Experimental reduction of codon bias in the Drosophila alcohol dehydrogenase gene results in decreased ethanol tolerance of adult flies

Authors


David B. Carlini, Department of Biology, American University, 4400 Massachusetts Avenue, NW, Washington, D.C. 20016, USA.
Tel.: 202 885 2184; fax: 202 885 2182; e-mail: carlini@american.edu

Abstract

The ethanol tolerance of adult transgenic flies of Drosophila containing between zero and ten unpreferred synonymous mutations that reduced codon bias in the alcohol dehydrogenase (Adh) gene was assayed. As the amino acid sequences of the ADH protein were identical in the four genotypes assayed, differences in ethanol tolerance were due to differences in the abundance of ADH protein, presumably driven by the effects of codon bias on translational efficiency. The ethanol tolerance of genotypes decreased with the number of unpreferred synonymous mutations, and a positive correlation between ADH protein abundance and ethanol tolerance was observed. This work confirms that the fitness effects of unpreferred synonymous mutations that reduce codon bias in a highly expressed gene are experimentally measurable in Drosophila melanogaster.

Introduction

There are two types of point mutations in protein-coding genes: nonsynonymous and synonymous mutations. The functional effects of nonsynonymous mutations have been studied extensively, because of their effects on the primary structures of encoded polypeptides. Nonsynonymous mutations result in changes in the amino acid sequence, which can alter important properties such as the activity, specificity, or stability of the encoded protein (Kimura, 1983; Li, 1997). Synonymous point mutations, on the contrary, do not alter the encoded amino acid sequence, and have therefore received comparatively little attention. Although synonymous codons encode the same amino acid, they are not used with equal frequency, a phenomenon known as codon bias (Ikemura, 1981). Codon bias is apparent in DNA sequences from a wide array of organisms including unicellular prokaryotes and eukaryotes, and multicellular eukaryotes (Akashi, 2001).

Although synonymous mutations do not directly alter the primary structure of the encoded protein, they can affect the accuracy and efficiency of protein synthesis (Akashi, 2001 and references therein); consequently, such variation can also affect fitness and is therefore subject to natural selection. Several lines of evidence suggest that codon bias is due to natural selection for translational accuracy and/or efficiency at synonymous sites. Optimal codons, synonymous codons used more frequently in highly expressed genes than in weakly expressed genes, tend to match the most abundant species of isoaccepting tRNA (Ikemura, 1981, 1982; Grosjean & Fiers, 1982; Moriyama & Powell, 1997). This observation, coupled with the fact that codon bias is most extreme in highly expressed genes (Gouy & Gautier, 1982; Sharp & Li, 1986; Duret & Mouchiroud, 1999), supports the hypothesis that codon bias is due to natural selection for translational accuracy and efficiency (Bulmer, 1991). Codon bias maximizes translational accuracy by reducing the number of amino acid misincorporations into the nascent polypeptide. In Escherichia coli, optimal codons are translated approximately 10 times more accurately than suboptimal codons (Precup & Parker, 1987). In Drosophila, codon bias is significantly higher in functionally constrained codons and proteins (Akashi, 1994). Codon bias also maximizes the speed of translation by minimizing the time required to attach the correct aminoacyl-tRNA during protein synthesis. In general, optimal codons exhibit perfect Watson–Crick base pairing with the tRNA anticodon (Ikemura, 1981, 1982; Bennetzen & Hall, 1982; Grosjean & Fiers, 1982). Ribosomes spend less idle time when encountering optimal codons, so that the use of optimal codons maximizes fitness by increasing the cellular concentration of free ribosomes (Bulmer, 1991). The relationship between codon bias and rate of gene expression has been experimentally confirmed in unicellular organisms such as E. coli, where optimal codons are translated three to six times faster than suboptimal codons (Robinson et al., 1984; Varenne et al., 1984; Sörensen et al., 1989; Andersson & Kurland, 1990).

In multicellular organisms the multitude of tissue types and developmental stages make it difficult to generalize which species of tRNA is most abundant. Furthermore, cell-specific patterns of gene expression make it difficult to relate codon bias and gene expression. The importance of codon bias in enhancing mRNA translation rates and fidelity in multicellular organisms has been established largely through indirect evidence, such as population genetic analyses of molecular sequence data (Bulmer, 1991; Akashi, 1994, 1996; Hartl et al., 1994; Duret & Mouchiroud, 1999; McVean & Vieira, 2001). Population genetics theory predicts that selection for codon bias is relatively weak in Drosophila melanogaster, with estimates of selection coefficients on the order of |s| = 10−6, approaching the limits of natural selection in finite populations (Akashi, 1995; McVean & Vieira, 2001). The effects of synonymous mutations with such minor selective differences were therefore assumed to be difficult to measure in vivo in Drosophila, and no attempts were made to do so until recently.

