The rimP gene ranks first in our list of genes most often regulated by attenuation (Figure 1). rimP, previously known as yhbC in E. coli and ylxS in B. subtilis, encodes a protein recently shown to be involved in 30S ribosomal subunit maturation 40 . It is the first gene of an operon encompassing the nusA and infB genes, which are present in almost all bacteria. While infB encodes the translation initiation factor IF-2, the NusA protein is characterized as a transcriptional pausing, readthrough, termination and anti-termination factor, and is shown to participate in the Rho-dependent anti-termination complex 41 .

Analysis of predicted attenuators in rimP-nusA-infB operons (Figure 2b) revealed the presence of a short and highly conserved motif corresponding to the terminator, the 'rimP-leader', but gave no evidence of larger conserved elements characteristic of riboswitches, T-boxes or ribosomal protein-dependent attenuators. The terminator stem contains an unusual highly conserved GGGc(...)gCCC motif. We were unable to detect such a conserved motif in any other 5' or 3' terminator, suggesting this sequence signature is specific to the rimP-leader. We could find, however, the same motif along with the downstream U-stretch in many 5' UTRs of rimP-nusA-infB operons where no terminator structure was detected (Supplementary data 1 in Additional file 1). These additional motifs were missed because the potential hairpin was too short for detection with our programs, consistent with our previous observation that regulatory terminators are less stable than regular terminators. The distance separating the terminator from the gene start varies between 4 and 130 nucleotides, and may consequently encompass the ribosome-binding site (RBS). In several Gammaproteobacteria (listed in Supplementary data 1 in Additional file 1), the rimP-leader appears more complex, with a second terminator found in tandem and upstream of the former (Figure 2a), and presenting a clear potential anti-terminator structure. Interestingly, this terminator presents a CCCg(...)cGGG motif, inverse to the downstream motif.

High sequence conservation in the rimP-leader terminator stem and the absence of any visible antiterminator structure argue for regulation involving a termination or anti-termination protein that can specifically recognize a nucleic-acid motif 12 13 . If feedback control by RimP, now known to interact with rRNA 40 , may be hypothesized, the best candidate for this direct interaction is probably the NusA protein, the expression of which was shown 25 years ago to repress expression of the operon 42 43 . NusA contains an amino-terminal domain that interacts with RNA polymerase, an S1 domain frequent in RNA-associated proteins, and two RNA-binding K homology (KH) domains 44 . It was already shown to be involved in the attenuation of the Trp, His and S10 operons 45 by interacting with the upstream arm of the terminator hairpin, but the RNA motif we describe here was not observed in these instances.

While this manuscript was under review, a deep sequencing study of 5' regulators in B. subtilis 46 observed that transcripts encoding certain core transcription elongation subunits, including ylxS (that is, rimP), appear to contain a long 5' leader region. The authors suggested these regions may contain elements regulating the associated genes. The 180-nucleotide leader they observed by deep sequencing in B. subtilis rimP transcripts indeed covers the attenuator we predict for this gene (Figure 2).

The rpsL gene also appears at the top of the list of genes frequently regulated by transcriptional attenuation (Figure 1). Like rimP, it belongs to an operon of largely conserved genes, rpsL and rpsG, encoding the ribosomal proteins S12 and S7 respectively, and fusA and tufA, encoding the elongation factors EF-G and EF-Tu, respectively. Sequence-based clustering allowed us to identify variants among the different 'rpsL-leaders' (Table S2 in Additional file 1). We derived consensus structures for several of these elements and were able to find potential alternative antiterminator structures in each case (Figure 3).

The translation of rpsL and rpsG was already shown to be controlled by S7, the product of rpsG, in E. coli 47 , and Merino and Yanofsky 17 predicted putative transcription attenuators upstream of rpsL in 24 species. However, their results scantly overlap our own predictions: we have only four common candidates, while we scanned 22 of their 24 species. The occurrences we missed with our protocol involved non-canonical terminators (with a long, GC-poor and/or bulged hairpin), or terminators that were too far from the ATG to meet our distance filter (8 of 18 cases).

