Cell wall stress responses in Bacillus subtilis: the regulatory network of the bacitracin stimulon



In response to sublethal concentrations of antibiotics, bacteria often induce an adaptive response that can contribute to antibiotic resistance. We report the response of Bacillus subtilis to bacitracin, an inhibitor of cell wall biosynthesis found in its natural environment. Analysis of the global transcriptional profile of bacitracin-treated cells reveals a response orchestrated by two alternative σ factors (σB and σM) and three two-component systems (YvqEC, YvcPQ and BceRS). All three two-component systems are located next to target genes that are strongly induced by bacitracin, and the corresponding histidine kinases share an unusual topology: they lack about 100 amino acids in their extracellular sensing domain, which is almost entirely buried in the cytoplasmic membrane. Sequence analysis indicates that this novel N-terminal sensing domain is a characteristic feature of a subfamily of histidine kinases, found almost entirely in Gram-positive bacteria and frequently linked to ABC transporters. A systematic mutational analysis of bacitracin-induced genes led to the identification of a new bacitracin-resistance determinant, bceAB, encoding a putative ABC transporter. The bcrC bacitracin resistance gene, which is under the dual control of σX and σM, was also induced by bacitracin. By comparing the bacitracin and the vancomycin stimulons, we can differentiate between loci induced specifically by bacitracin and those that are induced by multiple cell wall-active antibiotics.


The bacterial cell envelope is the first and major line of defence against threats from the environment. It is also the target of numerous antimicrobial substances, many of which are produced by soil microbes, presumably to inhibit the growth of competitors. Bacillus subtilis is a ubiquitously distributed soil microorganism and provides an excellent model system to investigate the responses of bacteria to antimicrobial compounds made by other soil bacteria. It is estimated that more than two-thirds of all antibiotics currently in use are natural products of Streptomycetes and other actinomycetes that are abundant in the soil ecosystem (Bentley et al., 2002). In previous studies we identified a fosfomycin resistance gene, fosB, regulated by the antibiotic-inducible σW extracytoplasmic function (ECF) σ factor (Cao et al., 2001) and bcrC, a bacitracin resistance gene, regulated by two other ECF σ factors, σM and σX (Cao and Helmann, 2002).

To further define the transcriptional responses induced by antibiotics in B. subtilis, we have investigated gene induction by bacitracin. Bacitracin is a branched cyclic dodecylpeptide antibiotic produced by B. licheniformis and some strains of B. subtilis (Azevedo et al., 1993; Ishihara et al., 2002). It is synthesized as a mixture of up to 50 different closely related congeners (Kang et al., 2001) and needs a divalent metal ion (most efficiently Cu2+, Mn2+ or Zn2+) for its biological activity (Adler and Snoke, 1962). Bacitracin inhibits bacterial cell envelope biosynthesis by binding very tightly to the long-chain C55-isoprenol (i.e. undecaprenyl) pyrophosphate (Stone and Strominger, 1971; Storm and Strominger, 1973). Undecaprenyl pyrophosphate (UPP) is the lipid carrier responsible for the translocation of cell envelope building blocks from the cytosol to the external side of the cytoplasmic membrane, where they are incorporated in the macromolecular network of the cell envelope (i.e. peptidoglycan, teichoic acids and polysaccharide capsule). Binding of bacitracin to UPP prevents its recycling by dephosphorylation to the monophosphate form that is normally reloaded on the inner face of the membrane.

Bacitracin is a potent narrow spectrum antibiotic directed primarily against Gram-positive cocci and bacilli. It is widely used in topical ‘triple antibiotic’ ointments (along with neomycin and polymyxin B) in the treatment of minor cuts and burns. Its nephrotoxicity limits its systemic use and it is considered only as a last resort for the treatment of gastrointestinal infections (Arky, 1997). However, it is widely used as an animal feed additive in the livestock industry (Huyghebaert and De Groote, 1997). Despite its widespread use, bacitracin resistance is still scarce (Ming and Epperson, 2002).

Three different bacterial bacitracin resistance mechanisms have been described. The self-resistance of the producer B. licheniformis is mediated by an ABC-transport system, encoded by the bcrABC locus (Podlesek et al., 1995). Homologues of the bcrC gene, coding for the membrane-spanning domain of an ABC transporter have been described as resistance determinants in B. subtilis and Escherichia coli (Harel et al., 1999; Cao and Helmann, 2002; Ohki et al. 2003a). A second mode of bacitracin resistance is through de novo synthesis of active undecaprenyl phosphate by an undecaprenol kinase. This mechanism has been found in E. coli, Streptococcus pneumoniae and Staphylococcus aureus (Cain et al., 1993; Chalker et al., 2000). A role of exopolysaccharide production in bacitracin resistance has been reported for Xanthomonas campestris, Sphingomonas sp., E. coli and S. mutans (Pollock et al., 1994; Tsuda et al., 2002).

In this paper, we report that treatment of B. subtilis with bacitracin leads to the transcriptional induction of numerous genes controlled by at least two alternative σ factors and three two-component regulatory systems, including a second bacitracin resistance locus. Bacitracin is preferentially sensed by a subfamily of histidine kinases that appear to lack a significant extracytoplasmic sensing domain. Comparison of the bacitracin and vancomycin stimulons reveals extensive overlap, but also uniquely regulated systems.


The yvqIH operon is induced by cell wall antibiotics

In a previous study, we examined the response of B. subtilis to the glycoside antibiotic vancomycin, an inhibitor of bacterial cell wall biosynthesis (Cao et al., 2002). Most induced genes were part of the regulons controlled by two ECF σ factors, σM and σW. However, the most strongly induced gene in these experiments was yvqI, showing a more than 100-fold increase in expression level 10 min after vancomycin addition (Cao et al., 2002). For this reason, we chose this locus for further investigation.

