To date, more than 150 modified nucleosides have been found in RNAs from the three domains of life [1]. Transfer RNA contains numerous modified nucleosides [2,3] and the majority of modified nucleosides in tRNA are introduced by site-specific tRNA modification enzymes. Transfer RNA modification enzymes frequently contain one or more distinct domains in addition to the catalytic domain, although small tRNA methyltransferases such as TrmL [4,5] and TrmH [6,7] are mainly composed of the catalytic domain [8,9,10]. The existence of the thiouridine synthetase, methyltransferases and pseudouridine synthase (THUMP) domain was originally predicted in a bioinformatic study [11]. in 2001, Aravind and Koonin reported that tRNA 4-thiouridine synthetase-like proteins, conserved RNA methyltransferases, archaeal pseudouridine synthases and several uncharacterized proteins share a predicted RNA binding domain, which adopts an α/β fold [11]. At that time, although the Escherichia coli thiI gene product had already been identified as a tRNA 4-thiouridine synthetase [12], functions of the other proteins were unknown. Furthermore, no structures for any of the proteins, including ThiI, had been reported. In 2004, the Pyrococcus abyssi PAB1283 protein was firstly identified as a tRNA methyltransferase, which contains a THUMP domain [13]. Because the PAB1283 protein possesses enzymatic activity for the formation of N²-methylguanosine (m²G) and N², N²-dimethylguanosine (m²₂G) at position 10 in tRNA, nowadays, the PAB1283 protein is called archaeal Trm11 (arcTrm11). At the same time, the Saccharomyces cerevisiae tan1 gene product was found to be an essential protein for the formation of N⁴-acetylcytidine at position 12 (ac⁴C12) in tRNA^Leu and tRNA^Ser [14]. Although Tan1 contains a THUMP domain, this protein itself does not possess tRNA acetyltransferase activity [14] and does not contain a catalytic domain [15]. Later, Tan1 was identified as a partner protein of S. cerevisiae tRNA acetyltransferse (Kre33) [16]. Since the prediction of the THUMP domain more than two decades ago, many tRNA modification enzymes containing a THUMP domain have been identified. Among them, in addition to tRNA 4-thiouridine synthetases, tRNA methyltransferases, tRNA pseudoridine synthases, tRNA deaminase [17] and a partner protein of tRNA acetyltransferases [16] have been identified.

When the existence of the THUMP domain was predicted [11], ThiI was the only identified tRNA modification enzyme in the list of predicted THUMP-related proteins. ThiI is a tRNA s⁴U synthetase [12]. s⁴U is found at positions 8 and 9 in tRNAs from eubacteria and archaea [1,2,3]. The biosynthesis pathways of s⁴U are different in eubacteria and archaea [39,40,41,42]. In E. coli, the sulfur atom in L-cysteine is activated by cysteine desulfrase (IscS) and is then transferred to tRNA by ThiI in the presence of ATP [43,44,45]. Cysteine residues at positions 344 and 456 in E. coli ThiI are essential for the reaction and these residues are considered to form a disulfide bond in the catalytic turnover [46,47]. In contrast, the iscS gene is not encoded in the majority of archaea genomes [48]. In the case of Methanococcus maripuludis, ThiI contains an Fe-S cluster and S²⁻ is used as a sulfur donor instead of L-cysteine [22,48]. However, the Fe-S cluster type thiI gene is not present in some archaea genomes and the biosynthesis pathways in these organisms are still unknown [39,48,49]. During the submission of this manuscript, it was reported that M. maripuldis and P. furiosus ThiI proteins possess a [4Fe-4S] cluster [50]. Furthermore, it has been proposed that these enzymes be renamed TtuI [50].

