Translational Efficiency of Redundant ACG Initiator Codons Is Enhanced by a Favorable Sequence Context and Remedial Initiation*

Earlier studies showed that the redundancy of ACG initiation codons enhanced the efficiency of translation initiation by 3- to 6-fold. Evidence presented here shows that this “redundancy effect” can be attributed to a favorable sequence context and, to a lesser extent, remedial initiation. In the case of redundant ACG initiator codons, the second ACG not only acts as a remedial initiation site for scanning ribosomes that skip the first ACG but also enhances the activity of the preceding initiator by providing a preferable “A” at its relative +4 position. Hence, non-successive ACG codons can be as effective as successive ACG codons in initiation, if positioned within a similar context. In contrast, redundant GUG initiation codons (GUG/GUG) bear an unfavorable “G” nucleotide at both the +4 and –3 positions relative to the first and second GUGs, respectively, such that redundant GUG codons act more poorly as translation initiation sites than does a single GUG with a favorable “A” nucleotide in the +4 position (∼2.5-fold). Thus, the sequence context plays a much more important role than remedial initiation in modulating the efficiency of translational initiation from redundant non-AUG codons.

Aminoacyl-tRNA synthetases are a group of primordial enzymes, each of which catalyzes the attachment of a specific amino acid to its cognate tRNAs. Aminoacyl-tRNAs are then delivered by elongation factor-1 to ribosomes for protein translation. Typically there are 20 aminoacyl-tRNA synthetases in prokaryotes, one for each amino acid (1)(2)(3)(4). In eukaryotes, protein synthesis occurs not only in the cytoplasm, but also in organelles, such as mitochondria and chloroplasts (5). Thus, eukaryotes, such as yeast, need two distinct sets of enzymes for each aminoacylation activity, one localized to the cytoplasm and the other to the mitochondria. Each set aminoacylates the isoaccepting tRNAs within its respective cell compartment and is sequestered from the isoacceptors confined in other com-partments. In most cases, cytoplasmic and mitochondrial synthetase activities are encoded by two distinct nuclear genes, regardless of the cell compartments to which they are confined. However, two Saccharomyces cerevisiae genes, HTS1 (the gene encoding histidyl-tRNA synthetase) (6) and VAS1 (the gene encoding valyl-tRNA synthetase (ValRS) 3 ) (7), specify both the mitochondrial and cytosolic forms through alternative initiation from two in-frame AUG codons. Each of these genes encodes more than one mRNA, and the mRNA species produced differ only at their 5Ј-ends. The mitochondrial form of the enzyme is translated from the first AUG on "long" mRNAs, while the cytosolic form is translated from the second AUG on "short" mRNAs, the 5Ј-ends of which are located between the first and second AUG initiator codons. Hence, mitochondrial enzymes have the same polypeptide sequences as their cytosolic counterparts, except for a short N-terminal mitochondrial targeting sequence. The transit peptide is subsequently cleaved away upon being imported into mitochondria. Because the isozymes are targeted to different compartments, the two isoforms of ValRS, for example, cannot be substituted for each other in vivo (7,8). A similar scenario has been observed for genes encoding the mitochondrial and cytoplasmic forms of Arabidopsis thaliana alanyl-tRNA synthetase (AlaRS), threonyl-tRNA synthetase, and ValRS (9). Paradoxically, although two isoforms of AlaRS are generated in a similar manner in Candida albicans, the longer form per se appears to be dualtargeted and is thus functional in both compartments (10). Recently, two rare cases of one gene encoding both activities have been reported in yeast in which the protein isoforms are produced by alternative use of two in-frame initiation codons: an upstream non-AUG initiator and a downstream AUG initiator (11,12).
Translation initiation in eukaryotes is a stringent process requiring not only initiator tRNA but also many protein factors, including eukaryotic initiation factors eIF1, eIF2, eIF3, eIF4F, and eIF5. Upon binding to the cap structure, the 43 S preinitiation complex, composed of the 40 S ribosome and initiation factors, moves along the mRNA in a 3Ј direction until it encounters the first AUG codon. At this point, GTP hydrolysis that leads to dissociation of the eIF2-GDP complex from the initiator tRNA in the preinitiation complex and subsequent P i release signifies a 3-bp codon-anticodon interaction between Met-tRNA(i)(Met) and the start codon (13). It has recently been shown that eIF1 plays a critical role in start codon selection. Mutations of eIF1 may lead to reduced interaction of this initiation factor with 40 S subunits and increased initiation at UUG codons (14).
Previous studies on CYC1 (15) in S. cerevisiae suggested that AUG is the only codon recognized as a translational initiator, and that the AUG codon nearest the 5Ј-end of mRNA serves as the start site for translation. If the first AUG codon is mutated, then initiation can begin at the next available AUG from the 5Ј-end of the message. The same rules apply to all eukaryotes. However, there are many examples in higher eukaryotes, where cellular and viral mRNAs can initiate from "non-AUG" codons that differ from AUG by just one nucleotide (16). The relatively weak base pairing between a non-AUG codon and the anticodon of an initiator tRNA appears to be compensated for by interactions with nearby nucleotides, in particular a purine (A or G) at position Ϫ3 and a "G" at position ϩ4 (17,18). Similarly, mutations in the sequence region surrounding the first AUG can lead to its inefficient utilization as an initiator and subsequent use of AUG at a downstream location. In addition to the sequence context, a stable hairpin structure located 12-15 nucleotides downstream from the initiator can also facilitate recognition of a weak initiator by the 40 S ribosomal subunit (19).
