The Specificities of Four Yeast Dihydrouridine Synthases for Cytoplasmic tRNAs* □ S

Dihydrouridine is a highly abundant modified nucleoside found widely in tRNAs of eubacteria, eukaryotes, and some archaea. In cytoplasmic tRNA of Saccharomyces cerevisiae , dihydrouridine occurs exclusively at positions 16, 17, 20, 20A, 20B, and 47. Here we show that the known dihydrouridine synthases Dus1p and Dus2p and two previously uncharacterized homologs, Dus3p (en-coded by YLR401c) and Dus4p (YLR405w), are required for all of the dihydrouridine modification of cytoplasmic tRNAs in S. cerevisiae . We have mapped the in vivo position specificity of the four Dus proteins, by three complementary approaches: determination of the molar ratio of dihydrouridine in purified tRNAs from different dus mutants; microarray analysis of a large number of tRNAs based on differential hybridization of uridine-and dihydrouridine-containing tRNAs to the complementary oligonucleotides; and the development and use of a novel dihydrouridine mapping technique, employ-ing primer extension. We show that each of the four Dus proteins has a distinct position specificity: Dus1p for U 16 and U 17 , Dus2p for U 20 , Dus3p for U 47 , and Dus4p for U 20a

A ubiquitous feature of tRNAs is the presence of numerous base and ribose modifications (1). More than 80 different RNA modifications have been described (2), 25 of which are found in cytoplasmic tRNAs of the yeast Saccharomyces cerevisiae; these occur at 35 positions, leading to about 11 modifications in an average yeast tRNA (3).
Dihydrouridine is among the most abundant modified nucleosides found in tRNA (3); the 905 dihydrouridines found in the 561 curated tRNAs are second in number only to the 1,234 pseudouridines. Consistent with its abundance, dihydrouridine is found widely in tRNAs of eubacteria and eukaryotes (3), although it is less common in archaebacteria (4). Furthermore, dihydrouridines are found at one or more of multiple different positions in tRNA, the vast majority of which are in the D loop at positions 16, 17, 20, 20a, and 20b and at the base of the variable arm at position 47. Only in six exceptional cases is dihydrouridine found elsewhere, and in these cases it occurs at one of four other positions in the D loop (15, 17a, 19, and 21) and at position 48 in the variable arm. The persistent occurrence of dihydrouridine in these positions in so many different organisms underscores the evolutionary importance of the modification and of the sites of modification.
To begin to address the roles of dihydrouridine modifications, an important first step is to decipher the substrate specificity rules for the various modified dihydrouridine residues of tRNA. The yeast S. cerevisiae is highly suitable for this analysis for three reasons. First, yeast tRNAs are modified with dihydrouridine at most of the sites of modification that have been found in characterized tRNAs. Thus, each of the six most common dihydrouridine modification sites are found in multiple yeast cytoplasmic tRNAs (Table I), and two of the five minor dihydrouridine modification sites are found in its sequenced mitochondrial tRNAs, at positions 15 and 17a. Second, formation of most dihydrouridines in yeast cytoplasmic tRNA requires primarily that a uridine is encoded at the appropriate site. Thus, all 83 encoded uridine residues at positions 16, 17, 20, 20a, and 20b of characterized yeast cytoplasmic tRNAs are modified to dihydrouridine, and 17 of the 20 U 47 residues in the variable arm are modified (Table I). In contrast, for many other modifications, only certain tRNA species with the corresponding residue are modified. Thus, m 1 G 9 modification occurs at only 10 of the 22 G 9 residues of characterized yeast tRNAs (5), ⌿ 1 modification of yeast cytoplasmic tRNAs occurs only rarely at encoded uridines at 6 of the 15 positions at which ⌿ is found (6), and dihydrouridine modifications in tRNA of other organisms are much more sporadic at uridine residues (Table I). Third, the genes encoding dihydrouridine synthases have recently been identified. We previously identified two yeast dihydrouridine synthases (Dus1p and Dus2p) that were responsible for some of the dihydrouridine modification of yeast bulk tRNA in vivo and of certain tRNAs in vitro (7). The yeast Dus proteins are part of a widespread family of likely dihydrouridine synthases found in eukaryotes and prokaryotes, including two other members found in yeast (encoded by open reading frames YLR401c and YLR405w) (7) and three related dihydrouridine synthases identified in Escherichia coli (8).
We have investigated the specificity of dihydrouridine modification of cytoplasmic tRNAs in yeast, by construction of strains containing every possible combination of deletions of the known or likely dus genes and examination of dihydrouridine modification of tRNAs from these strains. By quantification of the dihydrouridine levels of purified tRNA species, we developed a preliminary map to describe the specificity of the Dus proteins. Microarray analysis of individual dus deletion strains provided an independent map of the specificities that showed good agreement with the one generated from calculating total dihydrouridine content. The specificity model was confirmed and extended by the development and use of a new dihydrouridine mapping method suitable for analysis of individual tRNAs in bulk RNA. Our results indicate that each the four yeast Dus proteins is specific for dihydrouridine modification at one or two distinct positions in cytoplasmic tRNAs.
