Genetic robustness at the codon level as a measure of selection

Abstract

Selection at the DNA level is usually detected by analysing substitution rates from multiple-species comparisons. It has been suggested that measures of genetic robustness at the codon level, which can be measured by analysing a single coding sequence, can be used to estimate selection, but the validity of these measures has been questioned. Here I test the efficiency of different measures of genetic robustness at the codon level to estimate the level of selection acting on a gene. I find that volatility and other measures of robustness are correlated with dN/dS, and that this is not simply the effect of a preference for translationally optimal codons. I discuss the possible implications and the possible problems of these methods based on single-sequence codon usage analysis.

Introduction

Natural selection at the DNA sequence level is usually assessed by comparing the number of synonymous substitutions per synonymous site (dS) and the number of nonsynonymous substitutions per nonsynonymous site (dN) from an alignment of at least two homologous protein-coding DNA sequences (Li, 1997, Nei and Kumar, 2000).

Two methods have been proposed, independently (“volatility” by Plotkin and Dushoff, 2003, Plotkin et al., 2004 and “degree of error minimization” by Archetti, 2004a, Archetti, 2004b), that measure genetic robustness of a coding sequence, and could be used to detect selection by analysing a single sequence, with no need of comparative analysis. These methods rely on the fact that synonymous codons, though neutral at the protein level, have different mutant codons with different fitness, and therefore different rates of back mutations (Archetti, 2004b, Plotkin et al., 2004) or because they affect differently the fitness of competing individuals (Archetti, 2006). The main difference between the two methods is that volatility (Plotkin et al., 2004) is a relative measure (relative to the other genes of the genome), whereas the degree of error minimization (Archetti, 2004a, Archetti, 2004b) is an absolute measure (depending only on the genetic code used).

Both methods rely on codon usage analysis of single coding sequences, with no need of comparative analysis. Since homologous sequences are not always available, a method to detect selection based on direct single-sequence analysis would be a useful tool in the study of molecular evolution. Both Archetti, 2004b, Plotkin et al., 2004 found a correlation between genetic robustness and substitution rates in a limited set of genes. Therefore the importance of these methods as a complement to dN/dS is worth investigating.

Volatility (Plotkin et al., 2004) has been criticised on different grounds, and virtually all the subsequent analyses (Hahn et al., 2005, Nielsen and Hubisz, 2005, Sharp, 2005; Chen et al., 2005, Dagan and Graur, 2005, Friedman and Hughes, 2005, Stoletzki et al., 2005, Zhang, 2005, Pillai et al., 2005) seem to suggest the rejection of volatility as a method to detect selection, whereas the degree of error minimization (Archetti, 2004b) has received less attention, primarily in studies concerning the evolution of the genetic code (Goodarzi et al., 2005, Marquez et al., 2005).

The critiques to volatility belong to three broad categories: (i) theoretical or conceptual uncertainties on the method, (ii) existence of confounding factors, and (iii) doubts on the validity of the results (correlation with dN/dS). Clearly, as Stoletzki et al. (2005) point out, critiques of the first and second kind are rather unimportant if the method does provide an alternative valid measure to dN/dS.

The critiques put forward by Hahn et al., 2005, Nielsen and Hubisz, 2005, Sharp, 2005, Chen et al., 2005, Zhang, 2005, Pillai et al., 2005 and also Dagan and Graur (2005), belong to the first and second category. Stoletzki et al., 2005, Friedman and Hughes, 2005, and in part also Dagan and Graur (2005), belong to the third. Stoletzki et al. (2005) show that a correlation between dN/dS and volatility actually exists but suggest that it is a byproduct of a correlation with preference for codons that optimize translation efficiency, while Dagan and Graur, 2005, Friedman and Hughes, 2005 question the very existence of a correlation.

I develop new measures of genetic robustness and I compare these and previous measures with dN/dS values derived from classical comparative analysis. I also re-analyse the data for which volatility has been questioned, to test whether the other measures described here can provide a valid alternative to dN/dS or whether they have the same problems as volatility. I then discuss whether the theoretical critiques that have been put forward about volatility can be applied to other measures as well.