The Transcriptional Responses of Mycobacterium tuberculosis to ...

THE JOURNAL

OF

BIOLOGICAL CHEMISTRY

Vol. 279, No. 38, Issue of September 17, pp. 40174 –40184, 2004 Printed in U.S.A.

The Transcriptional Responses of Mycobacterium tuberculosis to Inhibitors of Metabolism NOVEL INSIGHTS INTO DRUG MECHANISMS OF ACTION*□ S Received for publication, June 17, 2004, and in revised form, July 9, 2004 Published, JBC Papers in Press, July 9, 2004, DOI 10.1074/jbc.M406796200

Helena I. M. Boshoff‡§, Timothy G. Myers¶, Brent R. Copp储, Michael R. McNeil**, Michael A. Wilson¶, and Clifton E. Barry III‡ From the ‡Tuberculosis Research Section, NIAID, National Institutes of Health, Rockville, Maryland 20852, the ¶Research Technologies Branch, NIAID, National Institutes of Health, Bethesda, Maryland 20892, the 储Department of Chemistry, University of Auckland, Auckland 1020, New Zealand, and the **Department of Microbiology, Colorado State University, Ft. Collins, Colorado 80523

The differential transcriptional response of Mycobacterium tuberculosis to drugs and growth-inhibitory conditions was monitored to generate a data set of 430 microarray profiles. Unbiased grouping of these profiles independently clustered agents of known mechanism of action accurately and was successful at predicting the mechanism of action of several unknown agents. These predictions were validated biochemically for two agents of previously uncategorized mechanism, pyridoacridones and phenothiazines. Analysis of this data set further revealed 150 underlying clusters of coordinately regulated genes offering the first glimpse at the full metabolic potential of this organism. A signature subset of these gene clusters was sufficient to classify all known agents as to mechanism of action. Transcriptional profiling of both crude and purified natural products can provide critical information on both mechanism and detoxification prior to purification that can be used to guide the drug discovery process. Thus, the transcriptional profile generated by a crude marine natural product recapitulated the mechanistic prediction from the pure active component. The underlying gene clusters further provide fundamental insights into the metabolic response of bacteria to drug-induced stress and provide a rational basis for the selection of critical metabolic targets for screening for new agents with improved activity against this important human pathogen.

Despite the introduction of directly observed therapy, short course, in 1995, millions of tuberculosis patients continue to perish, and fully one-third of the world’s population is infected with the causative agent of this disease, Mycobacterium tuberculosis. New drugs are urgently needed to shorten the duration of tuberculosis chemotherapy and treat the increasing number of infections with drug-resistant organisms. Target selection is critical to the development of new drugs but is hampered by a lack of understanding of the dynamics of the metabolic response to interruption of target function even by current * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. □ S The on-line version of this article (available at http://www.jbc.org) contains supplemental data. § To whom correspondence should be addressed: Twinbrook II, Rm. 239, 12441 Parklawn Dr., Rockville, MD 20852. Tel.: 301-4519438; Fax: 301-4020993; E-mail: [email protected].

agents. Predicting targets that would manifest a cidal activity, therefore, is limited by our incomplete understanding of the physiology of the bacilli and its ability to adapt to disruption of metabolism. An organism responds to changes in its environment by altering the level of expression of critical genes that transduce such signals into metabolic changes favoring continued growth and survival. Analysis of the transcriptional response by microarray can, in theory, provide clues to such adaptive responses, but thus far gene expression profiles have only been used to contrast the mechanisms of action of a small number of related drugs (1–3). Coordinately regulated sets of genes (regulons) are often controlled by single transcriptional regulators that function as genetic master switches, committing the bacterium to a major alteration in metabolism. In M. tuberculosis, examples of such regulatory mechanisms have been reported recently from studies using genetic approaches, including the dormancy regulon (4) and the stringent response (5). The complexity of the cellular transcriptional response to drug-induced stress makes it very difficult to derive this sort of information solely from microarray analysis of a limited number of agents affecting the same metabolic pathway (6, 7). However, by analyzing a wide variety of conditions, groups of genes have been identified that appear co-regulated under many different conditions in yeast (8). In this study, we applied genome-wide expression profiling to diverse environmental changes, including many different drug types, to begin to map the adaptability of the bacilli to interruption of specific arms of metabolism. This allowed us to identify clusters of coordinately regulated genes both diagnostic for drug mechanism of action and useful for a more rational approach to the selection of critical drug targets. EXPERIMENTAL PROCEDURES

M. tuberculosis Growth Conditions, RNA Isolation, and Hybridization—M. tuberculosis (H37Rv, ATCC 27294) was grown in Middlebrook 7H9 supplemented with albumin/dextrose/NaCl/glycerol/Tween 80, Dubos medium, or defined minimal medium as previously described (9). Carbon sources were either 10 mM glucose, 10 mM succinate, or 0.05 mM sodium palmitate, and cultures were grown from an A650 of 0.005– 0.3 before RNA isolation. Cultures grown under a self-depleted oxygen gradient (NRP-1) were set up as described (10), and RNA was isolated after 3– 6 days. Nutrient starvation cultures were set up as previously described (5) in phosphate- or Tris-buffered saline containing 0.05% Tween 80 (PBST or TBST). The organic extract of Eudistoma amplum was prepared as follows: the frozen, ground invertebrate was extracted with water at 4 °C, and the pellet was freeze-dried and then extracted at room temperature with methanol/methylene chloride (1:1). The solvent was evaporated and the extract dissolved in Me2SO to 9 mg/ml.

