A significant proportion of non-immunoglobulin genes also have codon bias consistent with low mutability under somatic hypermutation

  

Figure 4.3: Histogram of the normalized ranks of the 140 non-immunoglobulin genes among their translationally neutral variants.

In the previous chapter, I showed that the codon usage of framework and complementarity-determining regions of immunoglobulin genes is biased, inducing lower mutability of a FR nucleotide compared to a CDR nucleotide. This can be inferred by comparing the mutability of the germline sequence with a set of variants with identical amino acid sequence, but unbiased codon usage. I will apply a similar technique to the set of non-immunoglobulin sequences. Briefly, for each sequence in the data set, I generate a set of 10⁴ variants as follows. I translate the nucleotide sequence into its corresponding amino acid sequence. Then, for each amino acid, I choose, with uniform probability, one of the codons that can encode it. I generate 10⁴ such variants for each non-immunoglobulin sequence in the initial data set. I calculate their average replacement mutability per nucleotide, and then determine the rank of the mutability of the germline sequence relative to its translationally neutral variants. Fig. shows the frequency distribution of the normalized ranks of the 140 genes. Approximately half of the genes in the set have a mutability that is in the low 5% compared to their variants with the same amino acid translation, and unbiased codon usage. As I mentioned, I ruled out any obvious genealogical relationship between these sequences. If their codon usage of the genes was unbiased, we would expect that the distribution of ranks would be uniform. The fact that it is not could indicate two things:

• That a codon usage bias is present in these genes for other reasons, and somatic hypermutation picks out this signal.
• That these genes share a codon usage bias consistent with low mutability because the somatic hypermutation mechanism borrows components from a mechanism that operates with a much larger scope in the genome than immunoglobulin genes.

I attempted to decide between these alternatives using the following test. Let us generate a different codon usage bias. Let C_ij be the set of codons, where i denotes the amino acid that the codon j is specifying. Let P_i(j) be a random permutation of the codons encoding amino acid i. Then to construct a sequence under this new codon usage bias, I replace each codon in the sequence C_ij by C_iPi(j). The set P_i(j), with i = 1...20 constitutes the new codon usage bias. For each codon bias thus constructed, I re-generate the set of 140 gene sequences, and calculate their replacement mutability under somatic mutation. Due to computational constraints, I only generated 100 different permutations of the codons.

As I showed previously (Fig. ), 73 of the 140 non-immunoglobulin sequences that I studied have codon usage that places them in the lowest 5% in mutability among their translationally invariant variants. In fact, 66 of the 140 sequences are in the lowest 1% among their neutral variants. I generate similar sets of translationally neutral variants for each sequence under each codon usage bias. I then determine how many of these codon usage biases give us as many significantly low mutable sequences. It turns out that if I set the significance level at the normalized rank of 1% among the neutral variants, none of the codon usage biases can produce as many low mutable sequences as the original codons usage bias.

This result allows me to conclude that it is not a random codon usage bias that the somatic hypermutation mechanism would pick out of these sequences. It is specifically the codon bias present in the set of germline genes that I used for this study. Thus, there is a significant correlation between the sequence specificity of the somatic hypermutation mechanism and the codon bias present in human genes. This may be due to:

• Somatic mutation being derived from a more general mutation mechanism that operates on the level of the whole genome.
• Somatic mutation having evolved to exploit a codon usage bias already present in the genome.

At the moment, I cannot decide between these two alternatives. On a simplicity argument, I tend to favor the first hypothesis. If the somatic mutator is derived from a more general-purpose mutation mechanism, we would expect that gene sequences, including immunoglobulin gene sequences, already had low mutability at the time when the mutator appeared. Only codon bias in CDRs would need to be evolved to arrive at the current mutability data. On the other hand, if a whole new mutation mechanism was evolved, it had to first adjust to the codon usage bias of the genome, and then the CDRs evolved a different codon usage. Whether the simplest path was indeed taken in the evolution of the immune system remains to be seen.