Dual pathways of tRNA hydroxylation ensure efficient translation by expanding decoding capability

In bacterial tRNAs, 5-carboxymethoxyuridine (cmo5U) and its derivatives at the first position of the anticodon facilitate non-Watson–Crick base pairing with guanosine and pyrimidines at the third positions of codons, thereby expanding decoding capabilities. However, their biogenesis and physiological roles remained to be investigated. Using reverse genetics and comparative genomics, we identify two factors responsible for 5-hydroxyuridine (ho5U) formation, which is the first step of the cmo5U synthesis: TrhP (formerly known as YegQ), a peptidase U32 family protein, is involved in prephenate-dependent ho5U formation; and TrhO (formerly known as YceA), a rhodanese family protein, catalyzes oxygen-dependent ho5U formation and bypasses cmo5U biogenesis in a subset of tRNAs under aerobic conditions. E. coli strains lacking both trhP and trhO exhibit a temperature-sensitive phenotype, and decode codons ending in G (GCG and UCG) less efficiently than the wild-type strain. These findings confirm that tRNA hydroxylation ensures efficient decoding during protein synthesis.

R NA modifications confer chemical diversity on simple RNA molecules, expanding their functional repertoires. To date, more than 140 species of RNA modifications have been identified in RNA molecules from all domains of life 1 .
In protein synthesis, tRNA serves as an adapter molecule to connect codons on mRNA with the corresponding amino acids. After they are transcribed, tRNAs undergo chemical modifications mediated by site-specific tRNA-modifying enzymes. These modifications play critical roles in stabilizing tRNA tertiary structure and fine-tuning the decoding process 2,3 . A wide variety of modifications are present at the first (wobble) position of the anticodon in tRNA (position 34). The wobble modification modulates codon recognition, thereby promoting accurate decoding during protein synthesis 3,4 .
In contrast to xm 5 (s 2 )U-type modifications, 5-hydroxyuridine derivatives (xo 5 U) are present at the wobble positions of tRNAs responsible for NYN family boxes, and serve to expand decoding capacity in most bacterial species (Fig. 1a, b). To date, four species of xo 5 U modifications have been reported. 5carboxymethoxyuridine (cmo 5 U, also called uridine-5-oxy acetic acid) and 5-methoxycarbonylmethoxyuridine (mcmo 5 U, also called uridine-5-oxy acetic acid methyl ester) are found in Gram-negative bacteria, including Escherichia coli and Salmonella enterica 2,12 . 5methoxycarbonylmethoxy-2'O-methyluridine (mcmo 5 Um) was recently detected in E. coli tRNA Ser as a minor modification 13 . In addition, 5-methoxyuridine (mo 5 U) is present in tRNA Thr from Gram-positive bacteria, including Bacillus subtilis 14 . Using mass spectrometry, we revealed that cmo 5 U34 is present as a major wobble modification in tRNA Leu3 and tRNA Val1 , whereas mcmo 5 U34 is primarily present in tRNA Ala1 , tRNA Ser1 , tRNA Pro3 , and tRNA Thr4 in E. coli 13 (Fig. 1b). The xo 5 U-type modifications facilitate non-Watson-Crick base pairing with guanosine and pyrimidines at the third positions of codons [15][16][17][18] (Fig. 1b), thereby contributing to expansion of decoding capability. Moreover, the terminal methylation of mcmo 5 U contributes to decoding ability, at least in tRNA Ala1 13 .
A structural study of cmo 5 U nucleoside in solution revealed that cmo 5 U adopts the C2'-endo ribose pucker conformation, providing mechanistic insight into base pairing with pyrimidines 9 . The crystal structures of codon-anticodon interactions at the ribosomal A-site revealed that the O5 atom of cmo 5 U is involved in an intramolecular hydrogen bond that pre-structures the anticodon loop, enabling cmo 5 U to pair with pyrimidines at the third position of the codon 19 . In addition, cmo 5 U pairs with G in standard Watson-Crick geometry, rather than classical U-G wobble geometry, indicating that keto-to-enol tautomeric conversion of the uracil base is involved in this base pairing interaction 19 .
Multistep reactions are involved in the biosynthetic pathway of cmo 5 U and mcmo 5 U in bacterial tRNAs (Fig. 1c). Three enzymes, CmoA, CmoB, and CmoM, are involved in the second to last step in the pathway. Initially, U34 is hydroxylated to form 5-hydroxyuridine (ho 5 U) in a reaction catalyzed by unidentified factors. Chorismate, an end product of the shikimate pathway, is involved in this step, but the source of the oxygen atom remains unknown 20 . On the other hand, the subsequent steps have been elucidated. In the second step, a unique carboxymethyl donor, Scarboxymethyl-S-adenosyl-L-homocysteine (SCM-SAH, also called Cx-SAM) is synthesized from AdoMet and prephenate in a reaction catalyzed by CmoA 21,22 . CmoB then transfers the carboxymethyl group of SCM-SAH to ho 5 U34 to form cmo 5 U34 on tRNA 21 . In four tRNA species (for Ser1, Pro3, Thr4, and Ala1), cmo 5 U is further methylated by CmoM to form mcmo 5 U34 13 . The terminal methylation of mcmo 5 U in tRNA Pro3 dynamically alters its modification frequency in a growth phase-dependent manner 13 .
Hydroxylation of RNA molecules is a major posttranscriptional modification that plays roles in multiple biological contexts. In addition to xo 5 U modification of bacterial tRNAs, RNA hydroxylation is also responsible for the biogenesis and function of eukaryotic RNA modifications. The JmjC domaincontaining protein TYW5 is responsible for biogenesis of hydroxywybutosine (OHyW) at position 37 of mammalian tRNA Phe23 . ALKBH1 catalyzes tRNA hydroxylation to form 5hydroxymethyl-2'O-methylcytidine (hm 5 Cm) and 5-formyl-2'Omethylcytidine (f 5 Cm) in cytoplasmic tRNA Leu , as well as 5formylcytidine (f 5 C) in mammalian mitochondrial tRNA Met24 . ALKBH1-knockout cells exhibit respiratory defects, indicating that ALKBH1 is required for efficient mitochondrial activity. Recently, we discovered hydroxy-N 6 -threonylcarbamoyladenosine (ht 6 A) in echinoderm mitochondrial tRNA; 25 this modification alters the genetic code in echinoderm mitochondria. N 6methyladenosine (m 6 A) and 1-methyladenosine (m 1 A) are demethylated by ALKBH family proteins via hydroxymethyl formation 26,27 . The Tet protein is responsible for forming 5hydroxymethylcytidine (hm 5 C) in Drosophila mRNAs 28 . All of the hydroxylases, including the JmjC, ALKBH, and Tet families, are Fe(II)-and 2-oxoglutarate (2-OG)-dependent oxygenases that use a molecular oxygen as a substrate for hydroxylation 29 . Recently, we identified RlhA as a factor responsible RNA hydroxylation to form 5-hydroxycytidine (ho 5 C) at position 2501 in E. coli 23S rRNA 30 . RlhA does not have any characteristic motifs conserved among the known RNA oxygenases, but it does belong to a family of proteins bearing the peptidase U32 motif. This finding encouraged us to explore other peptidase U32containing proteins responsible for hydroxylation of biological molecules.