Although some reports have suggested that sequences immediately preceding the initiation codon may also play a role in modulating the efficiency of AUG translation initiation in yeast, the magnitude of this context effect appears relatively insignificant (20 -22). Perhaps for that reason, yeast cannot efficiently use non-AUG codons as the translation start site (23,24). Nonetheless, three yeast genes, GRS1 (one of the two glycyl-tRNA synthetase (GlyRS) genes in S. cerevisiae) (11), ALA1 (the only alanyl-tRNA synthetase (AlaRS) gene in S. cerevisiae) (12), and CARP2A (the gene coding for acidic ribosomal protein P2A in Candida albicans) (25), have recently been reported to use naturally occurring non-AUG triplets as translation initiators. Moreover, a very recent study suggested that the translational efficiency of non-AUG initiation codons is significantly affected (up to 32-fold) by nucleotides at its relative Ϫ3 to Ϫ1 position, and, to a lesser extent, ϩ4 (26). The nucleotide at position Ϫ3 is the most influential one. AARuugA (R represents A or G; uug represents a non-AUG initiation codon) appears to be the most favorable sequence context for a non-AUG initiation site (26).
In the case of CARP2A, a non-AUG codon, i.e. GUG, serves as the exclusive translation initiator, whereas in the cases of ALA1 and GRS1, non-AUG codons act as alternative translation initiators that are accompanied by a downstream in-frame AUG initiation codon. Although two homologous GlyRS genes have been identified in the yeast genome (GRS1 and GRS2), only GRS1 is functional by possessing both cytoplasmic and mitochondrial aminoacylation activities, while GRS2 appears to be dispensable for growth (27). Further studies indicated that two functionally exclusive protein isoforms are alternatively generated from GRS1. A short form that is responsible for the cytoplasmic activity of GlyRS is translationally initiated from a classic AUG initiator, whereas a longer isoform that provides the mitochondrial activity is initiated from an upstream in-frame UUG codon (11). Expression of ALA1 essentially follows a similar mechanism. However, it is noteworthy that the mitochondrial form of AlaRS is initiated from two successive in-frame ACG codons 69 nucleotides upstream of the AUG initiator of the cytoplasmic form (12,28). Further studies showed that redundant ACGs contain stronger initiation activity than does a single ACG and can functionally substitute for the alternative AUG initiator codons of VAS1 (coding for mitochondrial and cytoplasmic isoforms of ValRS) in vivo. This feature of redundancy of non-AUG initiator codons may in itself represent a novel mechanism to improve the overall efficiency of a poor initiation event (29). So far, it is not clear whether this mechanism is a general feature of all possible non-AUG initiation codons. Moreover, because the sequence context also plays a role in modulating the efficiency of translation initiation from a non-AUG codon (26), redundancy of non-AUG initiation codons may run into a situation in which the initiation codons are flanked by an unfavorable sequence context imposed by themselves. For example, redundant GUG codons (GUG/ GUG) bear an unfavorable nucleotide G at both the ϩ4 and Ϫ3 positions relative to the first and second GUGs, respectively. Hence, the redundancy of GUG codons might not improve the efficiency of translation at all, but instead may impair the initiating activity of the preceding GUG codon and the overall efficiency of translation. In the work described here, we tested this hypothesis by comparing the translational efficiencies of several pairs of redundant and single non-AUG initiation codons and determined the effect of the sequence context on the translational activity. Our results suggested that a redundancy of non-AUG initiation codons does not always significantly enhance the translational efficiency and does so only when the first nucleotide of the initiation codons is an "A."

EXPERIMENTAL PROCEDURES
Construction of Plasmids-Various ALA1-lexA fusions for the Western blot analysis were constructed as previously described (11,29). Briefly, an initiator mutant of lexA was amplified by PCR as an SpeI-XhoI fragment and cloned in appropriate sites of pADH (11). To construct various ALA1-lexA fusions, a wild-type or mutant ALA1 sequence containing base pairs Ϫ105 to Ϫ24 relative to ATG1 was amplified by PCR as a PstI-SpeI fragment and cloned in-frame into the 5Ј-end of the lexA mutant. The expression of these constructs was under the control of a constitutive ADH promoter (30). Note that, because the native initiator codon of lexA was rendered inactive, the only translational initiation sites for these fusion constructs are those which exist in the ALA1 portion.
