Comparative transcriptomics reveals key gene expression differences between the human and bovine pathogens of the Mycobacterium tuberculosis complex

The Mycobacterium tuberculosis complex can be viewed as a set of ecotypes with distinct host preferences (Smith et al., 2006). Hence, while M. tuberculosis can only be sustained in human populations, Mycobacterium bovis can infect and be maintained in populations of wild and domesticated animals such as cattle, deer and badgers. The molecular origin for these phenotypes is encoded in the genomes of the pathogens. The genome sequences of M. tuberculosis H37Rv and M. bovis 2122 are >99.95 % identical at the nucleotide level, with M. bovis containing no unique genes per se compared to M. tuberculosis (Cole et al., 1998; Garnier et al., 2003). Comparative analyses of the M. tuberculosis and M. bovis genomes have revealed the basis for distinguishing phenotypes such as the pyruvate requirement of M. bovis in glycerol-based media, or the reason for eugonic/dysgonic colony morphology (Keating et al., 2005). However, comparative genomics in itself does not reveal the basis for the complexity of phenotype between M. tuberculosis and M. bovis. Extra information needs to be layered onto the genome data, such as gene expression profiling, metabolic network analyses, signalling pathways, etc., to fully explore the biology of these pathogens.

To explore the hypothesis that transcriptome differences play a role in the differing ecotypes of the M. tuberculosis complex, we set out to define the key gene expression differences between the human and bovine tubercle bacilli. To achieve this, the culture conditions for both organisms must be identical. Clearly, the nature of batch culture makes the system highly dynamic, complicating interpretation of expression profiles. Chemostat systems on the other hand allow tight control of bacterial growth in a defined, reproducible system (Hoskisson & Hobbs, 2005). We therefore chose to define the transcriptomes of the genome sequenced strains M. tuberculosis H37Rv and M. bovis 2122 under physiological steady-state conditions achieved through growth in carbon-limited chemostat cultures.

During the pathogenic process, M. tuberculosis and M. bovis are thought to be exposed to a number of different stress conditions, including acidic pH, reactive nitrogen and oxygen species, and nutrient starvation (Chan et al., 2005). Many of the proteins induced in response to stress are thought to be involved in the survival of the pathogen inside the host. We therefore also chose to study the effect of acid-shock on the steady-state transcriptomes of M. tuberculosis and M. bovis as a means to probe the response of both strains to a perturbation.

Hence, the aim of this study was to identify key gene expression differences between M. tuberculosis H37Rv and M. bovis 2122 under both steady-state and perturbation conditions. We found that the expression of genes encoding a range of functional activities varied between the strains, and we discuss the possible biological impact of this variation on strain phenotype.

Bacterial strains and growth conditions.
For the continuous culture experiments, M. tuberculosis H37Rv and M. bovis AF2122 were grown at 37 °C in defined minimal medium, CAMR Mycobacteria Media (CMM) Mod6 (pH 7.0), at a dissolved oxygen tension of 10 % (50 % air saturation), using the approach described previously for M. tuberculosis H37Rν (Bacon et al., 2004). CMM Mod6 was a modified version of CMM (James et al., 2000) in which all amino acids were removed except for L-asparagine. Glycerol and glucose were omitted and Tween 80 (0.2 %, v/v) was included as sole carbon source. As is standard practice, BSA (0.5 % w/v) was included to ameliorate the toxic side effects of Tween 80 degradation. The cultures were performed in 1 l glass fermentation vessels, which were operated as chemostats by controlling medium addition and effluent removal to maintain a working volume of 500 ml. The chemostats were operated at a constant dilution rate of 0.03 h^–1, which corresponds to a mean generation time of 23 h. The cultures were maintained at a constant pH 7.0±0.1 and a temperature of 37 °C±0.1. Cultures were monitored for viability by measuring optical density at 540 nm and total viable count as described by Bacon et al. (2004).

Acid-shock experiments.
Acid-shock studies were performed in a chemostat. Three independent aerobic steady-state cultures of M. tuberculosis and M. bovis were established at pH 7.0, then when the pH had dropped to 5.5 the samples were rapidly removed for transcriptome analysis. At the start of each pH perturbation experiment, maximum and minimum pH levels were set on the controller unit in the range of pH 5.48–5.52. HCl (1 M) was automatically added dropwise to the culture until the desired pH of the culture was reached, i.e. pH 5.5. The addition of the first drop of acid was taken as time point zero and the time taken for the pH to fall from 7.0 to 5.5 was in the range of 2.5 to 5 min. The culture was then maintained at pH 5.5 via the automatic addition of 1 M NaOH or 1 M HCl. Samples (20 ml) were rapidly withdrawn from the culture at 5, 15, 30, 60 min and 24 h after time point zero and RNA was extracted as described previously (Bacon et al., 2004). Residual DNA was removed with the use of a DNAfree kit (Ambion) and the integrity of the RNA was checked by non-denaturing gel electrophoresis.