All deletion strains were confirmed by PCR analysis of genomic DNA prepared from mutant strains, using primers homologous to the region outside the integrated area in chromosome.
Preparation of Low Molecular Weight RNA from Yeast-Low molecular weight RNA was extracted by phenol at 55-60°C essentially as described (13), from yeast cells grown in 125 ml of medium to an A 600 of 3-4. RNA was precipitated twice with ethanol, resuspended in 1 ml of H 2 O, and stored at Ϫ80°C.
Purification of Individual tRNAs from Yeast-tRNA Phe , tRNA CAA Leu , and tRNA Tyr were purified as previously described (5) from low molecular weight RNA of wild type and various dus-⌬ strains, using the following 5Ј-biotinylated DNA oligomers complementary to the 3Ј side of corresponding tRNAs: 5Bio-F1 for tRNA Phe (5Ј/Biotin/GTGGATC-GAACACAGGACCT); 5Bio-Y2 for tRNA Tyr (5Ј/Biotin/TGGTCTC-CCGGGGGCGAGTCGAACG); and 5Bio-L2 for tRNA CAA Leu (5Ј/biotin/ GAGATTCGAACTCTTGCATCTTACG). Purity of the tRNA was examined by both quantitative primer extension (using a 50-fold excess of primer to tRNA input) and by HPLC, as described below.
HPLC Analysis of Dihydrouridine-Bulk RNA (ϳ20 g/injection) and purified tRNA (ϳ1 g/injection) from different yeast strains were digested with 5 g of P1 nuclease in buffer containing 30 mM sodium acetate (pH 5.2) and 0.2 mM ZnCl 2 at 37°C overnight and then with 5 units of calf intestinal alkaline phosphatase (Roche Applied Science) in 1ϫ alkaline phosphatase reaction buffer at 37°C for 3 h. The resulting nucleosides were resolved by HPLC (Waters Alliance model 2690, equipped with a Waters 996 photodiode array detector) on a reverse phase C18 column (Supelco LC-18; 30 cm ϫ 4.0 mm; 5 m, 120 Å; Supelco, Inc.) essentially as described (14). Individual spectra of the nucleosides retrieved by the photodiode array detector were used to assign the peaks. The number of moles of dihydrouridine and other modified nucleosides per tRNA were calculated as described (5).   (15) were grown in 50 ml of SC medium to ϳ1 ϫ 10 7 cells/ml in parallel with a wild type (BY4743) culture. Total RNA was extracted using hot phenol as previously described (16), ethanol-precipitated and quantified by measuring the A 260 . Contaminating DNA was removed by treatment with DNase I (Fermentas) according to the manufacturer's instructions followed by phenol extraction and precipitation. Ten micrograms of total RNA was directly labeled with Alexa Fluor 546 or 647 (Molecular Probes "Ulysis" kit), ethanol-precipitated, and hybridized to a commercially produced microarray (Agilent Technologies) as previously described (17), except that the formamide concentration in the hybridization buffer was reduced to 25%. Arrays were scanned on an Axon 4000B instrument and quantitated with GenePix software. The raw log ratios for each oligonucleotide were calculated from the median intensity at each spot, and the data were normalized by applying a Lowess smoother to the ratios of each experiment over intensity. All microarray data are compiled in the Supplemental Table S1. Detailed probe sequences and other data are compiled on the World Wide Web at hugheslab.med.utoronto.ca/Xing.
Dihydrouridine Mapping-ϳ20 g of bulk cellular RNA was incubated in 8-l reaction mixtures with 0.1 N KOH at 37°C for 5 min or longer as indicated, and the pH was neutralized by the addition of 7 l of 5ϫ annealing buffer (250 mM Tris-HCl (pH 8.3), 150 mM NaCl, and 50 mM dithiothreitol). Then RNA was subject to a primer extension assay as described (5), except that ϳ8 g of RNA (containing ϳ8 pmol of individual tRNA) and 25 pmol of 5Ј-32 P-labeled primer were used in an annealing reaction of 9 l. Primers used for dihydrouridine mapping the 5Ј side of tRNA Phe , tRNA Tyr , and tRNA CAA Leu and for the 3Ј side of tRNA Tyr are illustrated in Fig. 5. Other primers used in dihydrouridine mapping are 5V-1 for tRNA Val (5Ј-GCGTGTTAAGCAGATG), 5E-1 for tRNA Glu (5Ј-GGTCTCCACGGTGAAAGCG), and 5F-1b for the 3Ј side of tRNA Phe (5Ј-GGTGCGAATTCTGTGGATC). Primer extension products were resolved on 12% polyacrylamide (19:1), 4 M urea gels and visualized using a PhosphorImager (Amersham Biosciences).