To generate mutations in the 5Ј leader region of the ALA1 gene, a short ALA1 sequence containing bp Ϫ300 to ϩ60 (relative to ATG1) was PCR-amplified from the full-length ALA1 clone as an EagI-XbaI fragment and cloned in appropriate sites of pBluescript II SK (Stratagene, La Jolla, CA). This construct was subsequently used as a template for mutagenesis. Mutagenesis was carried out following standard protocols provided by the manufacturer (Stratagene). After mutagenesis, the ALA1 fragment carrying the desired mutation was recovered from the plasmid by EagI/XbaI digestion and then fused in-frame to the 5Ј-end of a truncated version of ALA1 lacking the EagI-XbaI segment, yielding a full-length ALA1 clone. Note that the XbaI site is a native restriction site between bp ϩ55 and ϩ60 of ALA1. To generate mutations in the 5Ј leader region of the VAS1 gene, a similar approach was followed (29).
Complementation Assays for the Mitochondrial Function of VAS1-The yeast VAS1 knockout strain, CW1, was previously described (8). To test the mitochondrial function of the wildtype and mutant VAS1 constructs, CW1 was first transformed with a test plasmid (carrying a LEU2 marker) and plated onto selection medium lacking uracil and leucine. Following 5Ј-fluoroorotate selections, a single colony of transformants was selected and grown to the stationary phase in synthetic medium lacking leucine. Starting from a cell density of 1.0 A 600 , cultures were 10-fold serially diluted, and 5-l aliquots of each dilution were spotted onto the designated YPG plates. The plates were incubated at 30°C for 3-5 days. Photos were taken of the complementation assays at day 3 following incubation. Because a yeast cell cannot survive on glycerol without functional mitochondria, the transformants did not grow on the YPG plates unless a functional mitochondrial ValRS was provided by the test plasmid. Complementation assays for the mitochondrial function of ALA1 followed a similar protocol (12).
Western Blot Analysis-The protein expression patterns of the ALA1-lexA fusions were determined by a chemiluminescence-based Western blot analysis. The lexA fusion constructs that were expressed under the control of a constitutive ADH promoter were first transformed into INVSc1 (Invitrogen), and the transformants were subsequently grown in selection medium lacking leucine. Total protein extracts were prepared from each transformant with a buffer containing 50 mM Tris (pH 7.4), 150 mM NaCl, 0.5% SDS, 0.5% Triton X-100, 10 mM EDTA, and 1 mM phenylmethanesulfonyl fluoride. Aliquots of the protein extracts (10 -40 g) were loaded onto a gel containing 10% polyacrylamide and electrophoresed at 100 V for 1-2 h. Following electrophoresis, the resolved proteins were transferred using a semi-dry transfer device to a polyvinylidene difluoride membrane in a buffer containing 30 mM glycine, 48 mM Tris base (pH 8.3), 0.037% SDS, and 20% methanol. The membrane was probed with a horseradish peroxidase-conjugated anti-LexA antibody (Santa Cruz Biotechnology, Santa Cruz, CA) and then exposed to x-ray film following the addition of the appropriate substrates. Data were obtained from at least three independent experiments (only one blot is shown in each figure), and the relative intensity of the protein bands shown is presented as a percentage using the mean value of the intensity of the protein band translated from the native initiator (or its equivalent) as a reference.
Degradation Assay-To determine the turnover of the fusion proteins, analogous ALA1-lexA constructs were made and expressed under the control of an inducible GAL1 promoter. Transformants carrying these constructs were first grown in medium lacking leucine with 2% raffinose to a cell density of ϳ1.0 A 600 and then induced with 2% galactose for 1 h. Afterward, the cells were washed twice and then grown in medium lacking leucine with 2% glucose and 100 g/ml cycloheximide. Cells were harvested at 0, 2, 4, 8, 16, and 32 h postinduction and lysed. Forty-microgram samples of the cellular lysates were resolved on 10% polyacrylamide and electrophoresed at 100 V for 1-2 h, and the proteins were transferred to a polyvinylidene difluoride membrane and immunoblotted with the appropriate antibody.
RT-PCR-To determine the relative levels of specific ALA1-lexA mRNAs derived from these fusion constructs, a semiquantitative reverse-transcription (RT)-PCR experiment was carried out following the protocols provided by the manufacturer (Invitrogen). Total RNA was first isolated from the transformant, and aliquots (ϳ1 g) of the RNA were then reversetranscribed into single-stranded complementary DNA (cDNA) using an oligo(dT) primer. After RNase H treatment, the singlestranded cDNA products were amplified by PCR using a pair of specific primers. The forward and reverse primers contained sequences complementary to nucleotides Ϫ90 to Ϫ70 of ALA1 (5Ј-TATGAAAGCAGTTTGATTGAA-3Ј) and nucleotides ϩ370 to ϩ390 of lexA (5Ј-CAAGTCACCATCCATAATGCC-3Ј), respectively. To obtain more reliable data, two different cycle numbers of PCR amplification were carried out for each cDNA preparation as indicated in the figure. As a control, the relative levels of actin-specific mRNAs in each preparation were also determined using a set of primers complementary to nucleotides ϩ537 to ϩ560 (5Ј-ACCAACTGGGACGATATG-GAAAAG-3Ј) and nucleotides ϩ696 to ϩ719 (5Ј-TTGGATG-GAAACGTAGAAGGCTGG-3Ј) of actin, respectively, and only the cDNA products of 21 cycles of PCR amplification are shown.