40174

This paper is available on line at http://www.jbc.org

Transcriptional Analysis of M. tuberculosis S-Nitrosoglutathione (GSNO)1 was used as a NO source. Cultures were grown from an A650 of 0.07– 0.3 before adding either drug or solvent (1000-fold dilutions from Me2SO, ethanol, or water), and RNA was isolated at selected intervals thereafter (11). For each drug-treated culture, a parallel culture was treated with an equivalent amount of vehicle (Me2SO, ethanol, or water) for the same amount of time. RNA from the latter culture was used as the reference sample to which the drug-treated sample was compared. Each treatment condition and each drug concentration was repeated a minimum of two independent times. M. tuberculosis carrying an integrated copy of the Mycobacterium smegmatis amidase pzaA, which hydrolyzes several aromatic amides (12), was used for cultures treated with pyrazinamide, 5-chloropyrazinamide, nicotinamide, or benzamide. This strain was also used to investigate the transcriptional response during extracellular pH stress. Treatment was done in Middlebrook 7H9-based medium adjusted to the required pH (4.8, 5.2, or 5.6) with H3PO4 and referenced to cells grown in the same acidified medium without amide addition. Since pyrazinamide is only effective at low cell densities, cultures were grown to an A650 of 0.05 before treatment was initiated. MICs were measured using the microbroth dilution technique (13) using H37Rv or M. tuberculosis ⌬mbtB (14). Iron preloading of cells was done for 3 h in 7H9-based medium containing a 10-fold excess of Fe3⫹. RNA labeling and hybridization was as previously described (11). Microarray Preparation and Data Analysis—Microarray preparation is described under GEO accession number GSE1642. Expression ratios were calculated as the feature pixel median minus background pixel median for one color channel divided by the same for the other channel. In cases where more than 10% of the feature pixels were saturated, the feature pixel mean was used instead of the median. When the feature pixel mean did not exceed the background pixel mean by more than two S.D. values (calculated from the background pixel distribution), the feature pixel median is used in the ratio without background subtraction. In cases where both color channels were near background (same criterion), the ratio value was set to “missing.” Expression ratios were transformed to the log base 2 for all further calculations. Standardized gene expression ratio patterns were calculated by subtracting the mean expression ratio and dividing by the S.D. statistics calculated from all ratios (all microarrays) for that gene. Standardizing in this way corrected for scale differences between the response patterns for different genes. The resulting z-scores were averaged according to the drug treatment name, resulting in a single value for each drug name for each gene (see Supplementary Data). These gene patterns where then clustered using a K-means algorithm (SAS Proc Fastclus) using the Euclidean distance as the dissimilarity metric. Two rounds of K-means clustering were conducted. The first with the subset of genes showing the highest treatment-dependent variation in expression as judged by one-way analysis of variance (SAS Proc ANOVA) on the original log ratio vectors, using treatment name as the class variable. The second round used all genes, but without allowing the cluster number to increase or the cluster centroids to drift (assigning the remaining genes to the existing clusters formed in the first round of clustering). We arrived at the Fastclus “maxclusters” parameter, the maximum number of clusters to define, value of 150 clusters, by multiplying the number of class levels (treatments), 75, by 2. The number of genes selected for the first round of clustering (1650) was limited to 11 times the number of clusters and were those with the most statistically significant one-way analysis of variance score. A single pattern of response for each gene cluster was calculated as the mean of all standardized gene patterns assigned to the cluster by Proc Fastclus. These cluster centroids were themselves clustered using average linkage algorithm calculated in Microsoft Excel VBA using a one minus the Pearson correlation coefficient for the distance metric (15) to arrive at the ordering of rows in Fig. 3 (for details, see Supplementary Data). Patterns of response to each treatment were clustered using the same method to arrive at the column order in Fig. 3. The array data have been deposited in the Gene Expression Omnibus at NCBI (GEO; available on the World Wide Web at www.ncbi.nlm.nih.gov/geo) with GEO accession number GSE1642. Real Time, Quantitative Reverse Transcription-PCR Assay—The expression of iniB, Rv3161, kasA, efpA, fadE23, rplJ, rplN, dnaE2, radA,

1 The abbreviations used are: GSNO, S-nitrosoglutathione; MIC, minimum inhibitory concentration; MTT, 3-(4,5-dimethylthiazol-2-yl)2,5-diphenyltetrazolium bromide; GC, gene cluster; TRC, triclosan; CPZ, chlorpromazine; TRZ, thioridazine; CCCP, carbonyl cyanide chlorophenylhydrazone; DNP, dinitrophenol; CCO, cytochrome c oxidase; EMB, ethambutol.

40175

mbtB, csd, narH, narG, cydA, and ald was quantitated after normalization of RNA levels to the expression of the sigA gene as previously described (11), and results are available in Supplementary Data. Enzyme Assays—Enzyme assays were done on proteins from M. smegmatis. Purification and determination of NADH dehydrogenase and succinate dehydrogenase activities was performed as previously described (16). Oxygen Consumption Assays—The effect of drugs on oxygen consumption by M. tuberculosis was done in parafilm-sealed, glass screwcap tubes that were filled with midlog phase culture containing 0.001% methylene blue. Decolorization typically occurred after 12 h. The rate of oxygen consumption was measured in M. smegmatis using midlog stage cultures treated with drug for 1 h before adding 0.01% methylene blue and monitoring decolorization at 665 nm. Sugar Analysis—Cell walls were prepared, and glycosyl compositions were determined by the alditol acetate methods as described (17). NADH/NAD⫹ Determination—NADH and NAD⫹ levels were determined by a sensitive cycling assay (18). Briefly, M. tuberculosis cultures were grown to an A650 of 0.3 and treated with drugs for 3 h. At this stage, cells were rapidly harvested (two 2-ml samples) and resuspended in 0.2 M HCl (NAD⫹ determination) or 0.2 M NaOH (NADH determination). Nucleotide extraction was further facilitated by bead beating of the suspensions with 0.2-ml glass beads (40 s, maximum speed). Extracts were further prepared, and enzyme assays were performed as previously described (18). All determinations were repeated in at least three independent experiments. Menaquinol/Menaquinone Analysis—Cultures were grown to an A650 of 0.3 and treated for 3 h with drug or solvent alone. Menaquinone and menaquinol were extracted as described (19) and quantified by liquid chromatography-mass spectrometry (Hewlett-Packard 1100) using a C18 column with detection by DAD at UV 266 nm. All extractions were repeated at least three independent times. Intracellular ATP Determinations—These assays were done on cultures of M. tuberculosis containing an integrated luciferase gene from pMV306-groELluc grown to an A650 of 0.1– 0.2 and treated with drug for 20 –120 min. ATP levels were determined by bioluminescence as previously described (20). MTT Assays—M. smegmatis at an A650 of 0.2 was treated with drug or vehicle alone for 15 min (100 ␮l/well in 96-well plates in quadruplicate) before the addition of 25 ␮l of 2 mg/ml MTT. The reaction was stopped after 30 min by the addition of 25 ␮l of 10% SDS, and the absorbance at 595 nm was recorded. The assay was repeated two independent times. RESULTS

De Novo Analysis of Expression Data Results in Mechanism of Action-based Clustering of Transcriptional Responses— Transcriptional profiling of M. tuberculosis was performed using 430 whole-genome microarrays to measure the effects of 75 different drugs, drug combinations, or different growth conditions at various times relative to a sample of logarithmically growing M. tuberculosis. The drug concentrations and time points (see Supplementary Data) were chosen such that 10% or less of the total number of genes were differentially (2-fold or more) regulated and through consulting previously published studies (2, 3). Expression of highly responsive genes within certain drug groups was confirmed by quantitative reverse transcription-PCR (see Supplementary Data). The quality of these data were tested using Pearson rank tests, which verified that arrays within each treatment group were highly correlated (Fig. 1). Log-transformed expression ratios were standardized according to the pan-array distribution for each gene and then averaged according to treatment. To analyze these data, unsupervised clustering methods were applied to the 345 expression profiles. Expression data sets of genes that were upor down-regulated at least 3-fold in four or more experiments were analyzed by agglomerative hierarchical clustering method (21) to identify drug groupings based on gross analysis of coordinately regulated genes (Fig. 2). This revealed that groups of drugs clustered separately based on known mechanisms of action. Thus, protein synthesis inhibitors, transcriptional inhibitors, aromatic amides, cell wall synthesis inhibitors and agents that damage DNA fell into distinct groups.

