Structural basis for tRNA-dependent cysteine biosynthesis

Cysteine can be synthesized by tRNA-dependent mechanism using a two-step indirect pathway, where O-phosphoseryl-tRNA synthetase (SepRS) catalyzes the ligation of a mismatching O-phosphoserine (Sep) to tRNACys followed by the conversion of tRNA-bounded Sep into cysteine by Sep-tRNA:Cys-tRNA synthase (SepCysS). In ancestral methanogens, a third protein SepCysE forms a bridge between the two enzymes to create a ternary complex named the transsulfursome. By combination of X-ray crystallography, SAXS and EM, together with biochemical evidences, here we show that the three domains of SepCysE each bind SepRS, SepCysS, and tRNACys, respectively, which mediates the dynamic architecture of the transsulfursome and thus enables a global long-range channeling of tRNACys between SepRS and SepCysS distant active sites. This channeling mechanism could facilitate the consecutive reactions of the two-step indirect pathway of Cys-tRNACys synthesis (tRNA-dependent cysteine biosynthesis) to prevent challenge of translational fidelity, and may reflect the mechanism that cysteine was originally added into genetic code.

T he fidelity of protein synthesis is based on the accurate aminoacylation of tRNAs in which specific amino acids are attached to cognate tRNAs. A group of enzymes named aminoacyl-tRNA synthetases (aaRSs) are responsible for this reaction. It was believed that there is a matching aaRS for each of the 20 amino acids, which specifically ligates an amino acid to the 3′ terminal adenosine of its cognate tRNA 1,2 . However, the advent of whole genome sequencing has revealed that several assigned aaRSs are absent in some organisms 3,4 . In these cases, a non-cognate amino acid is initially attached to the tRNA by the first enzyme, which is subsequently converted to the cognate amino acid by the second enzyme in a tRNA-dependent manner. This indirect pathway is adopted for the synthesis of Gln-tRNA Gln and Asn-tRNA Asn in many bacteria and archaea [5][6][7] , Cys-tRNA Cys in most of the methanogenic archaea 8 and Sec-tRNA Sec in many species across all three domains of life 9 . Interestingly, the presence of the indirect pathways is accompanied with the absence of the enzymes responsible for canonical amino acids biosynthesis, where the indirect pathways are the sources of the amino acids biosynthesis by tRNA-dependent mechanism 8 .
The synthesis of Cys-tRNA Cys (and also cysteine) by an indirect pathway involves two enzymes, O-phosphoseryl-tRNA synthetase (SepRS) and Sep-tRNA:Cys-tRNA synthase (SepCysS), which are respectively responsible for the attachment of a noncanonical amino acid O-phosphoserine (Sep) to tRNA Cys and the conversion of Sep-tRNA Cys to Cys-tRNA Cys 8 . Phylogenetic studies have suggested that the indirect pathway for Cys-tRNA Cys synthesis is of ancient origin 10 . This system also acts as the sole route for cysteine biosynthesis for these organisms 8 . More recently, it was reported that a third protein SepCysE (SepRS/ SepCysS pathway enhancer) is necessary for coupling the consecutive two reactions and for enhancing tRNA Cys binding 11 . Thus, the SepRS, SepCysE, and SepCysS ternary complex (named the transsulfursome) is responsible for the whole process. Sep-CysE is preserved in class I methanogens, which evolved before other methanogens and may have appeared as the ancestral Euryarchaeota 11,12 . Therefore, elucidation of the functional mechanism of this system is of vital importance as it may explain the mechanism by which cysteine was added to the genetic code.
SepRS is a homotetramer and structurally characterized as a Class II aaRS, closely related to PheRS 10,13,14 . SepCysS is a homodimer with pyridoxal-5′-phosphate (PLP) bound in the active sites that shares structural similarity to the PLP-dependent cysteine desulfurase 10,15 . The structural analysis of SepCysS in complex with the N-terminal domain of SepCysE (SepCysE (NTD)) showed that the dimer of SepCysE is bound to the dimer of SepCysS with their dimer axes aligned 11 . However, it is not yet known how SepRS is bound to this binary complex to form a ternary complex (transsulfursome) and how Sep-tRNA Cys is transferred between these two enzymes safely. Furthermore, while structural and biochemical analyses showed the recognition of tRNA Cys by SepRS 13,16,17 , the specific interaction between SepCysS and tRNA Cys is unclear, which is essential to guarantee the fidelity of the two-step indirect pathway in translation.
