The DB case: pattern matching evidence is not significant

The existence of a downstream box (DB) able to enhance translation efficiency in leaderless transcripts has been reported in several Escherichia coli genes. This putative element occurs downstream of the start codon, and it is complementary to the 16S rRNA sequence AGUACUUAGUGUUUC, with which it has been proposed to interact. According to the model of Sprengart and Porter (1997, Mol Microbiol 24: 19±28), this should compensate for the lack of stabilizing interaction of the rRNA with a weak or absent Shine±Dalgarno sequence. In particular, some indication about the existence of such elements has been reported: notably, its requirement for the transcription of lcI mRNA (Shean and Gottesman, 1992, Cell 70: 513±522); its role in the expression of the lysyl t-RNA synthetase gene lysU (Ito et al., 1993, Proc Natl Acad Sci USA 90: 302±306); and its role in the expression of the cold shock protein CspA (Mitta et al., 1997, Mol Microbiol 26: 321±335). Nevertheless, several other works have refuted the former analysis (Resch et al. ,1996, EMBO J 15: 4740± 4748; Tedin et al., 1999, Mol Microbiol 31: 67±77). Following the latter publication, two MicroCorrespondences were published regarding this subject. In one of these, BlaÈsi and colleagues described a large number of biochemical experiments and arguments against the DB hypothesis (see BlaÈsi et al., 1999, Mol Microbiol 33: 439± 441 and references therein), namely: (i) it is difficult to reconcile how the anti-DB region can be brought into alignment with the mRNA track as shown in the model proposed by Sprengart and Porter (1997, Mol Microbiol 24: 19±28); (ii) the interaction DB±anti-DB has not been supported by chemical footprinting or cross-linking studies; (iii) chemical protection studies have failed to reveal protection of the putative DB; (iv) the DB does not appear to increase the affinity of mRNA for 30S subunits; (v) the anti-DB stretch of rRNA is very well conserved among different species because of the stability of the helix in which it participates, but not because of its primary structure; (vi) finally, although the putative DB of Grampositive bacteria is rather different, the leaderless lcI mRNA was found to be expressed efficiently in vivo in Bacillus subtilis. Since the publication of these MicroCorrespondences, O'Connor et al. (Proc Natl Acad Sci USA 96: 8973±8978) have presented important results concerning the effect of mutating the anti-DB at the rrnB ribosome of E. coli. This mutation, although disrupting the DB±anti-DB interaction, did not modify gene expression significantly. Therefore, this work reinforces the idea that DB is irrelevant for gene expression. Resch et al. (1996, EMBO J 15: 4740±4748) have shown that the deletion of the downstream box of the lcI leaderless mRNA does not result in decreased expression levels of the transcript. To this, Etchegaray and Inouye (1999, Mol Microbiol 33: 438±441) responded that the deletion mentioned above created a new DB. This should invalidate the conclusions of Resch et al. (1996, EMBO J 15: 4740±4748). Although Etchegaray and Inouye (1999, Mol Microbiol 33: 438±441) admit that cross-linking studies failed to reveal DB±antiDB interactions, they argue that, as the signal exists, an interaction is also likely to exist. In fact, they maintain that their work reveals a correlation between the existence of the signal and gene expression. In this letter, we contend that such a correlation is not significant, as the DB signal itself is not statistically significant. There is no evidence for the evolutionary selection of the DB signal. In fact, we shall show that the constraints given to define a DB sequence are so poor that any deletion would most likely result in the production of another putative DB site. We have recently analysed factors affecting translation in B. subtilis, using its complete genome, and failed to identify a consensus sequence or a weight matrix for this signal (Rocha et al., 1999, Nucleic Acids Res 27: 3567± 3576). As the absence of the ribosomal protein S1 in this organism significantly increases the requirements for a good ribosome binding site (RBS) (Vellanoweth and Rabinowitz, 1992, Mol Microbiol 6: 1105±1114), we expected that this would also be the case for the DB. In order to check for the statistical significance of the DB pattern found in lysU, lysS (Ito et al., 1993, Proc Natl Acad Sci USA 90: 302±306) and tetR, lcI and P2V (Etchegaray and Inouye, 1999, Mol Microbiol 33: 438± 441), we devised the following simple in silico experiment. Let us consider one sequence fragment (e.g. the first 30 bases of lysU) in which the DB pattern is observed by the authors (Ito et al., 1993, Proc Natl Acad Sci USA 90: 302± 306) with k exact matches (e.g. k is 8 for lysU). Let us then generate a large number (5000) of random sequences with the same length and nucleotide composition, and let us count how many of them have a `best' hit of the DB pattern scoring at least k matches. The fraction of this