The persistence of an attenuator element upstream of rpsL in widely divergent species argues for a common origin. However, we found no globally conserved feature, neither in sequence nor in structure, associated with this attenuator. This probably explains why no common structure has been proposed for the rpsL-leader previously. In the Streptococcus genus, we found no attenuator upstream of rpsL, contrary to Merino's analysis, which found one (this terminator is too far from the ATG codon (189 nucleotides) to satisfy our distance filter). However, we found a candidate further downstream between rpsG and fusA in streptococci (Figure 3b). It is possible that two similar elements are present in this operon, since Meyer et al. 48 found two similar RNA structures upstream of rpsL and fusA in the Proteobacteria Candidatus Pelagibacter ubique. This would be consistent with co-regulation of the two genes. The element identified in 48 , however, does not meet our terminator criteria and does not present any common sequence feature with any of our predicted attenuators.

Is the structure and sequence diversity in rpsL-leaders compatible with an interaction with the same S7 protein partner in all species? The highly flexible RNA binding portion of S7 49 could tolerate some variation in RNA targets or, alternatively, the leader may bind different protein partners. In the current view of ribosomal protein leaders, each leader family displays a characteristic motif that mimics corresponding binding sites in ribosomal RNA. This view may be too restrictive and the example of rpsL suggests that the modalities of interaction may differ across distant phyla.

ATP-binding cassette (ABC) transporters constitute one of the largest and most ancient protein families, with hundreds of paralogs transporting a wide variety of substrates across the plasmic membrane, including ions, amino acids, lipids and drugs 50 51 . In Bacteria, these multiproteic complexes are encoded by operons comprising genes for ATPase, permease and periplasmic components. A number of them were shown to be regulated by transcriptional factors 52 , whereas very few are known to be subject to transcriptional attenuation 53 . ABC transporters do not appear in Figure 1, where scores are weighted for family size; however, this family presents the highest absolute number of genes regulated by attenuation (Table S0 in Additional file 1), with a total of 205 candidates in our study, and an enrichment P-value of 8.8e-05. Further scrutiny of these candidates is important because of the great diversity of transporters and potential variability of regulatory elements controlling them. We found different sequence motifs associated with these 'ABC-leaders' (listed in Table S3 in Additional file 1).

Figure 4 shows five ABC-leaders associated with transporters of either known (Figure 4a-d) or unknown (Figure 4e) substrates. Each is able to form an antiterminator structure and the conserved sequence/structure motif (see alignments in Supplementary data 2 in Additional file 1) suggests that it responds to a unique substrate. The candidate shown in Figure 4c is a probable T-box, but the other candidates do not resemble any known cis-regulator. Their regulatory mechanisms thus remain to be determined. The size of the conserved structure is sufficient to form an aptamer that could directly detect a substrate, thus defining new classes of riboswitches; however, we cannot exclude an indirect regulation involving a protein factor. Indeed, a number a RNA-binding proteins target palindromic RNA 12 13 that may also act as a terminator hairpin.

The analysis of such a multiple paralog family raises interesting evolutionary questions on the origin of associated attenuators, that is, whether they all derive from an ancient attenuator that would have regulated the ancestral ABC transporter, or if certain genes of this family tend to 'attract' attenuators for their regulation. The relatively low proportion of attenuated ABC transporter genes and the ability of attenuators to 'hop' from one gene to another (see below) argue for the second hypothesis.

Syntenic attenuation systems do not necessarily use the same sensing system. There are well known examples of switches from a T-box or a riboswitch in certain species (for example, Firmicutes) to a leader peptide in others (for example, Proteobacteria) 9 10 . Our protocol, which does not require a conserved RNA sequence or structure, is well suited to detect such exchanges, and at least five are present in our list of frequently attenuated genes (Figure 1).