In the present study, the structures of the binary complex of SepRS and SepCysE, and the ternary complex of SepCysE, Sep-CysS, and tRNA Cys from Methanocaldococcus jannaschii have been solved. Combined with the SAXS, EM, and biochemical results, the recognition of tRNA Cys by SepCysS was determined and the architecture of the transsulfursome was established. On the basis of these results, we propose a mechanism for tRNA channeling, whereby a hinge motion of the SepCysE-SepCysS complex relative to SepRS and a swinging motion of SepCysE (CTD) with bound tRNA Cys facilitate the transport of the mischarged Sep-tRNA Cys to the second, distant active site.

Results
Structure of the SepCysE-SepCysS-tRNA Cys ternary complex. A crystal structure of the SepCysE-SepCysS-tRNA Cys ternary complex was determined at a resolution of 2.6 Å by a molecular replacement method (Table 1, Supplementary Fig. 1a). A Fo-Fc map calculated after several rounds of refinement cycle showed a bulk of density adjacent to the end of one SepCysE(NTD), which represents the linker chain and the C-terminal domain of SepCysE [SepCysE (CTD)], which was disordered in the binary complex previously determined 11 . Thus, this crystal revealed most of the structural features of SepCysE ( Supplementary Fig. 1b), as well as the binding manner with tRNA Cys and SepCysS (Fig. 1a). A dimer of the Nterminal domain of SepCysE [SepCysE(NTD), residues 35-101] was tightly bound to a dimer of SepCysS, with their twofold axes aligned. From one of the NTDs of the SepCysE dimer, a linker chain (residues 102-110) extended to the C-terminal domain of SepCysE [SepCysE(CTD), residues 111-213] to which tRNA Cys was bound with its CCA terminus in contact with SepCysS (Fig. 1a). The SepCysE(CTD) of the second subunit was not visible in the crystal, probably because of the absence of bound tRNA Cys . The stoichiometry of SepCysE:SepCysS:tRNA Cys = 2:2:1 is in agreement with the result of electrophoresis mobility shift assay (EMSA) 18 .
The structure of the SepCysE(CTD) consisted of threestranded parallel β sheets with four peripheral alpha helices ( Supplementary Fig. 1b). The connectivity of the secondary structures was β1, α4, β2, α5, α6, β3, and α7. It is basically an α/β fold with short insertion helix α5. Although the primary sequence of SepCysE(CTD) is not related to any known structures in the PDB, a Dali server search 19 based on the present structure identified that the fold of SepCysE(CTD) is related to the Toprim (topoisomerase-primase) domain, which is found in proteins where < I(hkl) > and N (hkl) are the mean intensity of a set of equivalent reflections and the multiplicity, respectively b Rwork = Σhkl ||Fobs|−|Fcalc|| / Σhkl |Fobs|, Rfree was calculated for 5% randomly selected test sets that were not used in the refinement related to DNA metabolism such as primases, topoisomerases, nucleases, and DNA repair proteins and believed to be an ancestral domain 20 .
C-terminal domain of SepCysE non-specifically binds tRNA. SepCysE(CTD) was bound to the elbow region of tRNA Cys mainly via van der Waals and ionic interactions. The shape of the SepCysE(CTD) surface was highly complementary to tRNA Cys , and the central region of SepCysE(CTD) consisting of secondary structures of α4, β2, α5, and α6 interacted with the TΨC and D loops of tRNA Cys (Fig. 1a). Especially prominent were the α4 and α6 helices in the groove of the tRNA, and the side-chains in these helices rich in lysine residues extended into this groove (Fig. 1b). Although the poor density map at the region of α6 hindered building of precise model of the side chains, most of the residues  is likely to be a non-specific tRNA binding domain. This idea was confirmed by a binding assay of SepCysE(CTD) with other tRNAs (tRNA Sec , tRNA Phe and tRNA His ), which indicated that SepCysE(CTD) can bind these tRNAs with affinities almost at the same level as tRNA Cys (Supplementary Fig. 3a). Therefore, we characterized SepCysE(CTD) as an ancient, non-specific tRNA binding domain. However, we cannot rule out the possibility that SepCysE(CTD) exhibits binding preference to tRNA Cys . It was reported that, in E. coli, the interaction between ribose-phosphate backbone of tRNA Cys and CysRS is essential for recognition (indirect readout) 21 . The same strategy may also be adopted by the present system. It was previously noted that SepCysE possessed similar binding affinities as SepRS for tRNA Cys , but SepCysS did not show measurable binding to tRNA Cys and the formation of binary complex with SepCysE significantly increased the affinities of SepCysS for tRNA Cys 11 . In the present ternary complex, tRNA Cys was in contact with both SepCysE(CTD) and SepCysS, while no direct interaction was observed between SepCysE(CTD) and SepCysS (Fig. 1a). Thus, it is possible that a mobile SepCysE (CTD) connected to the SepCysE(NTD)-SepCysS complex by a flexible linker, mediated the binding of tRNA Cys to SepCysS. Interactions of SepCysS to the acceptor stem of tRNA Cys may reinforce the binding of tRNA Cys to the SepCysE-SepCysS complex. Consistent with this model, the tRNA Cys binding affinity of the full complex (SepCysE-SepCysS: K d = 0.35 µM) is somewhat stronger than that of SepCysE(CTD) (K d = 0.67 µM), while the affinity of SepCysE(del_CTD)-SepCysS for tRNA Cys could not be detected ( Supplementary Fig. 3b).