In 2006, the crystal structure of Bacillus anthracis ThiI (PDB code: 2C5S) was the first of the THUMP-related proteins to be reported [51]. B. anthracis ThiI contains three domains, an N-terminal ferredoxin-like domain (green), a THUMP domain (red) and a C-terminal PP-loop domain (blue). This structure revealed that the THUMP domain is composed of α-helices and β-strands as predicted. A tRNA binding model was also constructed [51]. In the model, the THUMP domain of ThiI was placed near the CCA-terminus of tRNA because it was reported that the CCA-terminus was essential for the sulfur-transfer reaction of ThiI [52]. Later, this idea was experimentally verified by biochemical and structural studies of truncated tRNA [53] and ThiI-truncated tRNA complex [54]. The N-terminal ferredoxin-like domain functions to maintain the distance and angle between the THUMP and PP-loop domains. The PP-loop was originally found as a P-loop-like sequence motif, which had been observed in ATP pyrophosphatases [55]. The PP-loop domain in ThiI binds ATP and activates tRNA by adenylation [56,57]. At the same time that the crystal structure of B. anthracis ThiI was solved, the structure of Pyrococcus horikoshii PH1313 protein (PDB code: 1VBK) was released as a protein of unknown function [58]. In the Pyrococcus genera, multiple genes for ThiI homologs are often encoded in their genomes [22]. Because ThiI is involved in thiamine biosynthesis in addition to s⁴U modification in tRNA [12,59,60,61], the ThiI homologs in Pyrococcus may not have a dual function but instead individual proteins have single roles. Although the structure of the PH1313 protein resembles other ThiI proteins, the PH1313 protein lacks several conserved amino acid residues of ThiI proteins. To date, the enzymatic activity of the PH1313 protein has not been confirmed. Furthermore, modified nucleosides in tRNAs from P. horikoshii have not been analyzed [62].

Transfer RNA modification enzymes often recognize local structure(s) in tRNA [63]. Therefore, tRNA modification enzymes are frequently able to modify a truncated tRNA. For example, E. coli TrmA [64,65], E. coli TruB [66], E. coli Tgt [67,68], T. thermophilus TrmFO [69], T. thermophilus TrmI [70] and A. aeolicus TrmD [71] can modify a micro-helix RNA, which mimics the T-arm or anticodon-arm of substrate tRNA. TrmA, TruB, Tgt, TrmFO, TrmI and TrmD are tRNA (m⁵U54) methyltransferase [72], tRNA (Ψ55) synthase [73], tRNA guanine-transglycosylase [67,74,75,76], N5, N10-methylenetetrahydrofolate-dependent-tRNA (m⁵U54) methyltransferase [77], tRNA (m¹A58) methyltransferase [78] and tRNA (m¹G37) methyltransferase [79], respectively. Furthermore, E. coli TrmJ [80], A. aeolicus TrmB [81] and T. thermophilus TrmH [82] can methylate a truncated tRNA. TrmJ, TrmB and TrmH are tRNA (Cm32/Um32) methyltransferase [83], tRNA (m⁷G46) methyltransferase [84] and tRNA (Gm18) methyltransferase [6,85], respectively.

Lauhon et al. have reported that a truncated tRNA^Phe is a minimum substrate for E. coli ThiI [52]. This truncated tRNA^Phe is also recognized by Thermotoga maritima ThiI as a substrate [54]. The crystal structure of the complex of the minimum substrate RNA and T. maritima ThiI has been reported [54]. T. maritima ThiI forms a dimer and two minimum substrate RNAs bind to this dimer. The THUMP domain in one subunit captures the CCA terminus of one minimum substrate RNA and the PP-loop domain in this subunit accesses the modification site (U8) in another minimum substate RNA. Thus, this complex structure demonstrates that ThiI acts as a dimer. The disulfide bond, which acts in the catalytic cycle, in E. coli ThiI is formed within a single subunit [86]. Furthermore, this structure proposes a concept that the THUMP domain recognizes the 3′-end of RNA (in the case of tRNA, the CCA terminus).

M. kandleri is a hyper-thermophilic archaeon in which position 8 in 30 tRNA genes is encoded as C [87,88]. This C8 is modified to U8 by deamination (C to U editing) [17]. For further information about deamination in tRNA [89]. The enzyme responsible for deamination of C8 is CDAT8. CDAT8 can modify C8 in a micro-helix RNA. A crystal structure of CDAT8 has been reported (PDB code, 3G8Q) [17]. The domain arrangement of CDAT8 is different from that of ThiI. From the N-terminus to the C-terminus, the order of the domains is deaminase, ferredoxin-like and THUMP. However, the structure of the ferredoxin-like and THUMP domains is very similar to that of ThiI. From the model of the complex between CDAT8 and tRNA, it was predicted that the THUMP domain of CDAT8 captures the CCA terminus of substrate tRNA [17].

