Real-time RT-PCR was used to investigate the effect on mRNA stability of sequence changes in the 5′ UTR of ΔermC mRNA, consisting of either the native ermC 5′ UTR or the SP82 42 nt 5′ UTR. The ease with which mRNA half-life could be assayed using this system allowed us to measure the half-lives of many different mutants in a reasonable amount of time, with measurements made in triplicate (at least) and using only data from experiments in which linear regression of mRNA decay at time points after rifampicin addition gave an R2 value greater than 0.9. Interestingly, when measured by real-time RT-PCR, the base-line half-life of ΔermC mRNA was always around 6 min (this study), whereas ΔermC mRNA half-life measured by Northern blot analysis in multiple previous experiments was closer to 8 min (Sharp and Bechhofer, 2003). Many differences in the Experimental procedures might explain this discrepancy, but what was striking was the internal consistency of the observed half-lives in both types of experiments.
We can divide the present findings into five different elements of a 5′ UTR that may affect mRNA stability: (i) strength of SD sequence; (ii) distance from 5′-terminus to SD sequence; (iii) unpaired 5′ nucleotides; (iv) strength of 5′-terminal secondary structure; (v) distance of SD sequence from 5′-terminal secondary structure.
Strength of SD sequence
We found that the predicted SD sequence strength did not correlate with mRNA stability (Table 1, constructs 1–4). This is in agreement with the findings of Hambraeus et al. (2002), who concluded from a study of the stable aprE mRNA that the strength of a B. subtilis SD sequence is not predictive of mRNA stability. Our findings are also consistent with the results of Hambraeus et al. (2003), who examined decay of a large number of B. subtilis mRNAs using a microarray system and found no correlation between strength of SD−16S rRNA interaction and mRNA stability. On the other hand, it was found for Bacillus thuringiensis cryIIIA mRNA expressed in B. subtilis (Agaisse and Lereclus, 1996) and gsiB mRNA (Jürgen et al., 1998) that stability depended on a strong SD sequence. We note that in these cases, the SD sequence is located only four (cryIIIA) or nine (gsiB) nucleotides away from the 5′-end. Taken together with our earlier results regarding the effect of ternary complex formation on mRNA stability (Sharp and Bechhofer, 2003), we propose that SD sequence strength by itself will determine stability only when ribosome binding occurs in close proximity to the 5′-terminus. Ternary complex formation at such a site would sequester the 5′-end, and higher-affinity SD−16S rRNA interaction would enhance protection. When occurring further downstream, however, the strength of the SD sequence would not be a determining factor in mRNA stability.
Distance from 5′-terminus to SD sequence
The data in Table 2, regarding mRNA half-lives of a large set of ΔermC 5′ UTR insertions, are informative. We tested the hypothesis that distance between the 5′-end and the SD sequence could account for the difference in SP82 and ΔermC half-life. The basis for this hypothesis was our earlier study on the importance of ternary complex formation for mRNA stability (Sharp and Bechhofer, 2003). If ribosomes bound in a ternary complex could restrict access to the 5′-end, perhaps location of the SD sequence a particular distance from the 5′-end would allow, at once, optimal ribosome binding and optimal interaction of 5′-proximal nucleotides with the ribosome. This would provide maximal protection of the 5′-end. In fact, the data on constructs in Table 2 did not reveal a consistent relationship between distance from 5′-end to SD sequence and stability. Rather, the conclusion from these constructs was that 5′-terminal secondary structure above a minimal ΔG0, located less than 4 nts from the 5′-end, was the key element for stabilization. Thus, the most stable construct was ΔermC +8 (Table 2, construct 6; t1/2 = 19.6 min), which had a 5′-terminal secondary structure with a ΔG0 of −2.8 and no unpaired nucleotides. Constructs with weak 5′ secondary structure or 4 or more unpaired 5′ nucleotides had short half-lives similar to that of ΔermC. Apparent exceptions to this conclusion were ΔermC +6 and ΔermC +9 (Table 2, constructs 5 and 7), both of which had 5′-terminal secondary structures with ΔG0 of −3.1 and no unpaired nucleotides. The mRNAs of these constructs were significantly more stable than ΔermC (t1/2 = 10.8 and 9.1 min respectively, vs. 6.0 min for ΔermC) but not nearly as stable as ΔermC +8 (t1/2 = 19.6 min). Another exception was SP82 Ver. 1 (Fig. 2F), which had a stronger predicted 5′-terminal structure than SP82 itself, but still had only a 13.7 min half-life. In the case of ΔermC +9, we found that mfold predictions gave an alternative structure, with a slightly lower ΔG0 of −2.7 (not shown), and one might argue that the possibility of forming closely related alternative structures may be destabilizing. However, this would not explain the relative instability of ΔermC +6 and SP82 Ver. 1. A more likely explanation is that we cannot rely completely on predicted secondary structures. Such predictions are based on RNA sequence alone, and this may not be accurate for an mRNA in vivo with ribosomes binding in close proximity. The smaller the number of base–pairing interactions, the less reliable such predictions would be. However, the good correlation otherwise between predicted 5′-terminal structure and mRNA stability engenders confidence that such predictions are good approximations in general. The results of in vivo methylation studies (Fig. 5), which largely confirmed the RNA secondary structure predictions, also suggests that, in many cases, the computer-predicted structures do exist in vivo.
