Ambushing the Ambush Hypothesis: predicting and evaluating off-frame codon frequencies in prokaryotic genomes

Abstract

Background: In this paper, we address the evidence for the Ambush Hypothesis. Proposed by Seligmann and Pollock, this hypothesis posits that there exists a selection for off-frame stop codons (OSCs) to counteract the possible deleterious effects of translational frameshifts, including the waste of resources and potential cytotoxicity. Two main types of study have been used to support the hypothesis. Some studies analyzed codon usage and showed that codons with more potential to create OSCs seem to be favored over codons with lower potential; they used this finding to support the Ambush Hypothesis. Another study used 342 bacterial genomes to evaluate the hypothesis directly, finding significant excesses of OSCs in these genomes.

Results: We repeated both analyses with newer datasets and searched for other factors that could explain the observed trends. In the first case, the relative frequency of codons with the potential to create OSCs is directly correlated with the GC content of organisms, as stop codons are GC-poor. When evaluating the frequency of OSCs directly in 1,976 bacterial genomes we also detected a significant excess. However, when comparing the excess of OSCs with similarly obtained results for the frequency of out-of-frame sense codons, some sense codons have a more significant excess than stop codons.

Conclusions: Two avenues of study have been used to support the Ambush Hypothesis. Using the same methods as these previous studies, we demonstrate that the evidence in support of the Ambush Hypothesis does not hold up against more rigorous testing.

Figure 1

Example of OSC formation. There are six ways in which an arginine AGT codon at position i can form OSCs with codons at position i-1 or i+1. OSCs are given in upper case and red. X represents any base for the codons in positions i-1 and i+1.

Figure 2

GC content versus slope of regression between potential to form OSCs and usage of a codon. A linear regression gives the line y = -0.0251x + 0.013 with R² = 0.978, indicating that GC content explains 97.8% of the variation in the correlations. Organisms in which the correlation between a codon usage and its potential to form OSCs are positively or negatively significant (p < 0.05) are given in blue, while those with non-significant correlations are given in red. The Ambush Hypothesis predicts that organisms should have significant positive correlations between the usage of a given codon and its potential to form OSCs. There are no significant positive correlations in organisms with GC content higher than 46% and only three non-significant correlations in organisms with GC content lower than 37%.

Figure 3

Histograms of percent excess of selected out-of-frame codons in the 1,976 bacterial genomes analyzed for the 2–3 Markov model. The first graph shows the percent excess in the observed number of all stop codons out-of-frame. The other graphs show the percent excess for the 3 stop codons TAA, TAG and TGA; and for 3 TGN codons, TGC, TGG, and TGT. For the stop codons, only out-of-frame TGA codons have a systematic excess in the organism studied. All out-of-frame TGN codons have a systematic excess and are present out-of-frame in larger excesses than the set of all stop codons. The 5–3 Markov model gives similar results (data not shown).

Figure 4

Correlation between GC content of genomes and excess of certain off-frame codons calculated using the 2–3 Markov model. The top graph shows the correlation between the GC content and the excess of OSCs in the organisms studied; as predicted by the Ambush Hypothesis this graph has a significantly positive correlation. The graphs in the second row show the correlation between the GC content and the excess with which each stop appears out-of-frame; it should be noted that only TGA shows a positive correlation as predicted by the hypothesis, while TAA and TAG show negative correlations in opposition to the hypothesis’ prediction. The graphs in the third row show the same correlation but for the TGG, TGC and TGT codons; all three of these codons show significant positive correlations, with TGG and TGC showing even stronger and more significant correlations than TGA. The 5–3 Markov model gives similar results (data not shown).