␤-Galactosidase Assay-Yeast cells were pelleted by centrifugation at 12,000 ϫ g for 30 s and resuspended in 100 l of breaking buffer (100 mM Tris-HCl (pH 8.0), 1 mM dithiothreitol, 10% glycerol, and 2 mM phenylmethanesulfonyl fluoride) and 100 l of beads. Cells were then lysed at 4°C using a bead beater, followed by centrifugation at 12,000 ϫ g for 2 min. Aliquots of the supernatants (25 g) were diluted to 0.8 ml with Z buffer (60 mM Na 2 HPO 4 , 40 mM NaH 2 PO 4 , 10 mM KCl, 1 mM MgSO 4 , and 50 mM 2-mercaptoethanol). ␤-Galactosidase activity assays were initiated (at 37°C) by adding 0.2 ml of o-nitrophenyl ␤-D-galactoside (4 mg/ml). The reaction mixtures were incubated with constant shaking at 37°C for 20 min and then terminated by the addition of 0.4 ml of 1 M Na 2 CO 3 . The reaction mixtures were centrifuged at 12,000 ϫ g for 2 min, and the absorbance (A 420 ) of the supernatants was determined. Relative ␤-galactosidase activities were calculated from A 420 readings normalized to protein concentrations.

Translational Efficiencies of Redundant versus Single Non-AUG Initiation
Codons-Our previous study showed that two distinct protein isoforms are generated from the yeast ALA1 gene via alternative use of two in-frame initiation sites. The longer form that accounts for the mitochondrial activity is initiated from ACG(Ϫ25)/ACG(Ϫ24) (the numbers "-25" and "-24" in parenthesis refer to the 25th and 24th codon triplets that precede AUG1), whereas the short form that is responsible for the cytoplasmic activity is initiated from AUG1 (11). Further study suggested that the redundancy of ACG initiation codons may represent a novel mechanism to enhance the over-all efficiency of translation (29). To investigate whether this is a common feature of all possible non-AUG initiation codons (that differ from AUG by a single nucleotide), redundant ACG initiation codons were mutated to various redundant or single non-AUG initiation codons (Fig. 1A), and the translational efficiencies of the resultant constructs were tested. We used an initiator mutant of lexA as a reporter to monitor the relative initiating activities and thus levels of expression of various ALA1-lexA fusions (see "Experimental Procedures") (12). The first nucleotide of all initiator candidates chosen for this assay differed (Fig. 1A), so that we could simultaneously examine "remedial initiation" and the "context effect" (particularly the Ϫ3 and ϩ4 nucleotides relative to the second and first initiator codons, respectively). Consistent with previous observations (29), Fig. 1B shows that redundant ACG codons acted better than ACG/ACC and ACG/TCG as a translational initiation site by 2.5-and 6.0-fold, respectively (Fig. 1B, numbers 1-3). In contrast to the case scenario of ACG initiation, redundant GTG codons did not exhibit the expected "redundancy effect" at all. Instead, the translational efficiency of GTG/GTG was almost equivalent to that of a single GTG (GTG/GTC) with an unfavorable G nucleotide in its relative ϩ4 position, and 2.5-fold lower than that of a single GTG (GTG/ACC) with a favorable "A" nucleotide in its relative ϩ4 position (Fig. 1B, numbers 4 and 5). This result suggested that the second GTG in redundant GTG codons is essentially inactive as a remedial initiation site, probably due to the presence of an unfavorable G nucleotide in its relative Ϫ3 position (Fig. 1B, numbers 4 and  6). As a result, a single GTG codon with a favorable A nucleotide in its relative ϩ4 position functioned better as a translational initiation site than did redundant GTG codons (Fig.  1B, numbers 4 and 5). Similar to the case of redundant GTG codons, redundancy of TTG initiation codons failed to enhance the protein yield (Fig.  1, A and B, numbers 7-9). The initiating activity of TTG/ACC was slightly higher than that of redundant TTG codons (ϳ1.5fold) (Fig. 1B, numbers 7 and 8). Strangely enough, the initiating activity of TTG/TTC was also slightly higher than that of redundant TTG codons (ϳ1.5-fold) (Fig. 1B, numbers 7 and 9). Also, TTG/TTC and TTG/TTT expressed similar levels of proteins (Fig. 1E, numbers 9 and 13). Note that both TTT and TTC code for Phe. Although this result suggests that the second amino acid may affect the stability of the protein product (31), the possibility that the nucleotide at position ϩ6 plays a minor role in modulating the efficiency of translation initiation cannot be ruled out at the moment. As for CTG initiation, the initiating activity of redundant CTG codons was equivalent to that of a single CTG codon with a favorable ϩ4 A nucleotide, and was 2-fold higher than that of a single CTG with an unfavorable ϩ4 C nucleotide (Fig. 1, A and B, numbers 10 -12). Therefore, it appears that a favorable nucleotide at position ϩ4 is just as good as redundancy of initiator codons in this instance. To check whether the ALA1-lexA fusions shown in Fig. 1B expressed similar levels of mRNAs, a semiquantitative RT-PCR was conducted. Two different cycle numbers of PCR amplification were carried out for each construct and compared. Fig. 1C shows that similar levels of cDNA products were generated from these fusions, suggesting that these mutations did not impair the stability of the specific mRNAs in vivo.