In conflict with predictions based on population genetic theory, experimental manipulation of codon bias in D. melanogaster demonstrated that the reduction in ADH expression associated with unpreferred synonymous mutations is experimentally detectable (Carlini & Stephan, 2003). Here, the phenotypic effects of reduced codon bias in the Drosophila alcohol dehydrogenase (Adh) gene are further examined through ethanol tolerance assays. The purpose of the present study is to determine if these differences in protein expression, a consequence of unpreferred synonymous mutations, affect the fitness of adult flies in the laboratory. The ethanol tolerance of adult flies, rather than that of larvae, was assayed because previous work has shown that there is no relationship between the amount of ADH protein and the ethanol tolerance of larvae, but that there is a positive correlation between the two variates in adult flies (Schmitt et al., 1986; Geer et al., 1993).

Because synonymous changes do not affect the amino acid sequence of the ADH protein, but rather ADH abundance, the question remains as to whether the differences in ADH abundance actually affect fly fitness (i.e. ethanol tolerance). It is possible that the level of ADH expression in wild-type flies is more than the minimum level required to metabolize ethanol at ecologically relevant concentrations. In this case, a slight decrease in the level of ADH protein expression would not affect the fitness of adult flies, consistent with the idea that the fitness effects of alternative codon usage are very weak. On the contrary, if differences in ADH abundance affect fitness of adult flies, this would support the results of Carlini & Stephan (2003), who used saturation theory to estimate a rather large value of |s| ≈ 4 × 10−5 for the synonymous substitutions in the leucine codons of the Drosophila Adh gene. Because the estimate of the strength of natural selection at synonymous leucine codons obtained in that study was over one order of magnitude greater than previous estimates, the fitness effects of such substitutions would be expected to be much larger and therefore experimentally tractable under such a scenario. This work confirms their results: the fitness effects of unpreferred synonymous mutations that reduce codon bias in a highly expressed gene are experimentally measurable in D. melanogaster.

Methods

Construction of transformant lines

Methods for site-directed mutagenesis, plasmid construction, P-element mediated germline transformation, and construction of transformant lines are described in detail in Carlini & Stephan (2003). The three mutant classes of transformant lines used in the present study, designated 1 Leu, 6 Leu, and 10 Leu lines, along with control lines containing a wild-type Adh allele, were constructed previously (Carlini & Stephan, 2003). These lines differed in the number of synonymous substitutions incorporated at leucine codons relative to the wild-type Wa-F Adh allele (Kreitman, 1983). The 1 Leu lines contained a single synonymous substitution at codon 16 (CTG to CTA). The 6 Leu lines contained a total of six synonymous substitutions at codons 5 (TTG to CTA), 16 (CTG to CTA), 21 (CTG to CTA), 27 (CTG to CTA), 28 (CTC to CTA), and 32 (CTG to CTA). The 10 Leu lines contained a total of 10 synonymous substitutions at the same positions as the 6 Leu lines (with the exception of codon 5), along with an additional five substitutions at codons 35 (CTG to CTA), 38 (CTC to CTA), 50 (CTG to CTA), 76 (CTG to CTA), and 77 (CTG to CTA). The control transformant lines contained an unaltered copy of the Wa-F Adh allele (Parsch et al., 1997). All transgenes (control included) were introduced into Adh flies (Adhfn6, splicing defect), such that only the Adh gene introduced via germline transformation produced a functional gene product (Benyajati et al., 1982). To avoid differences in ADH expression because of differences in copy number and dosage compensation mechanisms (Laurie-Ahlberg & Stam, 1987), only lines containing single insertions on autosomes, as identified through Southern blotting and mapping crosses, were used in the ethanol tolerance assays.

Ethanol tolerance

Adult ethanol tolerance was assayed for each of seven randomly selected independent insertion lines within each of the four genotypes according to the methods described by Merçot et al. (1994). Flies heterozygous for the Adh insertion were obtained by crossing transformed males to the y w; Adhfn6 stock. For each cross, 20 male and 20 female flies were collected over the course of 3 days and aged separately until they were 4–6-days old, at which point they were transferred to the ethanol tolerance chambers. Each ethanol tolerance chamber consisted of a 50 mL airtight conical plastic vial containing a compressed cotton ball at the bottom of the vial to absorb the ethanol/sucrose solution. All chambers contained 2.5 mL of a solution of a given concentration of ethanol and 3% sucrose to prevent starvation and mortality. Ethanol tolerance was assayed at the following six concentrations: 2.5, 5, 6.25, 7.5, 8.75 and 10%. At each concentration, four replicate sets of 20 flies of each sex for each line were assayed. Mortality was calculated as the number of dead flies after 24 h of treatment at a temperature of 20 °C. Differences in mortality between the sexes were negligible, so the data for females and males were pooled to estimate the 50% lethal dose (LD50), a commonly used toxicity metric. For each line, the LD50 was estimated by performing a linear regression of mortality on ethanol concentration.