To search for genes responsible for RNA modifications, we developed a method called ribonucleome analysis to screen knockout strains for uncharacterized genes by liquid chromatography-mass spectrometry (LC/MS) 31 . If a target RNA modification is absent in a certain knockout strain, we can identify the gene dedicated to biogenesis of the target RNA modification via reverse genetics. Using this approach, we have discovered dozens of genes responsible for RNA modifications in tRNAs 13,23,[32][33][34][35][36][37][38][39] and rRNAs 30,40,41 . In this study, we apply ribonucleome analysis in conjunction with comparative genomics to identify two genes, trhP and trhO, which are responsible for formation of ho 5 U at the initial step of xo 5 U biogenesis at the wobble position of bacterial tRNAs. TrhP, a peptidase U32 family protein, is involved in prephenate-dependent ho 5 U34 formation. TrhO, a rhodanese family protein, catalyzes oxygen-dependent ho 5 U34 formation under aerobic conditions. These two pathways play redundant roles in ho 5 U34 formation. Double knockout of both enzymes causes a temperature-sensitive phenotype and decreases the efficiency with which codons ending in G (GCG and UCG) were decoded, indicating that ho 5 U34 formation ensures efficient decoding during translation.

Results
E. coli yegQ is involved in tRNA hydroxylation. We recently reported that RlhA is responsible for ho 5 C formation at position 2501 in E. coli 23S rRNA 30 . RlhA belongs to a family of proteins that contain the peptidase U32 motif. This finding prompted us to speculate that other paralogs of peptidase U32-containing proteins are involved in ho 5 U34 formation in tRNAs. The E. coli genome contains four paralogs of peptidase U32-containing proteins: rlhA, yegQ, yhbU, and yhbV ( Supplementary Fig. 1, Supplementary Data 1). We extracted total RNA from each of the respective knockout strains, digested with RNase T 1 , and subjected the digests to capillary LC-nano-ESI-mass spectrometry (RNA-MS) to detect tRNA fragments containing cmo 5 U (shotgun analysis). RNA fragments were detected as multiply charged negative ions (Supplementary Table 1). We clearly detected an anticodon-containing RNA fragment of tRNA Val1 from total RNA of wild-type (WT) E. coli cells (Fig. 2a). As reported previously 13 , cmo 5 U34 was present as a major wobble modification in this tRNA, whereas little unmodified fragment (U34) was detected. In a knockout strain lacking yegQ (ΔyegQ), which encodes a peptidase U32-containing protein, cmo 5 U34 frequency decreased markedly, to about 30% of the WT level, and a corresponding fragment containing unmodified U34 appeared, indicating that yegQ is partially responsible for the initial step of cmo 5 U34 formation. We reasoned that other paralogs of peptidase U32-containing proteins might be involved in this process. Accordingly, we constructed a quadruple mutant strain, ΔyegQ/ ΔyhbU/ΔyhbV/ΔrlhA, lacking all paralogs of peptidase U32containing proteins, and analyzed the RNA fragment of tRNA-Val1 . However, the cmo 5 U34-containing fragment persisted in this strain (Fig. 2a)  Anticodon-codon pairing in six codon boxes decoded by tRNAs bearing cmo 5 U and mcmo 5 U. Each of the three tRNA genes in parenthesis was individually deleted. In the family boxes (four codons specify a single amino acid), many codons are redundantly decoded by two or three isoacceptors (tRNAs charging with the same amino acid). For example, CUU codon is decoded by tRNA Leu2 with GAG anticodon as well as tRNA Leu3  To analyze the modification status of each tRNA species in the ΔyegQ strain, we used the reciprocal circulating chromatography (RCC) method 42 to isolate six tRNA species (tRNA Ala1 , tRNA Leu3 , tRNA Pro3 , tRNA Ser1 , tRNA Thr4 , and tRNA Val1 ) bearing a cmo 5 U34 or mcmo 5 U34 modification from both WT and ΔyegQ strains. Each tRNA was digested by RNase T 1 and subjected to RNA-MS to analyze the anticodon-containing fragments (Fig. 2b, Supplementary Fig. 2, Supplementary Table 1). Hypomodified fragments were further sequenced by collisioninduced dissociation (CID) analysis ( Supplementary Fig. 3). In tRNA Ala1 , mcmo 5 U34 was present as a major wobble modification in the WT, but its levels were reduced to about 50% (concomitant with appearance of the unmodified fragment) in ΔyegQ (Fig. 2b). Similarly, mcmo 5 U34 levels were reduced in    Supplementary  Fig. 2). The level of cmo 5 U34, a major modification in WT tRNA Val1 , was also reduced about 50% in ΔyegQ (Fig. 2c, Supplementary Fig. 2). By contrast, the levels of cmo 5 U34 of tRNA Leu3 and mcmo 5 U34 of tRNA Pro3 dropped sharply, to less than 5% of WT levels, in the ΔyegQ strain ( Fig. 2c, Supplementary  Fig. 2). These results indicate that yegQ is involved differently in ho 5 U34 formation in each tRNA. In particular, tRNA Leu3 and tRNA Pro3 are major targets of YegQ. Accordingly, we renamed yegQ as trhP (tRNA hydroxylation P).
Biogenesis of mo 5 U34 in Bacillus subtilis. Instead of cmo 5 U34 or mcmo 5 U34, B. subtilis uses mo 5 U34 as an xo 5 U-type wobble modification. Upon depletion of intracellular AdoMet, mo 5 U is replaced by ho 5 U 43 , indicating that ho 5 U is the precursor and AdoMet is the methyl donor for mo 5 U formation. We identified two orthologs of trhP in B. subtilis, yrrN and yrrO (Supplementary Fig. 1, Supplementary Data 1), both of which are encoded tandemly in the same operon (Fig. 2d). YrrO has a peptidase U32 motif and a characteristic C-terminal motif also found in E. coli TrhP, whereas YrrN only has a peptidase U32 motif. We investigated whether these two genes are actually involved in mo 5 U34 formation by shotgun analysis of total RNAs extracted from B. subtilis ΔyrrN and ΔyrrO strains. Similar to our observation in the E. coli ΔtrhP strain (Fig. 2a), mo 5 U34 frequency in tRNA Val1 decreased, and an RNA fragment containing unmodified U34 appeared, in both ΔyrrN and ΔyrrO strains (Fig. 2e), indicating that both yrrN and yrrO were necessary for hydroxylation of U34 to form ho 5 U34 in B. subtilis. Accordingly, we renamed yrrO and yrrN as trhP1 and trhP2, respectively. To determine whether these two paralogs act redundantly to form ho 5 U34, we constructed the doubledeletion strain ΔtrhP1/ΔtrhP2 and analyzed the modification status of tRNA. The same level of mo 5 U34 was detected in this strain as in the single-knockout strains (Fig. 2e), suggesting that both paralogs are involved in synthesizing about 50% of mo 5 U34 in the cell, and another pathway is required for the remainder, as also observed for formation of cmo 5 U34 in E. coli.
