Conservation and Diversification of tRNA t6A-Modifying Enzymes across the Three Domains of Life

The universal N6-threonylcarbamoyladenosine (t6A) modification occurs at position 37 of tRNAs that decipher codons starting with adenosine. Mechanistically, t6A stabilizes structural configurations of the anticodon stem loop, promotes anticodon–codon pairing and safeguards the translational fidelity. The biosynthesis of tRNA t6A is co-catalyzed by two universally conserved protein families of TsaC/Sua5 (COG0009) and TsaD/Kae1/Qri7 (COG0533). Enzymatically, TsaC/Sua5 protein utilizes the substrates of L-threonine, HCO3−/CO2 and ATP to synthesize an intermediate L-threonylcarbamoyladenylate, of which the threonylcarbamoyl-moiety is subsequently transferred onto the A37 of substrate tRNAs by the TsaD–TsaB –TsaE complex in bacteria or by the KEOPS complex in archaea and eukaryotic cytoplasm, whereas Qri7/OSGEPL1 protein functions on its own in mitochondria. Depletion of tRNA t6A interferes with protein homeostasis and gravely affects the life of unicellular organisms and the fitness of higher eukaryotes. Pathogenic mutations of YRDC, OSGEPL1 and KEOPS are implicated in a number of human mitochondrial and neurological diseases, including autosomal recessive Galloway–Mowat syndrome. The molecular mechanisms underscoring both the biosynthesis and cellular roles of tRNA t6A are presently not well elucidated. This review summarizes current mechanistic understandings of the catalysis, regulation and disease implications of tRNA t6A-biosynthetic machineries of three kingdoms of life, with a special focus on delineating the structure–function relationship from perspectives of conservation and diversity.

Biosynthesis of tRNA t 6 A requires two universally conserved protein families of TsaC/Sua5 (COG0009) [59] and TsaD/Kae1/Qri7 (COG0533) [60], which belong to the minimal gene-sets and the last universal common ancestors (LUCA) [61][62][63][64]. Comparative genomics and proteomic analyses have identified a complete ensemble of enzymes capable of catalyzing tRNA t 6 A biosynthesis in the three domains of life (Table 1 and Figure 1C) [65]. As conventionally represented, tRNA t 6 A biosynthesis in bacteria requires TsaC (YrdC), TsaD (YgjD), TsaB (YeaZ) and TsaE (YjeE) [66,67]; in archaea and eukaryotic cytosol, it requires a Sua5/YRDC and KEOPS complex that contains Kae1, Bud32, Cgi121, Pcc1 and Gon7 [68][69][70][71][72]; in eukaryotic mitochondria, it requires Sua5/YRDC and Qri7/OSGEPL1, which are nuclear-encoded and imported into mitochondria [29,60,61,73,74]. The enzymatic synthesis of tRNA t 6 A takes place in two steps ( Figure 1C): in the first step, TsaC/Sua5 protein independently catalyzes the condensation of L-threonine, HCO 3 − /CO 2 and ATP to form an intermediate Threonylcarbamoyl-adenylate (TC-AMP) and an inorganic pyrophosphate (PPi) [32,67,75,76]; in the second step, the TC-moiety is catalytically transferred onto the N6 nitrogen of tRNA A37 by TsaD/Kae1/Qri7 protein with support of a varying number and types of accessory proteins in bacteria, archaea and eukaryotes [65,66,[68][69][70][71][72][73][74]. The molecular mechanisms of tRNA t 6 A-modifying enzymes vary among different biological systems. This review work is focused on analyzing the structure-function relationship and portrays an overview of conservation and diversification pertaining to molecular workings of these tRNA t 6 A biosynthetic machineries in the three domains of life. Deletion of sua5 and KEOPS (kae1, bud32, cgi121 and pcc1) abolishes the tRNA t 6 A formation and leads to pleiotropic lethal phenotypes. [41,65,68,70] S. cerevisiae Sua5 and Qri7 (MT) Qri7 is essential for yeast life. Deletion of qri7 leads to mitochondrial genome instability and abnormal morphology. [18,30,73,78] H. sapiens YRDC and KEOPS (CT) Deletion or mutations of YRDC and KEOPS (OSGEP, TP53RK, TPRKB, LAGE3 and GON7) leads to depletion of tRNA t 6 A and is implicated in Galloway-Mowat syndrome. [42,56,79,80] H. sapiens YRDC and OSGEPL1 (MT) Deletion of YRDC and OSGEPL1 causes mitochondrial dysfunction (interferes with protein synthesis and respiratory activities) and is clinically associated with an onset of myoclonus epilepsy with ragged-red fibers (MERRF). [29,74]  Two homologs of TsaC/Sua5 family are found in almost all sequenced organisms ( Figure 1C) [59,81]. While TsaC and YRDC are single domain proteins comprising 200 amino acids, Sua5 consists of an N-terminal YrdC/TsaC domain and an extra Sua5 domain (~100 amino acids), which are connected by a 40 amino acid-long segment ( Figure 1D). The phylogenetic distribution shows that, with few exceptions, the single-domain