40176

Transcriptional Analysis of M. tuberculosis

FIG. 1. Prediction matrix using Pearson correlation over the entire array of genes to assess accuracy of classification of individual arrays. Individual arrays from each treatment group (columns, n ⫽ 430) were compared with the collection of averaged array representation computed to be characteristic of each treatment group (rows). The treatment group with the top average score was taken as the “prediction” for that array. Matrix numbers are counts of individual arrays predicted to match with the averaged array representing a treatment group. Major treatment groups (class of inhibitor) are color-coded as follows: violet, growth in minimal medium with palmitate or succinate as carbon sources as compared with glucose as carbon source; lavender, growth in acidified medium; blue, aromatic amides that can be hydrolyzed intracellularly; light blue, agents that inhibit cell wall synthesis; pale blue, agents that affect DNA integrity or topology; green, inhibitors of protein synthesis; yellow, growth under conditions that were associated with expression of the dosR regulon; light green, agents besides NO that inhibit respiration, and TRC; pink, transcriptional inhibitors; orange, nutrient starvation in PBST or TBST; red, pyridoacridones and iron scavengers. The black boxes are correct assignments within the treatment class. DIPED, diisopropylethylenediamine; DTNB, dithiobis(nitrobenzoic) acid.

These observations were further supported by a prediction matrix using the Pearson correlation over the entire array of genes to predict the treatment group (Fig. 1). This showed that the individual profiles of drug treatments were accurately classified to groups of agents with similar predicted modes of action. The majority of apparent “mispredictions” of well characterized inhibitors in this matrix, usually correlated with agents within the same treatment group. Identification of Co-regulated Genes Reveals Validated Regulons—To identify biologically meaningful groups of genes, profiles were partitioned into 75 drug groups with each drug group corresponding to a single type of treatment, and genes were analyzed by K-means clustering to uncover those with

similar expression patterns across these sets (Fig. 3). Many of these clusters contained genes that were functionally related, but many also contained genes that encoded proteins of unknown function. Genes previously shown to be members of the DosR- and RecA-controlled regulons (4, 11, 22) were independently identified by this process of gene clustering. Gene cluster 39 (GC39), for example, contained 21 of the 48 members of the dormancy regulon, and three closely linked clusters (GC126, -56, and -137) contained 38 genes previously reported to be up-regulated by DNA damage. The Metabolic Response to Inhibition of Translation—Analysis of the cellular response to translational inhibition re-


40177

vealed a general down-regulation of macromolecular synthesis, as expected, although there was an evident attempt to increase synthesis of the translational apparatus. Up-regulated genes included those implicated in ribosomal architecture and translation (e.g. GC28, -36, -70, -71, -90, and -118) whereas down-regulated genes included aspects of macromolecular metabolism similar to those responsive to starvation (e.g. ppk and relA). Regulation of these genes did not appear to be mediated by the stringent response through ppGpp, since relA was down-regulated. Interestingly, Rv1026 (encoding a possible pppGpp-5⬘-phosphohydrolase that would hydrolyze any residual mediator of the stringent response (23)) was up-regulated during translational inhibition and downregulated during starvation. The observed up-regulation of the inorganic pyrophosphatase encoded by ppa would probably slow ribosomal degradation (24) and also contrasts with the down-regulation of this gene during starvation. Not surprisingly, ppa is part of a regulon (GC71) containing genes implicated in translation, and, combined with the observed down-regulation of ppk (a polyphosphate kinase), this suggests an important role for polyphosphate in mycobacterial adaption to translational inhibition (25). A gene cluster containing the gene encoding the regulatory protein of pyrimidine biosynthesis (pyrR) (GC69) was also upregulated, consistent with the observed down-regulation of expression of several genes involved in pyrimidine biosynthesis, whereas genes involved in purine and pyrimidine salvage (apt, gmk, prsA, thyA, and cdd) and conversion of nucleotides to deoxyribonucleotides (nrdF1, nrdF2, nrdH, and nrdI) were upregulated upon translational inhibition. Aminoglycosides were associated with an up-regulation of heat shock proteins (GC134), presumably resulting from mistranslation-induced aberrant peptides in the cytoplasm as has been observed for other bacteria (26). Tetracycline and roxithromycin did not induce this response, consistent with the fact that they block release of the nascent peptide during translational inhibition. Our analysis also suggested that translational inhibition results in inhibition of DNA replication and the processing of replication forks. The down-regulation of several genes supports this hypothesis, including the following: Rv1708 (possible role in initiation of replication); the major replicative DNA polymerase (11); and DnaA, which plays a role in initiation of chromosomal replication. Likewise, genes implicated in turnover of DNA were up-regulated, including nth, recR, hupB, recF, and ssb. This did not result in a signal that was relayed as DNA damage, however, since recA and dnaE2 (11) were not up-regulated. The Metabolic Response to Inhibitors of DNA Transcription and Gyrase Function—Unsurprisingly, the mode of action of transcriptional inhibitors such as rifamycins could best be described as a global down-regulation of most gene clusters, including the ribonucleotide reductase genes (GC49), heat shock proteins (GC134), and several ribosomal genes. Despite this, some transcript levels were elevated, but this was probably due to differential mRNA stabilities. Fluoroquinolones bind gyrase and topoisomerase IV on DNA, blocking transcription and replication and resulting in DNA damage (27). DNA damage also results from treatment with UV irradiation, H2O2, and mitomycin C. All of these treatments resulted in the up-regulation of the previously characterized (11, 28) SOS gene clusters (GC56, -126, and -137) as FIG. 2. Cellular transcriptional responses cluster by drug mechanism of action. Average linkage clustering of expression profiles of genes that were up- or down-regulated at least 3-fold in four or more experiments. Profiles were clustered using a modified uncentered Pearson correlation coefficient as the similarity metric. The major drug groups are color-coded as in Fig. 1. Pyridoacridone clusters 1 and 2

correspond to low (5⫻ MIC) and high (10⫻ MIC) concentrations, respectively. CPZ and TRZ profiles correspond to concentrations of 1–2⫻ MIC (1) and 2–3⫻ MIC (2).

40178


FIG. 3. Transcriptional responses to drug treatment reveal underlying logically associated gene clusters. Shown is a heat maprendered table of gene expression changes. In A, each column is the average of several microarray experiments using the same drug/treatment, whereas each box within a row is the average for all genes assigned to the same gene cluster. Major drug classes are color-coded (bars in top