The genetic organization indicates that yvqI is part of a two-gene operon with yvqH. The gene yvqI codes for a hydrophobic, 126-amino-acid protein probably localized to the cell membrane. The product of the yvqH gene shows significant similarity to phage-shock proteins such as PspA of E. coli. No obvious promoter structure was found upstream of yvqI and no significant yvqIH expression could be detected in uninduced cultures.

Disc-diffusion assays were used to determine the spectrum of stimuli that induce the expression of yvqIH. The B. subtilis strains BFS2469 and BFS2470, harbouring pMUTIN-insertions in yvqH and yvqI, respectively, were plated on a medium containing the indicator dye Xgal and overlaid with filter disks impregnated with various antibiotics that interfere with cell wall biosynthesis. The pMUTIN-insertion results in a disruption of the targeted gene and generates a transcriptional fusion to lacZ, thereby allowing detection of both antibiotic sensitivity (as defined by the zone of inhibition around the disk) and inducibility (by the formation of a blue ring at the edge of the zone of inhibition) (Cao et al., 2002). The disruption of yvqI and yvqH had no effect on the sensitivity of the mutants to any of the antibiotics tested. However, vancomycin, bacitracin and, to a lesser extent, tunicamycin induced the expression of both genes (data not shown). β-Galactosidase assays were performed in liquid culture to quantify induction of yvqI. The resulting ∼65-fold induction by vancomycin (234 Miller units after vancomycin induction versus 3.7 Miller units uninduced) correlates very well with the data previously obtained by microarray analysis (Cao et al., 2002). Addition of tunicamycin only resulted in a threefold induction under these conditions (11.6 Miller units). The strongest effect was obtained with bacitracin (∼ 2000 Miller units). Therefore we chose bacitracin for further studies.

The global transcriptional response of B. subtilis to bacitracin

We used DNA microarray technology to define the overall transcriptional response of B. subtilis to bacitracin. RNA was extracted from cultures of wild-type strain CU1065 and the yvqH mutant HB0920 grown in LB medium to mid-logarithmic growth phase (OD600≈ 0.45) that were either left untreated or treated with 100 µg ml−1 bacitracin for 5 or 15 min. The latter strain, a yvqH-deletion mutant, was included in the analysis because the well-characterized YvqH-homologue in E. coli, PspA, functions as a repressor of transcription (Adams et al., 2003; Bordes et al., 2003). Because yvqH is strongly induced by bacitracin, we reasoned that it might be possible to identify target genes for this putative transcriptional regulator. The results of bacitracin treatment for HB0920 are shown in Fig. 1 and summarized in Table 1. Similar results were obtained with CU1065 (see below).

Figure 1.

Graphical presentation of the bacitracin stimulon. The fold changes of gene-expression levels of the data-set 5 min after bacitracin addition (on the x axis) are plotted against the fold-changes 15 min post-induction (on the y axis; values of both data-sets relative to the expression level in the uninduced control experiment). The scatter-plot represents the data sets obtained for the yvqH deletion strain HB0920 (yvqH was still detected as bacitracin inducible due to the presence of a remaining region near the 5′ end of the gene). The most strongly induced gene signals are highlighted and circled (see text). Members of the σM (□) and σB (▴) regulons are highlighted. All other gene signals are represented as small grey diamonds. Most of the genes that were significantly repressed are involved in sugar uptake and general carbon metabolism. These probably reflect metabolic consequences of antibiotic stress and will not be further considered here. Note that for the generation of this graphical display the data sets were not filtered (see Experimental procedures), so some of the background signals (small grey diamonds) that appear to represent highly regulated genes may not be significant.

Table 1. Genes induced by zinc-bacitracin in HB0920 that were not part of the σB regulon.
Gene(s)Fold inductionaRegulatorBacitracinb MIC (µg ml−1)(put.) Functions
5 min15 min
  • a

    . Highest fold change of transcriptional units (usually the first gene). Except as noted, most of these changes have not been confirmed by independent techniques.

  • b

    . The MIC is shown for strains with a deletion mutation in the indicated gene. The reference value for the wild-type strain CU1065 is 500 µg ml−1.

  • c

    . Due to the deletion of yvqH in this mutant, the strong induction of this locus was only detectable for the yvqI gene.

  • d

    . In a published study on this locus, YtrA was identified as the regulator of this operon, binding to a σA.-type promoter upstream of ytrA (Yoshida et al., 2000). Treatment of cultures with bacitracin, as well as with vancomycin (Cao et al., 2002), only induced ytrBCDEF, but not ytrA, suggesting an alternative and inducible promoter in front of ytrB.

Genes that were induced (≥five fold) after 5 min
yvqIH c 7721062YvqEC500unknown
bceAB (ytsCD)282 365BceRS (YtsAB) 10ABC transporter
czcD  61   3.8ZneRNDZn-efflux
cadA (yvgW) 58   9.8ZneRNDZn-efflux
czcO (trkA) 33   2.8ZneRNDK-uptake, Zn-resistance
yhcYZ-yhdA 21  14YvqEC500two-component system
yvcRS  12  25YvcPQ500ABC transporter
ydhK  12   4.4 500unknown
bcrC  10   9.5σMX 25bacitracin resistance prot.
cotY   9.4   5.9 NDspore coat protein
ytrBCDEF d   9.1   9.7YtrAd500acetoin utilization
gerAA/AB   7.1   4.2 NDspore germination
ycgRQ   6.7   7.9 500unknown
yvqGFEC   6.1   7.0YvqEC500unknown, unknown, TCS
ykvS   5.9  10 500unknown
yybR   5.4   3.7 NDunknown
yjbIH   5.2   7.1 NDunknown
Genes that were induced (≥ten fold) after 15 min
ygaCD   2.0  12 500unknown, ABC-ATP
ykuNOP   1.4  12FurNDflavodoxin homologue
ytzB   2.8  11 500unknown
yetG   1.4  10 NDunknown

