Experimentally Increased Codon Bias in the Drosophila Adh Gene Leads to an Increase in Larval, But Not Adult, Alcohol Dehydrogenase Activity

Abstract

Although most amino acids can be encoded by more than one codon, the synonymous codons are not used with equal frequency. This phenomenon is known as codon bias and appears to be a universal feature of genomes. The translational selection hypothesis posits that the use of optimal codons, which match the most abundant species of isoaccepting tRNAs, results in increased translational efficiency and accuracy. Previous work demonstrated that the experimental reduction of codon bias in the Drosophila alcohol dehydrogenase (Adh) gene led to a significant decrease in ADH protein expression. In this study we performed the converse experiment: we replaced seven suboptimal leucine codons that occur naturally in the Drosophila melanogaster Adh gene with the optimal codon. We then compared the in vivo ADH activities imparted by the wild-type and mutant alleles. The introduction of optimal leucine codons led to an increase in ADH activity in third-instar larvae. In adult flies, however, the introduction of optimal codons led to a decrease in ADH activity. There is no evidence that other selectively constrained features of the Adh gene, or its rate of transcription, were altered by the synonymous replacements. These results are consistent with translational selection for codon bias being stronger in the larval stage and suggest that there may be a selective conflict over optimal codon usage between different developmental stages.

DESPITE the redundancy of the genetic code, synonymous codons are not used with equal frequency—a phenomenon known as codon bias (Ikemura 1981). Codon bias is apparent in the genomes of a wide array of organisms including eubacteria, archaea, and both unicellular and multicellular eukaryotes; it is essentially a universal property of genomes. The two main hypotheses that have been proposed to account for synonymous codon bias are (1) mutational bias (including biased gene conversion) and (2) natural selection for translational accuracy and/or efficiency (reviewed by Akashi 2001; Duret 2002).

In Drosophila, several lines of evidence suggest that codon bias results from natural selection for translational accuracy and/or efficiency. The lack of a significant association between intronic and synonymous site base composition indicates that mutational bias cannot account for codon bias (Vicario et al. 2007). Optimal codons, those synonymous codons whose usage shows a statistically significant increase in frequency with increasing gene expression (Duret and Mouchiroud 1999), tend to match the most abundant species of isoaccepting tRNA (Moriyama and Powell 1997). Codon bias is most extreme in highly expressed genes (Sharp and Li 1986; Duret and Mouchiroud 1999) and is significantly higher in the functionally constrained codons of proteins (Akashi 1994). Additionally, in Drosophila codon bias is highest in genes with maximal expression in the larval stage, which is the stage requiring the fastest rate of protein synthesis (Vicario et al. 2008). These observations support the hypothesis that codon bias results from natural selection for translational accuracy and efficiency (Bulmer 1991), referred to herein as the translational selection hypothesis.

Although there is a substantial body of indirect evidence for translational selection driving synonymous codon usage in Drosophila, direct experimental evidence for the translational selection hypothesis is comparatively sparse. Experimental reduction of codon bias in the leucine codons of the alcohol dehydrogenase (Adh) gene, the most highly biased codon family in one of the most highly expressed genes in the Drosophila melanogaster genome, resulted in a significant reduction in ADH activity (Carlini and Stephan 2003) and rendered flies less tolerant to ecologically relevant levels of environmental ethanol (Carlini 2004). However, to date no studies have been conducted to examine the functional effects of experimentally increased codon bias in Drosophila. This is a significant consideration, because the levels of codon bias observed in the most highly expressed genes rarely approach the theoretical maximum. At present, it is unclear whether this reflects the shape of the fitness curve for codon bias (i.e., diminishing returns due to tRNA saturation), interference from adaptive amino acid substitutions within the same gene (Betancourt and Presgraves 2002; Comeron and Kreitman 2002; Hambuch and Parsch 2005), or some trade-off between translational selection and other factors that influence synonymous codon usage such as mRNA stability (Carlini et al. 2001; Chamary and Hurst 2005a), exonic splice enhancers (Willie and Majewski 2004; Chamary and Hurst 2005b; Parmley and Hurst 2007), and/or transcription-driven mutagenesis (Hoede et al. 2006).

