Exploring the antimicrobial action of a carbon monoxide-releasing compound through whole-genome transcription profiling of Escherichia coli

Carbon monoxide (CO) formed in nature results from the incomplete oxidation of organic matter and as the by-product of the activity of the haem oxygenase enzyme, which oxidizes haem to biliverdin (Li et al., 2007). In the latter process, which can occur in bacteria, plants and animals, the amount of CO endogenously produced by the various cell types is low. In humans, the CO formed has several physiological functions in the neuronal and cardiovascular systems (Wu & Wang, 2005). However, high concentrations of CO have a toxic effect in mammals due to the binding of CO to haemoglobin; the consequent impairment of oxygen storage and transport by haemoglobin causes hypoxia (Wu & Wang, 2005). CO binds preferentially to transition-metal-containing proteins, including haem-based proteins/sensors, giving rise to structural modifications and alteration of their biological function which may also account for the toxic effects of CO. However, it has been shown that living cells can tolerate CO at the concentration of 0.01 % (100 p.p.m.) for several hours (Otterbein & Choi, 2000), and rodents could be exposed continuously for up to 2 years to 500 p.p.m. CO without deleterious effects (Otterbein & Choi, 2000; Stupfel & Bouley, 1970; Verma et al., 1993).

Following the discovery that carbon monoxide is physiologically important, molecules acting as CO-carriers, namely CO-releasing molecules (CORMs), were developed. In order to mimic the role of CO derived from haem oxygenase (Ryter et al., 2006), these molecules have to be able to liberate CO in a biological environment. This liberation is triggered by some physical, chemical or enzymic stimulus. For many CORMs that carry molecular CO as a ligand to a central transition metal, the lack of a pharmacological effect of a similar compound devoid or depleted of CO has been taken as a positive proof for the biological action of CO (Dulak et al., 2002). Following this principle, the beneficial therapeutic action of several CORMs in a wide range of vasoactive and anti-inflammatory situations and models has been attributed to CO (Motterlini et al., 2005).

Recently, we reported that four transition-metal-based CORMs at micromolar concentrations have antibiotic activity, reducing the viability of bacteria such as Escherichia coli and Staphylococcus aureus, under both aerobic and anoxic conditions (Nobre et al., 2007). In that work we also showed that the bactericidal action of those CORMs was only effective when CO was taken up by the cells. In fact, we demonstrated that one of the CORMs led to extensive accumulation of the transition metal molybdenum inside bacterial cells. In all cases, the addition of haemoglobin to the medium, before addition of the CORM, prevented any bactericidal action. Separate experiments (Motterlini et al., 2002; J. D. Seixas, personal communication) showed that these CORMs transfer CO to haemoglobin at a very fast rate. Taken together these results strongly suggest that the bactericidal action of a given CORM depends on its ability to enter bacterial cells and enable CO to reach specific targets.

We have now extended these studies to the analysis of the global transcriptome of E. coli treated with CORM-2, which reflects the organism's immediate response to the environment. Additionally, we have examined the phenotype of 12 mutant strains to test the role of each deleted gene in the E. coli CORM-2 response. Regulatory genes and genes involved in methionine metabolism and biofilm formation were found to play an important role when the micro-organism is exposed to the bactericidal action of CORM-2.

Bacterial strains and growth conditions.
Escherichia coli MG1655 was grown, at 37 °C and 150 r.p.m., in minimal medium salts (MS) (Nobre et al., 2007) in the presence or absence of oxygen. Aerobic experiments were performed in flasks filled with 1/5 of their volume and anaerobic experiments in rubber-sealed flasks that, once filled with medium and closed, were extensively flushed with nitrogen gas. Cultures were grown to an OD₆₀₀ of 0.3, and at this point cells were left untreated or treated with CORM-2 (Sigma).