SepCysS recognizes the discriminator base U73 of tRNA Cys . SepCysS is a pyridoxal-5′-phosphate (PLP)-dependent enzyme and a homodimer with a large domain (about 300 residues, residues Gly43-Lys295) and a small domain (about 100 residues, residues Trp296-Lys396) per monomer. PLP was covalently bound to the side chain of the conserved Lys234 located deep in the cleft at the dimer interface. The inner surface of the cleft was highly positively charged, along which the negatively charged acceptor stem of the tRNA Cys had access to the catalytic site deep in the cleft (Fig. 1c).
Two regions were involved in tRNA Cys binding. First, the Nterminal chain of SepCysS extended over the stem of the TΨC loop of tRNA Cys with three residues (Asn19, Asp21, Asn25) interacting with G51, C62, and C63 (Fig. 1c). Deletion of the 30 N-terminal residues (del_1-30) slightly reduced the binding of tRNA Cys (Supplementary Fig. 2b), suggesting its role in tRNA Cys binding is minor. Nevertheless, it may help direct the acceptor stem of tRNA Cys to the active site. Second, a helix-loop-helix-loop region (Pro333-Glu368) at the tip of the small domain held the acceptor stem of tRNA Cys via hydrogen bonds and electrostatic interactions (Fig. 1c). Although most of the interactions were nonspecific between the basic side chains of SepCysS and the phosphate ions of tRNA Cys , the discriminator base U73 was strongly recognized by the residues Arg345, Gly346, Phe347 and Gly364; the side chain of Arg345 interacted with the phosphate moiety of U73, and Gly346 and Gly364 together created room exclusively for U73 by specifically interacting with the uracil, and allowing the phenyl ring of Phe347 to stack with U73 and the G1-C72 base pair (Fig. 1c). These residues together acted as a gate for the discriminator base U73 while excluding non-cognate tRNAs. The residues responsible for the recognition of U73 are conserved in methanogenic archaea species (Supplementary Fig. 4a). Deletion of residues Arg344, Arg345, and Phe347 in the recognition region (del_RRF), and Ala substitution mutations at Gly346, Phe347, and Gly364 of SepCysS, reduced the binding affinity of SepCysE-SepCysS to tRNA Cys ( Supplementary Fig. 2b).
An in vivo genetic study of Methanococcus maripaludis further confirmed the importance of this region. As M. maripaludis contains a canonical CysRS 22 and can take up free cysteine from the medium, the inactivation of the whole indirect pathway by deleting both the sepS (encoding SepRS) and the pscS (encoding SepCysS) genes was not lethal. As expected, the double deletion ΔsepRSΔsepcysS mutant strain was a cysteine autotroph (Supplementary Fig. 5a). However, the deletion of pscS by itself was lethal 23 , possibly because the accumulation of Sep-tRNA Cys was toxic to the cell. Complementation of the ΔsepRSΔsepcysS strain with SepRS and wild-type SepCysS expressed from a shuttle vector restored growth without cysteine ( Supplementary Fig. 5b). However, attempts to transform the ΔsepRSΔsepcysS strain with a vector expressing SepRS and SepCysS(del_RRF) did not result in viable colonies (Supplementary Table 1), similarly to the pscS deletion experiment. This result suggested that SepCysS(del_RRF) was unable to convert Sep-tRNA Cys to Cys-tRNA Cys and that this mutated region was vital for the recognition of tRNA Cys .
