Characterization of Mycobacterium tuberculosis Rv3676 (CRP_Mt), a Cyclic AMP Receptor Protein-Like DNA Binding Protein

Abstract

Little is known about cyclic AMP (cAMP) function in Mycobacterium tuberculosis, despite its ability to encode 15 adenylate cyclases and 10 cNMP-binding proteins. M. tuberculosis Rv3676, which we have designated CRP_Mt, is predicted to be a cAMP-dependent transcription factor. In this study, we characterized CRP_Mt's interactions with DNA and cAMP, using experimental and computational approaches. We used Gibbs sampling to define a CRP_Mt DNA motif that resembles the cAMP receptor protein (CRP) binding motif model for Escherichia coli. CRP_Mt binding sites were identified in a total of 73 promoter regions regulating 114 genes in the M. tuberculosis genome, which are being explored as a regulon. Specific CRP_Mt binding caused DNA bending, and the substitution of highly conserved nucleotides in the binding site resulted in a complete loss of binding to CRP_Mt. cAMP enhanced CRP_Mt's ability to bind DNA and caused allosteric alterations in CRP_Mt conformation. These results provide the first direct evidence for cAMP binding to a transcription factor in M. tuberculosis, suggesting a role for cAMP signal transduction in M. tuberculosis and implicating CRP_Mt as a cAMP-responsive global regulator.

Tuberculosis (TB) remains a serious global health problem that is growing at an estimated rate of 3% per year (49). This TB epidemic is exacerbated by an unexplained synergy with human immunodeficiency virus and steadily increasing rates of drug resistance that are a by-product of lengthy treatment regimens (15, 20). A better understanding of Mycobacterium tuberculosis biology is needed to improve treatment and develop a more effective vaccine. A key area of interest is how M. tuberculosis senses and responds to the environments it encounters during host infection.

Cyclic AMP (cAMP) is a critical signaling molecule in many bacterial and eukaryotic cells. The role of cAMP signal transduction in mediating catabolite repression has been well characterized in Escherichia coli, and this forms the paradigm for cAMP-mediated gene regulation in prokaryotes (7, 10, 11, 16, 33, 36). A class I adenylate cyclase (AC) in E. coli catalyzes the synthesis of cAMP, which then transduces the signal by binding cAMP receptor protein (CRP) and activating it as a transcription factor (18).

cAMP signaling is also critical for virulence in a diverse range of pathogens, including yeast, fungi, bacteria, and parasites (3, 12, 19, 25, 37, 38, 42, 50, 70). In some cases, cAMP regulates virulence genes within the pathogen (3, 38, 42). For example, CRP-cAMP signaling is essential for virulence in Salmonella enterica serovar Typhimurium (17) and has recently been shown to control virulence-associated type III secretion systems in Pseudomonas aeruginosa and Yersinia enterocolitica (50, 70).

The M. tuberculosis genome contains 15 putative class III adenylate cyclase genes (46). The activity of at least 10 of these cyclases has been confirmed with biochemical assays (13, 26, 40, 41, 61, 64), making it likely that cAMP contributes substantially to signal transduction in M. tuberculosis. We recently identified the first cAMP-regulated genes in M. tuberculosis by using an exogenous cAMP culture model (24). Some of these genes are upregulated during intracellular growth in macrophages (29), suggesting that cAMP signaling may be important to M. tuberculosis during its interaction with the host. This observation is intriguing in light of a previous study that reported elevated levels of cAMP in macrophages that showed an impairment of phagosome-lysosome fusion upon infection with Mycobacterium microti (44).

The mechanism of cAMP-mediated gene regulation in M. tuberculosis has not been explored. We previously reported that the M. tuberculosis Rv3676 protein belongs to a superfamily of proteins that contain both cAMP binding and helix-turn-helix (HTH) DNA binding motifs (46), and we hypothesized that the Rv3676 protein is a CRP-like transcriptional regulator in M. tuberculosis. For this study, we used experimental and computational approaches to define Rv3676's DNA binding sequence and characterize its interactions with DNA and cAMP. We designated the M. tuberculosis Rv3676 gene crp and the encoded protein CRP_Mt, based on the results. We also identified 114 members of a putative CRP_Mt regulon, implicating CRP_Mt as a biologically relevant global regulator of transcription in M. tuberculosis.

MATERIALS AND METHODS

Expression and purification of CRP_Mt.

crp was PCR amplified from M. tuberculosis H37Rv using primers KM977 (5′-GGAATTCGACGAGATCCTGGC CAGGGC) and KM963 (5′-AAGCTTGCGTGCGACTCTGTGTCTGC) with EcoRI and HindIII restriction sites (underlined). The amplified DNA fragment was cloned into pET28a+ (Novagen) using the EcoRI and HindIII sites to generate pMBC372, verified by sequencing, and maintained in E. coli BL21(DE3). CRP_Mt expression was induced from a bacterial culture (optical density at 600 nm, 0.6) with 1 mM isopropyl-β-d-thiogalactopyranoside (IPTG) for 3 h and confirmed by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and Western blot analysis using an anti-His6 monoclonal antibody (Clontech). For purification, a 200-ml culture was pelleted and resuspended in 1 ml lysis buffer containing 50 mM Tris-HCl (pH 8.0), 50 mM NaCl, 1 mM dithiothreitol, and 1% protease inhibitor cocktail (Sigma). Bacteria were lysed by two freeze-thaw cycles (−70°C-37°C) and sonication for 5 min, followed by two additional freeze-thaw cycles, and the lysate was cleared by centrifugation at 13,000 × g for 10 min. CRP_Mt was purified using a Hitrap affinity column (Amersham Biosciences) per the manufacturer's instructions, and the eluted protein was dialyzed against phosphate-buffered saline with 10% glycerol. The protein concentration was measured with a NanoOrange protein quantitation kit (Molecular Probes) and diluted to 2 mg/ml before being stored in aliquots at −70°C.