RNA isolation.
The hot-phenol method was used to extract total RNA from two independent cultures of cells grown either aerobically or anaerobically and treated with 250 µM CORM-2 for 15 min. The concentration of the drug added to the cultures is equivalent to 0.5 times the minimal inhibitory concentration of 500 µM which was previously determined (Nobre et al., 2007). After the isolation procedure, RNA was incubated with DNase (Ambion), and once the absence of any residual DNA was confirmed, concentration and purity were determined in a Nanodrop ND-1000 UV–visible spectrophotometer. RNA integrity was checked in an Agilent 2100 Bioanalyser coupled to an RNA Nano-Assay (Agilent Technologies).

Microarray analysis.
Total RNA (10 µg) was processed for use in the Affymetrix GeneChip E. coli Genome 2.0 Arrays, according to the manufacturer's instructions. Arrays were scanned in an Affymetrix GeneChip scanner 2500 and analysed first with Affymetrix MAS 5.0 software to obtain Absent/Present calls and to ensure that all quality parameters were within the recommended range. Hybridization, scanning and detection procedures were done at the Genomics Unit of the Instituto Gulbenkian de Ciência (Portugal). The advantage of using the Affymetrix GeneChip is that it contains 20 probe pairs to detect each ORF, providing multiple independent measurements for each transcript. The subsequent analysis was performed with the dChip software program (Li & Wong, 2001). The invariant set method (Li & Hung Wong, 2001) was used to normalize arrays at the probe cell level for comparison purposes, and the model-base (Perfect Match-only model) was used to probe selection and computing expression values. The expression profiles of 4306 genes encoding proteins of E. coli K-12 MG155 were analysed (intergenic regions were not included) and genes were considered to have significant differential expression when they showed a transcriptional fold change >2 (with a 90 % confidence interval) and a P-value <0.05 for a paired Student's t-test. To minimize occurrence of false positives the comparison criterion was carefully chosen to ensure that the empirical false discovery rate (FDR) was low (random permuting of the samples 200 times led to a median of FDR of 0 % and 1.1 % for the arrays acquired for cells grown anaerobically and aerobically, respectively).

Gene ontology analysis.
To investigate the biological relevance of the differential gene expression profile, the microarray data were analysed with the FatiGO algorithm (Al-Shahrour et al., 2004, 2006). Briefly, FatiGO considers two lists of genes (in this case the genes up- or downregulated by CORM-2 anaerobically or aerobically were compared with the unchanged genes) and converts them into two lists of gene ontology (GO) terms. Subsequently, FatiGO determines separately the percentage of the relative frequency of genes in each group that is annotated with a particular GO term (biological process or molecular function), and finally, a Fisher's exact test is used to determine if the GO term is significantly over- or under-represented in one of the groups with respect to the other one. Hence, it retrieves information about the predominant biological process or molecular function for the genes that are differently expressed in a certain experimental condition. In this study, the GO terms were considered differently represented when they exhibited a FDR-adjusted P-value <0.2. FatiGO is implemented in a web tool at: and within the Babelomics environment ().

Quantitative real-time RT-PCR.
DNA microarray data were validated by quantitative real-time RT-PCR. To this end, 2.5 µg E. coli total RNA, derived from two independent samples grown aerobically or anaerobically and treated with 250 µM CORM-2 or with 500 µM Ru(II)Cl₂(DMSO)₄, was used to synthesize cDNA, using the first-strand synthesis protocol of the Universal RiboClone cDNA Synthesis System (Promega). Each cDNA synthesis reaction was performed using 40 µg ml^–1 of random primers, 40 U RNasin RNase inhibitor, 4 mM sodium pyrophosphate and 30 U AMV reverse transcriptase. After amplification, cDNA was purified by phenol extraction and ethanol precipitation. Real-time PCRs were performed in a LightCycler Instrument using the LightCycler FastStart DNA Master SYBER Green I kit according to the manufacturer's instructions (Roche Applied Science). For each target gene, specific oligonucleotides were designed to amplify nucleotide fragments of 200–580 bp (see Supplementary Table S4, available with the online version of this paper, for oligonucleotide sequences). The amplification reactions were carried out with equal amounts of cDNA (120 ng), as initial template, and each reaction contained 0.5 µM primers, 2 mM MgCl₂, and hot-start PCR mix (Roche Applied Science). The expression ratio of the target gene was determined relative to a reference gene, the E. coli glyceraldehyde-3-phosphate dehydrogenase gene (gapA), which did not show variation in transcription abundance under any of the conditions tested by microarray analysis or real-time RT-PCR experiments.

