INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion Appendix References

Escherichia coli K-12 has been exhaustively studied for over 50 years. Early experiments measured the molecular fluxes from small compounds into macromolecular constituents (33). These studies were followed by others in which small molecule pools of central metabolic building blocks (21), nucleotides (3), and amino acids were enumerated. The levels of several macromolecular components, including individual species of proteins (26), have been measured. Such measurements of the steady state provide a census of the cellular content, while changes upon imposition of a stress catalogue the cell's fight for survival. This response to an insulting or adverse condition can take many forms, from relieving end product inhibition to derepressing transcription (20).

In E. coli, experiments to define stress-related, global regulatory responses have often relied upon either the isolation of operon fusions induced by a particular stress (16) or proteomic measures in which the protein fractions from stressed and unstressed cultures are separated by a two-dimensional method prior to comparison (37). Each method has an inherent technological hurdle; the map location of responsive gene fusions must be ascertained precisely, while induced or repressed proteins excised from the two-dimensional gels must be correctly identified.

Alternatively, mRNA measurements utilizing techniques such as hybridization to DNA and primer extension have allowed the monitoring of individual gene's expression profiles. Recently, expression profiling of most yeast genes has been reported (8, 40); such measurements were facilitated by high-density arrays of individual genes and specific labeling of cDNA copies of eukaryotic mRNA by using poly(A) tail-specific primers. Thus, the lack of a poly(A) tail and the extremely short bacterial mRNA half-life represent hurdles for the application of DNA microarray technology to prokaryotic research. Nonetheless, early attempts at comprehensive expression profiling using large DNA fragments from an ordered  library of E. coli genomic fragments as a capture reagent and radiolabeled cDNA as a probe suggested that these problems were not insurmountable (6).

Here we present a means to successfully perform microarray-based comprehensive gene expression profile analyses with E. coli. We show that such experimentation can be informative by examination of (i) differences in gene expression profiles caused by growth of E. coli in either minimal or rich medium, (ii) changes in gene expression associated with the transition from exponential-phase to stationary-phase growth in minimal medium, and (iii) the specificity of induction mediated by isopropyl--D-thiogalactopyranoside (IPTG), the classic lac operon inducer. Moreover, a method for determining the relative abundance of each transcript was developed and used to provide a census of the mRNA composition of E. coli under each of the growth conditions mentioned above.

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion Appendix References

Microbiological methods. E. coliMG1655 (1) was cultured with aeration in either the minimal medium, M9 (23), supplemented with 0.4% glucose or in the rich medium, Luria-Bertani (LB) (23), at 37°C. The overnight culture was diluted 250-fold into fresh medium and aerated at 37°C. Samples of the minimal medium culture were harvested at A₆₀₀s of 0.40 (exponential phase; just under five generations) and 1.6 (transition to stationary phase, just less than seven generations) prior to RNA isolation. An IPTG induction (23) was performed to examine the specificity with which it affects gene expression. The LB medium-grown culture was split when it achieved an appropriate density (A₆₀₀ of 0.40). To one portion was added IPTG to a final concentration of 1 mM; the untreated sample served as a control. Incubation of both samples was continued with aeration at 37°C for another 15 min (A₆₀₀ of 0.45 for both cultures) before RNA isolation was initiated.

RNA isolation. Shaved ice was added to 50-ml samples which were pelleted immediately in a refrigerated centrifuge by spinning at 10,410 × g for 2 min. Each resultant pellet was resuspended in a mixture containing 100 µl of Tris-HCl (10 mM, pH 8.0) and 350 µl of -mercaptoethanol-supplemented RLT buffer (Qiagen RNeasy Mini kit; Valencia, Calif.) that was kept on ice. The cell suspension was added to a chilled 2-ml microcentrifuge tube containing 100 µl of 0.1-mm-diameter zirconia-silica beads (Blospec Products Inc., Bartlesville, Okla.). The cells were broken by agitation at room temperature for 25 s with a Mini-Beadbeater (Biospec Products, Inc.). Debris was pelleted by centrifugation for 3 min at 16,000 × g and 4°C; the resultant supernatant was mixed with 250 µl of ethanol. This mixture was loaded onto Qiagen RNeasy columns from the Qiagen RNeasy Mini kit. RNA isolation was completed by using the protocol supplied with this kit. Incubation for 1 h at 37°C in 40 mM Tris (pH 8.0), 10 mM NaCl, 6 mM MgCl₂ with RNase-free RQ1 DNase (1 U/µl; Promega, Madison, Wis.) digested any genomic DNA contaminating the RNA preparation. The digestion products were purified by a second passage through an RNeasy column (Qiagen). The product was eluted from the column in 50 µl of RNAse-free water prior to determining sample concentration by an A₂₆₀ reading. RNA preparations were stored frozen at 20°C until use.