TsaC and YRDC are present in bacteria and higher eukaryotes, while Sua5 is found in archaea and yeast ( Figure 1C) [81]. The lethal phenotype of E. coli tsaC deletion and loss of tRNA t 6 A could be rescued and restored by either B. subtilis ywlC (tsaC) or S. cerevisiae SUA5, and vice versa [59,81,82], substantiating an independent catalytic role in TC-AMP synthesis. E. coli TsaC folds into an α/β twisted open sheet structure with antiparallel adjacent β-strands in the center and α-helices flanking on the exterior side, forming a baseball glovelike structure. The structure of TsaC aligns well with the YrdC domain of Sua5, of which the Sua5 domain adopts a Rossmann fold with an inner β-sheet composed of five β-strands and framed by three α-helices [31][32][33]76]. The binding and coordination of TC-AMP substrates (ATP, L-threonine and HCO 3 − ), TC-AMP and the byproduct PPi have been determined in the catalytic centers of TsaC/Sua5 proteins ( Figure 1D and Table 2) [32,33,75,76]. The structure reveals that the interdomain segment wraps over the catalytic center and functions as an essential gating loop to regulate the catalytic activity of Sua5 ( Figure 1D) [31,32]. The structure-function analysis provides a mechanistic basis underscoring the chemical transformations of L-threonine, HCO 3 − and ATP en route to TC-AMP: 1) L-threonine is first bound to react non-enzymatically with HCO 3 − in the closed catalytic pocket, generating an intermediate N-carboxy-L-threonine; 2) the subsequent opening of the gating loop permits the entry of ATP for adenylation with N-carboxy-L-threonine, of which the carboxyl group performs a nucleophilic attack, giving birth to TC-AMP and PPi [31,32,76]. Though many essential cellular roles have been associated with TsaD/Kae1/Qri7 family proteins [60,78,81,86], the central and convergent one is the catalytic function in tRNA t 6 A biosynthesis ( Figure 1C). Crystal structures of TsaD/Kae1/OSGEP/Qri7 proteins exhibit a typical "acetate and sugar kinase/heat shock protein 70/actin" (ASKHA) superfamily characteristic of comprising two similar ancestral oligonucleotide-binding (OB) domains on either side of a large cleft with a canonical ATP-binding site at the bottom ( Figure 1E), except for local variations incurred by insertion or deletion of connecting loops [35,37,38,[40][41][42][43]73,[87][88][89]. The overall structural architecture and configuration of the catalytic sites are conserved among TsaD/Kae1/OSGEP/Qri7 proteins. The binding and coordination of nucleotides (ATP, ADP and AMP) have been determined in crystal structures ( Figure 1E) and verified in biochemical assays (Table 2). In addition, different divalent cations (Fe 2+ , Zn 2+ and Mn 2+ ) are bound in coordination with ATP or ADP in the catalytic sites [35,37,38,40,41,73,87,90]. The metal ions are octahedrally coordinated by a highly conserved HxxH···D motif and two water molecules and play a canonical role in coordinating the γand β-phosphate of ATP and stabilizing the structural conformation of loops on the edge of the α-helices ( Figure 1E) [35,37,40,41,91]. Biochemical assays exhibited differentiated effects of metal ions on stimulating the t 6 A-catalytic activities of TsaD/Kae1/Qri7 proteins, demonstrating that TsaD/Kae1/Qri7 proteins are versatile in making the best use of these metal cations [65,74].
However, TsaD/Kae1/OSGEP/Qri7 does not necessarily consume ATP throughout the chemical transformation in tRNA t 6 A catalysis. The ATP-binding site must be used for binding TC-AMP or A37 of tRNA ( Figure 1C). As expected, crystal structures revealed that KG4 (analog of carboxy-AMP) and BK951 (analog of TC-AMP) are bound in the same ATP-binding sites of TsaD [36,38]. In addition, the bound TC-AMP analogs adopt a similar anti-conformation as that seen for ATP and ADP bound in catalytic sites of TsaD/Kae1/Qri7 proteins ( Figure 1E). Thus, the ATP-binding site is, in fact, used for TC-AMP binding and is involved in catalyzing the transfer of TC-moiety. The structure revealed that TC-moiety is bound to the active site Zn 2+ atom via an oxygen atom from both the phosphonate and the carboxylate moieties, as the binding of A37 of tRNA in the catalytic sites of TsaD/Kae1/Qri7 proteins is presently unknown. Functional analysis combined with the computational modeling of the A37 of tRNA suggested that TC-AMP and A37 could concomitantly accommodate in a manner that TC-moiety is bound to the metal ion and A37 is perpendicularly positioned to the TC-AMP plane. The catalysis exploits a putative metal oxyanion hole, in which the incoming N6 of tRNA A37 performs a nucleophilic attack at the carbamoyl bond of TC-AMP [36,38].