Most of the significantly induced genes were targets of two alternative σ factors: σM, an ECF σ factor, and σB, the regulator of the general stress response. The most strongly induced gene of the σM regulon was bcrC, showing a ∼10-fold increase in expression both at 5 and 15 min post-induction. BcrC is homologous to membrane spanning proteins of ABC transporters and has been reported to be a bacitracin resistance determinant in two independent studies (Cao and Helmann, 2002; Ohki et al., 2003a). Whereas the σM regulon showed similar induction ratios 5 and 15 min after bacitracin addition, the general stress response was more strongly induced at the early time point, with a significant decrease in expression by 15 min after addition of bacitracin. The same bias towards an early response was observed for the three genes of the ZneR (formerly YozA) regulon that codes for zinc-efflux functions (A. Gaballa and J.D.H., unpubl. data). This induction was probably due to the use of the zinc salt of bacitracin for this study, which is biologically one of the most active forms of bacitracin (Adler and Snoke, 1962). The final concentration of Zn-bacitracin (100 µg ml−1) corresponds to 67.3 µM Zn2+. The ZneR regulon is induced above 10 µM Zn2+ (A. Gaballa and J.D.H., unpubl. data).

In addition to yvqIH, the genes most strongly induced by bacitracin were bceAB (formerly ytsCD) (280-fold at 5 min/360-fold at 15 min) and yvcRS (12-/24-fold), both coding for ABC transporters, and the yhcYZ-yhdA locus (20-/14-fold), encoding a two-component system and an azoreductase homologue respectively. The four genes directly downstream of yvqIH, yvqGFEC, were also significantly induced (six-/sevenfold) by bacitracin.

A number of additional genes showed a delayed response to bacitracin addition, with significant induction only after 15 min (Table 1). The ygaCD genes encode proteins with homology to the membrane-spanning and nucleotide-binding domains of bacterial ABC transporters (Higgins, 2001). There are no defined homologues in the database for the products of ytzB and yetG. The ykuNOP locus is weakly induced, together with other members of the Fur regulon (Baichoo et al., 2002). This induction is probably an indirect effect due to elevated zinc levels (A. Gaballa and J.D.H., unpubl. data).

The bacitracin stimulon in the wild-type strain CU1065 appears to be very similar to that of the yvqH-deletion mutant HB0920 as judged by a comparison of mRNA levels at 5 min after treatment (See Experimental procedures and Supplementary material, Fig. S1). The most noteworthy difference was the yhcYZ-yhdA locus, which was not induced by bacitracin in the wild type, but showed a 20-fold increase of expression in HB0920. This locus is therefore a candidate for negative regulation by YvqH. In addition, expression of the bceAB and yvcRS loci was threefold greater in CU1065 than in the yvqH mutant when both were measured 5 min after bacitracin treatment. These minor differences (compared with the overall induction rate for these loci) may be due to slight variations between samples.

The regulatory network of the bacitracin stimulon includes three two-component systems

The microarray experiments demonstrated a very strong transcriptional response of the yvqIH (GFEC), bceAB and yvcRS loci to bacitracin. Candidate regulators for these genes were previously identified based on the observation that in each case adjacent genes encode a two-component regulatory system (Joseph et al., 2002) (Fig. 2A). These authors demonstrated that overexpression of each response regulator induced the linked genes coding for the ABC-transporters. This finding was confirmed in a second comprehensive study, which analysed the genome-wide transcriptional changes in mutants overexpressing response regulators using DNA microarrays (Kobayashi et al., 2001). This study additionally suggested that YvcP might regulate the bceAB genes and that YxjL might contribute to the induction of the yvqIH(GFEC) locus.

Figure 2.

A. Graphic representation of the yvqIHGFEC, bceRSAB and yvcPQRS regions on the chromosome. Hatched arrows represent response regulators (RR), dotted arrows histidine kinases (HK) and checkered/striped arrows represent the nucleotide-binding and membrane-spanning domains of ABC transporters respectively. Unknown genes are shown in grey and the genes flanking the region of interest are white. Predicted rho-independent transcription terminators are indicated by stem-loops. The line corresponds to a size of 7.5 kb for all three regions.
B. Northern blot analysis to identify the regulators of bacitracin- inducible genes. Expression of bcrC (lanes 1–4), yvqH (lanes 5–10), bceR (lanes 11–16) and yvcR (lanes 17–20) was measured using 10 µg of total RNA from each sample separated on a 1% formaldehyde gel. RNA was transferred to a nylon membrane and hybridized with a radiolabelled DNA fragment containing ∼500 nucleotides of the coding region of each gene. ‘–’ represents the uninduced control lane, ‘+’ the RNA sample from cultures induced with bacitracin for 15 min (final concentration 100 µg ml−1). ‘wt’ indicates the wild-type strain CU1065. The isogenic mutants bearing deletions in transcriptional regulators are indicated. Arrows indicate the major transcript(s) for each gene, their sizes correspond to transcripts covering bcrC (lanes 1–4), yvqIH (black arrow) and yvqIHGFEC (grey arrow) (lanes 5–10), bceAB (lanes 11–16) and yvcRS (lanes 17–20).

To investigate the role of these regulators in the observed bacitracin induction, we constructed allelic replacement mutants of the four response regulator genes (bceR, yvcP, yvqC and yxjL) by long-flanking homology (LFH)-PCR (Wach, 1996), resulting in strains HB0927, HB0931, HB0933 and HB0936 respectively (Table 2). Each mutant strain, as well as HB0031 (sigM::kan) and CU1065, was grown in LB medium to mid-logarithmic phase (OD600≈ 0.45) and treated with 100 µg ml−1 bacitracin for 15 min. RNA was isolated from treated and untreated control cultures and gene expression monitored by Northern blot hybridization (Fig. 2B). For all four genes analysed (bcrC, yvqH, bceA and yvcR), the bacitracin-dependent induction was verified in the wild-type strain. In the case of yvqH and bceA (Fig. 2B lanes 5–10 and 11–16 respectively), no residual expression could be detected in the untreated cultures, consistent with the fact that expression in the untreated controls was below the detection limit in the DNA microarray experiment (data not shown).