Synthesis of fluorescent cDNA from total RNA. To a volume brought to 22 µl with double-distilled water (ddH₂O) were added 6 µg of total RNA template and 12 µg of random hexamer primers (Operon Technologies, Inc., Alameda, Calif.). Annealing was accomplished by incubation for 10 min at 70°C followed by 10 min at room temperature. cDNA probes were synthesized with SuperScript II reverse transcriptase (10 U/µl; Life Technologies, Inc., Gaithersburg, Md.) in the presence of deoxynucleoside triphosphates (dNTPs) (dATP, dGTP, and dTTP, each at 0.1 mM; dCTP at 50 µM) and Cy3- or Cy5-dCTP at 25 µM. In order were added 8 µl of 5× SuperScript II reaction buffer (Life Technologies, Inc.), 4 µl of 0.1 M dithiothreitol, 2 µl of the dNTP mix (2 mM dATP, 2 mM dGTP, 2 mM TTP, 1 mM dCTP), 2 µl of 0.5 mM Cy3- or Cy5-labeled dCTP (Amersham Pharmacia Biotech, Arlington Heights, Ill.), and 2 µl of SuperScript II reverse transcriptase. cDNA synthesis proceeded at 42°C for 2.5 h before the reaction was terminated by heating at 94°C for 5 min. The RNA templates were hydrolyzed with 0.25 M NaOH. The reaction was then neutralized by adding HCl and Tris-HCl (pH 6.8). The labeled cDNA was purified with a PCR purification kit (Qiagen), dried, and stored at 20°C. Labeling efficiency was calculated by using the A₂₆₀ and either A₅₅₀ for Cy3 incorporation or A₆₅₀ for Cy5 labeling measurements.

Fluorescent copying of genomic DNA. Genomic DNA was isolated from strain MG1655 by a standard procedure (38). Genomic DNA, sheared with a nebulizer to approximately 2-kbp fragments, was used to prepare labeled DNA. Three micrograms of this DNA was mixed with 6 µg of random hexamer primers (Operon Technologies, Inc.) in 33 µl of ddH₂O. DNA was denatured by heating at 94°C prior to annealing on ice for 10 min. Fluorescent copying of the genomic DNA was accomplished with the Klenow fragment of DNA polymerase I (5 U/µl; Promega, Madison, Wis.). To the DNA mixture was added 6 µl of 10× Klenow buffer (supplied with the enzyme), 3 µl of the dNTP mix described above, 12 µl of ddH₂O, 3 µl of 0.5 mM Cy3-dCTP (Amersham Pharmacia Biotech), and 3 µl of the Klenow fragment of DNA polymerase I. After a static, 2.5-h incubation at room temperature, the labeled DNA probe was purified with a PCR purification kit (Qiagen) before being dried in a speed vacuum.

Amplification of 4,290 E. coli genes. Our amplification method was based on a previously described protocol (31). Specific primer pairs (Sigma Genosys, The Woodlands, Tex.) for each protein-specifying gene of E. coli were used in two consecutive PCR amplifications. Two amplifications were performed to prevent contaminating genomic DNA within the initial PCR product from being spotted to the microarray. Any such carried-over material was eliminated by the "dilution" associated with the second amplification reaction. Genomic DNA (30 ng) was used as the template in the first round of PCR amplification, and 500-fold-diluted PCR products served as templates for PCR reamplification. Duplicate 50-µl scale reactions were performed in the reamplification. The PCRs were catalyzed with ExTaq polymerase (Panvera, Madison, Wis.) with the four dNTPs (Amersham Pharmacia Biotech) present at 0.2 mM and the primers at 0.5 µM. Twenty-five cycles of denaturation at 95°C for 15 s, annealing at 64°C for 15 s, and polymerization at 72°C for 1 min were conducted. A 2-µl aliquot of each PCR product was sized by electrophoresis through agarose gels. More than 95% of these resultant second PCR products displayed visible bands of the correct size. Second-round PCR mixtures, devoid of templates and primers, were saved to be spotted onto slides to serve as negative controls for hybridization experiments. Each second-round PCR mixture was purified with 96-well PCR purification kits (Qiagen). The eluant was dried with a vacuum centrifuge.

