MGK simultaneously tracks the relative abundance of individual mutants in gene inactivation libraries grown in a reference and experimental condition. This is achieved by hybridizing polymerase chain reaction (PCR)-amplified flanks of the inactivated genes to specifically designed DNA microarrays (Figure 1). The approach utilizes random or defined gene knockout libraries, in which individual genes are inactivated by either transposon insertion or gene replacement (kanamycin resistance [Km^r] cassette in Figure 1).

The mixed library of knockout mutants is grown in a reference and selective condition for several generations. Genomic DNA isolated from each population serves as a template in a primer extension reaction. The chromosomal regions flanking the gene replacement cassette or sites of the transposon insertion are linearly amplified by repeated rounds of time-controlled extension of biotinylated primers specific for the Km^r cassette (Table 1). The yield of amplified flanks of the Km^r cassette corresponds to the relative abundance of individual gene knockout mutants in the population. Streptavidin-coated magnetic beads are used to isolate the biotinylated flanks, which are polyadenylated at the 3'-ends using terminal deoxynucleotidyl transferase. The flanks are then exponentially PCR amplified using a nested Km^r cassette-specific primer and an oligo-dT primer, yielding 'MGK targets'. At this step, amino-allyl modified dUTP is incorporated into the MGK targets and subsequently conjugated with fluorescent dyes. A mixture of labeled targets is then hybridized to a custom designed oligonucleotide microarray, and the relative abundance of individual mutants present in the library after growth in the reference and selective conditions is assessed. Once a preliminary list of conditionally essential genes is generated, the phenotypes of individual mutants (either picked from the arrayed gene inactivation defined libraries or engineered de novo in the case of random transposon insertion libraries) is verified.

The design of the DNA microarray for MGK depends on the type of mutant library being analyzed. For the E. coli random transposon insertion library employed in this study, a microarray was designed to contain unique oligonucleotide sequences (34-mer, on average) spaced approximately every 500 base pairs (bp) in the E. coli genome. As a result, each gene knockout was represented by one to three probes. For the E. coli defined deletion mutant library, 34-mer oligonucleotide sequences were selected from a region about 100 bp upstream and 100 bp downstream of each gene, so that each knockout was represented by two flanking probes. For clarity, the random transposon library and defined deletion library used in this study are referred to as 'random library' and 'defined library', respectively.

The aromatic amino acid biosynthesis pathway has been well characterized in E. coli 2. Decades of painstaking experiments have identified 18 genes that are involved in the production of aromatic amino acids when they are not readily available in the environment (Figure 2a). Thirteen genes belonging to this pathway encode nonredundant enzymes and are expected to be essential for cell growth in medium lacking phenylalanine, tryptophan, and tyrosine. To evaluate the applicability of MGK for identification of conditionally essential genes, we used MGK to identify mutants (in both a random and a defined library) that are unable to grow in medium lacking aromatic amino acids. The E. coli random library of about 1.2 × 10⁵ mutants was generated using random mini-Tn10 transposon mutagenesis 3. The defined library consisted of 3,985 E. coli gene replacement mutants 4; mutants in this library were mixed at equal ratio (see Materials and methods, below). In addition to demonstrating the flexibility of the method, the use of two types of libraries provided an opportunity to test the versatility of MGK and assess the extent to which mutant representation affected the sensitivity of the MGK screen.

For the MGK selection, libraries were grown for 10 generations in defined medium either containing or lacking aromatic amino acids (see Materials and methods, below, for details). MGK targets were prepared from each library and hybridized to corresponding microarrays. Experiments were performed twice, with dye swapping (correlation coefficient between two experiments was 0.84 for the defined library and 0.93 for the random library). (For the entire set of microarray raw data and intensity ratios, see Additional data files 1 and 2.)