In this study we build on previous work (Carlini and Stephan 2003), again focusing on leucine codons in the Adh gene because of the high levels of codon bias observed in Drosophila leucine codons (Table 1). Overall, Adh has a frequency of optimal codon usage (Ikemura 1981) of 75%, and 20 of its 27 (74%) leucine codons are the optimal CTG. To investigate the effect of increasing codon bias on ADH protein expression, we performed site-directed mutagenesis to replace the seven suboptimal leucine codons with the optimal CTG codon. The in vivo ADH activity imparted by the mutant allele was compared to that of the wild-type allele in stable transformed lines that otherwise lacked a functional Adh gene. Using standard transformation methods, we were unable to detect a difference in ADH activity between wild-type and mutant transformants. However, the use of a more sensitive transformation method that eliminates genomic position-effect variation on transgene expression (Siegal and Hartl 1996, 1998; Parsch 2004) revealed developmental stage-specific differences between wild-type and mutant transformants, with the introduction of optimal leucine codons leading to an increase in larval ADH activity, but a decrease in adult ADH activity. These results suggest that the intensity of translational selection varies over the course of development and that there may be an evolutionary conflict over optimal codon usage between life history stages.

TABLE 1.

Codon usage bias in the leucine codons of D. melanogaster

Codon	Genomewide usagea (%)	Genomewide RSCUb	Weakly expressed RSCUc	Highly expressed RSCUd	ΔRSCUe
TTA	4.30	0.26	0.38	0.21	−0.17
TTG	17.70	1.06	1.18	1.08	−0.10
CTA	9.20	0.55	0.57	0.44	−0.13
CTC	15.40	0.92	0.93	0.90	−0.03
CTG	43.50	2.61	2.33	2.81	0.48
CTT	9.80	0.59	0.61	0.56	−0.05

^aGenomewide codon usage (n = 13,464 genes) from Hambuch and Parsch (2005).

^bRSCU, relative synonymous codon usage (Sharp et al. 1986).

^cGenes in the lowest 5% of expression determined by microarray hybridization (Gibson et al. 2004).

^dGenes in the highest 5% of expression determined by microarray hybridization (Gibson et al. 2004).

^eThe difference in RSCU between highly expressed and weakly expressed genes. Codons for which ΔRSCU > 0 are defined as optimal codons (Duret and Mouchiroud 1999).

MATERIALS AND METHODS

Site-directed mutagenesis:

The Adh Wa-F allele (Kreitman 1983) was used as the wild-type allele for all experiments. For mutagenesis, an 8.6-kb BglII fragment containing the complete Adh transcriptional unit and ∼5.5 kb of upstream flanking region was excised from the plasmid pΔWaf2a (Choudhary and Laurie 1991) and inserted into the BamHI site of the vector pBluescript SK+ (Stratagene, La Jolla, CA). To facilitate subsequent cloning steps, an XhoI linker sequence (New England Biolabs, Beverly, MA) was then inserted into the SpeI site. The resulting plasmid, designated as pBSX2a, served as the template for site-directed mutagenesis.

The wild-type Adh allele contains 27 leucine codons, 7 of which are not the optimal CTG (Figure 1). All 7 of these codons (either TTG or CTC) were changed to CTG using the QuikChange XL site-directed mutagenesis kit (Stratagene). The primer pairs used for mutagenesis were as follows (given are the forward primers in 5′−3′ orientation, and the reverse primers are complementary; the mutated nucleotide is underlined): Leu5_T→C, CC ATG TCG TTT ACT CTG ACC AAC AAG AAC GTG ATT TTC GTG GCC G; Leu28_C→G, C ACC AGC AAG GAG CTG CTG AAG CGC GAT CTG AAG GTA AC; Leu38_C→G, G AAC CTG GTG ATC CTG GAC CGC ATT GAG AAC CCG GC; Leu198_T→C, G CAC ACG TTC AAC TCC TGG CTG GAT GTT GAG CCT CAG G; Leu208_C→G, GTT GCC GAG AAG CTG CTG GCT CAT CCC ACC CAG C; Leu217_T→C, ACC CAG CCC TCG CTG GCC TGC GCC GAG AAC; and Leu240_T→C, CC ATC TGG AAA CTG GAC CTG GGC ACC CTG GAG GC. The resulting Adh allele with 7 suboptimal codons replaced by optimal codons was designated as 7up and its sequence was confirmed using a MegaBACE automated sequencer and the DYEnamic ET terminator cycle sequencing kit (Amersham Biosciences, Piscataway, NJ).

Figure 1.—

Seven suboptimal leucine TTG or CTC codons in the wild-type Adh allele were replaced with optimal CTG codons to construct the Adh 7up allele. Boxes indicate exons, horizontal lines indicate introns. The locations of the suboptimal codons within the coding sequence are indicated by solid rectangles and the amino acid positions are given below.