A particularly striking case of switch between sensing systems is provided by the hisS gene (Figure 5). In the Bacillus genus, hisS, encoding the histidyl-tRNA synthetase, has been long known to be regulated by a T-box, like many other tRNA synthetases 9 . This gene, however, underwent two successive duplications: a recent one that appears specific to bacilli, and a more ancestral one that occurred before divergence of the Proteobacteria and Firmicutes. Interestingly, all three paralogs now found in bacilli are predicted to be regulated by a different type of attenuator, as shown in Figure 5. In addition to the known hisS T-box, we found novel attenuators upstream of hisS* (the Bacillus hisS paralog) and hisZ, encoding an ATP phosphoribosyltransferase regulatory subunit.

We performed a sequence-based clustering of the 5' UTR regions in the hisS family to identify sets of related motifs (Figure 5). The 5' UTRs of hisZ and of hisS* have highly similar sequences that include short ORFs encompassing a stretch of histidine codons. This strongly argues for the presence of a histidine leader peptide regulating both genes. To our knowledge, no leader peptide system had been shown to exist outside of Proteobacteria 22 and Actinobacteria 54 , if we exclude the atypical ermC leader 55 that controls translation and has no amino acid specificity but senses a global slowdown in translation. This result thus strengthens the evolutionary relevance of leader peptide systems and expands the range of RNA-based regulation in Gram-positive bacteria.

Leader peptides may have spread to Firmicutes by horizontal transfer. However, we may also hypothesize that short reading frames may have emerged repeatedly from random sequences, especially in the favorable cellular environment of species such as those in the Firmicutes phylum. In support of convergent evolution, no similarity in the non-coding or leader peptide sequence is observed between the Firmicutes and Proteobacteria. That the two hisS gene duplications are observed only in certain Firmicutes species suggests a recent event: a gene resulting from the first duplication may have evolved or captured an attenuation system (for example, from a Gram-negative bacteria) before undergoing a second duplication.

Sequence analysis of hisS attenuators also reveals a T-box element in Lactobacillales (Figure 5a). Furthermore, we observed one possible horizontal transfer of attenuator elements from Firmicutes to Gammaproteobacteria: an unknown proteobacterial attenuator, which corresponds to a sequence encompassing a putative short ORF, clearly resembles Firmicute T-boxes (Figure 5c, red box). This illustrates the remarkable lability of attenuator elements, which can be acquired from other species and subsequently evolve to fit the preferred regulatory mechanisms of their new host.

The greA/rnk gene family ranks seventh in our list of genes frequently regulated by attenuation. Although the propensity of greA/rnk genes for transcriptional attenuation was detected previously 17 , experimental evidence for an E. coli greA attenuator is recent 4 . The gene family includes two major paralogs, greA, which encodes a transcription elongation factor, and rnk, which encodes a regulator of nucleoside diphosphate kinase. We found attenuators upstream of these genes in species ranging from Proteobacteria to bacilli and Clostridia. In several species, we identified attenuators in both genes. Figure 6 shows the result of a sequence-based clustering of greA/rnk attenuators and secondary structure models for different sequence clusters. In each case, we were able to detect a clear antiterminator structure; however, no common feature could be detected between the greA- and rnk-leaders.

Very interestingly, the limited experimental evidence available on these two putative cis non-coding RNAs argues for a second mechanism in trans. Potrykus et al. 4 showed that overexpression of the short form of the greA transcript, released after attenuation has occurred, leads to repression of several genes. Furthermore, an intergenic region corresponding to the rnk-leader was found in a systematic screening 56 to co-immunoprecipitate with Hfq, a conserved bacterial protein known to facilitate interaction between small RNAs and their target mRNA. Although leaders doubling as trans-acting RNAs is a recent and, for the time being, rare finding 57 , we may have found here two such cases with the greA- and rnk-leader families.