We then focused on yrrM, which is encoded within the same operon as trhP1 and trhP2 (Fig. 2d). B. subtilis YrrM has high sequence similarity to the AdoMet-dependent catechol Omethyltransferase that catalyzes methylation of the hydroxyl group on the aromatic ring, suggesting that this protein is a methyltransferase responsible for mo 5 U34 formation. As expected, mo 5 U34 in tRNA Val1 disappeared and was converted to ho 5 U34 in a ΔyrrM strain (Fig. 2e). We then generated recombinant YrrM protein and performed in vitro reconstitution of mo 5 U34 in E. coli tRNA Thr4 containing ho 5 U34, which had been isolated from a ΔcmoB strain. mo 5 U34 was successfully synthesized in the presence of both YrrM and AdoMet ( Supplementary Fig. 4). The product was confirmed by CID analysis (Supplementary Fig. 4). Taken together, these findings indicate that YrrM is an AdoMet-dependent methyltransferase that converts ho 5 U34 to mo 5 U34 in B. subtilis. During the preparation of this manuscript, B. subtilis yrrM was demonstrated to be a ho 5 U34-methyltransferase and renamed trmR 44 .
Alternative pathway for tRNA hydroxylation. E. coli trhP and B. subtilis trhP1/trhP2 are partially involved in the initial step of cmo 5 U34 and mo 5 U34 formation, respectively, indicating the existence of another redundant pathway for ho 5 U34 formation in both species. To search for the gene(s) responsible for this pathway, we used a comparative genomic approach. In bacteria, we found several organisms with cmoA and cmoB homologs, but no trhP homolog (Fig. 3a). Similarly, some bacteria had a trmR homolog, but no trhP homologs (Fig. 3a). Given that ho 5 U34 is a common precursor for cmo 5 U34 synthesis mediated by cmoA and cmoB, and mo 5 U34 synthesis mediated by trmR, the bacterial species lacking trhP homologs should have another gene responsible for ho 5 U34 formation independent of the trhP pathway. We identified seven bacterial species with cmoAB or trmR homologs but no trhP or trhP1/trhP2 homologs (Fig. 3a). Among 4746 E. coli ORFs, we selected 141 genes ( Fig. 3b) commonly present in all seven species, as well as in B. subtilis. We then narrowed down the list of candidates to seven genes with unknown functions (Fig. 3b). Among them, a yceA homolog was identified as a strong candidate because its genomic locus is close to that of cmoA in cyanobacteria, and yceA is encoded as a fusion protein with trmR in three bacterial species, Phytophthora sojae, Phytophthora ramorum, and Phaeodatylum tricornutum (Fig. 3b). To determine whether yceA is responsible for the second pathway of ho 5 U34 formation, we isolated tRNA Val1 from E. coli knockout strains of trhP and yceA, and analyzed the status of wobble modifications. The residual cmo 5 U34 observed in the ΔtrhP strain completely disappeared in the double-deletion strain ΔtrhP/ΔyceA (Fig. 3c), indicating that yceA is responsible for the second pathway of ho 5 U34. Hereafter, we refer to yceA as trhO (tRNA hydroxylation O). However, the amount of unmodified U34-containing fragment increased slightly in the single-deletion strain ΔtrhO (Fig. 3c), indicating that the trhP-mediated pathway plays the predominant role in ho 5 U34 formation especially in the absence of trhO.
B. subtilis ybfQ is an ortholog of E. coli trhO. To determine whether ybfQ is involved in the second pathway for ho 5 U34 formation in B. subtilis, we constructed a triple-knockout strain, ΔtrhP1/ΔtrhP2/ΔybfQ, and analyzed the status of tRNA wobble Fig. 2 Identification of trhP responsible for tRNA hydroxylation. a Mass-spectrometric shotgun analysis of total tRNAs in E. coli strains. Extracted ion chromatograms (XICs) show multiply charged negative ions of the anticodon-containing fragments of tRNA Val1 with U34 (upper panels) and cmo 5 U34 (lower) in total tRNAs from wild-type (left), ΔyegQ (center), and ΔyegQ/ΔyhbU/ΔyhbV/ΔrlhA strains (right). Sequence, m/z value, and charge state of each fragment are shown on the right. Asterisks indicate nonspecific peaks with the same m/z values. b Mass-spectrometric analysis of the wobble modification in E. coli tRNA Ala1 isolated from WT (left panels) and ΔyegQ (right) strains. XICs show anticodon-containing fragments of tRNA Ala1 with U34 (top panels), cmo 5 U34 (middle panels), and mcmo 5 U34 (bottom panels). The cleavage sites of RNase T 1 are shown in Fig. 1a. It is assumed that cmo 5 U detected in this tRNA was generated by artificial hydrolysis of mcmo 5 U during tRNA isolation 13 . The black arrowhead indicates a small peak corresponding to the U34-containing fragment detected in the WT. c Modification frequencies of cmo 5 U derivatives at the wobble position of six tRNA species isolated from WT and ΔyegQ strains. Relative compositions of each modification were calculated from the peak area ratio of mass chromatograms of RNase T 1 digest fragments containing mcmo 5 Um (red), mcmo 5 U (green), cmo 5 U (blue), or U (gray) at the wobble position ( Supplementary Fig. 2 . trhP orthologs were mainly detected in γand β-proteobacteria, in addition to some desulfobacteria in δ-proteobacteria. trhP1/trhP2 orthologs always co-occurred, and were present in phylum Firmicutes and some members of Tenericutes and Cyanobacteria. Given that TrhP1 is a related family with TrhP ( Supplementary  Fig. 1), TrhP1 might have branched out from TrhP, and evolved to require a paralogous protein TrhP2 that might be generated by gene duplication. Supporting this speculation, trhP1 and trhP2 are tandemly encoded in the same operon in B. subtilis (Fig. 2e). Also, we can explain the reason why trhP orthologs and trhP1/trhP2 orthologs show a mutually exclusive distribution in bacteria ( Supplementary Fig. 6). Intriguingly, over half of organisms bearing trhP or trhP1/trhP2 orthologs also harbor trhO orthologs [49 of 84 (58%) organisms bearing trhP, P = 0.0003 (Fisher's exact test); 22 of 32 (69%) organisms bearing trhP1/trhP2, P = 0.0007 (Fisher's exact test)] ( Supplementary  Fig. 7). This significant overlap suggests that harboring both pathways for tRNA hydroxylation might help organisms to adapt to two different environments, i.e., aerobic and anaerobic conditions.