Transformation vector construction:

For standard P-element-mediated germline transformation we used the YES transformation vector, a P-element vector containing the D. melanogaster yellow (y) gene as a selectable marker (Patton et al. 1992). The YES vector was used in previous experiments involving the reduction of codon bias in the Drosophila Adh gene (Carlini and Stephan 2003) and its use in the present study thus provides a means of directly comparing the effects of increasing codon bias with previous results. To introduce the 7up allele of Adh into the YES vector, an 8.6-kb ClaI fragment containing 7up was excised from the plasmid pBSX2a (described above) and ligated into the ClaI site of the YES vector. The sequence of the 7up allele in the YES vector was confirmed by DNA sequencing, using a LI-COR 4300 automated sequencer and the SequiTherm EXCEL II DNA cycle sequencing kit (Epicentre Biotechnologies). The final transformation vector was designated as pP[YES-7up].

For transgene coplacement, an 8.6-kb BglII fragment containing the wild-type Adh gene was excised from the plasmid pΔWaf2a and cloned into the BamHI site of the vector pP[wFl] (Siegal and Hartl 1996). This vector contains two cloning sites for inserts that are to be compared, each flanked by target sequences for a different site-specific recombinase, FLP or Cre. The cloning and recombination sites also flank the mini-white (w) gene of D. melanogaster, which serves as a selectable eye-color marker. The BamHI site is located upstream of the w gene and is referred to as cloning site 1. The 7up allele of Adh was excised from the pBSX2a mutagenesis vector as an 8.6-kb XhoI fragment and inserted into the XhoI site of pP[wFl] (cloning site 2 located downstream of the w gene), which already contained the wild-type allele at cloning site 1. This final vector was designated as pP[wFl-2a-7up]. In this construct, the two alleles of Adh are arranged in a head-to-head orientation, meaning that they are transcribed from opposite strands of the DNA. This is not expected to affect their relative expression in a systematic way, as long as pairs of coplaced alleles are compared (see below).

Germline transformation:

Germline transformation using the pP[YES-7up] vector was performed by microinjection of y w; Adh^fn6; Δ2–3, Sb/TM6 embryos. Adh^fn6 is a null allele (splicing defect) that produces no detectable ADH protein (Benyajati et al. 1982). The Δ2–3 P insertion on the third chromosome served as the source of transposase (Robertson et al. 1988). Following injection, surviving adults were crossed to y w; Adh^fn6 flies and transformant offspring were identified by their wild-type body color. Additional lines with inserts at unique chromosomal locations were generated through mobilization crosses as follows. Transformants carrying insertions on the X chromosome were crossed to the y w; Adh^fn6; Δ2–3, Sb/TM6 stock and transformants carrying insertions linked to the Sb marker (i.e., those with insertions linked to the source of transposase) were crossed to the y w; Adh^fn6 stock. Mobilized insertions were identified as y⁺ offspring where the y⁺ marker was not segregating with the same chromosome as the parental insert. When necessary, further crosses to the y w; Adh^fn6 stock were performed to remove the Δ2–3 source of transposase and establish stable transformed lines. Southern blots were performed to confirm that transformant lines contained a single insertion of the transgene (one line was found to contain a double insert and was not used in subsequent analyses). Only autosomal-insertion lines were used for subsequent analysis. For comparison, previously described transformants carrying the wild-type Adh allele were used (Parsch et al. 1999, 2000a,b; Carlini and Stephan 2003).

Germline transformation using the pP[wFl-2a-7up] vector was performed by microinjection of y w; Δ2–3, Sb/TM6 embryos. This strain carries the endogenous Adh gene that was later removed through crossing (see below). Successfully transformed flies showing red eye color were crossed to y w flies to remove the source of transposase (if still present) and establish stable transformed lines. In cases where the transgene inserted onto the third chromosome carrying the transposase gene, the insert was immediately remobilized by crossing with y w flies and selecting for offspring with red eyes and lacking the Sb marker (indicating the absence of the chromosome carrying the transposase gene). Transformed lines were then crossed to a strain with multiple phenotypic markers (y w; CyO/Sco; Sb/TM6, Ubx) to determine which chromosome contained the insertion and to establish homozygous lines for each independent insertion. Only lines with insertions on the third chromosome were used for subsequent transgene coplacement.