A binding assay of SepCysE-SepCysS to tRNA Cys variants confirmed the importance of U73. The tRNA Cys (U73G) had a weakened binding affinity to SepCysE-SepCysS, while tRNA Phe (A73U) acquired binding ability towards SepCysE-SepCysS ( Fig. 1d), suggesting that U73, but not the anticodon of tRNA Cys , was the major identity element for SepCysS. U73 in tRNA Cys is conserved in the three domains of life, and has been reported as a discriminator for CysRS and SepRS 16 . Here we found that SepCysS also utilizes this recognition element. The fact that the discriminator base U73 is cross-recognized by SepRS and SepCysS may explain, at least partly, the relative insensitivity of SepRS to mutation of this nucleotide base compared with CysRS 17 . SepCysE dimer binds to SepRS by a pair of N-terminal helices. During the study of the quaternary complex SepRS-SepCysE-SepCysS-tRNA Cys , we serendipitously obtained the crystal of the ternary complex SepRS-SepCysE-SepCysS. The structure was solved by the molecular replacement method (Table 1). In the electron density map, whole SepRS and the N-terminal helix (Met1-Lys24) of SepCysE(N-helix) were visible ( Supplementary  Fig. 6a). However, the remaining components of SepCysE and SepCysS were disordered in the crystal, although there was enough space in the unit cell ( Supplementary Fig. 6b) to accommodate the entire complex. SepRS was organized as a tetramer with 222 local symmetry (one of the two-fold axes is crystallographic), as previously observed 13,14 . A pair of Nterminal helices (N-helices) of the dimeric SepCysE wound around each other to form a coiled-coil structure and were tightly bound to the SepRS with their two-fold axes aligned (Fig. 2a). This N-helix was not visible in the structures of the ternary complex SepCysE-SepCysS-tRNA Cys , suggesting that an intervening stretch of about 10 residues (Ile25-Ile34) between N-helix and NTD was a flexible linker. This view was consistent with a secondary structure prediction that a long loop connects N-helix and NTD.
The highly conserved hydrophobic residues Leu9, Ile10, Ile17, and Leu20 ( Supplementary Fig. 4b) in the N-helix contributed to the hydrophobic interactions within the left-handed coiled-coil structure. Most of these residues (Leu9, Ile10, and Ile17) plus Ile14 were clustered with the hydrophobic residues in three pairs of loops ((loop 1: 376-380 ILDEF), (loop 2: 411-414 FNGE), and (loop 3: 494-498 ALVSN)) locating at the interface of two anticodon domains of the SepRS dimer (Fig. 2b) and contributed to the formation of the complex together with some hydrogen bonds (Fig. 2c). To verify the physiological importance of these interactions, several deletion mutants were made and the binding ability was checked by gel filtration experiments. Whereas deletion mutants of each of the three loops of SepRS showed no significant effect on binding, the double-loop deletion mutants did not bind SepCysE, suggesting these loops together contribute to the interaction with SepCysE ( Supplementary Fig. 7). Likewise, N-helix deletion mutant SepCysE(ΔN-helix) completely lost the ability to bind SepRS ( Supplementary Fig. 7). This structural and mutational evidence suggested that the SepCysE dimer tightly binds to the SepRS dimer by inserting its coiled-coil N-terminal helices into the interface of the SepRS dimer, which reinforces the stable architecture of transsulfursome. Dynamic architecture of the transsulfursome. Two complex structures solved in the present work were combined to show that SepCysE consisted of an N-terminal helix (N-helix), an Nterminal domain (NTD) and a C-terminal domain (CTD). These components were connected by two linkers, linker1 and linker2, each composed of about 10 residues (Fig. 3, Supplementary  Fig. 4b). Each component of SepCysE bound to each of the three molecules, SepRS, SepCysS, and tRNA Cys . Both the N-helix and NTD bound as dimers to the dimeric molecules of SepRS or SepCysS, while the CTD worked as "two monomers" of which only one "monomer" bound to monomeric tRNA Cys (Fig. 3). The two intervening linkers (linker1: between N-helix and NTD, and linker2: between NTD and CTD) made SepCysE highly flexible, enabling the movement of the SepCysE(NTD)-SepCysS complex with respect to the SepRS-SepCysE(N-helix) complex and also allowing a large motion of the SepCysE(CTD) with bound tRNA Cys during tRNA transfer from SepRS to SepCysS (see below). The flexibility of SepCysE(NTD)-SepCysS relative to SepRS via linker1 was further confirmed by SEC-SAXS and EM analyses.