Phenotypes of E. coli strains lacking tRNA hydroxylation. We then measured the growth rate of a series of E. coli knockout strains involved in (m)cmo 5 U34 modifications. No growth reduction was observed in the ΔcmoM strain, as reported previously 13 (Fig. 4a), indicating that the terminal methylation of mcmo 5 U34 has little impact on cell growth. A slight increase of doubling time was observed in the ΔcmoB strain, in which ho 5 U34 accumulated, indicating that the carboxymethyl group of (m)cmo 5 U34 contributes to efficient growth of E. coli cells. Notably, the ΔtrhP/ΔtrhO strain grew more slowly than the WT and ΔcmoB strains, providing a clear evidence for the functional importance of 5-hydroxyl group of (m)cmo 5 U34 in cells. To characterize phenotypic features of the ΔtrhP/ΔtrhO strain, we further knocked out tRNA genes responsible for G-ending codons [serU (tRNA Ser2 ) for the UCG codon, thrW (tRNA Thr2 ) for the ACG codon, and proK (tRNA Pro1 ) for the CCG codon], because these codons are redundantly deciphered by the respective tRNA and the isodecoder with the (m)cmo 5 U34 modification in each codon box. Growth reduction of the ΔtrhP/ΔtrhO strain relative to the ΔcmoB strain was observed upon knockout of serU, thrW, and proK (Fig. 4a), indicating that the 5-hydroxyl group of (m) cmo 5 U34 plays a critical role in deciphering G-ending codons, especially in the absence of the respective isodecoder.
Next, we examined the temperature sensitivity of a series of knockout strains involved in (m)cmo 5  on LB plates at 30°C, but slowly at 37˚C and not at all at 42°C (Fig. 4b). The growth defect of this strain at 42˚C was restored by introduction of plasmid-encoded trhP ( Supplementary Fig. 8), indicating that the temperature-sensitive phenotype of this strain can be attributed to hypomodification of tRNA Ser1 . Curiously, the ΔtrhP/ΔserU strain did not exhibit temperature sensitivity, although the frequency of mcmo 5 U34 actually decreased in tRNA Ser1 (Fig. 2c), indicating that trhO-mediated hydroxylation compensates for the growth defect when the trhP-mediated hydroxylation pathway is impaired. This result highlights the importance of redundant hydroxylation pathways for formation of xo 5 U. Similarly, we observed a severe growth reduction in the ΔtrhP/ΔtrhO/ΔthrW strain cultivated on M9 minimum plates, even at 40°C (Fig. 4c)   and ΔcmoB strain exhibited a growth defect. These results clearly demonstrated that the absence of the 5-hydroxyl group of (m) cmo 5 U34 causes a temperature-sensitive phenotype, especially in the absence of the isodecoder responsible for G-ending codons.
Decoding properties without tRNA hydroxylation. To investigate the functional role of (m)cmo 5 U34 modification in terms of decoding efficiency, we conducted dual-luciferase reporter assays based on the RF2 recoding system 18,45 . The reporter constructs consisted of Renilla luciferase (Rluc) fused with firefly luciferase (Fluc) in a +1 frame via a slippery linker derived from the +1 frameshift signal of the RF2 recoding site, so that Fluc expression requires a +1 frameshift at the linker sequence. The UGA codon at the recoding site was substituted with GCG and UCG as test codons to examine their ability to be decoded by tRNA Ala1 and tRNA Ser1 , respectively (Fig. 4d). We also prepared a control reporter in which the recoding site was replaced with GG (zero frame). These reporters were introduced to a series of E. coli knockout strains lacking genes involved in (m)cmo 5 U34 modifications. The decoding ability of tRNAs with different wobble modifications at each test codon of the frameshift site was negatively correlated with the +1 frameshift activity, measured by Fluc activity, because in this system the +1 frameshift activity is promoted by the hungry A-site. The +1 frameshift activity was calculated by normalizing the Fluc signal against the Rluc signal (F/R value). No difference in F/R value in the control construct (zero frame) was observed in any strains (Fig. 4e).
Characterization of trhP-mediated tRNA hydroxylation. According to our recent study 30 , the shikimate pathway is associated with rlhA-mediated ho 5 C2501 formation in 23S rRNA. A series of genetic studies revealed that prephenate is an essential metabolite for the first step of this modification. Given that TrhP belongs to a family of peptidase U32-containing proteins, we asked whether prephenate is also required for trhP-mediated ho 5 U34 formation. Consistent with this possibility, previous studies reported that the initial step of cmo 5 U34 formation is associated with chorismate biogenesis in E. coli, B. subtilis, and Salmonella typhimurium 20,46 . Chorismate is an end product of the shikimate pathway and a common precursor for aromatic amino acids and vitamins in bacteria and plants 47 . Shotgun analyses of E. coli total RNA revealed that cmo 5 U34 formation was significantly impaired in an ΔaroC strain, in which no chorismate was produced ( Supplementary Fig. 9). However, ho 5 U34 was still present in this strain because ho 5 U34 was redundantly synthesized by the trhO-mediated pathway. By contrast, as expected, no ho 5 U34 was detected in an ΔaroC/ΔtrhO strain ( Supplementary Fig. 10). In E. coli, chorismate is converted into five metabolites: isochorismate (catalyzed by the products of entC and menF), 4-hydroxybenzoate (ubiC), 4-amino-4deoxychorismate (pabB), anthranilate (trpE), and prephenate (pheA and tyrA) (Fig. 5a). Among these pathways, the ubiCmediated pathway was excluded because 4-hydroxybenzoate does not restore cmo 5 U34 formation in Salmonella ΔaroD strain 46 . As observed in the ΔaroC strain, cmo 5 U34 formation was only impaired in the ΔpheA/ΔtyrA strain, but not in the ΔentC/ ΔmenF, ΔpabB, and ΔtrpE strains ( Supplementary Fig. 9), indicating that prephenate or its downstream metabolites are required for ho 5 U34 formation.
