INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion References

Reporter gene fusions have been widely and successfully used to monitor gene expression in microbes, leading to important fundamental discoveries (35) and numerous applications (19). The facile methods for generating (4) and screening (10, 17, 51) such fusions have allowed genomewide surveys of gene expression in a response to a variety of stimuli. However, typically only the gene fusions having the desired property are identified in such surveys. With the availability of complete genomic sequences for many microbes, use of large sets of precisely defined reporter gene fusions is practical. Recently, defined sets of transcriptional gene fusions using gfp as a reporter in Saccharomyces cerevisiae (9) and luxCDABE (46) as a reporter in Escherichia coli have been described. The advantages of easily assayed reporter end products, such as fluorescence or light production, become increasingly critical for high-density analyses. For bacteria, bioluminescence as a reporter of gene expression is particularly useful because of the sensitivity and large dynamic range (5). Furthermore, use of the five-gene luxCDABE operon allows facile monitoring of kinetic responses because all the components necessary for light production are present in the cell, thus obviating the need for cell lysis and substrate addition (21).

Cellular arrays of reporter genes can be considered an alternative, complementary approach to other currently used genomewide transcriptional analyses such as those involving DNA arrays. Despite the success and broad use of DNA arrays, there are limitations, such as artifacts from RNA isolation (38) and cross hybridization (31). Here we describe a new solid-phase assay system consisting of 689 reporter strains, each containing a different, precisely sequenced fusion. The reporter strains were chosen from a collection of random, genomewide luxCDABE gene fusions in E. coli that had been sequenced to determine the identity and orientation of the genomic fragment upstream of the reporter genes; this collection includes representative members of several global stress response regulons and thus mirrors the E. coli transcriptional wiring diagram (46). The essence of this assay is to collect an image of the signal generated from reporter constructs arrayed in high density, such that the signal intensity can be subsequently quantified. Thus, bioluminescence from a high-density array of luxCDABE gene fusions in E. coli was measured by creating an image of the entire reporter array and quantitating the pixel density. Differences in the pixel density measurements in the presence or absence of a perturbation defined gene expression alterations.

The utility of this solid-phase cellular array was demonstrated by detection of responses to nalidixic acid, an inhibitor of DNA gyrase. Treatment of E. coli with this compound is known to result in selective transcriptional upregulation. The predominant response to nalidixic acid treatment is induction of the LexA-controlled SOS response (49). In addition to the SOS response, upregulation of the ³²-controlled heat shock response (18, 40, 45), increased activity of promoters responsive to DNA supercoiling (25, 37), increased expression of emr genes (20), and other responses (13, 36) are induced by nalidixic acid treatment. Accordingly, upregulation of gene fusions in two classes, fusions to genes of the LexA regulon and fusions to genes that are not part of the SOS response, was expected and was found. The genes included several novel nalidixic acid-upregulated genes, some of which were demonstrated to be members of the LexA regulon.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion References

E. coli strains, genetic nomenclature, growth medium, and chemicals. The construction (42), sequencing, and characterization (46) of a collection of random, 1.8-kbp-average-size E. coli genomic fragments upstream of Photorhabdus luminescens luxCDABE in moderate-copy-number plasmid pDEW201 (44) have been described. The host strain for transformants in this collection is E. coli DPD1675 (ilvB2101 ara thi [proAB-lac] tolC::miniTn10). The list of the 689 transcriptional fusions between predicted E. coli promoters and the luxCDABE operon (described in Results) is available upon request. Individual strains from this set are available to members of the scientific community for noncommercial purposes by contacting the corresponding author. Strain and plasmid names, in addition to the lux clone identification number, were assigned for strains containing gene fusions that were reproducibly upregulated by nalidixic acid, as shown in Table 1. Plasmid DNA isolated using a QiaPrep Spin kit (Qiagen Corp.) was used for transformation of E. coli DM800 (F metA28 lacY1 or lacZ4 thi-1 xyl-5 or xyl-7 galK2 tsx-6) and otherwise isogenic lexA1 strain DM803 (24), selecting for ampicillin resistance.