Evolutionary Implications of a Functional Cooperation between YrdC-like Domain and Kae1-like Domain
Interestingly, a YrdC-like domain and a Kae1-like domain are consecutively fused and functionally associated in S. tenebrarius TobZ [34] and E. coli HypF [44,92]. Crystal structures exhibit that TobZ consists of a Kae1 domain at the N-terminus and a YrdC domain at the C-terminus ( Figure 1F) [34]. By contrast, HypF comprises an N-terminal acylphosphatase (ACP) domain, a Zn finger-like (ZF) domain, a YrdC domain and a C-terminal Kae1 domain ( Figure 1G) [44,92]. The overall structures of both YrdC and Kae1 domains of TobZ and HypF are well matched to TsaC ( Figure 1D) and Kae1 ( Figure 1E), respectively [34,92]. Furthermore, configurations of the catalytic centers are conserved, including catalytic motifs in TsaC (KxR···SxN) and Kae1(HxxH···D) [31,34,40,92]. Although the organizing order of Kae1 domain and YrdC domain in TobZ and HypF is reversed, they cooperate to catalyze a two-step carbamoylation reaction: first, the YrdC domain catalyzes the formation of an intermediate carbamoyladenylate (CA) using substrates of carbamoyl phosphate (CP) and ATP; second, the Kae1 domains in TobZ and HypF catalyze the transfer of carbamoylmoiety from CA onto tobramycin via an O-carbamoylation ( Figure 1F) [34] and Cystine 351 of HypE via an S-carbamoylation ( Figure 1G) [44,92], respectively. These two evolutionarily ancient carbamoylation reactions are analogous to the threonylcarbamoylation reaction in tRNA t 6 A biosynthesis ( Figure 1C). The two types of reaction involve a classic adenylation transformation (R-COOH+ATP → R-C=O-AMP+PPi). In TobZ and HypF, the YrdC domain and Kae1 domain interact to create an electrostatic potential gradient channel for efficient transfer of CA ( Figure 1F,G) [34,44,92]. In contrast, TsaC/Sua5 proteins are capable of catalyzing the formation of TC-AMP independent of the t 6 A-catalysis by TsaD/Kae1/Qri7, and vice versa [67,73]. However, TC-AMP is a chemically unstable intermediate, and it decomposes spontaneously into AMP and L-threonine isocyanate (L-Thr-NCO) in minutes under physiological conditions [29,32,67,73]. Therefore, it is tempting to hypothesize that TsaC/Sua5 and TsaD/Kae1/Qri7 proteins could interact to facilitate an efficient delivery of TC-AMP to the catalytic sites of TsaD/Kae1/Qri7 proteins.

The Diversification of tRNA t 6 A Biosynthetic Systems
Genetic and biochemical validations demonstrated that Qri7/OSGEPL1 forms a minimal tRNA t 6 A biosynthetic system in combination with Sua5/YRDC [73,74]. Whereas bacterial TsaD requires TsaB and TsaE to perform t 6 A-catalysis [38,66,67], archaea and eukaryote Kae1 catalyze the t 6 A biosynthesis in the form of a KEOPS complex [70][71][72]79]. It appears that TsaD/Kae1/OSGEP proteins, in comparison to Qri7/OSGEPL1 proteins, have evolved to adopt diversified and easy-to-tune mechanisms underlying the t 6 A-catalytic activation in coping with the increasing biological complexity.
TsaB is concomitantly present with TsaD in all bacterial genomes but has no close orthologs in eukaryotes [81,93]. Structures reveal that TsaB also belongs to the ASKHA superfamily and adopts a canonical two-lobed HSP70/actin-like fold, each of which comprises a five-stranded mixed β-sheet surrounded by three α-helices [95,96]. While the Nlobe could be superimposed onto that of TsaD (Figure 2A), the truncated C-lobe is less conserved than that of TsaD and does not form a similar OB fold in support of binding nucleotides and metal ions [35,37,87]. TsaD interacts with TsaB via a conserved "helical bundle" interacting interface that is created by two pairs of N-terminal α-helices of TsaD and TsaB (Figure 2A) [35,37,87]. In addition, TsaD-TsaB dimer formation involves the C-terminal tail of TsaB, which extends over the interacting interface onto TsaD ( Figure 2A). Furthermore, the hyperthermophilic T. maritime TsaD-TsaB dimer further oligomerizes into a tetramer depicted as TsaD-TsaB-TsaB-TsaD, which is mediated by an extension of the β-stranded sheet of the C-terminal domains of TsaB ( Figure 2A) [85,87,97,98]. Structural comparison revealed that a conserved motif GPGXXTGXR located on the "helical bundle" region of TsaB is re-configured to create an atypical nucleotide-binding site, which is occupied by an ADP or AMP in the structures of an E. coli TsaD-TsaB complex [35,36]. TsaE folds into a structure comprising a seven-intertwining stranded mixed β-sheet with three connecting α-helices on either side [85,87,102]. TsaE is structurally related to the P-loop NTPase characterized by mono-nucleotide-binding fold that canonically catalyzes hydrolysis of the β-γ phosphate ester bond of the nucleotide [103,104]. The ATP-binding site is located at the C-edge of the parallel portion of the β-sheet and involves two canonical Walker motifs (A/B) and two Switch (I/II) loops between the central β-strand and the following α-helix [85,87,102]. The binding of ATP or ADP to TsaE has been observed in crystal structures of TsaE [87,102] and characterized in in vitro analyses (Table 2) [35,85]. In-solution small-angle X-ray scattering (SAXS), sedimentation velocity and gel filtration analyses confirmed that TsaE exists in an equilibrium of monomers and homodimers [35,84,103]. SAXS analysis showed that only the monomeric TsaE binds to the TsaD-TsaB heterodimer but not TsaD-TsaD or TsaB-TsaB homodimers in the presence of ATP [35,85]. The crystal structure of the T. maritime TsaD-TsaB-TsaE complex revealed that TsaE is bound at the TsaD-TsaB interface ( Figure 2A) and ATP is bound at the interface of TsaE and TsaD with its γ-phosphate group being coordinated by side chains from both subunits ( Figure 2C) [85,87]. Functional analyses demonstrated that a dynamic assembly of a TsaD-TsaB-TsaE complex is essential for tRNA t 6 A catalysis [35,85], during which TsaE plays a role in promoting the catalytic turnover via "re-setting" the catalytic conformation of TsaD [38].