Transcriptional Analysis of M. tuberculosis well as several DNA repair-associated genes that were not correlated with this regulon. The gyrase inhibitor novobiocin does not induce double-stranded breaks (29) and did not cluster with those agents that did. Novobiocin affected the expression of a more limited subset of DNA repair or structural maintenance genes including the up-regulation of the RecA-independent, Y family polymerase member encoded by dinP. The effects of fluoroquinolones (including novobiocin) could be distinguished from the other forms of DNA damage employed in this study by the unique up-regulation of the class Ib ribonucleotide reductase genes (GC49) as well as nrdF1. Deoxyribonucleotide pools are regulated by the activity of ribonucleotide reductase and are intricately linked to DNA replication (30). Repair of fluoroquinolone-induced doublestranded DNA breaks may provide the signal to elevate DNA synthetic machinery, as has been reported in E. coli (31), or such a signal may be generated during the stalling of chromosomal DNA replication. The up-regulation of nrdF1, the alternate nonessential ribonucleotide reductase small subunit (32), suggests that this subunit may play a role in supplying dNTPs during DNA turnover or repair. Gene Signatures for Inhibitors of Cell Wall Biosynthesis— Inhibition of cell wall biosynthesis by any known agent revealed that there was one unique set of genes broadly responsive to this insult. These genes comprised two regulons (GC27 and -128). GC27 consists of secreted cell wall-associated proteins such as Rv1987 encoding a putative glycohydrolase, lprJ, fbpC, murD, dacB1, and Rv3717 encoding a putative N-acetylmuramoyl-Lalanine amidase and may be regulated by SigD. GC128 included the iniBAC operon, an operon of unknown function that has previously been shown to be responsive to such inhibitors (3, 33) and several cell wall-associated genes. This gene cluster was also linked to two others (GC79 and -89) that contained several cell wall biosynthetic and turnover genes. The ␤-lactam antibiotics induced unique genes consistent with their known transpeptidase-inhibiting properties. These included Rv3717, a putative N-acetylmuramoyl-L-alanine amidase that may correspond to the enzyme implicated in penicillin-induced autolysis in other bacteria (34). Ethambutol (EMB), which inhibits the arabinosyltransferases that decorate arabinogalactan and lipoarabinomannan (35, 36), has recently been extensively analogued with some success (13). A comparison of the transcriptional profiles of two potent analogs with EMB showed that, despite many similarities in transcriptional profiles, EMB differentially affected a regulon (GC82) containing genes within the FAS-II pathway as well as a regulon (GC17) that contained genes implicated in fatty acid modification. This mechanistic divergence was confirmed by the observation that whereas M. tuberculosis cells treated with EMB rapidly lost acid-fastness, cells treated with the analogs did not (data not shown). Further, cells treated with EMB contained significantly less arabinose than controls, whereas cells treated with analogs of EMB did not (Fig. 4). Isoniazid, ethionamide, and thiolactomycin are all thought to inhibit enzymes in the repetitive catalytic cycle of the FAS-II pathway that elongates fatty to mycolic acids, and an analysis of the transcriptional response to all three drugs is very consistent with that mechanism. These drugs were also found to cluster closely with cerulenin, which inhibits both the FAS-II and FAS-I systems (FAS-I is responsible for de novo synthesis of fatty acids from acetate). The signature profile for effects on fatty acid syn-

40179

FIG. 4. Sugar content of mycobacterial cell wall during treatment with various diamines. M. tuberculosis cell walls were prepared 45 h after drug addition, and glycosyl compositions were determined by the gas chromatography analysis of alditol acetate derivatives. The bars indicate molar ratios of rhamnose (black), arabinose (gray), and galactose (hashed) residues after normalization to rhamnose. The inset shows arabinose/galactose ratios.

thesis includes the acpM-kasA-kasB-accD6 operon as well as efpA. There were also effects on several polyketide synthases as well as polyketide synthase-associated genes and proteins implicated in fatty acid transport that distinguished the fatty acid synthesis inhibitors from other groups of drugs as well as triclosan (TRC). Analysis of the gene clusters regulated by cerulenin, ethionamide, thiolactomycin, and isoniazid indicated that the combined effects on GC27, -79, -120, -82, and -128 provide a signature profile diagnostic for inhibition of fatty acid synthesis. These gene clusters all contain genes involved in aspects of cell wall metabolism and secreted proteins. Notably, GC82 contains the FAS-II operon (Fig. 3B) as well as pks16 and Rv0241c encoding a protein with homology to a fatty acid synthetase ␤ subunit, as well as fadA2, suggesting that these gene products are involved in interlinked metabolic pathways. In addition, the closely linked GC120 contains fas, efpA, accA3, accD4, pks13, and fadD32 (Fig. 3B). A recent report describing the role of pks13 in the final step of mycolate condensation (37) is further evidence for the metabolic relatedness of genes in many of the gene clusters such as GC82 and GC120. Differential expression of distinct sets of gene clusters distinguishes FAS-II from FAS-I inhibitors, whereas other gene clusters can distinguish the effects of isoniazid and ethionamide from thiolactomycin, showing that these various fatty acid synthesis inhibitors affect biochemical pathways that are connected to distinct transcriptional regulators. The Transcriptional Profile of TRC Suggests That the Primary Mode of Action Is on Respiration—TRC, a potent inhibitor of the FAS-II system in vitro, did not appear to elicit a similar response in vivo. This broad spectrum antibacterial agent inhibits the bacterial fatty acid biosynthetic enzyme, enoyl-(acylcarrier protein) reductase in vitro (38). Counterintuitively, TRC apparently stimulates degradation of fatty acids, since enzymes corresponding to every step of ␤-oxidation are upregulated. TRC concurrently up-regulated citrate synthase (gltA1), which controls flux through the tricarboxylic acid cycle, and the enzymes of the pyruvate dehydrogenase complex, which control levels of acetyl-CoA. Because TRC clustered with known respiratory inhibitors (Fig. 2), we investigated the activity of components of the respiratory chain in vitro. TRC was found to cause a dose-dependent inhibition of the membranebound quinol reductase succinate dehydrogenase (Fig. 5A). The effect of TRC on the membrane-bound electron transport chain was also manifested in a rapid drop in intracellular ATP levels in a reporter strain of M. tuberculosis expressing the luciferase gene, which was not observed with other cell wall inhibitors

dendogram) as in Fig. 1. The inset shows a color scale for z-scores. By interrogating the frequency at which boxes appear to be regulated by independent treatments, the specificity of the gene cluster for a series of stresses can be determined. The top dendogram shows relatedness of drug treatments based on gene clusters. The right dendogram indicates relationship between gene clusters. Further detail is provided in Supplementary Data. B, detail of gene clusters 82 and 120 showing z-scores for individual genes for each drug group. Ami, amikacin; Cap, capreomycin; SM, streptomycin; Rox, roxithromycin; Tet, tetracycline.

40180


FIG. 5. Effect of phenothiazines and triclosan on respiratory enzymes and cellular respiration. Shown are integral membrane succinate dehydrogenase (A and B) and type II NADH dehydrogenase (C) activities in the presence of 25 (⫻) and 50 (Œ) ␮g/ml TRC (A) and 25 (⫻) and 50 ␮g/ml (Œ) ␮g/ml TRZ (B and C) in comparison with control Me2SO (⽧)-treated assays. D, oxygen consumption of methylene blue-treated M. smegmatis cultures in the absence (1) or presence (2) of 20 (⫻) and 50 (Œ) ␮g/ml TRZ. The inset shows oxygen consumption of methylene blue-treated M. tuberculosis cultures in the absence (1) or presence (2) of 25 ␮g/ml TRZ.

FIG. 6. Effect of respiratory inhibitors and drugs on cellular ATP levels in luciferase-expressing M. tuberculosis. Cells in quadruplicate wells were treated for the 20 min before measurement of bioluminescence after the addition of luciferin. Shown is a typical result of one of three independent experiments.

