Introduction

In their analysis of a complete mitochondrial genome of an ostrich, Härlid et al. (1997) reported an apparent truncation of the nad3 (NADH dehydrogenase subunit 3) gene, due to the insertion of a single nucleotide at position 174 in the gene, altering the reading frame. Only the first 58 of 116 residues of the nad3 protein would be translated correctly. The rest of the gene is present, but not accessible in the starting reading frame. Mindell et al. (1998) examined that region in the mitochondrial genomes of a large number of bird species and in the painted turtle. They found the extra base in 46 of 61 bird species and in the turtle. More surprising, the entire reading frame could be restored in all taxa by deleting the extra nucleotide from the data file. There was nothing else wrong with these sequences. They examined the pattern of evolution among the taxa and concluded that the sequences are functional. Evidently the extra nucleotide is skipped during translation. They suggested that the site was subject to a +1 programmed frameshift during translation, allowing correct translation of a full-length, functional protein. Their report was the first evidence of programmed translational frameshifting in mitochondrial genomes.

During a phylogenetic study of 30 species of the ant genus Polyrhachis, using the mitochondrial cytb (cytochrome b) gene, we discovered a single nucleotide insertion of a G in P. dives at approximate position 333 within the gene. As the study proceeded, a total of 12 species were found to have single nucleotide insertions at one or two sites in the cytb gene, with four different sites affected. Two of the sites are present in more than one species, allowing examination of the pattern of substitution in the molecule subsequent to the insertion.

In this paper, we report the finding of four new sites of +1 frameshift insertions in animal mitochondrial genes. The finding of additional sites gives us the opportunity to examine the sequences in the region of the insertions for common features that may give some insight into the mechanism. The results give additional evidence that the genes are translated in the original reading frame and point to +1 translational frameshifting as the mechanism that restores the correct reading frame at each insertion site. We present evidence that these sequences are functional and provide a model for their translation, consistent with the mechanisms hypothesized for translation of programmed frameshift sites in the TY1 and TY3 retrotransposons in yeast.

Materials and Methods

DNA Extraction, Amplification, and Sequencing

Total genomic DNA was extracted from a single leg of a representative of each species using organic solvents and ethanol precipitation, as described by Beckenbach et al. (1993). Selected regions of the mitochondrial genome were amplified using primers listed in Table 1. PCR products were purified either by gel isolation, followed by a freeze/squeeze extraction (Tautz and Renz 1983), or by using the QIAquick PCR Purification Kit (QIAGEN). The purified products were sequenced directly using the amplification primers. The PCR products were sequenced manually using 33P-labeled dideoxy cycle sequencing with Thermal Sequenase (Amersham) or using the automated sequencing services of the DNA Sequencing Lab, University of Calgary Core DNA & Protein Services.

Table 1 List of primers used for PCR amplification and sequencing of Polyrhachis mitochondrial genes
Full size table

All sequences were determined from both strands. Those showing insertions were sequenced from at least two different overlapping PCR products, using different PCR primer pairs. The gels (for manual sequencing) or sequencing scans were carefully examined for accuracy in the region of the insertions. In no case was a conflicting result observed, nor was there evidence of double sequence, as might be observed if multiple copies of the gene were present.

Analysis

Sequences were aligned manually. Aside from the single nucleotide insertions at four sites in the cytb gene, there were no indels in either of the genes. Protein coding sequences were translated using the Drosophila mitochondrial code (Clary and Wolstenholme 1985). Regions surrounding the frameshift insertions, including 4 codons upstream, the insert region, and 12 codons downstream (17 codons; 51 nucleotides [nt]), were scanned for potential secondary structure at 25°C in the mRNA using the mfold 2.3 web server (http://www.bioinfo.rpi.edu/applications/mfold [Zuker 2003]). A sliding window of fragments of the same length (51 nt), at 20-nt intervals, was also scanned for potential secondary structure at 25°C for the entire cytb gene of Polyrhachis turneri, to provide a basis for interpreting those results.