Results

Adult mortality

The average mortality vs. ethanol concentration varied across genotypes (Fig. 1). For each genotype, average mortality at 2.5% ethanol was very low and mortality increased linearly with ethanol concentration. The degree of difference among genotypes in sensitivity to ethanol varied at different concentrations of ethanol. At 2.5%, differences among genotypes were negligible; at 5%, the mortality of the 6 Leu and 10 Leu lines was approximately twice that of the Wa-F and 1 Leu genotypes. The response at 6.25% was unusual in that the control lines were more sensitive than the 6 Leu or 1 Leu lines, although the 10 Leu lines remained the most sensitive. At 7.5–10%, the relative sensitivity to ethanol among genotypes was consistent with expectations based on the genotypic ADH activities, with a minor exception at 10% where the 1 Leu lines were more sensitive than all other genotypes.

Figure 1.

Average percentage mortality of each genotype at each of six ethanol concentrations. 1 Leu = lines containing a single unpreferred synonymous substitution. 6 Leu = lines containing a total of six unpreferred synonymous substitutions. 10 Leu = lines containing a total of 10 unpreferred synonymous substitutions. Wa-F = control transformant lines containing an unaltered copy of the Wa-F Adh allele (Parsch et al., 1997). For each genotype, average mortality was calculated as the mean mortality of seven independently transformed lines within each genotype at a particular ethanol concentration. Error bars indicate ±1 SE. Average mortality increased with increasing ethanol concentrations, and the differences in the average mortality among genotypes were significant (P < 0.01; Table 1), as were the differences in average mortality because of ethanol concentration (P < 0.0001; Table 1).

A two-way Model I anova on the average mortality of each line with genotype and ethanol concentration as fixed treatment effects indicated that differences in the average mortality among genotypes were significant (P < 0.01; Table 1), as were the differences in average mortality because of ethanol concentration (P < 0.0001; Table 1). The genotype × % ethanol interaction was not significant (P = 0.54; Table 1).

Table 1.  Model I analysis of variance as a function of genotype and ethanol concentration.
Sourced.f.SSMSFP
Genotype34816.651605.554.10.0079
% Ethanol5106 589.7621 317.9554.45<0.0001
Genotype × % Ethanol155408.19360.550.920.54
Error14456381.4391.54  

Ethanol tolerance comparisons

Differences in the slopes of the least squares regression were not statistically significant because of substantial among-line variation in mortality within each genotype (Fig. 2a–d). For each line, a separate regression was calculated based on the four replicate measurements of mortality at each of six ethanol concentrations. The average LD50 within each genotype, which was calculated as the average of the seven LD50s, was obtained from the independent regressions (Fig. 3). The differences in LD50 among genotypes were highly significant (P < 0.01, Kruskal–Wallis test). Post hoc comparisons between the control vs. 10 Leu and 1 Leu vs. 10 Leu were significant (P < 0.01, Tukey's method); all other paired comparisons between genotypes were not statistically significant.

Figure 2.

Linear regression of adult mortality vs. percentage ethanol for each of seven lines within each of the four genotypes. 1 Leu = lines containing a single unpreferred synonymous substitution. 6 Leu = lines containing a total of six unpreferred synonymous substitutions. 10 Leu = lines containing a total of ten unpreferred synonymous substitutions. Wa-F = control transformant lines containing an unaltered copy of the Wa-F Adh allele (Parsch et al., 1997). Mortality increased linearly as a function of ethanol concentration for all genotypes. Higher-order regression analysis did not improve the fit to the data. Differences in the slopes of the regression were not significant.

Figure 3.

Average LD50 of each genotype. For each line, a separate regression was calculated based on the four replicate measurements of mortality at each of six ethanol concentrations. The average LD50 within each genotype, calculated as the average of the seven LD50s, was obtained from the independent regressions. The differences in LD50 among genotypes were highly significant (P < 0.01, Kruskal–Wallis test). Post hoc comparisons between the control vs. 10 Leu and 1 Leu vs. 10 Leu were significant (P < 0.01, Tukey's method); all other paired comparisons between genotypes were not statistically significant. Error bars indicate 95% confidence intervals.

