The Effects of Mutation and Natural Selection on Codon Bias in the Genes of Drosophila Richard

Codon bias varies widely among the loci of Drosophila melanogaster, and some of this diversity has been explained by variation in the strength of natural selection. A study of correlations between intron and coding region base composition shows that variation in mutation pattern also contributes to codon bias variation. This finding is corroborated by an analysis of variance (ANOVA), which shows a tendency for introns from the same gene to be similar in base composition. The strength of base composition correlations between introns and co on third positions is greater for genes with low codon bias than for genes with high codon bias. This pattern can be explained by an overwhelming effect of natural selection, relative to mutation, in highly biased loci. In particular, this correlation is absent when examining fourfold degenerate sites of highly biased genes. In general, it appears that selection acts more strongly in choosing among fourfold degenerate codons than among twofold degenerate codons. Although the results indicate regional variation in mutational bias, no evidence is found for large scale regions of compositional homogeneity. F OR many organisms, examination of nucleotide sequences of multiple genes has revealed varying levels of the extent to which synonymous codon usage departs from equanimity. In a simple conception, this variation in codon bias is caused by variation in two primary evolutionary forces: (1) mutation, which generates codon diversity, and (2) natural selection against “suboptimal” codons, which reduces codon diversity. The goal of this report is to describe our findings on the relative contributions of mutation and natural selection to the variation in codon bias among the genes of Drosophila melanogaster. In a variety of organisms, studies have shown that variation in codon bias is partly caused by variation among genes in the action of natural selection. In prokaryotes, such as Escherichia coli, there is a clear relationship between the extent of codon bias and gene expression level, with more highly expressed genes displaying greater codon bias (Gowand GAUTIER 1982) ; a similar pattern has been observed in Bacillus subtilis (SHIELDS and SHARP 1987). As would be expected if natural selection limits codon choice, there is also a negative correlation between codon bias and divergence at synonymous sites in E. coli and Salmonella typhimurium (SHARP and LI 1987). A tendency toward high codon bias in highly expressed genes has also been observed in eukaryotes, such as the yeast Saccharomyces cereuisiae (BENNETZEN and HALL 1982) and the slime mold Dictyostelium discoideum (SHARP and DEVINE 1989). It appears that natural selection for efficient translation is primarVirginia 24142. ’Present address: Department of Biology, Radford University, Radford, ily responsible for biased codon usage, with more highly expressed genes subject to stronger selection pressure than less highly expressed genes (BENNETZEN and HALL 1982; IKEMURA 1985). For D . melanogaster, the evidence for natural selection comes from multiple sources (SHIELDS et al. 1988; KLIMAN and HEY 1993). In the first major study on codon bias in this species, SHIELDS et al. (1988) presented several pieces of evidence supporting the case that natural selection acts on synonymous codon usage: (1) G + C content is lower in genes with low codon bias, consistent with the expected increase in the influence of the general mutational bias in D . melanogaster toward A and T on low biased genes ( i . e . , selection overcomes mutation pressure in highly biased genes) ; (2) there is a tendency toward high codon bias in loci homologous to highly expressed (and highly biased) loci in yeast and E. coli; (3) among members of a few multigene families, there are anecdotal reports that the more highly biased genes also generate their products in greater abundance; (4) for a small number of cases for which anticodon sequences and tRNA abundances are known, there is a preference in highly biased genes for codons translated by the most abundant iso-accepting tRNA; and ( 5 ) the divergence between D . melanogaster sequences from homologous D . pseudoobscura sequences at synonymous sites is higher in three low biased genes than in three highly biased genes, analogous to the observation in prokaryotes (SHARP and LI 1987). Similarly, in a study that compared 16 homologous sequences from D . melanogaster and D . pseudoobscura, MORIYAMA and G~JOBORI (1992) found a negative correlation between silent site Genetics 137: 1049-1056 (August, 1994) 1050 R. M. Kliman and J. Hey divergence and codon bias. While there may be natural selection for efficient translation in highly expressed genes in D. melanogaster, a recent report that conserved amino acid positions have higher codon bias supports a model in which natural selection also acts on codon usage to maximize the accuracy of translation (&HI 1994). A different tack was taken by KLIMAN and HEY (1993), who used codon bias in D. melanogaster to test a population genetics prediction that natural selection is less effective when crossing over is reduced. The central idea is that natural selection could not simultaneously choose from a population the best variants at multiple sites if those sites are tightly linked (HILL and ROBERTSON 1966; FELSENSTEIN 1974; CHARLESWORTH et al. 1993). We