Transgene coplacement:

Following the protocol of Siegal and Hartl (1996), females of the transformed fly strains homozygous for pP[wFl-2a-7up] insertions were mated with males from a stock carrying both the FLP and cre recombinase genes (y w; MKRS, FLP/cre, TM6B), thereby producing offspring with one of the recombinase genes on one third chromosome and the transgenic insert on the other. These two types of flies were separated and treated independently. In the first treatment, cre expression was induced by rearing the flies at 25° to excise the wild-type Adh allele along with the w gene. In the second treatment, FLP expression was induced by heat shock at 38° for 1 hr during the first larval instar stage, which resulted in the removal of the 7up Adh allele together with the w marker gene. In both cases, successfully excised alleles generated flies with white eyes. Additional crosses to the above marker strains were performed to remove the recombinase genes and establish lines homozygous for their respective Adh inserts. This resulted in matched pairs of fly strains with homozygous third chromosome insertions of either the wild-type or the 7up allele of Adh.

Since the original injection stock and all other flies used in the above crossing scheme carried the endogenous Adh gene on chromosome 2, we performed additional crosses to remove the endogenous gene so that the two introduced Adh alleles could be tested in an otherwise Adh-null background. This was done with crosses to a stock of y w; Adh^fn6 flies and the above-mentioned strain y w; CyO/Sco; Sb/TM6, Ubx.

ADH activity assays:

Males of all transformed lines were crossed to y w; Adh^fn6 females to produce offspring heterozygous for their respective Adh insertion in an otherwise Adh-null genetic background. These offspring (individual third-instar larvae or males aged 6–8 days) were used for ADH assays following standard protocols (Maroni 1978), using isopropanol as the substrate. ADH activity units were defined as micromoles of NAD⁺ reduced per minute per milligram of total protein (multiplied by 100). For the P[YES-7up] transformants, the total protein concentration of the crude extracts was determined using the RC DC Protein Assay kit (Bio-Rad, Hercules, CA). For the P[wFl-2a-7up] transformants, total protein concentration was estimated by the method of Lowry et al. (1951).

For the P[YES-7up] adult ADH assays, the above crosses were repeated in two separate blocks, and from each cross two independent cohorts of five flies each were used. This resulted in a total of four adult ADH activity measurements for each transformed line. A one-way nested ANOVA (cross nested within line nested within genotype) was used to test the null hypothesis of no difference in adult ADH activity between genotypes. Larval ADH assays on P[YES-7up] transformants were performed on four to eight individual larvae per transformant line. Two replicate assays were performed on each larva, resulting in a total of 8–16 ADH activity measurements for each transformed line. A one-way nested ANOVA was used to test the null hypothesis of no difference in larval ADH activity between genotypes.

For the P[wFl-2a-7up] transformants ADH activity assays were performed on individual adult flies or third-instar larvae with two technical replicates per individual. Five to seven individual flies (or four to eight individual larvae) were assayed in each of nine lines and a one-way nested ANOVA (with genotype nested within chromosomal location) was used to test the null hypothesis of no difference in ADH activity between genotypes. A simulated F distribution was generated by permuting all values of wild-type and 7up ADH activity to generate 10,000 data sets.

Quantitative RT–PCR on P[wFl-2a-7up] transformants:

Total RNA was isolated from each of four individual 6- to 8-day-old male flies per line per genotype or six third instar larvae per genotype using the TRIzol reagent (Invitrogen). cDNA was synthesized using ∼70–360 ng of DNase I-treated total RNA and diluted to either 1:50 for adults or 1:5 for larvae in the subsequent quantitative (q)RT–PCR. qRT–PCR was performed using TaqMan assays (Applied Biosystems, Foster City, CA) per the manufacturer's recommendations. The Adh probe that covered the exon–exon junction of exons 2 and 3 of the Adh gene and the constitutive ribosomal protein RpL32 gene were used as an endogenous control. All assays were performed in duplicate and expression was quantified using the comparative cycle threshold (ΔCt) method by subtracting the average RpL32 control gene Ct value from the average Adh Ct value for each individual fly.

RESULTS

We used site-directed mutagenesis to create an allele of Adh in which the seven suboptimal leucine codons present in the wild-type sequence were replaced by the optimal codon, CTG (Figure 1). This mutant allele was designated as 7up and was compared to the wild-type allele in transformed lines of D. melanogaster that otherwise lacked a functional Adh gene. Because the amino acid sequences encoded by the wild-type and 7up alleles were identical, any differences in ADH activity could be attributed to differences in the efficiency or accuracy of ADH protein production.

