Skip to main content

Triad pattern algorithm for predicting strong promoter candidates in bacterial genomes



Bacterial promoters, which increase the efficiency of gene expression, differ from other promoters by several characteristics. This difference, not yet widely exploited in bioinformatics, looks promising for the development of relevant computational tools to search for strong promoters in bacterial genomes.


We describe a new triad pattern algorithm that predicts strong promoter candidates in annotated bacterial genomes by matching specific patterns for the group I σ70 factors of Escherichia coli RNA polymerase. It detects promoter-specific motifs by consecutively matching three patterns, consisting of an UP-element, required for interaction with the α subunit, and then optimally-separated patterns of -35 and -10 boxes, required for interaction with the σ70 subunit of RNA polymerase. Analysis of 43 bacterial genomes revealed that the frequency of candidate sequences depends on the A+T content of the DNA under examination. The accuracy of in silico prediction was experimentally validated for the genome of a hyperthermophilic bacterium, Thermotoga maritima, by applying a cell-free expression assay using the predicted strong promoters. In this organism, the strong promoters govern genes for translation, energy metabolism, transport, cell movement, and other as-yet unidentified functions.


The triad pattern algorithm developed for predicting strong bacterial promoters is well suited for analyzing bacterial genomes with an A+T content of less than 62%. This computational tool opens new prospects for investigating global gene expression, and individual strong promoters in bacteria of medical and/or economic significance.


Efficient promoter recognition is crucial in the synthesis of the gene-encoded products required by bacteria to allow them to grow rapidly and to adapt to different environmental conditions. The general architecture and protein-DNA interaction interfaces appear to be conserved in RNA polymerases of different bacteria, to judge by a comparison of the resolved structures of the multi-subunit protein or its subunits [1]. This structural information suggests that the principles of DNA recognition by RNA polymerases are universal, and this constitutes a basis for in silico prediction of promoters that are recognized by families of sigma factors. Research in bioinformatics has developed approximate matching methods to detect conserved sequences in nucleic acids [25], including promoter-specific sequences that are invaluable in helping to elucidate the overall organization of transcriptional signals and regulatory circuits in evolutionarily distant bacteria [614]. Most promoter prediction programs so far proposed use statistical or motif-based methods, and take into consideration what is known about experimentally defined promoter architectures, and extract conserved sequences from the genomes under analysis. Attempts have been made to improve promoter prediction by introducing statistical mechanical methods to measure the stress-induced destabilization or bendability of the duplex DNA region located upstream of the transcription initiation site required for the local dissociation of strands to start mRNA synthesis [1517]. The steady increase in the number of sequenced bacterial genomes of medical and economic significance means that there is an increasing need for computational tools to predict promoters, especially those responsible for high-level gene expression in organisms, of which there has been little experimental investigation.

Many housekeeping genes in Escherichia coli are transcribed from promoters possessing the recognition elements referred to as -35 and -10 sites (boxes), for which the TTGACA and TATAAT consensi, respectively, have been identified by compiling characterized RNA polymerase-binding regions in the DNA [18, 19]. The -35 and -10 sites, which are separated from each other by a 15–20-bp spacer [20], are specifically recognized by Eσ70 RNA polymerase, an RNA polymerase holoenzyme bearing the group I σ70 factor [21]. Experimental data have shown that high transcription rates of genes correlate with the level of conservation of three promoter parameters, with the consensus -35 and -10 hexanucleotide boxes, and with the 17 ± 1-bp spacer separating them [22]. This fact has been widely used to construct vectors for protein overexpression in bacterial cells [23].

However, the strength of strong promoters also depends on a fourth parameter, an AT-rich UP element 17–20 bp length, which is located just upstream of the -35 site, and which is recognized by the α subunit of Eσ70 RNA polymerase, which was first discovered for ribosomal RNA promoters [24]. The C-terminal domain of this subunit binds both to the UP element and to transcription regulation proteins, whereas the N-terminal domain makes contact with other subunits during the assembly of RNA polymerase [25]. A 17-bp consensus 5'AAAWWTWTTTTNNNAAA (where W is A or T, and N can be any of the four bases) has been identified for the UP element by analyzing the patterns, selected by the SELEX method, which mediate increases of between 10- and 300-fold in gene expression in E. coli cells [2628]. Two preferred sub-sites have been identified within the UP element. They are centered approximately at the -42 and the -52 positions respectively, and appear to be specifically recognized by one or two monomers of a dimeric α subunit in the RNA polymerase.

It is noteworthy, that a virtual analysis of patterns located upstream from the consensus -35 had since long time suggested their functional significance [3]. The sequences reminiscent of UP element have been detected in the E. coli genome by the algorithms PWM [29] and PlatPram [30]. A software-based (GCG, version 9.0) dissection of regions located upstream of the E. coli promoters had made it possible to detect putative promoters with ≤ 4 mismatches in the full UP element consensus [28]. Several UP elements have also been visually identified, and characterized by their ability to direct high level gene expression in vivo or in vitro in Bacillus subtilis [31], Geobacillus (formerly Bacillus) stearothermophilus [32] and Vibrio natrigens [33]. Recently, a comparative analysis of Eσ70-specific promoter and non-promoter regions indicated that upstream regions of E. coli ribosomal and T4 phage early promoters possess electrostatic elements that could be responsible for modulating promoter activities due to ADP-ribosylation of RNA polymerase α subunit [34]. However, no specific algorithms have yet been proposed to detect strong promoters in bacterial genomes, and so this remains an important task for genomic and proteomic research in microbiology.

In this study, we have developed a triad pattern algorithm that detects strong promoter candidates composed of a UP-element, and two consensi, -35 and -10 boxes, which are optimally distanced from each other. All four parameters are required for efficient DNA recognition, and the initiation of mRNA synthesis by an Eσ70-like RNA polymerase. The data presented indicate that the frequency of strong promoters is a function of the A+T content of the corresponding genomes. The proposed prediction program is flexible, and can be modified by users to modulate the search stringency criteria depending on the promoter features of the genome under analysis. The accuracy of detection has been experimentally validated for putative strong promoters predicted in a hyperthermophilic bacterium Thermotoga maritima.


Overview of the approach

The promoter activity in cells is determined by regulatory proteins (activators and repressors) that can recognize overlapping sequences specific for Eσ70 RNA polymerase sites, and thereby mask the true promoter strength. In addition, almost 20% of E. coli RNA polymerase Eσ70-specific promoters possess an extended -10 sequence that might compensate for the absence of a clear -35 site [35]. Different prediction programs based on statistical and motif-searching approaches have been developed to detect a variety of binding sites in DNA, and both position-specific weight matrices [36] and hidden Markov models [37] have been used to improve the accuracy of the prediction of promoter sequences in bacterial genomes [3840]. These programs usually detect hexanucleotide dyad patterns of RNA polymerase-promoter binding sites, such as -35 and -10 boxes, and none of them is free of false-positives, which correspond to similar, non-promoter sequences in bacterial genomes [for a review, see [41]].

In this study, we exploited the strengths of the "triad pattern" approach to develop an algorithm able to detect strong promoters by matching three nucleotide sequences recognized by the σ70 and α subunits of bacterial RNA polymerase. Theoretically, the presence of a UP element may not be essential for relatively strong promoter activity if two -35 and -10 boxes are well conserved and optimally distanced. Similarly, the presence of a well conserved UP element may compensate for a poor -35 box in some promoters. However, it seems very likely that the strongest promoters probably possess all three essential sequences. The specific interaction between the UP element and the α subunit significantly amplifies the association of RNA polymerase with promoter DNA [27, 28]. Therefore, to improve the filter to exclude possible false-positive due to short hexanucleotide similar sequences scattered throughout the genome, our algorithm starts by first matching the UP element, and only then identifying the -35 and -10 boxes located further downstream.