(Fig. 6). The effect of the rapid depletion of ATP levels was evidenced by the concomitant up-regulation of relA, which can be directly linked to the expected decrease in synthesis of charged tRNAs. In addition, TRC treatment resulted in a decline in the intracellular redox potential of M. smegmatis as measured by reduction of MTT by intracellular dehydrogenases (Fig. 7). The decline in the intracellular redox status was reflected in a decrease in the NADH/NAD⫹ ratio and an increase in the menaquinol/menaquinone ratio of the major isomer MK9(H2) (39) (Table I), an effect that was not observed with isoniazid. TRC did not, however, directly inhibit oxygen consumption as measured by the rate of methylene blue decolorization in normally growing cells (results not shown). Regulation of Respiratory Chain Components and the Mode of Action of Phenothiazines and Azoles—The phenothiazines chlorpromazine (CPZ) and thioridazine (TRZ) are thought to directly affect respiration (40), whereas the azoles econazole and clotrimazole have been proposed to inhibit growth via interaction with cytochrome P450-containing monooxygenases (41). CPZ has been proposed to inhibit respiration (42, 43), although other mechanisms have been proposed (44). Azoles are known to bind to the heme iron in cytochrome P450s, specifically the CYP51 sterol demethylase in fungi and possibly the CYP121 in M. tuberculosis as has been suggested (41, 45). The phenothiazines and azoles shared many similarities in regulated gene clusters, including one that contained known components of the respiratory chain (GC149), which included the

alternative terminal oxidase encoded by the cydA and cydB genes. To unambiguously demonstrate an effect on respiration, we performed methylene blue decolorization assays to measure the rate of oxygen consumption in treated and untreated cells. The results (Fig. 5D) indicated that the phenothiazines inhibited oxygen consumption in both M. tuberculosis and M. smegmatis (Fig. 5D), whereas TRC, azoles, carbonyl cyanide chlorophenylhydrazone (CCCP), dicyclohexylcarboxidiimide, KCN, and dinitrophenol (DNP) did not (results not shown). The activity of two quinone reductases, type II NADH-ubiquinone dehydrogenase and succinate dehydrogenase, were also assessed in membrane preparations in the presence of these drugs. The phenothiazines were potent inhibitors of both the type II NADH-ubiquinone dehydrogenase and the integral membrane succinate dehydrogenase (Fig. 5, B and C), whereas the azoles were inhibitors of succinate dehydrogenase activity (results not shown). The effects of the phenothiazines on respiration were further manifested in a rapid drop in intracellular ATP levels in M. tuberculosis (Fig. 6) as well as a decline in intracellular redox potential in M. smegmatis (Fig. 7), effects that were also observed with known modulators of the proton motive force such as protonophores (DNP, CCCP, and nigericin). These respiratory inhibitors all up-regulated relA expression, which can be ascribed to the expected decrease in charged tRNAs due to ATP depletion. The decreased intracellular redox potential was reflected by a decrease in the intracellular NADH/NAD⫹ ratio as well as a decrease in the cell-associated


40181

FIG. 7. M. smegmatis reductive power during treatment with respiratory inhibitors and drugs. Cells in quadruplicate wells were treated with drugs or vehicle alone for 15 min before initiation of the MTT reduction assay. The reduced formazan product was measured colorimetrically at 595 nm. Results shown are one of two similar independent experiments.

TABLE I Changes in redox status of NADH/NAD⫹ and the predominant isoprenoid quinol/quinone pair (MK9H2/MK9) during treatment of M. tuberculosis with various respiratory modulators Values represent the averages and S.D. values of three independent experiments. For values without S.D., only a single determination was performed. Drug treatment

-Fold change relative to control NADH/NAD⫹

MK9H2/MK9

-fold

10 ␮g/ml KCNa 20 ␮g/ml CPZa 20 ␮g/ml TRZa 50 ␮M CCCPa 0.2 mM GSNO 50 ␮g/ml TRCa 50 ␮M nigericina 5 ␮g/ml clofaziminea 0.5 mM DNPa 10 ␮g/ml menadione NRP-1a 0.2 mM GSNO, 10 ␮g/ml menadione 0.2 mM GSNO, 20 ␮g/ml CPZa 0.2 mM GSNO, 10 ␮g/ml KCNa 1 mM NaN3a 100 ␮M dicyclohexylcarboxidiimidea 0.04 ␮g/ml isoniazid a b

0.7 ⫾ 0.1 0.7 ⫾ 0.1 0.4 ⫾ 0.2 1.7 ⫾ 0.4 1.4 ⫾ 0.1 0.8 ⫾ 0.1 0.5 ⫾ 0.3 1.6 ⫾ 0.2 1.3 0.2 ⫾ 0.1 0.4 0.4 ⫾ 0.1 0.9 ⫾ 0.1 0.7 ⫾ 0.1 1.8 ⫾ 0.1 1.9 1.0

1.95 ⫾ 0.04 0.69 ⫾ 0.02 0.68 ⫾ 0.02 2.02 1.3 ⫾ 0.1 1.12 ⫾ 0.05 1.16 ⫾ 0.01 0.84 1.74 0.90 ⫾ 0.01 3.1 ⫾ 0.5 0.93 0.73 ⫾ 0.03 1.75 ⫾ 0.05 1.67 ⫾ 0.26 NDb 0.98 ⫾ 0.01

Treatments that resulted in up-regulation of the cyd operon. ND, not done.

menaquinol/menaquinone ratio (Table I). To further define the effect of respiratory inhibitors such as the phenothiazines on the regulation of respiratory chain components, we analyzed transcriptional profiles of cells treated either alone or with combinations of known respiratory modulators including cyanide, azide, dithiothreitol, ZnSO4, uncouplers, redox cycling agents (menadione and clofazimine), and NO. We also explored the effect on respiration for cells grown on palmitate as the sole carbon source. This analysis revealed that two distinct gene clusters (GC149 and -39) were independently associated with alterations in electron flux through the respiratory chain (Fig. 8). One of these was the previously described NO-inducible dosR (4), whereas the other was a cluster of genes containing the cyd operon (GC149). Inhibition of both terminal oxidases, cytochrome c oxidase (CcO) and cytochrome bd oxidase, by NO (but not CcO-specific inhibitors like cyanide or azide) or by depletion of oxygen during adaption to hypoxic conditions resulted in up-regulation of the dosR regulon, as did growth on the reduced carbon source palmitate. The effect of NO on the dosR regulon could be fully reversed by