M. tuberculosis–M. bovis microarrays.
The M. tuberculosis–M. bovis composite microarrays used in these experiments were developed by the Bacterial Microarray Group ( µG@S, St Georges, University of London) in collaboration with the Veterinary Laboratories Agency (Weybridge). The array consists of 4410 PCR products (size range 60–1000 bp) that represent all the genes in the genomes of M. tuberculosis strains H37Rv and CDC1551 and M. bovis strain AF2122. Primer pairs for each ORF were designed with Primer 3 software and selected by BLAST analysis to have minimal cross-homology with all other ORFs. PCR products were generated and purified with the aid of a RoboAmp 4200 robot (MWG Biotech) and arrayed in duplicate onto UltraGap aminosilane-coated glass slides (Corning) by a Biorobotics Microgrid II microarrayer (Genomic Solutions). The quality of spots on the microarrays was evaluated using the Spot Check kit (Genetix Ltd). The array design is available in µG@Sbase, accession no. A-BUGS-31 () and also ArrayExpress, accession no. A-BUGS-31.

cDNA labelling and microarray hybridizations.
The method used to synthesize labelled cDNA probes from RNA was similar to that described previously with some modifications (Stewart et al., 2002). Complementary DNA synthesized from RNA was fluorescently labelled in a reverse transcription reaction in the presence of Cy5-CTP (Amersham Biosciences). In a total volume of 16 µl, RNA (3–5 µg) was mixed with 5 µg random hexamers, heated to 70 °C for 10 min and then snap-cooled on ice. To the chilled mixture, 6 µl 5x first strand buffer, 3 µl 100 mM DTT, 15 nmol each of dATP, dGTP and dTTP, 6 nmol dCTP and 2.5 nmol Cy5-dCTP (Amersham Biosciences) were added. The labelling reaction was initiated by the addition of 400 units of Superscript II (Life Technologies), and the reactions were incubated for 10 min at 25 °C and then for 90 min at 42 °C.

cDNA synthesized using DNA as template was fluorescently labelled using Cy3-dCTP (Amersham Bioscience) and Klenow DNA polymerase. DNA (3 µg) was mixed with 3 µg random primers in a total volume of 41.5 µl, heated at 95 °C for 5 min and then cooled on ice. To the cooled mixture, 5 µl 10x Klenow buffer, 5 nmol each of dATP, dGTP and dTTP, 2 nmol dCTP and 1.5 nmol Cy3-dCTP were added. The labelling reaction was initiated by the addition of 5 units Klenow fragment of DNA polymerase (New England Biolabs) and the reaction was incubated at 37 °C for 1.5 h. The Cy5- and Cy3-labelled cDNAs from the labelling reactions were co-purified in water using a mini elute purification kit (Qiagen). The volume of the purified labelled cDNAs was adjusted to 60 µl in 4x SSC and 0.3 % SDS. Prior to use, the microarrays were prehybridized at 65 °C for 20 min in 50 ml of a solution containing 3.5x SSC, 0.1 % SDS and 10 mg BSA ml^–1. The slides were washed at room temperature with H₂O for 1 min, propan-2-ol for 1 min and then dried by centrifugation. The labelled cDNAs were heated for 3 min at 95 °C, before being applied to the array under two 22x22 mm coverslips (LifterSlip; Erie Scientific).

Scanning and image analysis.
Hybridized arrays were scanned using an Affymetrix 428 scanner with the photomultiplier tube gain set in the range 40–50, so that spots with the highest signal intensities were just below the level of saturation; this generated two images per microarray (one per channel) in TIF format. Fluorescent spots on each scanned image were quantified using Imagene 5.1 (Biodiscovery) using a segmentation signal percentage set to a high of 95 % and a low of 5 %. The signal for each spot was taken as the median of all pixel intensities within the defined area of the spot. Genomic DNA from M. tuberculosis was hybridized in the Cy3 channel, and was used as a common reference for all arrays.

Data normalization and statistical analysis.
Data processing employed Mathematica Version 5.2 (Wolfram Research). For each channel background, values were first subtracted from signal values for each spot and values less than 0.01 were set to 0.01. For each microarray the log ratio of the Cy5 to Cy3 signal was calculated for every spot; each log ratio was then divided by the median of the log ratios of the microarray so as to centre the gene expression distributions of the whole set of microarrays to the same point (zero). After this step the microarrays for the M. bovis acid-shock experiment presented a similar scale; however, some of the arrays for the M. tuberculosis acid-shock data showed a wider distribution. To set all arrays to a common scale, a median absolute deviation scale transformation was therefore applied. As a final normalization step, duplicate spots for each gene on every microarray were averaged, and then the means of every gene across all technical replicate microarrays were calculated.

Means of the three biological replicates were used to compare gene expression between time points. Genes whose mean normalized expression value was larger than threefold different in one or more samples collected after the addition of acid compared to the zero time point sample were selected. A t test with a significance of 0.05 and with the Benjamini Hochberg correction was applied to verify statistical differential expression; those genes verifying this latter requirement were selected for further study. Files containing normalized data were imported into GeneSpring 7.1 (Silicon Genetics) for a graphical representation of the microarray data. Fully annotated microarray data has been deposited in µG@Sbase (; accession no. E-BUGS-37) and ArrayExpress (; accession no. E-BUGS-37).

