We have systematically examined the mRNA profiles of 36 two-component deletion mutants, which include all two-component regulatory systems of Escherichia coli, under a single growth condition. DNA microarray results revealed that the mutants belong to one of three groups based on their gene expression profiles in Luria–Bertani broth under aerobic conditions: (i) those with no or little change; (ii) those with significant changes; and (iii) those with drastic changes. Under these conditions, the anaeroresponsive ArcB/ArcA system, the osmoresponsive EnvZ/OmpR system and the response regulator UvrY showed the most drastic changes. Cellular functions such as flagellar synthesis and expression of the RpoS regulon were affected by multiple two-component systems. A high correlation coefficient of expression profile was found between several two-component mutants. Together, these results support the view that a network of functional interactions, such as cross-regulation, exists between different two-component systems. The compiled data are avail-able at our website (http:ecoli.aist-nara.ac.jpxpanalysis 2components).
Two-component systems (TCSs) are required for innumerable adaptive responses in bacteria (Nixon et al., 1986; Hoch and Silhavy, 1995). These systems are widespread and exist not only in nearly all prokaryotes and many Archaea but also in eukaryotes such as plants, fungi and yeasts (West and Stock, 2001). A typical TCS includes a histidine kinase (HK) and a partner response regulator (RR). The HK (also called transmitter protein) contains an invariant histidine residue that is autophosphorylated. This phosphoryl group is accepted by a particular aspartate residue on the RR (also called receiver protein). Accordingly, the HK is autophosphorylated in response to an input signal, and histidine-to-aspartate (His-Asp) phosphotransfer to the RR results in a cellular output response. More complex systems can include a separate histidine containing phosphotransmitter (HPt) protein for phosphotransfer between an HK and RR.
An analysis of the Escherichia coli K-12 genome revealed the presence of 29 HKs, 32 RRs and a sole HPt (Mizuno, 1997). TCSs mediate responses to a variety of environmental signals, including nitrogen, oxygen and Pi limitations and osmolarity. To adapt and survive complex environmental changes in nature, it may be especially important that different TCSs form regulatory networks and show dependencies and regulatory hierarchies. However, little is known about whether functional interactions between different His-Asp phosphotransfer systems form signalling networks in E. coli. To address these issues, mRNA profiles were generated using DNA microarrays on a set of two-component regulatory mutants in which each system is deleted individually (K. A. Datsenko, H. Aiba, L. Zhou, K. Zhang, J. A. Masella, T. Mizuno and B. L. Wanner, unpublished data). Experiments were done under a single growth condition to allow direct comparison of all genes over all profiles.
Results and discussion
Construction and preliminary characterization of TCS mutants
Escherichia coli encodes 23 orthodox HK proteins, one unorthodox HK protein (CheA), 32 RR proteins, five hybrid HK proteins (containing both HK and RR domains) and one HPt protein ( Mizuno, 1997 ). In many cases, pairs of HK and RR proteins are encoded by adjacent or nearby genes that are often arranged in operons. Several (including ArcA, ArcB, BarA, FimZ, NarQ, NarP, RssB and UvrY) are encoded by genes separated on the chromosome. In order to delete genes for all TCS proteins, it was necessary to construct a series of 36 different mutants. In many mutants, the partner HK and RR genes were deleted simultaneously. In others, the HK and RR genes were deleted individually. All mutants grew on glucose minimal (M9) medium agar plates at temperatures ranging from 16°C to 42°C (data not shown). Before preparing RNA, we analysed cell growth in LB medium at 37°C and found that cell growth was impaired in the ArcB/ArcA (HK/RR) system mutants (doubling times of 49 and 44 min for the Δ arcA and Δ arcB mutants respectively) and the UvrY mutant (doubling time 45 min) compared with wild-type cell (doubling time 27 min) ( Table 1 ).
Table 1. . Summary of microarray data and strains used in this study.