Using standard P-element transformation (YES vector), we compared the adult ADH activity of 10 independent transformed lines with the wild-type Adh allele and 11 independent transformed lines with the 7up allele (Figure 2). We observed no difference in ADH activity between the two genotypes (Nested ANOVA, P = 0.953). The average ADH activity of the wild-type lines (112.79 ± 16.83) was virtually identical to that of the 7up lines (112.09 ± 31.44). Due to the random insertion location of the Adh transgenes in the Drosophila genome using the YES vector, substantial position-effect variation was observed among lines within genotypes (nested ANOVA, P = 0.043). We performed the same experiments on individual larvae and again found that there was no significant difference in ADH activity between the two genotypes (Figure 2, nested ANOVA, P = 0.297). Position-effect variation was slightly less pronounced in the larval assays (P = 0.068).

Figure 2.—

ADH activity (μmol NAD⁺ reduced/min/mg total protein × 100) of wild-type (open bars) and 7up (shaded bars) lines obtained from standard P-element transformation. ADH activity did not differ between the two genotypes in adults (top, nested ANOVA, P = 0.953), or in larvae (bottom, P = 0.297). There was significant position-effect variation among lines within genotypes for adults (P = 0.043) but not for larvae (P = 0.068). Error bars indicate ±1 standard deviation from the mean.

To avoid the problem of position-effect variation and increase our power to detect a difference between the wild-type and 7up alleles, we repeated the above experiment using the method of transgene coplacement (Siegal and Hartl 1996). This method allows us to introduce both alleles into the same chromosomal location and then remove one or the other allele through site-specific recombination. As a result, the two alleles can be compared in an otherwise identical genomic context. In total, we obtained nine pairs of transformed lines with coplaced wild-type and 7up alleles on the third chromosome. Overall, in adults the wild-type transformants had on average ∼10% higher ADH activity than 7up transformants, with the effect of genotype being significant (nested ANOVA, P = 0.046; Figure 3). In contrast, in larvae the wild-type transformants had on average ∼10% lower ADH activity than 7up transformants, with the effect of genotype being highly significant (nested ANOVA, P = 0.0002). There was a significant positive correlation between adult and larval ADH activity for all coplaced pairs of wild-type and 7up transformants (R = 0.57, P = 0.01), indicating that the chromosomal context of the insertion site had a similar effect on both larval and adult expression.

Figure 3.—

Ratio of 7up to wild-type ADH activity in adults (top) and larvae (bottom) for nine independently coplaced pairs of alleles. The dashed lines indicate the mean ratio while the shaded lines indicate the null hypothesis expectation of 1. Error bars indicate the 95% confidence interval for each line. For adults, the overall mean ratio of 7up to wild-type ADH activity is 0.90 (nested ANOVA, P = 0.046); for larvae, the overall mean ratio of 7up to wild-type ADH activity is 1.08 (nested ANOVA, P = 0.0002).

Because the differences in ADH activity between wild-type and 7up lines were significant using transgene coplacement in adults and larvae, qRT–PCR using TaqMan assays was performed on adults and larvae to determine if this result might be due to differences in mRNA quantity rather than translational effects. We were unable to detect any significant differences between wild-type and 7up mRNA expression for any of the coplaced pairs in adults or larvae (Figure 4). Due to the high variance among biological replicates, none of the comparisons within lines were significant. Average ΔCt values across locations were nearly identical in larvae (wild type = 0.924, 7up = 1.027) and in adults (wild type = –2.910, 7up = −2.532).

Figure 4.—

qRT–PCR measurement of Adh mRNA abundance in coplaced pairs of wild-type (solid circles) and 7up (shaded circles) transformants. (Top) The ΔCt (relative to the control RpL32 gene) in adult males for four biological replicates each of all nine coplaced pairs. (Bottom) The ΔCt in third-instar larvae for six biological replicates each of four coplaced lines that showed a significant difference in ADH activity between alleles (see Figure 3). Note that lower numbers indicate higher expression and negative numbers indicate that Adh shows higher expression than RpL32. No significant differences in Adh expression were detected between coplaced alleles at any location or developmental stage (t-test, P > 0.05 in all cases).