Preparation of reporter arrays. Duplicate cultures of the E. coli strains in the LuxArray were grown overnight at 37°C in 40 µl of LB medium supplemented with 100 µg of ampicillin/ml in a set of 16 96-well plates. These cultures were used as the cell source to manufacture the arrays. Porous nylon membranes (8 by 12 cm; Biodyne B; Nunc) were sterilized by UV illumination for 10 min and then placed in contact with solid LB growth media in a culture dish (50 ml; OmniTray; Nunc) that had been prewarmed to 37°C. Printing of 1,536 spots in four-by-four subarrays was accomplished using a BioMek 2000 (Beckman Coulter) equipped with a high-density replication tool. Sterilization between transfers was accomplished by soaking the pins successively in 0.2% sodium dodecyl sulfate in water, sterile water, and 70% ethanol. After sterilization, the pins were air dried prior to the next transfer. The E. coli reporter strains and control strains were printed in triplicate for each treatment.

Data analysis from reporter arrays. Spot intensity of each image was determined using ArrayVision (Imaging Research, Toronto, Canada), and the resultant pixel density measurements were imported into a template with identifiers for each spot. The background signal, which results from cross illumination of neighboring spots, was estimated by finding the median of the 24 spots containing a strain with the parental plasmid on each of the triplicate arrays followed by calculation of the average of the three medians. The background signal at each of the time points was subtracted from each reporter strain measurement at the corresponding time point. All negative numbers were converted to zero. The average signal for each of the triplicate spots was calculated.

Kinetic analysis of bioluminescence using liquid cultures. Kinetic analyses of bioluminescent response to chemicals was done as previously described (42, 43), except that a Luminoskan Ascent microplate luminometer (Labsystems) was used to measure light production at 37°C. Briefly, fresh overnight cultures for each strain grown in LB medium supplemented with ampicillin were diluted into LB medium and grown to mid-exponential phase at 37°C. These actively growing cultures were divided at the initiation of chemical treatment in 100 µl (total volume) in the wells of white, 96-well microplates (Microlite; Dynex). Pipetting the cell cultures was done rapidly at room temperature, after which the microplate was placed immediately into a prewarmed luminometer. Nonetheless, some cooling of the cell cultures occurred during pipetting, leading to an initial rise in the bioluminescence of both the control and chemically treated cultures as they warmed toward the temperature optimum of the Lux proteins. A typical liquid culture experiment used seven concentrations of the chemical of interest such that the range included a sublethal dose at which stress responses were induced without severe toxicity that would inhibit the light production reactions (41). Nevertheless, some growth inhibition may occur, and thus the response ratio, which is the bioluminescence of the chemically treated culture divided by the bioluminescence of the control at each time point (41), represents a minimum estimate of gene expression responses because it does not include a correction for reduced cell density in the presence of the added chemical.

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion References

Selection of a maximal nonredundant set of lux gene fusions. A random luxCDABE gene fusion collection of 4,988 plasmid-borne gene fusions, each with boundary information relative to the E. coli genome and the orientation relative to the lux operon (46) was the starting point from which an optimum subset was selected. A functional construct was defined as one consisting of a genomic fragment encompassing a promoter adjacent to the promoterless lux operon in an orientation that causes transcription initiated at the promoter to proceed into and through the lux operon. Therefore, to identify the functional subset of the collection, Perl scripts were used to computationally filter the list of gene fusions first for functionality and second for redundancy. A list of definitions of documented and predicted operons (39) was used to define genomic coordinates of the operons as the translational start codon position of the first open reading frame (ORF) in the operon and the translational stop codon position of the last ORF in the operon. Additional information included the strand on which the operon is contained (direction of transcription) and gene names. The lux gene fusions were filtered computationally using the following criteria to select functional transcriptional fusions: (i) the genomic fragment must start more than 50 bp upstream of the start codon of the first ORF and end anywhere between the start codon of the first ORF in the operon and the stop codon of the last ORF in the operon, thereby eliminating the occurrence of a transcriptional stop signal in the construct between the promoter and the lux operon; (ii) the promoter contained in the genomic fragment must be orientated such that it directs transcription into the lux operon. Finally, when more then one gene fusion fit the criteria for a single operon, only the construct containing the genomic fragment representing the greatest amount of upstream sequence was retained, thereby eliminating redundancy. The resulting 689 selected gene fusions represent 27% of the 2,584 known and predicted transcriptional units in the E. coli genome (3, 39). The working hypothesis based on the predicted operon structures is that bioluminescence from cells containing each of the selected gene fusions reports on the expression of the operon to which luxCDABE is joined.