Arraying amplified genes. Twenty microliters of 6 M Na₂SCN or 50% dimethyl sulfoxide was added to each dried DNA sample (0.1-ng/nl final concentration). A generation II DNA spotter (Molecular Dynamics, Sunnyvale, Calif.) was used to array the samples onto coated glass slides (Amersham Pharmacia Biotech). Two aliquots of approximately 1 nl from 1,536 resuspended PCR products were arrayed in duplicate on each slide; three slides were used to order all amplified E. coli genes. To serve as controls, 76 specific E. coli PCR products, 8 amplified genes of Klebsiella pnuemoniae, and 12 plant cDNA clones were also spotted onto each slide. Arrayed glass slides, after baking at 80°C for 2 h, were stored in a desiccator at room temperature under vacuum.

Hybridization and washing. Arrayed slides were placed in isopropanol for 10 min, boiled in ddH₂O for 5 min, and dried by passage of ultraclean N₂ gas prior to prehybridization. The prehybridization solution (PHS) was 3.5× SSC (1× SSC is 0.15 M NaCl plus 0.015 M sodium citrate) (BRL, Life Technologies Inc., Gaithersberg, Md.), 0.2% sodium dodecyl sulfate (SDS; BRL, Life Technologies, Inc.), 1% bovine serum albumin (fraction V; Sigma, St. Louis, Mo.). The hybridization solution (HS) contained 4 µl of ddH₂O, 7.5 µl of 20× SSC, 2.5 µl of 1% SDS (BRL, Life Technologies Inc.), 1 µl of 10 mg of salmon sperm DNA per ml (Sigma), and 15 µl of formamide (Sigma). The slides were incubated at 60°C for 20 min in PHS to block nonspecific binding of probe. The slides were next rinsed five times in ddH₂O at room temperature and twice in isopropanol before being dried by the passage of nitrogen. The dried probe was resuspended in the HS and denatured by heating at 94°C for 5 min. Thirty microliters of the probe containing HS was applied to a dried, prehybridized slide, covered with a coverslip (Corning, Corning, N.Y.), and put into a sealed hybridization chamber containing a small reservoir of water to maintain moisture. Hybridization occurred for approximately 14 h at 35°C. Coverslips were removed in washing buffer I (2× SSC-0.1% SDS) warmed to 35°C prior to incubation for 5 min. Next, the slides were washed sequentially for 5 min in 1× SSC-0.1% SDS and 0.1× SSC-0.1% SDS. Slides were then passed through three baths, each passage lasting 2 min, of 0.1× SSC. The slides were dried with a nitrogen gas flow.

Data collection and analysis. Hybridization to each slide was quantified with a confocal laser microscope (Molecular Dynamics, Sunnyvale, Calif.) the photomultiplier tube of which was set to 700 and 800 V for the Cy-3 and Cy-5 signals, respectively. The images obtained were analyzed with ArrayVision 4.0 software (Imaging Research, Inc., Ontario, Canada). The fluorescent intensity (I_i) associated with each spotted gene (i) was reduced by subtracting the fluorescence (N_i) of an adjoining, nonspotted region of the slide. These readings (R_i = I_i  N_i) were exported to a spreadsheet for further data manipulation. The four "no-DNA" spots derived from PCR mixtures devoid of template were controls used to determine the noise (background signal) level.

