate the presence of these sites with nitrate-regulated promoters. The full data are shown in supplemental Table S2; 92 operons were differentially regulated. The key points are summarized in Table 4, where operons have been divided into six groups, depending upon their pattern of regulation. For each group, the regulation of one or more operons by NarL and/or NarP had been characterized previously.
Fifty one operons were more highly expressed in the parental strain than in the narXL mutant. These operons can be clustered into two groups. Forty four operons were induced by nitrate in the parental strain but not in the narXL mutant (Table 4, part A). This group includes well characterized promoters that are activated by NarL but not by NarP, such as narK and narG (7,23). Note, however, that this group also includes sodA (6.7-fold induction), which is regulated by FNR and other transcription factors, and cyoA, which is known to be repressed by ArcA during anaerobic growth. Loss of active nitrate reduction in the narXL mutant would change the redox state of the ubiquinone/ubiquinol pool and thus indirectly affect their transcription control by ArcA (26,27). The ogt gene encoding a methyltransferase that repairs alkylation lesions in DNA was induced nearly 13-fold by nitrate in the parental strain but 1.5-fold in the narXL mutant.
The second cluster of 7 NarXL-activated operons is also significantly induced by nitrate even in the narXL mutant (Table 4, part B). The hcp promoter is typical of this group, being induced 30-fold by nitrate in the parental strain and nearly 14-fold in the narXL mutant. Previous studies have shown that hcp expression is dependent upon both NarXL and NarQP (41). Even larger nitrate induction ratios are observed for other genes in this cluster, 90-fold for yeaR-yoaG, and 50-fold for ytfE, which encodes a protein involved in resistance to nitric oxide stress (28). Also in this group are hmpA (activated 20-fold) and a gene encoding a hypothetical protein, yedF. These operons are likely to be regulated by NarL and either NarP, as is the case of hcp, or other transcription factors in response to nitrate or other stimuli.
A further 11 operons were more highly expressed during anaerobic growth in the presence of nitrate than in its absence but were not significantly dependent upon NarXL (Table 4, part C). The nirB promoter has been identified as being activated by either NarL or NarP; the NarQP gene products fully compensate for the narXL deletion (9). Other operons in this group might also be similarly activated by either NarL or NarP.
The 41 operons that appeared to be repressed by NarL, directly or indirectly, could be assigned to three subgroups. Three operons, nap, nrf, and yjiM, were induced by nitrate only in the narXL mutant (Table  4, part D). The regulation of nap and nrf has been characterized previously (6,10,24,29,30); NarL antagonizes transcription activation in the presence of nitrate, but in the narXL mutant this repression is lifted, and NarP activates transcription. Therefore, these are good candidates in this group for repression by NarL but activation by NarP.
Four operons were repressed not only by NarL but also by nitrate both in the parental strain and the narXL mutant (Table 4, part E). The hyaA promoter has been reported to be repressed by both NarL and NarP; the other operons in this group possibly share this pattern of regulation.
The contiguous fumB and dcuB genes are also in this group and were already known to be regulated by FNR, NarL, and ArcA (31); thus, at this promoter, regulation by NarL might be both direct and indirect (via the ubiquinol pool and ArcA). Two of the most obvious possible explanations for these results are either that these genes are repressed by the alternative nitrate-responsive transcription factor, NarP, in response to nitrate or that the metabolites required to induce transcription from these promoters are less available during anaerobic growth in the presence of nitrate.
Finally, 34 operons not falling into the above two groups were repressed by NarXL (Table 4, part F). The typical pattern of regulation for this group was a significant difference between the parental strain and narXL mutant but no significant activation by nitrate in either strain. Operons in this group include frdA and dmsA, both of which are known to be subject to hierarchical control by FNR and NarL in response to oxygen and nitrate (32,33). This group also includes operons such as hybO that are thought to be repressed by ArcA during anaerobic growth, so the effects of NarL at these promoters might again be direct, indirect (via the ubiquinol pool), or both.