Of the different modified nucleosides in tRNA, methylated nucleosides are the most abundant [1,2,90]. Consistent with this, numerous tRNA methyltransferases have been identified [90]. Transfer RNA methyltransferases can be divided into two types according to the methyl group donor. The majority of tRNA methyltransferases use S-adenosyl-L-methionine as a methyl group donor whereas mnmG (previous name, GidA) [91,92,93,94,95,96] and TrmFO [69,77,97,98] are an exception and use N5, N10-methylenetetrafolare. S-adenosyl-L-methionine-dependent tRNA methyltransferases are further classified on the basis of their catalytic domain [9,90,99]. The majority of S-adenosyl-L-methionine-dependent tRNA methyltransferases possess a Rossmann fold catalytic domain [9,99]. The second group of S-adenosyl-L-methionine-dependent tRNA methyltransferases belong to a SpoU-TrmD (SPOUT) superfamily, which possess a SPOUT catalytic domain [9,100]. In addition, TrmO is an exception and has a b-barrel type catalytic domain [101].

All THUMP-related tRNA methyltransferases reported possess a Rossmann fold catalytic domain and synthesize only m²G (and m²₂G). Several enzymes synthesize m²₂G from m²G by a second methylation and act on multiple positions. Although classification of tRNA (m²G/m²₂G) methyltransferases is complicated, the THUMP-related tRNA (m²G/m²₂G) methyltransferases can be divided into two types according to their methylation sites. Thus, Trm11/arcTrm11/arcTrm11-arcTrm112/TRMT11-TRMT112 act on position 10 in tRNA, whereas TrmN/Trm14/THUMPD3-TRMT112 act on position 6 and an additional site. It should be mentioned that tRNA (m²G/m²₂G) methyltransferases, which do not possess a THUMP domain, do exist. One major group of such tRNA (m²G/m²₂G) methyltransferases is the Trm1 family [102,103,104,105,106,107,108,109,110]. S. cerevisiae Trm1 catalyzes the methylation of G26 in tRNA and synthesizes m²G26 and m²₂G26 [102,103]. Mammalian and Aquifex aeolicus Trm1 enzymes form m²G27 and m²₂G27 in addition to m²G26 and m²₂G26 [105,107]. Crystal structures of P. horikoshii [109] and A. aeolicus [110] Trm1 proteins demonstrate that these proteins possess a distinct C-terminal domain instead of a THUMP domain.

Trm112, TRMT112 and arcTrm112 are hub-proteins, which regulate multiple methyltransferases [23,24,27,111,112,114,115,116]. In the case of human TRMT11-TRMT112, formation of the complex has been reported [111]. However, the modification, position and substrate tRNAs of human TRMT11-TRMT112 have not been experimentally confirmed. For T. kodakarensis Trm14, tRNA^Trp from a trm14 gene deletion strain loses the m²G67 modification [113]. However, subunit composition and enzymatic activity of T. kodakarensis Trm14 have not been confirmed with a purified enzyme. In addition, recently, RNA fragments from tRNA mixtures purified from M. Jannaschii [117], M. maripaldis, P. furiosus and Sulfolobus acidocaldarius [118] were analyzed by mass-spectrometry. m²G6 and m²G67 were observed in several tRNAs from M. Jannaschii [117], and thus Trm14 is probably involved in the formation of these modifications. Furthermore, in the case of P. furiosus, several tRNAs were shown to possess a m²₂G6 modification in addition to m²G6 and m²G67 modifications [118]. Therefore, archaeal Trm14 proteins may possess broader positional specificity than was previously thought.

As described in the Introduction, the P. abyssi PAB1283 protein (arcTrm11) was the first tRNA methyltransferase identified as containing a THUMP domain [13]. The THUMP domain of P. abyssi arcTrm11 has been expressed in E. coli cells, purified and analyzed [119]. This [119] reported that the THUMP domain autonomously folds and that the affinity of the THUMP domain for tRNA is very weak. In 2005, it was reported that S. cerevisiae Trm11 requires a partner subunit, Trm112 [23]. Furthermore, the S. cerevisiae Trm11-Trm112 complex only produces m²G10 in tRNA [23] whereas arcTrm11 produces m²G10 and m²₂G10 [13,24,34]. Moreover, in several archaea, arcTrm11 requires arcTrm112 for enzymatic activity as seen with S. cerevisiae Trm11 [24,112].

T. thermophilus TrmN is the only eubacterial THUMP-related tRNA methyltransferase reported [25]. TrmN methylates G6 in tRNA^Phe and produces m²G6 [25]. Methanococcus jannaschii Trm14 is an archaeal homolog of TrmN and produces m²G6 (and m²₂G6) in tRNA^Cys [26]. Furthermore, in in vitro experiments, the second methylation from m²G6 to m²₂G6 in the tRNA^Cys transcript was observed [26]. The human THUMPD3-TRMT112 complex methylates G6 and G7 in several tRNAs and produces m²G6 and m²G7 [27].