Deleted two component system (HK/RR)
No. of up- regulated genes
No. of down- regulated genes
Total no. of altered genes
Doubling time (min)
Identified function in this experiment
. In BW28079, the large chromosome region was deleted from position 8380 base of AE000281 to position 2480 base of AE000283. The mutation was denoted as
TCA cycle, chemotaxis, osmotic adaptation, protein peptide secretion, purine ribonucleotide biosynthesis, rpoS regulon, enterochelin, maltose transport
DNA microarray analysis of all TCS mutants
For the systematic analysis, the transcript profiles of 36 TCS deletion mutants were generated by means of a two-colour cDNA microarray analysis (Schena et al., 1995). In seven mutants (ΔcreABCD, ΔcusRS, ΔevgAS, ΔnarXL, ΔphoBR, ΔuhpAB and ΔygiXY), significant alterations in gene expression (twofold alteration) were found for less than 10 genes (Table 1). Because many of these systems operate under specific growth conditions (Kadner, 1995; Stewart and Rabin, 1995; Wanner, 1995; Kato et al., 2000; Munson et al., 2000), these systems were probably not operating strongly under the growth condition we used. Six mutants (ΔkdpABCDE, ΔnarP, ΔntrBC, ΔyedWV, ΔyfhK, ΔyehUT) displayed significant alterations for 10–20 genes. In this class, KdpD/KdpE, NarP and NtrB/NtrC are known to be involved in response to turgor pressure, nitrate and nitrogen availability respectively (Laimins et al., 1981; Darwin and Stewart, 1995; Zimmer et al., 2000). Among genes that have been known to be under the control of NtrB/NtrC, only the glnP gene encoding glutamine transport protein was downregulated in the ΔntrBC mutant. Other NtrC-regulated genes were not significantly altered, as their expression was also turned off in the wild type (Porter et al., 1995). The ΔkdpABCDE and ΔnarP mutants also had small effects under these conditions. The results are consistent with the idea that these systems were not operating under the growth condition we used. Therefore, we classified those mutants showing alteration in less than 20 genes as a class with no or little change (Table 1).
Nineteen mutants displayed significant alterations for more than 21 but less than 100 genes (Table 1). Clearly, these TCSs are likely to have a physiological function under these conditions. Downregulation of mgtA, the gene encoding a magnesium transporter, was observed in the ΔphoPQ mutant. This gene has been shown previously to be regulated by the PhoQ/PhoP system in a Mg2+-dependent manner (Kato et al., 1999). Thus, our array approach identified expected genes already known to be part of the regulon, lending credence to our methodology. Our analysis suggests that the PhoQ/PhoP system directly or indirectly controls the expression of over 20 genes. Downregulation of ppiA and cpxP, encoding peptidyl-prolyl cis-trans isomerase A and a periplasmic protein involving in feedback inhibition, respectively, was also observed in the ΔcpxRA mutant. These genes have been known to be under the control of the CpxA/CpxR system, which senses aggregated or unfolded proteins in the E. coli envelope (Danese and Silhavy, 1997; 1998; Pogliano et al., 1997; Raivio et al., 1999). Our data also showed many σs-regulated genes to be upregulated in the ΔrssB mutant. RssB (also called SprE) is an orphan response regulator. As RssB is indispensable for σs turnover, these data support previous findings that σs accumulates in the ΔrssB mutant, resulting in elevated expression of the RpoS regulon (Muffler et al., 1996; Pratt and Silhavy, 1996). Accordingly, the results for the ΔphoPQ, ΔcpxRA and ΔrssB mutants provide further support for the accuracy of our microarray analyses.
Interestingly, we found that the expression of several genes was affected by the deletion of more than one TCS as follows. Upregulation of the ent operon (encoding enzymes for enterochelin biosynthesis) was seen in the ΔcpxRA, ΔfimZ and ΔrstAB mutants (Table 1; a complete list of upregulated genes can be found on our web page; http:ecoli.aist-nara.ac.jpxpanalysis2componentsbenn. cgi). Enterochelin (also called enterobactin) is the catecholate siderophore and participates in iron uptake in E. coli (Earhart, 1996). Upregulation of flagellar genes (flgC, flgG and flgI) was seen in the ΔcitAB, ΔrcsB and ΔypdAB mutants, and downregulation of these genes was seen in the ΔatoSC mutant. Upregulation of maltose transport genes (malE and lamB) was also detected in the ΔcitAB and ΔrcsB mutants. These results suggest that many TCSs have not only unique physiological roles (PhoQ/PhoP for mgtA expression; CpxA/CpxR for ppiA and cpxP expression; and RssB for rpoS regulon expression), but also common physiological roles (CpxA/CpxR, FimZ and RstA/RstB for enterochelin biosynthesis; AtoS/AtoC, CitA/CitB, RcsB and YpdA/YpdB for flagellar gene expression; and CitA/CitB and RcsB for maltose transport gene expression). These data imply that several different sets of genes are controlled by different networks of TCSs. We classified these mutants that displayed alterations in 21–100 genes as ones with significant change (Table 1).