Using the cut-off criteria described in Materials and methods (below), eight genes were identified as putatively essential for E. coli growth in the absence of aromatic amino acids in the random library, and 37 genes were identified from the defined library (Table 2). As mentioned above, there are 13 genes whose inactivation is expected to cause aromatic amino acid auxotrophy in E. coli 2. All 13 of these genes were among the 37 genes identified in the MGK screen applied to the defined library, whereas five of the anticipated 13 auxotrophic mutants were among the eight genes found in the random library (Figure 2a). This finding demonstrates that MGK can successfully be applied to both types of libraries but that it provides a more complete dataset when it is used with the defined library.

Because several of the genes identified by MGK (three from the random library and 24 from the defined library) were previously unknown to be important for aromatic amino acid biosynthesis, the phenotypes of these gene deletion strains were tested. The disruption of epd and pdxA, found in the defined library, did cause a growth defect in medium lacking aromatic amino acids (Figure 2b). The encoded enzymes are involved in biosynthesis of pyridoxine (vitamin B₆), which is an essential co-factor of transaminase steps in the aromatic biosynthesis pathway 5. Finding these genes in our MGK screen was not surprising because the defined medium used in this study lacks vitamin B₆. Indeed, growth of the epd and pdxA mutants was restored to wild-type levels in the medium supplemented with vitamin B₆ (data not shown). Three other mutants, namely tktA identified in the random library, and rpe and ygdD found in the defined library, also exhibited reduced growth in medium lacking aromatic amino acids (Figure 2b). The encoded enzymes TktA (transketolase) 6 and Rpe (ribulose phosphate 3-epimerase) 7 are both involved in sugar phosphate interconversion in the nonoxidative branch of the pentose phosphate pathway; the function of YgdD is unknown. Although it is not clear why disruption of these genes reduces cell growth in the absence of aromatic amino acids, the phenotypes of these mutants confirm that they were legitimately identified by MGK. Disruption of the rest of the genes recovered in MGK screens (one from the random library and 20 from the defined library) caused no growth defect in the absence of aromatic amino acids, suggesting that either they were false-positive hits or that our conditions for testing individual mutants did not adequately reproduce the selection pressure experienced by mutants in the mixed library culture.

The results of this model MGK screen demonstrate that the method is well suited to genome-wide identification of conditionally essential genes. In addition, the comparison of results obtained from two libraries shows that the use of a defined library in which mutants are well represented increases the sensitivity of MGK screen.

Bacteriostatic antibiotics inhibit cell growth but they do not significantly decrease the number of viable cells. Proteins whose genetic knockout leads to bacterial cell death upon treatment with bacteriostatic antibiotics may serve as new targets for drug potentiators and may provide important insights into mechanisms of bacterial response to antibiotic stress. Identification of such mutations generally requires the near impossible task of plating thousands of mutant cultures onto multiple plates after exposing them to bacteriostatic antibiotics.

MGK provides a much better way to identify such mutants. As a proof of concept, we used MGK to identify genes required for survival of E. coli during challenge with chloramphenicol, which is a classic bacteriostatic antibiotic that prevents bacterial growth by interfering with the activity of the ribosomal peptidyl transferase 8. The pooled defined library was exposed for two consecutive rounds of 18-hour incubations in the presence of chloramphenicol (80 μg/ml, which is ten times the minimum inhibitory concentration). MGK targets prepared from libraries with or without chloramphenicol selection were hybridized to the microarray. (For entire set of microarray raw data and intensity ratios, see Additional data file 3.)

We identified 35 genes that exhibited at least threefold reduced signal intensity after cells were exposed to chloramphenicol (Table 3). Some of the identified genes were known to be co-transcribed within an operon (pstC and pstS; ptsH and ptsI; sufB and sufD; and tolQ, tolR, and tolA), or their gene products constituted a functional pair (arcA and arcB). We verified the phenotypes by testing survival of 29 individual mutants upon chloramphenicol treatment (among these 29 mutants, each of the aforementioned operons were represented by one mutant). Of these, 12 mutants exhibited more than fivefold reduced number of viable cells after exposure to chloramphenicol, and therefore they carried deletions of genes that are critical for survival of bacteria upon treatment with a bacteriostatic antibiotic. In comparison, the viable cell count of the wild-type was not affected by chloramphenicol (Figure 3).