Dihydrouridine Is Undetectable in Bulk RNA from a Yeast Strain Lacking All Four Dus Protein Family Members-
We previously found that dus1-⌬ and dus2-⌬ strains each had substantially less dihydrouridine than did control wild type strains, whereas strains lacking the conserved family members encoded by open reading frames YLR401c (now named DUS3) and YLR405w (now named DUS4) had marginally reduced dihydrouridine (7). To investigate this more fully, we have improved the resolution of our initial HPLC methods with longer columns and have quantified the dihydrouridine contribution of each of the four Dus proteins. To this end, we have constructed a series of haploid deletion strains with every possible combination of deletions of the four DUS family members (Table II) (see "Experimental Procedures"). Each of the 15 dus-⌬ strains is viable in rich medium (YPD) and, surprisingly, grows at comparable rates in this medium at different temperatures.
To evaluate the contribution of the four Dus protein family members to dihydrouridine modification, we extracted RNA from a wild type control strain and several of the dus mutant strains and then compared their nucleoside content by HPLC analysis of appropriately digested RNA. Fig. 1A shows a comparison of the nucleosides in RNA from a wild type strain with those from a dus1-⌬, dus2-⌬, dus3-⌬, dus4-⌬ quadruple mutant strain. The position of dihydrouridine in the HPLC elution profile is indicated by the arrow and verified according to its spectrophotometric profile (14); as shown in Fig. 1B, and described previously (7), dihydrouridine absorbs at 210 nm but not at 263 nm, whereas the adjacent nucleoside pseudouridine has absorption peaks at both of these wavelengths. It is clear from Fig. 1B that RNA from the dus1-⌬, dus2-⌬, dus3-⌬, dus4-⌬ quadruple mutant strain lacks any detectable dihydrouridine.
To quantify the contributions of individual Dus proteins to bulk dihydrouridine modification, RNA from a series of strains with different combinations of dus mutants was analyzed for its dihydrouridine content. As shown in Fig. 1C and calculated from the areas of the peaks, a dus1-⌬, dus2-⌬ strain retains ϳ35% of the normal dihydrouridine content, and further removal of either DUS4 or DUS3 drops the dihydrouridine level to ϳ20%. Finally, deletion of the fourth and last DUS homolog (dus1-⌬, dus2-⌬, dus3-⌬, dus4-⌬) generates the quadruple knockout, resulting in no detectable dihydrouridine. These results suggest that all four DUS genes contribute to dihydrouri-FIG. 1. HPLC separation of nucleosides from cellular RNA of wild type and dus-⌬ strains of yeast. Bulk tRNA prepared from wild type and dus mutant strains was digested to nucleosides and chromatographed on an HPLC column, as described under "Experimental Procedures." A, elution profile at 210 nm of nucleosides from wild type and dus1-⌬, dus2-⌬, dus3-⌬, dus4-⌬ strains. Nucleosides C, U, G, and A are labeled; the arrow indicates the elution positions of dihydrouridine (D) and pseudouridine (⌿). B, enlarged view of dihydrouridine and pseudouridine profiles from wild type and dus1-⌬, dus2-⌬, dus3-⌬, dus4-⌬ strains. Chromatograms are shown at 210 and 263 nm. C, comparison of dihydrouridine and pseudouridine profiles obtained from different multiple dus mutant strains. Nucleosides were separated as described above from wild type strains and double, triple, and quadruple dus-⌬ strains, as indicated, and profiles were monitored at 210 nm. dine modification of RNA in vivo.
Analysis of Purified tRNAs from dus Mutant Strains Yields a Preliminary Model for Specificities of the Dus Protein Members in S. cerevisiae-To begin to determine the specificities of the Dus proteins for the six dihydrouridine positions that are modified in yeast cytoplasmic tRNAs, we analyzed three individual tRNAs purified from different dus mutant strains for their dihydrouridine content. As shown in Fig. 2, these tRNAs (tRNA Phe , tRNA CAA Leu , and tRNA Tyr ) represent different dihydrouridine distributions. Each tRNA was purified by hybridization of bulk RNA to a 5Ј biotin-labeled DNA oligomer specific for the target tRNA, followed by binding of the hybridized tRNA to streptavidin beads and elution of the bound tRNA (see "Experimental Procedures"). This purification method yielded tRNA that was judged to be more than 95% pure, based on quantitative primer extension with specific probes, lack of detectable primer extension products with two other probes specific for other tRNAs, and comparison of the yield calculated from primer extension data with the known amount of RNA input (data not shown). To quantify the modified nucleosides in the purified tRNAs, tRNAs were digested to nucleosides and subjected to HPLC analysis, and the relative molarities of dihydrouridine and other modified nucleosides were calculated from the elution profile and the known extinction coefficients (see "Experimental Procedures").