DISCUSSION

Previous work has shown that experimentally decreasing codon bias in the D. melanogaster Adh gene leads to a reduction in ADH activity (Carlini and Stephan 2003). For example, the introduction of six suboptimal leucine codons reduced ADH activity by 19% in adult flies—a result consistent with the translational selection hypothesis. In the current study, we have further tested this hypothesis by performing the reverse experiment: codon bias was increased by replacing seven suboptimal leucine codons present in the wild-type Adh gene with the optimal leucine codon, CTG. We assayed ADH activity in adults as well as in third instar larvae. In adults, the introduction of these optimal codons did not lead to an increase in ADH activity. Instead, our transgene coplacement experiment indicates that it decreased ADH activity in adults. Conversely, in larvae these same synonymous replacements led to a highly significant increase in ADH activity in 7up lines transformed via transgene coplacement. In the following, we consider several possible explanations for these intriguing results.

One possibility is that the optimal codon substitutions improved the efficiency and/or accuracy of translation, but deleterious effects on mRNA stability and/or splicing obviated these beneficial effects. Such a scenario could account for the reduction of ADH activity observed in adult 7up transformants, but not for the increase in activity observed in larvae. Nevertheless, it is possible to examine the potential effects on mRNA stability by comparing the folding free energies of the most stable global mRNA secondary structures of the wild-type and 7up transcripts. Since the optimal substitutions involved four T → C changes and three C → G changes, it follows that the 7up version of Adh would have a more stable global secondary structure, with T → C changes increasing the opportunity for more stable G–C pairings and C → G changes allowing for additional G–U pairings at the RNA level. Using the program mfold (Zuker 2003), we found the most stable secondary structure of the adult primary transcript of the wild-type Adh allele to have a folding free energy of −588.5 kcal/mol. The folding free energy of the most stable structure for the 7up adult primary transcript was −594.9 kcal/mol, a difference of −6.4 kcal/mol. The average folding free energy of the 10 most stable structures for the 7up adult primary transcripts was significantly less than that of the 10 most stable wild-type adult primary transcripts (7up, −589.87 kcal/mol; wild-type, −584.00 kcal/mol; two-tailed t-test, P < 0.001). Similar results were obtained when considering only the coding regions of the wild-type and 7up sequences (two-tailed t-test: P < 0.01). Although statistically significant, it is unlikely that these differences in mRNA stability are large enough to have an effect on translation. Two recent genomewide studies have shown that global mRNA stability (measured as folding free energy of full-length mRNA) is not correlated with gene expression in Drosophila (Stenøien and Stephan 2005; Eck and Stephan 2008).

We also compared the local structures within each of the most stable global structures of the wild-type and 7up mRNAs and found no evidence for biologically significant differences among local structures. For example, the 13 most stable helices among those in the 7up and wild-type mRNAs were identical and ranged from −60.1 (24 bp) to −13.3 kcal/mol (7 bp). The remaining helices ranged from −13.3 (7 bp) to −1.3 kcal/mol (2 bp), and no differences >0.9 kcal/mol were observed in a ranked list of helices. None of these minor differences in local structures appear to be sufficient to differentially inhibit the helicase activity of the ribosome, which has been experimentally demonstrated to be capable of melting a highly stable 27-bp helix (−52.1 kcal/mol) without dissociation from the mRNA (Takyar et al. 2005). Furthermore, the 7up mutations did not alter any of the putative secondary structural elements identified by previous covariation analysis of multiple Drosophila species or by previous experimental manipulation (Kirby et al. 1995; Parsch et al. 1997; Carlini et al. 2001; Chen and Stephan 2003; Baines et al. 2004). We also determined a consensus mRNA secondary structure for Adh coding sequences of 12 Drosophila species from the recent 12 genomes project, using RNAalifold (Gruber et al. 2008). RNAalifold determines a consensus structure of a set of aligned sequences by averaging free energy contributions over all sequences while also scoring covariations to account for compensatory mutations. None of the seven nucleotides we altered were within stem regions of the consensus structure, lending support to the conclusion that the 7up mutations did not significantly alter secondary structure. However, we were unable to obtain a reliable alignment of Adh pre-mRNA sequences due to substantial variation in the noncoding nucleotides among the 12 sequences, so a possibility remains that the mutated nucleotides pair with noncoding portions of the pre-mRNA, although the analyses of the D. melanogaster wild-type and 7up pre-mRNA sequences described above do not indicate that this is the case.