The Effects of a narP Deletion Mutation on the E. coli Transcriptome-Many examples of nitrate-activated NarL functioning as a repressor have been reported, but there little evidence that nitrate-activated NarP can function as a repressor. It was therefore of particular interest that some operons were still repressed by nitrate even in a narXL deletion mutant. A narP deletion was therefore engineered into the ⌬narXL mutant, and RNA was isolated from the resulting double mutant after anaerobic growth in the presence of nitrate. Loss of NarP resulted in lower expression of 14 operons in response to nitrate and increased expression of a further 37 operons (supplemental Table S2). Only five operons were activated more than 3-fold by NarP; these were the previously identified nirB, nap, and nrf operons as well as yeaR and fdnH (Table 5). Operons repressed by NarP in response to nitrate included a subset of about half of the anaerobically induced genes involved in hydrogen and dicarboxylate metabolism. Although the frdA, dmsA, aspA, ansB, and hyb promoters were significantly repressed only by nitrate-activated NarL, the hycA and dcuB promoters were also repressed by nitrate-activated NarP, as predicted in Table 4, part E. Eight of the 37 significantly repressed operons gave repression ratios greater than 4-fold (Table 5, amount of transcript in the ⌬narXL ⌬narP strain relative to that of the ⌬narXL strain). These were the ynfEFGHI, chbF, hycA, yhiM, dcuB, yahL, prpB, and yaiV operons. This is the first report that there is an extensive regulon of 37 NarP-repressed operons,  the most striking examples being operons involved in dicarboxylate or hydrogen metabolism. Near-consensus NarP-binding sites are located close to the transcription start sites of some of these operons. Note, however, that many of these effects were small (2-3-fold repression by nitrate in the narP ϩ strain relative to the narP mutant) and that only four of these operons were repressed by nitrate in the parental strain. This observation might suggest that some of the operons observed to be NarP-repressed are indirectly regulated. Nevertheless, this study has provided the first definitive evidence that NarP can exert global negative control.
Confirmation by Immunoprecipitation of Key Assignments Based on Microarray Data-Attempts to define the FNR regulon revealed several unexpected results that were subsequently checked by chromatin immunoprecipitation (ChIP) experiments. FNR was cross-linked to its DNA-binding sites in cultures of bacteria during exponential growth under different conditions, and the precipitated protein-DNA complex was probed for the presence of target promoters. Both positive and negative controls were included in these assays: note, however, that although enrichment of a promoter in the immunoprecipitate is strong evidence that FNR binds to the targeted promoter, nothing should be concluded from a negative ChIP result, which might arise for many different reasons. More than 2-fold enrichment of the positive control, pdcuB, was obtained in almost every experiment, and the cspA promoter was always enriched, confirming that FNR binds to pcspA. Binding of FNR to the ygbA, cyoA, and ytfE promoters was confirmed, but only negative results were obtained for the yeaR, gltB, and ogt promoters.
Implication of Genes of Unknown Function in Detoxification of Nitrite, NO, or Other Reactive Nitrogen Species-The microarray data revealed four subgroups of genes that are either known to be required for or, on the basis of their regulation, can be implicated in the management of reactive nitrogen compounds (RNS). Selected examples are shown in Table 6, together with previously reported evidence for RNS induction. The genes fall into four subgroups, and transcription of the first two groups is dependent upon FNR activation. The nap and nrf operons encode periplasmic nitrate and nitrite reductases that scavenge for limited concentrations of nitrate or nitrite; expression of these genes is repressed by nitrate, but induced in the presence of nitrite (or very low concentrations of nitrate (29)). The NrfA protein is an active NO reductase with a very high V max and a low K m value for NO (34). Conversely, synthesis of the NADH-dependent nitrite reductase, NirBD, is induced by both nitrate and nitrite; transcription of the nir operon is equally activated by both NarL and by NarP. The hcp promoter is coordinately regulated with nirB, suggesting that the prismane (or hybrid cluster) protein encoded by hcp is involved in RNS management. Despite many previous studies that had assigned a possible hydroxylamine reductase function to the E. coli prismane protein, the physiological role of this fascinating protein remains enigmatic.