Three of the tRNA genes associated with the frameshift sites, tRNAGly, tRNAVal, and tRNAAla, were identified and folded using tRNAscan-SE (Lowe and Eddy 1997). We also sequenced the tRNASer-AGN genes from Polyrhachis phryne and from an outgroup ant species, Lasius alienus. The outgroup species was included to confirm identification of the gene. The genes were identified based on position in the genome, potential for forming a secondary structure consistent with those observed for each gene in other insects, and examination of the substitution pattern between the species. The sequences for tRNASer-AGN were folded using mfold 2.3 at 25°C with default settings for other parameters (Zuker 2003).

Results

We determined the complete sequences of the cytb and cox2 genes for Polyrhachis turneri. The cytb gene consists of 1116 nt, coding for a predicted product of 372 amino acids, while cox2 is 678 nt coding for 226 amino acids. Both genes are terminated by an inframe TAA stop encoded in the DNA sequence. We examined a 750-bp fragment of cytb and the first 579 bp of cox2 in 29 other species of the genus Polyrhachis. Each species gave a single, unique sequence for both genes. In all cases the cox2 sequences translated as expected. In 12 species, however, the cytb gene is not an open reading frame. In those species, a single nucleotide insertion at one or two sites disrupted the reading frame. These insertions were evident on both strands.

One insertion, at approximate position 333 in the cytb gene, was first discovered in Polyrhachis dives. After confirming it by sequencing both strands, we extracted DNA from three additional specimens and repeated the analysis. We designed additional PCR primers to examine the region, based on conserved regions in the Polyrhachis cytb sequences. The extra nucleotide was always present. As the survey of cytb sequences in Polyrhachis progressed, nine other species showed a single nucleotide insertion at the same site, eight of them an extra G. In all cases the extra nucleotide was present in sequences from both strands, and confirmed from at least two PCR products using different primer combinations.

In the course of the survey of Polyrhachis species, a total of 12 species showed single nucleotide insertions at one or two sites in the cytb gene (Table 2). A total of four sites were involved. If translated in the expected reading frame (0-frame), they would produce products of between 142 and 166 amino acid residues, less than half the expected length of the wild-type protein. The four sites are numbered according to the approximate position of the inserted nucleotide.

Table 2 Occurrence of frameshift insertions in the cytochrome b gene of 12 species of Polyrhachis
Full size table

Description of the Sites

The sequences at the four sites discovered in Polyrhachis, along with the site observed by Mindell et al. (1998) for several representative vertebrates, are given in Fig. 1. Included is a hypothetical “premutation” sequence obtained by removing the extra nucleotide.

Figure 1
figure 1

Sequence context of +1 frameshift sites in Polyrhachis species (this study) and selected vertebrates (Mindell et al. 1998). The “premutation” is the sequence that results after removal of the inserted nucleotide, based on sequence comparisons with species not carrying the frameshift.

Full size image

One site (FS333) at approximate position 333 within the gene was encountered in 10 species (Table 2). The sequence of the affected codon in 17 of the 20 other species is GGA, with third position G, T, and C appearing once each in the other three species. In nine of the FS333 species there is a third position G, so that the inferred mutation was GGA → GGG A (Fig. 1). The other FS333 species, P. sexspinosa, has an inserted T: GGA→GGT A. The insertion in this species could represent a recent G→T transversion, following an original insertion of a G or an independent insertion of a T at the same site. Species having an inserted nucleotide at this site include representatives of 5 of the 12 subgenera of the genus Polyrhachis. We analyzed two representatives of subgenera Hemioptica and Polyrhachis and three species of subgenus Myrmatopa. All seven members of these subgenera carried the insertion. Only one representative of subgenus Myrmothrinax showed the insertion; the other, P. queenslandica, did not. This subgenus is believed to be monophyletic (Rudy Kohout, personal communication). If so, the absence of FS333 in P. queenslandica must represent a secondary loss of the insertion. In addition, two of four representatives of subgenus Myrmhopla carried the FS333 insertion. This subgenus is divided into several species groups and is likely polyphyletic. A molecular phylogenetic study of these species will be published elsewhere.