Composite KEOPS models were built based on protein-protein interactions using crystal structures of KEOPS subcomplexes ( Figure 2B) [39,41,42,72,100,101]. In the architecture of KEOPS, Kae1/OSGEP binds to Bud32/TP53RK, which folds into a bi-lobal configuration characteristic of eukaryotic protein kinase, including an ATPase-dependent kinase Rio2 [101,112,113]. C-lobe of Bud32/TP53RK protrudes into the catalytic center of Kae1/OSGEP [42,101]. On one end of Kae1/OSGEP-Bud32/TP53RK, the N-terminal domain of Kae1/OSGEP forms a "helical bundle" interacting interface with Pcc1/LAGE3 ( Figure 2B), which further binds to Gon7/GON7 via forming an extended antiparallel β-sheet and contacts between α-helices; on the opposite end, the N-lobe of Bud32/TP53RK makes extensive contacts with Cgi121/TPRKB, which folds into a globular structure consisting of a four central antiparallel β-strands abutted by two α-helices on one side and five or six α-helices on the other [41,100,101]. Yeast or human KEOPS forms a pentameric complex that is linearly organized as Gon7-Pcc1-Kae1-Bud32-Cgi121 (Figures 2C and 3B) whereas archaean KEOPS forms a V-shaped dimer of a tetrameric Pcc1-Kae1-Bud32-Cgi121 complex via the dimerization of Pcc1 ( Figure 2C) [39,41,71,101]. The structure-function relationship and molecular workings of KEOPS have been described in detail in a recent review paper [116]. To put it simply, Kae1/OSGEP is the t 6 A catalytic subunit but remains catalytically inactive. It performs t 6 A catalysis in the form of a KEOPS complex. Bud32/TP53RK lacks structural motifs for being an atypical kinase but retains conserved elements required for ATP binding and hydrolysis. Bud32/TP53RK contributes to promoting the t 6 A-catalytic activity of KEOPS at the cost of consuming ATP and autophosphorylation that are positively regulated by Cgi121/TPRKB [88,89]. By doing so, Cgi121/TPRKB stimulates the t 6 A-catalytic activity of Kae1 in a positively cooperative manner [71,100]. Pcc1/LAGE3 folds into a KH-like domain that is frequently involved in RNA binding [41]. The fifth subunits Gon7 and GON7, which are found in yeast and humans, are intrinsically disordered proteins but acquire partial structure in the presence of Pcc1 and LAGE3, respectively [42,101]. The structured region of Gon7/GON7 folds into an antiparallel stranded β-sheet and two continuous α-helices. The connecting region between the second β-strand and first α-helix is disordered in the structures ( Figure 2B) [42,101]. Gon7/GON7 interacts with Pcc1/LAGE3 via the same interacting interface of Pcc1 homodimer and inhibits the dimerization of Pcc1/LAGE3 in yeast and humans [42,101].

Molecular Interactions between tRNAs and tRNA t 6 A-Modifying Enzymes
Both in vivo and in vitro analyses on tRNA t 6 A formation demonstrated that not only the frequencies of t 6 A modification vary among tRNAs of the same origin but the t 6 A biosynthetic enzymes of different systems exhibit different catalytic activities on the same tRNAs (Table 3) [27,29,65,74]. It is assumed that the frequency of t 6 A modification of tRNAs is determined largely by the molecular interactions between substrate and enzymes. However, very little is known about how tRNA t 6 A-modifying enzymes discriminate ANN-decoding tRNAs from non-substrate tRNA and how the A37 of tRNA is docked into the t 6 A-catalytic site.