Systematic evolution of ligands by exponential enrichment (SELEX)-based capture of CRP_Mt-binding ligand DNA.

Mycobacterium bovis BCG genomic DNA was digested to completion with Sau3AI and ligated to a linker sequence (5′CGAATTCAGGAAACAGCTATGTTAATTAA3′) prepared with a Sau3AI-compatible sticky end. A crude extract from 100 ml of His-CRP_Mt-expressing culture was applied to 100 μl His Mag agarose beads (Novagen), which were then washed and equilibrated with DNA binding buffer [10 mM Tris-HCl (pH 8.0), 50 mM KCl, 1 mM EDTA, 50 μg/ml bovine serum albumin, 1 mM dithiothreitol, 0.05% nonionic P-40 detergent, 100 μM cAMP, 20 μg/ml poly(dI-dC), and 10% glycerol] per the manufacturer's directions. Five micrograms of DNA fragments in 100 μl of DNA binding buffer was added to the CRP_Mt-bound beads for 15 min at room temperature. The beads were washed three times to remove nonspecific DNA before eluting CRP_Mt-DNA complexes in a volume of 100 μl. Ten microliters of this eluted protein was heated for 10 min at 95°C and used as a template for PCR with primer KM1040 (5′-CGAATTCAGGAAACAGCTATG). The resulting DNA product was cloned into the TA vector (Invitrogen). Individual E. coli DH5α transformants were picked into Luria broth containing 25 μg/ml kanamycin, grown overnight, and then used as templates for PCRs with primer KM1040 to amplify the captured insert fragments. PCR products of 200 to 300 bp were selected for further electrophoretic mobility shift assay (EMSA) analysis. The purity of the protein used for capturing DNA was also evaluated by SDS-PAGE.

EMSA.

A ³³P-end-labeled DNA probe (0.05 pmol) was used in each 10-μl binding reaction mixture as described by others (60), with modifications. Briefly, purified His-CRP_Mt (at specified concentrations) and DNA probes were incubated for 30 min at room temperature in DNA binding buffer prior to electrophoresis on a nondenaturing 8% polyacrylamide gel for 2 to 3 h at 14 V/cm in 0.5× Tris-borate-EDTA. A 200- to 500-fold excess of unlabeled DNA fragments was used for competition experiments. Gels were vacuum dried, exposed on a phosphor screen, scanned with a Storm 860 PhosphorImager (Molecular Dynamics), and analyzed with ImageQuant software.

Additional assays.

The effects of cAMP on CRP_Mt conformation were examined by treating 5 μg of CRP_Mt with 0.2 or 1 μg trypsin (Sigma) for 10 min at 37°C, as described by others (35). Half of the digested protein was assayed in a 15% SDS-PAGE gel, and a portion of the remainder was diluted for use in EMSA.

For DNA-bending experiments, five 156-bp DNA fragments were PCR amplified from different locations within the Rv0884c-Rv0885 intergenic region. Fragment end points were as follows: F1, positions 982623 to 982778; F2, positions 982590 to 982745; F3, positions 982554 to 982709; F4, positions 982528 to 982683; and F5, positions 982495 to 982650. The forward primer of each fragment was phosphorylated with [γ-³³P]ATP to generate end-labeled probes using PCR. CRP_Mt (35 nM) was used for EMSA to compare the mobilities of the binding complex for each DNA fragment.

Sequence analysis.

We identified the set of promoter-containing regions in the M. tuberculosis genome as those sequences that were upstream of a gene and contained at least 20 bp of intergenic sequence, as defined by M. tuberculosis H37Rv annotation (GenBank accession no. NC_000962.1) (14). This set of M. tuberculosis intergenic promoter regions consists of 2,066 sequences, totaling 346,025 bp of searchable sequence after masking a 43-bp repeat sequence.

Regulatory motifs were predicted using the Gibbs sampler (67). For applications to subsets of intergenic sequences or DNA fragments from CRP_Mt trap experiments, where most sequences were believed to contain a common pattern, the Gibbs recursive sampler was used with the following parameters: either one or two motif models were specified, where each model was 16 bp allowed to fragment to 24 bp and allowed to choose an even- or odd-width palindromic model, based on the sequence evidence; a position-specific background model (43) was incorporated; uninformed priors (i.e., no prior information on the motif models) were used; a maximum of two sites per sequence was allowed; and run parameters consisted of 5,000 iterations with a plateau period of 500 iterations and reinitializing 50 times using a random seed. For applications to the genome-scale data set, the Gibbs motif sampler was used with the following parameters: the motif model was 16 bp allowed to fragment to 24 bp and allowed to choose an even- or odd-width palindromic model, a position-specific background model (43) was incorporated, prior information was specified by providing a motif model and was weighted to provide three to five pseudocounts per motif position, the estimated number of sites was 50, and run parameters were as described above. The two motif models used as prior information were (i) an alignment of 87 experimentally verified (DNase I footprinted) E. coli CRP binding sites, weighted to provide five pseudocounts per motif position; and (ii) an alignment of putative CRP_Mt binding sites from the M. tuberculosis genome, weighted to provide either three, four, or five pseudocounts per motif position.

RESULTS

Identification of CRP_Mt binding sites.