In 2012, crystal structures of P. abyssi Trm14 and T. thermophilus TrmN were reported [120]. Both enzymes methylate G6 in tRNA and produce m²G6. The crystal structures revealed that these enzymes possess a N-terminal ferredoxin-like domain, a THUMP domain, a Rossmann fold methyltransferase (methylase) domain and a linker region. In the same study, it was reported that several positively charged amino acid residues are involved in tRNA binding [120]. Furthermore, the structures of the ferredoxin-like domain and the THUMP domain of Trm14 and TrmN are remarkably similar to those of ThiI and CDAT8. In 2016, the crystal structure of T. kodakarensis arcTrm11 was solved [34]. The arrangement of the domains of arcTrm11 is the same as that of Trm14 and TrmN. However, the distance between the THUMP and methylase domains in arcTrm11 is longer than that in Trm14 and TrmN due to structural differences in the ferredoxin-like domain and the linker region. This difference is important for the selection of the modification site (G10 or G6). A site-directed mutagenesis study showed that the THUMP domain in arcTrm11 captures the CCA terminus of substrate tRNA [34]. The distance between the CCA terminus and G10 in tRNA is longer than the distance between the CCA terminus and G6. Thus, these crystal structures led to the idea that the methylation site (G6 or G10) is determined by the distance from the THUMP domain to the catalytic pocket.

Eukaryotic and some archaeal Trm11 proteins require a partner subunit (Trm112, TRMT112 or arcTrm112) for enzymatic activity [23,24,27,111,112,114,115,116]. It should be mentioned that eukaryotic Trm112 homologs activate multiple methyltransferases. For example, S. cerevisiae Trm112 activates Trm9 [121], Bud23 [122,123] and Mtq2 [124,125] in addition to Trm11. Furthermore, a human ortholog of Trm112, TRMT112 interacts with at least seven human methyltransferases (WBSCR22 (responsible for formation of 7-methylguanosine at position 1636 in 18S rRNA) [126], METTL5 (formation of N⁶-methyladenosine at position 1832 in 18S rRNA) [127], HEMK2 (methylation of a glutamine side chain of eRF1 protein) [128], ALKBH8 (responsible for 5-methoxycarbonylmethyluridine derivatives at position 34 in tRNA) [129,130,131,132], TRMT11 [111], THUMPD2 (function unknown) [111] and THUMPD3 (production of m²G6 and m²G7 in tRNA)) [27].

Several tRNA modification enzymes form protein complexes [90,91,96,116,133,134,135,136]. The partner subunit(s) is frequently involved in the substrate tRNA recognition. Consequently, the binding sites of these modification enzymes are often extended over the whole tRNA molecule. For example, as described in Section 4.1., bacterial tRNA (m⁷G46) methyltransferase (TrmB) can methylate a truncated tRNA, in which the interaction between the T-arm and D-arm is disrupted [81]. However, in contrast, eukaryotic tRNA (m⁷G46) methyltransferase (Trm8-Trm82) [136] requires the interaction between the T-arm and D-arm for methylation [137]. Thus, the existence of Trm82 seems to act on recognition of the L-shaped tRNA structure. In the case of S. cerevisiae Trm7, the partner subunits (Trm732 and Trm734) decide the modification positions: Trm7-Trm732 and Trm7-Trm734 catalyze 2′-O-methylations at position 32 and position 34, respectively, in tRNA [138]. The biochemical and structural studies of Trm7-Trm734 suggest that Trm734 captures the D-arm in substrate tRNA and controls the accession of the modification site (ribose at position 34) in tRNA to the catalytic pocket in Trm7 [139]. A conserved motif (RRSAGLP sequence) in Trm732 is involved in the methylation of position 32 in tRNA^Phe [140]. Thus, the presence of a partner subunit is frequently involved in substrate tRNA recognition.

S. cerevisiae Trm11-Trm112 does not methylate truncated tRNAs [141]. This observation suggests that the binding sites of Trm11-Trm112 in tRNA are spread over the whole tRNA molecule. Biochemical and biophysical studies of S. cerevisiae Trm11-Trm112 resulted in the proposal of a model in whichTrm112 is accessible to the anticodon-loop region in tRNA dependent on the movement of the THUMP domain [142]. The required elements in tRNA for methylation by Trm11-Trm112 have been clarified: the CCA terminus, G10-C25 base pair, regular size (5 nt) variable region and ribose-phosphate backbone around purine38 in tRNA are essential for methylation by S. cerevisiae Trm11-Trm112 [141]. Thus, the biochemical study [141] supports the model referenced [142] because the ribose-phosphate backbone around position 38 is recognized by S. cerevisiae Trm11-Trm112. Furthermore, the crystal structure of A. fulgidus arcTrm11-arcTrm112 has been reported [24]. When the THUMP domain in arcTrm11 captures the CCA terminus in substrate tRNA, arcTrm112 accesses the anticodon-loop. Therefore, tRNA recognition mechanisms of eukaryotic and archaeal Trm11-Trm112 seem to be basically common. Human THUMPD3-TRMT112 requires the CCA terminus for methylation and does not methylate a mini-helix RNA [27]. Therefore, TRMT112 in THUMPD3-TRMT112 may also be involved in the anticodon-loop recognition as per Trm11-Trm112.

