We’ve seen that the genetic code channels cytosine deamination (one of the most common mutations) such that a rather random pool of amino acids is converted to a cluster of hydrophobic residues while at the same time is quite exceptional at buffering against deleterious mutations. But let’s consider the three termination codons that function as stop signals during the synthesis of proteins.
The three stop codons are as follows:
The first thing to note about all three stop codons is that none of them contain cytosine (C). In other words, these three are perfectly immune to cytosine deamination.
But remember that DNA is double-stranded. Is the complementary sequence on the other strand of DNA likewise immune to cytosine deamination?
The stop codons shown above are in the RNA format, so let’s put them in the DNA format:
Let’s now use the Watson-Crick base pair rules to write out the sequence on the complementary DNA:
As you can see, the first triplet has no C’s, so no worry there.
Oh, oh, the other two do have cytosine and are thus susceptible to deamination.
So what happens to the second when C is deaminated to T?
ATC —> ATT
Hey, that’s going to generate TAA on the other strand. In other words, a stop codon.
But what about the third one?
ACT —> ATT
The same thing!
In other words, these three stop codons are perfectly immune to the effects of cytosine deamination, meaning that one of the most common DNA mutations will not touch them. This, in turn, means that cytosine deamination will not create mutations that extend the C-terminal end of a protein. Yet more evidence of the exceptional nature of the universal genetic code.
Clearly, the choice of these three stop codons does not look arbitrary nor like a frozen accident. In fact, back in 2005, David Orren published a paper about this entitled, “The irresistible resistance of nonsense: Evolutionary adaptation of termination codons to minimize the effects of common DNA damage” (DNA Repair Vol: 4 Issue: 10, September 28, 2005, pp: 1208-1212). He noted:
Thus, termination codon sequences are resistant or well-adapted to TM resulting from not only guanine lesions but also cytosine deamination. Intriguingly, the predominance of the ochre codon as the termination signal for prokaryotic genes  might be explained by cytosine deamination events that ultimately result in convergence of ATC and ACT sequences to ATT in the template DNA.
I might also mention that yours truly was on the right track about this before Orren’s paper. Back on the Brainstorms forum, on November 25, 2002, I posted the following:
So let me offer a brief preview. Each of the three stop codons (UAG, UGA, and UAA) can be reached by a single cytosine deamination event. If we consider both strands, the CG:UA transitions can reach 7/9 positions. In contrast, the stop codons themselves are not nearly as prone to mutation through deamination (none contain cytosine). This asymmetry suggests that base substitutions are more likely to generate nonsense mutations than chain elongation mutations (not to mention that the codon pool to reach nonsense mutations is larger than the codon pool to reach chain elongation mutations). This makes sense from a front-loading perspective, as premature termination might unleash a subset of domains roughly analogous to Force’s DDC hypothesis. This could then set up selective pressure for recombination of domains. Chain-elongators, on the other hand, simply end up translating noise, which is probably less useful from an evolutionary perspective. This thus intersects with error correction and my thesis, where apoB is a good proof-of-concept example. Anyway, I’ll expand on this in a latter essay.