Experimental and computational approaches were combined to identify CRP_Mt binding sites. One hundred sixty-six M. bovis BCG genomic DNA fragments were captured using a SELEX-based technique and then screened by EMSA with purified CRP_Mt, as described in Materials and Methods. Five of these DNA fragments showed obvious mobility retardation in the presence of CRP_Mt (data not shown). These DNA sequences were analyzed using the Gibbs sampler (see Materials and Methods), and a common sequence pattern consistent with the E. coli CRP motif model was identified in all five DNA fragments (Table 1; Fig. 1A).

TABLE 1.

Sequence alignment of DNA ligands of CRP_Mt captured by SELEX and EMSA

Clone	Aligned sequencea	Genome coordinatesb
1D1	ccgtc TGTGAGCAAGATCACA tagct	4110823-4110835 (cr)
1B5	gtatc TGTGACTAAGGTCACA gacgc	13463-13478 (cr)
2D3	ctcta TGTGACGAAGCCCACA tcgac	4329936-4329951 (nc)
S16	tgcaa TGTGATATTAGCCTCA ttttg	1178962-1178977 (nc)
3G1	gttgt GGTGACCGCCGTCACA gcgac	2518157-2518172 (cr)

^aSites detected by Gibbs sampling in each clone are shown in capital letters; the underlined sequence for clone 1D1 is from the DNA linker.

^bM. tuberculosis (entry NC_000962.1) genome coordinates of the aligned sites (capital letters). The locations of these regions with respect to annotated protein coding regions or noncoding regions are indicated by “(cr)”and “(nc),”respectively.

FIG. 1.

Sequence logos of the E. coli CRP motif model (A) and the putative CRP_Mt motif model (B). The E. coli CRP motif model represents 87 experimentally identified (DNase I footprinted) binding sites; this motif model was used as prior information in the prediction of CRP_Mt binding sites (see Materials and Methods). The putative CRP_Mt motif model represents 58 predicted sites (see Table 2) from the set of intergenic regions from the M. tuberculosis genome. Sequence logos depict the relative frequency of each base at each position of the motif. The y axis indicates the information content measured in bits, and error bars represent standard deviations at each position due to the limited sample size (58).

Binding of CRP_Mt to the predicted sites within the recovered 1B5 and 2D3 DNA fragments (Table 1) was confirmed by EMSA. A 40-bp intergenic DNA fragment upstream of Rv1624c was used as a nonspecific control. The presence of CRP_Mt retarded, in a dose-dependent manner, the migration of 28-bp synthetic oligonucleotides containing the predicted DNA binding site of either 1B5 or 2D3 (Fig. 2A). The reappearance of the free probes in the presence of a 500-fold excess of unlabeled probe DNA confirmed the specificity of these CRP_Mt-DNA interactions (Fig. 2A). In addition, binding of the 2D3 probe could be competed with an unlabeled 2D3 or 1B5 oligonucleotide, but not by the same concentration of the nonspecific DNA control (Fig. 2B).

FIG. 2.

EMSA experiment showing binding of CRP_Mt with motifs identified in captured DNA sequences. (A) The mobility of 28-bp synthesized probes containing motifs of 1B5 and 2D3 fragments was retarded by CRP_Mt, and the free probes reappeared in the presence of a 500-fold excess of unlabeled probe DNA, as specified. (B) The shift of the 2D3 28-bp motif probe by CRP_Mt could be competed by either 2D3 or 1B5 unlabeled motif DNA, but not by the 40-bp intergenic DNA upstream of Rv1624c (1624c) that was used as a negative control. This control Rv1624c DNA probe also failed to bind to CRP_Mt. The labeled DNA probe, unlabeled competitor DNA (cold DNA), and amount of CRP_Mt (nM) that was used are specified for each lane.

Together, these results indicate that specific CRP_Mt binding sites are contained within the 28-bp oligonucleotides and that the same protein domain was responsible for CRP_Mt binding to both 1B5 and 2D3 DNA. However, only two of the predicted binding sites in the SELEX-trapped fragments are within annotated intergenic regions, so we pursued additional computational analyses to refine the CRP_Mt binding motif and identify potential regulon members.

An alignment of the helix-turn-helix regions of CRP_Mt and E. coli CRP and fumarate nitrate reductase regulator (FNR) (Pfam 17.0, PF00325) (4) showed several conserved amino acids in the DNA recognition helix (Fig. 3). We also found using EMSA that CRP_Mt could bind the CRP binding site in the E. coli lac promoter (5′-TAATGTGAGTTAGCTCACTCAT-3′), but not the FNR binding site in the E. coli ndh1 promoter (5′-AAACTTGATTAACATCAATTTT-3′) (data not shown). These observations were consistent with our identification of a CRP-like pattern for the CRP_Mt SELEX DNA fragments and suggested that information on the E. coli CRP binding motif could be used to predict the CRP_Mt binding motif.

FIG. 3.

Sequence alignment of helix-turn-helix DNA recognition domains of CRP_Mt and E. coli CRP (Ec CRP) and FNR (Ec FNR) (Pfam 17.0, PF00325) (4). CRP and FNR amino acid residues that form hydrogen bonds with DNA bases are underlined.