Both hmpA and ytfE encode defense mechanisms against stress imposed by NO or other RNS (28,36,37). Transcription from the hmp and ytfE promoters appeared to be repressed by FNR but strongly induced by both nitrate and nitrite. Note, however, that there is no near-consensus FNR-binding site in the ytfE regulatory region, so it is unlisted in Table 3. The similar regulation of the ygbA promoter suggests that this gene of unknown function might also encode a defense mechanism against RNS; indeed, ygbA expression was shown to be induced by NO donors (38). If so, there must be a regulatory mechanism that is responsive to NO or RNS stress. Immediately before this paper was submitted for publication, the NsrR protein encoded by nsrR (formerly yjeB) was identified as an NO-inactivated repressor of hmpA, ytfE, and ygbA (39,40).
The fourth group includes four operons (assuming that yeaR-yoaG and yibI-yibH are in operons), some of which show the greatest responses to the presence of nitrate or nitrite. Although this subgroup includes the ArcA-repressed cyoA operon encoding the main cytochrome oxidase during oxygen-sufficient growth, it also includes genes for a DNA methyltransferase, Ogt, and proteins of unknown function, YeaR-YoaG, YibH, and YibI. Of the genes in this group, only expression of yeaR was shown previously to be induced by NO donors (28). A striking feature of the regulation of these genes is that nitrite induction is most apparent only in the absence of FNR. However, an FNR-binding site was found only in the yibIH promoter (Table 3). This would be most important physiologically under extreme RNS stress, when NO is sufficiently abundant to inactivate FNR (37). We therefore propose that these four subgroups of genes enable E. coli to survive a wide variety of conditions under which RNS stress would otherwise be detrimental to growth or even lethal.

DISCUSSION
In this study we have attempted to map the major biochemical changes mediated by the transcription factor FNR that occur when E. coli K12 MG1655 adapts from aerobic to anaerobic growth. However, no microarray experiment can reveal the entire regulon of a transcription factor because different growth conditions will reveal different regulated genes, and small effects will often be rejected by rigorous statistical analysis. These problems are especially acute for studies of global regulators such as FNR at promoters where FNR regulation is co-dependent upon another factor, for example NarL or NarP. At such promoters, the choice of growth condition is crucial for detecting FNR regulation. By growing anaerobic cultures in the presence of glycerol as the nonrepressing carbon source and both TMAO and fumarate as the terminal electron acceptors, the fnr mutant grew as rapidly as the parental strain. Alternative ways to overcome problems of artifacts associated with changes in growth rate are available, but each of them generates far greater problems than they solve. For example, growth rates can be controlled rigorously by growing bacteria in a continuous culture, but even at reasonably high growth rates, selection of mutants defective in the stationary phase factor, RpoS, is so rapid that the genetic compo- sition of the culture is substantially changed well before a steady state can be achieved (43,44). Any apparent advantages of using a chemostat are therefore illusory. The FNR-binding site at the frdA promoter is a poor match to the consensus sequence, so moderately high levels of frd transcript are synthesized even in an fnr mutant. This explains the anaerobic growth of the fnr mutant even in the absence of TMAO (Fig. 1). As TMAO induces a small regulon mediated by TorS, TorR, and TorT (41,42) and the frd operon is activated by the two-component system, DcuRS (31,32), it was possible that addition of fumarate and TMAO to all of the growth media might have distorted global expression profiles. However, the effects, if any, of FNR, nitrate, nitrite, NarXL, and NarP on torCAD and frdABCD transcription, and also the responses of the large majority of operons known to be subject to FNR control, were entirely consistent with previous data from other studies, so we conclude that this was not a problem.