In addition to the frameshift insertion at position 333, P. sexspinosa showed a second frameshift insertion at approximate position 411 (FS411). The consensus codon of the other 29 Polyrhachis species is GGA, although the third position is variable. The inferred insertion was GGN → GGG A (Fig. 1). The insertion could have been either a G prior to a third position A or an A following a GGG codon.

Two species showed a single nucleotide insertion at approximate position 474 (FS474; Fig. 1). These mutations appear to be an insertion of a G. The codon at that position is TGA in the other 28 species, so the inferred mutation was TGA → TGG A. One of these two species, P. phryne, also carries a second insertion at approximate position 773 (FS773). The consensus at that site is ATT but the site is variable. The mutation appears to be GTA → GGT A.

Potential for Secondary Structure Near the Frameshift Sites

We examined the regions surrounding the insertion sites for the potential to form stable secondary structures. Optimal folding energies at 25°C for 51-nt fragments, including 12 nt (4 codons) prior to the insertion site, the codon sustaining the insertion, and 36 nt (12 codons) downstream from each insertion, ranged from −8.2 to −12.4 kcal/mol for the 10 FS333 sites. The optimal folding energies for the other sites were −8.2 for FS411 of P. sexspinosa, −9.5 and −10.3 for FS473, and −7.9 for FS773 of P. phryne. For comparison, we examined the folding energies at 25°C for fragments of the same length in a 20-nt sliding window along the length of the entire cytb gene of P. turneri. Optimal folding energies across the gene ranged from 0.9 to −17.8, with a mean of −7.0 (more negative values indicating increased stability). Although the stabilities of the regions surrounding the frameshift insertions were above average, the optimal folding structures and energies varied among species. We conclude that there is no evidence for unusual secondary structure in the regions immediately downstream from the insertion sites that might account for translational stalls.

Codons Involved in the +1 Insertion Sites

Examination of the sequences in the regions of the four frameshift sites in ants, and that in vertebrates, reveals several similarities (Fig. 1). The insertion at all five sites occurred just prior to a GUA or GCA codon (or both). At four of the five sites, the insertion results in an AGY as the first downstream 0-frame codon. The frameshift insertion observed in vertebrates at approximate position 174 in the nad3 gene (FS174; Fig. 1) has AGY as the first two downstream 0-frame codons.

The likely site of the insertion in two of the ant examples (FS333 and FS411) is a glycine (GGA) codon. The site of the FS474 insertion appears to be a second or third position of tryptophan (UGA). We therefore focus on these codons and the tRNA genes responsible for their translation.

tRNA Genes

We sequenced the tRNA gene involved in translation of GGN (glycine) codons from two Polyrhachis species. One species, P.nr. abnormis, carries a frameshift at position 333, in a glycine codon. The other species, P. thusnelda, is apparently free of frameshift mutations. We were particularly interested in asking whether there were any differences between the glycine tRNAs that might account for alternative translation of these codons. Suppressor mutations of +1 frameshifts have been studied in Salmonella (Riddle and Roth 1970). In their study, at least four of six genes identified involve altered tRNA structures, and two of those are alterations in tRNAGly-GGG genes (Riddle and Carbon 1973). The tRNAGly sequences of the Polyrhachis species are compared in Fig. 2. In both species, the anticodon gives an exact Watson–Crick match to GGA codons. No differences in the anticodon loops were found. The few differences in other parts of the molecules give no clues concerning a possible mechanism for +1 frameshift read-through. This result is not surprising, of course, since this tRNA must correctly translate glycine codons throughout the mitochondrial protein coding genes.

Figure 2
figure 2

Sequences of mitochondrial tRNAGly, tRNAVal, and tRNAla genes in ants.