The 7up synonymous substitutions could also affect ADH activity by altering one or more exonic splicing enhancers (ESEs). These cis-acting motifs tend to occur near exon–intron boundaries and are enriched in A's and diminished in C's, precisely opposite the pattern observed for optimal codons in Drosophila (Vicario et al. 2007). A recent genomewide survey of D. melanogaster exons provided evidence of a trade-off between the use of translationally optimal codons and the regulation of splicing (Warnecke and Hurst 2007). Because three of the 7up mutations involved C → G changes, it can be reasoned that they favored the creation of ESEs. However, the other four mutations involved T → C changes, presumably resulting in the disruption of ESEs. We used two software applications to determine if the 7up mutations altered splicing motifs in Adh. To date most work on the identification of ESE motifs has focused on mammals but many of the SR proteins, which recognize and bind the mRNA at ESEs, are strongly conserved within the metazoa (Barbosa-Morais et al. 2006); furthermore, in Drosophila patterns of codon enrichment/depletion at exon–intron boundaries are similar to those in mammals (Warnecke and Hurst 2007). Thus, many of the ESE motifs identified in mammals are likely to be functional in Drosophila. ESEfinder3.0 (Cartegni et al. 2003) was used to locate these comparatively well-characterized ESEs. Recently a set of ESEs was identified in Drosophila using both the RESCUE-ESE approach of Fairbrother et al. (2002), which was successfully used to identify human ESEs, and ELPH, a general purpose Gibbs sampler for finding sequence motifs (Pertea et al. 2007). The SEE ESE software application (http://www.cbcb.umd.edu/software/SeeEse/index.html) was used to determine whether any putative Drosophila ESEs were disrupted or created by the 7up mutations. The results from these analyses indicate that differences between wild type and 7up in ESE content were minimal and that, overall, the 7up mutations led to a roughly twofold increase in the number of ESEs in the Adh gene (Table 2). Thus, if anything, these differences are biased in favor of increased splicing efficiency of the 7up allele, which cannot account for the observed reduction in ADH protein activity. Furthermore, the results from qRT–PCR assays of expression are not consistent with splicing-related effects because the TaqMan probe for the Adh gene spanned the exon 2/exon 3 junction of the mature mRNA. We did not observe any consistent differences in the wild-type vs. 7up signal within transgenic lines (Figure 4); such differences would be predicted if differential splicing efficiency affected mature mRNA levels.

TABLE 2.

Total number of exonic splicing enhancers (ESEs) in the wild-type and 7up Adh coding sequences as predicted by two methods

	ESEfinder 3.0		SEE ESE
Codon	Wild type	7up	Wild type	7up
5	0	1	0	0
28	0	2	0	0
38	1	1	0	0
198	0	0	1	0
208	2	2	1	3
217	1	2	0	0
240	0	0	0	0
Total	4	8	2	3

Another possibility is that the suboptimal leucine codons present in the wild-type Adh gene play a functional role in translational pausing, which has been implicated as a requirement for proper protein folding (Buchan and Stansfield 2007). If so, we would expect that the degree of functional constraint at these codons would be comparable to that at optimal leucine codons. We evaluated this by comparing levels of overall sequence divergence (including synonymous and nonsynonymous substitutions) at homologous positions in the Adh genes of 12 Drosophila species (Drosophila 12 Genomes Consortium 2007). In pairwise comparisons among the 12 Adh homologs, the average nucleotide sequence divergence at the seven suboptimal leucine codons exceeded the average for the coding sequence as a whole and also that at the 20 preferred CTG leucine codon positions (Table 3). Because these comparisons involved a relatively wide range of divergence times, we also calculated sequence divergence for two more restricted subsets of taxa, (i) in the subgenus Sophophora and (ii) in the melanogaster subgroup for Adh as a whole, for the 7 suboptimal and the 20 optimal leucine positions and found that the pattern was more extreme in the more restrictive taxonomic groups (Table 3). Thus, if anything, there appears to be less functional constraint at the 7 suboptimal leucine codon positions, consistent with Akashi (1995) and inconsistent with the idea that these suboptimal codons are adaptively positioned to ensure proper cotranslational folding of the nascent polypeptide.

TABLE 3.

Average uncorrected pairwise sequence divergences (%) for the entire coding region, at the 20 optimal leucine codons and at the 7 suboptimal leucine codons of the Adh gene

Region compared	melanogaster subgroupa	Subgenus Sophophorab	12 Drosophila speciesc
Entire Adh coding region	3.61	11.70	16.76
20 optimal CTG codons	0.67	11.25	19.67
7 suboptimal codons	6.67	24.34	25.76

^aD. melanogaster, D. simulans, D. sechellia, D. yakuba, and D. erecta.