Quantitative real-time PCR (qRT-PCR).
In a total volume of 10 µl, RNA (0.5–1.0 µg) was mixed with 3 µg random hexamers, heated to 70 °C for 10 min and then snap-cooled on ice. To the chilled mixture, 4 µl 5x first strand buffer, 2 µl 100 mM DTT, and 20 nmol each of dATP, dCTP, dGTP and dTTP were added. The labelling reaction was initiated by the addition of 400 units Superscript II, and the reactions were incubated for 10 min at 25 °C and then for 90 min at 42 °C. The synthesized cDNA was purified with a DNA Clean and Concentrator kit (Zymo Research) and serial dilutions of the purified cDNA were made. qRT-PCR reactions were performed using the QuantiTect SYBR Green PCR kit (Qiagen) in a total volume of 25 µl containing 5 µl diluted cDNA, 12.5 µl 2x QuantiTect Master Mix and 0.5 µl gene-specific primers. PCR amplification was performed using a RotorGene 3000 (Corbett Research) and a programme consisting of 1 cycle at 95 °C for 15 min, and then 40–50 cycles at 95 °C for 15 s (denaturation), 55 °C for 30 s (extension step) and 80 °C for 30 s (signal acquisition). Melting curve analysis and gel electrophoresis were performed to confirm the specificity of the qRT-PCR products. The sequences of the gene-specific primer pairs were designed to show minimal cross-hybridization to all other genes within the genome. Sequences of the primer pairs are listed in supplementary Table S1 (available with the online version of the journal). The relative expression of each of the genes was normalized to sigA.

Establishing continuous cultures of M. tuberculosis and M. bovis
A key issue in setting up the culture system was the choice of carbon source. Previously in batch culture, glycerol has been used as the carbon source for M. tuberculosis, and pyruvate has been used for M. bovis; clearly, comparative transcriptome profiling required both strains to be grown on the same carbon source, so neither glycerol nor pyruvate were used in this study. Instead, we exploited the presence of the oleic acid ester, Tween 80, in mycobacterial liquid media where it acts as a detergent to ensure dispersed growth; however, it can also be used as a carbon source by mycobacteria (Dubos & Davis, 1946). Tween 80 (0.2 %, v/v, in water) was therefore used as sole carbon source. The composition of the medium was manipulated to ensure that Tween 80 was the growth-limiting nutrient, and this was confirmed by adding a pulse of Tween 80 (0.4 %, v/v) to the steady-state cultures which resulted in an increase in OD₅₄₀ over 48 h from 1.05 to 1.50 for the M. bovis culture, and from 1.325 to 1.55 for the M. tuberculosis culture. Hence, as well as its action as a detergent to prevent clumping, Tween 80 can be used as a carbon source by both M. tuberculosis and M. bovis under continuous culture conditions.

Cultures of M. tuberculosis and M. bovis were maintained in steady-state at a dissolved oxygen tension of 10 % (equivalent to 50 % air saturation) at a dilution rate of 0.03 h^–1, giving a constant mean generation time of 23 h. The mean OD₅₄₀ and viable counts for three M. bovis cultures (Mb1–3) under steady-state conditions were, respectively, as follows: Mb1, 0.95 and 3.4x10⁸ c.f.u. ml^–1; Mb2, 0.8 and 3.5x10⁸ c.f.u. ml^–1; Mb3, 1.16 and 1.8x10⁸ c.f.u. ml^–1. Mean OD₅₄₀ and viable counts for three M. tuberculosis cultures (Mtb1–3) under steady-state conditions were, respectively, as follows: Mtb1, 1.29 and 3.7x10⁸ c.f.u. ml^–1; Mtb2, 1.23 and 3.9x10⁸ c.f.u. ml^–1; Mtb3, 1.32 and 3.1x10⁸ c.f.u. ml^–1.

Differential gene expression in M. tuberculosis H37Rv and M. bovis AF2122 during steady-state growth
Ninety-two genes were identified that showed a minimum of threefold differential expression between M. bovis and M. tuberculosis during steady-state growth in continuous culture. Forty-three M. bovis genes showed higher expression than their M. tuberculosis orthologues, while 49 genes showed higher expression in M. tuberculosis (Tables 1 and 2, respectively). To focus our discussion we highlight differences in the key categories of cell wall and secreted proteins, lipid metabolism, gene regulators, PE/PPE family, and toxin–antitoxin (TA) gene pairs, and examine the potential role of these variably expressed genes in phenotypic differences between the human and bovine bacilli.

Table 1. Genes expressed at higher levels in M. bovis than M. tuberculosis under steady-state growth

Table 2. Genes expressed at higher levels in M. tuberculosis than M. bovis under steady-state growth

Cell wall and secreted proteins
Two of the most highly expressed genes in M. bovis compared to their orthologues in M. tuberculosis were mpb83 and mpb70, encoding the serodominant antigens MPB83 and MPB70 which have been shown to be the major protein differences between M. tuberculosis and M. bovis (Hewinson et al., 1996). Using a combination of expression profiling and comparative genomics, Behr and colleagues have shown that the inactivation of a negative regulator is the basis for high expression in M. bovis (Charlet et al., 2005; Said-Salim et al., 2006). Hence, expression of mpb83 and mpb70 is controlled by an anti-sigma factor/sigma factor pair, RskA/SigK; a mutation in the RskA (Mb0452c) anti-sigma factor in M. bovis prevents repression of SigK activity, leading to constitutive mpb70 and mpb83. Another animal-adapted member of the M. tuberculosis complex, the oryx bacillus (Mostowy et al., 2005; van Soolingen et al., 1994), is also a high producer of MPB70 and MPB83, and has an independent mutation in RskA that abolishes its activity (Said-Salim et al., 2006). These independent mutations would suggest positive selection for overexpression of MPB70 and MPB83 in M. bovis and the oryx bacillus, the phenotypic advantage of which remains to be determined. The genes Mb0455c, Mb0456c, Mb0457c, dipZ, Mb2901 and Mb2902c are also part of the SigK regulon, and show higher expression in M. bovis than in M. tuberculosis; however, as with MPB83 and MPB70, the precise functions of their encoded proteins are unknown.

