To construct target-specific screening strains, we amplified a kanamycin-resistance gene cassette flanked by loxP sites^{16, 17} by PCR using primers with long tails (50 base pairs). The tails share homology with regions immediately upstream and downstream of an essential target gene (Fig. 1D). Direct transformation of the PCR product into cells expressing redgam followed by kanamycin selection results in replacement of the target gene with the lox²–kan cassette. When the target is essential for bacterial growth, 100% of the transformants show the appropriate deletion only if the plasmid copy of the gene is induced (data not shown). PCR confirmed the location of the allele exchange on the chromosome (Fig. 1E, lanes 7 and 10).

Using the site-specific recombinase from bacteriophage P1, we removed the kan cassette from deletion strains. We transiently expressed the P1 Cre protein (cloned into a low-copy-number, temperature-sensitive plasmid) in cells, and subsequently eliminated the plasmid by incubation at 37°C. The result was excision of the antibiotic-resistance marker, leaving a single lox site in its place¹³. Replica plating and PCR confirmed loss of kanamycin resistance (Fig. 1E, lanes 8 and 11).

In the last step before screening, we removed the temperature-sensitive prophage from each strain. ^- recombinants were selected at 42°C following transformation of a 4 kilobase PCR fragment generated from an att⁺ strain (Fig. 1C). PCR (Fig. 1E, lanes 4 and 5) and phage plating (data not shown) confirmed removal of the cI₈₅₇ prophage. A partial list of target-specific strains engineered by this methodology is shown in Table 1.

Characterization of hypersusceptible strains. Reducing expression of an essential gene should affect a cell's growth characteristics and render it hypersusceptible to specific inhibitors. As a first test of this hypothesis, we analyzed the growth rate of the engineered strains in microtiter plates. A matrix of conditions was developed to monitor the effects of arabinose concentration, inoculum size, and inhibitor concentration. Results in Figure 2 show that both metG and fabI strains were similarly affected by lowering arabinose concentrations. (A downward arrow () directly preceding a gene name indicates a bacterial strain in which expression of that gene is under inducible control.) Growth defects appeared as a delayed escape from lag phase followed by exponential growth comparable to the reference strain. In fact, doubling times at low and high inducer concentrations were nearly identical (2.5–3.0 h). At 65 M arabinose, exponential growth of the fabI strain began only after an 8 h delay. In contrast, the metG strain showed a similar delay, but at 420 M arabinose. The conditions necessary to demonstrate significant growth defects were unique for each strain and required experimental determination (Fig. 2).

At the lowest inducer concentrations, growth rate drops off significantly for both the metG and fabI strains. Doubling time for the metG strain increased from 3 h (at 420 M arabinose) to 5 h at 70 M arabinose. A nearly identical increase in doubling time was seen for the fabI strain (3 h at 65 M, 5 h at 6.5 M).

For the murA strain, reducing arabinose did not lengthen the period before exponential growth until the lowest concentrations of inducer were tested. There was, however, a slight reduction in growth rate at 1.3 mM arabinose (3.7 h compared with 3 h for the reference strain). Lag phase was extended only after a tenfold reduction in inducer concentration. Lowering arabinose concentration to 13 M affected both lag and exponential growth (as seen for fabI and metG). When cultures were prepared under optimal conditions, none of the engineered strains were able to grow in the absence of inducer (Fig. 2).

From these data, we chose growth conditions for screening. We tested each target-specific strain for hypersusceptibility to known antibacterial agents by monitoring changes in the minimum inhibitory concentration (MIC; Table 2). murA is the target of the cell wall biosynthesis inhibitor, fosfomycin¹⁸. Arabinose-dependent downregulation of murA shifts the MIC for fosfomycin at least eightfold to 0.03 g/ml (as compared with the reference strain). High expression of MurA results in a >32-fold increase in the fosfomycin MIC (Table 2). Other antibiotics (for instance, norfloxacin and trimethoprim) do not exhibit such a shift. Likewise, the fabI and tufA strains demonstrate an MIC shift only with their cognate inhibitors: triclosan¹⁹ and kirromycin²⁰, respectively. For the metG strain, no shift in MIC was detected upon testing the strain against commercially available inhibitors of bacterial growth (Table 2).