A small angle X-ray scattering (SAXS) experiment was performed to study the structure of the transsulfursome in solution. For simplicity, CTD-truncated SepCysE was used for this experiment. A plot of the intensity of the scattering data (I), as a function of scattering angle (Q) was compared with the transsulfursome model derived from the crystal structures. To take account of the structural variability of the transsulfursome due to flexible linker1, a number of models were generated by changing the distance and orientation of these two rigid structures, and the theoretical curves calculated from the models were compared with the observed plot. As shown in Fig. 4a, the observed plot matched well to a theoretical curve with χ 2 agreement of 1.24 for "a straight model" after adjusting the distance between the two rigid structures, SepRS-SepCysE(Nhelix) and SepCysE(NTD)-SepCysS connected by linker1. Whereas the fitting was sensitive to the distance between the two rigid structures, it was rather insensitive to the tilt angles and even better fitted models were obtained when the angle was also varied (Fig. 4b). Thus, an ensemble distribution of the structures with various tilt angles by linker1 was expected in solution.
The structure of the transsulfursome complex was further characterized by negative stain electron microscopy and class averaging. The particles were adsorbed onto the EM grid in side orientation. Over 20,000 particles were selected from the EM micrographs and subjected to image alignment and classification. This analysis confirmed that the complex possessed an overall architecture, roughly resembling a "three sectional staff". Tetrameric SepRS was positioned in the center of the complex and dimeric SepCysE-SepCysS at the opposite ends (Fig. 4c). The SepCysE-SepCysS domains did not occupy stable positions but instead flexed independently of the SepRS domain. Thus, the complex could adopt straight and/or curved conformations without changing its overall composition (Fig. 4d). The wildtype complex underwent a tilting-motion from a curved to an Sshaped conformation ( Supplementary Fig. 8a, Supplementary Movie 1) and the distribution of the center-to-center distance between the SepCysE-SepCysS components at the opposite ends of the complex ranged from 170 to 290 Å (Fig. 4d). This suggests that a hinge exists at the site of linker1 connecting SepRS-SepCysE(N-helix) and SepCysE(NTD)-SepCysS, and allows movement of these domains relative to each other. Because of the flexibility of the complex and preferred orientation on the EM grid, we were not able to determine a low resolution 3D volume.
To support this notion, we truncated 10 residues of linker1 (transsulfursome (del_linker1)) and analyzed the resulting complex by negative stain EM and image classification. As expected, shortening of the flexible connector, resulted in reduction of the overall flexibility of the complex. Most of the class-averages, obtained from~12,000 particles of this complex, appeared straight and the tilting-motion of SepCysE-SepCysS relative to SepRS was greatly reduced (Fig. 4d, Supplementary  Fig. 8b, Supplementary Movie 2). The increased rigidity was also reflected in the narrower distribution of the distance between the SepCysE-SepCysS domains (230-260 Å, Fig. 4d). Thus, motion of SepCysE-SepCysS relative to SepRS is linker1 dependent.

Discussion
The ternary complex SepCysE-SepCysS-tRNA Cys solved in the present work presents the structure of which the 3′-end moiety of tRNA Cys is at the active site for the second reaction, while SepRS-tRNA complex solved previously shows the structure for the first reaction 13 . Distance between these two active sites in the transsulfursome is not fixed due to the conformational change caused by the flexible linker1. However, model building shows that the distance is no less than 90 Å. Therefore, tRNA Cys must move a long distance between these two sites during two step reactions. Furthermore, as both SepRS and SepCysS access to the same side of tRNA Cys , it is not possible for SepRS and SepCysS to bind to the tRNA Cys simultaneously and to make a ternary complex. In contrast, as SepCysE(CTD) accesses to tRNA Cys from opposite side, it binds to the tRNA Cys without disturbing the binding of the SepRS or SepCysS to tRNA Cys . These considerations suggest a scenario for this consecutive reactions where SepCysE(CTD) carries tRNA Cys from the first active site to the second site ( Fig. 5a). At first, tRNA Cys bound to SepCysE(CTD) binds to the active site of SepRS while SepCysE(N-helix) remains fixed to SepRS. The model building shows such complex is possible by extending the two linkers of SepCysE (Fig. 5a, left). This is consistent with the experiment that SepCysE enhances the binding of SepRS and tRNA Cys 11 . After Sep is attached at the first reaction site, CCA terminus of the tRNA Cys leaves from the active site (Fig. 5a, middle) and rotates with the SepCysE(CTD) to reach