Rv0987 and Rv0988 encode part of an ABC transporter and a putative secreted hydrolase, respectively. Both genes showed higher expression in M. tuberculosis H37Rv (Table 2), and in M. bovis 2122 the Rv0987 orthologue is frameshifted (Mb1013/Mb1014). Inactivation of Rv0987 in M. tuberculosis reduces the ability of the mutant to bind to THP-1 macrophages, but does not reduce virulence in the mouse model (Rosas-Magallanes et al., 2007). An Rv0987 mutant was also found to be incapable of invasion across the blood–brain barrier (Jain et al., 2006). Whether mutation and decreased expression of the M. bovis Rv0987-0988 orthologues reflect inactivation of factors no longer needed for full virulence of the bovine bacillus remains to be verified.

Lipid metabolism
Sulfolipid is a trehalose-containing sulfatide which is absent from M. bovis, but present in M. tuberculosis. It is therefore not surprising that the mmpL8-papA1-pks2 locus responsible for sulfolipid synthesis (Bhatt et al., 2007; Converse et al., 2003; Sirakova et al., 2001) showed approximately fourfold higher expression in M. tuberculosis (Table 2). The cluster pks4-Rv1184c-Rv1185c involved in the synthesis of another group of non-mycolated trehalose esters, the diacyltrehalose (DAT) and polyacyltrehaloses (PATs) (Dubey et al., 2002), also showed higher expression in M. tuberculosis (Table 2). Both of these gene clusters have been shown to be under control of the PhoPR regulon (Gonzalo Asensio et al., 2006; Walters et al., 2006). It is therefore interesting that phoR contains two non-synonymous single nucleotide polymorphisms in M. bovis, with glycine at position 71 of the M. tuberculosis PhoR replaced by isoleucine in M. bovis and proline-172 replaced by leucine. Hence, mutation of PhoR may lead to the divergent expression of sulfolipid, DAT and PAT biosynthetic genes in M. bovis.

M. bovis showed higher expression of the lppX-pks1 pair, which are involved in the transport (Sulzenbacher et al., 2006) and synthesis (Constant et al., 2002) of phthiocerol dimycoserosate (PDIM), respectively (Table 1). TLC analyses did not reveal significant differences in PDIM production between M. bovis 2122 and M. tuberculosis H37Rv (data not shown), suggesting the expression difference may relate to the synthesis of the PDIM-derived mycoside B by M. bovis, a glycolipid which is lacking in M. tuberculosis H37Rv. Rv2958c, encoding a glycosyltransferase involved in the synthesis of the trisaccharide phenolic glycolipid (PGL) that is derived from PDIM, was also found to be expressed at higher levels in M. bovis (Table 1). PGL is not produced by M. tuberculosis H37Rv, so the higher expression of this gene in M. bovis may simply be down to low expression of the synthetic pathways in M. tuberculosis. However, it should also be noted that Rv2958c is non-functional in M. bovis, so the higher expression of its gene may simply be due to loss of negative feedback inhibition. As the regulation of PDIM and PGL synthesis is currently poorly understood, comparisons between M. bovis and M. tuberculosis offer an ideal starting point to tease the system apart.

Mb0222 showed 17-fold higher expression in M. bovis compared with the M. tuberculosis orthologue Rv0216, with the adjacent lipW also up-regulated (Table 1). The similarity of Mb0222 to acyl dehydratases, and its location close to genes encoding a fatty acyl-CoA ligase (Mb0220c), an acyl dehydrogenase (Mb0221c) and an esterase (Mb0223c), indicates a role in fatty acid catabolism. Rv0216 has been shown to be required for virulence in M. tuberculosis (Sassetti & Rubin, 2003), while Mycobacterium leprae has retained a functional orthologue, ML2625 (Cole et al., 2001); the M. bovis orthologue may also, therefore, play a role in virulence. Up-regulation of Mb0222 may indicate that the encoded product is primed for immediate activity on M. bovis entering the host.

Gene regulators
Transcriptional regulators that showed variable expression were Mb2651, Mb2654c, Mb3109c/virS, Mb3477c (expressed at higher levels in M. bovis; Table 1), and Rv0196, Rv0275c and Rv2160A-Rv2160c (higher in M. tuberculosis; Table 2). The latter gene pair, Rv2160A-Rv2160c, are frameshifted in M. tuberculosis H37Rv, with the intact orthologue in M. bovis (Mb2184c) showing similarity to TetR-family transcriptional repressors. As the flanking genes Rv2159c and Rv2161c also show higher expression in M. tuberculosis H37Rv, it suggests that loss of the repressor from M. tuberculosis leads to up-regulation of this locus.