Parallel high-throughput screening (HTS). We screened the reference strains against a compound library in 96-well microtiter plates using 100 l Luria–Bertani (LB) broth containing 1% dimethyl sulfoxide at 37°C. Plates were read by absorbance at 600 nm after overnight incubation. Hits scored positive if the A₆₀₀ value was <0.1, and no visible growth could be detected in a well. Figure 3 shows the results from screening a reference strain against a minor fraction of compounds from our library. Three compounds from this set were able to prevent the growth of the E. coli reference strain (green circles). Upon screening the same set of compounds against the murA and metG strains, we identified the same three compounds (Fig. 3). In addition, four hits were found that were specific for the murA strain (pink triangles). Most striking is the identification of five new hits specific only to the metG strain (orange squares). These compounds were unable to prevent growth of either the reference strain or the murA strain.

Hits from the primary screens were retested for antibacterial activity using the engineered strains grown under different concentrations of inducer. If a hit is target-specific, upregulation of the target (by increasing inducer concentration) should render the bacteria less sensitive to the compound. Data from such a secondary screen appear in Table 3. We determined the absolute MICs for five compounds identified as hits in HTS of the murA strain. When the screening strain was grown at high inducer concentrations, the MICs for the compounds increased 16-fold for compound 1, 8-fold for compound 2, and 6-fold for both compound 3 and compound 4. Only a marginal shift in MIC is recorded for compound 5. For comparison, fosfomycin shows an increase of 100-fold under these same conditions. No change in the MIC for norfloxacin was recorded. When tested in an in vitro assay, these five compounds also inhibited the enzymatic activity of MurA (Table 3). IC₅₀ values for the compounds ranged from 1.4 to 6.2 M. The IC₅₀ of fosfomycin is 0.2 M (Table 3).

DISCUSSION

We describe here the development and application of a screening strategy for antimicrobial drug discovery. This approach combines the attributes of more traditional discovery approaches^21-23 into a simple, efficient, and high-throughput methodology that incorporates target specificity and antimicrobial activity into a primary screen. The cornerstone of this approach is the ability to generate an array of bacterial strains engineered for low-level expression of a particular essential gene. Each strain is then screened against a chemical library, using growth inhibition as an end point.

Strain construction was facilitated by the use of the recombination systems of bacteriophages and P1. The -redgam system is highly efficient, and allows the use of linear DNA molecules as recombination substrates^{24, 25}. This avoids cumbersome genetic manipulations, such as repetitive plasmid cloning steps and the use of counterselectable markers to resolve recombination intermediates. In fact, recombination mediated by -redgam was highly reliable, often yielding 100% recombination efficiencies (data not shown). The portability of the redgam prophage was highly advantageous. It allowed the use of plasmid-based systems for controlled expression of target genes, providing flexibility with regard to incompatibility groups and selectable markers.

Removal of the prophage from the chromosome of each engineered strain prevented functions from interfering with screening (Fig. 1D). During HTS at 37°C, expression of -kil and other potentially toxic products should be repressed by the repressor, cI. Temperature variations can, however, occur in incubators harboring large numbers of multiwell plates. A 2°C increase in temperature (to 39°C) is known to induce the cI₈₅₇ temperature-sensitive allele²⁶. Killing as a result of prophage expression and the potential of false positives during screening necessitated removal of the prophage before HTS.

Expression of essential target genes was controlled by the well-characterized arabinose regulon²⁷. We chose episomal expression of target genes as a general strategy (Fig. 1A). Fusion of P_BAD to coding sequences on the chromosome was avoided because the majority of bacterial genes are found in operons. Such a strategy might subject flanking genes to arabinose control and misregulation, calling into question the specificity of any active compounds identified during HTS.

Plasmid-based expression has its advantages. Using plasmids not only precludes interference with expression of neighboring genes, but also provides an additional level of expression control—either by using plasmids with different copy numbers or by using mutants in nonessential host genes that alter plasmid copy number. In fact, the pcnB^- host was required to render murA hypersusceptible for HTS (Figs 2, 3; Table 2). One drawback of using episomally expressed genes may be sensitivity to inhibitors of plasmid replication. Such inhibitors would show up as false positives during HTS, but would be easily sorted out by secondary testing under high-expression conditions (Table 3). In addition, inhibitors of plasmid replication would consistently score positive during the screening of many hypersusceptible strains, and would be easily identified by a routine analysis of the array HTS database.