Table 2. Strains used in this study.
StrainGenotypeaReference, source, or primers used for constructionb
  • a

    . All strains are derivatives of Bacillus subtilis strain CU1065 (W168 trpC2 attSP β).

  • b

    . Underlined primer sequences correspond to the 5’ and 3’ end of the resistance cassettes used to replace the gene(s) by LFH-PCR (see Experimental procedures).

HB0031 sigM::kan Cao and Helmann (2002)
HB0106 bcrC::pMUTIN Cao and Helmann (2002)
BFS2469 yvqH::pMUTINZoltan Pragai (Harwood lab)
BFS2470 yvqI::pMUTINZoltan Pragai (Harwood lab)
HB0937 bcrC::pMUTIN, bceAB::kanThis study

The bcrC gene is under the dual control of two ECF σ factors, σX and σM, but bacitracin induction is mediated by σM only (Cao and Helmann, 2002; Ohki et al., 2003a). This finding is confirmed by this Northern analysis. A single band corresponding to a transcript of ∼0.6 kb size can be detected in all four lanes (Fig. 2b, lanes 1–4). This corresponds to the size of bcrC, which forms a monocistronic transcription unit. The bacitracin-dependent induction is lost in the sigM deletion mutant. While the σX regulon is not part of the bacitracin stimulon, the residual bcrC expression is σX-dependent, as shown previously (Cao and Helmann, 2002).

Two major transcripts were detected with the yvqH probe (Fig. 2B, lanes 6 and 10). The strongest signal corresponds to a ∼1.1 kb band consistent with the hypothesis that the yvqIH genes form an operon. A larger partially degraded transcript of about 4.2 kb probably represents the whole yvqIH-yvqGFEC locus. Both transcripts were only present in bacitracin-treated cultures of CU1065 and the yxjL mutant HB0936. No signal was detected in the yvqC mutant strain indicating that expression of yvqIH is completely dependent on YvqC in the presence of bacitracin (Fig. 2B, lanes 5–10) or vancomycin (data not shown). Similarly, the bacitracin-dependent induction of bceAB (∼2.7 kb transcript) was completely dependent on BceR (Fig. 2B, lanes 11–16) and yvcRS was dependent on the linked response regulator, YvcP (Fig. 2B, lanes 17–20). We note that induction of the bceAB operon was unaffected in the yvcP mutant and induction of yvqIH was unaffected in the yxjL mutant. These results are in contrast to the overlapping regulation observed in DNA microarray experiments under conditions of response regulator overexpression (Kobayashi et al., 2001).

The bacitracin-sensing histidine kinases share an unusual N-terminal sensing domain and are linked to ABC transporters in Gram-positive bacteria with a low G+C content

Most histidine kinases are modular transmembrane proteins with an extracytoplasmic input domain and a cytoplasmic transmitter domain (Parkinson, 1993). Due to the diversity of stimuli sensed, the N-terminal domains of histidine kinases show almost no sequence similarity. In contrast, the highly conserved C-terminal transmitter domain harbours the phosphorylation activity common to all members of this protein family. Analysis of the three bacitracin-sensing histidine kinases (BceS, YvqE and YvcQ) revealed an unusually short N-terminal domain (∼60–70 amino acids) due to the almost complete lack of a linker region between the two deduced transmembrane helices (BceS, YvqE: <5 amino acids, YvcQ: ∼15 amino acids).

The apparent lack of an extracytoplasmic signal input domain in the bacitracin-sensing kinases results in an overall protein length of 360 amino acids or less, compared with approximately 450 amino acids for most EnvZ-like histidine kinases. To identify other sensor kinases with a similar domain structure, we used the simple modular architecture research tool (SMART, http://smart.embl-heidelberg.de) (Schultz et al., 1998). BceS and YvqE belong to two different subclasses of histidine kinases (Grebe and Stock, 1999) and only YvqE contains a HAMP domain (Aravind and Ponting, 1999). These two protein sequences were used as matrices to identify histidine kinases with a similar overall domain organization. Two partially overlapping groups of 1905 and 667 histidine kinases, respectively, were retrieved. The resulting pool of proteins was filtered to identify those of less than 400 amino acids total length with an N-terminal domain of not more than 100 amino acids, including two putative transmembrane helices separated by no more than 20 amino acids. Only 45 sensor kinases fit these criteria (see Supplementary material, Table S1), including BacS and MbrD, two histidine kinases known to be linked to bacitracin resistance in B. licheniformis and S. mutans, respectively (Neumuller et al., 2001; Tsuda et al., 2002), and VraS, an antibiotic-responsive sensor kinase from S. aureus (Kuroda et al., 2003). Based on their unusual topology, we suggest that these 45 proteins define a unique subfamily (see Discussion). Remarkably, 90% of these histidine kinases (40/45) were found in Gram-positive bacteria (see Supplementary material, Table S1).

We noted that, like the B. subtilis bceRS-bceAB and yvcQP-yvcRS systems, the bacitracin-sensing two-component systems of B. licheniformis and S. mutans are genetically linked to genes coding for an ABC-transporter. To determine if this genetic linkage is a common feature, we analysed the adjacent chromosomal regions for all 45 identified two-component systems (Supplementary material, Table S1). Consistent with a previous analysis (Joseph et al., 2002), we found adjacent ABC transporter gene clusters in the Bacillus/Clostridium group (low G+C) of Gram-positive bacteria (24/33), but not in the high G+C group (0/7). Multiple sequence alignment and the resulting phylogenetic tree demonstrated a close evolutionary relationship among this subset of histidine kinases that are encoded adjacent to ABC transporters (Supplementary material, Figs S2 and S3). However, the three histidine kinases involved in bacitracin resistance in B. licheniformis, S. mutans and B. subtilis are not tightly clustered within this group (Neumuller et al., 2001; Tsuda et al., 2002).