Extensive use of microarrays in many laboratories has resulted in a consensus view that biological replication in any study is of far greater value than technical replication, and only rarely do reverse transcription-PCR experiments or operon fusion studies yield data that challenge the value of microarray data, a point extensively documented in the previous studies of the FNR regulon (1,2). In this study, not only were highly reproducible data obtained from rigorously performed biological replications, but in addition four additional approaches were used to validate data from the initial microarray experiments with the fnr mutant. First, we used a bioinformatics approach to locate appropriately positioned binding sites in the regulatory regions of every operon that appeared to be regulated by FNR. Second, we demonstrated that increasing the range of growth conditions used to analyze a regulon can be more productive than operon fusion or reverse transcription-PCR experiments that simply confirm what has already been established convincingly in the microarray experiments. Data for anaerobic cultures supplemented with nitrate or nitrite frequently yielded much clearer evidence of regulation than the unsupplemented cultures alone. Our third approach to validate and extend the initial microarray data was to repeat experiments with nitrate-supplemented cultures first with a narXL deletion strain and then with a double mutant that lacks narX narL and narP. Finally, a few of the most interesting results were confirmed by chromatin immunoprecipitation experiments targeted to specific regulatory regions. The combined data, summarized in Fig. 2, have established that at least 103 operons are regulated by FNR, 68 of which are activated by FNR and 35 are repressed. Twenty four of these were also regulated by NarXL, 12 by NarP and 7 by both. Nitrate induced transcription of 42 operons, of which only 5 were FNR-dependent and appear in either Table 2 or Table 3. Transcription of 51 operons was activated by NarXL, 30 of these also being induced by nitrate. Eight of these operons were induced more than 10-fold by nitrate, only four of which were also FNR-dependent. Nitrate-stimulated NarP activated expression of 14 operons; all of those most strongly activated by NarP were part of the FNR regulon and had been characterized independently. Only five operons were induced by nitrate, NarXL and NarP, two of which were also activated by FNR. Nine operons were regulated in different directions by NarXL and NarP (five being NarXL-repressed and NarP activated); three of these were also FNR-dependent.
To assess whether many other FNR-dependent operons might have been missed, it is instructive to consider why 11 of the previously described FNR-dependent operons were not shown to be FNR-dependent in the current study. Two of the five reportedly "FNR activated" operons, cydDC and glpTQ, lack even a remotely credible consensus FNR-binding site, and the data implicating FNR in their regulation is weak. Conversely, although the arcA and adhE promoters include almost consensus FNR sites located appropriately to activate transcription, regulation by FNR is masked by other more significant factors that also regulate transcription positively or negatively. The adhE gene is expressed in the stationary phase from two promoters, but the FNRbinding site is located at the downstream promoter that is inactive during exponential growth, conditions tested in the current experiments (45). Similarly, the hlyE promoter includes a 9 of 10 fit to the core FNR-binding site, but this is also a binding site for CRP, the cAMP receptor protein, and H-HNS (heat-stable nucleoid-structuring protein) interferes with FNR activation. Thus in all five cases, FNR is not the major factor, or even a significant factor, that determines the levels of these transcripts. Similar explanations apply to the six operons that we failed to confirm to be FNR-repressed; no credible FNR sites are located in the hemA or norVW promoters (the latter is an RpoN-dependent promoter), and putative sites in the ubiCA, yeiL, and tpx promoters diverge from the consensus. Most of these operons are also regulated strongly by other factors, including by the FNR-regulated ArcA. The function of the consensus FNR-binding site in the narXL regulatory region is to activate transcription from the divergent narK promoter; it is inappropriately located to affect narXL transcription. On this basis, we predict that relatively few operons that are strongly activated or repressed by FNR were not detected in the current experiments, and the biochemical relevance of any small effects of FNR missed in the current experiments is likely to be minimal.
Secondary effects are generated not only by regulation of the expression of other transcription factors but also by the accumulation or depletion of metabolites caused by differences in expression of enzymes, mediated either directly or indirectly by FNR. Thus, processes occurring many steps downstream from an initial FNR-mediated effect can generate secondary effects. Fig. 3 illustrates one possible scenario by which indirect regulation might occur, in this case why nitrate activates but FNR represses transcription of cyoA; similar regulatory circuits explain why Kang et al. (2) included many ArcA-regulated operons in their lists of genes activated or repressed by FNR. Ubiquinone inhibits the kinase activity of ArcB (26,27) and hence determines the phosphorylation state of ArcA and its ability to regulate gene expression. Loss of ability of FIGURE 2. Overview of metabolic regulation in response to oxygen and nitrate mediated by FNR, NarXL, and NarQP. The number of operons rather than the number of genes that fall into each subclass are listed. A, overlap between the FNR, NarXL, and NarP regulons. B, subgroups of operons activated directly or indirectly by nitrate, NarXL, or NarP. C, subgroups of operons repressed directly or indirectly by nitrate, NarXL, or NarP.