Design of the triad pattern algorithm

We designed an algorithm able to detect the triad nucleotide patterns in bacterial genomes. The core of the algorithm is the FIND_TRIAD procedure, which given an input nucleotide string, s, returns the substring s' of s, which is the best approximation of a given triad pattern of the form (pat(1),L1)-(l1,l2)-(pat(2),L2)-(d1,d2)-(pat(3),L3), where each pat(i), i = 1,2,3, is a nucleotide string, Li is its length, l1 and l2 are the minimum and maximum distances respectively between the first and the second patterns, and d1 and d2 are the minimum and maximum distances respectively between the second and the third patterns. To avoid making a "bad" approximation, three scores Sc1, Sc2 and Sc3 are used as input parameters for the procedure. The resulting substring, s', can then be represented as (spat(1),Ls1)-ls1-(spat(2),Ls2)-ls2- (spat(3),Ls3), where each spat(i), i = 1,2,3, is a substring of s aligned to pat(i), Lsi is its length, ls1 is the distance between spat(1) and spat(2), and ls2 is the distance between spat(2) and spat(3). This result for s' satisfies the following conditions:

  1. (1)

    for each i = 1,2,3 the similarity score (weight) Wi of the match or alignment of pat(i) and spat(i) is not less than Sci (or the number of "mismatches" does not exceed (Li - Sci));

  2. (2)

    (l1 ≤ ls1 ≤ l2) and (d1 ≤ ls2 ≤ d2).

For each of the three patterns, one can either forbid insertions/deletions or allow them. In the former case, Lsi = Li and the weight = W i are computed as the sum of matching pairwise symbols, whereas in the latter case, the difference |Lsi - Li| between spat(i) and pat(i) is bounded by a value Ri for the permissible deletions/insertions (gaps), an optimum alignment, and its weight, Wi, are computed by the standard dynamic programming method for global string alignment [42]. In both cases, a symbol scoring matrix Mi(x,j) is used to define the weight of the symbol x in the position j, 1 ≤ j ≤ Lsi, of spat(i). If symbol x occurs in position j of pat(i), then Mi(x,j) = 1, otherwise Mi(x,j) ≤ 1. To choose the best approximation of the triad pattern from substrings satisfying conditions (i) and (ii), FIND_TRIAD uses a total score function with the form:tot_sc = C1*nsc1(L1,W1)+D12*nsc_dist12(11,l2,ls1) + C2*nsc2(L2,W2) + D23*nsc_dist23(d1,d2,ls2) + C3*nsc3(L3,W3),

where nsci(Li,Wi), i = 1,2,3, are normalized scores of matching (alignments) of pat(i) and spat(i), 0 ≤ nsci(Li,Wi) ≤ 1, and nsc_dist12(11,l2,ls1) and nsc_dist23(d1,d2,ls2) are the normalized scores of the distances between the first and the second, and the second and the third patterns, respectively, and 0 ≤ nsc_dist12(11,l2,ls1), nsc_dist23(d1d2,ls2) ≤ 1. The linear coefficients C1, C2, C3, D12, and D13 are chosen so that their sum is equal to 1. They indicate the relative importance of the corresponding sub-patterns of the triads; and the distances between them. So, the best value of tot_sc is 1.

Application of the algorithm to searching for strong promoter candidates

Here we describe the main parameters of the FIND_TRIAD procedure used to detect strong promoter candidates in bacterial genomes. In this study, a bacterial promoter is assumed to be a nucleotide sequence, located upstream from genes encoding proteins, tRNAs or rRNAs that could be recognized by an RNA polymerase holoenzyme containing a major σ factor (using E. coli70 RNAP as the reference). The triad patterns defined for strong promoter candidates include three specific nucleotide sub-regions: (i), a UP element, which is a 17-nt prefix of the strong promoter, and has the following consensus pattern: pat(1) = P UP = aaaWWtWttttNNNaaa; (ii) the -35 site, which is located downstream of the UP element, and has the pattern pat(2) = P 35 = tc ttgaca t (underlining indicates a commonly used consensus for group I σ70 factors; however, the σ4 domain of these factors appears to be in contact with 9 nucleotides in the region extending from -30 to -38 [43, 44]); (iii) the -10 site, which is located downstream of the -35 site, and has the pattern pat(3) = P 10 = tataat (this site is highly conserved). We used the following boundaries for the distances between the sub-regions: l1 = 0, l2 = 5 (these boundaries were extracted from the examples of UP-elements in [2527]), d1 = 14, d2 = 20 (these boundaries are standard for the distance between the -35 site and the -10 site). To search for the first pattern pat(1) of the UP-element, the simple matching algorithm was chosen with an a and t mismatch score of 0.5. The reason is that in the full UP-element consensus and the consensuses of two of its subsides – distal and proximal – in some places do not distinguish between a and t. We assumed that the consensus for the -35 site of length 9 is less conserved than that of the -10 site, and so in order to detect the second pattern pat(2) of the -35 site we used a dynamic programming algorithm to search for optimal alignment, with boundaries for the number of permissible deletions/insertions of R2 = 2. For the most of -35 sites, which were detected by algorithm, no insertions/deletions were applied. However, this scoring system allowed us to identify some stronger promoter candidates. Thus, the insertion of C between two AA in the sequence TCTTGAAT of TM1016, increases the score of a putative promoter (see below). The -10 site is better conserved, and so we used the straightforward matching algorithm to detect this site.

To define the total score function, tot_sc (formula 1), we chose the following normalized scores for the three patterns and for the distance between the -35 site and -10 site (no information was available about the best values for the distance between the UP element and the -35 site):nsc1(17,W1)= nsc_up = 1 - (17 - W1)/20,nsc2(9,W2)= nsc_35 = 1 - (9 - W2)/10,nsc3(6,W3)= nsc_10 = 1 - (6 - W3)2/10,

and the values of the normalized distance score, nsc_dist23(14,20,ls2) = nsc_dist, are defined as follows:

l s 2 : distance between the -35 and -10 sites, nt 17 16 , 18 15 , 19 14 , 20 n s c _ d i s t 1 0.95 0.85 0.7 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaqbaeaabiqbeaaabaGaemiBaWMaem4CamNaeGOmaiJaeiOoaOJaeeizaqMaeeyAaKMaee4CamNaeeiDaqNaeeyyaeMaeeOBa4Maee4yamMaeeyzauMaeeiiaaIaeeOyaiMaeeyzauMaeeiDaqNaee4DaCNaeeyzauMaeeyzauMaeeOBa4MaeeiiaaIaeeiDaqNaeeiAaGMaeeyzauMaeeiiaaIaeeyla0Iaee4mamJaeeynauJaeeiiaaIaeeyyaeMaeeOBa4MaeeizaqMaeeiiaaIaeeyla0IaeeymaeJaeeimaaJaeeiiaaIaee4CamNaeeyAaKMaeeiDaqNaeeyzauMaee4CamNaeeilaWIaeeiiaaIaeeOBa4MaeeiDaqhabaGaeGymaeJaeG4naCdabaGaeGymaeJaeGOnayJaeiilaWIaeGymaeJaeGioaGdabaGaeGymaeJaeGynauJaeiilaWIaeGymaeJaeGyoaKdabaGaeGymaeJaeGinaqJaeiilaWIaeGOmaiJaeGimaadabaGaemOBa4Maem4CamNaem4yamMaei4xa8LaemizaqMaemyAaKMaem4CamNaemiDaqhabaGaeGymaedabaGaeGimaaJaeiOla4IaeGyoaKJaeGynaudabaGaeGimaaJaeiOla4IaeGioaGJaeGynaudabaGaeGimaaJaeiOla4IaeG4naCdaaaaa@892E@

We also chose linear coefficients C1 = 0.3, C2 = C3 = 0.25, D12 = 0, and D23 = 0.2. These coefficients indicate the relative importance of corresponding sub-regions for evaluating the total score of a candidate sequence. They were chosen empirically, after preliminary tests with several annotated genomes, assuming a higher significance of the UP element, equal significance of the -10 and -35 boxes, and lower significance of the distance between them. In this application, the value D12 = 0 means that we ignore the variations of the distance between a putative UP element and -35 box because a priory it is not known what value is the best in the interval 0–5 nt.