Full size image

We were interested in the valine and alanine tRNAs since GUA and GCA may be involved in the +1 translation at these sites. In the Polyrhachis mitochondrial genome, the tRNAVal anticodon in both of the species we sequenced is UAC (5′ → 3′), giving an exact codon–anticodon match to GUA valine codons (Fig. 2). The anticodon of tRNAAla in the two ant species we examined is UGC (Fig. 2), giving an exact match to GCA codons.

In most of the mitochondrial +1 frameshift examples in ants and vertebrates, codons AGY appear in the immediate downstream position(s). The tRNA responsible for decoding these codons is somewhat challenging to locate, since it lacks the DHU stem and does not form a standard cloverleaf structure (Clary and Wolstenholme 1985; Crozier and Crozier 1993; Stewart and Beckenbach 2003). Structures representing the minimum energy folds for this gene from P. phryne and Lasius alienus are shown in Fig. 3. These species differ by over 23% of nucleotide sites in the cox2 and cytb genes. Comparison of these sequences shows only a single substitution (G ↔ A transition) in the paired stem regions, yielding a G–U versus an A–U pair at the base of the anticodon stem. There are no changes in the anticodon loop, but the other unpaired regions are not alignable between these species. The apparent structural conservation between these fairly distant species and the appropriate UCU anticodon predicted from these structures indicate that these sequences are the genes responsible for decoding AGN codons in ants.

Figure 3
figure 3

Inferred secondary structures of tRNASer-AGN of two species of ants. Optimal folds for Lasius alienus (ΔG = −13.4 kcal/mol) and P. phryne (ΔG = −13.2kcal/mol) predicted by mfold (Zuker 2003).

Full size image

Discussion

We have presented evidence for single nucleotide frameshift insertions at four sites within the mitochondrial cytb gene of 12 species of the ant genus, Polyrhachis. The principal questions are whether these sequences are functional mitochondrial cytb genes, rather than pseudogenes, and if they are functional, what the likely mechanism for their translation is.

Evidence That the Frameshifted Sequences AreFunctional

Two of the insertion sites, FS333 and FS474, were observed in 10 and 2 species, respectively. Nine of the ten FS333 insertions are G. The possibility that these insertions are independent insertions of G at this site during the recent creation of a pseudogene in each species is remote. Thus we assume the FS333 sequences are descendants of a single sequence in a common ancestor of these 10 species. The evolution observable among these sequences must have occurred subsequent to the insertion event. Evolution of functional protein coding genes is easily distinguishable from evolution of pseudogenes. Thus the pattern of substitutions subsequent to the frameshift mutation provides clear evidence of their function. We use a similar logic for the FS474 sequences.

We compare the sequence evolution in the two genes, cox2, where no indels were observed, and cytb, in species with and without frameshift insertions. Table 3 gives the pairwise distances for these two genes, corrected for multiple substitutions using Jukes–Cantor. Across the genus, cox2 shows a mean of about 18% sequence divergence between species, while cytb shows about 21%. The variation observed in cytb is higher than that observed in cox2 in 393 of 435 comparisons. The pattern of substitution in both genes is similar, with a ratio of about 2:1:4 by codon position in the normal reading frame for both sequences. This pattern is typical of insect mitochondrial protein coding genes evolving under constraints required to maintain a full-length, functional protein product (Beckenbach and Borkent 2003).

Table 3 Mean pairwise Jukes–Cantor distances between Polyrhachis species
Full size table

The observed genetic distance between the two species carrying FS474, P. trapezoidea and P. phryne, is 13.4% for cox2 and 18.2% cytb (Table 3). Evidently a substantial amount of sequence evolution has occurred in both genes since these two species separated. If we “repair” the cytb gene by deleting the extra nucleotide from the data file to restore the original reading frame, and examine the pattern of substitution by codon position, it is evident that the two genes have a similar pattern of substitution. In particular, the substitution pattern in cytb is approximately 2:1:4. Thus the sequence evolution in cytb has continued in these sequences under constraints expected in the original reading frame.