Prephenate is converted to downstream metabolites via three pathways (Fig. 5a). cmo 5 U34 formation was unchanged in ΔtyrA and ΔtyrB strains (Supplementary Fig. 9). pheA encodes a fusion of chorismate mutase (CM) and prephenate dehydratase (PDT), which synthesize prephenate and phenylpyruvate, respectively (Fig. 5a). To dissect these two enzymatic activities, we constructed a pheA variant possessing only the CM activity [pheA(CM)] by introducing an active-site mutation in the PDT domain 48 . To determine whether prephenate is responsible for ho 5 U34 formation, we constructed the quadruple-knockout strain ΔpheA/ΔtyrA/ΔcmoA/ΔtrhO, and then introduced plasmidencoded pheA(CM), resulting in accumulation of prephenate. ho 5 U34 levels were restored relative to those in a mock Fig. 4 Phenotypes of E. coli strains lacking tRNA hydroxylation. a Growth rates of E. coli strains with different wobble modifications. Doubling time of WT (mcmo 5 U34 for tRNA Ser1 , tRNA Thr4 , and tRNA Pro3 ), ΔcmoM (cmo 5 U34), ΔcmoB (ho 5 U34), and ΔtrhP/ΔtrhO (U34) in the presence (left) or the absence of tRNA isodecoders at 37°C in liquid LB medium: tRNA Ser2 (middle left), tRNA Thr2 (middle right), or tRNA Pro1 (right). Individual data (dot plots) and their means ± SD (bar graph) are presented (n = 4). *P < 0.01; **P < 0.001 between two data series (Student's t-test, one-sided). b Growth of WT, ΔcmoM, ΔcmoB, ΔtrhP/ΔtrhO, ΔtrhP, and ΔtrhO strains in the absence of tRNA Ser2 (ΔserU). The expected wobble modification status of each strain is shown on the right. Each strain was serially diluted (1:10), spotted onto LB agar plates, and cultivated for 11 h (30°C), 8 h (37°C) or 8 h (42°C). c Growth of WT, ΔcmoM, ΔcmoB, ΔtrhP/ΔtrhO, ΔtrhP, and ΔtrhO strains in the absence of tRNA Thr2 (ΔthrW). The expected wobble modification of each strain is shown on the right. Each strain was serially diluted (1:10), spotted onto M9 minimum agar plates, and cultivated for 31 h (30°C), 21 h (37°C), 26 h (40°C) or 31 h (42°C). d Schematic of the reporter construct for the dual-luciferase assay, based on the RF2 recoding system. SD, Shine-Dalgarno sequence. Renilla and firefly luciferases were fused with a linker containing the +1 frameshift inductive signal of the RF2 recoding site. The frameshift target site was replaced with a GCG codon for tRNA Ala1 , a UCG codon for tRNA Ser1 , or GG (zero frame) for the control. e Relative pausing activities at the frameshift site with a GCG codon (left), a UCG codon in the presence (middle left) or absence (middle right) of isodecoder tRNA Ser2 , or zero frame (right) were calculated by normalizing Fluc activity vs. Rluc activity and further normalizing against the WT activity in each graph. Individual data and their means ± SD (n = 4) are presented. *P < 0.01 between two data series (Student's t-test, two-sided). All source data for Fig. 4 are provided as a Source Data file transformant (Fig. 5b). Furthermore, when prephenate was directly added to a culture medium of the quadruple-knockout strain, ho 5 U34 clearly appeared (Fig. 5c). These results demonstrated that prephenate is required for ho 5 U34 formation.
To characterize the peptidase U32 domain of trhP, we mutated each of six conserved residues in this domain (Fig. 5d,  Supplementary Fig. 11) and examined their activities in vivo by complementation of the ΔtrhP/ΔtrhO strain. Shotgun analyses revealed that cmo 5 U34 was fully restored by WT trhP, but not by any of the mutants examined in this study (Fig. 5e) (Fig. 6a), whereas no reduction in cmo 5 U34 was observed in WT cells ( Supplementary Fig. 12), implying that O 2 is necessary for trhO-mediated ho 5 U34 formation. We then metabolically labeled ho 5 U34 using 18 O-labeled O 2 . To accumulate ho 5 U34, we cultured the E. coli ΔtrhP/ΔcmoB strain in mixed gas containing 20% 18 O 2 . Total RNA extracted from this culture was digested into nucleosides and subjected to LC/MS. We clearly detected a deprotonated ho 5 U34 nucleoside with molecular mass (m/z 261) 2 Da greater than that of the naturally occurring nucleoside (Fig. 6b). CID analysis of the nucleoside revealed that the 18 O atom was present in the base moiety of ho 5 U ( Supplementary  Fig. 13). No oxygens in the uracil base were labeled with 18 O under the 18 O-air condition ( Supplementary Fig. 13), demonstrating that trhO-mediated tRNA hydroxylation utilizes O 2 as an oxygen atom donor. Next, we generated recombinant E. coli TrhO and subjected it to biochemical characterization. Electrophoretic mobility shift assay (EMSA) revealed that TrhO interacted with tRNA Ala1 , but not with tRNA Leu3 (Fig. 6c). This result is consistent with our observation that cmo 5 U34 of tRNA Leu3 is mainly synthesized via the trhP-mediated pathway, whereas mcmo 5 U34 of tRNA Ala1 is redundantly synthesized by both the trhP-and trhO-mediated pathways. We then attempted to reconstitute ho 5 U in vitro, and successfully detected ho 5 U on tRNA Ser1 in the presence of recombinant TrhO. When TrhO was present in excess, 75% of the tRNA had ho 5 U (Fig. 6d). CID analysis of the modified fragment confirmed that ho 5 U was formed at position 34 of the tRNA (Supplementary Fig. 14). This result demonstrated that TrhO is an RNA hydroxylase that can form ho 5 U34 on tRNA by utilizing O 2 as an oxygen donor.
TrhO is a rhodanese family protein characterized by the CXGGXR motif 49 , with an active-site cysteine in a glycine-rich loop (Fig. 6e, Supplementary Fig. 15). The crystal structure of Legionella pneumophila Lpg2838, a TrhO homolog, was solved by structural genomics (Supplementary Fig. 16) 50 . In the rhodanese domain, the active-site loop structure with the CTGGIR sequence (positions 200-205 in E. coli numbering) is formed by hydrogen bonds between T201 and R205 (corresponding to T178 and R182 in Supplementary Fig. 16). To characterize this motif, we constructed a series of E. coli trhO mutants in which C200, T201, G202, G203, R205, or C206 was mutated to alanine, and then examined their activities in vivo by complementation of the ΔtrhP/ΔtrhO strain. The level of cmo 5 U34 in tRNA Val1 was partially restored by WT trhO (Fig. 6f). For the trhO mutants, C200, T201, G203, and R205 were essential for trhO-mediated tRNA hydroxylation, whereas G202 and C206 were not ( fig. 6f). In the TrhO structure, a positively charged β-sheet is present near the active-site loop ( Supplementary Fig. 16), providing a surface capable of recognizing tRNAs. The trhO mutants K112A and R114A, which lack positively charged residues on the β-sheet surface, hardly rescued or did not rescue cmo 5 U34 formation, respectively (Fig. 6f), suggesting that this surface is required for trhO-mediated tRNA hydroxylation.

Discussion
In this study, we identified two independent pathways, mediated by trhP and trhO, involved in tRNA hydroxylation in the early steps of (m)cmo 5 U34 formation. We confirmed that (m) cmo 5 U34 was completely converted to unmodified U34 in a ΔtrhP/ΔtrhO strain. This finding enabled us to analyze the physiological roles of (m)cmo 5 U34. This strain grew more slowly than a ΔcmoB strain, which has ho 5 U34, confirming the physiological importance of the O5 oxygen atom of (m)cmo 5 U34. In addition, the ΔtrhP/ΔtrhO strain exhibited severe growth defects and temperature-sensitive phenotypes when each of the tRNA genes (serU, thrW, and proK) responsible for G-ending codons was simultaneously deleted. These genetic interactions strongly indicate that (m)cmo 5 U34 plays a functional role in efficiently deciphering G-ending codons. The luciferase reporter assay revealed that decoding of UCG was significantly impaired in the ΔtrhP/ΔtrhO strain relative to the ΔcmoB strain in the absence of tRNA Ser2 , suggesting that the ho 5 UGA anticodon decodes the UCG codon more efficiently than the UGA anticodon. These observations demonstrate the direct involvement of the O5 oxygen of the xo 5 U34 modification in codon recognition in vivo.
