Identification, variation and transcription of pneumococcal repeat sequences
© Croucher et al; licensee BioMed Central Ltd. 2011
Received: 7 November 2010
Accepted: 18 February 2011
Published: 18 February 2011
Skip to main content
© Croucher et al; licensee BioMed Central Ltd. 2011
Received: 7 November 2010
Accepted: 18 February 2011
Published: 18 February 2011
Small interspersed repeats are commonly found in many bacterial chromosomes. Two families of repeats (BOX and RUP) have previously been identified in the genome of Streptococcus pneumoniae, a nasopharyngeal commensal and respiratory pathogen of humans. However, little is known about the role they play in pneumococcal genetics.
Analysis of the genome of S. pneumoniae ATCC 700669 revealed the presence of a third repeat family, which we have named SPRITE. All three repeats are present at a reduced density in the genome of the closely related species S. mitis. However, they are almost entirely absent from all other streptococci, although a set of elements related to the pneumococcal BOX repeat was identified in the zoonotic pathogen S. suis. In conjunction with information regarding their distribution within the pneumococcal chromosome, this suggests that it is unlikely that these repeats are specialised sequences performing a particular role for the host, but rather that they constitute parasitic elements. However, comparing insertion sites between pneumococcal sequences indicates that they appear to transpose at a much lower rate than IS elements. Some large BOX elements in S. pneumoniae were found to encode open reading frames on both strands of the genome, whilst another was found to form a composite RNA structure with two T box riboswitches. In multiple cases, such BOX elements were demonstrated as being expressed using directional RNA-seq and RT-PCR.
BOX, RUP and SPRITE repeats appear to have proliferated extensively throughout the pneumococcal chromosome during the species' past, but novel insertions are currently occurring at a relatively slow rate. Through their extensive secondary structures, they seem likely to affect the expression of genes with which they are co-transcribed. Software for annotation of these repeats is freely available from ftp://ftp.sanger.ac.uk/pub/pathogens/strep_repeats/.
Small interspersed repeats, spatially separated genomic regions of similar sequence typically < 200 bp in length, are frequently found in bacterial chromosomes . These can be classified as either 'simple', when consisting of a single repeated unit, or 'composite', when comprised of a combination of different subsequences arranged in particular patterns . For example, a number of enterobacterial species harbour many instances of the simple 127 bp Enterobacterial Repetitive Intergenic Consensus (ERIC) sequence  and hundreds of composite Bacterial Interspersed Mosaic Elements (BIMEs), which include multiple copies of the Palindromic Unit in a regular configuration. Similarly, Neisseria meningitidis genomes host simple 183 bp AT-rich Repeats and two families of more common, composite elements: 70-200 bp Neisserial Intergenic Mosaic Elements (NIMEs) and Correia Elements (CE), comprised of internal sequences up to 156 bp long delimited by 26 bp inverted repeats .
Many such repeat families are likely to be non-autonomous mobile parasitic elements, termed Miniature Inverted-repeat Transposable Elements (MITEs). These are characterized as being AT-rich, possessing terminal inverted repeats (TIR), having highly base-paired secondary structures and generating target site duplications (TSDs) on insertion . In a number of cases, it has been proposed that repeats are mobilized by the transposases encoded by IS elements within the same host, based on similarities between the TIR of the MITE and the IS sequence. For instance, the Nezha MITE found in cyanobacteria is proposed to be mobilized by ISNpu3-like elements .
The tightly folded secondary structure characteristic of putative MITEs means they can impact on gene expression when they insert into transcribed regions. Some BIMEs, when inserted into operons, have been found to decrease the expression of downstream CDSs through acting as transcriptional attenuators . By contrast, regions upstream of ERIC elements integrated into operons may be destabilised by the presence of the repeat when in a specific orientation, as it appears to trigger transcript cleavage through introducing a putative RNase E target site . Similarly, there is evidence that CE act as a target site for RNase III-mediated endoribonucleolytic cleavage when transcribed [8, 9]. CE insertions have also been found to influence gene expression through generating functional promoters in N. meningitidis . As well as affecting transcriptional regulation, repeat sequences can alter the sequences of genes without disrupting their function. For instance, in Rickettsia, repeat element insertions have been found in both coding and non-coding genes that appear still to be functional [11, 12].