Mb3706c/whiB4 and Mb3829c/whiB6 were expressed at higher levels in M. bovis 2122 (Table 1). The WhiB family were originally designated as transcription factors based on the presence of putative DNA-binding motifs (Davis & Chater, 1992), but direct proof for this activity is lacking. Recent work has shown that this family bind Fe–S clusters (Jakimowicz et al., 2005), and indeed WhiB4 binds Fe–S and functions as a disulfide reductase (Alam et al., 2007). Furthermore, comparing expression of all seven whiB genes in M. tuberculosis to a range of stress conditions, Bishai and colleagues found that whiB6 expression was the most stress-responsive (Geiman et al., 2006). The WhiB proteins of tubercle bacilli may therefore function in redox signalling during oxidative and nitrosative stress, conditions which prevail in the in vivo milieu. Up-regulation of whiB4 and whiB6 in M. bovis under in vitro conditions may therefore indicate an increased need to sense and respond quickly to oxidative stress.

The level of expression of virS was 10-fold higher in M. bovis compared with M. tuberculosis (Table 1). In M. tuberculosis, VirS has been shown to regulate expression of the genes Rv3083-Rv3089, the products of which could be involved in the modification of fatty acids in the cell envelope (Singh et al., 2003). However, no differential expression of these latter genes between M. tuberculosis and M. bovis was observed.

Clearly, a full description of the regulatory networks controlled by the transcription factors named above is required if their role in the phenotypic discrimination of human and bovine bacilli is to be defined. The application of genome-wide location studies, such as chromatin-immunoprecipitation assays, will aid in the identification of these gene networks (Rodrigue et al., 2007).

PE and PPE family
The PE and PPE proteins are two large protein families with >160 members in M. tuberculosis and M. bovis. Approximately 60 % of the PE and PPE proteins differ across the two strains, a feature clearly at odds with the high sequence identity across the majority of encoded proteins in human and bovine bacilli (Cole et al., 1998; Garnier et al., 2003). While the precise function of the PE and PPE proteins remains to be defined, members have been implicated in virulence and antigenic variation. As well as variation at the nucleotide level, it is therefore interesting that the genes encoding PPE33, PE35, PPE51 and PPE55 showed variation in expression between M. tuberculosis H37Rv and M. bovis AF2122 (Tables 1 and 2). The higher expression of the gene encoding PPE33, Rv1809, in M. tuberculosis may be due to the M. bovis orthologue being frameshifted (Mb1838/Mb1839), thus disrupting its correct regulation. Rv3136, which encodes PPE51, has been shown to be one of 22 genes that are consistently up-regulated in M. tuberculosis H37Rv compared to the attenuated derivative H37Ra (Mostowy et al., 2004), a result that parallels its decreased expression in M. bovis. The differential expression of PPE genes between human and bovine bacilli points to subtle variation in the control of these genes among strains, the biological implications of which remain to be defined.

TAs
TA modules were originally described as plasmid maintenance systems that trigger post-segregational killing of plasmid-free cells (Gerdes et al., 2005). The TA pair is transcriptionally coupled, with the antitoxin typically acting both to repress transcription of the locus and inhibiting the action of the toxin. In response to various stresses, toxin release can lead to growth arrest and antibiotic tolerance (Correia et al., 2006). At least 60 TA genes have been discovered in the genome of M. tuberculosis, a very high number compared to most other sequenced bacteria (Arcus et al., 2005; Pandey & Gerdes, 2005).

TA pairs showed variation in expression between M. tuberculosis and M. bovis. Rv3407-Rv3408 were up-regulated in M. tuberculosis H37Rv (Table 2) and encode a TA pair (belonging to the vapBC family) designated vapBC-44 by Pandey & Gerdes (2005); Rv3407 encodes the antitoxin, and Rv3408 the PilT amino terminus (PIN) domain toxin. M. tuberculosis also showed up-regulation of Rv1397c-Rv1398c, another TA pair designated vapBC-20 (D. P. Pandey & K. Gerdes, unpublished). Comparative proteome analysis of BCG and M. tuberculosis showed that Rv3407 is present at lower levels in BCG than M. tuberculosis (Jungblut et al., 1999). Inactivation of four of the five M. tuberculosis rpf genes, encoding glycolytic hydrolases implicated in resuscitation of dormant bacilli, led to up-regulation of Rv3407/3408; however, inactivation of the one rpf gene which is different between M. bovis and M. tuberculosis, rpfA, had no effect on Rv3407/3408 expression (Downing et al., 2004). Interestingly, an RD1 knockout of M. tuberculosis showed a 10-fold up-regulation of Rv3407 and Rv3408 (Mostowy et al., 2004). However, BCG is deleted for RD1, yet shows low expression of its Rv3407/Rv3408 orthologues (Jungblut et al., 1999), again pointing to defective regulation of the vapBC-44 locus in M. bovis strains.

In M. bovis, Mb0066, encoding a PIN domain protein, was expressed at higher levels than in M. tuberculosis; however, no upstream antitoxin had been annotated (Garnier et al., 2003). Using exhaustive searches for TA loci, Pandey & Gerdes (2005) identified a putative CDS upstream of Mb0066/Rv0065 that they designated vapB-1 and that encodes the antitoxin for Mb0066/Rv0065 (vapC-1). In keeping with M. bovis genome nomenclature, we designated this upstream CDS Mb0066A. Using qRT-PCR analyses we confirmed both the transcription of the Mb0066A/vapB-1 CDS in M. bovis and M. tuberculosis and the increased expression of Mb0066A in M. bovis compared to M. tuberculosis (Fig. 1).