We used the Gibbs motif sampler (see Materials and Methods) with prior information on the E. coli CRP motif model (Fig. 1A) to detect a putative CRP_Mt binding motif in the intergenic regions of the M. tuberculosis genome. We initially identified a subset of 29 M. tuberculosis intergenic regions that contained CRP-like sites in at least 3 of 20 independent Gibbs sampling runs. The Gibbs recursive sampler was then used to detect a motif model de novo (i.e., without prior information) for this set of 29 intergenic regions. This preliminary CRP_Mt model was then used as prior information to analyze the intergenic regions of the M. tuberculosis genome again (see Materials and Methods). These analyses provided a set of 55 M. tuberculosis intergenic regions that contained a site in at least 1 of 30 independent Gibbs sampling runs, and the Gibbs recursive sampler was used to detect a motif model de novo for these 55 sequences (Fig. 1B).

The resulting motif model consists of 58 sites from the 55 input sequences and resembles the central core of the E. coli CRP binding motif, with a palindrome of two 5-bp half-sites separated by a 6-bp spacer (Fig. 2B; Table 2). Several of these 55 intergenic regions are positioned between divergently transcribed genes, such that a total of 73 genes are immediately flanking and downstream of these intergenic regions. Furthermore, several of these genes are likely to be the first gene of a polycistronic operon, suggesting that the 55 intergenic regions contain promoters that regulate the expression of ∼114 genes. These regulon candidates include genes reported to be starvation (34 genes) or hypoxia (16 genes) regulated, members of the PE and PPE families (10 genes), or essential for M. tuberculosis growth in culture medium (9 genes) (5, 8, 57, 62). The remaining 55 genes are mostly uncharacterized.

TABLE 2.

CRP_Mt binding sites at the intergenic region of the M. tuberculosis H37Rv genome

Probabilitya	Sampling frequencyb	Site sequence	Genomic coordinates	Flanking gene(s)c	Distance(s) upstreamd
0.92	1.00	CAGGTCACACCAGTCACAGA	1571-1590	Rv0002 (dnaN)	−462
0.98	1.00	CACGTCAGCAAAGTCACGAT	23788-23807	Rv0019c	−51
0.97	0.69	ACCGTGAATCTGGCGACGCC	161521-161540	Rv0134 (ephF)	−229
0.59	1.00	AATGTGATCCTCTTCGCGTG	301656-301675	Rv0249c	−3
0.92	1.00	ACGGTCGCAACAGTCACATG	309913-309932	Rv0256c (PPE2)	−368
0.99	0.91	CGCGTCGCAGGATTCACACT	518531-518550	Rv0429c (def) / Rv0430	−137/−181
0.92	1.00	AGTGTAACGCATATCACGTG	542072-542091	Rv0451c (mmpS4) /Rv0452	−164/−49
0.97	1.00	TAGGTGACCAAACTCACGCT	542901-542920	Rv0453 (PPE11)	−252
0.46	1.00	AGTGTGAGCTGTATTACATG	549630-549649	Rv0457c/Rv0458	−25/−24
0.97	0.53	ATTGTGAATCTGGCGACGCG	747949-747968	Rv0651 (rplJ)	−306
0.98	0.60	GGCGTCGCGGGATTCACACT	747981-748000	Rv0651 (rplJ)	−274
0.78	1.00	ATCGTGGGCTTGCTGACGCG	754621-754640	Rv0658c	−214
0.96	1.00	GATGTCAGCTACTCCACAAC	780543-780562	Rv0680c/Rv0681	−129/−157
0.97	1.00	ATCGTGGAACGGCTGACAAC	898731-898750	Rv0805	−79
0.95	1.00	AGCGTGATTCTGGCGACGCC	948516-948535	Rv0851c/Rv0852 (fadD16)	−50/−22
0.93	0.86	TATGTGGGTAATGTCACATG	965958-965977	Rv0867c (rpfA)	−425
1.00	1.00	AGTGTGAACAAGCTCACATG	982625-982644	Rv0884c (serC)/Rv0885	−73/−116
0.99	1.00	ATTGTGAGTTGGATCACGTT	1061787-1061806	Rv0950c/Rv0951 (sucC)	−135/−156
0.97	1.00	ACTGTGACTGGCGCGACACG	1112239-1112258	Rv0996	−124
0.70	1.00	CTCGTGACCTACGTGGCAGC	1220408-1220427	Rv1092c (coaA)/Rv1093 (glyA1)	−224/−145
1.00	1.00	GATGTGACTCAAGTGACACG	1284911-1284930	Rv1158c/Rv1159	−51/−60
0.98	1.00	GCCGTGAGGGGCGTCACGGT	1327660-1327679	Rv1185c (fadD21)	−150
0.47	0.66	TTCGTGAACTAGATCACCAT	1374256-1374275	Rv1230c	−61
0.97	0.85	ATTGTGAATCTGGTGACGCG	1404643-1404662	Rv1256c (cyp-130)	−42
0.97	0.55	CGCGTCGTGCGGTTCACACT	1404675-1404694	Rv1256c (cyp-130)	−74
0.90	1.00	CGTGTGACCAAACTCACCGC	1561327-1561346	Rv1386 (PE15)	−116
0.96	1.00	ATTGTGGCCCACGCCACGTC	1582120-1582139	Rv1405c/Rv1406 (fmt)	−153/−25
1.00	1.00	AATGTGATCTAGGTCACGTG	1757393-1757412	Rv1552 (frdA)	−267
0.99	0.98	GTTGTGATTTCAGTCACGCG	1774835-1774854	Rv1566c	−217
0.99	0.83	GCCGTGACGCATGTGACGTT	1784400-1784419	Rv1581c	−101
0.99	0.68	TGCGTCACATGCGTCACGTT	1784468-1784487	Rv1581c	−169
0.99	1.00	ACCGTCACAGCTGTCACACG	1992785-1992804	Rv1759c (wag-22)/Rv1760	−210/−347
0.91	1.00	CTTGTGGTGTTGTTCACAAC	2014548-2014567	Rv1779c/Rv1780	−71/−130
0.99	1.00	GCCGTGAGATTCGTCACGTC	2052776-2052795	Rv1810	−136
0.96	1.00	AATGTCAGTGATCTGACGGC	2219565-2219584	Rv1976c/Rv1977	−316/−168
0.99	1.00	ATCGTCACGCACGTCACGGC	2234038-2234057	Rv1990c	−403
0.95	0.63	ACCGTCACCTGCGTCACCGC	2334300-2334319	Rv2077c/Rv2078	−8/−738
0.97	1.00	GACGTCGCATCGGTCACAGC	2386129-2386148	Rv2124c (metH)/Rv2125	−64/−143
0.97	1.00	GCTGTCAAATCCGTCACGAA	2610746-2610765	Rv2336	−70
0.85	1.00	GTCGTGAATTGGGTGACCAA	2620124-2620143	Rv2342	−127
0.91	1.00	CGGGTCACTTCGATCACGCG	2677687-2677706	Rv2385 (mbtJ)	−21
0.98	1.00	AGTGTGGCCAAGGTGACAGC	2692237-2692256	Rv2396 (PE_PGRS41)	−541
0.97	1.00	GTGGTGAGCTGGTTCACACC	2704451-2704470	Rv2406c/Rv2407	−16/−225
0.96	1.00	GCGGTGATCGGCGTCACGCC	2705800-2705819	Rv2408 (PE24)	−196
1.00	1.00	CGCGTGACATGTGTCACATG	2921509-2921528	Rv2591 (PE_PGRS44)	−21
0.97	1.00	AGTGTGATTTACATCACATA	3015060-3015079	Rv2699c/Rv2700	−98/−121
0.91	0.95	AATGTGAGGTTCACGACGCG	3412018-3412037	Rv3050c	−64
0.99	1.00	GGCGTCGCCAGATTCACGCT	3427147-3427166	Rv3063 (cstA)	−74
0.96	1.00	GACGTCGGCCGCGTCACGTT	3675815-3675834	Rv3295	−228
0.99	1.00	AACGTCATCTTTGCCACGTT	3874764-3874783	Rv3454	−35
0.99	1.00	GATGTGATGCACTTGACATC	4091145-4091164	Rv3650 (PE33)	−66
0.99	1.00	GTCGTGAACGGTTTGACGGT	4092932-4092951	Rv3652 (PE_PGRS60)	−678
0.92	1.00	GCTGTGAACCCTGCGACGCA	4169322-4169341	Rv3724A (cut5a)	−262
0.99	1.00	ACTGTGACCACGGCCACGCT	4178090-4178109	Rv3729	−173
0.97	1.00	GTTGTGACTTACCTCACCGT	4317080-4317099	Rv3843c/Rv3844	−487/−1673
0.96	1.00	TATGTGACGAAGCCCACATC	4329934-4329953	Rv3857c	−323
0.99	1.00	CATGTCAAAACATTGACAAT	4360091-4360110	Rv3879c	−312
0.98	1.00	GATGTGAAATCTGCGACGCC	4376153-4376172	Rv3893c (PE36)	−161