Streptococcus pneumoniae is a nasopharyngeal commensal and major respiratory pathogen estimated to have caused almost 15 million cases of disease in 2000 . The genome, typically around 2 Mb in size, is known to contain two types of small interspersed repeat. The first to be discovered was the BOX element, a composite repeat consisting of boxA and boxC sequences usually separated by a variable number of boxB elements arranged in a tandem array . The variation in different strains' complements of these repeats has allowed them to form the basis of a PCR-based epidemiological typing scheme . An early hypothesised function of BOX elements, based on their proximity to a number of genes involved in competence and pathogenesis, was that they might act as regulatory motifs , and subsequent experiments have shown that boxA and boxC elements are able to stimulate the expression of downstream genes, although boxB elements can have an opposing inhibitory effect, depending on their orientation . A BOX element has also been hypothesised to increase the frequency of pneumococcal phase variation through affecting the regulation of neighbouring genes . Similarity between the TIR of BOX elements and ISSpn2, a transposon found in S. pneumoniae, has been proposed as the basis for mobilization of these elements. Likewise a second repeat also present in high copy number in the pneumococcal genome, the simple 107 bp long Repeat Unit of Pneumococcus (RUP), has TIR similar to those of IS630-Spn1, another transposon commonly found in S. pneumoniae . RUP were proposed to preferentially insert into or near IS elements, based on their distribution in a draft of the S. pneumoniae TIGR4 genome , leading to the suggestion that these elements may serve to limit the number of functional transposase genes in the chromosome .
Here we present an analysis of the distribution of pneumococcal small interspersed repeats throughout the publicly available streptococcal genomes and outline how these elements may have impacted upon the evolution of pneumococcal coding and non-coding genes.
Following refinement of the models (see Methods), the final HMMs used to identify the repeats are represented as logos in Figure 1. Overall, 125 BOX (composed of 422 modules), 110 RUP and 30 SPRITE elements were found in the ATCC 700669 genome; in addition, 17 lone box modules were found. All of the original examples used to define BOX and RUP elements were identified by this approach [14, 18]. It seems likely that the lower frequency of the SPRITE repeat is the explanation as to why it was not characterised prior to the availability of complete genome sequences.
Each of the three families of repeats share at least some features of MITEs. All are typically < 200 bp in length; unsurprisingly, the modular BOX elements are the most variable in size, ranging from 67 bp to 637 bp. Both RUP and SPRITE are AT-rich relative to the S. pneumoniae genome (GC content of 39.5%), with mean GC levels of 27.5% and 28.1% respectively. Both BOX and RUP have been previously shown to have TIR and cause TSDs on insertion [14, 16, 18]. SPRITE repeats have comparatively shorter and simpler TIR (the tetranucleotide AAAA and the complement TTTT; Figure 1c). Any TSD produced by SPRITE insertions could not be established from the current dataset, because no instances of the repeat with an easily comparable empty site could be found in the available collection of sequences, and no clear evidence could be identified by examining the regions flanking insertions.
All three elements are predicted to form stem-loop structures if transcribed into an RNA form (Figure 2). The structure of BOX elements was generated from those elements with a canonical A1B1C1 sequence; notably, the folding of the boxB element is predicted to involve few interactions with the boxA and C elements that form the rest of the structure. If this folded RNA is functional, this characteristic may be permissive in allowing boxB to be absent, or present in multiple copies, without causing much disruption to the overall form of the transcript.
The SPRITE structure is less tightly folded than that of BOX or RUP, and consists of an 18 bp duplex followed by a relatively uridine-rich (~48% uridine) tract, seeming likely to imbue it with the properties of a Rho-independent terminator. However, the repeat's structure is distinctive in that both the stem duplex and T-rich tract are much longer than the ~10 bp size of both these features in typical streptococcal Rho-independent terminators . Hence it appears that SPRITE are distinct from normal Firmicute terminators, although they may be able to function in such a capacity.
Overrepresentation of repeats between convergently transcribed genes
No. Upstream of ≥ 1 CDS
No. Between Convergently Transcribed CDSs
3.4 × 10-7
1.7 × 10-9
5.1 × 10-8
Across the pneumococcal chromosome, the size of intergenic distances follows a gradually decaying distribution (Figure 3b). A similar pattern is observed with the distances between BOX elements and the nearest gene, whereas the density of RUP elements is greatest 50-150 bp from the nearest gene. IS elements have an even more pronounced tendency to be distant from neighbouring CDSs; this may reflect the greater potential disruption to gene expression caused by these longer repeats should they insert within, or near, functional transcripts. SPRITE sequences tend to be close to adjacent CDSs, with only one SPRITE found >200 bp from the nearest gene. This enrichment of SPRITE close to the 3' termini of CDS suggests they may have been co-opted by the pneumococcus into acting as functional transcriptional terminators.