The bceAB operon encodes a new bacitracin resistance determinant

Bacitracin is synthesized non-ribosomally by B. licheniformis and some strains of B. subtilis (Azevedo et al., 1993; Ishihara et al., 2002). The self-resistance mechanism that was reported for B. licheniformis consists of an ABC transporter, BcrABC, which is thought to function as a bacitracin pump. The expression of this transporter is induced by bacitracin and regulated by a two-component system BacRS (Neumuller et al., 2001). Prior to this work, only one resistance determinant had been reported for B. subtilis: BcrC, homologous to the eponymous protein of B. licheniformis (Cao and Helmann, 2002; Ohki et al., 2003a).

The presence of two loci encoding bacitracin-inducible ABC transporters suggested that B. subtilis may contain additional resistance determinants. To test this hypothesis, we engineered deletion mutants of strongly induced genes with homology to ABC transporters or of unknown function using LFH-PCR (Table 2). Measurements of the minimal inhibitory concentration (MIC) using a microtitre plate-based assay confirmed the bacitracin sensitivity of the sigM and bcrC mutants (Fig. 3). Note that the bcrC mutant is more sensitive to bacitracin than the sigM mutant due to the residual σX-dependent expression of bcrC (Cao and Helmann 2002; Fig. 2B, lanes 1–4). Additionally, these studies revealed a role in bacitracin resistance for the bceR and bceAB genes (Fig. 3), consistent with the recent results of Ohki et al. (2003b). BceA and BceB show high sequence similarity to nucleotide-binding and membrane-spanning proteins of ABC transporters from many closely related Gram-positive bacteria, but no specific homology to BcrA or BcrB from B. licheniformis (Neumuller et al., 2001). Both the bceR and bceAB deletions led to a 50-fold reduction in MIC compared with CU1065. Significantly, a bcrC bceAB double mutant (HB0937) was significantly more sensitive than either single mutant with a 200-fold decrease of bacitracin resistance compared with the wild-type strain (2.5 versus 500 µg ml−1; Fig. 3). This indicates that bcrC and bceAB form two independent bacitracin resistance pathways, as also suggested by Ohki et al. (2003b).

Figure 3.

Effects of bacitracin on growth of Bacillus subtilis strains. All strains were grown in microtitre plates for 4 h after dilution into LB medium containing the indicated concentrations of bacitracin. The experiment was done in triplicate and a representative result is shown. Strains used: CU1065 (▪), ‘bceR’ (HB0927, ◊), ‘bceAB’ (HB0928, ◆), ‘sigM’ (HB0031, ○), ‘bcrC’ HB0106 (•), ‘bcrC/bceAB’ (HB0937, ▴) Compare also with Table 1.

Comparison of the bacitracin and vancomycin stimulons

In a previous study, we analysed the vancomycin stimulon (Cao et al., 2002). To gain some understanding of the spectrum and specificity of the cell wall stress responses of B. subtilis, we compared the vancomycin and bacitracin stimulons in a scatter plot (Fig. 4). Both antibiotics induce the σM and the σB regulons and, in both cases, yvqIH was the most strongly induced locus. In contrast, the σW regulon is induced by vancomycin but not bacitracin, whereas bceAB, yvcRS and the ZneR-regulon specifically respond to bacitracin (Fig. 5).

Figure 4.

Graphical comparison of the vancomycin (x axis) and bacitracin (y axis) stimulon of CU1065. The most strongly induced gene signals are highlighted and circled (see text). Members of the σM (□), σW (○) and σB (▴) regulons are highlighted. All other gene signals are represented as small grey diamonds. Note that for the generation of this graphical comparison, the data sets could not be filtered to remove low-quality and non-reproducible signals, so some of the background signals (small grey diamonds) that appear to represent highly regulated genes are not significant.

Figure 5.

Graphical overview of the regulatory network of the cell wall stress response of B. subtilis. The regulatory pathways are indicated by arrows. Dotted arrows and grey regulator symbols indicate regulation not linked to the cell wall stress response. Regulons shared by both stimulons are underlaid in light grey. Identified bacitracin resistance determinants are framed.


Treatment of bacteria with antibiotics elicits numerous changes in gene expression. With the advent of tools for the global analysis of gene expression, it has become possible to survey antibiotic-induced stimulons and efforts are underway to define the component regulons. Significantly, different mechanistic classes of antibiotics result in characteristic profiles of gene expression, as monitored by either transcriptome (Wilson et al., 1999; Goh et al., 2002; Hong et al., 2003; Ng et al., 2003) or proteome studies (Bandow et al., 2003), and this provides a tool for defining the mechanism of action of novel compounds (Rosamond and Allsop, 2000).

In general, the stimulon induced by antibiotic treatment is controlled by regulators that either directly sense the presence of antibiotic or sense the biochemical consequences of antibiotic action. In addition, antibiotics elicit a plethora of indirect effects. Examples of regulators that directly sense antibiotic(s) include the S. aureus penicillin-binding proteins BlaR1 and MecR (Zhang et al., 2001; Golemi-Kotra et al., 2003), and regulators of antibiotic efflux including the TetR (Berens and Hillen, 2003), QacR and BmrR proteins (Godsey et al. 2002). Other resistance genes are regulated by biochemical changes resulting from drug treatment. One well-characterized example is Citrobacter freundii AmpR, a regulator that senses antibiotic-induced changes in the pools of peptidoglycan precursors (Jacobs et al., 1997). In many other well-studied antibiotic-inducible systems the nature of the inducing signal is not yet known. Examples include the VanRS two-component system regulating inducible vancomycin resistance (Ulijasz et al., 1996; Pootoolal et al., 2002) and regulons controlled by alternative σ factors (Michelle et al., 1999; Cao et al. 2002; Thackray and Moir, 2003).

The bacitracin stimulon

In this study, we examined the bacitracin stimulon of B. subtilis using global transcriptome analysis. Under the conditions tested, approximately 60–70 genes were induced greater than fivefold. The most strongly induced genes were bceAB (formerly ytsCD) and yvcRS, coding for putative ABC transport systems and yvqIH, encoding a putative transmembrane-protein and a phage-shock protein homologue respectively (Fig. 1). Interestingly, two BceAB homologues (VraDE and VraFG) are upregulated in vancomycin-resistant S. aureus (Kuroda et al., 2000), and the vraDE locus is inducible by vancomycin (Kuroda et al., 2003). Thus, there is evidence linking this family of ABC transporters to antibiotic resistance in several species.