Analysis of tRNA Phe suggests that its two dihydrouridines at positions 16 and 17 are modified by Dus1p. From the chromatograms shown in Fig. 2A, it is clear that tRNA Phe from a dus1-⌬ strain has little detectable dihydrouridine, whereas tRNA Phe from a wild type or a dus2-⌬ strain has comparable amounts of dihydrouridine. The quantification shown in Table III confirms this result. tRNA Phe from the wild type and the dus2-⌬ strain contains 1.4 and 1.5 mol of dihydrouridine/mol of tRNA, and tRNA Phe from dus1-⌬ strains has no detectable dihydrouridine. By contrast, quantification of most of the other modified nucleosides present in this tRNA (⌿, m 2 G, m 2 2 G, m 5 C, and m 1 A) shows that they are all close to their expected values and are unchanged in tRNA Phe from all three strains. This close correspondence of expected nucleoside composition to observed composition independently validates our claim that these tRNAs are highly pure. This result demonstrates that Dus1p is responsible for modifying positions 16 and 17 of this tRNA, consistent with our previous observation that Dus1p modifies U 17 of tRNA Phe in vitro (7).
tRNA CAA Leu contains two dihydrouridines, D 20 and D 20b , which appear to be modified by Dus2p and Dus4p. Fig. 2B and Table  III show that tRNA CAA Leu from a dus1-⌬ strain (2.2 D/tRNA Leu ) and a dus3-⌬ strain (2.1 D/tRNA Leu ) retains full levels of dihydrouridine in tRNA CAA Leu compared with the wild type strain (2.2 D/tRNA Leu ). However, deletion of either DUS2 (1.0 D/tRNA Leu ) or DUS4 (1.2 D/tRNA Leu ) causes loss of one dihydrouridine residue from this tRNA, and deletion of both DUS2 and DUS4 completely eliminated dihydrouridine from this tRNA species. Thus, we conclude that Dus2p and Dus4p each modify one position of tRNA CAA Leu , although neither protein can be assigned to a specific residue based on this analysis. These results are also consistent with our previous in vitro results, which demonstrated that Dus2p and Dus4p each had activity with a tRNA CAA Leu substrate, although only marginal activity was detected with Dus4p (Ylr405p) (7).
tRNA Tyr is of particular interest, because dihydrouridine is present at all six known positions of the modification that are found in yeast cytoplasmic tRNA (Table I). As shown in Fig. 2C and Table III, tRNA Tyr from the wild type strain has 5.2 D/tRNA Tyr , whereas tRNA Tyr from a dus1-⌬, dus2-⌬, dus3-⌬ strain has nearly 2 dihydrouridine residues (1.8 D/tRNA Tyr ). This suggests that Dus4p activity accounts for formation of two of the six dihydrouridine residues in tRNA Tyr . These sites could be D 20 and D 20A or could be D 20A and D 20B but not D 20 and D 20B , since our analysis of tRNA CAA Leu suggests that Dus2p acts at either D 20 or D 20B in this tRNA species (assuming that site specificity is conserved in different tRNA species). tRNA Tyr also has a dihydrouridine at position 47 in the extra loop. As shown in Fig. 2C and Table III, one dihydrouridine residue (0.9 D/tRNA Tyr ) is present in tRNA Tyr from a dus1-⌬, dus2-⌬, dus4-⌬ strain. Thus, we tentatively assign Dus3p to formation of D 47 .
Microarray Analysis Largely Confirms the Preliminary Model of Dus Protein Specificity-It was previously observed that the absence of dihydrouridine in tRNAs from a dus1 mutant strain could be detected by microarray, using a protocol in which the fluors are coupled directly to the cellular RNA and hybridized to an array of oligonucleotides (16). It was noted that only oligonucleotides covering tRNA positions 16 and 17 reported significant preferential hybridization to RNA from a dus1 mutant relative to RNA from a wild type strain (positive ratios), and it was concluded that the loss of dihydrouridine modification in the dus1 mutant results in higher binding affinity.