Viability assays and complementation studies.
In order to compare the sensitivity to CORM-2 of the E. coli parental strain versus the various mutant strains, cell viability was determined by evaluation of the number of colony-forming units (c.f.u. ml^–1), after 7 h growth in the presence or in the absence of CORM-2. The percentage survival was calculated as the number of colonies obtained from CORM-2-treated cultures divided by the number of colonies formed upon plating untreated cultures. The experiments were performed in duplicate, with a minimum of two independent cultures of the mutant strain and of its corresponding parental strain (see Supplementary Table S5 for details of bacterial strains); the results are presented as means±SE. For complementation purposes, ibpAB and metR genes were amplified from E. coli genomic DNA using the appropriate oligonucleotides (Supplementary Table S4), and cloned into pUC19 to generate plasmids pibpAB and pmetR (see Supplementary Table S6 for details of plasmids). These plasmids or the empty vector were transformed into the correspondent mutant strain and the phenotype was analysed by plating on agar successive dilutions of the cultures treated with CORM-2 or left untreated. For ΔsoxS, ΔoxyR, ΔtsqA and ΔbhsA, the complementation experiments were performed using plasmids pWB31, pAQ17, pCA24NydgG⁺ and pCA24NycfR⁺, respectively (Supplementary Table S6).

Biofilm assays.
Biofilm formation was quantified by the crystal violet method (Ren et al., 2005; Zhang et al., 2007). Briefly, overnight LB-grown cultures of wild-type E. coli, ΔtqsA and ΔbhsA mutant strains were used to inoculate fresh medium to an OD₆₀₀ of 0.05 and then incubated for 6 h, at 37 °C. At this stage, 250 µM CORM-2 was added and the plates were incubated for another 18 h, at 37 °C. The total biofilm formed was stained with a solution of 0.1 % crystal violet for 30 min and quantified by measuring the A₅₄₀. Relative biofilm formation was calculated as the total biofilm formed by the cultures treated with CORM-2 divided by the total biofilm obtained in untreated cultures; each data point represents the mean of six replicated wells from four independent cultures.

Analysis of the E. coli transcriptome of cells treated with CORM-2
Microarray technology was utilized to analyse the genome-wide transcriptional pattern of E. coli cells, grown aerobically and anaerobically, in the presence of CORM-2. The complete list of genes that exhibited a twofold, or higher, increase or decrease (with P<0.05) in transcription is presented in Supplementary Tables S1 (anaerobic growth) and S2 (aerobic growth), available with the online version of this paper. In order to independently confirm the microarray results, quantitative real-time PCR was carried out on ten selected genes that were found to be upregulated or downregulated by the action of CORM-2. The values acquired in the real-time PCR experiments were in agreement with the fold variation obtained in the microarrays (Table 1). To facilitate subsequent analysis, the differentially expressed genes were divided into functional categories, based on the NCBI () and EcoCyc (http://ecocyc.org) gene annotations and on the protein databases that were used to search for homologues of hypothetical gene products. Since a better correlation between gene expression and function is found for groups of genes instead of individual genes, the FatiGO algorithm (Al-Shahrour et al., 2004, 2006, 2007) was also used to interpret the data.

Table 1. Quantitative real-time RT-PCR analysis performed in cells exposed to CORM-2 and to the CO-free Ru(II)Cl2(DMSO)4

Anaerobically grown cells.
The microarray data acquired for cells grown under anaerobic conditions and treated with CORM-2 showed that 396 genes had their transcription altered (∼9 % of the total genome). Of these, the transcription of 228 genes was repressed (∼5 % of the transcriptome); these genes were dispersed through nearly all functional categories, particularly amino acid transport and metabolism (Fig. 1A). The FatiGO analysis showed that the genes downregulated are over-represented in cellular metabolic processes (which correspond to 84 % of all repressed genes), namely in catabolic processes, nucleotide metabolism, and energy production by oxidation of organic compounds (Fig. 2A), indicating that CORM-2 induces metabolic adaptation.