cyanide, menadione, and clofazimine, all of which could reoxidize the nitrosylferroheme of cytochromes (Fig. 8). Notably, all three also were found to result in resumption of oxygen consumption (results not shown). In contrast, the CcO-specific inhibitors (cyanide and azide) and agents that affect maturation of CcO (ZnSO4 and dithiothreitol) resulted in up-regulation of the cyd operon (cydA, cydB, cydD, and cydC) encoding the non-proton-pumping cytochrome bd oxidase (Fig. 2). Unlike other bacterial systems (54), we did not observe up-regulation of the cyd regulon during H2O2 or menadione treatment. The cyd regulon was also upregulated during adaption to hypoxic conditions as has been reported for M. smegmatis (46) as well as during growth on palmitate, indicating that up-regulation of the dosR regulon and cyd genes was not mutually exclusive. The cyd operon was highly responsive to alterations in the transmembrane proton gradient by treatment with protonophores such as CCCP, DNP, and nigericin. Intracellular acidification would also affect maintenance of the transmembrane proton gradient, and indeed amides such as pyrazinamide also resulted in up-regulation of the cyd operon. Since changes in the pyridine nucleotide or respiratory quinol redox poises are associated with modulation of expression of respiratory components in a variety of bacteria (47– 49), the reduced versus oxidized forms of these molecules were measured with a variety of respiratory inhibitors that affected the expression of the cyd operon or the dosR regulon (Table I). This indicated that the regulation of these gene sets could not be simply correlated with the redox state of these electron carriers. High Information Content Screening: Transcriptional Profiling for de Novo Mechanism of Action Determination—Ascididemin is a marine pyridoacridone alkaloid that has cytotoxic activity to tumor cells as well as showing antiparasitic activity (50). The mechanism of action of these compounds in eukaryotic cells has been attributed to inhibition of DNA topoisomerase and direct cleavage of DNA (51). Ascididemin has also been reported to have potent antimycobacterial activity (Table II). By transcriptional profiling, ascididemin was shown to induce up-regulation of the mycobactin biosynthetic genes and affected transcription of several iron-associated genes. The mycobactin genes and a few putative functionally related genes formed a gene cluster (GC108) that constituted a signature profile for this group of drugs that included other iron scavengers such as dipyridyl and desferoxamine. The contribution of iron scavenging to the mode of action of ascididemin was confirmed by an increase in MIC when the

40182


FIG. 8. The cyd and dosR regulons are independent cellular responses to inhibition of respiration. Shown is the transcriptional response of two gene clusters associated with respiration to all treatments. The mean z-score over all of the genes within the dormancy regulon (GC39, gray) and GC149 (pink) is plotted for each drug group. The dashed lines represent the median expression of the regulons (GC39 (gray) and GC149 (pink)) over all of the drug treatments.

TABLE II Minimum inhibitory concentration of ascididemin and the ascididemin-containing natural product against M. tuberculosis strains in the presence or absence of excess iron Values are in ␮g/ml. H37Rv

H37Rv ⫹ Fe3⫹

Ascididemin

0.12

0.5

0.062

0.5

Organic extract

0.56

2.3

0.28

1.13

Compound

⌬mbtB

⌬mbtB ⫹ Fe3⫹

␮g/ml

cells were preloaded with iron (Table II). Further, a mycobactin-deficient mutant (14), which is impaired for iron uptake, was found to be hypersusceptible to this compound (Table II). The up-regulation of the cyd operon was not due to an inhibition of the respiratory dehydrogenases (results not shown) but probably indicates decreased electron transport through the respiratory chain due to the decreased synthesis of mature cytochromes and other iron-containing respiratory proteins. Ascididemin could be distinguished from other iron scavengers by the up-regulation of at least two gene clusters (GC44 and -116) unique to this compound. To assess whether microarray analysis could be used to identify the major mode of action in a crude extract containing a natural product with antimycobacterial activity, an organic extract of Eudistoma amplum which produces ascididemin, was used to treat M. tuberculosis and the resulting transcriptional profile compared with that of ascididemin. The crude extract produced an almost identical profile to that of the pure molecule. High Information Content Screening: Predicting Detoxification—Molecules with aromatic character in general were potent inducers of monooxygenases, dioxygenases, certain methylases, efflux systems, and the associated carboxylic acid degradation genes. There was, for example, a striking up-regulation of potential drug detoxification and efflux mechanisms during TRC treatment, and many of the gene clusters upregulated in common between the phenothiazines, azoles, DNP, CCCP, clofazimine, and TRC included genes potentially involved in drug detoxification and efflux. Gene clusters 53 and 51 consisted largely of potential detoxification mechanisms. Microarray analysis indicated that the diamine analogs were potent inducers of potential drug detoxification mechanisms and that the apparent lack of effect on cell wall arabinogalactan composition could also be ascribed to detoxification of the analogs.

DISCUSSION

In this study, we analyzed the transcriptional response of M. tuberculosis to diverse environmental alterations. Analysis of this data allowed us to select clusters of genes that were coordinately regulated across multiple different treatments and to examine the responses of these gene clusters to each class of agents. Not only did unbiased grouping successfully cluster inhibitors of known similar mechanisms of action, but the underlying metabolic logic for the effect of these inhibitors was consistent with historical studies of mechanism of action of such agents. As expected, genes within operons were predominantly associated with the same gene cluster. The co-regulation of unknown genes with genes of known and related function allows inference to be drawn about their putative metabolic roles. In general, inhibitors of protein translation induce the cell to attempt to synthesize more ribosomes, reduce the turnover and degradation of existing ribosomes, and reduce the de novo synthesis of nucleotides while enhancing nucleotide recycling and salvage. Although somewhat similar to the genetic response to starvation (5) this response is unique and independent of ppGpp regulation. Importantly, these studies suggest an as yet unexplored role for polyphosphate metabolism in determining the overall status of the mycobacterial translational apparatus. The cellular response to interrupting DNA supercoiling was directly related to the ability of the inhibitor to induce doublestranded breaks in the chromosome. Fluoroquinolones, which induce such damage, strongly induce the SOS response, whereas novobiocin, which does not induce such damage, does not. Disruption of DNA supercoiling levels by either fluoroquinolones or novobiocin induces genes involved in DNA synthesis and the synthesis of DNA precursors like deoxyribonucleotides. The unique regulation of nrdF1 suggests either a general role of this subunit in the regulation of DNA synthesis levels or a specific role in DNA synthesis during DNA turnover or repair. Pyrazinamide-elicited transcriptional profiles clustered with other amides such as nicotinamide and benzamide, supporting the hypothesis that these agents exert their antimycobacterial effect by imposing stress on the intracellular pH homeostasis mechanisms (12). These aromatic amide-elicited profiles were in turn distinct from the transcriptional responses of the organism due to extracellular pH stress during growth in an acidic environment. Inhibition of cell wall synthesis represents a major mechanism of action for many existing antituberculars and has been