Formulas 2, 3 and 4 reflect the lack of exact matching for the different sub-regions. If the -10 box is highly conserved and is essential for initiation of transcription [22], then the penalty for its mismatches is higher than for those of the other parameters. For example, for 2 mismatches, the penalty is (6 - 4)2/10 = 0.4 for the -10 site, whereas it is (9 - 7)/10 = 0.2 for the -35 site, and (17 - 15)/20 = 0.1 for the UP element. The choice of the normalized score functions in equations 2, 3 and 4 is based on empirical observations, and on common sense, and may seem to be arbitrary. We want to stress that, in fact, the total score function tot_sc also has a further role: it does not significantly change the set of the best candidates identified by the algorithm. This set is defined by the three score bounds Sc1 = scup for UP element, Sc2 = sc35 for -35 site, and Sc3 = sc10 for -10 site. The total score affects only the ordering of these candidates amongst themselves.

The general scheme of the algorithm is as follows. It has the following input: (i) the name of a genome file in GenBank format; (ii) three parameters of scores: scup, sc35 and sc10, determining the minimum acceptable similarity between candidate sequences of the UP element, the -35 box, and the -10 box, respectively, and the E. coli consensus patterns. For each gene in the genome input file that is not inside an operon, the algorithm runs in two steps:

  1. (i)

    it extracts a 300-bp DNA region, s, upstream of the annotated coding sequences for tRNA, rRNA or proteins (we limited the search to 300 bp, since most E. coli promoters fall within this length inter-gene space [41, 45]);

  2. (ii)

    then it uses the FIND_TRIAD procedure to identify the best strong promoter candidate within s that satisfies conditions (1) and (2) above. If such a candidate is found, it is added to the output list of strong promoters.

We recommend to read attentively the "ReadMe" information [see Additional file 1] before to start proceeding the "strong_promoters.doc" software [see Additional file 2]. The algorithm is implemented by a program that produces the results in two forms: (i) a Text-format table which lists all strong promoter candidates in a genome, and provides additional information about the operon organization of genes located downstream (for example, see Fig. 1); (ii) a Word-format table which lists strong promoter candidate sequences. A 20-nt sequence preceding a possible initiation codon of each ORF is also included in the annotation, as this could be useful for the visual examination of the translation signals of the corresponding genes. Lastly, the user can select a convenient score for each sequence-specific motif taking into consideration the promoter features of the annotated genome if they differ from the E. coli-specific patterns used to create the algorithm (for example, a weakly conserved -35 or -10 box).

Figure 1

Text-format presentation of strong promoter candidates.


Construction of recombinant linear DNAs

Putative promoter regions in the T. maritima genome, identified by the algorithm described above, were amplified by PCR using appropriate oligonucleotide primers connected to the previously-described G. stearothermophilus argC gene [46]. This reporter gene encodes N-acetyl glutamylphosphate reductase, a thermostable and soluble protein that is easily detectable after exposing E. coli cleared lysates to 65°C. In order to increase protein yield, the ribosome-binding site of G. stearothermophilus argC was modified to the sequence GGAGG GGGAACATATG (the modified Shine-Dalgarno site and the initiation codon are underlined), and the distance between the -10 promoter site and the Shine-Dalgarno site was shortened to 15 bp (Fig. 2). The DNA fragment carrying the argC gene was connected to T. maritima or control promoters by two consecutive PCR steps, as described previously [47]. The quantity and quality of the amplified DNAs were determined with a 2100 Bioanalyzer (Agilent Technologies).

Figure 2

Diagram of the fusion DNA constructs used to express the G. stearothermophilus argC -reporter gene from putative strong promoters of T. maritima in a cell-free system. The argC gene was amplified with forward 5'-GGAGGGGGAACATATGATGAA and reverse 5'-GGACCACCGCGCTACTGCCG primers from pHAV2 [32] by conserving a 112-bp downstream region carrying transcriptional terminators of the vector DNA.

Two well-characterized strong promoters, Ptac and PargC, were used as references to compare the strength of the putative promoters of T. maritima. The strong promoter Ptac contains an AT-rich nucleotide sequence upstream of a -35 site [48], which has no defined UP element; it was obtained from the vector pBTac2 (purchased from Boehringer Mannheim). PargC, a strong promoter of G. stearothermophilus, contains the UP element, as demonstrated both in vivo and in vitro, and was amplified from the plasmid pHAV2 [32].

Cell-free protein synthesis

PCR-generated linear DNA fragments carrying a promoter region fused to the argC reporter gene were used to evaluate the promoter strength in a coupled transcription-translation system, as described previously [49]. The cell-free extracts were prepared from the E. coli strain BL21 (DE3) Star recBCD (our laboratory construction) as described by Pratt [50]. Protein synthesis was carried in the presence of pyruvate oxidase to generate ATP [51]. Typically, 50 ng of PCR-amplified DNA was added to a pre-mix containing all necessary compounds and 10 μCi of [α35S]-L-methionine (specific activity 1000 Ci/mmol, 37 TBq/mmol, Amersham-Pharmacia Biotech), and E. coli S30 cell-free extracts. The reaction mixture was incubated at 37°C for 90 min, and heated to 65°C for 10 min. After centrifuging, the supernatant was precipitated with acetone, and then protein samples were separated by SDS-PAGE and bound to 3 MM paper. The ArgC protein synthesized in vitro was quantified by counting the radioactivity of the corresponding band with a PhosphorImager 445 SI (Molecular Dynamics).

The bacterial genome sequences were extracted from available data banks. The logo of T. maritima promoter consensus sequences was generated at the WebLogo site as described in [52, 53].


The number of strong promoters reflects the A+T content of bacterial genomes

In our algorithm, 26 of the 32 symbols used to evaluate matches in the three promoter-specific patterns, namely in the UP element and the -35 and -10 boxes, are a and t. One could expect the number of genes transcribed from potential strong promoters to depend on the A+T content of a given genome. To find out whether this is indeed the case, we compared the frequency of candidates in 300-bp regions located upstream of genes of annotated bacterial genomes and in random sequences of the same regions generated by computing. First, we calculated the (A+T)% in all 300-bp regions preceding each gene or operon in the annotated genomes (Table 1). The A+T content in these DNA regions was found to be slightly higher than that of the entire genomes of almost all bacteria that have been analyzed. Next, we generated 10.000 random sequences with the same A+T content for all the 300-bp regions of each genome. The algorithm was applied to detect strong promoter candidates in the 300-bp real genomic and random-generated regions of 43 bacterial genomes.