(11K): Fig. 1. Up-regulation of a TA pair in M. bovis compared with M. tuberculosis. (a) Map of the chromosomal location of Mb0066A/Mb0066 in M. bovis. (b) -Fold increase in expression of Mb0066A/Mb0066 in M. bovis (black bars) compared with M. tuberculosis (white bars) during steady-state growth as measured by qRT-PCR. Data are the means of two independent experiments.

The biological importance of TA pair differential expression is unclear. However, evidence from E. coli has shown that some TAs play roles in the generation of drug-resistant persister populations (Correia et al., 2006; Keren et al., 2004). Given the increased emergence of drug-resistant populations of M. tuberculosis, exploration of the function of TA proteins is a high priority, and further studies of the pairs that show differential expression between M. tuberculosis and M. bovis may offer an ideal starting point to elucidate function.

Verification by qRT-PCR
To confirm gene expression differences predicted by array analysis, real-time qRT-PCR was performed. All of the 92 genes showing differential regulation between M. tuberculosis and M. bovis could not be analysed due to cost implications, so a subset of nine was selected that included genes involved in transcriptional regulation (Mb2654c, virS, whiB4), lipid metabolism (Rv1184c, Mb0222), and a TA pair (Rv3407). Fig. 2 compares the -fold changes in expression measured by microarray and qRT-PCR for these nine genes. The results showed that both methods display broad agreement in their measurements of expression for each gene, confirming that experimental procedures and data handling had not introduced any significant bias to the dataset.

(17K): Fig. 2. Confirmation of microarray results using qRT-PCR. Selected genes that show higher levels of expression in M. bovis compared with M. tuberculosis (a), or higher levels of expression in M. tuberculosis compared with M. bovis (b) in steady-state samples, as measured by microarray (black bars) and qRT-PCR (white bars). Data are the means of two independent experiments.

Acid-shock response of M. tuberculosis and M. bovis
Our chemostat experiments defined a core gene set that showed differential expression between M. tuberculosis and M. bovis during steady-state growth. However, we also wished to determine if the response to a perturbation of steady-state was significantly different between the strains. We chose acid shock as this perturbation, a biologically relevant stress experienced by mycobacteria during infection. Cultures were grown to steady-state, then the pH dropped to 5.5 and gene expression was monitored over a 24 h period at 5, 15, 30 and 60 min and 24 h after acid shock. We identified 150 M. tuberculosis genes and 66 M. bovis genes that showed a minimum of a threefold change in expression post-acid-shock compared with the steady-state, zero time point (see supplementary Tables S2 and S3, available with the online version of this paper, for a full listing of all M. tuberculosis and M. bovis genes that were responsive to acid shock).

While M. tuberculosis showed repression of 44 genes in response to acid shock across all time points (Table S2), M. bovis repressed only six genes (Table S3), with just Rv3269/Mb3298, encoding a protein of unknown function, common to both species. In terms of gene induction, 106 M. tuberculosis genes showed increased expression, compared to 60 M. bovis genes, with 15 genes in common (Tables S2 and S3). Temporal changes in gene expression were evident in M. tuberculosis and M. bovis, with some genes showing early induction (5–60 min), while others were only induced at the late (24 h) time point. At the global level it was apparent that M. tuberculosis and M. bovis showed significant differences in their response to acid stress. Below we highlight the commonality and differences in this temporal response.

Common acid-shock response
Two genes encoding putative methyltransferases, Rv1403c/Mb1438c and Rv1405c/Mb1440c, were induced early after acid shock in both M. tuberculosis and M. bovis (Table 3). Rv1404/Mb1439 lies between the genes and encodes a MarR family transcriptional repressor that regulates expression of Rv1405c and Rv1403c (P. Golby, unpublished). Previous work has shown that Rv1405c was strongly up-regulated in macrophages (Schnappinger et al., 2003), while transposon site hybridization studies identified Rv1405c as one of 194 genes predicted to be essential for survival of M. tuberculosis in vivo (Sassetti & Rubin, 2003). The target(s) of Rv1405c and Rv1403c are unknown. Methyltransferase-mediated modification of unsaturated fatty acids to produce cyclopropane fatty acids is a known acid resistance mechanism in E. coli (Grogan & Cronan, 1997). The gene encoding the cyclopropane fatty-acyl synthase, cfa, is also up-regulated in response to low pH in E. coli (Maurer et al., 2005). However, Rv1403c and Rv1405c show low similarity to cyclopropane fatty-acyl synthases, suggesting this is not their function. Identification of the methylated target will help to shed light on the function of Rv1403c/Mb1438c and Rv1405c/Mb1440c both in virulence and in response to stress.

Table 3. M. tuberculosis and M. bovis genes up-regulated in response to acid-shock

Genes involved in PDIM synthesis were up-regulated in both M. tuberculosis and M. bovis, with fadD26, ppsA and Rv2959c showing maximal induction after 5–10 min, with expression then decreasing over the 24 h period (Table 3). ppsB and ppsE also showed up-regulation over the same time frame in M. tuberculosis (Table 4). It has previously been shown that mutants lacking PDIM have increased permeability (Camacho et al., 2001), suggesting that the early up-regulation of PDIM biosynthesis may be a general stress response that attempts to remodel the cell wall, decreasing permeability to damaging agents. This is similar to the suggested role for increased levels of cyclopropane fatty acids in membrane lipids of acid-tolerant E. coli, with conversion of unsaturated to cyclopropane fatty acids leading to a decrease in proton permeability (Grogan & Cronan, 1997).