RNA abundance. To convert normalized equivalent readings into measures of transcript abundance (AB), a further correction was needed. That correction required the hybridization signal arising from an equimolar concentration of all transcripts. The surrogate for this correction factor was the fluorescent intensities arising from hybridization with the fluorescent copy of genomic DNA. Thus, the fluorescent intensities from hybridization with RNA-derived probes were corrected by using fluorescent intensities arising from genomic DNA-derived probes. The abundance of each gene's transcription product(s) was determined by dividing the normalized equivalent reading of a genomic DNA-derived sample into the normalized equivalent reading from the RNA-derived sample (AB = norm ER_transcripts/ norm ER_genome). The convention of Riley and Labedan (32) was followed in grouping genes into functional sets.

-Galactosidase content. -Galactosidase content was measured by the method of Miller using the combined action of chloroform and SDS to disrupt the cell envelope, thus allowing entry of the substrate (23). Two independent cultures of MG1655 were grown in LB medium at 37°C to densities (A₆₀₀s) of 0.43 and 0.47, split, treated with 1 mM IPTG for 15 min, and assayed. The data from these two independent experiments performed on different days were averaged.

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion Appendix References

Array quality. Preliminary experiments (data not shown) indicated that the fluorescent signals obtained with labeled cDNA derived from random priming of E. coli genomic DNA (3 µg) were not saturating and were well within the linear range of the instrumentation. The noise level was determined by averaging the readings of the four control spots derived from PCR mixtures lacking a DNA template after subtracting the background signal derived from an adjacent area that had not been spotted.

IPTG induction. The effect of 1 mM IPTG upon expression of the arrayed genes was investigated. Duplicate RNA preparations of the control and induced cells were each labeled with Cy-3 and Cy-5 by cDNA synthesis. Averaging of measurements was essential for optimal signal detection (Fig. 1). lacZYA induction above the background was detected when the results of a single hybridization experiment in which Cy3-labeled cDNAs derived from treated and control cells were separately hybridized to individual slide sets as viewed in a log-log plot (Fig. 1A). Variation in the measurement of other transcripts was also significant, as indicated by the width of the spread in the data points falling along the diagonal of this scatter plot. Thus, dual-labeling experiments were performed. Improvements were observed by labeling the control sample with Cy-5 and the induced sample with Cy-3 before hybridizing to a single set of three slides (Fig. 1B). However, there was a skewing of the data away from the abscissa (x axis) and towards the ordinate (y axis).

View larger version (19K): [in this window] [in a new window]   FIG. 1.   Analysis of IPTG induction. Basal expression levels expressed as normalized equivalent readings were plotted on the ordinate, and induced levels, also normalized equivalent readings, were plotted on the abscissa. (A) Results obtained when two Cy-3-labeled probes were hybridized to duplicate whole genome array sets. (B) Experiment in which the Cy-5-labeled cDNA copy of control DNA and the Cy-3-labeled copy of induced RNA were coannealed to a single slide set. The RNAs used to generate the results in panel B were each labeled with the other dye to allow a "reciprocal" hybridization. The resulting data were averaged with the data presented in panel B to yield the scatter plot depicted in panel C. A second independent set of RNA samples were isolated, their cDNAs were labeled with both dyes, and the products were hybridized in both possible combinations to generate the results depicted in panel D. (E) Averaged results of the two independent experiments depicted in panels C and D.

Estimate of steady-state transcript levels. The percentage of RNA that programs protein synthesis has been determined under a wide variety of growth regimens (4). Here we estimated the fraction of those protein-specifying transcripts devoted to each arrayed gene. Hybridization signals arising from annealing of RNA-derived Cy-3 labeled cDNA populations were normalized by dividing by the signal generated with Cy-3 fluorescent cDNA arising from copying of sheared E. coli genomic DNA as a probe. Each spot's corrected signal from RNA-derived cDNA hybridization reflected the amount of RNA in the sample. Three RNA samples were thus measured; they were isolated from cells growing exponentially in rich medium, growing exponentially in minimal medium, and cells in minimal medium making a transition from the exponential phase to the stationary phase (for culture conditions, see Materials and Methods). RNAs from certain central metabolic (gapA and ptsH), defense (ahpC and cspC), DNA metabolic (hns), surface structure (acpP, ompACFT, and lpp), translation (rplBCKLMPWX, rpmBCl, rpsACDHJNS, trmD, fusA, infC, and tufAB), transcription (rpoAB), and unassigned (b4243) genes (32) were abundant (>0.1%, among the top 100 transcripts) in all three samples.