Initial characterization of the LuxArray. The bioluminescence over time from a LuxArray consisting of 1,536 spots arrayed at a density of 16 spots/cm² of medium is shown in Fig. 1A. Each array in Fig. 1 was a side-by-side replicate, such that the left half was identical to the right half. The intensity of bioluminescence at each location reflected the relative activity level of each individual promoter controlling expression of the luxCDABE reporter. The increased bioluminescent signal at each spot location with time was, in general, due primarily to cell growth.

View larger version (71K): [in this window] [in a new window]   FIG. 1.   Images of duplicate LuxArray 1.0 cellular bioluminescent reporter arrays. Following the spotting of the E. coli strains containing reporter gene fusions onto membranes on LB agar plates and growth for 6 h, the membranes were moved to LB medium plates (A) or LB medium plates containing 5 µg of nalidixic acid/ml (B). The images were taken as described in Materials and Methods immediately after moving the membrane and subsequently at 2, 4, 6, and 8 h and after overnight incubation. A magnification of the 16 spots in the D-4 primary location from panels A and B is shown in panel C. The spot at the secondary location of row 3, column 2, containing cells with upregulated gene fusion dinB-luxCDABE is boxed.

Identification of nalidixic acid-responsive gene fusions using LuxArray. The antibiotic nalidixic acid, an inhibitor of DNA gyrase known to be an effective inducer of the SOS DNA damage stress response (49), was used to test the utility of this array for detecting changes in gene expression after perturbation. Figure 1B displays the bioluminescence over time of the LuxArray treated with 5 µg of nalidixic acid/ml. In contrast to that of the control array (Fig. 1A), the bioluminescence levels of the majority of the spots did not increase substantially after 2 h of nalidixic acid treatment (Fig. 1B). This general inhibition of bioluminescence increase was due to growth inhibition or metabolic inhibition, which lowers cellular ATP levels and reducing power required for the bioluminescence reactions. This inhibition of bioluminescence increase was also evident from the sum of all bioluminescent signals in the array plotted as a function of time (Fig. 2). Thus, each bioluminescence measurement was normalized to correct for the lower cellular density in the nalidixic acid-treated array (see Materials and Methods). A scatter plot of this normalized data comparing the nalidixic acid-treated spots to untreated spots at the 4-h time point is shown in Fig. 3. It is evident that the signals from vast majority of the spots in the array were not affected by nalidixic acid treatment but that spots with increased signal in response to nalidixic acid, as well as those with decreased signal, were present. Also evident from this scatter plot is that there was a wide range of signal strengths and that the data points for low signals showed more scatter than those for high signals.

View larger version (18K): [in this window] [in a new window]   FIG. 2.   Total light production from LuxArray 1.0 over time. The averaged pixel density of each spot on triplicate membranes was summed over the entire array and plotted at each time point for each condition. Squares, control LB plates; diamonds, LB plates containing 5 µg of nalidixic acid/ml.

View larger version (23K): [in this window] [in a new window]   FIG. 3.   Scatter plot of showing the relationship of normalized signal intensity with and without nalidixic acid treatment at the 4-h time point. The data from all signals greater than 1.0 for both treatments are plotted.

Validation of nalidixic acid-upregulated gene fusions with cultures in liquid medium. Each of the eight newly identified nalidixic acid-upregulated gene fusions and three of the known SOS gene fusions were tested in liquid medium using a range of nalidixic acid concentrations (80 µg/ml in twofold dilutions to 1.2 µg/ml). Table 2 shows the results expressed as ratios of the signal from the nalidixic acid-treated cultures to that from the untreated control at 2 h without correction for growth inhibition caused by nalidixic acid. The concentration that yielded the maximal response and the range of nalidixic acid concentrations that resulted in response ratios of 1.5 or greater are also given. By the criteria of a maximal response ratio of at least 1.8 and of more than one concentration tested yielding response ratios of at least 1.5, six of the eight putative novel nalidixic acid-upregulated gene fusions were reproduced in liquid medium.