The functional categories of the verified 12 genes varied widely, including peroxide detoxification (AhpC), redox regulation (ArcA), proteolysis (ClpP and Prc), membrane integrity (Lpp), and transport (TolQ, OmpA, and YbeX). From this diverse set, we can only tentatively rationalize the importance of a few genes for cell survival upon antibiotic treatment (see Discussion, below). This finding underscores the advantage of an unbiased global gene-screening technique such as MGK for identifying potential new drug targets as well as targets for drug potentiators. Taken together, the results presented here demonstrate the power of MGK for identifying loss-of-function mutations in complex mutant libraries.

MGK microarray target preparation includes six steps: preparation of genomic DNA, linear amplification of single-stranded DNA flanking the gene-inactivation cassette, separation of single-stranded flanks from genomic DNA, polyadenylation of the 3'-ends of single-stranded flanks, PCR amplification of the DNA flanks, and fluorescent dye conjugation. Individual steps are described below in detail. Parameters optimized for MGK microarray target preparation are shown in Additional data file 6.

CombiMatrix™ DNA microarrays (CombiMatrix Corporation, Mukilteo, WA, USA) were custom designed for the detection of knockout mutants in the defined or the random library. For the defined gene-replacement library, microarray probes represented the chromosomal regions adjacent to the deletion cassette. To select these probe sequences, CombiMatrix™ DNA microarray design software scanned for 32-36-mer sequences within 100 bp regions upstream and downstream of each gene 27, and a total of 8,942 probes was synthesized on the microarray. For the random transposon insertion library, the microarray contained 11,579 unique 32-36-mer oligonucleotide probes, spaced approximately every 500 bp in the E. coli genome. Both types of microarrays also contained about 500 negative control probes corresponding to Arabidopsis thaliana, Agrobacterium tumifaciens, and phage lambda DNA sequences. Additional data files 1 and 2 contain information about the microarray design and oligonucleotide sequences of probes for each library.

CombiMatrix™ microarrays were hybridized with 5 μg each of differentially labeled target DNA for 20 hr at 50°C in a rotating oven. All hybridization and washing steps were performed in accordance with the manufacturer's protocol.

Microarrays were scanned using a PerkinElmer confocal dual-laser microarray scanner equipped with ScanArray Lite software (PerkinElmer, Boston, MA, USA). CombiMatrix Microarray Imager™ analysis software was used to obtain raw signal intensities. After background subtraction, intensity values were initially normalized on the basis of individual contribution to the total intensity of the channel. These values then underwent a second normalization based on contribution to intensity within a range of 50 probes upstream and downstream of each probe using custom designed Python software (available upon request). This second, 'sliding scale' normalization method accounted for the localized variation in DNA copy number caused by the varying rates of chromosome replication between the two conditions 2829. Resulting signal intensities for probes were used to calculate the intensity ratios (values in reference/values in selection). Intensity ratios were considered significant if they were greater than or equal to 3 in two independent experiments, with dye swapping.

The defined collection of E. coli gene replacement mutants 4 was obtained from Hirotada Mori (Nara Institute of Science and Technology, Nara, Japan). This collection is comprised of 3,985 E. coli deletion strains derived from wild-type BW25113 (F^- λ^- rph-1 ΔaraBAD_AH33 lacI^q ΔlacZ_WJ16 rrnB_T14 ΔrhaBAD_LD78 hsdR514). In each deletion strain, the coding region (except for seven codons at the carboxyl-terminus) of a nonessential gene is replaced by in-frame insertion of a kanamycin resistance gene 30. For application of MGK, individual deletion mutants were grown in 96-deep-well plates overnight at 37°C to an optical density (OD) of about 1.3 at 600 nm in Luria-Bertani (LB) medium containing 30 μg/ml kanamycin. An equal volume of each culture was combined. Cells were harvested, washed with LB, and re-suspended in LB supplemented with 15% glycerol. Aliquots containing about 1 × 10⁹ cells in 500 μl were frozen and stored at -80°C.