One way in which BOX elements are observed to vary quite considerably is in their size (Figure 5b). Several mechanisms have been proposed to explain the fluctuation in the length of tandem repeat arrays, including slipped strand mispairing, unequal crossover during homologous recombination and circular excision followed by reinsertion . Plotting the mean size of each BOX element insertion against the range of the lengths of the insertion in different genomes reveals a positive linear correlation (R 2 = 0.74, p < 2.2 × 10-16). This implies that the greater the average number of boxB repeats in a BOX element, the more likely that element is to vary by losing or acquiring these modules. Notably, all BOX elements with a large mean size exhibit considerable variation in length between strains. This result indicates that at the disparate loci at which BOX elements are found, there is significant variation in the rate of mechanisms that change the number of boxB modules in these arrays, or greatly differing levels of selection pressure constraining the size of these composite repeats.
The application of the HMMs to the genomes of other nasopharyngeal commensals (Haemophilus influenzae, Neisseria meningitidis and Staphylococcus aureus) failed to identify any cases where the repeats had been horizontally transferred. A similar investigation of all publicly available complete streptococcal genomes, encompassing twelve species other than S. pneumoniae, also detected few instances of these repeat elements (Additional file 1). The sole representative genome of the most closely related species to S. pneumoniae, S. mitis B6 , contained 104 BOX elements (a mean density of 0.048 kb-1), slightly lower than the mean of 122 in the pneumococcal chromosomes (a mean density of 0.057 kb-1). By contrast, the density of SPRITE sequences in S. mitis is about half that of the pneumococcus, and there are only 9 detected instances of RUP in S. mitis B6. As S. mitis and S. pneumoniae are able to exchange DNA, it is not clear whether the repeats were present in their last common ancestor, or whether they have been acquired after speciation and subsequently spread horizontally. By contrast, all three repeat types are almost entirely absent from the genome of S. sanguinis, the only other mitis group streptococci to have been sequenced. Hence the most parsimonious conclusion is that these elements have spread in the pneumococcal chromosome subsequent to the divergence of the more distantly related members of the mitis group.
The only other streptococcal species to have a comparatively high number of detected repeats was S. suis, all genomes of which had 11 boxC elements. These were found to coincide with previously discovered repeats, annotated as 'RepSU1', on the complementary strand of the genome in strains SC84, P1/7 and BM407 . Further analysis revealed the presence of two novel families of BOX-type elements in these genomes, composed of a total of seven different subsequences in particular permutations. One is bounded by boxA and C modules, both of which are around 50 nt long, as are the pneumococcal equivalents. The RepSU1 elements accounted for only the smallest BOX-type repeats of this type, equivalent to A1C1 BOX sequences. The other family has a boxE sequence at the 5' end and a boxF module at the 3' end; these motifs are comparatively large, having mean sizes of 115 nt and 133 nt respectively. Both types are found surrounding the same type of intervening boxB modules; however, the boxAC-flanked elements are also sometimes found having boxD modules, always in addition to boxB modules. Hence the diversity of S. suis BOX elements appears to be greater than that of the S. pneumoniae equivalents.
BOX, RUP and SPRITE elements are frequently found together in clusters, and appear to have inserted into one another on a number of occasions. These spatial groupings may reflect a common preference for insertion sites, or a general tolerance of insertions in certain regions of the chromosome. However, repeats are also found interspersed within pseudogenes and regulatory sequences. It is known that BOX insertions can affect the expression of nearby genes [16, 17]; another example where they might impact on the transcription of an operon is upstream of the trp gene cluster. In many Gram positive species, this operon is regulated by two copies of the T box riboswitch, which binds uncharged tRNA. Whilst streptococci have previously been thought to only have a single copy , in fact the pneumococcus has two, separated by a A1B2C1 BOX element. This results in the formation of a compound 5' untranslated region nearly a kilobase long, composed of three elements that, given their individually tightly folded structures, seem likely to fold largely independently.