(29K): Fig. 1. Effect of CORM-2 on the E. coli transcriptome analysed according to the distribution of the genes with altered expression into functional categories. The differentially regulated genes in cells grown under anaerobic (A) and aerobic (B) conditions were divided into 21 functional categories according to the NBCI and EcoCyc databases. The black and white bars represent the number of genes whose transcription was up- and downregulated by CORM-2, respectively. The functional categories that include genes up- and downregulated by CORM-2 in both growth conditions are highlighted in bold, and the eight functional classes containing genes that were repressed by CORM-2 under anaerobic conditions and are absent in aerobically grown cells are shown in italics.

(21K): Fig. 2. Biological processes modified by CORM-2. The genes down- and upregulated by CORM-2 anaerobically (A and B) or aerobically (C and D) were compared with those unaltered by CORM-2 (reference genes), using the FatiGO algorithm (see Methods). The y-axis shows the biological processes that were considered differentially represented between the two groups compared in each case. The specificity of the biological process increases from the top to the bottom of the y-axis, as indicated by the numbers in parentheses. The x-axis represents the differences in percentages of the relative frequency of genes annotated for each biological process. The right and left part of the x-axis corresponds to the biological processes over- and under-represented, in the downregulated genes (A and C) and the upregulated genes (B and D), respectively. TCA, tricarboxylic acid cycle.

The largest class of genes that were upregulated belongs to the category hypothetical proteins (∼39 %). Apart from these, the genes that showed increased transcription are divided essentially into three classes: inorganic ion transport, post-translational modification and transcription (Fig. 1A). Further analysis revealed that the genes upregulated by CORM-2 were significantly over-represented within the classes transcriptional regulation and protein folding (Fig. 2B). Among the latter, genes encoding two heat-shock proteins that are directly connected with protein stability, ibpA and ibpB, were strongly induced (19- and 40-fold, respectively). The data also showed that, under anaerobic conditions, the addition of CORM-2 had effects on iron metabolism, since several genes related to this function were altered, e.g. those encoding ferritin (ftn), and bacterioferritin (bfr) (Supplementary Table S1).

Aerobically grown cells.
E. coli cells grown aerobically and exposed to CORM-2 displayed altered transcription of 175 genes, ∼4 % of the global gene expression profile, with 46 genes repressed and 129 genes induced. Apart from genes encoding hypothetical proteins, three functional categories exhibited a similar number of transcriptionally repressed genes, namely the classes energy production, and amino acid and carbohydrate transport and metabolism (Fig. 1B). The FatiGO algorithm revealed over-representation of repressed genes in coenzyme catabolic processes, tricarboxylic acid cycle and aerobic respiration classes (Fig. 2C). This trend would ultimately lead to the inhibition of aerobic respiratory metabolism.

The genes whose expression was induced by CORM-2 were spread through several functional classes, the inorganic ion transport category containing the second-highest number of genes transcriptionally modified (Fig. 1B). Interestingly, the FatiGO analysis indicates that under aerobic conditions the addition of CORM-2 to E. coli cells led to a significant upregulation of genes involved in sulfur metabolism, such as tauABC, ssuAD, cysWA and sbp, and in methionine metabolism, like the gene clusters metNI and metBLF (Fig. 2D).