Transcripts of exponential-phase cells cultured in rich medium. High-density microarrays were used to measure the transcriptional content of cells growing in rich medium. A total of 1,776 genes were apparently not expressed; their hybridization signals did not exceed the noise by a factor of 2. Of these nonexpressed genes, function has been ascribed to 465. They are listed in the Appendix. Each gene having an assigned function was expressed in at least one of the two tested stages of cultivation in minimal medium.

View larger version (27K): [in this window] [in a new window]   FIG. 2.   Distribution of expressed genes. The histogram plots number of genes as a function of expression range. Diagonally striped, solid, and horizontally striped bars reflect distributions observed in RNAs derived from cells growing exponentially in minimal medium, cells transitioning to the stationary phase in minimal medium, and cells growing exponentially in rich medium, respectively. Expression of 766, 1,030, and 1,776 genes was not detected under the three respective conditions.

View larger version (18K): [in this window] [in a new window]   FIG. 3.   Fractional expression. The extent of ORF transcripts is plotted as a function of genes summed. The order in which genes were summed was based upon expression level, with the most highly expressed gene summed first.

Transcripts of exponential-phase cells cultured in defined minimal medium. The gene expression pattern of cells growing exponentially in minimal medium was also examined. At this cell density, the pH remained at 7.0. Expression levels varied from 0.001 to 0.7%. Apparently, as illustrated in Fig. 2, biosynthetic requirements mandated that a significantly greater fraction of the genome was expressed in minimal medium. Only 776 genes were expressed negligibly (signal/noise ratio of <2); 149 of these genes have presumed or demonstrated function. They are listed in the Appendix.

Transcripts of cells making a transition from the exponential phase to the stationary phase in defined minimal medium. During this transition, at a point where the pH had dropped slightly to 6.7, significant changes in gene expression were expected and observed. Expressed gene levels were from 0.0023 to 1.6%. A total of 1,030 genes, of which 110 have a defined role (Appendix) did not appear to be expressed at this transitional phase of growth.

Compilation. The observed expression patterns are summarized in Table 5, where gene products were grouped by metabolic function according to an established classification scheme (32). Exponential growth in minimal medium elevated the amount of pyrimidine and amino acid biosynthetic transcripts with respect to growth in the rich broth, LB. In contrast, cofactor and purine transcripts did not appear to accumulate relative to growth in LB broth. Expression of glyoxylate shunt and other glucose metabolism-related transcripts was also elevated in minimal medium; the seven-fold elevation of glyoxylate shunt transcripts exceeded the average of that observed for amino acid biosynthetic mRNAs. Expression of genes involved in sulfur fixation was also elevated during growth in minimal medium.

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion Appendix References

Comprehensive expression profiling has been performed previously with the yeast Saccharomyces cerevisiae (8). Here we report that such profiling can also be accomplished and refined by using a dye-swapping, high-density microarray technology when the prokaryote E. coli is used as an experimental system. Adaptation of RNA isolation and labeling protocols from eukaryotes to prokaryotes is not straightforward, because eukaryotic mRNA manipulations often exploit the specific 3'-polyadenylation of this molecular species. The short half-life of bacterial mRNA is another obstacle. We chose to isolate RNA by a standard procedure that included a centrifugation step and to reverse transcribe bulk prokaryotic RNA to prepare our hybridization probe. Thus, the reported measurements are subject to possible systematic errors, including differential mRNA stabilities (18). Despite the large amount of stable RNA in the sample, hybridization to protein-encoding genes was readily detected. Recently, independent studies of E. coli (31, 34) successfully applied nylon-based medium-density DNA array or glass-based high-density DNA array technologies to assess gene expression changes in response to growth medium and heat shock.