^aProbability that the site belongs in the alignment.

^bFrequency with which the site was sampled.

^cGene(s) encoded downstream of the site; for sites found between divergently transcribed genes, the two genes are listed separated by a slash.

^dDistance upstream of the start codon of the gene listed in the previous column.

CRP_Mt binding to computationally predicted sites within the M. tuberculosis genome.

Two of the 58 sites that were computationally predicted in the M. tuberculosis genome (Table 2), including a “good” site and a “poor” site, were examined for CRP_Mt binding. The good site in the Rv0884c-Rv0885 intergenic region was selected because of its reproducibility. This site was consistently sampled (frequency, 1.00) during 5,000 Gibbs sampling iterations and had a high probability (1.00) of belonging to the CRP_Mt motif model (Fig. 1B). In contrast, the poor site upstream of Rv1230c was sampled much less frequently (0.66) and had a relatively low probability (0.47) of belonging to the model, making it the weakest of the putative regulon members. Both sites were capable of binding to CRP_Mt when tested by EMSA as 20-bp DNA probes, although the Rv0884c-Rv0885 site bound CRP_Mt more efficiently (Fig. 4).

FIG. 4.

EMSA showing CRP_Mt interactions with representative DNA binding sites identified by Gibbs sampling. DNA probes are as follows: A, the full-length Rv0884c-Rv0885 intergenic region, which contains a motif with a high probability of belonging to the CRP_Mt motif model; B, the 20-bp predicted binding site in the Rv0884c-Rv0885 intergenic region; and C, the 20-bp predicted binding site upstream of the Rv1230c open reading frame, which has a low probability of belonging to the motif model. For each lane, the CRP_Mt concentration (nM) is noted, and “cold DNA” specifies unlabeled competitor DNA fragments used at a 500-fold excess relative to the labeled probe DNA. The 40-bp Rv1624c nonspecific DNA was used as a negative control throughout.

Substitution of highly conserved nucleotides in CRP_Mt-binding DNA motif.

Contact between CRP and specific nucleotides within the DNA binding site is crucial for CRP-DNA interactions (30), and we reasoned that this might also be true for CRP_Mt. We used the sequence of the Rv0884c-Rv0885 site to examine the specificity of CRP_Mt binding when the strongly conserved nucleotides at positions 4 and 17 (Fig. 1B) were altered. Native and modified sites within 20-bp oligonucleotides and a 250-bp DNA fragment encompassing the entire Rv0884c-Rv0885 intergenic region were labeled and subjected to EMSA with CRP_Mt.