Transcriptional alterations caused by CORM-2 are dependent on CO release
Administration of any CORM produces some metabolite(s) along with the CO that is liberated. Therefore, we determined whether the biological effect of a given CORM actually results from the liberated CO and not from any of its metabolites or the CORM's molecular skeleton itself. The commercially available tricarbonyldichlororuthenium(II) dimer, [Ru(CO)₃Cl₂]₂, also known as CORM-2, is by far the most tested CORM and we have reported its bactericidal activity (Nobre et al., 2007). The dimer complex is insoluble in aqueous media and its solubilization, usually carried out in DMSO, entails a number of chemical transformations. According to Motterlini et al. (2002), 23 min after dissolution three species are already formed, namely RuCl₂(CO)₃(DMSO) and two isomers of RuCl₂(CO)₂(DMSO)₂. Of necessity, a certain amount of free CO is liberated into the DMSO solution and eventually lost to the atmosphere at an undetermined rate. More species appear upon standing or warming of the solution and further CO loss to give RuCl₂(CO)(DMSO)₃, and eventually RuCl₂(DMSO)₄ is to be expected since they are all synthetically interrelated in the presence of CO gas (Alessio et al., 1995). Nothing is presently known about the biological activity of any of these pure species individually. However, the octahedral structure, the oxidation state as well as the ligands other than CO are retained along this series, strongly supporting our choice of Ru(II)Cl₂(DMSO)₄ as a control for the bactericidal activity of the CO-free Ru(II)Cl₂(DMSO)_x fragment. It must also be mentioned that, regardless of the species present in the DMSO solution of [Ru(CO)₃Cl₂]₂, when an aliquot of this solution is added to the cell culture medium no CO is released to the gas phase, as previously reported (Nobre et al., 2007). We challenged E. coli with 500 µM CO-free Ru(II)Cl₂(DMSO)₄ and analysed the expression of several genes. Table 1 shows that no significant alteration in transcription was observed when the CO-free compound was used, which allows us to conclude that the main biological activity is due to the release of CO from CORM-2.

Genes transcriptionally regulated in response to CORM-2, independently of oxygen
The E. coli microarray data showed that upon exposure to CORM-2 the cells grown anaerobically exhibited more repressed genes than the cells grown aerobically. In particular, eight functional classes containing genes that were repressed by CORM-2 under anaerobic conditions were not detected in aerobically grown cells (Fig. 1A, B). These results are in agreement with our previous observation that the decrease in cell viability caused by CORMs is higher for cells grown under anaerobic conditions (Nobre et al., 2007). However, a large set of genes were found in common between the aerobic and anaerobic conditions, and the fold induction or repression generated by CORM-2 did not vary significantly between them (Table 2). There is also no example of a gene that was repressed under anoxic conditions and induced under aerobic conditions, and vice versa. The major fold difference was observed for the genes encoding a heat-shock protein (ibpB), two hypothetical proteins (yncJ and ymgG), a flagellar repressor (lrhA) and an envelope-stress-induced periplasmic protein (spy). In all these cases, the genes showed a higher fold CORM-2-dependent induction in aerobically grown cells than in cells cultured anaerobically (Table 2).

Table 2. Genes of E. coli differentially regulated by CORM-2 under both anaerobic and aerobic conditions

Phenotypic analysis of E. coli regulators induced by CORM-2
To gain further insight into the function of the genes whose transcription was perturbed by treatment with CORM-2 we analysed the phenotype of 12 E. coli mutant strains to test if the deleted gene has a role in the E. coli CORM-2 response. The genes were chosen on the basis that they showed a high fold induction or repression and/or exhibited variation in transcript abundance in cells grown under both aerobic and anaerobic conditions (see Table 2). Hence, genes encoding regulators (soxS, oxyR, zntR, metR, gadX), a putative stress-combating protein (cpxP), heat-shock proteins (ibpA and ibpB), proteins involved in biofilm formation [ydgG (tqsA), ycfR (bhsA)], an antiporter protein (chaA), and genes involved in methionine metabolism (metN, metI) were studied.

Study of the E. coli single mutant strains deleted in the genes zntR, gadX and chaA revealed that these strains did not display a CORM-2 growth-induced delay when compared to the parent strain (data not shown).

The soxS gene, whose transcription is controlled directly by SoxR, showed a high fold induction upon treatment with CORM-2, under both aerobic and anaerobic conditions (Table 2). Together with SoxR, SoxS participates in the regulation of several genes involved in the response to oxidative stress (Wu & Weiss, 1992). However, with the exception of the marRAB operon, none of the other known members of the SoxRS regulon was induced. A similar behaviour, which so far remains unexplained, was previously observed in other microarray experiments with E. coli grown in the presence of nitric oxide and hydrogen peroxide (Justino et al., 2005; Zheng et al., 2001). The phenotypic study of the E. coli strain deficient in the soxS gene showed that the mutation leads to an increased sensitivity to CORM-2 of the cells grown under aerobic (>70 %) and anaerobic conditions (>35 %) (Fig. 3A, B).