We have extended this analysis by examining RNAs from all four dus deletion strains on a microarray with oligonucleotides designed to hybridize to multiple different portions of 70 genomic tRNA sequences in S. cerevisiae, comprising 42 different tRNA species (including 14 that have not previously been analyzed). Oligonucleotides with an average length of 18.4 and T m of 41°C were designed to tile across the tRNA sequence at 5-nucleotide intervals. We reasoned that only oligonucleotides specific for regions of tRNA spanning dihydrouridine modification sites would show preferential hybridization to RNA from dus mutant strains, allowing us to evaluate the specificity of each of the Dus proteins. An example of this approach is illustrated in Fig. 3A and Table IV, in which we evaluated hybridization data for oligonucleotides spanning the region around nucleotide 47 of tRNA GCA Cys . As can be seen, probes that span position 47 (probes 546 -549) yielded significantly more hybridization to RNA from a dus3 mutant compared with wild type RNA but not with RNA from any other dus mutant. By contrast, neighboring oligonucleotides (probes 545 and 550) that do not span position 47 of tRNA GCA Cys have similar hybridization for RNAs from all dus mutants and wild type cells. These data are most consistent with the conclusion that Dus3p is specific for postion 47 of tRNA GCA Cys . Analysis of the microarray data largely corroborates the specificity of Dus3p, further confirms our other tentative assignments of Dus protein specificity, and extends the results to other tRNA species. Fig. 3B lists the oligonucleotides with the highest ratio of preferential hybridization (over 2.83-fold) to RNA from one or more of the dus mutant strains (relative to wild type) with array results presented in a clustergram (the data from the entire microarray is shown as supplementary data in Table S1, and can be viewed on the World Wide Web at hugheslab.med.utoronto.ca/Xing). From these data, it is obvious that Dus3p is specific for modification at position 47 for all eight tRNAs for which a signal is observed, including one species (tRNA UGU Thr ) for which there is no previous analysis. Similarly, the assignment of Dus1p specificity to positions 16 and 17 is unequivocal based on 12 positive hybridization ratios observed in RNA from dus1 mutants, using oligonucleotides spanning positions 16 and 17 and terminating before position 20. This includes four tRNA species that have not previously been analyzed (tRNA CUG Gln , tRNA CGA Ser , tRNA GCT Ser , and tRNA GAG Leu ). The hybridization results also implicate Dus2p and Dus4p in modifications covering positions 20, 20A, and 20B, but further discrimination is not possible based on these results. Surprisingly, RNA from dus2 mutants also appears to preferentially hybridize to probes specific for the extra arm of tRNA CAA Leu , which is not known to have a dihydrouridine, as well as to oligonucleotides 6 -18 of tRNA GAG Leu and 6 -19 of tRNA GCA Cys , although the preponderance of data argues that tRNAs are modified by Dus1p before nucleotide 20.
A Primer Extension Method for Mapping Dihydrouridine Modification Sites in tRNA-The above analyses of the specificity of Dus proteins for dihydrouridine modification are incomplete in two respects; it is difficult to assign specificity for Dus2p and Dus4p with either approach, and both methods lack nucleotide resolution. Thus, we sought a convenient and rapid mapping technique to precisely determine the positions of dihydrouridines in a tRNA.
We developed such a dihydrouridine mapping technique based on the observation that dihydrouridine is moderately susceptible to alkaline hydrolysis, which results in breakdown at the 3-4 linkage of dihydrouracil ring (18 -20). The product of the ring-opening reaction, ␤-ureidopropionic acid (Fig. 4A), is therefore expected to lose the capacity for base pairing and to generate a primer extension stop (4B). To test this idea, we incubated 20 g of wild type bulk RNA (corresponding to ϳ1 pmol of a specific tRNA) with 0.1 N KOH at 37°C for 2-45 min, neutralized the sample, and then performed primer extension using a primer specific for the 3Ј-end region of tRNA Tyr (see Fig. 5E for positioning of the primer). As shown in Fig. 4C, a band at C 48 increases with increased time of alkali treatment, presumably because of the destruction of D 47 . The sequence ladder on the left confirms that the primer extension is from tRNA Tyr present in the bulk RNA. In the absence of alkali treatment, there is no stop product formed at this position; rather, the primer can extend through to position 29, 2 nucleotides before m 2 2 G 26 . Thus, this dihydrouridine mapping technique appears to provide a rapid and accurate determination of the site of dihydrouridines in tRNA.
To further validate this mapping method, we tested tRNA Phe . As shown ( Fig. 2A and Table III), the D 16 and D 17 modifications of tRNA Phe are due to Dus1p, based on analysis of the nucleoside content of purified tRNA Phe from a dus1 mutant strain. Primer extension analysis of alkali-treated RNA with a tRNA Phe primer confirms this result (Fig. 5B). As anticipated, primer extension stops are found at positions 17 and 18 of tRNA Phe from a wild type strain as well as from dus2-⌬, dus3-⌬, and dus4-⌬ strains, 1 residue downstream of the known dihydrouridine positions (Fig. 5B, lanes f and h-j). By contrast, these primer extension stops are both missing in alkali-treated RNA from a dus1-⌬ strain (lane g). This result confirms that Dus1p modifies U 16  tRNA Tyr has dihydrouridine residues at five positions in the D loop as well as at D 47 . We proposed above that Dus3p might be responsible for formation of D 47 , based on our quantitation of the dihydrouridine content of tRNA Tyr from a dus3-⌬ strain ( Fig. 2C and Table III) and the microarray analysis (Fig. 3A). Our data in Fig. 5F validate this claim, since the primer extension stop corresponding to position 47 (lane f) is absent in RNA from a dus3-⌬ tRNA Tyr (lane i) but not in the other three dus deletion strains (lanes g, h, and j).