Sequence Motifs and Structural Determinants of t 6 A tRNA Substrate
Mature L-shaped ANN-decoding tRNAs of different origins could be t 6 A-modified by enzymes of different organisms (Table 3) [27,72,74]. These sets of cross-validation assays revealed a number of tRNA substrate properties pertaining to sequence and structural motifs: (1) U36A37A38 motif is a universal determinant for all t 6 A-modifying enzymes [27,30,117]; (2) C32 is an essential determinant for KEOPSs [27]; (3) the C10U11 motif of the D-stem is a determinant for archaean KEOPS [72]; (4) C33 is an anti-determinant for bacterial TsaD-TsaB-TsaE and yeast KEOPS [27]; (5) the 3 CCA addition contributes to efficient t 6 A modification by KEOPS [72].

Structural Models of t 6 A-Modifying Enzymes in Complex with tRNA
Based on the only crystal structure of M. jannaschii Cgi121-tRNA Lys (UUU) ( Figure 2B) [72], composite structure models of four-subunit KEOPS-tRNA, eight-subunit KEOPS-tRNA, five-subunit KEOPS-tRNA, TsaD-TsaB-tRNA and Qri7-Qri7-tRNA are generated by means of structural alignment (Figures 2C and 3B). Except for the Archaean KEOPS-tRNA model that has been validated by electron microscopy and mutagenesis studies [72], other models have not been experimentally validated and may incur potential inaccuracy. These models are used in this work to provide an architectural overview of tRNA in complex with t 6 A-catalytic machineries of different biological systems.
All these models manifest that the binding of the tRNA A37 to the catalytic centers of TsaD/Kae1/Qri7 proteins necessitates contributions of the conserved "helical bundle" interacting interfaces of TsaD-TsaB, Kae1/OSGEP-Pcc1/LAGE3 and Qri7-Qri7 (Figure 2). Mutations disrupting the interface abolished the t 6 A-catalytic activities of TsaD-TsaB-TsaE [35], KEOPS [39] and Qri7 [73]. Judged from the structures and models (Figure 2), the interface is not only essentially involved in the formation of the complexes built around TsaD/Kae1/Qri7 but may play a contributing role in docking ASL of tRNAs into catalytic centers of TsaD/Kae1/Qri7 proteins. Crystal structures E. coli TsaD-TsaB [35,36] and T. maritima TsaD-TsaB-TsaE [38] revealed that the formation of the "helical bundle" interfaces creates an atypical ADP/AMP-binding site, which might bind adenines of substrate tRNA. Mutational studies of S. cerevisiae Qri7 showed that only one molecule of tRNA is sterically allowed to be lodged in Qri7-Qri7 dimer ( Figure 2C) [73]. In contrast, OSGEPL1 (the human counterpart of Qri7) functions as a monomer and does not form such a signature "helical bundle" interacting interface [74]. However, it was demonstrated that OSGEPL1 has evolved to make use of a post-translational modification mechanism (acylation of lysine residues) in regulating the t 6 A-catalytic activation [74].

Binding of tRNAs to TsaC/Sua5
Interestingly, high affinity interactions between tRNA and E. coli TsaC have been determined (Table 4) [31,59], though a direct binding of tRNAs to TsaC/Sua5 protein is dispensable in the in vitro reconstitution of tRNA t 6 A ( Figure 1C) [67,73]. It is still not known whether such an intriguing binding phenomenon is involved in the E. coli tRNA t 6 A modification in a cellular context. Mechanistically, E. coli TsaC folds into an α/β twisted open-sheet structure with a large positively-charged surface surrounding the catalytic center, which favors the binding of nucleic acids, preferably double-stranded RNA ( Figure 1D) [31]. Therefore, the binding of tRNA to E. coli TsaC is deemed promiscuous and is not related to tRNA t 6 A. Alternatively, tRNA mediates a transient interaction between TsaC and TsaD-TsaB in analogy to the interaction between YrdC domain and Kae1 domain in TobZ and HypF ( Figure 1F,G) [34,92]. By doing so, the two catalytic centers are brought in proximity for efficient delivery of TC-AMP from TsaC to TsaD. More puzzlingly, the activity of Sua5 in hydrolyzing ATP into AMP and PPi is potentiated by the addition of tRNA [65], though no interaction between Sua5 and tRNA has been determined. Kinetic analysis on tRNA t 6 A formation also suggested that a direct binding of TsaC to TsaD in the presence of tRNA might have played a role in protecting the bound TC-AMP from decomposing into AMP and L-Thr-NCO via shielding it from water [36]. Table 4. Summary of K D values of the binding between tRNA t 6 A-modifying enzymes and tR-NAs. Abbreviations: Ec, E. coli; Tm, T. maritima; Pa, P. abyssi; Mj, M. jannaschii; Sc, S. cerevisiae; Hs, H. sapiens; mt, mitochondrial; FQ, fluorescence quenching; EMSA, electrophoretic mobility shift assay; FP, fluorescence polarization; BLI, biolayer interferometry; TsaDB, TsaD-TsaB; TsaB 2 D 2 , TsaD-TsaB-TsaB-TsaD; TsaB 2 D 2 E, TsaD-TsaB-TsaB-TsaD-TsaE; KEOPS, Cgi121-Bud32-Kae1-Pcc1; KBC, Kae1-Bud32-Cgi121; BC, Bud32-Cgi121; * total tRNAs purified from E. coli.