Table 4. M. tuberculosis and M. bovis genes that showed differential expression between the strains after acid shock

Rv0263c/Mb0269c was up-regulated 30–60 min after acid-shock (Table 3) and its predicted protein sequence shows similarity to allophanate hydrolase subunit 2 which hydrolyses urea carboxylate to carbon dioxide and ammonia. Up-regulation would lead to increased ammonia production, hence acting to buffer the cytoplasmic pH during acid stress. Tubercle bacilli also contain the classic bacterial nickel-containing UreABC enzyme, but the genes encoding this complex were not up-regulated. While pathogens such Helicobacter pylori are known to produce high levels of urease as a means of protection against host acidic environments (Weeks et al., 2000), this is the first report of up-regulation of urease genes in tubercle bacilli in response to acid stress.

Differential acid-shock response
Among the 44 genes repressed in M. tuberculosis in response to acid-shock were two clusters of genes involved in energy transduction (Table 4). Five atp genes encoding components of the membrane-bound proton-translocating ATP synthase were down-regulated in M. tuberculosis. As this system serves to import protons, and hence acidify the cytoplasm, its repression under acid stress would aid in maintaining cytoplasmic pH near neutrality. However, seven nuo genes encoding subunits of the type I NADH dehydrogenase, which serves to export protons, were also down-regulated. As proton export would help to maintain cytoplasmic pH under acid-stress, the repression of the nuo operon is surprising. A previous analysis of the M. tuberculosis acid-shock response also noted down-regulation of genes encoding type I NADH dehydrogenase subunits nuoA and nuoG (Fisher et al., 2002). The repression of both proton influx and efflux systems in M. tuberculosis, and their differential expression between M. tuberculosis and M. bovis, suggests a complex dynamic in the control of these systems in response to acid stress. This does not of course mean that constitutive pH homeostasis systems that direct proton efflux, such as cytochrome respiratory chains (Booth, 1985), are not important.

The kdpABC operon encodes a high affinity K⁺ uptake system that is regulated by turgor pressure in E. coli (Malli & Epstein, 1998). In M. bovis 2122, this operon was found to be induced at the 24 h time point, while in M. tuberculosis H37Rv only kdpD, encoding the sensor kinase that controls kdpABC expression, was up-regulated after 24 h (Table 4). Under conditions of cytoplasm acidification, protons need to be removed from the cytoplasm; however, to maintain electrochemical neutrality, positively charged ions must be introduced. Hence, induction of K⁺ uptake systems allows the cell to maintain electrochemical gradients (Booth, 1985). As proton extrusion would need to occur quickly after acidification of the cytoplasm begins, it is therefore puzzling that kdpABC was induced only at the late time point in M. bovis. For example, for osmotic stress in Salmonella, it has been shown that up-regulation of kdpABC occurs less than 10 min after exposure to the stress (Balaji et al., 2005). This suggests that the up-regulation of kdpABC in M. bovis seen here is either not a direct response to acid stress, or that the temporal response is markedly different in M. bovis compared to other bacteria. Monitoring the expression of kdpABC in M. bovis and M. tuberculosis in response to osmotic stresses will help to unravel the role of K⁺ import for homeostasis in the tubercle bacilli.

Differential expression of amino acid biosynthesis genes in response to acid shock was seen. M. tuberculosis repressed the tryptophan biosynthetic genes trpCBA 30 min after the pH drop, but induced the expression of the rocD2-rocD1-Rv2323c locus over the same time period, reaching maximal expression by 24 h (Table 4). This latter frameshifted locus encodes an ornithine aminotransferase (RocD), and a putative amidinotransferase (Rv2323c), enzymes that are involved in the urea cycle and hence lead to ammonia production. As with up-regulation of ureases, increased production of ammonia will serve to increase cytoplasmic pH in the face of proton efflux.

While the pks4-papA3-mmpL10 locus involved in DAT and PAT synthesis was expressed at lower levels in M. bovis than M. tuberculosis under steady-state conditions (Table 1; see above), this region was up-regulated in M. bovis after acid shock, reaching maximum expression after 24 h (Table 4). In M. tuberculosis, acpM-kasA-kasB, involved in mycolic acid biosynthesis, were down-regulated across all time points, as was efpA; repression of these genes has been shown to be a signature for agents that effect fatty acid synthesis (Boshoff et al., 2004). As with up-regulation of PDIM synthetic loci that was seen in both strains in response to acid shock, altered expression of genes involved in cell wall synthesis suggests remodelling of cell wall architecture in response to stress. Indeed, the hydrophobic cell wall acts as a major protective barrier for the bacillus, so maintaining its integrity is clearly of major importance for the cell.