Mitomycin C responses and effect of a noninducible lexA allele. Mitomycin C, a compound that damages DNA by intercalation and covalent modification, was used to determine if these newly discovered nalidixic acid-upregulated gene fusions were generally responsive to DNA damage. In addition, the effect of a lexA1 mutation, which renders the LexA repressor noncleavable by the RecA coprotease (24), was tested. The expectation was that expression of SOS-responsive gene fusions controlled by LexA-regulated promoters would be induced by both nalidixic acid and mitomycin C in a manner that is dependent on normal LexA function. As shown in Fig. 4, the expression of the gene fusion to yigN was markedly induced by both nalidixic acid and mitomycin C in the lexA⁺ host. The kinetics of this response, consisting of a 20-min lag time, where the bioluminescence of the nalidixic acid-treated cultures was not different from that of the untreated control, followed by a dose-dependent nalidixic acid-mediated increase in light production, were similar to the those of the response of other LexA-regulated gene fusions (46). This upregulation of gene expression was completely eliminated when the same plasmid was put into an otherwise isogenic lexA1 host strain (Fig. 4). Likewise, the expression of a gene fusion to oraA was also induced by both mitomycin C and nalidixic acid only in the lexA⁺ host (Table 3). In contrast, four other nalidixic acid-responsive gene fusions were not upregulated by mitomycin C in the lexA⁺ host, suggesting that they are not generally DNA damage responsive but rather are more specifically responsive to nalidixic acid. Consistent with this observation, the upregulation by nalidixic acid, which was weaker in general than the LexA-controlled responses, was still present in the lexA1 mutant host (Table 3). Unexpectedly, the bioluminescence of these four gene fusions was slightly upregulated in response to mitomycin C in the lexA1 mutant host. Also shown in Table 3 are the levels of light production in the absence of chemical treatment. These levels, which reflect promoter strength, were largely unaffected by the lexA1 mutation. As expected, expression of ^S-controlled gene fusion yciG-lux (42), which was not expected to respond to DNA damage, was not induced by either nalidixic acid or mitomycin C (Table 3).

View larger version (32K): [in this window] [in a new window]   FIG. 4.   Response of the yigN-luxCDABE gene fusion to nalidixic acid (NA) and mitomycin C (MC) in lexA+ and lexA1 host strains. Actively growing cultures in LB medium were mixed with chemicals at time zero, and light production was measured in a Luminoskan Ascent microplate luminometer. (A and C) Plasmid pDEW634, containing the yigN-luxCDABE gene fusion, in E. coli strain DM800. (B and D) Plasmid pDEW634 in E. coli strain DM803. RLU, relative light units.

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion References

A high-density cellular array of transcriptional reporter gene fusions. LuxArray 1.0 consists of a set of 689 E. coli strains, each containing a reporter gene fusion to a different E. coli promoter element. Overall, 27% of the known or predicted transcriptional units in E. coli are represented in this array. It has previously been shown that several major stress-responsive regulons are each represented by one or more gene fusions in this collection (46). Thus, it is reasonable to assume that activation of most other major regulatory circuits will be reported by gene fusions in this array. Although the LuxArray does not afford a complete analysis of transcriptional alterations in E. coli, it provides a representative view of transcriptional changes because the set of strains originated in a sequenced collection of random gene fusions. Given that a larger collection of random gene fusions was culled using a compilation of known and predicted promoters in E. coli (39) as a reference and that predicting promoters and operons is not yet entirely accurate (15), the LuxArray may include some gene fusions to promoters that are not active despite predictions (46) and may have missed some promoters that were not predicted yet that are functional. Nevertheless, LuxArray 1.0 provides a genomewide collection that will yield a much more detailed and representative transcriptional pattern in response to perturbation than previously used panels of selected stress-responsive gene fusions (2, 27, 45).