A similar argument can be made for the ten species in the FS333 group (Table 3). Comparable bias of substitutions at the three codon positions is evident in both genes, suggesting that the sequences are the functional mitochondrial genes for cytb. The original reading frame appears to be maintained despite the frameshift insertion(s).

To rule out the possibility that a truncated portion of the gene, coding for either the amino-terminal or the carboxy-terminal regions, is evolving under constraints of a functional reading frame we also separate out the regions prior to the insertions and those beyond the insertions, in the original reading frame (Table 3). Similar substitution bias is evident both upstream and downstream from the insertion sites, for both FS333 and FS474. Therefore the entire gene is evolving as a functional protein coding gene requiring a shift to the +1 reading frame at the site of the insertion(s).

Mutational Effects of Frameshifts

There are several consequences of frameshift mutations that are distinct from those of replacement substitutions. Frameshift mutations affect not only the site of the mutation, but the translation of all residues downstream. A frameshift insertion results in a defective polypeptide product unless it is corrected either by processing of the mRNA to remove the frameshift or through some mechanism to read through the frameshift during translation, shifting to the correct reading frame. In the frameshifted sequences observed in this study, the predicted products would be truncated to less than half the length of the normal cytb protein if translated normally (Table 1). For the FS333 species, only the first 111 codons of the 372 required for a wild-type product would be translated correctly. In contrast, if a stall during translation causes a +1 translational frameshift at the site of the insertion, the entire 372-residue protein may be translated correctly, yielding a wild-type product. This result is quite different from the effect of a nonsynonymous nucleotide substitution, where translation of the mutated message produces no wild-type product.

In this context, it is important to note that under normal conditions, translation is error prone. A processivity error rate of 2.5 × 10−4 per codon has been estimated for wild-type E. coli under ideal laboratory growth conditions (Kurland 1992). For a protein of 372 residues, about 9% of products would fail to terminate correctly, through premature termination or read-through of the terminator. These errors include spontaneous translational frameshifts, which have been estimated to occur at about 5 × 10−5 per codon (Kurland 1992). Improperly terminated products are subject to proteolytic degradation and are part of normal gene function. The frequency of frameshift errors during translation is strongly sequence dependent. Specific sequences (“programmed frameshift sites”) may undergo translational frameshifting at rates three or four orders of magnitude higher than sequences at nonprogrammed sites and may exceed 0-frame translation (Farabaugh 1996; Gesteland and Atkins 1996). A frameshift mutation in a protein coding sequence that generates a compensating (programmed) translational frameshift will produce at least some wild-type product. The efficiency of the frameshift translation depends on both the sequence and the physiological state of the cell. Experimental manipulation of the abundance of the cognate tRNA for 0-frame decoding of the first downstream codon can alter the efficiency of translational frameshifting for certain sequences in yeast, from a few percent to almost 70% (Sundararajan et al. 1999).

As noted above, a unique feature of frameshift mutations, compared to missense mutations, is that any full-length products generated from a compensating translational frameshift are wild-type. That is, all full-length, properly terminated products should be fully functional, not mutant. Products produced when the translation fails to shift into the +1 reading frame will truncate prematurely and be subject to degradation. Thus the fitness effect depends on the efficiency of the programmed frameshift in taxa carrying a frameshift, relative to the accuracy of translation of the wild-type message in taxa with a full-length open reading frame.

A Model For Decoding of the FS333 Sequences

In all of the well-studied systems, including the TY1 and TY3 frameshifts in yeast, there are two requirements for relatively efficient translational frameshift-ing (Sundararajan et al. 1999). First, there must be a pause at the frameshift site during translation. Second, the tRNA bound at the peptidyl site (P-site), carrying the nascent polypeptide, must have a poor wobble pairing in the third codon position. The translational pause may be caused by slow decoding of the codon in the amino acyl site (A-site) or by the presence of stable secondary structure just downstream from the frameshift site (Ivanov et al. 1998). We found little evidence, however, for unusually stable secondary structure in the mRNA sequences downstream from the Polyrhachis frameshift sites. Thus we focus on the codon–anticodon interactions.