The dipZ gene, one of the SigK regulon genes whose expression is higher in M. bovis than M. tuberculosis, was up-regulated threefold after acid shock only in M. bovis at the 24 h time point. As this gene is already highly expressed in M. bovis (Table 1), its up-regulation is intriguing. None of the other SigK regulon genes was expressed at higher levels after acid shock, indicating SigK-independent regulation. DipZ is a homologue of DsbD in E. coli, which functions to transfer reducing power from the cytoplasm to the periplasm and plays key roles in the cytochrome c assembly pathway and correct disulfide-bond formation in proteins (Kadokura et al., 2003). The crystal structure of the C-terminal domain of DipZ revealed a dimeric structure with an expected thioredoxin-like fold, but also an unexpected fold with similarity to carbohydrate-binding modules, suggesting a role in carbohydrate processing (Goldstone et al., 2005). As cytochrome respiratory chains are key mediators of pH homeostasis through proton efflux (Booth, 1985), up-regulation of dipZ may reflect an increased requirement for cytochrome biogenesis. However, the unique structural features of DipZ suggest it could play a novel role that remains to be elucidated.

Conclusions
The M. tuberculosis complex can be viewed as a set of ecotypes with distinct host preferences (Smith et al., 2006), the basis of which is encoded in the pathogen's genomes. Comparative analyses of the M. tuberculosis and M. bovis genomes has identified a set of genetic differences between these strains, some of which have been linked to known in vitro phenotypic differences (Keating et al., 2005; Said-Salim et al., 2006; Stermann et al., 2004). However, comparative genomics cannot provide a full picture of the impact of genetic change on phenotype. To accelerate our identification of candidate genes that contribute to the phenotypes of M. tuberculosis and M. bovis, we therefore sought to identify key gene expression differences between the strains.

In defining core gene expression differences it is essential that growth conditions are identical between cultures. We chose continuous culture since it allows environmental conditions and bacterial growth rates to be tightly and reproducibly controlled, reducing variation between biological replicates. A seminal example of the need to control growth rate in transcriptome experiments has been provided by a global analysis of central carbon metabolism in E. coli using gene expression, protein and metabolome data (Ishii et al., 2007). Using glucose-limited chemostat cultures of E. coli, it was found that the effect of variation in growth rate on the transcriptome and proteome was larger than the effects of most gene disruptions in carbon metabolism. We are therefore confident that our profile of steady-state gene expression in M. tuberculosis and M. bovis identifies key gene expression differences between the strains, rather than being simple growth rate or media effects.

In a recent publication, Rehren et al. (2007) described the transcriptome of M. tuberculosis H37Rv and M. bovis Ravanel grown in batch culture, identifying 278 genes with differential expression. Comparing their results with our data, 45 genes appear in both datasets, with 325 gene differences. Whether these differences are due to variation between M. bovis Ravanel and M. bovis 2122, or are down to technical differences between the experiments, is difficult to assess. For example, while we grew both M. tuberculosis H37Rv and M. bovis 2122 on Tween 80 as sole carbon source, Rehren and colleagues grew H37Rv in Middlebrook 7H9 medium with glycerol, while the Ravanel strain was grown in 7H9 medium with pyruvate; hence, some of their gene expression differences may be due to different carbon source utilization. The two studies also used different microarrays and data handling procedures, a further source of variation. Hence, the minimal set of differentially expressed genes between M. tuberculosis and M. bovis remains to be defined.

Prior to our work, Fisher et al. (2002) studied the effect of acid on the transcriptome of M. tuberculosis H37Rv, but in batch culture. Overlap between the genes differentially expressed in response to acid stress identified here and in their work is apparent, with Rv1130, gltA1, PE13, acpM, kasAB, ahpCD and efpA identified in both analyses. However, some key differences in the data were also apparent, such as the induction of a putative polyketide synthase operon (Rv3083-Rv3089) in their work which we did not see in M. tuberculosis or M. bovis. Conversely, the strong induction of the putative methyltransferase-encoding genes Rv1405c and Rv1403c seen by us in both M. tuberculosis and M. bovis was not reported by Fisher and colleagues, although they did see a slight induction of Rv1404, which encodes the putative regulator of Rv1405c and Rv1403c. It is difficult to determine whether the variation in expression profile between the two studies is due to choice of strain, growth condition, experimental procedures or data handling.

In summary, we have defined key gene expression differences between the sequenced M. tuberculosis H37Rv and M. bovis 2122 strains, and have identified candidate genes that may play pivotal roles in the pathogenic process. The focus now is to explore the role of this differential gene expression in the distinct host preference of the human and bovine tubercle bacilli.

This work was funded by the UK Department for Environment, Food and Rural Affairs (Defra), the Department of Health, and the Health Protection Agency. The authors would like to thank Paul Wheeler and Lisa Keating (VLA) for their help in the development of the chemostat growth medium for M. bovis, and for useful discussions on the microarray data. We would like to acknowledge Jason Sawyer (VLA) for developing a protocol for qRT-PCR, to thank Lucy Brooks and Adam Whitney (Bacterial Microarray Group, St Georges) for help with depositing data in µG@Sbase and ArrayExpress, and acknowledge Colorado State University for the provision of M. tuberculosis H37Rv genomic DNA produced under NIH contract HHSN266200400091C/ADB NO1-A1-40091 Tuberculosis Vaccine Testing and Research Materials Contract. The views expressed in the publication are those of the authors and not necessarily those of the Department of Health and Health Protection Agency.

Edited by: J. Parkhill

Footnotes

†Present address: Veterinary Exotic Diseases, Research and Official Controls Division (VEROD) Defra Area 101, 1a Page Street, London SW1P 4PQ, UK.

Supplementary tables of primer sequences used for qRT-PCR reactions, and M. bovis and M. tuberculosis genes regulated in response to acid-shock are available with the online version of this paper.