Interaction between tRNA and TsaD-TsaB-TsaE Complex
Structural characterization of the molecular interactions between tRNA and TsaD-TsaB-TsaE is scarce, apart from a set of quantitative binding analysis (Table 4). In-solution SAXS analysis confirmed that only one molecule of tRNA is bound to the dimer of the TsaD-TsaB heterodimer, forming an unsymmetrical complex depicted as tRNA-TsaD-TsaB-TsaB-TsaD [38]. The SAXS model of tRNA-TsaD-TsaB shows that tRNA occupies the binding region of TsaE in TsaD-TsaB and is therefore bound to TsaD-TsaB in a mutually exclusive manner to TsaE. Biochemical analysis confirmed that T. maritima TsaD-TsaB binds to E. coli tRNA Thr with a K D value of 1.3 µM and the presence of T. maritima TsaE prevents the TsaD-TsaB complex from binding tRNA [85].
TsaE is an intrinsically weak ATPase whose activity is stimulated by a TsaD-TsaB dimer [35,84,85,118]. TsaE-catalyzed ATP hydrolysis is required for additional enzymatic turnover and occurs after the release of t 6 A-modified tRNA. Once the bound ATP is hydrolyzed, the ADP-bound TsaE is dissociated from TsaD-TsaB [35], creating an access and time window for binding of TC-AMP and tRNA in the active site of TsaD for the next catalytic cycle [38,85]. However, the ATPase-null TsaE mutant (mutation in the Walker B motif) is still capable of binding ATP and stimulating t 6 A catalytic activity of TsaD-TsaB in a comparable level as wild type TsaE [35,38]. The binding affinity between TsaD-TsaB and ATP-bound TsaE (K D = 0.34 µM) is tri-fold higher than that of TsaD-TsaB and tRNA (K D = 1.3 µM) [38,79,85,91]. It is therefore hypothesized that TsaE-catalyzed ATP hydrolysis plays a role in dissociating TsaE from TsaD-TsaB and in re-setting the active conformation of TsaD-TsaB [38]. Structure and function analysis demonstrated that TsaE works as a "molecular switch" triggered by ATP hydrolysis. However, the t 6 A-catalytic activity of TsaD-TsaB could be prohibitively stalled by ADP-bound TsaE after the ATP hydrolysis, as TsaE exhibited higher affinity to ADP over ATP (Table 3) [35,102]. In in vitro tRNA t 6 A formation assay, this mechanistic model might work as ATP is supplied in surplus. However, how the inactive ADP-bound TsaE is changed into ATP-bound form to promote the t 6 A-catalysis of TsaD-TsaB in a cellular context remains to be uncovered.

Interaction between tRNA and the KEOPS Complex
The structural models of KEOPS in complex with tRNA demonstrate that all the subunits contribute to the binding of tRNA onto KEOPS ( Figure 2C). The models show that ASL of tRNA is lodged into the catalytic center of Kae1/OSGEP with support from Pcc1/LAGE3 and Bud32/TP53RK, which mainly interact with ASL and D-stem of tRNA, respectively. Cgi121/TPRKB could bind to tRNA via accommodating its 3 CCA end. EMSA assays quantitatively confirmed that P. abyssi Pcc1-Kae1 complex forms a binding core for tRNA. The K D value between tRNA and Pcc1-Kae1 was estimated to be 0.2-0.4 µM, which is comparable to that of the whole KEOPS complex [71]. Similar filter binding experiments with M. jannaschii KEOPS proteins showed a strong binding between Bud32-Cgi121 and tRNA, whose binding affinity is comparable to that of either the ternary subcomplex of Kae1-Bud32-Cgi121 or the KEOPS complex [72]. The crystal structure of M. jannaschii Cgi121-tRNA reveals that both the 3 CCA end of tRNA and Cgi121 play critical roles in forming a catalytic tRNA-KEOPS complex [72]. The binding surface and 3 CCA-coordinating residues ( Figure 2B) are conserved in S. cerevisiae Cgi121 and H. sapiens TPRKB [41,72,100,101]. However, Cgi121 is dispensable in tRNA t 6 A biosynthesis by P. abyssi KEOPS [71] or S. cerevisiae KEOPS [70]. In particular, Drosophila KEOPS does not even contain a Cgi121 subunit [50]. In addition, S. cerevisiae KEOPS catalyzes the formation of t 6 A using tRNAs devoid of 3 CCA end [27]. These data suggest that an initial recruitment of tRNA by Cgi121 is not strictly required for KEOPS.