According to the crystal structure of the ribosome 30S subunit in complex with the anticodon-stem loop (ASL) of tRNA, cmo 5 U-G pairing forms Watson-Crick geometry ( Supplementary  Fig. 17), which is more stable than U-G wobble geometry due to the stacking interaction with the neighboring base pair (i.e., the second letters of codon and anticodon) 19 . The O5 oxygen of cmo 5 U34 makes a hydrogen bond with 2' OH of U33 to prestructure the ASL, presumably reducing entropic cost to base pair with any codons. In addition, the O5 oxygen may induce keto-toenol tautomeric conversion of the uracil base to stabilize cmo 5 U-G pairing in Watson-Crick geometry. Moreover, because (m) cmo 5 U34 decodes G-ending codons more efficiently than Fig. 5 Characterization of trhP-mediated tRNA hydroxylation. a A shikimate pathway and related metabolism. Chemical structures of metabolites and the responsible genes (italicized) at each step are shown. Two or three genes at each step indicate a redundant pathway [e.g., prephenate is redundantly synthesized from chorismate mediated by pheA(CM) and tyrA]. Black or gray arrows represent pathways indispensable or dispensable for trhP-mediated ho 5 U34 formation, respectively ( Supplementary Fig. 9). White arrows represent pathways not examined in this study. b Genetic complementation of ho 5 U34 formation. Mass-spectrometric shotgun analysis of total tRNAs obtained from the E. coli ΔpheA/ΔtyrA/ΔcmoA/ΔtrhO strain transformed with a control plasmid (left panels) or pMW-pheA(CM) (right panels). XICs show multiply charged negative ions of anticodon-containing fragments of tRNA Val1 with U34 (upper panels) and ho 5 U34 (lower panels). Sequence, m/z value, and charge state of each fragment are shown on the right. c Metabolic complementation of ho 5 U34 formation. Mass-spectrometric shotgun analysis of total tRNAs obtained from the E. coli ΔpheA/ΔtyrA/ΔcmoA/ΔtrhO strain cultured in the absence (left panels) or presence (right panels) of 1 mM prephenate. XICs show multiply charged negative ions of anticodon-containing fragments of tRNA Val1 with U34 (upper panels) and ho 5 U34 (lower panels). Sequence, m/z value, and charge state of each fragment are shown on the right. d Domain organization of E. coli TrhP, which contains Peptidase_U32 (PF01136) and Peptidase_U32_C (PF16325) domains. Six residues in the Peptidase_U32 domain that are essential for TrhP-mediated hydroxylation are indicated. e Mutation study of trhP. Mass-spectrometric shotgun analysis of total tRNA in the E. coli ΔtrhP strain transformed with plasmid-encoded trhP WT or mutants, as indicated. XICs show multiply charged negative ions of the anticodon-containing fragments of tRNA Val1 with U34 (black lines) and cmo 5 U34 (red lines) in total tRNAs. Sequence, m/z value, and charge state of each fragment are shown on the right ho 5 U34 13,18 , the carboxymethyl group and terminal methylation of (m)cmo 5 U34 contribute further to efficient codon recognition. Structural analysis has shown that the carboxymethyl group of cmo 5 U34 forms a hydrogen bond with the O4 carbonyl oxygen of U in the first letter of the codon, implying that the cmo 5 U side chain is directly involved in codon recognition 19 . Because ho 5 U34 has a phenolic hydrogen, and the pK a value of O5 is 7.78 51 , ho 5 U34 is ionized to some extent under neutral pH conditions. Thus, the carboxymethylation of cmo 5 U34 and methylation of mo 5 U34 might confer efficient codon recognition by suppressing the ionization of ho 5 U34.
Our findings reveal that the TrhP-dependent pathway requires prephenate, whereas the TrhO-dependent pathway requires molecular oxygen. Thus, these two pathways are biochemically independent with respect to their requirement for metabolites. The existence of redundant and robust pathways for ho 5 U formation emphasizes that the xo 5 U34 modification is essential for bacteria to survive in a harsh environment. This is the unique instance of the RNA modification synthesized by two independent pathways in the same organism. According to phylogenetic distribution analysis in all domains of life ( Supplementary Fig. 6), some organisms possess both trhP and trhO genes, whereas other c d e f organisms possess just one of them. Considering that anaerobes preceded aerobes in the early evolution on Earth, the trhP pathway might have been established in anaerobic bacteria before the trhO pathway arose, assuming that xo 5 U34 was present in such ancestral organisms. Presumably, the trhO pathway was acquired by aerobic bacteria after the O 2 concentration increased on Earth. The trhP pathway is required for anaerobic bacteria, whereas the trhP and trhO dual pathways are useful for organisms that live in both anaerobic and aerobic environments. TrhP is a peptidase U32-containing protein. Phylogenetic analysis has shown that peptidase U32-containing proteins can be classified into 12 subfamilies (Supplementary Fig. 1) 30 . We showed previously that three of these families include the RlhA proteins (RlhA1, RlhA2a, and RlhA2b) responsible for ho 5 C formation in 23S rRNA 30 . It is plausible that other subfamilies are also involved in hydroxylation of RNA or other biomolecules. Clostridia species harbor a member of the PepU32#1 family and trmR, but no homologs of trhO or trhP, implying that mo 5 U34 is present and that PepU32#1 family proteins are functional homologs of trhP in these species. Helicobacter pylori, a representative of the ε-proteobacteria, possesses a PepU32#2 family protein (HP0169), cmoA, and cmoB, but no homologs of trhO or trhP, indicating that cmo 5 U34 is present and suggesting that HP0169 is responsible for prephenate-dependent ho 5 U34 formation in this species. Intriguingly, HP0169 is required for gastric colonization by H. pylori 52 . Similarly, in Salmonella enterica, a trhP ortholog is associated with chicken macrophage infection 53 . Together, these findings suggest that xo 5 U34 contributes to bacterial infection and pathogenesis.
We found that prephenate is required for trhP-dependent ho 5 U34 formation. Given that prephenate is also a substrate for cmo 5 U34 formation mediated by CmoA and CmoB 21 , it is a critical metabolite involved in the entire pathway of cmo 5 U34 biogenesis. Prephenate is generated from chorismate, which in turn is a common precursor of multiple metabolites, including aromatic amino acids, quinones, folate, and siderophores 47 . Thus, cmo 5 U34 modification might be tightly associated with the shikimate pathway and biogenesis of aromatic amino acids. The frequency of cmo 5 U34 might be regulated by the cellular concentration of prephenate under some environmental stress conditions.