To verify whether the data of Western blots shown in Fig. 1B truthfully reflect the translational efficiencies of the various initiation codons, we next used lacZ as a reporter gene. The lexA portion of constructs 1-6 used in Fig. 1B was replaced by a lacZ reporter gene where the native initiator codon ATG had been mutated to ACT. Therefore, the redundant ACG initiator codons (or their derivatives) in the ALA1 portion became the only initiation sites for the translation of the AlaRS-LacZ fusion proteins. To examine the initiating activities of various initiation codons, aliquots of the soluble protein extracts were prepared from the transformant and assayed for their ␤-galactosidase activity. As shown in Fig. 1D, redundancy of ACG initiator codons significantly enhanced the enzymatic activity (2-to 3-fold) (Fig. 1D, numbers 14 -16), while the redundancy of GTG initiation codons did not (Fig. 1D, numbers 17-19). In fact, redundant GTG codons exhibited enzymatic activity that was ϳ2-fold lower than that of a single GTG with a favorable ϩ4 nucleotide "A." This result is basically consistent with that of the Western blots shown in Fig. 1B. Thus, redundancy of non-AUG initiation codons does not always enhance the overall translational efficiency, and the redundancy effect manifested by the ACG initiator codons is in effect attributed to both a context effect and remedial initiation.
Translational Efficiency versus Turnover Rate-The steadystate levels of the AlaRS-LexA fusions are determined not only by their translational efficiencies but also by their turnover rates. Many mutations shown in Fig. 1 involved changes in the second amino acid residue. To investigate whether the second amino acid has a strong impact on the turnover or stability of the fusion proteins, we inserted AUG into the initiation site for some of the constructs to minimize the effect of sequence con-text (26) and then subcloned these fusions into a yeast shuttle vector with an inducible GAL1 promoter. Transformants carrying these constructs were first grown to a cell density of ϳ1.0 A 600 in medium lacking leucine with 2% raffinose and then induced with 2% galactose for 1 h. Afterward, the cells were washed twice and then grown in medium lacking leucine with 2% glucose and 100 g/ml cycloheximide. Aliquots of the cell cultures were harvested at 0, 2, 4, 8, 16, and 32 h postinduction and prepared for Western blotting. As shown in Fig. 2, these fusions expressed a similar level of protein after induction (see T 0 in Fig. 2B), and the second residue, Ser, Val, Leu, Thr, or Phe, had little effect on the turnover of the fusion proteins. The proteins having Ser, Val, Leu, Thr, or Phe as their second amino acid retained an almost constant level throughout the time period tested. Thus, changes in protein levels shown in Fig. 1 are a valid readout of initiation efficiency. It should be noted that the amino acids tested in Fig. 2 encompass all the penultimate N-terminal residues that appeared in Fig. 1.

Efficiencies of Translation from Various Combinations of Non-AUG Initiation Codons-
We have so far assayed the translational efficiencies of several pairs of redundant and single non-ATG initiation codons and shown that redundant non-ATG initiation codons are not always stronger than a single non-ATG initiation codon. In fact, ACG appears to be the only initiator codon among the four candidates that showed a strong redundancy effect. To provide a more in-depth picture on this characteristic, we next tested the translational efficiencies of various combinations of non-ATG initiator codons such as ACG/TTG, ACG/ATT, TTG/TTG, and TTG/ACG. These non-ATG initiation codons were individually introduced into the ALA1 sequence by mutagenesis to replace the native ACG/ ACG initiator codons (Fig. 3A), and the protein yields of the resultant constructs were assayed (Fig. 3B). As shown in Fig. 3, ACG/ACG and ACG/ATT possessed higher initiating activity than did ACG/TTG (2-to 3-fold) (numbers 1-3). Similarly, TTG/ACG and TTG/ATT possessed higher initiating activity than did TTG/TTG (ϳ1.5-fold) (numbers 4 -6). Despite the fact that TTG per se is a stronger initiation site than ACG in a similar context (Fig. 1B, numbers 2 and 8), TTG was not preferred as a remedial initiation site over ACG (Fig. 3). In conjunction with the results shown in Fig. 1, this result strongly favored the notion that the nucleotide at position ϩ4 relative to the preceding non-ATG initiator codon plays an important role in modulating the efficiency of redundant non-ATG initiation codons. In other words, the nature of the first nucleotide of the remedial initiator codon is far more important than the codon itself in modulating the efficiency of translation in these instances.