Unpaired 5′ nucleotides
The half-lives of constructs shown in Table 2 that have 5′ unpaired nucleotides suggested that the protective secondary structure needed to be less than 4 nts from the 5′-terminus. However, those constructs (e.g. 8–10) had different predicted structures, so it was difficult to make a conclusion about the length of unpaired nucleotides that could affect stability. To examine this more systematically, 3, 5 or 7 nts were added at the SP82 5′-end, and these were predicted to remain as single-stranded tails upstream of the identical secondary structure. The data (Table 3, constructs 6, 7 and 8) showed that the presence of three or five unpaired nucleotides gives a half-life not significantly different from that of SP82 itself, but the presence of seven unpaired nucleotides resulted in almost a complete loss of stabilization (t1/2 = 18.6 min for SP82 and 8.7 min for construct 8). This is consistent with the requirement for access to the 5′-end to initiate mRNA decay, which would be denied by 5′-terminal secondary structure but would be provided by upstream 5′ unpaired nucleotides.
These experiments speak to the proposed mechanism of a putative 5′-dependent endonuclease in B. subtilis. Based on a number of differences between the nature of 5′ stabilizers in E. coli and B. subtilis, Condon (2003) has suggested that 5′-end-dependent decay in B. subtilis requires 5′-end binding followed by tracking to an internal cleavage site, unlike the case in E. coli where the 5′-binding endonuclease (RNase E) may loop to an internal site. Our results suggest that providing an accessible 5′-end will result in a loss of stability, even though a strong secondary structure is present downstream. Either the 5′ binding activity can loop around this structure or its processivity in tracking from the 5′-end is not disturbed by RNA secondary structure.
Strength of 5′-terminal secondary structure
Once the results in Table 2 had demonstrated the importance of secondary structure at the 5′-end, the effect of the predicted strength of this structure on mRNA stability was analysed. The predicted strength of 5′-terminal structures of the stable SP82 and ΔermC +8 mRNAs were decreased by insertion of nucleotides, and these constructs gave mRNAs with much reduced half-lives (Fig. 3), as predicted. Interestingly, a single nucleotide addition to the 5′-terminal sequence of SP82, which resulted in a large increase in predicted stability of secondary structure, gave no additional increase in mRNA half-life (Table 3, construct 3). This latter result suggests that formation of a stable, 5′-terminal structure is sufficient to confer stability by blocking 5′-end-dependent decay initiation. Further increases in stability of such structures above a certain minimum would not result in increased mRNA stability. This was indeed the observation in comparing the half-lives of SP82 (t1/2 = 18.6 min) and ΔermC +8 (t1/2 = 19.6 min). Based on the significant difference in predicted free energy of formation for the respective stem–loop structures (−4.7 for SP82 and −2.8 for ΔermC +8), one might have expected that SP82 would be more stable than ΔermC +8. We suggest that the terminal secondary structure of ΔermC +8 is above the threshold required to block 5′-access. A similar conclusion was reached by Arnold et al., 1998) in their study of E. coli ompA mRNA.
The observation that abolishing ternary complex formation (by changing the AUG initiation codon to ACG; Table 3, construct 2) had a major effect on stability suggests that 5′ stem structure is not sufficient to confer stability. Perhaps both 5′-terminal structure and 5′-proximal ribosome binding contribute to protection against 5′-end-dependent initiation of decay, and the presence of both acts synergistically.