Forty-six terminator clusters in Enterobacteria and 96 clusters in bacilli are associated with transposable elements. We describe them as 'mobile attenuators'. Terminator sequences in these clusters show a high level of conservation, consistent with an origin from transposable elements of recent dissemination. They fall into two classes. The first class (Figure 7a) corresponds to terminators located upstream of transposases or other insertion sequence (IS)-related genes and is mainly species-specific. Of note, transposase or IS-related genes also rank high in the list of frequently attenuated gene families, when no normalization for family size is applied (Table S0 in Additional file 1). These genes belong to transposon families such as ISBma2 (mainly present in the Betaproteobacteria Burkholderia mallei, in Bacillus thuringiensis and in the Clostridia Symbiobacterium thermophilum, Thermoanaerobacter tengcongensis and Clostridium novyi), IS3/IS2/IS600/IS1329/IS407A (found sporadically in all phyla) and ISL3 (mainly present in different Firmicutes species). The second class of mobile attenuators (Figure 7b) represents families of related transposon-borne sequences located immediately upstream of different, unrelated cellular genes.

The emergence of mobile attenuators can be explained by the structure of the IS containing both 3' and 5' transcription terminators 59 60 (Figure 7). Terminators of the first class correspond to IS 5' terminators, whose function is to limit transposon proliferation when inserted in a coding region under control of an active promoter. Such transposition events would be deleterious for the host, and consequently for the element's own survival. Clusters of conserved attenuators located upstream of unrelated genes (Figure 7b) may correspond to 3' terminators of ISs. The significant proportion (25 to 35%) of conserved terminator clusters that result from IS transposition suggests they have a significant impact in genome evolution, particularly in terms of regulation. The possible pseudogenization of transposed genes may allow exaptation of their 5' and/or 3' terminators and give birth to new regulatory elements (Figure 7b). It should be noted that this exaptation of transposable elements for regulatory functions is probably a universal mechanism, as several instances are already described in eukaryotic genomes 61 62 .

Ten terminator clusters in Enterobacteria and 23 in bacilli are found in different species associated with the same set of non-homologous genes. They may therefore constitute attenuator regulons (Tables S4 and S5 in Additional file 1). Several known riboswitches fall in this class (15 out of 23 clusters in bacilli), which is consistent with the ability of riboswitches to occur upstream of unrelated genes from distant species. Analysis of these clusters thus seems promising for the identification of new regulons, although this class also contains sequences from repeated elements. For instance, the largest cluster detected in Enterobacteria (Figure 8) involves elements conserved upstream of the gatY, rbfA, cvpA and, in some cases, mscS, dppB, yidB and sbp genes. Several members of this cluster present significant similarities with the rtT transcript identified by Bösl and Kersten 63 , a by-product of the tyrT operon that could have a role in stringent response, supporting the biological significance of these candidate attenuators. Interestingly, we found these highly conserved elements to belong to bacterial interspersed mosaic elements (BIMEs) 64 . Even though this is a repeated element, it probably constitutes a bona fide attenuator as well. Indeed, BIMEs have been shown to act as bidirectional transcription terminators 65 and to induce transcriptional attenuation when placed upstream of a tRNA reporter gene 66 . The high conservation of these repeats, both in sequence and physical location, among Enterobacteria supports an exaptation scenario, possibly as an attenuator system. A second large attenuator cluster involving the genes artJ, ygdL, yhfZ and yijO was also found to correspond to BIMEs (Table S4 in Additional file 1).

Clusters of attenuators located upstream of different genes suggest coexpression, and therefore functional relationships between these genes. In bacilli, an interesting such element, which we called the 'vanR-leader', was found upstream of a gene encoding a flavin oxidoreductase and of the vanR gene, encoding a TCS transcription factor involved in vanillate catabolism (Figure 9; alignment in Supplementary data 3 in Additional file 1), suggesting a concerted regulation of these two genes. Another common element is found upstream of a sulfonate ABC transporter operon and upstream of an unknown gene. Its involvement in sulfonate metabolism may consequently be hypothesized (alignment in Supplementary data 3 in Additional file 1). Finally, a cluster of related attenuators, which we named 'fabH-leaders', involves elements upstream of the fabH gene, involved in fatty acid metabolism, and upstream of genes encoding a DNA binding protein, for which an association with the same pathway would be interesting to test (alignment in Supplementary data 3 in Additional file 1).