A number of protein coding genes are disrupted by repeat insertions. Instances found in genome annotations include orthologues of the S. pneumoniae TIGR4 CDS SP_0243, encoding the extracellular binding protein for a putative iron ABC transporter, which is disrupted by the insertion of a RUP element in all the other pneumococcal genomes except S. pneumoniae AP200, 670-6B and TIGR4 itself. However, another CDS encoding part of the same ABC transporter (SP_0241 in TIGR4) is disrupted through frameshift mutations in these three strains. Both of these CDSs appear to be intact in several incompletely sequenced S. mitis strains, which lack the alternative pit2 iron transport system found on Pneumococcal Pathogenicity Island 1 . SPN23F05190 (TIGR4 orthologues SP_0574 and SP_0575), encoding a restriction endonuclease in S. pneumoniae ATCC 70069, has a RUP insertion in S. pneumoniae TIGR4 and D39, whilst the orthologous gene in S. pneumoniae AP200 has been disrupted through the insertion of an IS element. Further examination of the repeat insertions reveals a RUP insertion that has knocked out a serine/threonine protein kinase, previously annotated as two separate CDSs (e.g. SPN23F18490 and SPN23F18500 in S. pneumoniae ATCC 700669; SP_1831 and SP_1832 in S. pneumoniae TIGR4), in all strains except S. pneumoniae Taiwan 19F-14 and TCH8431/19A. BOX elements can also cause gene disruption through insertion: a gene encoding a DNA alkylation repair protein is disrupted by a BOX insertion in all the available pneumococcal sequences, whilst an E1B1F1 element appears to have inserted into an acetyltransferase pseudogene in the sequenced S. suis genomes. Hence the mobility of these repeats has the potential to contribute to phenotypic polymorphism in the S. pneumoniae and S. suis populations.
Fifty-eight CDSs in the S. pneumoniae ATCC 700669 annotation overlap with BOX elements. In 36 cases, this corresponds to the extreme 3' end of a gene, with the BOX repeat encoding the stop codon; in some cases, these correspond to well-characterised genes such as folE, mtlD, dnaJ and glgP. However, alignments with non-pneumococcal orthologues do not provide strong evidence for truncation of the encoded polypeptide in any case, especially when the relatively weak conservation of the extreme C terminal portion of proteins is taken into account.
A further 19 sequences, which appear to encode proteins on the basis of GC frameplot and correlation scores , with little or no functional annotation were found to be mostly, or wholly, encoded by BOX elements. Pneumococcal BOX repeats can extend to over 500 bp in length, and these larger elements tend to encode an open reading frame on both strands. Of the CDSs encoded mainly by BOX sequence, all but two (SPN23F00880 and SPN23F08320) were annotated on the opposite strand of genome to that on which the BOX elements are marked. None of the translated BOX-encoded CDSs exhibited significant similarity with any sequence in the public databases other than matches to hypothetical proteins annotated in mitis group streptococcal genomes.
However, in all three cases, there is also an ORF in the transcribed direction; rather than the start codon being in boxC and boxA encoding the stop codon, as predicted, boxC instead encodes the start codon and the stop codon lies beyond the BOX element. These expressed, BOX-encoded potential CDSs are indicated as dashed boxes in Figure 7a. Further RT-PCR confirmed that the RNA extended not just to the end of these BOX elements, but extended as far as the stop codon of these ORFs (Figure 7b). However, the proteins encoded by these ORFs also failed to significantly match any sequences other than hypothetical CDSs from mitis group streptococci and lacked good candidate Shine-Dalgarno sequences. Nevertheless, this confirmed that these 5' and 3' operon adducts, formed by BOX elements, have the potential to become nascent protein coding sequences.
The three families of small interspersed repeats found in the pneumococcal chromosome are found, albeit at a reduced frequency, in the closely related species, S. mitis, and very infrequently in other streptococci. These include the previously unidentified SPRITE repeat, which resembles a Rho-independent terminator element in its secondary structure. This is quite unlike the structures of the BOX and RUP elements, which are much more tightly folded and include their TIR hybridised to one another as parts of duplexes. A likely consequence of this form is the observed strong enrichment of this element close to the 3' ends of convergently transcribed CDSs, such that it does not disrupt normal gene expression patterns.
Even the naturally transformable oral streptococcus S. sanguinis, also part of the mitis group, lacks these elements. This implies that the repeats are unlikely to fulfil any of the possible important functions that might be ascribed to repeated sequences: for instance, chromosome packaging, aiding with replication or incorporation of horizontally transferred DNA. Furthermore, their distribution within the S. pneumoniae ATCC 700669 chromosome, resembling as it does the pattern of IS elements in being enriched between convergently transcribed CDSs, is suggestive of the main alternative explanation of their prevalence: that they are parasitic, non-autonomously mobile elements.