Nonetheless, errors could be introduced in the many steps from RNA purification to analyses of hybridization signals. As shown in Fig. 1, conditions have been optimized to yield highly reproducible data. The scatter plot of the optimized protocol (Fig. 1E) illustrated that measurements of gene expression were still subject to considerable variation when the signal was in the lowest part of the detectable range. It was found that expression of only eight genes was effected by IPTG treatment; all were induced. It was reassuring that the expected lacZYA induction was observed; the significance of the weaker inductions awaits confirmation by complementary techniques, perhaps based upon enzyme assay of gene fusions or analysis of an isogenic pair of strains differing in lacI. Such a correlation has been provided for IPTG induction of melA expression (31). Thus, such comprehensive gene expression profiling generates hypotheses requiring further study for verification.

Having developed confidence in the technology, it was applied to monitoring expression as a function of growth stage and medium. For these experiments, normalization of signal intensity was essential. Probe, derived from replication of genomic DNA and used as a replica of equimolar transcription of the entire genome, allowed calculation of mRNA inventories. Thus, we have provided measures of steady-state transcript levels under prescribed sets of conditions rather than the fold change in mRNA titers that represents the difference in expression between two conditions. These inventories were satisfying in several ways. First, the most highly transcribed genes in actively growing cells cultured in LB medium often encoded proteins involved in translation. In contrast, cultures at a similar growth stage in glucose minimal medium expressed to a very high level several small molecule biosynthetic genes and the means to utilize glucose.

Thus, agreement between this molecular analysis and the accumulated understanding of E. coli physiology (24, 25) was observed. This agreement was underscored in the analysis of cells making a transition from the exponential growth phase to the stationary phase; the elevated expression of several rpoS-controlled genes corresponded to expectations. Nonetheless, some caution is necessary; potential effects of differential mRNA stability (18) have yet to be considered.

It is most unlikely that the technology is limited to highly expressed genes. First, reproducible expression measurements were obtained over a wide dynamic range (Fig. 1E). Second, the data from Fig. 3 and Table 1 illustrate that lac operon expression, although low before IPTG induction, was detected, suggesting that most transcripts can be readily measured by the techniques described. The lower limits of expression that can be observed may be defined by the analyses of well-characterized "promoter-down" mutants (30) or "spiking" experiments. In the latter, the templates for cDNA synthesis would contain constant amounts of total RNA derived from a deletion mutant to which various quantities of a corresponding transcript synthesized in vitro has been added.

Compiling of data (Table 5) into functional groups (32) has become one method for the analysis of gene expression profiles. Work such as that presented here indicates that these categories do not respond as a bloc; rather subsets act differently, as has been observed for amino acid biosynthesis during a study of yeast (14). Nonetheless, such analyses allow global trends to be observed and the integration of gene expression patterns with the cell's overall physiological status. The histogram (Fig. 2) allows one to appreciate the quantity of transcripts that falls within each expression range. Figure 3 provides an indication of how transcriptional capacity is distributed under the three distinct conditions that were examined.

Moreover, unexpected results worthy of further study are found within the compilation presented in Table 5. Unlike transcripts dealing with energy transfer, ribosomal proteins, translation and aminoacyl-tRNA formation which were elevated in LB broth-grown cells, the sum of mRNAs specifying cell division proteins did not vary under the three conditions that were investigated. In a similar vein, hybridization to genes involved in the synthesis of the cell envelope was not increased when the probe was derived from cells cultured in LB broth.

In such global analyses, the reliability of the data obtained is an issue. The method described here makes four measurements of the transcript level present in each RNA sample. Moreover, the organization of genes into operons provides internal benchmarking of the measurements. Analyses of transcripts from ribosomal protein, biosynthetic, and PTS operons (Tables 3 and 4) suggests that the data are of high quality, since the range of the measurements for each operon is rather small. Such an analysis is consistent with the methodological improvements illustrated in Fig. 1.

Thus, an initial mRNA inventory was compiled. We believe that our analyses are subject to errors in measures of transcripts smaller than about 300 nucleotides and failures in PCR amplification. The compilation illustrated several physiological points. The protein biosynthetic demand during growth in rich medium was noted, accounting for about 15% of the polypeptide-specifying transcripts; what may be as significant is that more than one-quarter of all protein-specifying transcripts under any of the measured conditions lack a functional assignment. Consequently, gene expression profiling provides a further impetus to the continued study of E. coli.