The random transposon insertion library was generated using mini-Tn10 transposon mutagenesis. The suicidal transposon delivery vector pBSL177 was used, which contains a Tn10 transposon that harbors a kanamycin resistance marker 3, and a mutant transposase with altered target specificity controlled by an isopropyl-beta-D-thiogalactoside (IPTG) inducible tac promoter 31. We first generated a random library of about 1 × 10⁵ transposon mutants in E. coli cells grown in LB, but transposon insertions were severely biased toward chromosomal regions close to the origin of replication (data not shown). Therefore, in order to reduce this bias, transposition was induced in slowly grown cells (see Additional data file 1). Electrocompetent wild-type E. coli MG1655 cells were prepared from a culture grown in M9 minimal medium supplemented with 0.2% sodium acetate and electrotransformed with the pBSL177 plasmid (1 μg; 1.7 kV, 200 Ω resistance, and 25 μF capacitance). Cells were diluted with 1 ml SOB medium (2% tryptone, 0.5% yeast extract, 10 mmol/l NaCl, 2.5 mmol/l KCl, 10 mmol/l MgCl₂, and 10 mmol/l MgSO₄) containing 1 mmol/l IPTG, incubated with shaking at 37°C for 30 minutes, and plated on LB agar containing 30 μg/ml kanamycin. The next day, about 1.2 × 10⁵ transformants were pooled, washed with LB, and re-suspended in 15% glycerol LB medium. Aliquots containing about 2 × 10⁹ cells in 500 μl were frozen and stored at -80°C.

All incubations were performed at 37°C with shaking. A glycerol stock of the mixed library (random or defined) was inoculated to OD₆₀₀ 0.02 into Neidhardt supplemented MOPS-defined medium 32 containing aromatic amino acids (0.4 mmol/l phenylalanine, 0.1 mmol/l tryptophan, and 0.2 mmol/l tyrosine) and grown overnight. Cultures were then diluted 100-fold in the same medium and grown to OD₆₀₀ 0.4. Cells were washed in 1× MOPS-Tricine buffer (pH 7.4; Teknova, Hollister, CA, USA), re-suspended in the same buffer, and inoculated into two flasks: one containing Neidhardt supplemented MOPS-defined medium complete with aromatic amino acids and the other containing the same medium but without aromatic amino acids. The starting density of the cultures was OD₆₀₀ 0.002. Cultures were grown to OD₆₀₀ 2, at which point the cells were collected and genomic DNA was isolated. The same conditions were applied for validation of individual mutant strains from the defined mutant library.

Minimum inhibitory concentration of chloramphenicol for wild-type BW25113 cells was determined to be 8 μg/ml. For chloramphenicol selection, a glycerol stock of the mixed defined library was inoculated to OD₆₀₀ 0.02 in 200 ml LB and grown to OD₆₀₀ 0.2, at which point the culture was split into two flasks. One of the flasks was supplemented with 80 μg/ml chloramphenicol, whereas the other flask served as a control. The chloramphenicol-containing culture was incubated for 18 hours with shaking at 37°C. The cells were harvested, washed with 100 ml LB, re-suspended in 100 ml LB at OD₆₀₀ 0.02, and grown to OD₆₀₀ 0.2. At this point chloramphenicol was again added at 80 μg/ml, and the culture was incubated for an additional 18 hours. After the second round of chloramphenicol selection, cells were harvested, washed with LB, re-suspended in LB to OD₆₀₀ 0.1, and grown to OD₆₀₀ 3. Cells were then harvested and genomic DNA was isolated. The control culture was grown to OD₆₀₀ 0.4 and diluted into fresh LB to OD₆₀₀ 0.02. After growing to OD₆₀₀ 0.4, the culture was diluted to OD₆₀₀ 0.1 and grown to OD₆₀₀ 3, at which point cells were harvested for genomic DNA isolation. For phenotype verification, individual mutants from the defined library were grown to OD₆₀₀ 0.2, at which point chloramphenicol was added at 80 μg/ml and incubated for 18 hours. The number of viable cells was determined at zero time point and after chloramphenicol treatment.