Transcriptional Analysis of M. tuberculosis historically a well studied area. Transcriptional profiles of all such inhibitors (except TRC) revealed a potent up-regulation of common cell wall gene clusters (GC27 and -128) that included the iniBAC operon, previously shown to be responsive to such inhibitors (3, 33). This and other genes regulated in common by this class (interestingly, many are also up-regulated by starvation) are strong candidates for genes involved in cell wall turnover, remodeling, or maintenance, possibly critical functions in nonreplicating organisms. Cell wall turnover is thought to play a role during adaption to microaerophilia, and although GC27 and GC128 were down-regulated during adaptation to oxygen limitation, gene clusters implicated in fatty acid modification and polyketide synthesis were up-regulated (GC17 and -66). TRC is the single outlier that does not appear to induce any of the cell wall-responsive genes that characterized the cell wall inhibitors, instead stimulating fatty acid degradation and clustering with respiratory inhibitors. This was confirmed by showing that TRC directly inhibits the membrane-bound succinate dehydrogenase but surprisingly this does not translate into a global effect on respiration since oxygen consumption appears normal in TRC-treated cells. Thus, despite evidence that TRC has a cell wall component to its mechanism of action, the mechanism of toxicity is clearly more complex than previously appreciated (20). The up-regulation of genes implicated in biogenesis of respiratory cytochromes and of the cydAB genes encoding the cytochrome bd quinol oxidase by CPZ, the azoles, and TRC suggested a common effect on electron transport. CYP121, suggested as the target of the azoles, seemed an unlikely target due to its absence from the susceptible M. smegmatis and its nonessentiality in M. tuberculosis (52). For the phenothiazines CPZ and TRZ, the similarities in transcriptional profiles with known uncouplers and respiratory poisons suggested a direct effect on respiration. We were able to support this by directly demonstrating that consumption of oxygen by M. tuberculosis was in fact inhibited by these compounds and that two different dehydrogenases of the respiratory chain were inhibited in vitro, strongly suggesting that respiratory inhibition is a major component of the mode of action of these agents. TRC and the azoles inhibited the respiratory succinate dehydrogenase but did not inhibit oxygen consumption. The effect of these drugs on respiration may be due to their hydrophobicity, which would tend to sequester them in the mycobacterial cell wall, consistent with their effect on the membrane-bound succinate dehydrogenase complex. However, the effects of TRC and the azoles on respiratory and other membrane-associated proteins may be nonspecific. Our studies with various inhibitors also suggest some fundamental principles underlying respiratory regulation. The dormancy (dosR) regulon, induced by inhibition of both terminal oxidases by NO, is regulated by DosR and mediates adaptation to hypoxia (4, 53). It has been suggested that the signal detected by the cognate sensor kinase of DosR might be transduced by CcO (4). However, in our studies, we found that the CcO inhibitors cyanide and azide do not induce up-regulation of this response, whereas growth on reduced carbon sources did. This argues against CcO as the transducer of the dormancy signal and points to a sensor that detects the redox status of the cell. In some bacteria, the redox balance between quinones and quinols transduces signals to flavin-containing sensor kinases (47, 48), whereas in some other bacteria, the redox poise is sensed through the NADH/NAD⫹ ratio (49). However, we found that neither the NADH/NAD⫹ nor the menaquinol/ menaquinone redox status was responsible for the differential regulation of these gene sets in M. tuberculosis. These findings do not rule out the possibility that the redox poise of another

40183

electron carrier is the signal that controls the regulation of at least one of these sets of genes. Inhibition of CcO specifically results in up-regulation of the cyd operon encoding the non-proton-pumping cytochrome bd oxidase (46). This operon was also up-regulated during adaption to hypoxic conditions as has been found in M. smegmatis (46) and during growth on palmitate. Such conditions would be expected to alter the transmembrane proton gradient due to intracellular accumulation of organic acids and a protonophore effect of fatty acids. This effect was verified using known protonophores such as CCCP, DNP, and nigericin, which specifically disrupts the proton gradient of the transmembrane electrochemical potential. Intracellular acidification would be expected to have a similar effect, and, in fact, amides such as pyrazinamide resulted in up-regulation of the cyd operon. Thus, the data base of transcriptional profiles described here for a very diverse set of drugs and growth-inhibitory conditions provides information that is highly consistent with historical studies. It has also proven useful for agents lacking historical information. Transcriptional profiling of the pyridoacridone ascididemin suggested that this compound interfered with iron acquisition, which we were able to validate directly. Moreover, transcriptional profiles were useful in highlighting key metabolic responses even in the face of the more complex responses observed with unpurified natural products. The signature of ascididemin, for example, was found to be entirely reproducible in the crude extract of the tunicate from which it was obtained. Since a primary bottleneck in the discovery of new agents from natural sources lies in the resource-intensive process of isolation of the active principle, transcriptional profiling offers the opportunity to prioritize such extracts to those with novel mechanisms of action prior to such a commitment. This concept of high information content screening also encompasses information regarding chemotypes that induce undesirable bacterial detoxification and efflux systems that could be used to prioritize hits from high throughput screening using responses of a small number of responsive gene clusters. The coordinately regulated gene clusters identified here represent the most extensive set of regulons to date defining the metabolic potential of this important pathogen. Understanding this potential and the plasticity of the pathogen’s response to challenge, is critical to understanding pathogen biology to a level sufficient to define targets against both actively replicating and nonreplicating organisms. Highly responsive genes and the list of those that are not suggest precise targets and intervention points for the development of a new generation of antituberculosis agents. Acknowledgments—We thank Mike Cashel, Valerie Mizrahi, and co-workers for stimulating discussions and Jose Ribeiro for providing the M. tuberculosis data base. We gratefully acknowledge the assistance of Michael Goodwin with the menaquinol analyses. We acknowledge the NCI, National Institutes of Health, Natural Products Branch for the sample from the Open and Active Repositories Program. REFERENCES 1. Agarwal, A. K., Rogers, P. D., Baerson, S. R., Jacob, M. R., Barker, K. S., Cleary, J. D., Walker, L. A., Nagle, D. G., and Clark, A. M. (2003) J. Biol. Chem. 278, 34998 –35015 2. Betts, J. C., McLaren, A., Lennon, M. G., Kelly, F. M., Lukey, P. T., Blakemore, S. J., and Duncan, K. (2003) Antimicrob. Agents Chemother. 47, 2903–2913 3. Wilson, M., DeRisi, J., Kristensen, H. H., Imboden, P., Rane, S., Brown, P. O., and Schoolnik, G. K. (1999) Proc. Natl. Acad. Sci. U. S. A. 96, 12833–12838 4. Voskuil, M. I., Schnappinger, D., Visconti, K. C., Harrell, M. I., Dolganov, G. M., Sherman, D. R., and Schoolnik, G. K. (2003) J. Exp. Med. 198, 705–713 5. Dahl, J. L., Kraus, C. N., Boshoff, H. I., Doan, B., Foley, K., Avarbock, D., Kaplan, G., Mizrahi, V., Rubin, H., and Barry, C. E., III (2003) Proc. Natl. Acad. Sci. U. S. A. 100, 10026 –10031 6. Huang, P., Feng, L., Oldham, E. A., Keating, M. J., and Plunkett, W. (2000) Nature 407, 390 –395 7. Savoie, C. J., Aburatani, S., Watanabe, S., Eguchi, Y., Muta, S., Imoto, S., Miyano, S., Kuhara, S., and Tashiro, K. (2003) DNA Res. 10, 19 –25