The  modifications in this region (Fig. 5G). Furthermore, for reasons that are unclear, a number of bands appeared even in the absence of alkali treatment of wild type tRNA Tyr (lane e), and the density of dihydrouridine residues precludes visualization of all five dihydrouridine modifications in treated wild type tRNA (lane f). Because it cannot be compared, there is no score for tRNA Tyr from the dus1-⌬ strain (lane g), although we expected to see neither D 17 nor D 16 . As expected, tRNA Tyr of the dus3-⌬ strain showed the same mapping pattern as the wild type (lane i), consistent with the assignment of Dus3p to position 47. Despite these difficulties, we can still use the mapping method to assign all of the dihydrouridines in the D loop of tRNA Tyr to specific Dus proteins, with the exception of D 16 , which cannot be mapped. We observe a missing band corresponding to D 20 in RNA from a dus2-⌬ strain (lane h) and two missing bands corresponding to D 20a and D 20b in RNA from the dus4-⌬ strain (lane j). Thus, it is clear that Dus2p modifies only U 20 , and Dus4p modifies both U 20a and U 20b . We further demonstrated the position specificity of Dus1p, Dus2p, and Dus4p for tRNA Tyr , using RNA from multiple knockout strains. Thus, RNA from a dus2-⌬, dus4-⌬ mutant strain is missing D 20 , D 20a , and D 20b (lane k); RNA from a dus1-⌬, dus4-⌬ mutant strain has D 20 because of the presence of DUS2 (lane l); and RNA from a dus1-⌬, dus2-⌬ mutant strain has D 20a and D 20b because of the presence of DUS4 (lane m). Since a dus1-⌬, dus2-⌬, dus4-⌬ mutant is missing all of its modification in the D loop (lane n), this demonstrates that D 17 is modified by Dus1p, as expected. We note that the signal for D 16 is too weak to ascribe to Dus1p or any other Dus protein.
Based on both the HPLC analysis of purified tRNAs from wild type and dus mutant strains and results from this new dihydrouridine mapping technique, we propose a model of Dus protein specificity for the different dihydrouridine positions of tRNA, as shown in Fig. 6. According to this model, Dus1p modifies U 16  tRNA UUC Glu contains only D 20a , which ought to be modified by Dus4p. Fig. 7B shows, as expected, that D 20a of tRNA UUC Glu is lost exclusively in the dus4-⌬ strain (lane j) but not in the dus2-⌬ strain (lane h). This result suggests that both Dus4p and Dus2p possess highly stringent specificities for dihydrouridine modification, even in the absence of other dihydrouridines in the D loop as reference.
Position Alone Does Not Confer Dihydrouridine Modification-The Dus specificity model assigns a specific position(s) for dihydrouridine modification by each Dus protein. Thus, a uridine that is present at each of the six consensus positions for dihydrouridine is modified by a specific Dus protein. There are three exceptions to this rule among sequenced yeast cytoplasmic tRNAs: tRNA Phe1 and tRNA Phe2 (which differ only in the base pair between positions 6 and 67) and tRNA Lys1 , each of which has a U 47 instead of dihydrouridine at this position (3). To determine whether the assignment of U 47 is correct for these tRNA species, we mapped the extra loop region of tRNA Phe1 . As shown in Fig. 7D, the wild type and the four single dus mutant strains all carry an unmodified U 47 , since there is no primer extension stop at position 48. The observed primer extension stop in all lanes occurs at residue 47, which corresponds to m 7 G 46 . Thus, position 47 of tRNA Phe1 is indeed a uridine residue as previously assigned and not a dihydrouridine. Thus, for modification of U 47 by Dus3p, other signal(s) appear to be required. FIG. 4. Dihydrouridine mapping approach. A, schematic of reaction of alkali with dihydrouridine to yield ␤-ureidopropionic acid ribose derivative. B, scheme of dihydrouridine mapping method. Alkali treatment of RNA causes hydrolysis of the ring of dihydrouridine residues, resulting in a primer extension block at the residue N immediately 3Ј of the hydrolyzed dihydrouridine residue X. C, time course of effect of alkali treatment on primer extension of tRNA Tyr through D 47 . Bulk RNA from yeast was treated with 0.1 N KOH at 37°C for 2-45 min, as indicated, neutralized, and annealed with 5Ј-32 P-labeled primer for tRNA Tyr (see Fig. 4E), and then the primer was extended as described under "Experimental Procedures." a-d, sequencing ladders of tRNA Tyr using dideoxynucleotides as indicated; e, no KOH treatment; f-j, primer extension resulting from KOH treatment for 2, 5, 10, 20, and 45 min. The sequence ladder is indicated at the left; the arrow indicates primer extension stop at position C 48 , corresponding to hydrolysis at D 47 .