binding site of SepCysS. The structure of the present ternary complex shows the structure at this stage. SepCysS recognizes the discriminator base U73, and possibly Sep moiety of tRNA Cys , and the tRNA Cys separates from the anticodon recognition region of SepRS when CCA terminus binds to the second reaction site (Fig. 5a, right). In this model, two linkers are critically important to satisfy the flexibility required for this long-distance swing of tRNA Cys . Therefore, we examined this model by making two SepCysE mutants of which each of two linkers were truncated. The importance of the linkers was studied by an in vivo genetic study. The ΔsepRSΔsepcysE double deletion mutant strain is a cysteine auxotroph, which can be complemented by expressing wild-type SepRS and SepCysE from a shuttle vector. However, expression of wild-type SepRS together with either SepCysE(del-linker1) or SepCysE(del-linker2) in ΔsepRSΔsepcysE still required cysteine for growth (Fig. 6). This result indicates that both linkers are essential for the function of the transsulfursome. The multiple domain architecture of SepCysE confers flexibility to the transsulfursome, which enables SepCysE-SepCysS moiety to bend to SepRS to facilitate the access of tRNA acceptor arm to active site of SepRS. However, this bending keeps SepRS out of reach for another half of the dimeric moiety of SepCysE-SepCysS. That is, the dimeric SepCysE-SepCysS can convey only one tRNA at a time (Fig. 5b). In all, the transsulfursome can afford to catalyze two tRNAs at a time although it possesses four sets of active sites. This is compatible with our result that one transsulfursome only binds two tRNAs ( Supplementary Fig. 9) and also to the previous results that the tetrameric SepRS binds two tRNAs and that only two of the four chemically equivalent subunits are active 24 . It is possible that the left and the right active sites of the transsulfursome are alternately used for reaction. As a reference, the alternate mechanism is believed to be adopted by HisRS 25 . The highly interdigitated structure of SepRS may facilitate the alternate catalysis 24 . On the basis of these results, we propose the scheme of the tRNA-dependent cysteine biosynthesis by the indirect pathway (Fig. 5b), where multidomain SepCysE binds SepRS and SepCysS to assemble into a functional unit (transsulfursome) and mediates the transfer of tRNA substrate between the active sites of SepRS and SepCysS via flexible linkers. The released final product Cys-tRNA Cys is delivered to ribosome by EF-1α to incorporate a cysteine into proteins.
Among the indirect pathways known today, the system for Gln/Asn-tRNA Gln/Asn synthesis is best studied. The complex named transamidosome is responsible for this synthesis. The transsulfursome studied here is very different from this system, although both systems are involved in the indirect aminoacylation. First, formation of the transsulfursome is tRNAindependent and could be stable throughout the two reactions, whereas transamidosome does not form a stable complex 26,27 . This is reasonable because dynamic dissociation of the complex is important for transamidosome, as non-discriminating GluRS/ AspRS show dual-specificity to tRNA Glu /tRNA Asp and tRNA Gln / tRNA Asn28, 29 , and the second enzyme amidotransferase (Gat-CAB) is also employed in the two kinds of transamidosome [30][31][32] . By contrast, the transsulfursome is specialized for synthesizing Cys-tRNA Cys , and stable formation of the transsulfursome is beneficial for the efficient function of this complex in a hightemperature habitat. Second, unlike the case of the transamidosome, both SepRS and SepCysS have access to the same site of the major groove of tRNA Cys to recognize the discriminator base U73 13,16,17 . This suggests that the two enzymes evolved independently and only at a later stage they were combined together to form a system for Cys-tRNA Cys synthesis. This scenario implies that it is difficult to form a ternary complex such that the CCA terminus of tRNA is close to both active sites of the two enzymes and that the system requires a large movement of tRNA rather than just flipping the acceptor arm of tRNA between the two active sites, as in the case of transamidosome. Third, whereas in the transamidosome the first enzyme (Glu/AspRS) appears to continuously recognize the anticodon of the cognate tRNA throughout the entire two-step reaction, our model of the transsulfursome suggests that the anticodon is recognized only at the first step. In fact, it is unlikely that SepRS could specifically recognize tRNA Cys in both steps because tRNA Cys needs to be accessed from completely different directions at each step. The recognition of U73 of tRNA Cys by SepCysS also implies that the anticodon recognition by SepRS is redundant in the second step, although we cannot rule out the possibility that tRNA Cys maintains some kind of interaction with SepRS to safeguard the transfer of tRNA Cys .