While the conditions noted above are required, the efficiency of a frameshift during translation may be enhanced if the tRNA bound at the P-site is able to slip to a +1 position having exact cognate recognition, followed by rapid decoding of the codon in the +1 position at the A-site (Hansen et al. 2003). Most of the Polyrhachis and vertebrate mtDNA frameshift sites meet these requirements. A model for decoding of the FS333 sites is depicted in Fig. 4. This model follows those presented in Sundararajan et al. (1999) and Hansen et al. (2003). In species without the frameshift, exact cognate recognition of the GGA (glycine) codon by tRNAGly (anticodon, UCC), followed by rapid decoding of the next codon (GUA) in the normal reading frame (0-frame), allows translation to proceed through the site without pause (Fig. 4A).

Figure 4
figure 4

Model for translational frameshifting: alternative outcomes of translation. A In species without the frameshift, cognate recognition at both the peptidyl (P) and the amino acyl (A) sites allows rapid continuation of translation in the current reading frame. B Species carrying the FS333 insertion require a near-cognate decoding of the GGG codon at the P-site, followed by near-cognate decoding of the poorly recognized AGU at the A-site. The resulting pause allows two competing paths. C Decoding of AGU at the A-site leads to a continuation in the 0 reading frame, resulting in a truncated product of only 165 residues. D Slippage of the ribosome to the +1 frame allows cognate decoding of the GGA at the P-site, and rapid cognate decoding of the GUA (valine), yielding a wild-type protein of 372 residues.

Full size image

In nine of the FS333 species, the frameshift site is GGG AGU A (Fig. 1). The GGG codon must be decoded by tRNAGly (anticodon, UCC). We hypothesize that slow decoding of the AGU codon in the A-site generates a pause (Fig. 4B) and sets up two competing pathways (Figs. 4C and D). If a charged tRNASer-AGN binds at the A-site, translation continues in the 0 reading frame, resulting in a defective polypeptide product (Fig. 4C). That product would be degraded. Alternatively, if the tRNAGly bound in the P-site slips to a +1 position prior to binding of a tRNA in the A-site, the next (+1) codon will be rapidly decoded by exact cognate recognition of tRNAVal (Fig. 4D). Translation can then continue in the +1 reading frame to produce a fully wild-type product. The fitness effect of the mutation must depend on the efficiency of the programmed frameshift—the proportion of translations that follow path 4D.

An almost-identical model can account for +1 decoding of the FS411 site (sequence GGG AGC A; Fig. 1), except that rapid cognate decoding of the +1 A-site requires tRNAAla instead of tRNAVal. Decoding of FS474 (sequence, UGG AGU A, in both species; Fig. 1) requires some modification of the model, since the +1 P-site codon (GGA) is not an exact match for tRNATrp. This site may be comparable to the frameshift site in the yeast TY3 retrotransposon. Alternative models have been proposed for decoding of the frameshift in TY3, where the tRNA bound at the P-site cannot slip to an exact cognate binding in the +1 position. Farabaugh et al. (1993) suggested that out-of-frame (+1) binding of an abundant tRNA at the A-site would allow frameshifted read-through of this sequence without P-site slippage, while Hansen et al. (2003) argued that slippage was the likely mechanism even when the tRNA bound at the P-site can form only two Watson–Crick base pairs at the +1 codon. Either of these models can explain translational frameshifting at the FS474 site.