(43K): Fig. 3. Analysis of the sensitivity of various E. coli mutant strains to CORM-2. (A–E) The parental strain (grey bars), and mutants (white bars) were grown aerobically (panels A and C) or anaerobically (panels B, D and E) and exposed to 150 µM or 100 µM CORM-2, respectively. The E. coli ΔcpxP mutant and its parental strain were treated with 125 µM CORM-2 (panel E). In panel F, the ΔmetR (striped bars), ΔmetI (light grey bars) and ΔmetN (white bars) mutant strains were grown aerobically and treated with 150 µM CORM-2. The error bars represent the standard error of the mean values obtained from at least two independent biological samples performed in duplicate (*P<0.01, ** P<0.05). Panels A1–F1: complementation analysis of ΔsoxS (A1, B1), ΔoxyR (C1), ΔibpAB (D1) and ΔmetR (F1) using the plasmids indicated in each panel.

The oxyR gene, whose encoded protein participates in the regulation of genes involved in the response to oxidative stress (Zheng et al., 2001), was upregulated by CORM-2 under aerobic and anaerobic conditions (≥4-fold) (Table 2). Under aerobic conditions and in the presence of CORM-2, the sensitivity of the oxyR mutant strain was found to be ∼70 % higher than that of the wild-type strain (Fig. 3C), while no differences were observed for growth under anaerobic conditions.

The microarray data showed that in aerobically grown E. coli the transcription of metR was upregulated by CORM-2. Analysis of the ΔmetR strain revealed that this mutant displays a sensitivity ∼80 % higher than that of the parental strain for cells grown aerobically (Fig. 3F), while no effect was detected in anoxic conditions (data not shown).

In all cases, the wild-type behaviour was restored upon expression of the genes from a plasmid containing SoxR, OxyR or MetR (Fig. 3A₁–C₁ and F₁), showing that these transcription factors have an important role in the E. coli regulatory mechanisms triggered by exposure to CORM-2.

Identification of other genes involved in the CORM-2 sensitivity of anaerobically grown E. coli cells
A significant variation of the expression of ibpA and ibpB genes was measured in aerobically and anaerobically grown cells of E. coli treated with CORM-2 (Table 2). These genes encode two small heat-shock proteins, IbpA and IbpB, that bind to protein aggregates and inclusion bodies formed during heterologous protein expression (Lethanh et al., 2005). The two proteins are known to cooperate with ClpB and DnaK, forming a functional triad of chaperones (Lethanh et al., 2005; Thomas & Baneyx, 1998). Interestingly, the transcription of clpB and dnaK as well as that of genes coding for other chaperones, such as dnaJ, grpE and htpG, was also upregulated anaerobically by CORM-2 (Supplementary Table S1). Under anaerobic conditions, inactivation of the ibpAB genes led to a 45 % increase in CORM-2 sensitivity relative to the parental strain (Fig. 3D), while no differences were observed under aerobic conditions (data not shown). As expected, complementation of ibpAB mutation induced the rescue of the wild-type phenotype (Fig. 3D₁).

The transcription of E. coli cpxP, which encodes a periplasmic protein putatively involved in combating extracytoplasmic protein-mediated toxicity (Danese & Silhavy, 1998), was upregulated 12–14-fold in cells exposed to CORM-2 and cultured in both aerobic and anaerobic conditions. CpxP belongs to the E. coli Cpx system, which senses perturbations in the bacterial cell envelope and responds through the upregulation of many gene products involved in protein folding and degradation (Fleischer et al., 2007). Accordingly, we verified that together with the induction of cpxP occurred a high number of transcriptionally modified genes involved in cell wall biogenesis and in protein folding, including genes encoding heat-shock proteins (ibpAB, hslJ, htpX), chaperones (dnaJ, htpG, clpB) and proteases (ftsH) (Supplementary Table S1), mainly under anaerobic conditions. These genes are under the control of rpoH (σ³²) (Zhao et al., 2005), a gene that was also found to be upregulated by CORM-2 under anaerobic conditions. The study of the growth behaviour of the ΔcpxP mutant revealed that this strain is approximately 35 % more sensitive to CORM-2 than the parental strain, but only under anaerobic conditions (Fig. 3E), not under aerobic conditions (data not shown).