Interestingly, an ATP hydrolysis by Bud32 is indispensably involved in t 6 A biosynthesis by yeast and archaean KEOPS [70,71,80,100]. Mutation of the strictly conserved aspartate acid residue (D127 of P. abyssi Bud32, D112 of M. jannaschii Bud32 and D161 of S. cerevisiae Bud32) abolishes ATPase activity of Bud32 and tRNA t 6 A-catalytic activity of KEOPS [70][71][72]. However, the transfer of the TC-moiety from TC-AMP onto A37 of tRNA does not require ATP hydrolysis in terms of chemistry. It turns out that, in an analogy to TsaE in the bacterial TsaD-TsaB-TsaE system, Bud32/TP53RK functions as a "molecular switch" in activating the t 6 A-catalytic activity of KEOPS. Cooperatively, the ATPase activity of Bud32 in KEOPS is strongly potentiated by binding of full-length tRNA but not tRNA devoid of 3 CCA end [71,72]. NTPases, including ATPases and GTPases, often undergo conformational transitions between the NTP and NDP bound states, of which the slight movement is driven by the energy-releasing hydrolysis of NTP into NDP and Pi [103,104]. In KEOPS machineries, ATP hydrolysis could exert a driving force on the conformational dynamics of the two lobes of Bud32/TP53RK, which may influence its interactions with the substrate tRNAs. Crystal structures and structural models of KEOPS reveal that the C-terminal α-helix of Bud32/TP53RK near the active site of Kae1/OSGEP provides an opportunity to couple the ATP hydrolysis by Bud32/TP53RK and tRNA binding by the KEOPS complex ( Figure 2B,C). Based on activity tests, it is therefore hypothesized that the ATP hydrolysis by Bud32/TP53RK drives the dissociation of bound tRNA from KEOPS via reducing its direct interaction between tRNA and itself, or/and promotes the release of t 6 A-tRNA from catalytic centers of Kae1/OSGEP via dislodging the C-terminal tail of Bud32/TP53RK (Figures 2C and 3B) [72,116].

Mitochondrial Diseases Caused by tRNA t 6 A
In human mitochondria, tRNA t 6 A functions to monitor hypocarbia and regulate a set of genes involved in the Warburg effect, which occurs in order to better metabolize glucose anaerobically in highly proliferating cancer cells [13,29]. Knockout or knockdown of OSGEPL1 and YRDC deprives five mitochondrial tRNAs (Ile, Lys, Asn, Ser and Thr) of t 6 A modification, leading to respiratory defects in mitochondria [29,125]. Specifically, the t 6 A modification frequency of mitochondrial tRNAs is down-regulated by low cellular concentration of CO 2 /HCO 3 − in conditions of hypoxia [29]. The pathogenic mutations, A37G in mitochondrial tRNA Lys and tRNA Asn , abolished t 6 A formation, caused codon-specific dysfunction in mitochondrial translation, resulting in mitochondrial cytopathy and mitochondrial encephalopathy, respectively [29,30,53]. Mitochondrial tRNA Thr isolated from a myoclonic epilepsy with ragged red fibers (MERRF) patient contains significantly low levels of t 6 A modification [29,55]. The mitochondrial tRNA Thr from individuals with MERRF harbors pathogenic mutations on the U36A37A38 motif, which was mutated into U36A37G38 and resulted in deficiency of t 6 A modification [29]. By contrast, mitochondrial tRNA Ile from individuals with Leigh syndrome contains so-called "gain-of-function" mutations-U36A37G38 to U36A37A38, which led to unwanted t 6 A-modification of tRNA Ile [29]. In addition, the mutation of CDKAL1 gene causes type 2 diabetes, for reasons that CDKAL1 dysfunction results in aberration of the ms 2 t 6 A modification of cytoplasmic tRNA Lys , which leads to misreading of lysine codons in proinsulin [54,57,58,119].

KEOPS Mutations and Neurological Disorders
KOEPS is universally expressed in the human body and plays an essential role in maintaining the fitness of humans [114,115,126,127]. In particular, KOEPS-encoding genes are highly expressed in pronephros and in neural tissue such as the developing brain, eye and cranial cartilage ( Figure 3A) [127]. In human podocytes, the knockdown of KEOPS subunit genes led to the loss of tRNA t 6 A, interfered with protein biosynthesis and led to activation of the unfolded protein response and endoplasmic reticulum stress [56]. In clinical studies, recessive pathogenic mutations in KEOPS-encoding genes have been found in sequence genomes of patients with GAMOS ( Figure 3A and Table 5) [42,56,80,111,121,124]. Individuals with GAMOS also contain pathogenic missense variants of YRDC [42] and WDR4-a tRNA 7-methylguanosine (m 7 G) synthetase [128,129]. These genotype-phenotype analyses suggested that the loss of tRNA t 6 A modification, and probably other essential tRNA modifications, is implicated in the pathogenesis of GAMOS via causing deregulation of the translation processes. Except for a number of nonsense mutations leading to translation termination, most GAMOS related missense mutations are projected onto the crystal structures of GON7-LAGE3-OSGEP [42] and TP53RK-TPRKB [100], and the structural model of the human KEOPS-tRNA complex. The exact positions and possible effects of mutations on the structure and function of KEOPS are summarized in Table 5 and Figure 3B. While a number of mutations in OSGEP and TP53RK may interfere with the t 6 A-catalytic activity of OSGEP and ATPase activity of TP53RK, others may affect the assembly of a catalytically active KEOPS and possibly an expanded protein interaction network involving KEOPS [56,79].