Based on their distribution between different streptococci, it appears that the repeats are likely to have been acquired subsequent to the divergence of the mitis group species. Two possible hypotheses may be advanced to explain the current distribution of repeats in the pneumococcus; one is that they may have been present in the last common ancestor of S. pneumoniae, and the position of some repeat insertions in this progenitor subsequently conserved amongst all pneumococcal strains. Alternatively, the repeats may have been acquired by S. pneumoniae and then spread horizontally through the population, resulting in the repeats being fixed at certain chromosomal loci over time. This second scenario is likely to be more sensitive to negative selection against the repeat insertions. In either case, a period of relatively rapid spread seems to have occurred in the population's past, which now seems to have abated. The proportion of repeats that are 'core' is similar to the proportion of 'core' CDSs in the pneumococcal pan-genome , and there are few insertions unique to any given chromosome that would indicate recent transposition events, contrasting with the distribution of IS elements between chromosomes.
The only other sequenced streptococcal species to have acquired BOX-type repeats is S. suis, which is also able to colonise the human nasopharynx, suggesting there may be a common source of these sets of elements. Although the S. suis BOX elements are present at a lower density in the chromosome, they are more diverse. It is difficult to assess how 'active' these elements are in this species, given the closely related nature of the currently sequenced S. suis genomes [25, 32], but in the current sample there is little evidence that they are more mobile than in S. pneumoniae. Hence in both species, these elements appear to be currently dormant.
One reason to suggest there may be selection against any mechanism that mobilises such elements is the disruption of CDSs by repeat insertion, which is evident in both S. pneumoniae and S. suis. However, there is also the potential for the formation of novel ORFs by BOX elements. Again, this is observed in both species; as well as the pneumococcal instances, there are two CDSs in the S. suis genomes that appear to be intact despite containing box modules (SSUSC84_0055 and 0899 in S. suis SC84) and three that are mostly, or entirely, encoded by BOX elements (SSUSC84_0048, 0112 and 0453 in S. suis SC84). The RNA-seq and RT-PCR data suggest that in some cases in S. pneumoniae such elements are transcribed, and have the potential to become nascent CDSs. Such instances appear to represent the consequences of three proposed properties of BOX elements: firstly, their mobility allowing them to insert into transcribed regions of the genome; secondly, the formation of an open reading frame on both strands of the element, and thirdly, their modular nature allowing them to expand to longer forms.
Whether the polypeptides they encode are actually expressed is not clear; it seems more likely that they are transcribed as untranslated regions. If so, they may influence the levels of expression of co-transcribed genes; those elements forming 3' adducts to operons are likely to form stem-loop structures that may impede the action of 3'→5' exonucleases, the primary RNA degradation pathway in bacteria, thereby stabilising the transcript. However, ERICs are capable of triggering endoribonucleolytic cleavage of transcripts, depending on the orientation of the element and the sequence of the operon, and CE can also trigger cleavage of mRNA. Hence the overall impact of a repeat insertion into an operon is difficult to predict, and is liable to change with the variation in the length of the BOX element and the context of the insertion site. Unfortunately, the sequence read coverage across operons with current RNA-seq techniques is too inconsistent to make any firm inferences about the impact of these BOX elements .
The simplest mechanism by which these repeats may affect transcription is through acting as terminators, especially given the resemblance of SPRITE sequences to such structures. Such a function has been previously been proposed to be performed by a BOX element . There is also a precedent for repeats having a similar impact in another nasopharyngeal commensal and pathogen: the 10 bp DNA uptake sequences (DUS) of N. meningitidis which, when found in close proximity to one another, tend to be inversely orientated, allowing them to form a stem loop structure predicted to act as a terminator . S. pneumoniae, although naturally transformable, is not known to have any DUS as N. meningitidis and H. influenzae do, and partial SPRITE sequences were not sufficiently abundant to suggest the element described here is a composite of DUS pairs. It seems likely, in fact, that the prevalence of the repeat families present in the pneumococcal chromosome exemplifies a potential disadvantage of the intrinsically competent lifestyle these three respiratory pathogens have adopted: the risk of acquiring genomic parasites that may cause considerable disruption whilst they remain mobile.