TrhP is a paralog of RlhA in the same family of peptidase U32containing proteins. RlhA is responsible for prephenatedependent ho 5 C2501 formation in E. coli 23S rRNA 30 , strongly implicating the involvement of the peptidase U32 motif in the C5-hydroxylation of pyrimidine base. Here, we showed that three conserved residues (E162, C170, and C177) in the motif of TrhP are essential for ho 5 U34 formation. Additionally, the corresponding residues (E161, C169, and C176) in RlhA were also required for ho 5 C2501 formation 30 , demonstrating that the peptidase U32 motif is directly involved in the hydroxylation of RNA. To date, we have no evidence that TrhP and RlhA directly catalyze the hydroxylation of RNA molecules. Given that RlhA is directly bound to the 50S subunit and its precursor in the cell, RlhA might be the hydroxylase responsible for ho 5 C2501 formation. By analogy, TrhP might be a hydroxylase for tRNA. Regarding the role of prephenate in ho 5 U34 formation, several possibilities should be considered. Prephenate might serve as an oxygen donor for ho 5 U34 formation, or alternatively as a coenzyme for the reaction. Moreover, we cannot exclude the possibility that unknown metabolites derived from prephenate are involved in ho 5 U34 formation. Further studies are necessary to elucidate the molecular mechanism underlying ho 5 U34 formation mediated by TrhP and prephenate.
trhO homologs are present in many aerobes and facultative anaerobes, but not in obligate anaerobes such as Bacteroides, Clostridium, and Bifidobacterium. trhO homologs are distributed in a wide range of bacteria, including α-, β-, and γ-proteobacteria, Bacilli, actinobacteria, the FCB group, cyanobacteria, and a subset of phylum Tenericutes (Supplementary Fig. 6). Intriguingly, trhO homologs are also widely distributed in vertebrates and other eukaryotes. This finding suggests the presence of an xo 5 U-type modification in eukaryotes.
We also showed that TrhO directly catalyzes oxygendependent ho 5 U34 formation. TrhO is related to rhodanese, which is involved in persulfide formation during detoxification of cyanide; however, the functions of most rhodanese family proteins remain unclear. In the context of RNA modifications, Tum1p is a rhodanese protein that mediates a persulfide sulfur for 2-thiouridine synthesis in eukaryotes 38 . Bacterial YbbB (also known as MnmH) is another rhodanese family protein responsible for biogenesis of 2-selenouridine (se 2 U) 54 and geranyl-2thioudirine (ges 2 U) 55 . Mutation study of TrhO revealed that the active-site loop of the rhodanese domain is responsible for ho 5 U34 formation, suggesting that the rhodanese domain plays a critical role in hydroxylation of uracil base. Future studies should seek to clarify the mechanism by which rhodanese catalyzes this reaction.
We now have a complete picture of xo 5 U34 formation in bacteria (Fig. 7). In the first step, U34 is redundantly hydroxylated by TrhP and TrhO to form ho 5 U34 in tRNAs responsible for decoding NYN codons. TrhP requires prephenate as a metabolite for ho 5 U34 formation, whereas TrhO uses a molecular oxygen for this purpose under aerobic conditions. In E. coli, TrhP Fig. 6 Characterization of trhO-mediated tRNA hydroxylation. a trhO-mediated cmo 5 U formation takes place under aerobic conditions. Mass-spectrometric shotgun analysis of total tRNAs obtained from E. coli ΔtrhP strain cultured under anaerobic (left panels) or aerobic (right panels) conditions. XICs show negative ions of the anticodon-containing fragments of tRNA Val1 with U34 (upper panels) and cmo 5 U34 (lower panels). Arrowheads indicate target peaks, and asterisks indicate unspecific peaks. b Molecular O 2 is the metabolic source of the hydroxyl group of ho 5 U34 generated by trhO. Mass-spectrometric nucleoside analyses of total RNAs obtained from the E. coli ΔtrhP/ΔcmoB strain cultured in mixed gas with 20% 18 O 2 (left panels) or in normal air (right panels). UV traces at 254 nm (upper panels) and XICs (lower panels) of deprotonated ho 5 U extracted from non-labeled (black line) or [ 18 O]-labeled (red line) cells are shown. c TrhO interacts with tRNA Ala1 specifically. Electrophoretic mobility shift assay (EMSA) was performed to detect direct interaction between recombinant TrhO and tRNA Ala1 (left panels) or tRNA Leu3 (right panels) isolated from E. coli. tRNAs and TrhO were stained with SYBR Safe (upper panels) and Coomassie brilliant blue (lower panels), respectively. TrhO-tRNA complexes are indicated by red traces. Arrowheads indicate unbound intact tRNAs and TrhO. Source data are provided as a Source Data file. d In vitro reconstitution of ho 5 U34 with recombinant TrhO. RNase T 1 digests of E. coli tRNA Ser1 transcripts (10 pmol) incubated with (10 pmol; 1 eq or 100 pmol; 10 eq) or without TrhO are subjected to RNA-MS. XICs show doubly charged negative ions of the anticodon-containing fragments of tRNA transcript with U34 (upper panels) and ho 5 U34 (lower panels). Frequencies of ho 5 U34 are indicated. e Domain organization of E. coli TrhO bearing Rhodanese (PF00581) and Rhodanese_C (PF12368) domains. The CXGGXR motif and six residues mutated in this study are indicated. f Mutation study of trhO. Mass-spectrometric shotgun analysis of total tRNAs in the E. coli ΔtrhP/ΔtrhO strain transformed with plasmid-encoded trhO WT or mutants, as indicated. XICs show multiply charged negative ions of the anticodon-containing fragments of tRNA Val1 with U34 (black lines) and cmo 5 U34 (red lines) from total tRNA is involved in ho 5 U34 formation of all six tRNA species, but has a preference for tRNA Leu3 and tRNA Pro3 , whereas TrhO mainly hydroxylates the other four species. CmoA employs prephenate and AdoMet to generate SCM-SAH, a metabolite used for carboxymethylation of ho 5 U34 to yield cmo 5 U34 catalyzed by CmoB. Four tRNAs (for Ala1, Ser1, Pro3, and Thr4) are further methylated by CmoM to yield mcmo 5 U34. mcmo 5 U34 in tRNA Ser1 is partially methylated by TrmL to yield mcmo 5 Um34 as a minor modification. In Gram-positive bacteria, including B. subtilis, ho 5 U34 is methylated by TrmR to yield mo 5 U34 instead of (m)cmo 5 U34.