Effect of Nucleotide at Position Ϫ3 on Non-AUG Initiation-In Figs. 1 and 2, we demonstrate the effect of the nucleotide at position ϩ4 on the translational activity of a non-AUG initiator codon. However, the effect of the nucleotide at position Ϫ3, presumably the most influential position, on the translational activity of a non-AUG initiator codon has not yet been fully illustrated in the ALA1 gene. To further address this issue, we next mutated the Ϫ3 nucleotide (Fig. 4A) and tested its effect on the translational efficiency (Fig. 4B). As shown in Fig. 4, mutation of Ϫ3 A to T, G, and C reduced the initiation activity by 3.2-, 1.6-, and 2.2-fold, respectively (Fig. 4, numbers 1-4). This result suggested that the nucleotide at position Ϫ3 is important to the ACG initiator codons, and that A is the most favorable nucleotide at this position. Nevertheless, this effect was not as profound as anticipated for nucleotide changes at Ϫ3 (26). Perhaps that is because the overall initiation was mediated by redundant ACG initiator codons, and changes at the nucleotide position Ϫ3 affected mainly, if not only, the first initiator codon. To test this hypothesis and also extend our assay to other non-ATG initiator codons, we next tested the effect of the nucleotide at this position on a single GTG initiator codon. As expected, the effect of the nucleotide at position Ϫ3 on a single GTG initiator codon was much stronger than on redundant ACG initiator codons. Mutation of the Ϫ3 nucleotide from A to T, G, and C lowered the protein yields by 2.6-, 3.5-, and 12-fold, respectively (Fig. 4, numbers 5-8). Together, these results suggested that the nucleotide at position Ϫ3 plays a key role in modulating the efficiency of a non-AUG initiator. So, it is conceivable that in cases of redundant non-AUG initiator codons, the Ϫ3 and ϩ1 nucleotides (relative to the first initiator codon) largely determine the efficiency of the preceding and remedial initiator codons, respectively.
Translational Efficiency of Three Successive ACG Initiation Codons-Because the translational efficiency of an ACG initiator codon can be enhanced by redundancy of the initiation codons, we wondered whether this effect could be further extended by inserting more in-frame ACG codons. To investigate this possibility, the codons preceding and following the native ACG initiator codons were individually mutated to an ACG codon and the translational efficiencies of the resultant constructs were tested. As shown in Fig. 5, whereas mutation of TCA(Ϫ23) to ACG (resulting in pSJ350) enhanced the protein yield by ϳ1.3-fold, mutation of AAG(Ϫ26) to ACG (resulting in pSJ349) slightly reduced the protein yield by ϳ1.3-fold. At first glance, these two outcomes appeared to be contradictory to each other. However, further inspection of the nucleotide sequences surrounding the ACG initiation codons suggests that these outcomes were actually caused by different sequence contexts and are still consistent with other examples shown in Figs. 1 and 3. In the case of pSJ349, mutation of AAG(Ϫ26) to ACG created only a poor initiation codon (ACG(Ϫ26)) with an unfavorable G nucleotide at its relative position Ϫ3 and, more importantly, provided an unfavorable nucleotide C at position Ϫ2 relative to ACG(Ϫ25) (26), such that ACG/ACG/ACG (codons Ϫ26 to Ϫ24) did not function better than ACG/ACG (codons Ϫ25 to Ϫ24) in translational initiation (Fig. 5, numbers  1 and 2). Conceivably, context effect plays a predominant role in modulating the translational efficiency of triple ACG codons in this case. In contrast, in the case of pSJ350, mutation of TCA(Ϫ23) to ACG changed the nucleotide at position ϩ4 relative to ACG(Ϫ24) from T to a preferable A and created a potential initiation site (ACG(Ϫ23)) with a favorable A nucleotide at both the relative Ϫ3 and ϩ4 positions. Hence, ACG/ ACG/ACG (codon Ϫ25 to Ϫ23) acted stronger than ACG/ ACG (codon Ϫ25 to Ϫ24) as a translational initiation site (Fig.  5, numbers 1 and 3). Translational Efficiency of Non-successive ACG Initiation Codons-In the case of redundant ACG initiator codons, the second ACG codon not only acts as a remedial initiation site, but also provides a preferable nucleotide A for the preceding ACG at its relative ϩ4 position ( Fig. 1) (29). This observation promoted us to ask whether an in-frame, non-successive ACG triplet can also serve as a remedial initiation site and enhance the overall efficiency of translation. To test this possibility, an ACG codon nine nucleotides downstream of ACG(Ϫ25) was introduced into the ALA1 sequence, and the translational efficiency of the resultant construct was tested. As shown in Fig. 6, mutation of ACG(Ϫ24) to ACC alone reduced the protein yield by 2.5-fold (Fig. 6, numbers 1 and 2), whereas a secondary mutation that changed ACC(Ϫ21) to ACG successfully restored the protein yield (Fig. 6, numbers 2 and 3). To ascertain that this compensatory effect actually resulted from the newly inserted non-successive ACG codon, the nucleotide at its relative Ϫ3 position was mutated from A to an unfavorable nucleotide C (resulting in pSJ352), and the initiating activity of the resultant construct was assayed. As expected, once the sequence context of ACG(Ϫ21) was compromised, the initiating activity of this non-successive initiation codon and the associated compensatory effect became negligible (Fig. 6, numbers 3 and 4). This result provides strong evidence that non-successive ACG codons can be as effective as successive ACG codons in initia-   tion, providing they contain a similar sequence context. Fig. 6C shows that similar levels of cDNA products were generated from these fusions, suggesting that mutations at these sites did not compromise the stability of the specific mRNAs in vivo. To substantiate the ability to create a non-successive remedial initiation site, we next used LacZ as a reporter gene. The results obtained from the ALA1-LacZ fusion constructs were consistent with those obtained from the ALA1-lexA constructs (Fig.  6D).