Table 1 A+T content of bacterial genomes and 300-bp regions located upstream of genes and the percentage of strong promoter candidates predicted in 300-bp real genomic and random-generated regions of the same content.

We tested different matching stringencies and empirically found that the score parameters sUP = 13, s35 = 5.5 and s10 = 4.5 satisfied the criteria required for scaled comparative analysis without grossly exaggerating the number of candidate sequences identified in the various genomes. This analysis revealed that the real genomes with an A+T content of less than 50% contained many more potential strong promoters than their simulated counterparts (see Table 1). The percentage of candidate sequences was very low in the bacterial genomes with an A+T content of between 33% and 47%, and these sequences were completely absent in the corresponding 300-bp, random-generated sequences. When the A+T content increased from 47% to 78%, the percentage of strong promoter candidates increased dramatically, whereas the difference between the real and random sequences decreased, and virtually disappeared when the A+T content exceeded 62%. There were two exceptions where the genomes analyzed did not display this pattern at an A+T content of less than 62%. One was M. pneumoniae, the genome of which had an A+T content of about 60%, and in which the promoters had no -35 consensus [54]. The other example is the hyperthermophilic species A. aeolicus (~58% AT-rich genome). This species is very close to the Archaea, and occupies a unique position in the bacterial kingdom [55].

Our data show that the number N(A+T) of strong promoter candidates in 300-bp random-generated sequences corresponding to upstream regions of bacterial genes satisfies the "exponential low" of the form N(A+T) = exp [c1 (A+T) + c2]. The distribution of strong promoter candidates in real genomes indicates that the critical point of the A+T content is close to 62% (Fig. 3). Above this level, the number of random sequences reminiscent of strong promoter patterns increases markedly.

Figure 3

The number of strong promoter candidate sequences is a function of the A+T content of bacterial genomes. For the score parameters sUp = 13, s35 = 5.5, s10 = 4.5 and constants c1 = 0.22 and c2 = -11.7, the picture displays a linear graph of the "exponential low" (thin line), which approximates fairly closely to the curve ln [N(A+T)], shown as a thick line. The logarithm of the percentage of strong promoter candidates in real genomes is shown by ().

Strong promoter candidate sequences are located upstream of gene-coding regions

Another important aspect of the quality of detection is the location of candidate sequences with regard to coding regions in the genome analyzed. We compared the frequencies of strong promoter-like patterns identified upstream and downstream of the initiation codon in all the genomes. The frequency of candidate sequences was clearly greater in the upstream region of ORFs in most of the genomes with an A+T content of less than 62% (Table 2). No difference was detected in T. pallidum (~47% AT-rich genome), which belongs to a distinct phylum of Spirochetes that appear to use different DNA patterns for the promotion and regulation of transcription [56].

Table 2 Number of sequences reminiscent of strong promoters in regions located upstream and downstream of the initiation codon of genes in bacterial genomes.

The fact that more candidate sequences were identified upstream of ORFs highlights the fact that they are not randomly distributed in bacterial genomes, which suggests that the detection of strong promoter candidates in genomes with an A+T content of less than 62% is fairly reliable.

Experimental validation of virtual prediction: analysis of putative strong promoters of T. maritima

Taking our cue from the results of the virtual prediction, we sought to find out whether, and if so, to what extent the putative promoters are functional in a biological context. To do this we used reporter-gene technology, which relies on the fusion of an assayable sequence with a promoter being investigated, and the subsequent evaluation of promoter strength in a cell-free system (see Fig. 2). The genome of the hyperthermophilic bacterium T. maritima [57] was used to evaluate the feasibility of the algorithm experimentally.

63 candidate sequences were detected in the T. maritima genome using the matching scores described above. We increased the penalty for mismatching of -35 and -10 boxes by raising the scores of s35 and s10 to 6 and 5, respectively. This reduced the number of candidate sequences to 34 (Table 3). In this shorter list, 28 T. maritima strong promoter candidates possessed a total score higher than the 0.8475 calculated for the reference strong promoter, Ptac, that does not have a typical UP element [48]. 15 of these candidates had a total score higher than 0.8775, as estimated for PargC, another reference strong promoter that has a well defined UP element [32, 49]. It is worth mentioning that 6 candidate DNA regions in T. maritima had a total score higher than 0.91, a value estimated for E. coli promoters that govern the transcription of 16S ribosomal RNA, and which were used as models for studying the stimulating effect of the UP element on gene expression [58].

Table 3 Strong promoter candidates identified in T. maritima MSB8*.

We selected 13 candidate promoter sequences for further analysis by evaluation of the ArgC thermostable protein production in a coupled transcription-translation system. These sequences all exhibited a total score ≥ 0.8475, apart from TM1490 (see Table 3). The amplified DNA regions were connected to the reporter gene argC, and used directly to assess promoter activity in vitro (see Fig. 2). All putative promoters of T. maritima were found to be active; the protein yield ranged from 0.3 to 2.7-times that of the reference Ptac promoter (Fig. 4). The gene expression from the promoter PTM1272 was similar to that of Ptac, whereas PTM0032 was reduced almost threefold. However, higher expression was detected from the other 11 promoters; the greatest expression level was observed for PTM0477, PTM1016, PTM1429 and PTMt45. Reporter gene expression was also higher for the strong promoter PargC, which carries the UP element.

Figure 4

Assessment of the strength of T. maritima strong promoter candidates in a cell-free system. Lanes 1 – Ptac (reference); 2 – PTM0032; 3 – PTM0373; 4 – PTM0477; 5 – PTM1016; 6 – PTM1067; 7 – PTM1271; 8 – PTM1272; 9 – PTM1429; 10 – PTM1490; 11 – PTM1667; 12 – PTM1780; 13 – PTMt45; 14 – PTMt11; 15 – PargC. Similar results were obtained in 3 experiments.

We next aligned experimentally analyzed promoters of T. maritima (Fig. 5). The most conserved sequence was the -10 box, which was identical to the E. coli consensus. The -35 box was also highly conserved, except that cytosine preceded the -35 site in 9 promoters, and no significant preference was detected for the nucleotides at the 5th and 6th positions. An 18-bp spacer appeared to be more representative than a 17-bp distance between the -35 and -10 boxes. Although all candidates possessed an AT-rich region upstream of the -35 site, some of them had only one A-rich tract, suggesting that they harbor only a single sub-site of a putative UP element. In any case, the high score attributed to 11 identified promoters was corroborated by elevated activity in vitro. Taken together, the alignment data and the expression data from the cell-free system, suggest that E. coli RNA polymerase efficiently recognizes putative strong promoters of T. maritima, and that the presence of an UP-like element might contribute to the strength of the promoter.

Figure 5

Organization of strong bacterial promoters. (A), Alignment of 13 promoter candidates of T. maritima; (B) consensus sequences of T. maritima and E. coli strong promoters; consensus of the E. coli UP element is described in [26, 27]; (C) the strong promoters Ptac and PargC were used as references in this study.

Two regions, (2.4 and 4.2) of the four domains of σ70 are involved in the recognition of the -10 and -35 boxes of E. coli promoters, respectively [59]. Several amino acids involved in contact with DNA have been also identified in the α subunit [60]. These DNA-binding regions in both σ70 and α subunits of E. coli and T. maritima RNA polymerases share high similarity (data not shown), which highlights the fact that -35 and -10 boxes and UP-like element all contribute to the high promoter activity in the thermophilic host.


