The mechanisms involved in translation termination are still not completely understood, partly because more interactions than those required for codon-anticodon recognition are involved. In particular, protein-RNA and protein-protein interactions are other potential targets of translation termination control. Since in vitro translation may not always accurately reflect the in vivo conditions, we chosed to perform such analyses in living mouse cells. This requires the use of a system that allows the detection of very low readthrough levels. Thus, we used a reporter system bearing: i) the highly sensitive luciferase reporter gene; ii) the powerful RSV LTR promoter [24
We used the sequence CAA UAG CAA UUA, that drives a high level of translational readthrough in mouse cells in culture [19
], and introduced targeted mutations in the sequences surrounding the stop codon. We found that single changes in this sequence were sufficient to decrease readthrough by up to 30 fold in vivo. We constructed a set of mutants where changes corresponded to synonymous codons. In this case, one can analyze separately at least some of the different parameters - amino acid identity, codon frequency, nucleotide identity - that may be involved in termination efficiency. We were also able to analyze nucleotide effects in different contexts, i.e. determine whether a given change would give the same effect in different sequence environments. This last point is crucial to assess the physiological significance of nucleotide usage found in stop signal regions.
We focused our analysis on two positions which have already been shown to be involved in termination or readthrough in eukaryotes, but which have not been studied in mammalian cells so far: the nucleotide preceeding the stop (-1) and the third nucleotide following the stop (+6) [15
Two sets of modifications were introduced in the CAA triplet preceeding the stop: CAN and GGN. We found that the same amino acid can drive very different readthrough efficiencies (5 fold differences between GGA and GGU), suggesting that the amino acid itself does not play a major role in readthrough efficiency. Availability of the tRNA preceeding the stop is another parameter that could modulate ribosome pausing [26
] and thus may affect readthrough. Since rare tRNAs generally correspond to rare codons [27
], we checked whether a bias in codon usage was correlated with readthrough efficiency. This seemed to be the case for some of the mutants presented here. For example, the rare CAA glutamine codon gave a higher readthrough than the frequent CAG codon. However, when we systematically examined codon frequency in parallel with readthrough efficiency, no correlation could be found (see Table ). Although we cannot rule out a marginal effect of codon frequency, it is unlikely that it represents the main determinant of readthrough efficiency.
Readthrough efficiency is clearly correlated with the nucleotide immediately preceeding the stop codon. Hence, an adenine in the -1 position is associated with a high readthrough while a pyrimidine is associated with a low readthrough. Taken together, these results show that there is a gradation in the observed effect on readthrough efficiency by which A>G>Y. The fact that a rule could be drawn suggests that relatively simple interactions are involved, possibly between the tRNA in the P site, the incomming potential suppressor tRNA and the message, as it has been shown in yeast [28
]. However, since the information on the Glycine-, Histidine- or Glutamine-isoacceptor tRNAs in mouse is not yet available (see M. Sprinzl, K.S. Vassilenko, J. Emmerich, F. Bauer "Compilation of tRNA sequences and sequences of tRNA genes". http://www.uni-bayreuth.de/departments/biochemie/trna/), this possibility cannot be explored further for the time being.
In a second series of experiments, two sets of mutations were introduced in the triplet following the stop codon: CAA was changed for CAN and for GGN. A drastic effect on readthrough, up to 100 fold, was observed. In this case however, no uniform rule emerged, the hierarchy being A>G>U=C in the CAN series, but C>U≥A≥G for the GGN series. These contrasted results imply that complex interactions are taking place downstream of the stop codon. Although the nature of these interactions cannot be deduced from our results, one possibility would be that mRNA structure is involved in this phenomenon, as it is in numerous translational events [29
]. Another obvious candidate potentially interacting with the stop context is the release factor eRF1 [30
]. If this is the case, one would predict to be able to obtain mutant forms of eRF1 showing a better recognition of poor codon context. The three dimensional structure of eRF1 has been published recently and could help to design mutants affected in stop codon context interaction [31
It is striking that the CAA UAG CAA UUA sequence directs a high readthrough in several very different eukaryotes. In plants, the readthrough efficiency reaches 4-5% [9
], a level similar to what is observed in mouse cells (2%). In the yeast S. cerevisiae, it is even more efficient, driving a readthrough of 15-25% in a [psi-] strain [18
] and up to 55% in a [PSI+] strain (Olivier Namy and Jean-Pierre Rousset, unpublished results). Furthermore, when our results are compared with those obtained in plant cells and yeast, a similar effect of the nucleotide context is observed [9
]. By contrast, this sequence is unable to drive a significant readthrough activity (4,5 10-4
) in E. coli (JPR, MC, unpublished results, see also [9