A complete loss of binding to CRP_Mt occurred with modified probes that contained either a G-to-C change at position 4 or a C-to-G substitution at position 17 (Fig. 5A). These modified DNA probes also failed to compete CRP_Mt binding to either the 20-bp native site probe (Fig. 5A) or the 250-bp intergenic DNA probe (Fig. 5B). These results indicate that in the context of the Rv0884c-Rv0885 site, these bases are essential for specific binding by CRP_Mt, and suggest that the highly conserved G4 and C17 positions in the motif model reflect the importance of these positions for CRP_Mt-DNA interactions.

FIG. 5.

Binding of CRP_Mt to native and modified sequences in the Rv0884c-Rv0885 intergenic sequence. (A) Sequences of the 20-bp native and modified sites are shown below the figure. Modified sites are marked with asterisks. The mobility of the native probe was retarded by CRP_Mt, and the reappearance of free probe was observed when a 500-fold excess of unlabeled native DNA or 1B5 or 2D3 DNA was present. The same concentration of unlabeled modified DNA and the Rv1624c 40-bp nonspecific DNA failed to compete the binding of the probe. Modified probes failed to bind CRP_Mt. (B) Binding of CRP_Mt to the Rv0884c-Rv0885 full-length intergenic DNA could be competed by a 500-fold excess of unlabeled 1B5 or 2D3 DNA, as well as by the native Rv0884c probe, but not the modified probes, as shown in panel A. “Cold DNA” refers to the competitor. The CRP_Mt concentration (nM) is noted for each lane in panel A and was 20 nM for panel B.

Evidence for allosteric alteration of CRP_Mt upon cAMP binding.

cAMP binding to CRP induces conformational changes that affect CRP's ability to interact specifically with DNA, resulting in cAMP-CRP-mediated gene regulation in E. coli (33). In this study, we used several approaches to test the ability of CRP_Mt to bind cAMP and to examine the effect of cAMP on CRP_Mt's DNA binding ability. The effect of the cAMP concentration on the CRP_Mt-DNA interaction was first measured by EMSA, using cAMP levels from 1 to 100 μM in the binding reaction buffer. The affinity of the protein-DNA complex was highest with 100 μM of cAMP (Fig. 6A).

FIG. 6.

Effect of cAMP on affinity of CRP_Mt-DNA interaction, as measured by EMSA. (A) Binding of CRP_Mt to the 2D3 intergenic DNA probe in the presence of different amounts of cAMP in the binding buffer, as specified. (B and C) Comparison of CRP_Mt binding affinities in the presence and absence of 100 μM cAMP. (B) 2D3 probe; (C) Rv0884c-Rv0885 intergenic region probe. CRP_Mt concentrations are shown in nM.

Complex formation was further examined with several additional probes in the presence or absence of 100 μM cAMP. CRP_Mt bound approximately twofold more efficiently to each of the DNA probes in the presence of 100 μM cAMP than in its absence (Fig. 6B and C). Similar results were obtained when 100 μM cAMP was also added to the gel and running buffer (data not shown). These results indicate that cAMP improves the binding of CRP_Mt to specific DNA sequences but is not absolutely required with any of the probes we tested.

CRP loses DNA binding activity after limited proteolysis with trypsin in the presence, but not the absence, of cAMP (35). This is due to the conformational change induced in CRP upon cAMP binding. We examined the effect of cAMP on CRP_Mt's conformation by using limited proteolysis of CRP_Mt with trypsin. Surprisingly, SDS-PAGE analysis of trypsin-digested samples showed less proteolysis in the presence of 100 μM cAMP than in the absence of cAMP (Fig. 7A). In particular, a large protein fragment was retained in the presence of cAMP that was missing when digestion was performed without cAMP. The addition of AMP had no protective effect, confirming the specificity of this effect for cAMP.

FIG. 7.

Evidence of allosteric alteration of CRP_Mt by cAMP. (A) SDS-PAGE of CRP_Mt after limited proteolysis with trypsin in the presence or absence of cAMP or AMP. M, molecular weight marker (Molecular Probes). CRP_Mt treatment was as follows: lanes 1 to 3, trypsin digestion; lanes 4 to 6, undigested controls; lanes 2 and 5, addition of 100 μM cAMP; lanes 3 and 6, supplementation with 100 μM AMP. (B) EMSA of CRP_Mt using intergenic Rv0884c-Rv0885 DNA probe after limited proteolysis with trypsin in the presence or absence of 100 μM cAMP or 100 μM AMP. The CRP_Mt treatment is shown at the top, with the digested CRP_Mt concentration indicated in mM. An EMSA with all samples was performed with 100 μM cAMP in the binding reaction buffer. The figure is representative of three experimental repeats.

On a functional level, trypsin digestion of CRP_Mt in the absence of cAMP nearly abolished its ability to bind the Rv0884c probes (Fig. 7B). However, CRP_Mt's DNA binding ability was partially protected when 100 μM cAMP was present during the digestion procedure. Together, these results indicated that the conformation of CRP_Mt was altered in the presence of cAMP, resulting in an increased resistance of CRP_Mt to trypsin cleavage and partial protection of its DNA binding domain.

DNA bending by CRP_Mt.

The binding of CRP to specific DNA sequences induces a sharp bend in the DNA (59, 71). This induced DNA bending is an integral part of the mechanism by which CRP activates gene transcription (37, 39). Bent DNA fragments, including those bound by CRP, typically display lower electrophoretic mobilities when the bend occurs near the center of the DNA molecule (34).