Four mutants that showed severe growth impairment displayed significant alterations for more than 100 genes (ΔarcA, ΔarcB, ΔompR–envZ, ΔuvrY). The ArcB/ArcA system is known to be involved in anaerobic respiratory control (Iuchi and Lin, 1995). The EnvZ/OmpR system is known to control the expression of outer membrane porins in response to medium osmolarity (Aiba and Mizuno, 1990; Pratt and Silhavy, 1995). In this class, downregulation of ompF and ompC was seen in the ΔompR–envZ mutant, and upregulation of TCA cycle genes was seen in both ΔarcA and ΔarcB mutants, in agreement with previous data (Aiba and Mizuno, 1990; Iuchi and Lin, 1995; Pratt and Silhavy, 1995). However, in this analysis, we did not detect regulation of some genes, such as bolA and malT, that were known to be under the control of OmpR–EnvZ. This may be because the effects of OmpR–EnvZ on these genes were detected under specific growth conditions (Case et al., 1986; Yamamoto et al., 2000) that differ from those we used. Cellular functions that were affected in these TCS mutants included genes for energy metabolism in ΔarcA, ΔarcB and ΔuvrY mutants, genes for cysteine and isoleucine biosynthesis in the ΔompR–envZ mutant, genes for enterochelin biosynthesis in the ΔarcB and ΔompR–envZ mutants, genes for maltose transport in the ΔarcB and ΔuvrY mutant, and genes for flagellar synthesis in the ΔarcA and ΔompR–envZ mutants (Table 1). Many of these functions (TCA cycle, amino acid metabolism, iron uptake, carbon source transport) are important for cell growth and are needed directly or indirectly for efficient energy metabolism. Improper regulation of such functions is probably responsible for the growth defects of these mutants. We have not determined whether the changes observed are direct or indirect consequences. For example, in the case of the ΔompR–envZ mutant, one can speculate that the transport of many compounds is affected by the loss of major outer membrane porin proteins. Many of the genes affected by the ΔompR–envZ mutant may therefore be indirect. Anyway, we classified these mutants showing changes in more than 100 genes as ones with drastic change. Tables summarizing these microarray data are given in the Supplementary material. Complete array data are available at our web page (http:ecoli.aist-nara.ac.jpxpanalysis2components).
Evidence of functional interaction between various TCSs
The term cross-regulation has been used to refer to the control found between two or more TCSs (Wanner, 1992). For example, the HK PhoR senses environmental Pi (directly or indirectly), is autophosphorylated and transfers the phosphoryl group to the RR PhoB for activation of the Pho regulon (Wanner, 1996). Under other conditions, the HK CreC detects a signal (as yet unknown), is autophosphorylated and transfers the phosphoryl group to both its partner RR CreB and PhoB. Although CreC has been shown previously to phosphorylate PhoB only in the absence of PhoR, overproduction of CreC can result in phosphorylation of PhoB in both the presence and the absence of PhoR (J. A. Masella and B. L. Wanner, unpublished data).
Additional evidence for cross-regulation has recently been found between the HK ArcB and the EnvZ/OmpR system (Matsubara et al., 2000). Expression of ompC and ompF (encoding the OmpC and OmpF porins) is mainly regulated by the osmoresponsive EnvZ/OmpR system. Under anaerobic growth conditions, however, the anaeroresponsive ArcB also participates in porin gene regulation in a manner requiring OmpR. That is, under anaerobic growth conditions, both ArcB and EnvZ appear to be involved in phosphotransfer to OmpR. Other apparent cases of cross-regulation appear to involve the central metabolic intermediate acetyl phosphate. Whereas acetyl phosphate may, under certain conditions, act by phosphorylating PhoB directly (that is in the absence of an apparent or known HK; Wanner and Wilmes-Riesenberg, 1992), the apparent phosphorylation of PhoB by EnvZ requires acetyl phosphate (Kim et al., 1996). With respect to the orphan RR RssB (which controls σs turnover), σs is stabilized in a cell unable to synthesize acetyl phosphate (Δpta-ackA mutant), suggesting a role for acetyl phosphate in the phosphorylation of RssB (Bouche et al., 1998; Cunning and Elliott, 1999; Zhou et al., 2001).