DISCUSSION
In this paper, we demonstrate that formation of dihydrouridine in yeast cytoplasmic tRNAs is carried out by a family of four dihydrouridine synthases (Dus1p, Dus2p, Dus3p, and Dus4p), each acting at specific positions in tRNAs. Dus1p modifies U 16 and U 17 , Dus2p modifies U 20 , Dus3p modifies U 47 , and Dus4p modifies U 20a and U 20b . These four proteins are responsible for all dihydrouridine modification of cytoplasmic tRNAs in yeast, since bulk cellular RNA from a dus1-⌬, dus2-⌬, dus3-⌬, dus4-⌬ strain (Fig. 1A) lacks any detectable dihydrouridine and since the specificities of the four proteins account for all known modifications of sequenced yeast cytoplasmic tRNAs. Consistent with these assignments, E. coli tRNAs have dihydrouridines exclusively in the D loop (at positions 16, 17, 20, 20a, and 21) and not at position 47, and a family of three Dus proteins is sufficient for dihydrouridine modification of E. coli tRNAs (8).
We determined the specificity of the Dus proteins in three steps, starting with a set of strains with various combinations of dus mutants. First, we developed a preliminary model of Dus protein specificity by HPLC analysis of nucleosides and quantification of dihydrouridine in purified tRNAs from these strains. This led to the assignment of Dus1p and a preliminary assignment of Dus3p in the same way as E. coli DusA protein was found to be responsible for D 21 modification of tRNA 2 fMet (8). Second, we used microarray analysis to confirm the preliminary model for Dus1p and Dus3p and to extend the results to multiple tRNA species. However, although both methods could partially define the specificities of Dus2p and Dus4p, we could not define them precisely with the tRNAs and the probes we employed. Third, we developed a new primer extension technique to precisely map all dihydrouridine modification sites, based on brief alkali treatment of bulk tRNA to break the dihydrouridine ring, followed by primer extension to map the newly created primer extension stop. This dihydrouridine mapping technique was used to confirm initial assignments and to assign the remaining specificities.
Each method of dihydrouridine evaluation has its specific advantages. Analysis of nucleosides from purified tRNAs yields accurate quantification of dihydrouridine and of all other modifications in one experiment. Microarray analysis gives an immediate global view of dihydrouridine modification of many species of tRNA and, in principle, of all unique species; indeed,  this experiment surveyed 70 of the 274 known tRNA genes (21). Moreover, the microarray method can be generalized to any other modification that yields a hybridization difference. Although we did see a small number of false positive signals in the microarray data, the frequency at which these occur is very low. All nine of the false positive signals were observed in analysis of dus2 mutants, and seven of these occurred within a specific region of tRNA CAA Leu ; perhaps these tRNA CAA Leu -specific oligonucleotides fortuitously hybridize to the region across nucleotide 20 of this or another tRNA species. The dihydrouridine mapping technique, like the pseudouridine mapping technique developed by Bakin and Ofengand (22,23), is specific, simple, and precise and generally applicable to any dihydrouridine in RNA, except perhaps at the extreme 3Ј-end. For each of 13 known dihydrouridine modifications in five tRNAs for which we could see a primer extension result, we were able to unequivocally determine the position of the dihydrouridine modification and the Dus protein responsible for it. However, we note that we could not map D 16 in tRNA Tyr because of the lack of a clear primer extension signal after alkali treatment. This weak signal could be due in part to the large number of dihydrouridine residues in this region, which could mask the primer extension stop corresponding to D 16 , and in part to enhanced natural resistance of this particular dihydrouridine residue relative to the other dihydrouridines. In support of this, we note that a weak signal is detected from D 16 in dus2, dus4 double mutants.
As with most modification enzymes, the Dus proteins do not appear to compensate for each other or to depend on other dihydrouridine modifications for their action at nearby residues. In all cases, the absence of a specific Dus protein leads to the absence of that modification in the tRNA and of no other dihydrouridine modification. In particular, Dus2p and Dus4p faithfully retain their specificity for position 20 and positions 20a and 20b, respectively, in the absence of the other protein (Fig. 7B). Similar nonredundant activities were also observed in analysis of dihydrouridine modification in E. coli.