A critical part of the model is the role of the tRNAs. For efficient +1 translational frameshifting, the tRNA that has already bound and shifted to the P-site (which therefore cannot be exchanged), must have a suboptimal codon–anticodon pairing. The codon in the A-site at that stage in translation must be slowly recognized to generate a stall. This second requirement has been experimentally studied by Vimaldithan and Farabaugh (1994) in yeast, where elimination of the tRNAArg-AGG gene required for cognate recognition of the AGG codon in the A-site increased the efficiency of programmed frameshifting by more than threefold. Thus translation pauses until a “near-cognate” tRNA binds or a +1 frameshift allows translation to continue in the new reading frame. In insect mitochondria there are only 22 tRNAs to decode 62 sense codons. Since most (or all) of these sense codons appear in mitochondrial genes, suboptimal (near-cognate) recognition is actually more common than exact cognate pairing (see Table 4 for examples). For the translation of nuclear genes there is a correlation between the availability of charged tRNAs and the number of copies of each tRNA gene in the genome (Percudani et al. 1997). In animal mitochondrial genomes there is exactly one copy of each tRNA gene per genome. Thus we cannot fall back on stoichiometry at the genic level to define “hungry” (slowly decoding) codons. Instead we have to look at relative codon usage under the assumption that codon usage reflects translational efficiency. Table 4 gives codon usage for codon families relevant to these sites for the cox2 and cytb genes of Polyrhachis and for all genes in two vertebrates. The codons that are decoded by exact cognate binding (where known) are in boldface.

Table 4 Codon usage in Polyrhachis and representative vertebrates
Full size table

The first point to note is that serine is more commonly coded by UCN rather than AGN (about a five fold difference in Polyrhachis cox2 and cytb). We also note that GGG (glycine—FS333 and FS411) and UGG (tryptophan—FS474) are rarely used. The frameshift sites themselves present an interesting contrast to these general observations. Among FS333 species there are 15 glycine (GGN) codons within the region of cytb studied here (Table 5). Across these species, the third position is most often A (53%), giving exact cognate recognition by the Polyrhachis glycine tRNA. There is, however, considerable variation at 14 of these sites among the FS333 species. The exception is the frameshift site itself (FS333). At this site (and at glycine FS411) the cognate codon (GGA) is not present in any of the sequences. Nine of the ten FS333 have GGG and the other, P. sexspinosa, carries GGU at the FS333 site and GGG at the FS411 site. Under the model for efficient programmed translational frameshifting (Fig. 4) a third position mismatch is required at these sites. A simple G → A transition would eliminate that mismatch and strongly favor 0-frame translation (Fig. 4C). This mutation would likely not be viable and is not observed at any of the sites. Evidently the conservation of a mismatch at the third position is required for efficient decoding of these cytb genes.

Table 5 Third position variation at GGN codons in 30 species of Polyrhachis
Full size table

The same argument can be made for the FS474 site (Table 6). A third position G is generally avoided in the UGR codons at the 12 sites in the two genes examined here. Both FS474 species have UGG at the frameshift site. Although we did not sequence the tRNA responsible for decoding tryptophan in ants, in most insects and vertebrates the anticodon is UCA, with cognate recognition of UGA codons. The conservation of UGG where frameshifting is required for translation of these genes is consistent with the requirement for “near-cognate” codon–anticodon interactions at the P-site at this position.

Table 6 Third position variation at UGR codons in 30 species of Polyrhachis
Full size table

Fitness Considerations

In all of the well-studied cases of translational frameshifting, the requirement for a frameshift during translation appears to serve a functional role: to down-regulate the gene except under certain physiological conditions. These examples include the vertebrate Ornithine Decarboxylase Antizyme gene, the E. coli Release Factor 2 gene, and the TY1 and TY3 retrotransposons in yeast (reviewed by Farabaugh 1996; Gesteland and Atkins 1996). Gurvich et al. (2003) recognized that the sequences that promote elevated levels of translational frameshifting may occur by chance in genes that do not require a frameshift. In these cases, any frameshift products would be nonfunctional, and the presence of such sequences may actually reduce fitness. They showed that such sequences undergo significant levels of translational frameshifting and are underrepresented E. coli genes. An alternative coding for the same amino acid sequence is overrepresented. The implication is that these sequences generate high levels of frameshifted product, whether desirable or not.