To test for evidence of functional interaction, such as cross-regulation, we first calculated a Pearson correlation coefficient between all pairs of TCS mutants using the microarray data for their expression profiles. If cross-regulation operates between different TCSs, similar gene expression profiles may be expected between such TCS mutants. This would result in a high correlation coefficient value. Indeed, a high value (0.360) was calculated between ΔarcB and ΔarcA mutants, which remove the HK and RR of the anaeroresponsive ArcB/ArcA system. We found a high correlation coefficient for the gene expression profiles among ΔrcsB, ΔuvrY, ΔcitAB and ΔypdAB mutants. The same was also found among the ΔarcB, ΔuvrY and ΔrssB mutants and between the ΔfimZ and ΔrstAB mutants. These results are consistent with cross-regulation occurring among these TCSs with involvements in the same cellular functions (Table 2). Alternatively, high correlation coefficients can also mean that the TCSs have common downstream targets. In addition, a high correlation coefficient was calculated between several other TCS mutants. For example, a high correlation coefficient (0.353) was detected between the PhoQ/PhoP and EvgS/EvgA systems, suggesting possible cross-regulation. The compiled data are given in the Supplementary ma-terial and at our web site (http:ecoli.aist-nara.ac.jpxpanalysis2componentsoverlaps.html).
Table 2. . Cellular functions affected by multiple two component systems.
Commonly affected genes
Flagellar synthesis (up)
ompR-envZ, rcsB, uvrY, citAB, ypdAB
Flagellar synthesis (down)
flgA, flgC, flgE, flgG fliZ
RpoS regulon (up)
arcB, uvrY, rssB
poxB, otsA otsB
fimZ, rstAB, cpxRA, arcB, yfhA, ompR-envZ
Maltose transport (up)
citAB, rcsB, uvrY, arcB
In Table 2, functions affected by multiple TCS mutants are summarized. Genes belonging to the RpoS (σs) re-gulon were upregulated in ΔarcB, ΔuvrY and ΔrssB mutants. Genes involved in flagellar synthesis were up- or downregulated in the ΔompR–envZ, ΔrcsB, ΔuvrY, ΔcitAB and ΔypdAB mutants or the ΔarcA and ΔatoSC mutants respectively. Genes involved in enterochelin biosynthesis were upregulated in the ΔfimZ, ΔrstAB, ΔcpxRA, ΔarcB, ΔyfhA and ΔompR–envZ mutants. Genes for maltose transport were upregulated in the ΔcitAB, ΔrcsB, ΔuvrY and ΔarcB mutants. Although many of these changes can result from indirect effects, one interpretation of these results is that functional interactions, such as cross-regulation, exist between these TCSs to control the cellular functions. The biological significance of why these genes are regulated by respective TCSs will require future investigation.
Interestingly, we also found that expression of atoSC (for the AtoS/AtoC TCS) was downregulated in the ΔompR–envZ mutant (Fig. 1A). Such ‘cascade regulation’ appears to be a common feature among other TCSs. We also observed downregulation of torS (encodes an HK) in the ΔbasRS and ΔntrBC mutants, downregulation of evgS (encodes an HK) in the ΔdcuSR mutant, downregulation of the rstAB (encodes the RstB/RstA TCS) in the ΔphoPQ mutant and upregulation of atoS (encodes an HK) in the ΔyfhA (RR) mutant. Although the physiological importance of these effects is unknown, they strongly suggest that regulatory cascades are also important for adaptation of E. coli to complex environmental signals.
Additional evidence supporting the mRNA profiling results
Our microarray data indicate that several TCSs may directly or indirectly control the expression of genes belonging to particular groups (Table 2). The genes affected can be searched at our web site (http:ecoli. aist-nara.ac.jpxpanalysis2componentsbenn.cgi), and their representatives are also shown in Table 2. This provides evidence in favour of cross-regulation (Wanner, 1992) between TCSs having an important role in cell physiology. We therefore tested whether individual TCS mutations resulted in phenotypes that are predictable on the basis of the mRNA profiling results.