Distance of SD sequence from 5′-terminal secondary structure
Increasing the distance between the SP82 5′-terminal secondary structure and the SD sequence had a statistically insignificant effect on SP82 half-life (Table 3, constructs 4 and 5). This is consistent with the idea that the major determinant for mRNA stability is a blocked 5′-end. It would be of interest to determine whether locating the SD sequence even further from the 5′-end would cause a decrease in half-life. This might be attributed to endonucleolytic cleavage between the 5′-terminal secondary structure and the SD sequence. Insertion of only 12 or 24 nts, as we have done, may not have provided a sufficiently large target for such cleavages.
The experiments shown in Fig. 4 demonstrated that even a highly stable 5′-terminal structure could not confer stability without adequate distance between the 3′-end of the structure and the SD sequence. We propose that the binding of ribosomes at an SD sequence that is located too close to the 5′-terminal structure will result in perturbation of the structure. Based on crystallography studies (Yusupova et al., 2001), 15 nts upstream of the initiation codon are protected by the ribosome bound in a ternary complex. In the case of ΔermC +14/5′T or ΔermC +14/PS (Fig. 4B and C), there are 18 nts between the downstream base of the stem–loop structure and the initiation codon. We suggest that this distance is short enough to allow ribosome binding to perturb formation of the protective 5′-terminal structure. Once the distance is increased to 25 nts (ΔermC +14/7A; Fig. 4D), ribosome binding no longer affects secondary structure formation, resulting in a very stable mRNA. The reverse could also be true, i.e. the presence of the secondary structure could affect translation, which is also required for stability. It has been shown that secondary structure located immediately upstream of an SD sequence reduces translational efficiency (Coleman et al., 1985).
A recent article (Hambraeus et al., 2003) reported the use of microarray analysis to examine genome-wide mRNA stability in B. subtilis. Fifty-two mRNAs were identified that had half-lives ≥ 15 min, and seven of these (abnA, acoA, aprE, cotG, gsiB, sunA and yhxD) were confirmed by Northern blot or primer extension analysis. It was stated that no common features were found that could explain the stability of these mRNAs. Based on our conclusions from the current study, however, we can explain the stability of five of these seven. Analysis of the +1-to-SD sequences by mfold predicted the presence of 5′-terminal secondary structure for three of the seven stable mRNAs (sunA, cotG and aprE; Fig. 6). acoA mRNA lacks any predicted 5′-terminal structure but its SD sequence is only 13 nts from the 5′-end. Thus, it may be similar to the case of gsiB, which has a +1-to-SD distance of 9 nts, as discussed above. For the final two cases, abnA has a 116 nt 5′ UTR, which may assume structures that cannot be predicted or may be bound by protein. The yhxD gene function is unknown, and its transcription start site has not been mapped.
Our results that identified specific 5′ elements that can confer stability were based on ΔermC mRNA, which is a relatively small mRNA (254 nts). While we would like to believe that the elements found in this study to confer stability are applicable to B. subtilis mRNAs generally, it is possible that similar 5′ elements will not confer stability to longer mRNAs that have increased target size for an alternative decay pathway. In this respect, however, we note that the 42 nt SP82 5′ UTR could confer stability when fused to lacZ mRNA, the fusion mRNA being at least 10-fold longer than ΔermC (Hue et al., 1995). On the other hand, we should also note that, based on our observations from the 5′ UTRs described here, we have constructed mRNAs with other 5′-terminal stem–loop structures and measured their stability. In one particular case (data not shown), a ΔermC mRNA with a strong 5′-terminal stem structure (ΔG0 = −9.5) had a surprisingly short half-life of only 5.8 min. The sequence of nucleotides in the stem and loop of this construct, as well as the nucleotides between the stem–loop structure and the SD sequence, differed from those of the stable constructs found in this study. Thus, there must be other factors, of which we are not aware, that can affect initiation of mRNA decay.
Finally, we suppose that a 5′-end-dependent pathway of mRNA decay would be the result of an RNase E-like activity. However, as stated above, there is no RNase E sequence homologue in the B. subtilis genome. We and others have tried unsuccessfully to complement a temperature-sensitive E. coli RNase E mutant, using shotgun-cloned B. subtilis chromosomal DNA (D.H. Bechhofer, unpubl. results). Elucidation of the basis for 5′-end-dependent mRNA decay in B. subtilis remains a high priority.