CORM-2 interferes with methionine biosynthesis
Addition of CORM-2 to aerobically grown cells of E. coli caused a marked increase in the transcription of several genes implicated in methionine biosynthesis and uptake, with metR being the gene that showed the highest induction (21-fold) (Supplementary Table S2). Apart from the genes listed in Supplementary Table S2, induction of the expression of metA (12.4-fold; P=0.0527) and metE (5.5-fold; P=0.0565) was also observed. To further clarify this issue, growth experiments were conducted aerobically in the presence of CORM-2 for the E. coli ΔmetR, ΔmetI and ΔmetN mutant strains. All these mutants were hypersensitive to CORM-2 (80 %, 66 % and 40 % higher sensitivity, respectively; Fig. 3F). Furthermore, the complementation assay showed that E. coli ΔmetR regains resistance to CO upon expression of the regulator MetR (Fig. 3F₁). These findings suggest that, under aerobic conditions, CORM-2 affects the metabolism of methionine by E. coli.

CORM-2 influences biofilm formation
Previous microarray studies revealed that apart from the E. coli genes that are directly implicated in the process of biofilm formation, such as tqsA, mqsR, bhsA, yceP (bssS) and yliH (bssR), genes involved in the response to stress conditions are also induced during biofilm formation, namely ibpAB, soxS, cpxP and spy (Beloin et al., 2004; Ren et al., 2004). Many of these genes were also found to be transcriptionally modified when E. coli is exposed to CORM-2 (Tables 2 and S3). In particular, comparison of the microarray data showed that approximately 40 % of the genes that are transcriptionally modified during the formation of E. coli biofilm (Ren et al., 2004) are in common with those altered by CORM-2 under anaerobic conditions, and 12 % of the common genes are observed when growth is performed aerobically (Supplementary Table S3).

To evaluate the effect of CORM-2 in the process of biofilm formation, quantification of the total biofilm formed during aerobic growth in LB and in the presence of the compound was performed. The results showed that CORM-2 increased by 1.6-fold the total biofilm of the E. coli wild-type strain (Fig. 4A). This indicates that E. coli produces more biofilm as a defensive response against CORM-2, as previously observed for other stress conditions (Zhang et al., 2007).

(31K): Fig. 4. Effect of CORM-2 on biofilm formation and cell viability of E. coli ΔtqsA and ΔbhsA mutant strains. (A) Biofilm was assayed in E. coli wild-type and in ΔtqsA and ΔbhsA mutant strains, in the absence or presence of 250 µM CORM-2. (B, C) Cell viability of E. coli parental strain (grey bars), ΔtqsA (white bars) and ΔbhsA (black bars) mutants grown anaerobically (b) or aerobically (c) in the presence of 100 µM or 180 µM CORM-2, respectively. The number of c.f.u. was determined for at least two biological samples and in duplicate. Error bars represent the standard error of mean values (*P<0.01, ** P<0.05). Complementation analysis of ΔtqsA and ΔbhsA (B1 and B2, anaerobic conditions) and ΔtqsA (C1, aerobic conditions) mutant strains, using the plasmids indicated in each panel, is also shown.

The bshA gene, which is related to biofilm formation (Zhang et al., 2007), was significantly induced under anaerobic conditions (26-fold). Although the ΔbhsA mutant showed an elevated resistance to CORM-2 anaerobically (Fig. 4B), and the complementation experiment performed with a multiple-copy clone containing the bhsA regulatory region rescued the wild-type behaviour (Fig. 4B₂), the values of total biofilm measured for the mutant strain exposed to CORM-2 matched those of the parental strain (Fig. 4A).