Because upregulation of RpoS regulon genes is probably caused by an increase in σs levels, we measured the effect of the TCS mutations on expression of the rpoS gene. For this, we analysed both the expression of rpoS–lacZ protein fusion and the amount of σs protein (Fig. 2). We have shown previously that expression of the rpoS–lacZ protein fusion is a reliable indicator of rpoS expression (Ueguchi et al., 2001). As shown in Fig. 2A and B, a high level of rpoS–lacZ expression was seen during the logarithmic growth phase in the ΔarcB, ΔuvrY and ΔrssB mutants. At the logarithmic growth phase, a high level of σs protein was also detected in these mutants, in agreement with the microarray data (Fig. 2C). Accordingly, upregulation of the RpoS regulon in ΔarcB, ΔuvrY and ΔrssB mutants results from upregulation of σs synthesis. RssB is known to be involved in controlling σs stability (Muffler et al., 1996; Pratt and Silhavy, 1996). However, these results provide the first evidence for an involvement of ArcB and UvrY in controlling σs regulation. Possibly, ArcB is an HK for UvrY and/or RssB, as well as for ArcA (Fig. 1B). Or, ArcB and UvrY affect σs regulation independently. The latter possibility is supported by the finding that phospho-ArcB is unable to donate its phosphoryl group to UvrY (Pernestig et al., 2001). Alternatively, a non-cognate HK or acetyl phosphate may be involved in this regulation. It is also possible that, in spite of the in vitro results, ArcB, UvrY and/or RssB act together under our in vivo conditions. Further experimentation is required to test these or other possibilities.
Our microarray data also revealed major effects on the expression of flagellar genes. We therefore tested the effects of TSC mutations on motility. As shown in Fig. 3, ΔcitAB, ΔompR–envZ and ΔuvrY mutants formed large rings (that is are more motile) on swimming agar than the wild type. In contrast, the ΔarcA and ΔatoSC mutants, like the chemotaxis-defective ΔcheABYZ mutant, were non-motile or showed reduced motility. These results are in agreement with expectation based on the motility phenotype and the microarray data. These results also suggest that multiple TCSs affect cell motility. In the motility assays, we also observed differences between the ΔarcA and ΔarcB mutants. Even though ArcA and ArcB belong to the same TCS, the motility of the ΔarcA mutant, but not of the ΔarcB mutant, was drastically reduced in comparison with wild type (Fig. 3). This difference between the ΔarcA and ΔarcB mutants suggests that ArcA may also be subject to cross-regulation by an unknown HK. Namely, ArcA may be phosphorylated by non-cognate HK or acetyl phosphate under these conditions (Fig. 1C). This hypothetical HK may ordinarily act redundantly with ArcB.
Our microarray analyses also indicated that the expression of the ent operon was upregulated in ΔfimZ, ΔrstAB, ΔcpxRA, ΔarcB, ΔyfhA and ΔompR–envZ mutants. The genes involved in maltose transport were also indicated as upregulated in ΔcitAB, ΔrcsB, ΔuvrY and ΔarcB mutants (Table 2; the list of upregulated genes can be found at http:ecoli.aist-nara.ac.jpxpanalysis2componentsbenn.cgi). These microarray results were confirmed directly by means of Northern hybridization analyses. For analysing the expression of the ent operon, entB and entE genes were used as representative probes. However, we did not detect signals for entB and entE transcripts in ΔfimZ, ΔrstAB, ΔcpxRA, ΔarcB, ΔyfhA and ΔompR–envZ mutants as well as in wild-type cells (data not shown). This suggests that the expression of the ent operon in these mutants is too low to be detected by Northern hybridization. Both malE and lamB genes were used as probes for analysing maltose transport genes. As shown in Fig. 4, the expression of both malE (Fig. 4A) and lamB (Fig. 4B) genes in ΔcitAB, ΔrcsB, ΔuvrY and ΔarcB mutants was high compared with wild-type cells. These results are consistent with the results of our microarray analyses and support the accuracy of our microarray data.