Substituting the ATG Initiator of the Mitochondrial Form of ValRS with Redundant GTG Codons-Our previous study showed that redundant ACG triplets can effectively substitute for the ATG initiator codon of the mitochondrial form of ValRS in vivo, while a single ACG triplet can hardly do so (29) (Fig. 7,  numbers 3 and 4). To test whether redundant GTG codons can assume a similar function in VAS1 and whether redundant GTG codons act more strongly as an initiation site than does a single GTG codon in this gene, the ATG1 initiator codon of VAS1 was substituted with a single GTG or redundant GTG triplets, and the ability of the resultant constructs to translate the mitochondrial form of ValRS and rescue a vas1 Ϫ yeast strain on YPG plates was tested. As shown in Fig. 7, redundant GTG codons could functionally substitute for the ATG1 initiator codon of VAS1; the resultant construct successfully rescued the growth defect of the knockout strain on a YPG plate (Fig. 7, number 6). In addition, redundant GTG codons had initiation activity equivalent to that of redundant ACG codons (Fig. 7, numbers 4 and 6). Paradoxically, a single GTG codon acted more strongly as an initiation site than did redundant GTG or ACG codons (Fig. 7, numbers 4 -6). This finding is in agreement with the data of the Western blots shown in Fig. 1B  (numbers 1, 4, and 5). Thus, the feature of a redundancy effect appears to be independent of the target genes used for analysis. We did not directly assay the relative levels of proteins initiated from these non-ATG initiator codons, because translational initiation from ATG1 (or these non-ATG codons) is likely to be blocked by an upstream, out-of-frame ATG triplet (nucleotides Ϫ5 to Ϫ3 relative to ATG1) (Fig. 7A), making the proteins practically undetectable by Western blot analysis.

DISCUSSION
In contrast to the scenario of ACG initiation, where redundancy of ACG initiation codons significantly enhances the translational efficiency, we present evidence herein that a redundancy of GTG initiation codons does not enhance the translational efficiency at all, but instead lowers the translational efficiency (Fig. 1). Thus, a redundancy of non-ATG initiation codons does not always improve the overall translational efficiency. If the first nucleotide of the non-ATG initiator codon is an A, redundancy of the initiator codons provides not A, schematic summary of VAS1 mutants. The VAS1 sequence shown includes nucleotides Ϫ6 to ϩ16 relative to ATG1. Non-ATG initiator codons are shaded, and nucleotides that were mutated are underlined. B, complementation assays for the mitochondrial function on YPG plates. CW1 was transformed with the wild-type and mutant VAS1 constructs and then its growth phenotypes were tested. In B, numbers 1-6 (circled) denote the VAS1 constructs shown in A. only a favorable nucleotide A at position Ϫ3 for the remedial initiator codon, but also a favorable nucleotide A at position ϩ4 for the preceding initiator codon (Figs. 1-3). Thus, the redundancy effect manifested by the ACG initiator codons is in effect attributed to a favorable sequence context and, to a lesser extent, remedial initiation (Figs. 1-6). In contrast, if the first nucleotide of the non-AUG initiator codon is a G, redundancy of the initiation codons provides both an unfavorable ϩ4 nucleotide G for the preceding initiator codon and an unfavorable Ϫ3 nucleotide G for the remedial initiator codon. As a result, the remedial initiator codon is essentially non-functional, and redundant GTG codons act worse than a single GTG codon with a favorable ϩ4 nucleotide A (Fig. 1). This is a unique and interesting scenario, where both "remedial initiation" and "sequence context" contribute to the translational efficiency of redundant non-AUG initiator codons. Conceivably, a portion of the scanning ribosome must be able to skip the first initiator codon and begin from the adjacent initiator codon under such circumstances. This feature is further supported by the observation that a non-successive ACG codon can also serve as an efficient remedial initiation site and enhance the overall translational efficiency (Fig. 6). On the contrary, if the first initiator codon is an ATG triplet, then leaky scanning or remedial initiation is unlikely to take place anywhere downstream of this initiation site. Therefore, no redundancy effect would normally be expected for an ATG initiator codon. In this sense, it is interesting to mention that the expressions of two yeast genes, MOD5 (coding for isopentenyl pyrophosphate: tRNA isopentenyl transferase) (32) and CCA1 (coding for ATP (CTP): tRNA nucleotidyltransferase) (33), also involve leaky scanning. However, leaky scanning occurs in these two genes not because the upstream initiator is a non-AUG codon or an AUG initiator with a suboptimal sequence context, but because the first AUG codon in both genes is positioned too close to the 5Ј-end of the mRNA. In contrast to the few cases of leaky scanning found in yeast, this mechanism has frequently been observed in mammals (34 -37).