reasoned that if this prediction is true and if natural selection limits codon usage in the highly biased genes of D. melanogaster, then codon bias should be lower, on average, in genes located in sections of the genome that experience low levels of crossing over. Alarge and highly significant difference was found between genes in regions of low recombination and those in the remainder of the genome (KLIMAN and HEY 1993). Until recently, there has been no evidence from D . melanogaster that variation in mutation contributes to variation in codon bias. Several studies have taken the approach of measuring base composition in genomic regions thought to be subject to very low levels of natural selection. In the absence of selection, base composition is a function solely of mutation; however, the function is complex and it is generally not possible to estimate variation in mutation rates among the different nucleotides from information on base composition. SHIELDS et al. (1988) compared the G + C content of introns, thought to experience little natural selection on base composition, with the G + C content of silent sites in flanking exons. They did not find a statistically significant correlation [for 36, genes the correlation coefficient was 0.23 ( P = 0.09)]. Recent studies by MORIYAMA and HARTL (1993) and by CARULLI et al. (1993) also found no significant correlation between G + C content at third positions of fourfold degenerate codons and either introns or flanking DNA, respectively. CARULLI et al. (1993) did find evidence for compositional heterogeneity across the D. melanogaster genome. However, the basis for the heterogeneity could not be established because the proportions of coding sequence and other genetic structures known to influence base composition were not known for the various YAC clones used (CARULLI et al . 1993). The study by KLIMAN and HEY (1993), on the other hand, did show that variation in intron base composition is associated with variation in codon bias. It happens that all of the preferred codons of D. melanogaster (i. e . , those that appear more frequently in loci with very unequal codon usage) have a G or C in the third position, so that codon bias is highly correlated with G + C content at silent sites (SHIELDS et al. 1988; also, see RESULTS). In an analysis of 142 D. melanogaster sequences, intron G + C content correlated significantly with codon bias (KLIMAN and HEY 1993). The correlation between intron G + C content and codon bias suggests that the base compositions of introns and silent sites share a common influence, presumably related to regional mutation patterns. This report further explores the role of mutation, as well as the relative contributions of mutation and natural selection, in the determination of codon usage in D . melanogaster. MATERIALS AND METHODS Nucleotide sequences: We obtained the complete amino acid-coding sequences of 428 D. melanogaster loci from GenBank/EMBL (a table of loci used in this study is available from the authors upon request). Intron sequence was available for 155 of these loci; sequence from more than one intron was available for 79 of the loci. Codon usage/base composition: Codon usage bias was estimated using the Codon Adaptation Index (CAI; SHARP and LI 1986). This index requires data on codon usage in genes for which natural selection strongly favors a subset of “preferred” codons. We use the codon frequencies presented by SHARP et ul. (1992) for the most highly biased D. melanogaster loci, as determined by their position on the first principal axis produced by correspondence analysis on codon frequencies. For each amino acid, the most common codon used in this subset of genes was assigned a relative frequency of 1.0, and the relative frequency of rarer synonymous codons was scaled accordingly. CAI is calculated as the geometric mean of the relative frequency values corresponding to each codon in a locus. Thus, a gene using only preferred codons would obtain the maximum CAI value of 1.0; genes that use all synonymous codons equally would have a CAI value around 0.2 (the exact value depending on the amino acid composition of the locus); genes with lower values ofCAI tend to be biased toward codons that are rare in genes subject to strong selection for optimal codon usage. The CAI differs in principle from two other commonly used indices of codon bias, Chi/L (SHIELDS et al. 1988) and Effective Number of Codons (WRIGHT 1990). The latter indices measure deviation from equal codon usage, regardless of the direction. It should be noted that our choice of CAI rather than the other indices should not substantially affect our results. The three indices are all highly correlated with each other in D. melanoguster ( r > 0.92, P < 0.0001 for all pairs of indices using the 428 loci). In general, the extent of deviation from equal codon usage in this species reflects the degree of usage of a particular subset of codons. In our analyses, fourfold degenerate codons include the fourfold degenerate classes of arginine, leucine and serine. G + C content at third positions of twofold degenerate codons was calculated by summing the number of Cending codons and the number of G-ending codons (where synonymous choices are either C/T or A/G) , and dividing by the total number of twofold degenerate codons. Prior to statistical analyses, all proportion values ( e . g . , intron G + C content) were arcsineroot transformed.