It is difficult to see a functional role for programmed frameshifts in animal mitochondrial genes. The same set of genes is present in nearly all animal mitochondrial genomes. The only examples that have been identified, where a translational frameshift is required to translate these genes, are in the nad3 gene of birds and a turtle and the cytb gene of ants. These cases appear to be exceptional. As well, there is phylogenetic evidence that the frameshift insertion has been lost multiple times during evolution. These observations argue against a functional role for the frameshifts we have observed. We suggest, instead, that certain sequence contexts can tolerate a single nucleotide frameshift insertion, without excessive harm. Since the mitochondrial genome is haploid, the organism and her descendants must live with it. For eusocial insects, such as ants, that includes the queen, her colony, and any descendant colonies.

Conclusions

The discovery of a single nucleotide +1 frameshift insertion at a single site in many birds and a turtle (Mindell et al. 1998) confronts us with two apparently unlikely alternatives: either (1) the mutation arose as a unique event in a common ancestor of birds and turtles or (2) identical deleterious mutations have been fixed two or more times independently during evolution. In the first case, we must infer that the mutation has been retained in many birds and a turtle for at least 250 million years, despite apparent harmful effects. The alternative appears even less likely: the fixation of a deleterious insertion at precisely the same site in a molecule of more than 16 kb, in two or more lineages independently.

Since the vast majority of frameshift mutations in essential genes are rejected by selection, the sites where a +1 insertion produces a viable result must be limited to sequences where the gene can still be correctly translated. We scanned the protein coding sequences of the ostrich (Haddrath and Baker 2001; GenBank Accession NC 002785) and painted turtle (Mindell et al. 1999; NC 002073) for occurrences of consecutive codons of NNA GTA GYN, where a single nucleotide insertion prior to the first A will generate consecutive AGU AGY codons immediately downstream in the transcribed mRNA. Aside from the nad3 site present in the putative premutation sequence, there are seven occurrences in the ostrich and eight in the turtle. If these are the only sites that can generate a viable frameshift mutation, then frameshifts occurring at all other sites will be filtered out by evolution.

The evidence presented here shows that four different sites in cytb of ants have independently sustained +1 insertions in the relatively brief history of the genus Polyrhachis. Evidently +1 mutations do arise from time to time, but the only ones we can observe will be those that can be correctly translated to produce an adequate level of full length, functional product.

A corollary to our hypothesis that consecutive AGU AGY codons can generate a high incidence of +1 frameshift products during translation in birds and turtles is that this sequence should be avoided in all mitochondrial genes. We examined all of the protein coding genes of both ostrich and turtle sequences. This sequence occurs in frame in only one place, in the ostrich nad4l gene. A +1 frameshift product of this gene would be severely truncated and probably not functional.

An unexpected result of our phylogenetic study of Polyrhachis was the finding of not just one, but multiple sites carrying a +1 frameshift insertion in different species of a single genus. We suggest that frameshift mutations continuously arise, but are rarely fixed. Ohta (1977) has shown that the likelihood of fixation of mildly deleterious mutations is inversely proportional to the effective population size. Eusocial insect groups, such as ants of the genus Polyrhachis, with an inherently small effective population size (Pamilo and Crozier 1997; Wilson 1963), may be much more susceptible to the fixation of deleterious, but survivable, mutations. Thus close study of these species may provide us with a wealth of examples that demonstrate the limits and constraints of evolution at the molecular level.

Supplementary Material

The sequences for P. turneri have been deposited in GenBank under accession numbers AY437886 (cytb) and AY437887 (cox2); partial sequences of the two genes for the other 29 Polyrhachis species are available from GenBank under accession numbers AY442226–AY442254 (cytb) and AY443390–AY443418 (cox2). Voucher specimens are retained in the collections of the Queensland Museum and S.K. Robson.