The present structure analysis revealed how SepCysE assembles two enzymes (SepRS and SepCysS) and a substrate (tRNA) together to create a mechanism of aminoacylation of tRNA in the indirect pathway. However, the indirect pathway for aminoacylation of tRNA is associated with the risk that the mischarged intermediate is used for protein synthesis. Therefore, the direct pathway is employed in most extant organisms. Indeed, in some archaeal lineages, CysRS was horizontally transferred and replaced the indirect mechanism 10,33 . Nonetheless, the ancestral indirect pathway for Cys-tRNA Cys synthesis has been preserved in methanogenic archaea, although the reason for this is unclear. One possible explanation is the energetic advantage by the direct link of amino acid synthesis and protein synthesis in methanogens where cysteine in the proteome is in a large demand 34 . Whatever the reason for this would be, the structure of transsulfursome reflects the most primordial feature of a multiprotein complex each of whose components have acquired individual functions. SepCysE, which is made up of three different binding    18 . X-ray diffraction of the crystal was collected on a beamline BL41XU at the SPring8 (Harima, Japan), and the data were processed using the XDS and CCP4 packages 18 . The crystal structure of SepCysE-SepCysS-tRNA Cys was solved by the molecular replacement method 18 and the C-terminal domain of SepCysE was manually constructed with Coot 37 . Initial protein models were fitted manually using Coot and tRNA models were automatically rebuilt by LAFIRE_NAFIT 38 . Structure refinement was performed using phenix.refine 39 . After refinement, R work and R free factors converged to 22.5% and 26.1%, respectively. Statistical information of the structure refinement is summarized in Table 1.
Crystallization and data collection of transsulfursome. In an attempt to solve the structure of the SepRS-SepCysE-SepCysS-tRNA Cys quaternary complex, 0.75 μL of solution containing 50 μM SepRS-SepCysE-SepCysS(del_RRF), 30 μM tRNA Cys , and 10 mM ATP were mixed with an equivalent volume of reservoir (0.1 M Tris-HCl pH 7.0, 0.3 M ammonium sulfate, 36% (w/v) MPD). The welldiffracted crystals were obtained after dehydration trials. Diffraction data were collected on a beamline BL-5A at the Photon Factory (Tsukuba, Japan). The data were processed using the XDS package 40 .
Structure determination and refinement of transsulfursome. The structure was solved by the molecular replacement method using AutoMR of the Phenix package. The structures of M. jannaschii SepRS (PDB entry 2DU7), M. jannaschii SepCysS-SepCysE(NTD) (PDB entry 3WKR), and E. coli tRNA Cys (PDB entry 1B23) were used as search models. However, only SepRS was found by this search. In the Fo-Fc map subsequently created, a pair of N-terminal helices of SepCysE was obtained. ATP was also found in the active site of SepRS. Structure refinement was performed using phenix.refine with secondary structure restraints, and final R work and R free factors converged to 18.5% and 20.9%, respectively. Statistical information of the structure refinement is summarized in Table 1.
Isothermal titration calorimetry experiment. Isothermal titration calorimetry (ITC) experiment was performed to investigate the interaction of proteins and tRNA. Protein and tRNA were dialyzed to buffer containing 20 mM HEPES-KOH pH 8.0, 150 mM KCl, 5 mM MgCl 2 , and 10% glycerol. The protein samples were loaded to the cell of Nano ITC (TA Instruments) and 25 aliquots of 2 µL of tRNA Cys were injected into the cell with a time interval of 180 s under a stirring speed of 250 rpm at 60°C. The data were analyzed using the program NanoAnalyze. The control (buffer to buffer, protein to buffer, and tRNA to buffer) was also performed to normalize the experiment.