Ethanol tolerance and ADH activity

The average ADH activity of each line had been measured in a previous study (Carlini & Stephan, 2003). Differences in the ADH activity among lines and genotypes are due to differences in the abundance of the protein. To determine if differences in protein abundance affected ethanol tolerance, the average ADH activity of each of the seven lines was plotted as a function of the LD50 of each line (Fig. 4). Although the correlation is relatively weak (R2 = 0.14), there is a positive relationship between ADH activity and ethanol tolerance (F1,26 = 4.09, P = 0.05). The 10 Leu lines expressed the lowest levels of ADH protein and were in fact the most sensitive to ethanol, whereas the Wa-F controls and 1 Leu lines expressed the highest levels of ADH protein and were the most tolerant. The 6 Leu lines were intermediate in both respects.

Figure 4.

Average ADH activity of each of the seven lines was plotted as a function of the LD50 of each line. The ADH activity was obtained from Carlini & Stephan (2003), and is expressed in standard units (μmoles NAD+ reduced per minute per milligram of total protein × 100) Genotypes differed only at synonymous sites, so differences in ADH activity among genotypes were because of differences in ADH protein abundance, not catalytic activity. Although the correlation is relatively weak (R2 = 0.14), there is a positive relationship between ADH abundance and ethanol tolerance tolerance (F1,26 = 4.09, P = 0.05). The 10 Leu lines expressed the lowest levels of ADH protein and were in fact the most sensitive to ethanol, whereas the Wa-F controls and 1 Leu lines expressed the highest levels of ADH protein and were the most tolerant.

Discussion

Population genetic analyses of molecular sequence data have established the importance of codon bias in enhancing mRNA translation rates and fidelity in multicellular organisms (Bulmer, 1991; Akashi, 1994, 1996; Hartl et al., 1994; Duret & Mouchiroud, 1999; McVean & Vieira, 2001). The joint effects of genetic drift and natural selection determine the fate of synonymous mutations, parameterized as Nes, the product of the effective population size (Ne) and the selection coefficient (s) for or against the mutation in question (Kimura, 1983). In this model, directional selection favours optimal codons (Nes > 0), whereas purifying selection removes sub-optimal codons (Nes < 0). In comparison with levels of selection for most adaptive nonsynonymous substitutions (10−4 < s < 10−3), current population genetics theory predicts that selection for codon bias is generally thought to be rather weak (s ≈ 10−6), such that optimal codons persist primarily as a consequence of mutation pressure and random genetic drift (Li, 1987; Bulmer, 1991; Akashi, 1995, 1996; McVean & Vieira, 2001).

However, experimental manipulation of codon bias in Drosophila resulted in an average 2.13% reduction in ADH expression per unpreferred synonymous substitution, a finding not consistent with population genetics theory (Carlini & Stephan, 2003). This work demonstrated that synonymous substitutions affect the abundance of the ADH protein, but did not address whether the decrease in ADH abundance affected fly fitness. The purpose of the present study was to determine if the differences in protein expression because of reduced codon bias affect the fitness (i.e. ethanol tolerance) of adult flies in the laboratory. Perhaps the level of ADH expression in wild-type flies is more than the minimum level required to metabolize ethanol at ecologically relevant concentrations. In this case, a slight decrease in the level of ADH protein expression would not alter the ethanol tolerance of adult flies. The data from this study clearly demonstrate that this is not the case: flies with 10 unpreferred synonymous mutations were more sensitive to ethanol than flies with <10 such mutations. This pattern underscores the significance of the ADH expression data (Carlini & Stephan, 2003) because it demonstrates that the differences in levels of ADH expression, attributable to differences in synonymous codon usage, are indeed evolutionarily significant.

Drosophila melanogaster is a fruit breeding drosophilid species (females oviposit in decaying fruit), and it has been shown that fruit breeding species exhibit a higher tolerance to ethanol than nonfruit breeders (David & Van Herrewege, 1983; Merçot et al., 1994). This is not surprising, given the fact that concentration of ethanol in the natural habitats of fruit breeders ranges up to 4% and often exceeds 10% in artificial habitats such as wineries (Briscoe et al., 1975; McKenzie & McKechnie, 1979; McKechnie & Morgan, 1982). There is substantial heritable variation in the ethanol tolerance among D. melanogaster populations, with the most tolerant populations coming from ethanol-rich and high latitude habitats (Kamping & van Delden, 1978; David & Van Herrewege, 1983; Vouidibio et al., 1989). Although ethanol tolerance is a complex genetic trait, the Adh gene is clearly a major locus affecting tolerance: approximately 90% of ethanol is degraded by the ADH pathway (Geer et al., 1985; Heinstra et al., 1987; Heinstra & Geer, 1991). There are two major classes of Adh alleles that account for much of this variation in ethanol tolerance: Adh-F and Adh-S. In general, Adh-F flies have higher levels of ADH activity and higher ethanol tolerance than flies of the Adh-S genotype (Vouidibio et al., 1989; Geer et al., 1990; Merçot et al., 1994), although there is also variation in tolerance within each of the genotypic classes (Merçot et al., 1994). The nonsynonymous difference between the Adh-F and Adh-S genotypes results in catalytic differences in the corresponding ADH proteins and may affect the ethanol tolerance of larval and adult flies (Geer et al., 1990).