We used Rv0884c-Rv0885 intergenic DNA sequences containing the CRP_Mt binding site to test the possibility that CRP_Mt binding causes DNA bending. Five identically sized (156-bp) DNA fragments were PCR amplified from overlapping regions of the Rv0884c-Rv0885 intergenic sequence, placing the motif at different positions relative to the center of each DNA molecule (Fig. 8A). The mobilities of the DNA fragments were compared by EMSA with CRP_Mt.

FIG. 8.

DNA bending by CRP_Mt. (A) Graphic showing the genetic structure of the Rv0884c-Rv0885 intergenic region. Five 156-bp subregions, designated F1 to F5, were amplified by PCR, with the binding site at a different location within each fragment, as shown. (B) Fragments F1 through F5 were labeled and used for EMSA with 35 nM of CRP_Mt. Unbound probes showed similar mobilities (left half of gel), while the mobility of each protein-DNA complex varied depending on the position of the CRP_Mt binding site within the fragment (right side of gel).

The mobilities of all five labeled DNA fragments were quite similar in the absence of CRP_Mt (Fig. 8B). In contrast, the mobility of the DNA-protein complexes was affected by the position of the motif within the fragment. Complexes formed with F3, in which the motif is centrally located, had the slowest mobility, while F1 and F5 complexes, which contain a motif within 6 bp of an end of the fragment, migrated the fastest. The mobility retardation of the complexes formed with F2 or F4 was intermediate, consistent with the location of their binding sites midway between the middle and end of each fragment (Fig. 8). These results indicated that CRP_Mt binding caused DNA bending and are consistent with CRP_Mt being a CRP-like gene regulatory factor.

DISCUSSION

Combination of genomic SELEX and Gibbs sampler for identifying DNA binding sites.

The identification of regulons and their cognate regulatory factors is one of the major challenges of the current genomic era. This study shows the power of combining experimental and computational approaches. SELEX has been successfully used on a genome-wide scale to identify transcription factors and their cognate binding sites in a variety of organisms (22, 32, 51, 55, 63, 68). However, a limitation of SELEX is that its tendency to bias towards the strongest binding sequences can reduce the diversity of the DNA sequences that are recovered. Although our SELEX studies provided a critical experimental foundation early on in this investigation, these results alone were not sufficient for making genome-scale predictions about CRP_Mt regulon membership. The computational approach was essential for extending the DNA binding data to make regulon predictions.

Computational methods designed to detect matches to a consensus sequence or motif model suffer from poor specificity or poor sensitivity (or both) and thus are of limited use when searching genome-scale sequence data. Therefore, we employed a sequence alignment (i.e., motif discovery) algorithm, Gibbs sampling, to predict CRP_Mt binding sites in the M. tuberculosis genome. Gibbs sampling has been widely used to detect transcription factor binding sites and their motif models when the model is not known beforehand, where the sequence data analyzed are typically from coregulated or coexpressed genes from a single species (21) or from the promoter regions of a single gene from several closely related species (i.e., phylogenetic footprinting) (45, 47). We took a novel approach in the advanced application of Gibbs sampling described here, using this motif prediction algorithm to detect a single regulon in genome-scale data in the absence of gene expression information, i.e., to detect a single, relatively rare, sequence pattern in nearly 350,000 bp of intergenic sequence data, using Bayesian prior information on an E. coli motif.

Given the size of the sequence search space and the stochastic nature of Gibbs sampling, we do not expect to have completely delineated the CRP_Mt regulon. For example, the electrophoretic mobility of the relatively low-probability Rv1230c binding-site probe was obviously retarded in the presence of CRP_Mt (Fig. 4C), suggesting that additional binding sites remain to be discovered. It is also important that the EMSA experiments were performed in vitro with purified protein and DNA and that additional cofactors and/or competing paralogs could alter CRP_Mt's behavior in vivo. Future in vivo expression-based studies are expected to refine and extend these predictions in a biological context. The identification of several additional CRP_Mt regulon candidates from a recent expression-based study is consistent with this prediction (54).

It remains possible that the predicted regulon represents a mixture of binding sites for paralogous transcription factors, but we think this is unlikely based on our experimental results. For example, such a Gibbs sampling approach with the E. coli genome might detect a mixed model containing CRP and FNR sites. However, all of the predicted DNA sites that we tested bound specifically to CRP_Mt, and CRP_Mt was able to readily discriminate between various closely related sequences (e.g., CRP and FNR or mutant CRP_Mt sites). In addition, the genome sequence of M. tuberculosis H37Rv encodes relatively few transcription factors that are likely to cause the type of mixed-model regulon predictions described above. Specifically, there are only three predicted transcription factors with putative CRP/FNR-family HTH regions (Rv3676, Rv1675c, and Rv1719). Two of these proteins (Rv3676 and Rv1675c) are likely paralogs with similar domain structures (cAMP-binding domain followed by an HTH domain), whereas the third (Rv1719) contains an N-terminal HTH domain followed by an IclR-type effector binding domain (Pfam PF01614) (4). This information and our unique Gibbs sampling approach provide a foundation for generating testable hypotheses to address such issues in future studies.

Putative CRP_Mt regulon membership in M. tuberculosis.