Both the bceAB and yvcRS systems are positively regulated by a linked two-component regulatory system (Fig. 2), consistent with previous transcriptome analyses of strains engineered to overexpress the corresponding response regulators (Kobayashi et al., 2001). However, the overproduction of response regulator proteins may also lead to non-specific effects: a number of the putative target genes as judged from the response regulator overexpression studies (Kobayashi et al., 2001) were not induced by bacitracin. In addition, we did not detect significant cross-regulation between these and other two-component systems as proposed by Kobayashi et al. (2001). Of the three bacitracin-responsive two-component systems only YvqEC seemed to slightly autoregulate its own expression and in each case the target genes studied were dependent on only a single two-component system.

In addition to these three two-component systems, bacitracin also induced the σB and σM regulons. Induction of both of these regulons was also detected upon vancomycin treatment. Whether the induction of the σB regulon was mediated by the energy or the environmental stress pathway remains to be investigated. The σM regulon, which is known to be induced by diverse stresses (Horsburgh and Moir, 1999; Thackray and Moir, 2003), is regulated by the yhdLK gene products that function as anti-σ factors (Thackray and Moir, 2003). These proteins are thought to be membrane-bound sensors that keep the σ factor in an inactive state through protein–protein interactions. An incoming signal results in the release of the corresponding σ factor, thereby activating the expression of its target genes (Helmann, 1999). The σM regulon includes bcrC, a previously characterized bacitracin resistance determinant (Cao and Helmann, 2002). The BceAB transport system defines a second, apparently independent bacitracin resistance pathway.

Bacitracin resistance genes in B. subtilis

B. licheniformis and B. subtilis are closely related organisms (Priest, 1993) and bacitracin-producing strains can be found in both species (Azevedo et al., 1993; Ishihara et al., 2002). The B. subtilis bcrC and bceAB genes encode components of ABC transporters. Overall, this is functionally analogous to the bcrABC self-resistance mechanism described for the bacitracin-producing strain B. licheniformis. It is therefore surprising that the B. subtilis bceRSAB system shows no significant sequence similarity to B. licheniformis bacRS-bcrAB, although the sensor kinases do share an unusual topology (see below). In B. licheniformis, the bcrABC genes are in one operon and encode an ABC transporter consisting of a monomer of each membrane spanning domain (BcrBC) and a dimer of the nucleotide-binding domain BcrA (Podlesek et al., 1995). This locus is under the negative control by the BacRS two-component system (Neumuller et al., 2001). The regulation of the B. subtilis bceAB and bcrC resistance genes by two separate regulatory systems suggests that they may function independently. Indeed, a bcrC bceAB double mutant is significantly more sensitive to bacitracin than either single mutant (Fig. 3). The ability of these systems to function independently is supported by the observation that overexpression in B. subtilis of B. licheniformis bcrC or bcrAB alone results in an intermediate increase in resistance compared with the expression of the complete bcrABC locus (Podlesek et al., 2000). Expression of bcrB and bcrA alone did not result in an increased resistance, whereas expression of a bcrAC fusion locus increased resistance almost to the level of the complete locus (Podlesek et al., 2000).

The role of ABC-transporters in mediating bacitracin resistance is still poorly understood. The bactericidal action of bacitracin is a result of interaction with undecaprenyl pyrophosphate (UPP). The inactive UPP is normally recycled to the monophosphate form by a specific, but so far unidentified, pyrophosphatase required to allow the reloading of the carrier lipid on the cytoplasmic side of the membrane. Bacitracin prevents recycling by titrating active lipid-carrier out of the cycle. It has been suggested that BcrABC functions according to the ‘hydrophobic vacuum cleaner’ model (Podlesek et al., 1995), analogous to multidrug-efflux pumps of tumour cells (Higgins and Gottesman, 1992). In this model, bacitracin is taken up by the transporter directly from the hydrophobic environment of the membrane. Surprisingly, the yvcPQRS locus (Fig. 2A) encodes proteins highly similar to the BceRSAB proteins (24–48% identity; Joseph et al., 2002) yet these genes play no role in bacitracin resistance. The function of this ABC transporter remains to be investigated, but the fact that it is inducible by bacitracin suggest that it may be involved in resistance to peptide antibiotics.

A family of intramembrane-sensing histidine kinases

We have identified a subfamily of sensor kinases that share the unusual domain organization noted for the bacitracin-inducible histidine kinases of B. subtilis. We speculate that these kinases may sense signals associated with the cell membrane and suggest the name of intramembrane-sensing histidine kinases for this group. The signal sensed by these kinases is not yet known, but it seems reasonable to suggest that those identified in this study may sense perturbations of lipid II-dependent processes. A fourth member of this subfamily in B. subtilis, YbdK, is not induced by bacitracin and the signal regulating this kinase is still unknown. Conversely, the CseC histidine kinase of Streptomyces coelicolor also senses bacitracin (as well as other cell wall antibiotics), without showing the features of an intramembrane-sensing histidine kinase (Hong et al., 2002). Intriguingly, the set of two-component systems identified in our analysis (based on the unusual topology of the histidine kinase sensor domain) corresponds closely to those shown previously to be genetically linked to ABC transporters, at least in low-G+C Gram-positive bacteria (Joseph et al., 2002).

This association may reflect a further functional link. We speculate that BceS may sense bacitracin–UPP complexes by interaction with the transmembrane helices, although an indirect mechanism involving sensing of perturbations of cell envelope structure cannot be ruled out. In addition to serving as a sensor for UPP-bound bacitracin, BceS may also deliver this substrate to the ABC transporter thereby facilitating removal. Such a mechanism could also offer an explanation for a puzzling finding in B. licheniformis: due to the negative regulation mechanism a bacRS-deletion mutant still expresses the BcrABC system, but it no longer confers bacitracin resistance (Neumuller et al., 2001).