There are three families of small interspersed repeats in the S. pneumoniae chromosome: BOX, RUP and SPRITE. BOX-type repeats are also prevalent in the S. suis chromosome. The pneumococcal repeats appear to be parasitic, non-autonomous mobile elements that seem to have spread mainly during a burst of transposition subsequent to the divergence of S. pneumoniae and S. mitis from the other mitis group streptococci. BOX elements vary in size significantly, and are found to form expressed open reading frames that may constitute potential novel protein coding sequences or untranslated adducts to pre-existing operons.
The sequence of S. pneumoniae ATCC 700669  [EMBL accession code: FM211187] was searched for repeats longer than 50 bp using RepeatScout . For each of the three families identified, multiple sequence alignments were produced with MUSCLE , which were used to generate Hidden Markov Models (HMM) using HMMER1.8 (more recent versions of HMMER have not been optimised for searching long nucleic acid sequences for short motifs) . In order to define the modular nature of BOX elements, HMMs representing boxA, B and C sequences individually were produced using available sequence data [14, 37]. Sequences identified with these initial models were then aligned and used to produce the final HMMs used in this study; cutoff score thresholds were determined empirically from the distribution of scores for all hits throughout the genome. A composite BOX element was defined as two or more adjacent boxA, B or C modules. The same approach was used to generate HMMs for the repeats identified in S. suis. Thorough de novo searches for novel interspersed repeats in other species were not conducted.
In a number of cases where annotated repeat sequences overlapped, it was evident that one element had inserted into another. In such cases, for each repeat in the pair, a realignment of one repeat with the appropriate HMM was attempted using the concatenated flanking sequences of the other repeat, effectively excluding the sequence of the other element. If one of the elements had a greater bitscore when realigned in such a manner, it was reannotated as a split feature into which the other repeat had inserted.
HMM logos were produced using LogoMat-M . Secondary structure predictions were produced from a multiple alignment of 30 repeat sequence examples (a random sample in the base of RUP elements; only BOX elements with the canonical A1B1C1 structure were used) using RNAalifold . The HMMs for the S. pneumoniae and S. suis repeats, and a program to automate their annotation for viewing in Artemis , are made freely available from ftp://ftp.sanger.ac.uk/pub/pathogens/strep_repeats/. The annotation of repeat elements in complete S. pneumoniae, S, mitis and S. suis genomes is also available from this site. A revised annotation of S. pneumoniae ATCC 700669 has been submitted to the EMBL database.
Of the 14 available complete pneumococcal genomes in the EMBL database (Additional file 1), all except S. pneumoniae R6 (a laboratory derivative of S. pneumoniae D39, the sequence of which is also available in the database) were analysed. For each annotated repeat element, 250 bp of upstream and downstream flanking sequence were concatenated into a single 500 bp string. All pairwise sequence comparisons between strings corresponding to repeats of the same type were performed using BLASTN . The alignments with an E value smaller than 10-25 were then used to cluster the strings into groups, corresponding to orthologous repeat insertions, using OrthoMCL . The inflationary parameter used in clustering was set to 3, the smallest integral value that did not cluster a pair of insertions within the same genome together (i.e. identify 'paralogous' insertions). The IS elements in Figures 3 and 4 correspond to all the annotated IS element transposase CDSs in the S. pneumoniae ATCC 700669 genome; however, in order to identify orthologous IS elements in different pneumococcal genomes, a consistent annotation across the genomes was required. To automate this, such repeats were identified as BLASTN matches to defined elements in the IS database  with a nucleotide identity >95% and a length >90% of that of the reference sequence. Insertions of the same IS element type were then clustered as described for the small interspersed repeats, and the results for all IS elements subsequently combined to generate the data used in the graph.
Sequences of primers used in this study
Primer Sequence (5'-3')
Sequence reads were mapped as paired end data using BWA . The orientation of the second read in correctly mapped pairs was reversed using Samtools  before producing coverage plots, in order to maintain the directional fidelity of the data. In order to generate the plots in Figures 6 and 7, the 'XA' note in the alignment file was used to identify alternative mapping locations. All reads were used to generated the fully redundant plot; reads with alternative mapping loci only within the displayed region were maintained in the set used to generate the 'locally redundant' plot, whilst those that mapped equally well to sequences outside of the displayed region were excluded. This allows reads that come from a specific BOX element, but cannot be unambiguously assigned to a particular boxB module therein, to be retained within the 'locally redundant' plot. The RNA-seq data has been submitted to the European Nucleotide Archive with accession code ERR015608.
We thank Andrew Berry and Thomas Dan Otto for their help with data analysis. This work was funded by the Wellcome Trust.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.