^bmelanogaster subgoup plus D. ananassae, D. pseudoobscura, D. peresimilis, and D. willistoni.

^cSophophora subgenus plus D. mojavensis, D. virilis, and D. grimshawi.

An important consideration is that the previous experiments decreased codon bias by replacing leucine codons of the wild-type Adh gene with the rarely used leucine codon CTA—a codon that does not occur naturally in the Adh gene. It may be that the introduction of CTA codons has a much stronger effect on protein expression than the introduction of CTG codons. In the wild-type Adh sequence, only 7 of the 27 leucine codons are not the optimal CTG. This limited our experimental options, as only CTC or TTG codons could be altered. Although both of these codons are used less frequently in highly expressed genes than CTG, they are not as strongly avoided as CTA (Table 1) (Chen et al. 1999). For a limited set of genes, the CTC codon even demonstrated a significant increase in its use as codon bias within a gene increases, causing it to be defined as one of two “preferred” leucine codons (Akashi 1994, 1995), although our genomewide analysis of codon usage and gene expression presented in Table 1 indicates that CTC codons are used less frequently in highly expressed genes. Regardless, there is likely to be an asymmetry in the effects of introducing CTA vs. CTG codons.

It is possible that there are diminishing returns to increasing codon bias with respect to translational efficiency in adults, but not in larvae, due to differences in overall protein production rates associated with the two life history stages. The wild-type Adh gene already shows very strong bias in synonymous codon usage, and increasing this bias further may have little or no impact on translation in adult cells if the tRNA pool is saturated, i.e., when the availability of charged tRNAs rather than free ribosomes limits the efficiency and/or accuracy of translation. Recent experimental evidence for tRNA pool saturation in human cells is provided by Coleman et al. (2008), who constructed artificial polio capsid proteins that were either enriched or depleted in codons that are preferentially used by the host (human) genome. Notably, the capsid protein construct that was most highly enriched in preferred codons was not more virulent than the wild-type protein. In contrast, constructs that were enriched in unpreferred codons were attenuated in human cells. This suggests that the degree of codon bias in the wild-type polio capsid protein has already been optimized over evolutionary time, and the observed lack of increased virulence was due to saturation of the tRNA pool.

Recently Kudla et al. (2009) proposed that codon bias arises from selection to make translation globally efficient, rather than at the level individual genes, on the basis of their results from experimental manipulation of codon bias in Escherichia coli. Under this hypothesis, high codon bias increases the overall elongation rate by increasing the pool of free ribosomes available to initiate translation, the rate-limiting step of protein synthesis. In Drosophila larvae, the ratio of free ribosomes to actively translating ribosomes is likely to be much lower than in adults because of higher rates of protein synthesis needed to account for the ∼300-fold increase in protein quantity from newly hatched first-instar to full grown third-instar larvae (Vicario et al. 2008). If the availability of free ribosomes, rather than that of charged tRNAs, limits translation rates in the larval stages of Drosophila, the introduction of preferred codons would increase the availability of free ribosomes to initiate translation because each transcript could be translated more rapidly. Although this scenario predicts that increasing codon bias should have a global effect on the rate of translation and not on the translation of any particular gene's mRNA, the high expression level of Adh may allow it to serve as a proxy for the overall rate of translation. This could explain why the introduction of optimal leucine codons increased ADH activity in larvae, but not in adults. Overexpression of Adh mRNA has been shown to slow progression through the larval stages (Parsch et al. 2000a,b), which is consistent with developmental rate being limited by the rate of translation. Furthermore, if the wild-type Adh gene is already optimized for synonymous codon usage, this would suggest that it represents a compromise between the optima in adults and that in larvae. Thus, there may be selective conflict over optimal codon usage at different life history stages that affects levels of codon bias in many genes with a developmentally broad pattern of expression.

In summary, our analyses suggest that the 7up mutations are unlikely to significantly alter the Adh mRNA secondary structure, splicing code, or translational pausing. The results from qRT–PCR assays do not point to any clear differences between wild-type and 7up transcipt levels. Thus, the observed differences in ADH activity are most likely the result of translational effects. Future experiments that alter codon bias and tRNA expression individually and in combination could test our hypothesis that the free ribosome pool is limiting in larvae, whereas the tRNA pool is limiting in adults, as well as shed light on the coevolutionary dynamics that led to the emergence of codon bias as a ubiquitous feature of genomes.