Responses to nalidixic acid perturbation. Twelve putative upregulated gene fusions were identified following perturbation of the LuxArray with nalidixic acid treatment. Ten of these were confirmed by demonstrated upregulation in response to nalidixic acid in liquid medium (Table 2) (46). Two with differing responses on solid medium and in liquid medium may be false positives or may represent responses that only occur when the cultures are grown on solid medium, as has been observed for aluminum-activated gene expression in E. coli (14). Another formal possibility is that overexpression of a full-length ORF contained in the plasmid with the gene fusion conferred increased resistance to nalidixic acid. Thus growth to a greater extent than that for the majority of the spots in the array could give the appearance of a greater signal after normalization to correct for growth inhibition. This explanation is ruled out for the 10 gene fusions that were demonstrated to be upregulated in liquid medium where the bioluminescent signal was greater in the presence of nalidixic acid than in its absence without correction for growth inhibition. Of the 10 gene fusions that yielded upregulation by nalidixic acid in both solid and liquid formats, 4 were fusions to known LexA-regulated genes. Also included in the LuxArray were three additional gene fusions to known members of the LexA regulon that did not meet the established criteria to be scored as upregulated. This rate of false negatives, 43%, is similar to the rate from a DNA array analysis where 7 of 19 SOS genes, 37%, were not scored as upregulated following mitomycin C treatment (46). Improvement in these rates of false negatives by optimization of conditions is possible and desirable for both methods. Even so, since all methods are likely to have associated errors, the importance of alternative, independent methods of genomewide gene expression for increased reliability of transcriptional response analyses is underscored.

New LexA-regulated SOS genes. Gene fusions to two genes not previously known to be members of the LexA-dependent SOS regulon were found to be upregulated by two DNA-damaging agents with different mechanisms of action in a LexA-dependent fashion (Table 3 and Fig. 4). These results suggest that yigN and oraA are new members of this regulon. In agreement with the results from the LuxArray analysis reported here, a very recent publication reports lexA-dependent upregulation of yigN and oraA transcription in response to UV light treatment as determined by DNA array (7). Conversely, these results conflict with another recent report in which a LexA binding site upstream of yigN was identified but no increase in mRNA formation upon mitomycin C treatment or regulation by LexA was observed (11). However, the substantial, lexA-dependent upregulation of a luxCDABE gene fusion to yigN observed in response to both mitomycin C treatment and nalidixic acid treatment (Fig. 4) as well as the data from Courcelle et al. (7) suggest that this gene, which encodes a conserved protein with two modules, both related to transcriptional repressors (32, 33), is in the LexA-controlled SOS regulon.

Novel nalidixic acid-upregulated genes that are not generally DNA damage inducible. Four gene fusions demonstrated a specificity of response to nalidixic acid compared with mitomycin C, thus suggesting that these are not part of the DNA damage-responsive SOS regulon. None of the promoters for these four transcriptional units are known to be controlled by ³², nor were any of the known ³²-controlled gene fusions in the LuxArray found to be upregulated by nalidixic acid treatment. Thus, this format of gene expression analysis does not report on the induction of the heat shock response by nalidixic acid, which has a magnitude lower than that of the SOS response (40, 45). These four nalidixic acid-specific gene fusions are likely responding to other signals, such as levels of DNA supercoiling (22), which are decreased in plasmid DNA upon inhibition of DNA gyrase with nalidixic acid (13, 26). However, currently available information about expression of these transcriptional units sheds little light on possible mechanisms of nalidixic acid-mediated regulation.

Comparison of reporter gene array and DNA array technologies. The results presented here demonstrate that a cellular array of reporter gene fusions can be used in a fashion analogous to that for DNA array hybridization assays to monitor transcriptional changes. The application of independent methods for genomewide transcriptional analysis is useful because there are advantages and disadvantages of each. An advantage of the current DNA array technology is the availability of comprehensive analysis of essentially all ORFs in several microorganisms (6, 31, 52, 54) due to the relative simplicity of their construction. Of course, comprehensive reporter arrays are also possible to construct by sequencing a larger collection of random gene fusions or by PCR amplification of promoter regions and subsequent cloning. However, not all applications require a comprehensive analysis. Thus, transcriptional fingerprints that distinguish chemicals with different modes of action are found with a representative subset of the genome, as shown here. DNA arrays are likely to be more useful than reporter gene arrays for genomewide transcription analyses of mutant strains because of the inconvenience in transferring a large set of reporter constructs to a mutant host. Implementation of both DNA arrays and reporter gene arrays requires a substantial amount of equipment; however, the automated liquid handling and camera required for the LuxArray are not specialized and thus do not need to be dedicated solely to this technology.