Bacterial promoters can be arbitrarily classified as weak, moderate and strong promoters, depending on the level of expression of mRNAs or of the corresponding proteins. We have developed an algorithm that can predict strong promoters in bacterial genomes by matching the triad pattern specific for the group I σ70 factor of E. coli RNA polymerase. The first step in the proposed triad pattern approach involves matching the UP element located 300 bp upstream of a gene-coding sequence, and then matching two optimally separated -35 and -10 boxes.

The accuracy of the computational prediction of bacterial promoters depends on the A+T content of the genomes, which means that the matrix has to be adjusted to account for this factor in the DNA under analysis [29]. The data presented highlight the fact that the detection accuracy is lower in genomes with a high A+T content. The number of potential strong promoters identified in 43 bacterial genomes, is a direct function of their A+T content; this implies that the accuracy of the prediction is lower for genomes with A+T content higher than 62%.

The choice of the matching score is yet another difficulty in identifying DNA-binding sites including promoters, as the highest score may not be the one most biologically relevant for genome-scale predictions [61, 62]. It is therefore helpful to use additional criteria to eliminate false-positives. It looks as if the total score of 0.8475, calculated for the reference promoter Ptac, can be used as an reasonable criterion for identifying real strong promoters recognized by an Eσ70-like RNA polymerase. In particular, using the scores applied to genomes analysis (see Tables 1 and 2), the algorithm detects 7 potential strong promoters in M. tuberculosis (~34% AT-rich genome) that encodes a variety of σ factors, including σA that recognizes the promoters possessing typical -10 and -35 boxes [63]. However, none of the predicted strong promoters had a total score in excess of 0.8475, and visual inspection indicated that none of these promoters possesses an UP-like sequence, suggesting that this gene expression-stimulating element is absent in M. tuberculosis.

The possibility of applying linear PCR-generated molecules for cell-free protein synthesis, without needing to perform DNA cloning in bacteria, is a prerequisite for assessing gene expression on a genome-wide scale. As a first step in this direction, we tested reporter-gene fusions to evaluate the strength of the promoters identified in the genome of T. maritima. Though this approach does not exclude possible masking effects of E. coli repressors or activators in the extracts, it is relatively simple, timesaving and informative, all of which are major advantages for evaluating computational predictions. Using the two well-characterized strong promoters (Ptac and PargC) as references, high activity has been demonstrated for 11 out of 13 candidate sequences of T. maritima. This is quite a low proportion; however, it suggests that the detection accuracy by the triad pattern algorithm might be close to 85%. The limitations of the algorithm in terms of specificity and sensitivity of the virtual prediction of putative strong promoters might be further experimentally evaluated by analysis of bacterial genomes with high-throughput methods.

This study offers the first insight into the organization and distribution of strong promoters in hyperthermophilic organisms, which probably constitute the longest lineage in the microbial world [64]. Overall, strong promoters of hyperthermophiles are similar to those of mesophilic origin. We have recently shown that the T. maritima RNA polymerase α subunit binds to the PargG promoter described here under PTM1780 [65]. It has been found that the substitution of arginine in the hyperthermophilic α subunit, corresponding to the position Arg265 in the E. coli subunit and crucial for DNA recognition [60, 66], or the deletion of an AT-rich sequence located upstream of the -35 site, decreases the binding affinity for DNA [65]. The PargG promoter harbors a UP-like element, and is able to direct high gene expression in vitro. Moreover, this element appears to compensate for a poor -35 box or non-optimal 20-bp spacer of this promoter (see Table 3 and Fig. 5). Hence, these observations, along with the data obtained using other T. maritima promoters, allow us to assume that the presence of a UP-like element with less than 5 mismatches out of 17 nucleotides is essential for the strength of most strong promoters. This is consistent with the conservation of DNA interaction amino acids in the α subunit of the hyperthermoiphilic RNA polymerase. However, sequence-independent upstream DNA interactions within the C-terminal domain of the α subunit could often be required to initiate transcription in E. coli cells [67]. Therefore, the functional significance of the UP-like element in gene expression remains to be proven experimentally in hyperthermophilic organisms.

The strong promoters of T. maritima direct the transcription of genes involved in tRNA, ribosome synthesis, energy metabolism, transport, and cell movement (see Table 3). However, to our surprise, we found that 15 of the 38 best candidates promote the transcription of hypothetical proteins. The previously uncharacterized hypothetical protein TM1016 (total score 0.9175) turns out to share 28% identity with a biopolymer transport protein of Vibrio vulnificus [68]. In this context, recent studies of the T. maritima transcriptome have indicated that ABC transporters could play a major role in its ecology [69]. Further characterization of highly expressed hypothetical genes identified in our study might help to elucidate their role in the biology of this hyperthermophilic organism.

The strong promoter candidates prediction could contribute to the wide-scale genome expression analysis of evolutionarily distant bacteria, especially of those that possess an A+T DNA content lower than 62%. As a complement to DNA microarrays, it could help to elucidate the overall response of bacterial genomes to various environmental stresses. Moreover, the triad pattern algorithm can be used to extract the DNA region that carries translational signals; this is useful for investigating ORFs located downstream from the corresponding strong promoters (see Table 3). Thus, almost half of the T. maritima ORFs transcribed from putative strong promoters are preceded by a highly conserved Shine-Dalgarno site located 7–9 nucleotides from the ATG initiation codon, which is a characteristic feature of elevated protein synthesis in gram-negative and gram-positive bacteria [70]. This information will be useful for comparing highly synthesized mRNAs with the production of the corresponding proteins using high-throughput transcriptomic and proteomic methods, which is an important challenge in the fields of basic and applied microbiology [71]. Furthermore, the characterization of proteins whose expression is governed by strong promoters looks like a promising approach to selecting candidate vaccines against microbial diseases and/or to identifying potential new antibacterial targets in the fight against nosocomial infections.

Further quantitative assessment of a dynamic and complicated mechanism of protein-DNA and protein-protein interactions involved in transcription might help to develop a more advantageous multi-pattern tool using both DNA and protein parameters to provide a comprehensive prediction of the strength of promoter activity in bacterial cells.


The triad pattern algorithm developed predicts strong promoter candidates by matching UP-like elements and identifying the presence of -35 and -10 boxes optimally distanced from each other in the annotated bacterial genomes. The presence of strong promoters is a function of the A+T content of the bacterial genome, and the number of false-positives is greater for genomes that have an A+T content higher than 62%. The prediction algorithm has been validated by cell-free experimental dissection of putative T. maritima promoters. The data indicate that strong promoters govern the transcription of genes coding vital functions, and of genes coding as-yet unknown functions in this hyperthermophilic bacterium. This algorithm is simple to use and flexible, and it could be further adapted to meet the requirements of a genome of interest if its promoter-specific motifs differ from consensi recognized by Eσ70-like RNA polymerase.

Availability and requirements

The algorithm is freely accessible for non-commercial use at the web-site It takes several seconds to analyze the annotated genome sequence available from databases.


  1. 1.

    Darst SA: Bacterial RNA polymerase. Curr Opin Struct Biol 2001, 11: 155–162. 10.1016/S0959-440X(00)00185-8

    CAS  Article  PubMed  Google Scholar 

  2. 2.

    Queen C, Wegman MN, Korn LJ: Improvements to a program for DNA analysis: a procedure to find homologies among many sequences. Nucleic Acids Res 1982, 10: 449–456. 10.1093/nar/10.1.449

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  3. 3.