It is striking that all uridine residues in yeast tRNA at positions 16, 17, 20, 20a, and 20b in the D loop are modified to dihydrouridine. This may be a consequence of the observed high solvent exposure of all of the dihydrouridine-modified positions in the three known structures of yeast tRNAs (tRNA Phe , tRNA Asp , and tRNA i Met ) (24 -27). For many other modifications, only a portion of the residues with the appropriate base at a particular position are modified by the modification enzymes. For example, 12 of 15 sites in the cytoplasmic tRNAs of S. cerevisiae are not complete for pseudouridine modification, leaving 35% of total uridines at these 12 positions unmodified (6). This suggests that there are other determinants for this and other modifications, including dihydrouri- dine modification in other organisms. Interestingly, mutations in tRNA Ser (G18C and G19C, separately) not only abolished Gm 18 modification but also blocked the formation of D 20 in tRNA Ser , suggesting the dependence of the tertiary structure for D 20 modification (28).
Both Dus1p and Dus4p are members of a small class of tRNA modification enzymes that are region-specific, in that they modify more than one nucleotide in the same region of the tRNA. In yeast, the only other known members of this class are Trm7p, which catalyzes 2Ј-O-methylation at positions 32 and 34 (29), and Pus3p, which forms ⌿ 38 and ⌿ 39 (30). Other modification enzymes act either at multiple sites, like Pus1p (6,31,32) and Trm4 (33), or, more commonly, at single sites. This region specificity argues either for conformational flexibility between the protein binding site and the active site or a flexible region of the tRNA substrate.
Although we have not yet examined mitochondrial tRNA from dus mutant strains to determine whether Dus proteins also act in the mitochondria, we think it is likely that Dus1p and Dus2p will have this role. Yeast mitochondrial tRNAs are modified extensively with dihydrouridine at positions 16 (13 D residues, 1 U residue, 3 others), 17 (3 D, 3 U), and 20 (17 D, 0 U) and not at all at positions 20a (6 U, 2 others), 20b (2 U, 2 others), and 47 (4 U). Based on the specificity assignments here, the lack of other obvious Dus homologs present in the yeast genome, and the known dual compartment specificity of enzymes such as Mod5p (34,35), Trm2p (36), Pus3p (30), and Pus4p (37) that modify tRNA at similar positions in mitochondria and cytoplasmic tRNA, it seems likely that Dus1p and Dus2p will be responsible for the corresponding modifications in mitochondria. Based on its known region specificity, it also seems likely that Dus1p will be responsible for the occasional modification observed in mitochondrial tRNA at positions 17a (1 D, 0 U) and 15 (1 D, 1 U, and 15 others). Unfortunately, our microarray was not designed in such a was as to be able to answer these questions; however, it should be possible to design an array on which to detect modifications to mitochondrial tRNAs.
The results presented here set the stage for two lines of future research. First, the nature of the substrate specificity of the Dus proteins remains unsolved. Although there must be specific determinants in each Dus protein that dictate where it will bind and modify tRNA, the nature of those determinants is unknown. In this regard, it is surprising that E. coli DusA protein (also called YjbN) is most homologous to Dus1p (3.6 e Ϫ9 ) and much less so to Dus2p (4.4 e Ϫ4 ) and Dus4p (3.2 e Ϫ3 ), since it acts at position 21 (8) and might therefore be expected to be more related to Dus2p or Dus4p. Second, the role of dihydrouridine remains enigmatic. Dihydrouridine is known to cause increased flexibility in loops that contain it because of the propensity of dihydrouridine residues to adopt the C2Ј-endo conformation (38). Consistent with this increased flexibility, dihydrouridine is found in high levels in cold-adapted (psychrophilic) bacteria and archaea but not in thermophilic archaea (4,39). Dihydrouridine modification of tRNA also causes less efficient hybridization of oligonucleotides to regions of tRNA with dihydrouridine (16). This could arise as a consequence of the increased propensity of dihydrouridine for the C2Ј-endo conformation or perhaps because of the loss of the aromatic ring structure of uridine in dihydrouridine. Thus, dihydrouridine may have a role in preventing the formation of alternative tRNA structures in the cell. In addition, overexpression of Dus2p (also called Smm1p) can suppress the temperaturesensitive phenotype caused by a mutation at position 61 in mitochondrial tRNA Asp that impairs 3Ј-end processing (40,41); however, it is unclear how this occurs. We observe little growth phenotype in each of our dus mutant strains or in a multiple dus-1, dus2, dus3, dus4 mutant strain on plates or in liquid media at a variety of temperatures (23,30,37). This lack of a growth phenotype suggests a subtle phenotype that may only be revealed by examination of dus mutants in combination with other mutants in tRNA genes or modification enzymes (42,43).