We chose the single growth condition (LB broth under aerobic conditions) for DNA microarray analysis and analysed the mRNA expression profiles of 36 TCS mutants to identify the physiological functions and to test for cross-regulation among TCSs. We showed that the cellular pathways affected can be determined by pattern matching, even for mutants displaying subtle profiling differences. A similar analysis of the yeast transcriptome was performed and revealed that a single growth protocol was sufficient to generate functional data for roughly half the mutants (Hughes et al., 2000). Based on our results, the same is likely to be true for E. coli.
Over half the TCS mutants showed prominent alterations in a small number of genes. The most drastic changes were detected in mutants of the anaeroresponsive ArcB/ArcA system, the osmoresponsive EnvZ/OmpR system and the response regulator UvrY. We also uncovered previously unknown effects of several TCSs on the regulation of σs synthesis, flagellar synthesis (motility), respiration, maltose transport and ion uptake. We found that multiple TCSs are involved in the regulation of cellular functions such as RpoS regulon, flagellar synthesis, enterochelin synthesis and maltose transport. High correlation coefficients were also detected between several TCS mutants. We also found evidence for the control of several TCS genes by other TCSs (cascade regulation). Together, these results provide evidence that cross-regulation and cascade regulation have important roles in cell physiology for co-operative functioning between multiple TCSs. It must be pointed out that the observed phenomena are not always the direct consequence of deletion of each TCS. Indeed, many can be very indirect, reflecting complex changes in cellular physiology. However, further studies to identify the nature of the suggested functional interactions between TCSs should shed light on the TCS network(s) in E. coli.
The bacterial strains used for microarray analyses are summarized in Table 1. All strains are derivatives of E. coli K-12 strain BW25113 (Haldimann and Wanner, 2001). Mutants were constructed using polymerase chain reaction (PCR) products and a standard one-step gene inactivation protocol for disruption of chromosomal genes that is based on the high efficiency of the phage λ Red recombinase (Datsenko and Wanner, 2000). These mutants will be described elsewhere (K. A. Datsenko, H. Aiba, L. Zhou, K. Zhang, J. A. Masella, T. Mizuno and B. L. Wanner, manuscript in preparation). Strain CU263 (Ueguchi et al., 2001) carrying an rpoS–lacZ protein fusion in single copy on the chromosome was used to test the effect of each TCS deletion on rpoS expression.
Cell growth and RNA isolation
Cells were grown aerobically in Luria–Bertani (LB) medium at 37°C by reciprocal shaking (120 r.p.m., stroke 5 cm) in 500 ml Sakaguchi flasks (Iwaki) containing 100 ml of me-dium. Cells were harvested during the logarithmic growth phase (OD600 = 0.4). Total RNA was isolated using the RNeasyR maxi kit (Qiagen) as recommended by the supplier.
Preparation of E. coli DNA microarrays
Custom glass slide microarrays (Takara Shuzo) were spotted with 4095 PCR products corresponding to full-length E. coli open reading frames (ORFs) and the human β-actin gene as a negative control. PCR products were generated using as template a clone bank containing E. coli genes in the Archive vector (Mori et al., 2000). Other details were the same as described previously (Oshima et al., 2002).
Preparation of labelled cDNA, hybridization and data capture
The RNA from wild-type cells and cells from each mutant were labelled with Cy-3 and Cy-5 respectively. Each preparation was then tested twice by microarrray analysis. Thus, two values were obtained for each gene (spot). cDNA labelling and microarray hybridization were performed basically in accordance with the M guide (http:cmgm.stanford.edupbrownmguideindex.html;DeRisi et al., 1997). Fluorescent-labelled cDNA probes were prepared by random priming. Reverse transcriptase reactions were performed in 40 µl of 1× reaction buffer (XL, Life Science) containing 30 µg of total RNA, 5.3 nmol of random hexamer primers, 0.5 mM each dATP, dCTP and dGTP, 0.2 mM dTTP, 45 units of reverse transcriptase (XL, Life Science) and 4 nmol of either Cy3-dUTP or Cy5-dUTP (Amersham Pharmacia). Reaction mixtures lacking reverse transcriptase were incubated at 65°C for 5 min and then cooled to room temperature for 1 min. After the addition of reverse transcriptase, the reaction proceeded for 120 min at 42°C. Labelled cDNA probes were then purified by Centri-sep (Princeton Separations), phenol–chloroform extraction and ethanol precipitation. After drying, the cDNA probe was resolved in 9 µl of water. Both Cy3- and Cy5-labelled cDNA probes were then added to a final volume of 23 µl of hybridization buffer (final concentration; 4× SSC, 0.2% SDS, 5× Denhardt's solution, 100 ng ml−1 salmon sperm DNA) and denatured at 98°C for 2 min. The denatured cDNA probe was applied to the microarray prehybridized by 100 ng ml−1 salmon sperm DNA under a coverslip. The hybridization was carried out at 65°C for 16 h. The slides were washed at 60°C with 2× SSC for 5 min, then at 60°C with 0.2× SSC containing 0.1% SDS and, finally, at room temperature with 0.2× SSC. The slides were scanned for fluorescent intensity using a GMS 418 array scanner (Genetic Microsystems) and recorded as 16-bit image files. The signal density of each spot in the microarray was quantified using imagene software (BioDiscovery).