    Galas DJ, Eggert M, Waterman MS: Rigorous pattern-recognition methods for DNA sequences. J Mol Biol 1985, 186: 117–128. 10.1016/0022-2836(85)90262-1

    CAS  Article  PubMed  Google Scholar 

  4. 4.

    Staden R: Methods for discovering novel motifs in nucleic acid sequences. Comput Appl Biosci 1989, 5: 293–298.

    CAS  PubMed  Google Scholar 

  5. 5.

    Bailey TL, Elkan C: Fitting a mixture model by expectation maximization to discover motifs in biopolymers. Proc Int Conf Intell Syst Mol Biol 1994, 2: 28–36.

    CAS  PubMed  Google Scholar 

  6. 6.

    Alexandrov N, Mironov A: Application of a new method of pattern recognition in DNA sequence analysis: a study of E. coli promoters. Nucleic Acids Res 1990, 18: 1847–1852. 10.1093/nar/18.7.1847

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  7. 7.

    Demeler B, Zhou G: Neural network optimization for E. coli promoter prediction. Nucleic Acids Res 1991, 19: 1593–1599. 10.1093/nar/19.7.1593

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  8. 8.

    Cardon LR, Stormo GD: Expectation maximization algorithm for identifying protein-binding sites with variable lengths from unaligned DNA fragments. J Mol Biol 1992, 223: 159–170. 10.1016/0022-2836(92)90723-W

    CAS  Article  PubMed  Google Scholar 

  9. 9.

    Horton PB, Kanehisa M: An assessment of neural network and statistical approaches for prediction of E. coli promoter sites. Nucleic Acids Res 1992, 20: 4331–4338. 10.1093/nar/20.16.4331

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  10. 10.

    Thieffry D, Salgado H, Huerta AM, Collado-Vides J: Prediction of transcriptional regulatory sites in the complete genome sequence of Escherichia coli K-12. Bioinformatics 1998, 14: 391–400. 10.1093/bioinformatics/14.5.391

    CAS  Article  PubMed  Google Scholar 

  11. 11.

    Vanet A, Marsan L, Labigne A, Sagot M-F: Inferring regulatory elements from a whole genome. An analysis of Helicobacter pylori σ80family of promoter signals. J Mol Biol 2000, 297: 335–353. 10.1006/jmbi.2000.3576

    CAS  Article  PubMed  Google Scholar 

  12. 12.

    Leung SW, Melish C, Robertson D: Basic gene grammars and DNA-chartparser for language processing of Escherichia coli promoter DNA sequence. Bioinformatics 2001, 17: 226–236. 10.1093/bioinformatics/17.3.226

    CAS  Article  PubMed  Google Scholar 

  13. 13.

    Gordon L, Chervonenkis AY, Gammerman AJ, Shahmuradov IA, Solovyev VV: Sequence alignment kernel for recognition of promoter regions. Bioinformatics 2003, 19: 1964–1971. 10.1093/bioinformatics/btg265

    CAS  Article  PubMed  Google Scholar 

  14. 14.

    Jacques P-E, Rodrigue S, Gaudreau L, Goulet J, Brzezinski R: Detection of prokaryotic promoters from the genomic distribution of hexanucleotide pairs. BMC Bioinformatics 2006, 7: 423. (doi:10.1186/1471–2105–7-423) 10.1186/1471-2105-7-423

    PubMed Central  Article  PubMed  Google Scholar 

  15. 15.

    Benham CJ: Sites of predicted stress-induced DNA duplex destabilization occur preferentially at regulatory loci. Proc Natl Acad Sci USA 1993, 90: 2999–3003. 10.1073/pnas.90.7.2999

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  16. 16.

    Kanhere A, Bansal M: A novel method for prokaryotic promoter prediction based on DNA stability. BMC Bioinformatics 2005, 6: 1–10. 10.1186/1471-2105-6-1

    PubMed Central  Article  PubMed  Google Scholar 

  17. 17.

    Wang H, Benham CJ: Promoter prediction and annotation of microbial genomes based on DNA sequence and structural responses to superhelical stress. BMC Bioinformatics 2006, 7: 248. (doi:10.1186/1471–2105–7-248) 10.1186/1471-2105-7-248

    PubMed Central  Article  PubMed  Google Scholar 

  18. 18.

    Hawley D, McClure WR: Compilation and analysis of Escherichia coli promoter DNA sequences. Nucleic Acids Res 1983, 11: 2237–2255. 10.1093/nar/11.8.2237

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  19. 19.

    Harley C, Reynolds R: Analysis of E. coli promoter sequences. Nucleic Acids Res 1987, 15: 2343–2361. 10.1093/nar/15.5.2343

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  20. 20.

    O'Neil M, Chiafari F: Escherichia coli promoters. II. A spacing-class dependent promoter search protocol. J Biol Chem 1989, 264: 5531–5534.

    Google Scholar 

  21. 21.

    Helmann JD: The extracytoplasmic function (ECF) sigma factors. Adv Microb Physiol 2002, 46: 47–110.

    CAS  Article  PubMed  Google Scholar 

  22. 22.

    deHaseth PL, Zupancic ML, Record MT Jr: RNA-polymerase-promoter interactions: the comings and goings of RNA polymerase. J Bacteriol 1998, 180: 3019–3025.

    PubMed Central  CAS  PubMed  Google Scholar 

  23. 23.

    Makrides SC: Strategies for achieving high-level expression of genes in Escherichia coli . Microbiol Rev 1996, 60: 512–538.

    PubMed Central  CAS  PubMed  Google Scholar 

  24. 24.

    Ross W, Gosink KK, Salomon J, Igarashi K, Zou C, Ishihama A, Severinov K, Gourse RL: A third recognition element in bacterial promoters: DNA binding by the α subunit of RNA polymerase. Science 1993, 262: 1407–1413. 10.1126/science.8248780

    CAS  Article  PubMed  Google Scholar 

  25. 25.

    Ross W, Ernst A, Gourse RL: Fine structure of E. coli RNA polymerase-promoter interactions: α subunit binding to the UP element minor groove. Genes & Dev 2001, 15: 491–506. 10.1101/gad.870001

    CAS  Article  Google Scholar 

  26. 26.

    Estrem ST, Gaal T, Ross W, Gourse RL: Identification of an UP element consensus sequence for bacterial promoters. Proc Natl Acad Sci USA 1998, 95: 9761–9766. 10.1073/pnas.95.17.9761

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  27. 27.

    Aiyar SE, Gourse RL, Ross W: Upstream A-tracts increase bacterial promoter activity through interactions with the RNA polymerase alpha subunit. Proc Natl Acad Sci USA 1998, 95: 14652–14657. 10.1073/pnas.95.25.14652

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  28. 28.

    Estrem ST, Ross W, Gaal T, Chen ZWS, Niu W, Ebright RH, Gourse RL: Bacterial promoter architecture: sub-site structure of UP elements and interactions with the C-terminal domain of the RNA polymerase α subunit. Genes & Dev 1999, 13: 2134–2147. 10.1101/gad.13.16.2134

    CAS  Article  Google Scholar 

  29. 29.

    Hertz GZ, Stormo GD: Escherichia coli promoter sequences: analysis and prediction. Methods Enzymol 1996, 273: 30–42.

    CAS  Article  PubMed  Google Scholar 

  30. 30.