Conclusions and Future Perspectives
As one of the most chemically and biosynthetically complex modifications, tRNA t 6 A plays an essential role in promoting translational efficiency and in regulating protein homeostasis. Depletion of tRNA t 6 A leads to pleiotropic phenotypes in single cell organisms and a variety of pathological consequences in higher eukaryotes. The biosynthesis of tRNA t 6 A is catalyzed by two ancient protein families of TsaC/Sua5 and TsaD/Kae1/Qri7, which belong to LUCA. Except for the prototypal mitochondrial system made up of TsaC/Sua5 and Qri7/OSGEPL1 proteins, TsaD/Kae1/OSGEP proteins require accessory proteins to catalyze the biosynthesis of tRNA t 6 A. While a conservation in the chemical transformation and catalytic pathway has been sustained in both TsaC/Sua5 protein and TsaD/Kae1/Qri7 proteins, the mechanisms in catalytic activation of TsaD/Kae1/Qri7 proteins have been diversified over evolution. It is hypothesized that higher orders of assembly enable more precise regulations of the catalytic machineries, which are definitely demanded to cope with the increasing complexity of biological systems. In particular, the ATP hydrolysis-based regulation of the TsaD-TsaB-TsaE and KEOPS complex allows more efficient catalysis and well-tuned regulations in a spatiotemporal fashion. However, a number of fundamental questions pertaining to the t 6 A catalysis, including the recognition of A37 and the catalytic events, await elucidation.
The evolutionary relationship and cellular roles of the KEOPS complex are still rather enigmatic. The composition of KEOPS in archaea and eukaryotes differs mainly in Gon7/GON7, which are only found in yeast and humans. Yeast Gon7 and human GON7 are intrinsically disordered proteins. They are not evolutionarily related but function as structural analogs in the KEOPS assembly. In addition, Drosophila KEOPS only consists of Kae1, Bud32 and Pcc1. The absence of Cgi121 in Drosophila defies the molecular working model in which Cgi121 plays an essential role in recruiting tRNA onto KEOPS.
Furthermore, it appears that KEOPS or its subunits may perform other functions independent of that in catalyzing the tRNA t 6 A modification. Unlike yeast KEOPS, which is imported into the nucleus and promotes telomere uncapping and elongation independent of tRNA t 6 A modification, human KEOPS is localized in the cytoplasm and is not functionally involved in telomere replication. Future work will be needed in investigating the seemingly diverse cellular roles of KEOPS and in dissecting the molecular pathways in mammalian development.
Protein synthesis is the connecting knot that links the molecular roles tRNA t 6 A to cellular life and pathological consequences. Deregulation in protein synthesis in both the cytosol and mitochondria can cause a plethora of disastrous consequences, which are also dependent on the codon usage and the tolerance of tRNA t 6 A depletion. However, it is hard to dissect the molecular layers at which the dysfunction in tRNA t 6 A biosynthetic machineries is causing problems and various pleiotropic phenotypes in bacteria, yeast and higher eukaryotes. Pathogenic mutations of human YRDC and KEOPS are implicated in the pathogenesis of the autosomal recessive GAMOS via interfering with the protein synthesis in the absence of tRNA t 6 A. Nonetheless, a dysfunction in protein synthesis could be attributed to the loss of other tRNA modifications (i.e., m 7 G) in patients with GAMOS. In addition, depletion and hypomodification of tRNA t 6 A have occurred in individuals with mitochondrial diseases such as MERRF. However, it is still challenging to convince medical doctors of the validity or feasibility of KEOPS as a target for medical interference of neurological disorders, which may involve mutations of other genes. Therefore, causative roles of YRDC, OSGEPL1 and KEOPS in specific types of diseases or physiological disorders await in-depth analyses. A t 6 A-based quantitative sequencing technique is desirable, analyzing the frequency of tRAN t 6 A under physiological and pathological conditions.
Last but not least, one would have marveled at the achievement of elucidating the tRNA t 6 A biosynthesis over the past five decades. In retrospect, a great number of talented colleagues have contributed to experimental data aggregation and have offered insightful ideas toward resolving the enzymatic pathways and molecular workings of tRNA t 6 A modification in the three domains of life. In particular, several breakthroughs have played key roles in advancing this line of research: (i) establishment of a biosynthetic relationship between the universally distributed but function-unknown LUCA families (TsaC/Sua5 and TsaD/Kae1) and the universally found tRNA t 6 A modification without known enzymes; (ii) discovery and structural analyses of KEOPS/EKC; (iii) biochemical and biophysical analyses on TsaD-TsaB-TsaE interaction network; (iv) clinical genotype and phenotype analyses on mutations of YRDC and KEOPS in human diseases. In the future, we expect to see more breakthroughs in the mechanistic understanding of the roles of tRNA t 6 A and its biosynthetic enzymes.