an fnr or narXL mutant to reduce nitrate will result in more reduction of the ubiquinone pool, and hence repression of components of the Arc regulon such as cyoA, so FNR should indirectly activate rather than repress cyoA. This effect is compounded by the fact that arcA transcription is directly activated by FNR (5), which in turn is essential for oxidation of ubiquinol by the cytoplasmic or periplasmic nitrate reductases, NarG or Nap. Even Fig. 3 is an oversimplification because it ignores the effects of other well documented transcription factors, for example, integration host factor and ModE at the narG promoter and GadE at the cyoA promoter. This example underlines the difficulty of identifying genes regulated by a global regulator such as FNR. The ChIP experiments established that FNR binds directly at the cyoA promoter, so the effects of FNR and nitrate were confirmed to be both direct and indirect. A second indirect effect can be observed at the norVW promoter, induced by nitrite in an fnr mutant ( Table 2, part D). The fnr strain cannot reduce nitrite to ammonia because the NirBD and Nrf reductases are FNR-dependent; therefore, norVW expression is induced by nitrite or nitrite-derived radicals.
Some of the most interesting new discoveries from this study are the wider than expected range of genes implicated in the metabolism of RNS. Three previous studies have reported E. coli genes activated by NO or NO donors (28,35,38). The lists of RNS activated genes in each of these studies are different, because of the use of different NO donors and growth conditions. Many genes appear in more than one study; hmpA expression is induced in all three data sets and ytfE in two. However, some genes implicated in RNS response in this study, for example ogt, cyoA, and yibIH, have not been reported previously to be induced by NO donors. These differences support the idea that RNS-responsive gene expression is highly dependent upon growth conditions. The ability of NarXL and NarQP to regulate stress-response genes might reflect the response to generation of endogenous RNS during nitrate and nitrite reduction or a response to conditions when E. coli is likely to encounter exogenous RNS. This is relevant to survival during passage through the gastrointestinal tract, where differences in oxygen, nitrate, and nitrite concentrations in different parts of the gut might act to "program" E. coli to be more resistant to a specific reactive species or stress. We propose that there are four subsets of responses to RNS, each of which is coordinated with particular growth conditions (Table 6). When nitrate is abundant, the prismane protein is synthesized coordinately with the cytoplasmic nitrate and nitrite reductases, suggesting that HCP might be involved in the RNS response. When nitrate is scarce, the periplasmic pathway for nitrate reduction via nitrite to ammonia is functional; NrfA has already been shown to be an effective nitric-oxide reductase under these conditions (34). Transcription of a further group of genes, including yeaR, yoaG, yibH, and yibI encoding proteins of unknown function, is partially FNR-dependent and strongly induced by nitrate but also strongly induced by nitrite in an FNR mutant, suggesting that multiple regulatory circuits regulate the response of these genes to nitrate and nitrite. Finally, there is a small group of genes, like the well characterized hmpA, that are extremely highly expressed when the fnr mutant is grown in the presence of nitrite. This implies that YgbA and YtfE might protect E. coli from damage or repair damage when the RNS stress is so severe that FNR cannot function, as has been established previously for HmpA (37). The fact that some of these potential NO-responsive genes (yeaR and ytfE) have no near-consensus FNR sites in their promoter regions, but are observed to be FNR-repressed in some growth conditions, may reveal a previously unobserved method for the modulation of FNR repression. FNR sites with a poorer match to the consensus sequence are present in these promoter regions, so that FNR repression might be lifted when NO deactivates a smaller percentage of the total FNR pool. A promoter repressed by FNR binding to a near-consensus site, such as hmpA, would become derepressed only when a larger percentage of the FNR pool was deactivated. The roles of all of the above genes in RNS metabolism are currently under investigation.