The putative CRP_Mt regulon is consistent with the size of the CRP regulon, which contains >100 experimentally verified and newly predicted genes (9, 33, 66, 72). Despite the resemblance of CRP_Mt and CRP binding motif models (Fig. 2), there does not appear to be a correspondence in regulon membership. The two largest groups of genes in the predicted CRP_Mt regulon are both associated with in vitro dormancy models, including 14% which are linked to hypoxia (62) and 30% from a nutrient starvation model (5). crp is upregulated in this starvation model (5), suggesting that CRP_Mt may be important for regulating this starvation response. However, it is difficult to predict a specific role for CRP_Mt in M. tuberculosis latency from these data. Both previous studies reported large numbers of genes, and the proportional representation of each group in our study is similar to the proportions of these genes in the overall genome. Nonetheless, future studies on the possible role of CRP_Mt in latency are clearly warranted.

Another gene of interest in the predicted regulon is Rv0805, which is annotated as a homolog of E. coli cpdA, encoding a 3′,5′-cyclic-nucleotide phosphodiesterase (http://genolist.pasteur.fr/TubercuList/). Rv0805 contains a strong CRP_Mt motif in its upstream intergenic region and is a putative member of our CRP_Mt regulon. This suggests a role for CRP_Mt in the regulation of cAMP levels within the cell, and we are currently exploring this possibility.

Recently, Spreadbury et al. proposed 15 genes as potential Rv3676 regulon candidates, although Rv3676 binding sites were not specified. Only Rv0867c (rpfA) and Rv1552 (frdA) are predicted by both their study and the present study to be members of the putative CRP_Mt regulon (65). The regulation of rpfA by CRP_Mt has also recently been experimentally confirmed (54). While both studies made use of information from the E. coli CRP regulon, differences between studies in the approaches used for regulon prediction could account for this limited overlap. In particular, we restricted our regulon predictions by searching for putative binding sites only within intergenic regions upstream of annotated open reading frames because we expect that these sequences are most likely to be involved in transcriptional regulation. In addition, our search was genome-wide, that is, not restricted by putative gene function, because we expect that regulon membership has likely diverged between these two very distantly related bacterial species.

Functional comparison of CRP_Mt with CRP.

In E. coli, activation of the lac promoter involves the binding of cAMP-CRP to its cognate site, which induces DNA bending that facilitates direct contact between CRP and RNA polymerase (23, 31, 53). An amino acid sequence alignment has shown that the cAMP binding (46) and DNA binding (Fig. 1) contact residues found in CRP are conserved in CRP_Mt. Our results experimentally confirm CRP_Mt's interactions with both cAMP and DNA, including a cAMP-induced conformational change and an ability to bend DNA upon binding. Although CRP_Mt undergoes a conformational change in the presence of cAMP, structural differences between CRP_Mt and CRP were also indicated by trypsin digestion experiments. Whereas cAMP binding opens up the CRP molecule, making it more vulnerable to proteolytic cleavage (35), cAMP binding decreased CRP_Mt's sensitivity to trypsin digestion. This suggests that the overall topologies of these two proteins differ, at least when bound to cAMP, and this may have functional implications.

CRP_Mt bears some similarity to CRP*, in that CRP_Mt does not require cAMP for specific DNA binding in vitro, although the presence of cAMP did enhance CRP_Mt's DNA binding ability. Native E. coli CRP requires a cAMP-dependent conformational change to bind DNA specifically and function as a transcriptional regulator. However, mutant crp* alleles encode cAMP-independent proteins that are functionally active in vitro and in vivo in the absence of cAMP (2, 6). In addition, both CRP* and CRP_Mt are highly susceptible to proteolysis in the absence of cAMP but are protected from cleavage when bound to cAMP (28, 52). In contrast, unliganded CRP is resistant to digestion but is easily degraded when in complex with cAMP. CRP_Mt's interaction with cAMP during gene regulation remains to be explored in vivo.

It was recently reported that CRP_Mt was not functional in E. coli (65), and we have made similar observations (G. Bai and K. A. McDonough, unpublished). Such interspecies functional studies can be difficult to interpret. For example, the P. aeruginosa vfr gene product, a homolog of E. coli CRP, complements the β-galactosidase- and tryptophanase-deficient phenotypes of an E. coli crp mutant (69), but the E. coli crp gene does not complement the vfr mutation. These results have been interpreted as a failure of CRP to interact properly with P. aeruginosa RNA polymerase. We hypothesize that functional interactions of CRP_Mt may be restricted to mycobacterial RNA polymerase, and this warrants further investigation.

Despite its apparent functional similarity with CRP, the regulation of CRP_Mt expression differs from that of CRP and does not appear to be autoregulatory. M. tuberculosis crp does not contain a CRP_Mt motif in its upstream intergenic region, and its intergenic sequences failed to bind CRP_Mt in EMSA (data not shown). In addition, the mechanism of crp (1, 27, 48) and vfr (56) autoregulation involves competitive expression of a divergently transcribed gene, and there is no divergent open reading frame annotated upstream of M. tuberculosis crp (http://genolist.pasteur.fr). We expect that the regulation of M. tuberculosis crp will be an interesting area of future investigation.

In summary, DNA binding sites of CRP_Mt were identified by multiple approaches, including genome-scale experimental and computational approaches. We identified a palindromic DNA motif with similarity to the E. coli CRP binding motif and predicted a CRP_Mt regulon containing ∼114 genes. The interaction of CRP_Mt with cAMP and specific DNA binding sites was experimentally confirmed, providing the first direct evidence for cAMP interaction with a transcription factor in M. tuberculosis. These results indicate an important role for cAMP signal transduction during global gene regulation in M. tuberculosis. Future studies of cAMP-mediated gene regulation are likely to contribute to an understanding of M. tuberculosis's response to the host environment during infection.