The yvqIH locus shows the most dramatic response to vancomycin and bacitracin without conferring resistance to either antibiotic. This locus is controlled by the YvqEC two-component system. The orthologous system in S. aureus, VraSR, has recently been found to control a large regulon (∼46 genes) also inducible by bacitracin and vancomycin (Kuroda et al., 2003). In contrast, the B. subtilis YvqEC system controls a significantly smaller regulon. Whereas YvqI is a small hydrophobic protein of unknown function, YvqH shows significant homology to E. coli PspA, which acts as a repressor by inhibiting the transcriptional activator of psp expression, PspF, through protein–protein interaction (Adams et al., 2003; Bordes et al., 2003). Like PspA, YvqH might play a regulatory role: the yhcYZ-yhdA operon is induced by bacitracin only in a yvqH-deletion mutant, but not in the wild-type strain (see Supplementary material, Fig. S1).

Concluding remarks

The comparison between the vancomycin and the bacitracin responses (Fig. 4) is a first step to differentiate between specific and general cell wall-stress responses in B. subtilis. The two stimulons share three regulons (shown graphically in Fig. 5): the σB-dependent general stress response, the σM-regulon and the YvqC-target genes yvqIH(GFEC). The general stress response is a strong but transient response to cell wall antibiotic stress, consistent with the known transient nature of σB activation in response to other stresses (Hecker and Volker, 2001; Petersohn et al., 2001; Price et al., 2001). In contrast, the σM-regulon shows a constant level of induction under all conditions tested. σM is one of seven ECF σ factors in B. subtilis (Helmann, 2002) and its regulon is induced by salt, heat shock, acid and ethanol stress (Thackray and Moir, 2003). Several other regulons appear to respond to antibiotic-specific signals. For example, the σW regulon is induced by vancomycin, but not by bacitracin. Conversely, the bceAB resistance determinant is selectively induced by bacitracin. Unravelling this complex regulatory network will require studies to identify the signals that activate each of these antibiotic-inducible regulons.

Experimental procedures

Bacterial strains and growth conditions

B. subtilis was routinely grown in Luria–Bertani (LB) medium at 37°C with aeration. All strains used in this study are derivatives of the laboratory wild-type strain CU1065 (W168 trpC2 attSPβ). All strains are listed in Table 2.

Determination of the minimal inhibitory concentration (MIC)

MIC-assays were performed in microtitre plates using a ‘Tecan Spectra Rainbow’ microtitre plate reader. Pre-cultures were inoculated from fresh overnight LB plates and incubated at 37°C with aeration until they reached an OD600≈ 0.45. Next, 10-fold dilution of these cultures were inoculated in a total volume of 200 µl well−1 with increasing quantities of bacitracin, ranging from 0.01 to 1000 µg ml−1 (final concentration). The plates were incubated at 37°C, and the OD600 was read after 4 h when cultures had reached their final cell density. All experiments were performed in triplicate.

Allelic replacement mutagenesis using long flanking homology (LFH-)PCR

We adapted LFH-PCR to generate chromosomal deletions of the genes listed in Table 2. The protocol is modified from the published procedure (Wach, 1996). In brief, resistance cassettes were amplified from a suitable vector as template (Guerout-Fleury et al., 1995; Youngman, 1990). Two primer pairs were design to amplify ∼1000 bp DNA fragments flanking the region to be deleted at its 5′ and 3′ end. The resulting fragments are here called ‘up’ and ‘do’ fragment. The 3′ end of the up-fragment as well as the 5′ end of the do-fragment extended into the gene(s) to be deleted in a way that all expression signals of genes up- and downstream of the targeted genes remained intact. Extensions of ∼25 nucleotides were added to the 5′ end of the ‘up-reverse’ and the ‘do-forward’ primers that were complementary (opposite strand and inverted sequence) to the 5′ and 3′ end of the amplified resistance cassette. All obtained fragments were purified using the PCR-purification kit from Qiagen. Next, 150–200 ng of the up- and do-fragments and 250–300 ng of the resistance cassette were used together with the specific up-forward and do-reverse primers at the normal concentration in a second PCR-reaction. In this reaction, the three fragments were joined by the 25 nucleotide overlapping complementary ends and simultaneously amplified by normal primer annealing. The PCR products were purified with the PCR-purification kit from Qiagen and directly used to transform B. subtilis. Transformants were screened by direct colony-PCR, using the up-forward primer with a reverse primer annealing inside the resistance cassette.

All PCR-reactions were done in a total volume of 50 µl using the HotStar DNA-Polymerase Mastermix from Qiagen. The primers used to amplify the flanking regions for this study are listed in Table 2. A detailed protocol for LFH-PCR (‘LFH-PCR.doc’) as well as a list of the templates and primers used (‘LFH-PCR.xls’) can be downloaded from: http://www.micro.cornell.edu/faculty.JHelmann.html.

Measurement of induction by β-galactosidase assays

Cells were inoculated from fresh overnight LB plates and grown in LB medium at 37°C with aeration until they reached an OD600≈ 0.45. Next, 2 ml of cultures were harvested (uninduced control) and the cell pellets were shock frozen and kept at −70°C. The cultures were induced by addition of antibiotics to a final concentration of: bacitracin (100 µg ml−1), tunicamycin (50 µg ml−1) or vancomycin (2 µg ml−1) and incubated for additional 30 min at 37°C. Next, 2 ml of the cultures were harvested as described above (induced sample). The pellets were resuspended in 1 ml of working buffer and assayed for β-galactosidase activity as described with normalization to cell density (Miller, 1972).

Preparation of total RNA for Northern blot and microarray analysis

For Northern analysis total RNA was extracted from 5 ml of B. subtilis culture, with and without bacitracin. Bacitracin was added to the culture at OD600≈ 0.45 (mid-log phase) and the cultures were incubated for 15 min at 37°C with aeration before the cells were harvested and shock-frozen. RNA preparation was performed using the RNeasy kit (Qiagen) according to protocol.