Many cellular and viral mRNAs have been shown to use non-AUG codons, such as AUU, CUG, and GUG, as translation start sites (38). Occasionally, non-AUG codons act as exclusive translation initiators, such as in mRNAs derived from the Arabidopsis AGAMOUS gene (39) and the latently expressed kaposin locus of Kaposi sarcoma-associated herpesvirus (40). However, in most cases, non-AUG codons serve as alternative translation start sites that are accompanied by a downstream in-frame AUG initiator (17). Recognition of the non-AUG initiators in these mRNAs by initiating ribosomes appears to be compensated for by interactions with nucleotide sequences flanking the initiators. A recent study suggested that components of the 48 S translation initiation complex, in particular eIF2 and 18 S ribosomal RNA, might be involved in specific recognition of the context nucleotides at the Ϫ3 and ϩ4 positions (18). So, an optimal sequence context is important to the efficient recognition of a non-AUG initiator. Likewise, recognition of a canonical AUG initiator can be severely compromised by a suboptimal sequence context. In contrast to the active participation of the sequence context in translation initiation in higher eukaryotes, many reports have argued that the sequence context does not substantially affect the initiating activity of an AUG initiator in yeast (21,22,41). Mostly for this reason, yeast was previously thought to be unable to use non-AUG codons as translation start sites (24). Recently, two genes in S. cerevisiae (ALA1 and GRS1) and one gene in C. albicans (CARP2A) were reported to use naturally occurring non-AUG triplets as translation initiators. Further investigation of the GRS1 genes of 14 yeast species suggested that, except for GRS1 of Cryptococcus neoformans, all of these genes contain an alternative non-AUG initiation site followed by an in-frame downstream AUG initiator codon. 4 Fascinatingly, two of these genes contain successive non-AUG initiator codons: Schizosaccharomyces pombe contains 5Ј-ugcAUAAUUucu-3Ј (nucleotides Ϫ69 to Ϫ58 relative to AUG1) and Yarrowia lipolytica contains 5Ј-gccAUU-GUGacu-3Ј (nucleotides Ϫ96 to Ϫ85 relative to AUG1). A functional assay further suggested that AUU and GUG are the primary initiator codons of the mitochondrial GlyRSs of S. pombe and Y. lipolytica, respectively, lending further support to our finding that the sequence context, in particular the Ϫ3 nucleotide, plays a key role in modulating the efficiency of non-AUG initiators in yeast (Fig. 4) (26).
In higher eukaryotes, the most critical nucleotide positions are Ϫ3 and ϩ4 relative to the initiator, and the best (or most favorable) context contains a purine (A or G) at position Ϫ3 and a G at position ϩ4 (38). Although sequences immediately flanking AUG initiation codons are somewhat preferred in yeast, the bias in the nucleotide distribution (5Ј-(A/Y)A(A/Y)AaugUCY-3Ј) is highly divergent from the higher eukaryotic consensus (5Ј-CACCaugG-3Ј), with the exception of a preference for A at the Ϫ3 position (20). Consistent with this observation, many highly expressed yeast genes were found to have an A at the Ϫ3 position and an A-rich leader region (20). However, many reports have also argued that the sequence context plays only a minor role in translation of yeast mRNAs initiated from AUG. Therefore, the yeast consensus deduced solely from sequence analysis might not actually represent the best context for translation initiation in yeast. We recently discovered that the best context for the UUG initiator of the yeast GRS1 gene is AAR (R represents A or G) at nucleotide positions Ϫ3 to Ϫ1, and the most critical nucleotide position is Ϫ3. Mutation of Ϫ3 A to C alone reduced the initiating activity of the UUG initiator by up to 12-fold (26). Interestingly, the UUG initiator of GRS1 happens to have AAA at its relative positions Ϫ3 to Ϫ1, which may explain why only this particular UUG triplet, and not other non-AUG codons that also differ from AUG by a nucleotide elsewhere in the leader sequence, can efficiently function as a translation start site (11). In addition, the redundant ACG initiator codons of the yeast ALA1 gene also contain the optimal 5Ј sequence context AAR (AAG in this case) (Fig. 1) (12), lending further support to our findings. In contrast to positions Ϫ3 to Ϫ1, the nucleotide at position ϩ4 appears to have a relatively mild effect on the translational efficiency of the UUG initiator. Mutation of the nucleotide at this position from A to any other nucleotide affected the initiating activity at most by 2-fold under the conditions used ( Figs. 1 and 3) (26).
Interestingly, several studies suggested that the nature of the penultimate N-terminal residue may influence the substrate preferences of the methionine aminopeptidase and N-␣-acetyltransferase and in turn affect the stability of the protein produced (31,42). It should be noted that the destabilization by the N-end rule only affects proteins processed by endoproteases and does not influence proteins processed by methionine aminopeptidase alone (43). The penultimate N-terminal residues that favor N-terminal Met removal (e.g. Val, Ser, and Thr) are also considered stabilizing N-terminal residues (44). Thus, any protein that has its N-terminal Met removed by methionine aminopeptidase will still retain a stabilizing N-terminal residue. In contrast, N-terminal residues that are considered destabilizing (e.g. Leu, Phe, and Tyr) are not good substrates for methionine aminopeptidase. Therefore, these proteins will not be processed by methionine aminopeptidase and will retain the stabilizing N-terminal Met. Perhaps for these reasons, our fusions with a different penultimate N-terminal residue showed a similar stability (Fig. 2).