Gel filtration assay. The binding affinity of SepCysE-SepCysS and tRNA mutants was analyzed by gel filtration. Structural analysis in solution by SEC-SAXS. Small angle X-ray scattering incorporated with size exclusion chromatography (SEC-SAXS) was used to analyze the structure of SepCysE(CTD)-truncated transsulfursome in solution. SEC-SAXS data collection was carried out under room temperature at the beamline BL-10C, Photon Factory (Tsukuba, Japan). A wavelength of 1.500 Å was used, and the specimen-to-detector distance was 2,010 mm, as calibrated with silver behenate. Sample was loaded onto a Superdex 200 Increase 10/300 GL column (GE Healthcare) and the collected fractions were directly exposed to X-rays. During data collection, the flow rate was set at 0.05 ml min −1 and total 180 X-ray scattering images were collected for 60 min. The scattering images were recorded on a PILATUS3 2 M detector (Dectris) and each two-dimensional scattering image was circularly averaged to convert to the one-dimensional scattering intensity data by using the software SAngler 41 . The pair distribution function P(r) was calculated using DATGNOM4 40 , and all results are summarized in Supplementary Fig. 10 and Supplementary Table 3. Simulation experiment using crystal structure was carried out using CRYSOL programs 42 . The experimental data used for comparison with a theoretical profile is the data measured at the peak position in SEC-SAXS measurement since the interparticle interference was not observed in the serial data of SEC-SAXS.
Negative stain electron microscopy and image processing. The 4 μL of wild type transsulfursome and mutant (del_linker1) were adsorbed for 45 s at room temperature to freshly glow-discharged carbon-coated copper-grids (Agar scientific; G2400C). After washing with a 10 μL drop of buffer composed of 20 mM HEPES-KOH (pH 8.0), 200 mM KCl, and 5 mM MgCl 2 and a 10 μL drop of 0.75% freshly prepared uranyl formate solution, the grids were negatively stained with another drop of uranyl formate for 45 s and air-dried. Images were collected using a FEI G Spirit TEM operated at 120 kV, at a nominal magnification of 67k on a 4k × 4k CMOS camera F416 (TVIPS). Particles were selected using e2boxer. The wild type and mutant transsulfursome data sets contained 29,266 and 11,972 single particles, respectively. Because of the overall flexibility of the complex, focused reference-free alignment and classification procedures implemented in SPHIRE 43 and EMAN2 44 were applied 45 . In particular, images were first centered and rotationally aligned, using the reference free approach implemented in SPHIRE. Subsequently, a k-means classification was performed, within a mask including only SepRS in the center of the complex and excluding the peripheral and flexible SepCysE-SepCysS densities. In total 10-20 class averages were produced showing sufficient amount of details for SepRS, whereas SepCysE-SepCysS appeared rather blurred. Then we extracted the members of each class-average and performed a second round of k-means classification for each resulting subset, this time using a mask including the complete complex. The final class-averages contained 24-167 members, showed improved features for all components and were used to measure the center-to-center distance between the SepCysE-SepCysS components at the opposite ends of the complex, as shown in Fig. 4c, d. A sequence of selected classaverages in random order was used to create a movie ( Supplementary Movie 1 and 2).
Mutagenesis of Methanococcus maripaludis. The replacement of the scsS gene (mmp1240, encoding SepCysS) with a puromycin resistance (pac) cassette 46 was generated in the markerless ΔsepS strain 11 by homologous recombination. We confirmed the genotype of the ΔsepRS/ΔsepcysS::pac strain (S700) by Southern hybridization. For expression of SepRS and SepCysS, the two genes, mmp0688 and mmp1240, were inserted into the vector pMEV2 under the control of the hmvA 47 and sla promoters, respectively. The sepRS + sepcysS + strain (S702) was then obtained by transformation of S700 with pMEV2-mmp0688-mmp1240. The Sep-CysS(del-RRF) variant was constructed with the plasmid pMEV2-mmp0688-mmp1240 using the QuikChange mutagenesis kit (Agilent Technologies) and then transformed into strain S700. The mutations of SepCysE (MMP1217), including the variants of SepCysE(del-linker1) and SepCysE(del-linker2), were constructed by using the plasmid pMEV2-mmp0688-mmp1217 and then transformed into the ΔsepRS/ΔsepcysE::pac strain (S761) 11 . We grew M. maripaludis at 37°C in 28-ml aluminum sealed tubes with 5 ml of McFAA medium (minimal medium + 0.4 M sodium formate + 10 mM sodium acetate + 1 mM L-Ala) buffered with 0.2 M glycylglycine (pH 8.0) and reduced with 3 mM dithiothreitol 11 . L-Cys (3 mM), puromycin (2.5 µg ml −1 ), and neomycin (0.5 mg ml −1 in plates and 1 mg ml −1 in broth) were added as needed. Antibiotics were omitted when comparing the growth of the WT and mutants. As the sulfur source, 3 mM sodium sulfide was added before inoculation 48 .