The Adh-F and Adh-S genotypes also differ in the amount of ADH protein expressed, with Adh-F homozygotes expressing approximately 1.5 times more ADH protein than Adh-S homozygotes (Laurie-Ahlberg & Stam, 1987). This difference has been shown to be due in part to a polymorphism in the first Adh intron that affects levels of ADH expression (Laurie & Stam, 1994). Quantitative differences in the level of ADH expression may also play a role in ethanol tolerance, and the results of this study (Fig. 4) are consistent with previous studies that have documented a positive correlation between ADH expression level and ethanol tolerance (Schmitt et al., 1986; Geer et al., 1993). However, in these previous studies, the association between expression level and tolerance was established without consideration of genotype, so that the effects of genetic background, catalytic differences among Adh genotypes, and differences in expression level were not separated in their analyses. Because the Adh-F genotype has both a higher catalytic activity and higher levels of ADH protein expression, it could be reasonably argued that the positive correlation between ADH expression level and ethanol tolerance was driven by catalytic differences among the ADH proteins, rather than by levels of ADH expression. In this study, the genetic backgrounds of the Wa-F, 1 Leu, 6 Leu, and 10 Leu genotypes were identical, as were the catalytic properties of the ADH proteins. The differences in the ethanol tolerance are therefore due solely to quantitative differences in the level of ADH protein, supporting the notion that both the catalytic activity and expression level contribute to ethanol tolerance.

It should be noted that, because of the artificial conditions imposed on the flies in the laboratory, the significance of this effect in natural populations remains unclear. This study represents an extreme scenario for several reasons. First, the types of synonymous changes were deliberately selected to maximize the probability of observing an effect: Adh is one of the most highly expressed genes, and exhibits some of the highest levels of codon bias, in the Drosophila genome. Furthermore, the leucine codon family is among the most highly biased codon families in the Drosophila genome (Duret & Mouchiroud, 1999; Nakamura et al., 2000; McVean & Vieira, 2001). Estimates of |s| = 10−6 for selection at synonymous sites from molecular sequence data are based on analyses of the average across all codon families of highly and weakly expressed genes (Akashi, 1995). Leucine codons exhibit significantly higher levels of codon bias than the majority of codon families that contribute to the 10−6 average. This difference is because of the fact that amino acids with large codon families (i.e. leucine, serine, and arginine) are probably under stronger directional selection because they have greater opportunity for unfavourable codon-tRNA pairing (McVean & Vieira, 2001). Evidence from the change in relative synonymous codon usage (ΔRSCU) in highly vs. weakly expressed genes supports this notion: the preferred codons of six-fold degenerate leucine, serine, and arginine codon families exhibit the greatest ΔRSCU among all codons (Duret & Mouchiroud, 1999). It is therefore conceivable that selection at leucine codons in the highly expressed Adh gene is approximately an order of magnitude stronger (i.e. |s| ≈ 10−5) than the codon-wide average. Consequently, unpreferred synonymous changes in the leucine codons of the Adh gene are most likely to result in significant changes in the level of ADH expression.

Secondly, the standard method of measuring ethanol tolerance, which was used in this study, involves confining adult flies to airtight containers and constant exposure to ethanol, is highly artificial. In their natural habitat adult flies are probably not exposed to ethanol for more than a few minutes at a time, although they do ingest significant amounts of ethanol in their diet (Geer et al., 1993). Clearly the majority of synonymous substitutions in the Drosophila genome would probably not result in measurable fitness differences. Nevertheless, this study demonstrates that the fitness effects of carefully selected unpreferred synonymous substitutions are measurable in the lab, and warrant further experimental investigation.

Acknowledgments

I thank Masaaki Onishi and Alexis Capozzoli for their assistance in conducting some of the ethanol tolerance assays, and two anonymous reviewers for improving the manuscript through their constructive criticism. This research was funded in part by an American University Senate Research Grant awarded to D.B.C.

Ancillary