40184


8. Segal, E., Shapira, M., Regev, A., Pe’er, D., Botstein, D., Koller, D., and Friedman, N. (2003) Nat. Genet. 34, 166 –176 9. Schnappinger, D., Ehrt, S., Voskuil, M. I., Liu, Y., Mangan, J. A., Monahan, I. M., Dolganov, G., Efron, B., Butcher, P. D., Nathan, C., and Schoolnik, G. K. (2003) J. Exp. Med. 198, 693–704 10. Wayne, L. G., and Hayes, L. G. (1996) Infect. Immun. 64, 2062–2069 11. Boshoff, H. I., Reed, M. B., Barry, C. E., III, and Mizrahi, V. (2003) Cell 113, 183–193 12. Boshoff, H. I., and Mizrahi, V. (2000) J. Bacteriol. 182, 5479 –5485 13. Lee, R. E., Protopopova, M., Crooks, E., Slayden, R. A., Terrot, M., and Barry, C. E., III (2003) J. Combinatorial Chem. 5, 172–187 14. De Voss, J. J., Rutter, K., Schroeder, B. G., Su, H., Zhu, Y., and Barry, C. E., III (2000) Proc. Natl. Acad. Sci. U. S. A. 97, 1252–1257 15. Kaufman, L., and Rousseeuw, P. J. (1990) Finding Groups in Data: An Introduction to Cluster Analysis, 9th Ed., John Wiley & Sons, Inc., New York 16. Miesel, L., Weisbrod, T. R., Marcinkeviciene, J. A., Bittman, R., and Jacobs, W. R., Jr. (1998) J. Bacteriol. 180, 2459 –2467 17. Deng, L., Mikusova, K., Robuck, K. G., Scherman, M., Brennan, P. J., and McNeil, M. R. (1995) Antimicrob. Agents Chemother. 39, 694 –701 18. San, K. Y., Bennett, G. N., Berrios-Rivera, S. J., Vadali, R. V., Yang, Y. T., Horton, E., Rudolph, F. B., Sariyar, B., and Blackwood, K. (2002) Metab. Eng. 4, 182–192 19. Krivankova, L., and Dadak, V. (1980) Methods Enzymol. 67, 111–114 20. Slayden, R. A., Lee, R. E., and Barry, C. E., III (2000) Mol. Microbiol. 38, 514 –525 21. Sherlock, G. (2000) Curr. Opin. Immunol. 12, 201–205 22. Davis, E. O., Dullaghan, E. M., and Rand, L. (2002) J. Bacteriol. 184, 3287–3295 23. Primm, T. P., Andersen, S. J., Mizrahi, V., Avarbock, D., Rubin, H., and Barry, C. E., III (2000) J. Bacteriol. 182, 4889 – 4898 24. Kuroda, A., Nomura, K., Ohtomo, R., Kato, J., Ikeda, T., Takiguchi, N., Ohtake, H., and Kornberg, A. (2001) Science 293, 705–708 25. Zhang, H., Ishige, K., and Kornberg, A. (2002) Proc. Natl. Acad. Sci. U. S. A. 99, 16678 –16683 26. Ng, T. B., Lam, S. K., and Fong, W. P. (2003) Biol. Chem. 384, 289 –293 27. Drlica, K. (1999) Curr. Opin. Microbiol. 2, 504 –508 28. Rand, L., Hinds, J., Springer, B., Sander, P., Buxton, R. S., and Davis, E. O. (2003) Mol. Microbiol. 50, 1031–1042 29. Gormley, N. A., Orphanides, G., Meyer, A., Cullis, P. M., and Maxwell, A. (1996) Biochemistry 35, 5083–5092 30. Elledge, S. J., Zhou, Z., and Allen, J. B. (1992) Trends Biochem. Sci. 17, 119 –123 31. Filpula, D., and Fuchs, J. A. (1977) J. Bacteriol. 130, 107–113

32. Dawes, S. S., Warner, D. F., Tsenova, L., Timm, J., McKinney, J. D., Kaplan, G., Rubin, H., and Mizrahi, V. (2003) Infect. Immun. 71, 6124 – 6131 33. Alland, D., Steyn, A. J., Weisbrod, T., Aldrich, K., and Jacobs, W. R., Jr. (2000) J. Bacteriol. 182, 1802–1811 34. Normark, B. H., and Normark, S. (2002) J. Intern. Med. 252, 91–106 35. Belanger, A. E., Besra, G. S., Ford, M. E., Mikusova, K., Belisle, J. T., Brennan, P. J., and Inamine, J. M. (1996) Proc. Natl. Acad. Sci. U. S. A. 93, 11919 –11924 36. Khoo, K. H., Douglas, E., Azadi, P., Inamine, J. M., Besra, G. S., Mikusova, K., Brennan, P. J., and Chatterjee, D. (1996) J. Biol. Chem. 271, 28682–28690 37. Portevin, D., De Sousa-D’Auria, C., Houssin, C., Grimaldi, C., Chami, M., Daffe, M., and Guilhot, C. (2004) Proc. Natl. Acad. Sci. U. S. A. 101, 314 –319 38. Heath, R. J., White, S. W., and Rock, C. O. (2001) Prog. Lipid. Res. 40, 467– 497 39. Yassin, A. F., Brzezinka, H., Schaal, K. P., Truper, H. G., and Pulverer, G. (1988) Zentralbl. Bakteriol. Mikrobiol. Hyg. Ser. A 267, 339 –356 40. Leung, D., van der Oost, J., Kelly, M., Saraste, M., Hill, S., and Poole, R. K. (1994) FEMS Microbiol. Lett. 119, 351–357 41. McLean, K. J., Marshall, K. R., Richmond, A., Hunter, I. S., Fowler, K., Kieser, T., Gurcha, S. S., Besra, G. S., and Munro, A. W. (2002) Microbiology 148, 2937–2949 42. Kalnenieks, U., Galinina, N., Bringer-Meyer, S., and Poole, R. K. (1998) FEMS Microbiol. Lett. 168, 91–97 43. Balijepalli, S., Boyd, M. R., and Ravindranath, V. (1999) Neuropharmacology 38, 567–577 44. Amaral, L., Kristiansen, J. E., Viveiros, M., and Atouguia, J. (2001) J. Antimicrob. Chemother. 47, 505–511 45. Munro, A. W., McLean, K. J., Marshall, K. R., Warman, A. J., Lewis, G., Roitel, O., Sutcliffe, M. J., Kemp, C. A., Modi, S., Scrutton, N. S., and Leys, D. (2003) Biochem. Soc. Trans. 31, 625– 630 46. Kana, B. D., Weinstein, E. A., Avarbock, D., Dawes, S. S., Rubin, H., and Mizrahi, V. (2001) J. Bacteriol. 183, 7076 –7086 47. Bock, A., and Gross, R. (2002) Eur. J. Biochem. 269, 3479 –3484 48. Grabbe, R., and Schmitz, R. A. (2003) Eur. J. Biochem. 270, 1555–1566 49. Brekasis, D., and Paget, M. S. (2003) EMBO J. 22, 4856 – 4865 50. Copp, B. R., Kayser, O., Brun, R., and Kiderlen, A. F. (2003) Planta Med. 69, 527–531 51. Matsumoto, S. S., Biggs, J., Copp, B. R., Holden, J. A., and Barrows, L. R. (2003) Chem. Res. Toxicol. 16, 113–122 52. Sassetti, C. M., Boyd, D. H., and Rubin, E. J. (2003) Mol. Microbiol. 48, 77– 84 53. Sherman, D. R., Voskuil, M., Schnappinger, D., Liao, R., Harrell, M. I., and Schoolnik, G. K. (2001) Proc. Natl. Acad. Sci. U. S. A. 98, 7534 –7539 54. Poole, R. K., and Cook, G. M. (2000) Adv. Microb. Physiol. 43, 165–224