    Tutukina MN, Shakunov KS, Masulis IS, Ozoline ON: Intragenic promoter-like sites in the genome of Escherichia coli discovery and functional implication. J Bioinform Comput Biol 2007, 5: 549–560. 10.1142/S0219720007002801

    CAS  Article  PubMed  Google Scholar 

  31. 31.

    Fredrick K, Caramori T, Chen Y, Galizzi A, Helmann JD: Promoter architecture in the flagellar regulon of Bacillus subtilis : high-level expression of flagellin by the ΣD RNA polymerase requires an upstream promoter element. Proc Natl Acad Sci U S A 1995, 92: 2582–2586. 10.1073/pnas.92.7.2582

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  32. 32.

    Savchenko A, Weigel P, Dimova D, Lecocq M, Sakanyan V: The Bacillus stearothermophilus argCJBD operon harbours a strong promoter as evaluated in Escherichia coli cells. Gene 1998, 212: 167–177. 10.1016/S0378-1119(98)00174-7

    CAS  Article  PubMed  Google Scholar 

  33. 33.

    Aiyar SE, Gaal T, Gourse RL: rRNA promoter activity in the fast-growing bacterium Vibrio natrigens . J Bacteriol 2002, 184: 1349–1358.

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  34. 34.

    Sorokin AA, Osypov AA, Dzhelyadin TR, Beskaravainy PM, Kamzolova SG: Electrostatic properties of promoter recognized by E. coli RNA polymerase Esigma70. J Bioinform Comput Biol 2006, 4: 455–467. 10.1142/S0219720006002077

    CAS  Article  PubMed  Google Scholar 

  35. 35.

    Mitchell JE, Zheng D, Busby SJ, Minchin SD: Identification and analysis of "extended" promoters in Escherichia coli . Nucleic Acids Res 2003, 31: 4689–4695. 10.1093/nar/gkg694

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  36. 36.

    Stormo GD, Schneider TD, Gold L, Ehrenfeucht A: Use of the "Perception" algorithm to distunguish translational initiation sites in E. coli . Nucleic Acids Res 1982, 10: 2997–3011. 10.1093/nar/10.9.2997

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  37. 37.

    Baldi P, Chauvin Y, Hunkapiller T, McClure MA: Hidden Markov models of biological primary sequence information. Proc Natl Acad Sci USA 1994, 91: 1059–1063. 10.1073/pnas.91.3.1059

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  38. 38.

    Jarmer H, Larsen TS, Krpgh A, Saxild HH, Brunak S, Knudsen S: Sigma A recognition sites in the Bacillus subtilis genome. Microbiology 2001, 147: 2417–2424.

    CAS  Article  PubMed  Google Scholar 

  39. 39.

    Petersen L, Larsen TS, Ussery DW, On SL, Krogh A: RpoD promoters in Campylobacter jejuni exhibit a strong periodic signal instead of a -35 box. J Mol Biol 2003, 326: 1361–1372. 10.1016/S0022-2836(03)00034-2

    CAS  Article  PubMed  Google Scholar 

  40. 40.

    Munch R, Hiller K, Grote A, Scheer M, Klein J, Schobert M, Jahn D: Virtual Footprint and PRODORIC: an integrative framework for regulon prediction in prokaryotes. Bioinformatics 2005, 21: 4187–4189. 10.1093/bioinformatics/bti635

    Article  PubMed  Google Scholar 

  41. 41.

    Vanet A, Marsan L, Sagot M-F: Promoter sequences and algorithmical methods for identifying them. Res Microbiol 1999, 150: 779–799. 10.1016/S0923-2508(99)00115-1

    CAS  Article  PubMed  Google Scholar 

  42. 42.

    Waterman MS: Sequence alignments in the neighborhood of the optimum with general application to dynamic programming. Proc Natl Acad Sci USA 1983, 80: 3123–3124. 10.1073/pnas.80.10.3123

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  43. 43.

    Campbell EA, Muzzin O, Chlenov M, Sun JL, Olson CA, Weinman O, Trester-Zediltz ML, Darst SA: Structure of the bacterial RNA polymerase promoter specificity sigma subunit. Mol Cell 2002, 9: 527–539. 10.1016/S1097-2765(02)00470-7

    CAS  Article  PubMed  Google Scholar 

  44. 44.

    Vassylyev DG, Sekine S, Laptenko O, Lee J, Vassylyeva MN, Borukhov S, Yokoyama S: Crystal structure of a bacterial RNA polymerase holoenzyme at 2.6 A° resolution. Nature 2002, 417: 712–719. 10.1038/nature752

    CAS  Article  PubMed  Google Scholar 

  45. 45.

    Salgado H, Santos-Zavalets A, Gama-Castro S, Millán-Zárate D, Díaz-Peredo E, Sánchez-Solano F, Pérez-Rueda E, Bonavides-Martínez C, Collado-Vides J: RegulonDB (version 3.2): transcriptional regulation and operon organization in Escherichia coli K-12. Nucleic Acids Res 2001, 29: 72–74. 10.1093/nar/28.1.65

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  46. 46.

    Sakanyan V, Charlier D, Legrain C, Kochikyan A, Mett I, Piérard A, Glansdorff N: Primary structure, partial purification and regulation of key enzymes of the acetyl cycle of arginine biosynthesis in Bacillus stearothermophilus : dual function of ornithine acetyltransferase. J Gen Microbiol 1993, 139: 393–402.

    CAS  Article  PubMed  Google Scholar 

  47. 47.

    Karaivanova IM, Weigel P, Takahashi M, Fort C, Versavaud A, Van Duyne G, Charlier D, Hallet JN, Glansdorff N, Sakanyan V: Mutational analysis of the thermostable arginine repressor from Bacillus stearothermophilus : dissecting residues involved in DNA binding properties. J Mol Biol 1999, 291: 843–855. 10.1006/jmbi.1999.3016

    CAS  Article  PubMed  Google Scholar 

  48. 48.

    De Boer HA, Comstock LJ, Vasser M: The tac promoter: a functional hybrid derived from the trp and lac promoters. Proc Natl Acad Sci USA 1983, 80: 21–25. 10.1073/pnas.80.1.21

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  49. 49.

    Snapyan M, Lecocq M, Guevel L, Arnaud MC, Ghochikyan A, Sakanyan V: Dissecting DNA-protein and protein-protein interactions involved in bacterial transcriptional regulation by a sensitive protein array method combining a near-infrared fluorescence detection. Proteomics 2003, 3: 647–657. 10.1002/pmic.200300390

    CAS  Article  PubMed  Google Scholar 

  50. 50.

    Pratt JM: Coupled transcription-translation in prokaryotic cell-free systems. In Transcription and translation: a practical approach. Edited by: Hames BD, Higgins SJ. New York: IRL Press; 1984:179–209.

    Google Scholar 

  51. 51.

    Kim DM, Swartz JR: Prolonging cell-free protein synthesis with a novel ATP regeneration system. Biotechnol Prog 2000, 16: 385–390. 10.1002/(SICI)1097-0290(1999)66:3<180::AID-BIT6>3.0.CO;2-S

    CAS  Article  PubMed  Google Scholar 

  52. 52.

    Schneider TD, Stephens RM: Sequence Logos: a new way to display consensus sequences. Nucleic Acids Res 1990, 18: 6097–6100. 10.1093/nar/18.20.6097

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  53. 53.

    Crooks GE, Hon G, Chandonia JM, Brenner SE: WebLogo: A sequence logo generator. Genome Res 2004, 14: 1188–1190. [] 10.1101/gr.849004

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  54. 54.

