Several lines of evidence suggest that uncontrolled factors in the process of transcription and translation elongation may have direct and indirect effects on protein activity and yield. First, fusion of open reading frames to any of several well expressed genes (e.g., glutathione reductase GST) does not always produce full length protein. Second, statistical analysis of codon usage indicates that organisms differ in codon abundance. Altering the gene to eliminate rarely used codons can alter the expression of the gene in a particular host.

The observation here is that highly used codons are predominant in abundantly expressed proteins for a particular host and that these are optimal for gene expression. When the goal is to direct host-cell resources to producing a recombinant protein, human codon usage, when translated in a heterologous host, may create a scarcity of the cognate tRNA iso-acceptors and virtual starvation of the ribosome. Apart from some improvements, it is important to note that the expression changes seen are a response to so-called silent mutations that do not change the protein composition itself. Unfortunately, codon optimization alone does not predictably dictate high protein production; sometimes the expression actually gets worse.

Recently, another approach has been taken in an attempt to account for the variability. This approach stems from the observation that pairs of codons appear to explicitly encode signals that control the rate at which nascent proteins are elongated as the gene is translated along its full length.⁸ If translation elongation rates can differ for a given amino acid sequence based on the underlying mRNA sequence as translated by a given host, this might account for a large degree of the unpredictability seen in protein expression.

One early suggestion of the ability of simple sequences to control translation kinetics is related to the effect of codon context on nonsense codon suppression in E. coli, with certain codon pairs having much higher or lower suppression frequencies. This observation coincides with the observation of highly improbable bias in the abundance of codon pairs encoded in an organism's transcriptome (the sum of the sections of DNA in an organism's genome that are transcribed).⁷ The observed frequency of some codon pairs is many standard deviations higher than the expected abundance, and this over-representation is independent of the abundance of each individual codon.⁷ This phenomenon is specific and directional; changing the order of the codons in a pair eliminates the effect. This statistical aberration cannot be accounted for by the abundance of the codons, the amino acid pair associations, dinucleotide abundance, or other factors. This statistical anomaly is present in all organisms tested, but the actual codon pairs in the over-represented group are different for each organism.⁹

Careful in vivo and in vitro translation experiments reveal a counter-intuitive result: Over-represented codon pairs in a gene's open reading frame have the effect of slowing translation, and the greater the degree of over-representation, the greater the pause.⁸ What is the biological relevance of this slowing? One analogy is that the pauses act like "punctuation marks" — i.e., like commas in written language. There are only a few hundred statistically over-represented codon pairs in a given transcriptome (out of 3,721 possible non-terminating pair combinations) and a lower number of highly under-represented codon pairs. Moreover, the codon pairs that are significantly over-represented vary widely by organism, so that pausing signals are different in different organisms.⁹

Figure 1. Codon pair bias mediated translational pausing. Incompatible tRNA isoacceptors of over-represented codon pairs affect translational step times at the levels of tRNA binding, trans-peptidization, and perhaps translocation.