As an additional precaution against polar effects and misregulation, we removed the antibiotic-resistant cassette used for allelic exchange from the chromosome. Presence of the kan cassette had the potential to alter the expression of neighboring genes and result in off-target effects. The lox sites flanking the selectable marker easily prevented such an occurrence: after recombination by Cre, the kan cassette was removed from the genome, leaving only the small loxP site (Fig. 1D, E).

Strain characterization before HTS revealed that low arabinose concentrations extended the time required for cells to initiate exponential growth. One interpretation of this phenomenon may be related to the concentration of an essential protein required for bacterial growth. Although a cell is viable under low-expression conditions, it seems that accumulation of an essential gene product must reach some critical level for the cell to begin normal growth and division. At 13 M arabinose, the fabI strain required 9 h to attain that level, whereas the murA strain required 12 h (Fig. 2). Upon reaching this level, strains maintained normal growth rates unless expression was further reduced. Gene-specific factors such as the timing of induction and expression, translation rate and efficiency, and mRNA and protein stability likely play a role in the observed differences between strains.

It is precisely this result that permits the parallel screening strategy to be successful. Compared with the reference strain, an engineered strain in culture at some low arabinose concentration will require more time to reach saturating levels because of the limiting amounts of a single essential gene product. Should even a weak, but specific, inhibitor of that product be present in the culture medium, the number of active target molecules will be further reduced and result in a further growth delay. This lack of growth within a fixed time period (compared with the reference strain cultured under identical conditions with the same inhibitor) registers as a hit during HTS.

This can be demonstrated with known inhibitors and their targets^{28, 29}. The murA strain has increased sensitivity to the specific inhibitor fosfomycin (Table 2). Inhibitors of the later steps in the cell wall biosynthesis pathway (ampicillin and bacitracin) did not show increased activity against the murA strain. This attests to the specificity and sensitivity of the screening approach.

HTS of these hypersusceptible strains identified new lead compounds that would not have been found through traditional cell-based screens (Fig. 3; Table 3). In parallel screening, compounds that inhibit the reference strain are temporarily excluded. At lower concentrations, these compounds can be rescreened against the strain array in an attempt to identify their mechanisms of action. Conversely, the same compounds can be tested against the strain array when each target is upregulated, scoring resistance as a hit. In fact, retesting the hits from the murA screen at higher expression levels shows differences between or within the lead classes (Table 3). Confirmed by biochemical analysis (Table 3), these results suggest that preliminary structure–activity relationships may be defined that can help identify chemically attractive candidates early in the discovery process.

In addition to well-characterized essential gene targets, new targets identified through recent advances in genomics and bioinformatics can also be evaluated by this screening paradigm⁹. Essentiality testing of gene targets occurs simultaneously as a strain is engineered for HTS. Any newly defined essential target can then be screened for inhibitors, even in the absence of a defined biochemical activity (Table 1).

In summary, the advantages of using a strain array for antibacterial discovery are that (i) the strategy can be used for any target, including proteins that are difficult to assay in vitro, (ii) compounds identified in the screens will have antibacterial activity, (iii) the molecular target of an inhibitor of a hypersusceptible strain can be inferred or easily confirmed with a functional assay of the biochemical target, (iv) the screen for each different target is run with the same format and conditions, thereby streamlining the HTS process, (v) engineered strains are permanent tools that can be screened at any time and at low cost, and (vi) data collected from several screens of various targets may reveal compounds that inhibit more than one target simultaneously.

Experimental protocol

Reagents. Antibiotics were purchased from Sigma Aldrich (St. Louis, MO). For the purpose of duplicating experiments described in this paper, compounds 1–5 are available from Bristol-Myers Squibb Co. (New York, NY).