Methods
Strains and media. A series of single-knockout strains of E. coli and their parent strain were obtained from the National BioResource Project (NBRP), National Institute of Genetics (NIG), Japan (Keio collection) 56 . Other knockout strains were generated by homologous recombination using λ-derived Red recombinase 57 with the chloramphenicol-resistance (Cm R ) or the kanamycin-resistance marker (Kan R ); all strains were selected with the appropriate antibiotics (20 µg per ml chloramphenicol or 50 µg per ml kanamycin). E. coli strains with multiple gene deletions were constructed by P1 transduction. The Kan R marker was removed by pCP20 transformation 56  Plasmid construction. For the genetic rescue study, CDSs of trhP, trhO, and pheA with their 200 bp upstream sequences (including native promoter regions) were PCR-amplified from the E. coli BW25113 genome and cloned into pMW118 (Nippon Gene) to yield pMW-trhP, pMW-trhO, and pMW-pheA, respectively. pMW-pheA(CM), which lacks prephenate dehydratase activity due to the T278A mutation 48 , and a series of point mutants of pMW-trhP and pMW-trhO were constructed by QuikChange site-directed mutagenesis (Agilent Technology). For expression vectors of TrhO (YceA) and TrmR (YrrM), CDSs of trhO and trmR were PCR-amplified from genomic DNAs of E. coli BW25113 and B. subtilis 168, respectively, and cloned into pET21b (Novagen) to yield pET-trhO and pET-trmR. All constructs were confirmed by Sanger sequencing. All primers used in this study are listed in Supplementary Data 3.
RNA extraction and tRNA isolation. Total RNA from each E. coli strain was extracted by phenol under acidic conditions 35 . The cells suspended in 1 × TE buffer [10 mM Tris (pH 8.0), 1 mM EDTA] was mixed with an equal volume of watersaturated phenol, followed by freeze and thaw twice and vigorous mixing for one hour at room temperature. The aqueous phase was separated by centrifugation, transferred to a new tube, washed with chloroform, and further purified using TRIzol reagent (Thermo Fisher Scientific, Waltham, MA, USA) and chloroform. RNA was recovered by 2-propanol precipitation. Individual tRNAs were isolated by reciprocal circulating chromatography (RCC) 42 using a series of DNA probes 13 (Supplementary Data 3). For RNA-MS shotgun analyses 13 , E. coli strains were cultured overnight, harvested, and resuspended in TE buffer to extract total RNA by TRIzol. The small RNA fraction was obtained as follows 13 . 50-250 μg of total RNA in 800 μl of 3 M NH 4 OAc (pH 5.3) was mixed with 640 μl (0.8 volume) of isopropanol at room temperature and centrifuged at 20,400×g for 10 min to precipitate long RNAs including rRNAs. The smaller RNA fraction was precipitated with ethanol from the supernatant. For RNA-MS shotgun analyses of B. subtilis tRNA modifications, the harvested cells were treated with 1 mg per ml lysozyme for 10 min on ice, and then subjected to TRIzol treatment to extract total RNA.  60 . Thus, we relatively quantified modification frequencies of RNA fragments with different chemical structures from their intensities of XICs.
Anaerobic cultivation. WT and ΔtrhP strains were precultured in a BioShaker G·BR-200 (TAITEC) at 37°C with rotation at 60 rpm for 24 h in 10 ml of degassed LB medium in a 10 cm dish doubly packed in Ziploc (Asahi-Kasei, Japan) with one bag of AnaeroPack-Anaero (Mitsubishi Gas Chemical, Japan) and an oxygen indicator (OXY-1, JIKCO). The preculture (100 µl) was inoculated into 10 ml of degassed LB medium, sealed with an AnaeroPack-Anaero, incubated at room temperature for 1 h to deoxidize completely, and then cultivated at 37°C overnight at 60 rpm.
Metabolic labeling analysis using 18 16 O 2 ), 80% N 2 ] was obtained commercially (Tatsuoka, Japan). The ΔcmoB/ΔtrhP stain was precultured at 37˚C overnight in LB medium containing 50 µg per ml kanamycin. The preculture (1 ml) was inoculated into 100 ml of degassed LB medium containing 1 mM uridine and 50 µg/ml kanamycin packed in a PVDF air-sampling bag (As One, Japan). Inside air was carefully removed, and then replaced once with N 2 and twice with 18 O 2 mixed gas. The bag was capped and sealed with Parafilm (Bemis), and then cultured at 37°C for 24 h at 100 rpm in a BioShaker G·BR-200 (TAITEC). Total RNA was extracted from the culture, digested into nucleosides, and analyzed by LC/MS as described above.
Purified protein was dialyzed in the individual lysis buffer, supplemented with glycerol to a final concentration of 30%, and stored at −20°C.
In vitro reconstitution of tRNA modification. For in vitro ho 5 U formation by TrhO, tRNA Ser1 transcript (10 pmol) and recombinant TrhO (10 or 100 pmol) were incubated at 37°C for 1 h in a reaction mixture (10 µl) containing 25 mM Tris-HCl (pH 7.0), 300 mM NaCl, 1 mM MgCl 2 , and 10 mM 2-mercaptoethanol. The tRNA was extracted with acidic phenol/chloroform and precipitated with ethanol, followed by RNase T 1 digestion and RNA-MS analysis as described above.
In vitro methylation of ho 5 U by TrmR was performed essentially as described 13 . E. coli tRNA Thr4 bearing ho 5 U34 was isolated from the ΔcmoB strain. The reaction mixture (10 µl) containing 10 pmol of tRNA Thr4 , 20 pmol of TrmR, 50 mM HEPES-KOH (pH 7.5), 100 mM KCl, 10 mM MgCl 2 , and 7 mM 2-mercaptoethanol was incubated at 37°C for 1 h in the presence or absence of 1 mM AdoMet.
Comparative genomics. The comparative genomics approach used to identify the trhO (yceA) gene was performed with the IMG database 61 . The gene occurrence profile was used to select seven organisms in which cmoBA or yrrM homologs were present and trhP (yegQ) or trhP1/trhP2 homologs were absent: Ehrlichia canis str. Jake, Erwinia pyrifoliae DSM12163, Alcanivorax borkumensis SK2, Candidatus Phytoplasma asteris onion yellows OY-M, Pontibacter actiniarum sp. BAB1700, Dactylococcopsis salina PCC 8305, and Psychrobacter arcticus 273-4. Using the phylogenic profiler, 141 E. coli genes that are conserved in these seven organisms and B. subtilis were identified. According to the UniProt gene annotation, seven uncharacterized genes were picked as candidates.
Phylogenetic analysis. The phylogenetic tree of peptidase U32 ( Supplementary  Fig. 1) was generated as described 30 . Species names matched to proteins were retrieved from UniProt. The occurrence profiles of trhO, trhP, trhP1, trhP2, cmoA, cmoB, trmR, and cmoM homologs ( Supplementary Fig. 6) were retrieved from GTOP 62 or the Interpro database 63 . To generate the phylogenetic tree, 584 organisms that are registered in the databases we used, i.e., species listed in GTOP, phyloT, and Pfam, were selected.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
A reporting summary for this Article is available as a Supplementary Information file.