    Weiner J III, Herrmann R, Browning GF: Transcription in Mycoplasma pneumoniae . Nucleic Acids Res 2000, 28: 4488–4496. 10.1093/nar/28.22.4488

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  55. 55.

    Deckert G, Warren PV, Gaasterland T, Young WG, Lenox AL, Graham DE, Overbeek R, Snead MA, Aujay M, Huber R, Feldman RA, Short JM, Olsen GJ, Swanson RV: The complete genome of the hyperthermophilic bacterium Aquifex aeolicus . Nature 1998, 392: 353–358. 10.1038/32831

    CAS  Article  PubMed  Google Scholar 

  56. 56.

    Giacani L, Hevner K, Centurion-Lara A: Gene organization and transcriptional analysis of the trpJ , trpI , trpG , and trpF loci in Treponema pallidum strains Nichols and Sea 81–4. J Bacteriol 2005, 187: 6084–6093. 10.1128/JB.187.17.6084-6093.2005

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  57. 57.

    Nelson KE, Clayton RA, Gill SR, Gwinn ML, Dodson RJ, Haft DH, Hickey EK, Peterson JD, Nelson WC, Ketchum KA, McDonald L, Utterback TR, Malek JA, Linher KD, Garrett MM, Stewart AM, Cotton MD, Pratt MS, Phillips CA, Richardson D, Heidelberg J, Sutton GG, Fleischmann RD, Eisen JA, White O, Salzberg SL, Smith HO, Venter JC, Fraser CM: Evidence for lateral gene transfer between Archaea and Bacteria from genome sequence of Thermotoga maritima . Nature 1999, 399(6734):323–329. 10.1038/20601

    CAS  Article  PubMed  Google Scholar 

  58. 58.

    Paul BJ, Ross W, Gaal T, Gourse RL: rRNA transcription in Escherichia coli . Annu Rev Genet 2004, 38: 749–770. 10.1146/annurev.genet.38.072902.091347

    CAS  Article  PubMed  Google Scholar 

  59. 59.

    Dove SL, Darst SA, Hochschild A: Region 4 of sigma as a target for transcription regulation. Mol Microbiol 2003, 48: 863–874. 10.1046/j.1365-2958.2003.03467.x

    CAS  Article  PubMed  Google Scholar 

  60. 60.

    Murakami K, Fujita N, Ishihama A: Transcription factor recognition surface on the RNA polymerase alpha subunit is involved in contact with the DNA enhancer element. EMBO J 1996, 15(16):4358–4367.

    PubMed Central  CAS  PubMed  Google Scholar 

  61. 61.

    Eskin E, Keich U, Gelfand MS, Pevzner PA: Genome-wide analysis of bacterial promoter regions. Pac Symp Biocomput 2003, 29–40.

    Google Scholar 

  62. 62.

    Huerta AM, Collado-Vides J: Sigma70 promoters in Escherichia coli : specific transcription in dense regions of overlapping promoter-like signals. J Mol Biol 2003, 333: 261–278. 10.1016/j.jmb.2003.07.017

    CAS  Article  PubMed  Google Scholar 

  63. 63.

    Manganelli R, Proveddi R, Rodrigue S, Beaucher J, Gaudreau L, Smith I: σ factors and global regulation in Mycobacterium tuberculosis . J Bacteriol 2004, 186: 895–902. 10.1128/JB.186.4.895-902.2004

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  64. 64.

    Woese CR, Kandler O, Wheelies ML: Towards a natural system of organisms: proposal for the domains Archaea, Bacteria, and Eucarya. Proc Natl Acad Sci USA 1990, 87: 4576–4579. 10.1073/pnas.87.12.4576

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  65. 65.

    Braun F, Marhuenda FB, Morin A, Guevel L, Fleury F, Takahashi M, Sakanyan V: Similarity and divergence between the RNA polymerase alpha subunits from hyperthermophilic Thermotoga maritima and mesophilic Escherichia coli bacteria. Gene 2006, 380: 120–126. 10.1016/j.gene.2006.05.020

    CAS  Article  PubMed  Google Scholar 

  66. 66.

    Gaal T, Ross W, Blatter EE, Tang H, Jia X, Krishnan VV, Assa-Munt N, Ebright RH, Gourse RL: DNA-binding determinants of the alpha subunit of RNA polymerase: novel DNA-binding domain architecture. Genes & Dev 1996, 10: 16–26. 10.1101/gad.10.1.16

    CAS  Article  Google Scholar 

  67. 67.

    Ross W, Gourse RL: Sequence-independent upstream DNA-αCTD interactions strongly stimulate Escherichia coli RNA polymerase- lacUV5 promoter association. Proc Natl Acad Sci USA 2005, 102: 291–296. 10.1073/pnas.0405814102

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  68. 68.

    Kim YR, Lee SE, Kim CM, Kim SY, Shin EK, Shin DH, Chung SS, Choy HE, Progulske-Fox A, Hillman JD, Handfield M, Rhee JH: Characterization and pathogenic significance of Vibrio vulnificus antigens preferentially expressed in septicemic patients. Infect Immun 2003, 71: 5461–5471. 10.1128/IAI.71.10.5461-5471.2003

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  69. 69.

    Johnson MR, Conners SB, Montero CI, Chou CJ, Shockley KR, Kelly RM: The Thermotoga maritima phenotype is impacted by syntrophic interaction with Methanococcus jannaschii in hyperthermophilic coculture. Appl Environ Microbiol 2006, 72: 811–818. 10.1128/AEM.72.1.811-818.2006

    PubMed Central  CAS  Article  PubMed  Google Scholar 

  70. 70.

    Vellanoweth RL, Rabinowitz JC: The influence of ribosome-binding-site elements on translational efficiency in Bacillus subtilis and Escherichia coli in vivo . Mol Microbiol 1992, 6: 1105–1114. 10.1111/j.1365-2958.1992.tb01548.x

    CAS  Article  PubMed  Google Scholar 

  71. 71.

    Boyce JD, Cullen PA, Adler B: Genomic-scale analysis of bacterial genes and protein expression in the host. Emerg Infect Dis 2004, 10: 1357–1362.

    PubMed Central  CAS  Article  PubMed  Google Scholar 

Download references


At the time of the study AM was a research fellow supported by Pays de la Loire. MD acknowledges support from the Conseil Régional des Pays de la Loire and ProtNeteomix for his visit to Nantes University. We should like to thank anonymous reviewers whose suggestions allowed us to improve the manuscript. This study was supported by the "Post-Génome programme des Pays de la Loire", by the EU project EUR-INTAFAR (n°LSHM-CT-2004-512138), and by the R&D program of ProtNeteomix.

Author information



Corresponding authors

Correspondence to Michael Dekhtyar or Vehary Sakanyan.

Additional information

Competing interests

The authors declare that they have no competing interests.

Authors' contributions

MD developed the algorithm and performed the computational analysis; AM conducted cell-free experiments; VS designed the project, contributed to the development of the algorithm and data analysis, and wrote the manuscript.

Electronic supplementary material

Authors’ original submitted files for images

Rights and permissions

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Reprints and Permissions

About this article

Cite this article

Dekhtyar, M., Morin, A. & Sakanyan, V. Triad pattern algorithm for predicting strong promoter candidates in bacterial genomes. BMC Bioinformatics 9, 233 (2008).

Download citation


  • Bacterial Genome
  • Putative Promoter
  • Strong Promoter
  • Candidate Sequence
  • Hyperthermophilic Bacterium