As described in the Introduction, S. cerevisiae Tan1 (human THUMPD1) contains a THUMP domain and acts as a partner protein of tRNA acetyltransferse, Kre33 (human NAT10) [16]. The Methanothermobacter thermautotrophicus Tan1 homolog is composed of N-terminal ferredoxin-like and C-terminal THUMP domains [15]. Although the crystal structure of Kre33 (or NAT10) has not been reported, a structural model (PDB code, 2ZPA) has been proposed [16] in which Kre33 (NAT10) contains DUF1726 (of unknown function), helicase, N-acetyltransferase and tRNA binding domains. In the case of T. kodakarensis TkNAT10 (the archaeal homolog of NAT10), the C-terminal region is missing [18]. Kre33 catalyzes the acetylation of 18S rRNA as well as acetylation of tRNA [16]. A random mutagenesis study of T. kodakarensis revealed that the disruption of the Tk0754 gene causes complete loss of ac⁴C modification in a tRNA mixture [143]. Detailed enzymatic activity of the Tk0754 gene product (TkNAT10) has been reported [18].

Pseudouridine (Ψ) is abundant in RNAs from the three domains of life [1,2,3] and is synthesized by C5-ribosyl isomerization from uridine, which is catalyzed by pseudouridine synthases [144,145,146,147,148,149,150]. Pseudouridine synthases can be classified into six families; however, PUS10 is the only THUMP-related enzyme [28,29,144,145,146,147,148,149,150]. In 2006, Ψ55 formation in tRNA catalyzed by archaeal Pus10 was reported [28]. Thus, this report demonstrates that one of the predicted THUMP-containing proteins [11] has pseudouridine synthase activity. In 2008, it was reported that archaeal Pus10 can synthesize Ψ54 in tRNA in addition to Ψ55 [29]. Furthermore, Methanocaldoccus jannaschii PUS10 can modify U54 and U55 in a micro-helix RNA, which mimics the T-arm [151].

In 2007, a crystal structure of human PUS10 was reported and showed that the THUMP-related structure is contained in the N-terminal accessory domain [20]. When the CCA-terminus in tRNA is placed onto the THUMP-related structure, the modification sites (U54 and U55) have access to the catalytic pocket of the pseudouridine synthase domain [20]. However, human PUS10 can modify U54 in a tRNA transcript without a CCA terminus [30]. Because human PUS10 strongly recognizes the sequences of the aminoacyl-stem and T-arm [30], the recognition of the CCA terminus by the THUMP-related structure may be not important for pseudouridine formation. The accessory domain of human PUS10 is large compared to a typical THUMP domain. This large accessory domain was gained in the process of evolution of eukaryotic PUS10 [143]. Furthermore, tRNA recognition by human PUS10 in living cells is complicated. Human PUS10 is expressed in both the nucleus and cytoplasm [30]. Human nuclear PUS10 does not have the pseudouridine synthesis activity and inhibits the activity of TRUB1 [human tRNA (Ψ55) synthase] by binding to specific tRNAs in the nucleus [31]. In contrast, human cytoplasmic PUS10 can synthesize Ψ54 in tRNAs, which possess an AAAU sequence from position 57 to position 60 in the T-loop, in addition to Ψ55 [31]. Moreover, it has been reported that human PUS10 is involved in microRNA processing [152]. In this process, PUS10 directly binds to primary microRNA and the catalytic activity of PUS10 is not required [152]. Thus, PUS10 may act as an RNA binding subunit in microRNA processing.

Based on the crystal structure of human PUS10, a structural model of archaeal PUS10 was constructed and several amino acid residues, which are required for enzymatic activity and tRNA binding, were identified [21]. Another mutagenesis study revealed that the thumb-loop in the catalytic domain and N-terminal cysteine residues are important for the Ψ54 formation activity of M. jannaschii PUS10 [151].