In E. coli, the tqsA gene is proposed to encode a putative transport protein involved in biofilm formation (Herzberg et al., 2006). This gene was upregulated in aerobically and anaerobically grown CORM-2-treated cells, with a 9- and 6-fold increase, respectively. Analysis of the E. coli ΔtqsA mutant revealed that deletion of tqsA yielded a strain with higher resistance to CORM-2 than the parental strain, under both aerobic and anaerobic growth conditions (Fig. 4B, C), and the phenotype of the tqsA mutant strain could be fully complemented by a clone containing the tqsA gene (Fig. 4B₁–C₁). This result is in agreement with those obtained by Herzberg et al. (2006), which showed that inactivation of tqsA increases the resistance to several antibiotics. Additionally, deletion of tqsA abolished the increase in biofilm formation observed in the wild-type strain upon exposure to CORM-2 (Fig. 4A).

Conclusion
In this study, we provide the first microarray analysis of a micro-organism treated with the new bactericide CORM-2. The choice of E. coli was based on the fact that, besides being a model bacterium, this Gram-negative micro-organism is able to live in aerobic environments and also adapts to anaerobic niches as part of its normal colonization–transmission cycle within a host. Hence, it allowed us to compare the transcriptional response of a bacterium when treated with CORM-2 in both anaerobic and aerobic environments. The broad effect of CORM-2 on gene expression levels is evident from the distribution of induced and repressed genes over the spectrum of all functional categories. In particular, the higher number of genes affected in anaerobically CORM-2-exposed cells illustrates fundamental differences between the way in which microbes control external stress under the two oxygen conditions, and highlights the importance of specific investigations to understand the different adaptation strategies.

Although it has been assumed that the targets of CO are haem-containing enzymes or proteins, they have not yet been identified and little is known about the trigger event or mechanism that mediates the transfer of CO from the CORM to the haem target. The present data show that the action of CO extends beyond the action on haem proteins since a wide range of transcriptional modifications is observed in cells grown under fermentative conditions and exposed, aerobically or anaerobically, to CORM-2.

The changes in the expression level of key transcription regulators together with the phenotypic analysis of the mutant strains reveal that CORM-2 triggers a complex network of responses. As observed in all other microarray studies performed in E. coli cells submitted to toxic chemicals such as hydrogen peroxide and NO, exposure to CORM-2 increases the transcription of the soxS regulator but it does not induce any of the genes of the regulon, a result that requires further studies for clarification. A particular observation, common to all transcriptional studies, is the large representation of unknown genes, which leaves a wide field to be still explored, until bacterial physiology is fully understood and indeed ready for the application of a systems biology approach. Nevertheless, the data provided by this study will certainly be valuable for guiding future research on the pharmacological application of CORMs.

The authors are grateful to Professor Bernard Weiss for providing E. coli ΔsoxS, E. coli BW748 and BW1068, Professor François Baneyx for E. coli ΔibpAB, Professor Thomas Silhavy and Dr Dan Isaac for the ΔcpxP mutant, Professor Robert Poole and Professor Frederick Blattner for the E. coli strains mutated in methionine-related genes (ΔmetR, ΔmetI and ΔmetN), Dr Kobayashi Hiroshi for the ΔchaA mutant, Professor John Foster for the ΔgadX mutant, Professor Robert Poole and Dr Andrew Morby for the ΔzntR mutant, Professor Thomas Wood and Dr Rodolfo Contreras for the ΔtqsA and ΔbhsA mutants and E. coli AG1 containing the pCA24N, pCA24NtqsA⁺ and pCA24NbhsA⁺ vectors, and Dr Gisela Storz for the pAQ17 vector. We thank Professor Miguel Teixeira and Professor Carlos Romão from ITQB for helpful discussions. This work was supported by FCT project POCI/SAU-IMI/56088/2004 and L. S. N. is recipient of PhD grant FCT SFRH/BD/22425/2005.

Edited by: S. C. Andrews