Bacterial strains and phage. E. coli DH5 was used for plasmid manipulations. Strains N99, MG1655, and W3110 were used as sources of genomic DNA and as controls in phage plating experiments. E. coli DY329 (W3110 lacU169, nadA:Tn10, gal490, cI₈₅₇ [cro-bio]) was a gift of Don Court²⁴. It contains a defective prophage expressing the gam, bet, and exo recombination functions from the cI₈₅₇-regulated pL promoter. cI^- and vir (gifts of Bob Weisberg) were used to test for the presence and absence of the defective prophage.

Plasmids. Essential target genes (together with their natural Shine-Dalgarno signals) were cloned for arabinose-regulated expression into the vector pBAD18 Cam¹⁴ using standard techniques³⁰. The gene for P1 Cre recombinase (along with its natural promoters, pR2 and pR3) was amplified from pRH104 (ref. 31) and cloned into a low-copy-number plasmid with a temperature-sensitive origin of replication (pJDP8).

Linear DNA substrate design and selection of recombinants. Construction of the nad⁺ derivative of strain DY329 was accomplished through PCR of the nadA region on the chromosome of W3110 using the primers 5'-TCCTGCACGACCCACCACTA-3' and 5'-CCGCCTGCCCCTATTGGTAT-3'. After purification, 100 ng of product was electroporated into E. coli strain DY329 in which expression of -redgam had been induced²⁴. Selection for nad⁺ prototrophs and screening for sensitivity to tetracycline were carried out following standard methods³².

The prophage was removed by linear DNA transformation of redgam-induced cells with a 4.1 kb PCR product containing the phage attachment site, followed by selection at 42°C. Primer sequences for attB amplification were 5'-CGAACCGTAGGCCGGATAAG-3' and 5'-GGCGAGGTGTCCAG-GTTG-3'. Temperature-resistant survivors were also tested for phage growth at 37°C (ref. 33) and for biotin prototrophy. The ara– and pcnB– derivatives of DY329 were constructed using a similar strategy.

Linear DNA substrates for deletion of target genes from the chromosome were synthesized by PCR using the loxP-kanamycin (loxP-kan) cassette from pRH43 (ref. 16) as template. The following primers were used: N₅₀-TATCACGAGGCCCTTTCGTCTT-3' and 5'-N₅₀-TTTTCACCGTCATCACCGAAAC-3', where "N" represents nucleotides composing the homology arms necessary for recombination. A 2.5 kb PCR product was generated from loxP-kan using primers with long tails (approximately 50 bp in length, represented by the curvy lines in Figure 1D) that share sequence homology with the regions of the chromosome flanking a target gene (also represented by curvy lines). This double-stranded DNA is used as a substrate for recombination in cells expressing -redgam via the homology tails. Kanamycin-resistant transformants are isolated in the presence of a plasmid containing the complementary copy of the target gene under arabinose control (Fig. 1B).

The kanamycin-resistant marker was removed by transformation with the temperature-sensitive Cre plasmid followed by selection on spectinomycin at 34°C. After incubation at 37°C, colonies were replica-plated on LB plates with and without kanamycin and grown at 37°C overnight. Kanamycin-sensitive colonies were screened by PCR to demonstrate loss of the marker (Fig. 1E). All chromosomal rearrangements were screened by PCR and confirmed by DNA sequencing (Fig. 1E). Primer details are available upon request.

Characterization of engineered strains. Growth analysis under different inducer concentrations was conducted in a Bioscreen C microbiology reader from Thermo Lab Systems (Helsinki, Finland). Cultures from each strain were grown overnight in medium containing various concentrations of arabinose. These served as seed cultures for testing growth in microtiter plates under different conditions, such as inoculum size and inducer concentration. After monitoring these and other parameters, culture conditions under which the strains would become limiting for the essential gene target were determined. Before screening, mutant strains were also tested for sensitivity to antibiotics and drugs with known mechanisms of action to assess the target specificity of each strain. For such experiments, standard MIC determinations were done³⁴.

MurA biochemical assays. The activity of the MurA enzyme was measured by quantitating the UDP-GlcNAc-dependent release of inorganic phosphate from phosphoenolpyruvate according to published methods³⁵.

Received 1 February 2002; Accepted 15 February 2002.