The data analysis of microarrays was carried out as described previously (Oshima et al., 2002). By subtracting the local background, we first corrected the intensity of each spot. In addition, a mean value of the intensity of the 24 negative control spots (human β-actin gene) was determined, together with the standard deviation (SD). Then, spots were classified into three groups. In group 1, both Cy3 and Cy5 signal intensities were higher than the mean +1 SD of the negative control. In group 2, either Cy3 or Cy5 signal intensity was higher than the mean +1 SD of the negative control. In group 3, both Cy3 and Cy5 signals were lower than the mean +1 SD of the negative control. Initially, all spots classified in group 1 were normalized by defining the mean of the ratios (Cy5/Cy3) of all spots as 1.0. The ratios of group 2 spots could not be determined by this method because of lack of either a Cy3 or a Cy5 fluorescent intensity. Then, spots in this group with high Cy3 or Cy5 intensity values (over 1000, i.e. of a sufficiently high intensity value to be detected precisely by the GMS 418 array scanner) were selected. By this method, spots in group 3 were ignored as undetectable. Genes with significantly different expression were then selected by the following criteria, provided that similar values were seen in two hybridizations. From group 1, we selected genes that showed relative Cy5/Cy3 ratios <0.5 or>2.0 in both hybridizations as being down- or upregulated respectively. For genes classified into groups 1 and 2 in two hybridizations, we selected genes that showed relative Cy5/Cy3 ratios <0.5 or>2.0 in one hybridization and high intensity values (>1000) of Cy3 or Cy5 in another hybridization as showing down- or upregulation respectively. Genes classified into group 2 in both hybridizations were recognized as those showing altered regulation. Finally, we investigated the random fluctuation and systematic bias of our analysis system. Control experiments were conducted in which two independent reference strain cultures of BW25113 were compared with each other, with one labelled with Cy5 and the other with Cy3. Reproducible twofold differences were seen for five genes, rfaJ, rfaS, rfaZ, yjgL (repression) and ykgJ (induction). We judged these alterations to be artificial errors and systematic biases. Importantly, such systematic biases were very low. Thus, genes with a reproducibly twofold alteration in a mutant were considered to show significant changes and were used for analyses. The Pearson correlation coefficient was calculated between every TCS mutant by assessing the total gene expression profiles (Oshima et al., 2002).
Northern hybridization analysis
Total RNA (20 µg) was separated using 1.0% agarose gel electrophoresis with formamide and transferred to a Hybond-N+ membrane (Amersham Biosciences). All DNA fragments used for this analysis were amplified by PCR using ORF-specific primers (see http:ecoli.aist-nara.ac.jpPRIMERindex.html). The DNA probes were labelled with 32P-dCTP by a Megaprime DNA labelling system (Amersham Biosciences). Hybridization was carried out at 65°C for 12 h in Rapid-hyb buffer, as recommended by the supplier (Amersham Biosciences). The data were visualized with a Fuji bioimaging analyser (BAS-2500, Fuji Film).
We are grateful to R. Utsumi (Kinki University), K. Kobayashi (Nara Institute of Science and Technology) and Y. Fujita (Fukuyama University) for helpful discussions. This work was supported by Grant-in-Aid for Scientific Research on Priority Areas (C) ‘Genome Biology’ (to H.A.) and ‘Genome Science’ (to H.M.) from the Ministry of Education, Culture, Sports, Science and Technology of Japan, CREST of JST (Japan Science and Technology) and the US National Science Foundation (to B.L.W.).