For microarray analysis, 100 ml of LB medium was inoculated from a fresh overnight LB plate and incubated at 37°C with aeration until the culture reached an OD600≈ 0.45, when the culture was split: 30 ml served as an uninduced control. To the remaining culture bacitracin was added to a final concentration of 100 µg ml−1 and 30 ml samples were taken 5 and 15 min after addition. The cells were harvested by centrifugation at room temperature, and cell pellets were shock-frozen and stored at −70°C for at least 30 min. RNA was extracted using the ‘hot phenol method’ as described previously (de Saizieu et al., 1998). After extraction, the RNA was purified using the RNA clean-up protocol of the RNeasy kit (Qiagen) with on-column DNase treatment in order to remove abundant small RNA molecules (tRNAs and 5S rRNA) and residual genomic DNA.

Probe preparation and Northern blot analysis

Internal fragments of 500–750 nucleotide lengths were amplified by PCR using the following primer pairs: bcrC (CCAAGCTTCAGAATCCCCCCAGAAA, AAGAATTCGAAG AAAACAAGAGAT), yvcR (TATCATACCAAGCGCTCAGCG, CTTGCTGCTGTGGCATCATGCG), bceA (CAGGAAGTGC TGAAGGGCATCG, CGTTGCGTTTTTGATTGAGCTGGCTC AGC), yvqH (GGAGGAATCAGGTATGG, CTTGACCGCAA ATCCTTCC). The PCR-fragments were purified using the Qiagen PCR-Purification kit and 100 ng of each fragment was labelled with [α-32P]-dATP (New England Nuclear; 3000 Ci mmol−1, 10 mCi µ−1l) by random oligonucleotide-primed synthesis using the Klenow-fragment of DNA-polymerase (3′→5′ exo, New England Biolabs) according to protocol (Current protocols, 3.5.9–10). Unincorporated [α-32P]-dATP was removed by NucAway spin columns (Ambion).

Northern analysis was carried out using the NorthernMax formaldehyde-based system (Ambion) according to the instruction manual. In brief, 10 µg total RNA was denatured and loaded on a 1% formaldehyde agarose gel. After electrophoresis, the RNA was transferred to Zeta-Probe blotting membrane (Bio-Rad) in a downward transfer setup. The RNA was cross-linked by exposing the damp membrane to UV-light (1 min at λ = 302 nm). The blot was pre-hybridized at 42°C for 30 min, and the labelled probe (preheated to 95°C for 10 min) was added to the hybridization tube. Hybridization was performed overnight at 42°C. On the next day, the membrane was washed twice with low stringency buffer (2× SSC) at room temperature for 5 min followed by two high-stringency washes (0.1× SSC at 42°C for 15 min). The blot was wrapped in plastic wrap, exposed to a phosphor screen (Molecular Dynamics) and analysed using a Phosphor Imager (Molecular Dynamics).

Microarray analysis

DNA microarrays contained 4020 B. subtilis genes and consisted of PCR products printed in duplicate onto glass slides (Amersham Pharmacia Biotech) as previously described (Ye et al., 2000). Each slide contains 9220 features corresponding to duplicate copies of each open reading frame (ORF), additional PCR products for some ORFs, rRNA genes and other controls. RNA preparations were used to synthesize Cy3- and Cy5-labelled cDNA and hybridization was performed as described (Ye et al., 2000, 2001). All comparisons were performed twice (once each with Cy3 and Cy5) to control for possible differences in labelling efficiency between fluorophores. Fluorescent signal intensity data were quantified using ArrayVision software (Molecular Dynamics) and normalized to the total detectable mRNA. Mean fluorescence intensity was set to 1.0 with a value of 0.1 corresponding to background. Each expression ratio is represented by at least four separate measurements (duplicate spots on each of two slides). Competitive hybridizations were performed for strain HB0920 at 5 and 15 min after bacitracin addition compared with an untreated control. Problems with the untreated (control) RNA samples from the wild-type strain precluded a direct determination of fold induction. However, competitive hybridizations indicate that the expression profile 5 min after treatment is very similar in both HB0920 and the wild-type parent and each of the strongly induced regulons (as determined in HB0920) were confirmed using Northern analysis in the wild-type parent strain (Fig. 2B).

For most analyses, the micorarray datasets were filtered to remove those genes that were not expressed at levels significantly above background in either condition (sum of mean fluorescence intensity < 0.30; this typically reduces the size of the data files from 4610 lines to ∼2800 lines). In addition, the mean and standard deviation of the fluorescence intensity were computed for each gene (based on two signals on each of two slides) and those where the standard deviation was greater than the mean intensity were removed (this removes another ∼30 to ∼80 genes; typically those with a strong signal for only one or two of the four spots). Finally, control spots corresponding to rRNA genes were removed. The remaining fluorescence values (e.g. Supplementary material, Fig. 1S) were used for data display. For some comparisons (e.g. Figs 1 and 4) this filtering was omitted for ease in aligning the two different datasets. Complete datasets are available at http://www.micro.cornell.edu/faculty.JHelmann.html.

Supplementary material

The following material is available from http://www.blackwellpublishing.com/products/journals/suppmat/mmi/mmi3786/mmi3786sm.htm and http://www.micro.cornell.edu/faculty.JHelmann.html

Fig. S1. Gene expression in the yvqH mutant HB0920 compared with gene expression in the isogenic wild-type strain CU1065 after bacitracin induction (5 min).

Fig. S2. Multiple sequence alignment of histidine kinases with a domain organization similar to BceS and YvqE of Bacillus subtilis.

Fig. S3. Phylogenetic tree derived from the alignment of Fig. S2.

Table S1. Histidine kinases with domain organization similar to BceS and YvqE of Bacillus subtilis.


We thank Dr Min Cao for helpful discussions and Zoltan Pragai for providing strains BFS2469 and BFS2470.

This work was supported by NIH grant GM-47446 (to J.D.H).