Transfer RNA nucleotidyltransferases (CCA-adding enzymes) are responsible for the maturation or repair of the functional 3′ end of tRNAs by means of the addition of the essential nucleotides CCA. However, it is unclear how tRNA nucleotidyltransferases polymerize CCA onto the 3′ terminus of immature tRNAs without using a nucleic acid template. Here we describe the crystal structure of the Archaeoglobus fulgidus tRNA nucleotidyltransferase in complex with tRNA. We also present ternary complexes of this enzyme with both RNA duplex mimics of the tRNA acceptor stem that terminate with the nucleotides C74 or C75, as well as the appropriate incoming nucleoside 5′-triphosphates. A single nucleotide-binding pocket exists whose specificity for both CTP and ATP is determined by the protein side chain of Arg 224 and backbone phosphates of the tRNA, which are non-complementary to and thus exclude UTP and GTP. Discrimination between CTP or ATP at a given addition step and at termination arises from changes in the size and shape of the nucleotide binding site that is progressively altered by the elongating 3′ end of the tRNA.
Transfer RNA nucleotidyltransferases (CCA-adding enzymes) catalyse the post-transcriptional addition of CCA onto the 3′ terminus of immature tRNAs using CTP and ATP as substrates1. Although the CCA end of tRNA is universally conserved and absolutely required for protein biosynthesis2,3, it is not encoded in many eubacterial or archaeal tRNA genes, or in nearly all eukaryotic tRNA genes4. In the few organisms that do have the CCA sequence included in their genes for tRNA (such as Escherichia coli), the CCA-adding enzyme serves to maintain the integrity of the CCA terminus of tRNA5.
CCA-adding enzymes belong to the nucleotidyltransferase superfamily6 whose members are divided into two classes on the basis of their sequences7. Examples of class I enzymes include the archaeal CCA-adding enzyme, DNA polymerase β (Pol β), eukaryotic poly(A) polymerase (PAP) and terminal deoxynucleotidyltransferase (TdT). Class II enzymes include the eubacterial and eukaryotic CCA-adding enzymes and eubacterial PAP. Crystal structures of CCA-adding enzymes from both classes confirm that they share only a common amino-terminal catalytic domain, which is also homologous to the polymerase domain found in all nucleotidyltransferase family enzymes8,9,10,11.
The ability of the CCA-adding enzyme to add specific nucleotides in the absence of a nucleic acid template makes it one of the most intriguing polymerases. Currently, only PAP, TdT and CCA-adding enzymes are known to function without a nucleic acid template, and between them the CCA-adding enzymes exhibit the highest degree of specificity for both sequence and length. Biochemical studies, sequence analyses and structural data have established that the underlining chemistry for nucleotide addition is the same two-metal-ion-catalysed reaction used by all polymerases12; however, insights into the source of the CCA-adding enzyme's specificity of nucleotide incorporation are scarce.
Several models explaining how the enzyme incorporates CTP and ATP with high specificity have been proposed. Early models proposed multiple binding sites for CTP and ATP13,14,15, but these were excluded owing to the existence of a single nucleotidyltransferase active site in the sequence6,16 and observations of a single nucleotide-binding site in the crystal structures of both class I and class II CCA-adding enzymes8,10. Subsequent models were based on additional biochemical and structural data. Both the ‘collaborative templating model’ and the ‘scrunching model’ postulated a progressive refolding of the growing CCA terminus of the tRNA, which allows for the re-use of the single nucleotide-binding pocket while altering the templating specificity of the enzyme8,17. The structure of the class II CCA-adding enzyme of Bacillus stearothermophilus (BstCCA) binds both CTP and ATP, but excludes UTP and GTP, by binding to the Watson–Crick base edge of both nucleotides using the same protein side chains8. Recently, a model of the complex between a class I CCA-adding enzyme and tRNA was proposed10 that required the 3′ terminus of tRNA to unstack from the acceptor stem in order to position the priming nucleotide correctly for reaction with the bound NTP. In contrast to the conclusion of another study11, this model also required the binding of the tRNA substrate for nucleotide selection, as all NTPs bound to the enzyme in its absence10.
Although these latter models could explain certain aspects of the experimental observations, the central question regarding the specificity of the CCA-adding enzyme remained unanswered: how does the CCA-adding enzyme achieve its specificity of nucleotide incorporation without using a nucleic acid template? The crystal structures of the class I CCA-adding enzyme of A. fulgidus (AfCCA) in complex with tRNA, as well as its complexes with two RNA duplex mimics of the tRNA acceptor stem that end with the nucleotides C74 or C75 in the presence of the appropriate incoming nucleotide triphosphates, now provide the structural basis of this specificity.
The co-crystal structure of AfCCA bound to full-length yeast tRNAphe was determined at 6.5 Å resolution by the molecular replacement method using the structure of the dimeric apo-AfCCA as a search model10. As a consequence of the unusually high solvent content of ∼84% in the crystal, cycles of solvent electron density flattening were extremely effective in improving the electron density map and, together with two-fold non-crystallographic symmetry (NCS) averaging, produced a map of exceptional quality. All parts of the complex including the carboxy-terminal histidine tag of the AfCCA enzyme, which was not included in the model, exhibited well-defined electron density (see Supplementary Fig. S1). The 3′ terminus of the tRNA was built into an unbiased electron density map that was calculated before its inclusion in the model (Supplementary Fig. 2a). The structure of this complex was refined as groups of rigid bodies to a free R-factor of 41%.
AfCCA was also crystallized with synthetic RNA duplexes mimicking the acceptor stem of the tRNA: they have the nucleotide corresponding to either C74 or C75 at their 3′ termini, and are named AC74 and ACC75, respectively (Fig. 1c). Self-complementary RNA oligonucleotides were used to ensure the binding of the appropriate terminus to the active site as well as the formation of homogeneous duplexes in solution. The ternary complexes with incoming nucleotides were prepared by soaking ATP or CTP into crystals of the binary complexes, or by co-crystallizing the enzyme with ACC75 and a non-reactive αβ-methylene ATP. The crystal structures of the AC74 and ACC75 ternary complexes were determined by molecular replacement using the apo-AfCCA dimer as a search model. The initial phases derived from the protein model were improved by NCS averaging within each crystal and by cross-crystal averaging between highly non-isomorphous ternary complex crystals as well as the apo-AfCCA crystal10. The structures of the incoming nucleotides and the 3′ terminal nucleotides were built into unbiased electron density maps that were calculated before their inclusions in the models (Supplementary Fig. S2b, c). The final models were refined to free R-factors of 29.1% and 24.1% at resolutions of 3.4 Å and 2.2 Å, respectively (Table 1).
Overall structures of the complexes
Two tRNA molecules bind to the AfCCA dimer with each tRNA contacting only one subunit (Fig. 1a, b). As predicted in our earlier model10, the tRNA acceptor stem and T-stem loop are inserted into the extended cleft on the enzyme, with the CCA terminus being in the active site of the head domain, the T loop contacting the tail domain and the anticodon stem loop pointing away from the enzyme (Supplementary Movie 1). Here, the head, neck, body and tail domains of the enzyme are defined as in the apo-AfCCA structure10. The role of the body and tail domains of the class I enzyme in binding tRNA supports the proposal10 that the corresponding domains also bind to tRNA in class II enzymes because their shape and dimensions are similar, and they are complementary to the shape of the acceptor and T stems of tRNA. The duplex RNA constructs that mimic the acceptor stem of the tRNA bind to AfCCA in an almost identical fashion to tRNA, and they superimpose on the bound tRNA particularly well in the regions that make contact with the enzyme (Fig. 1d, e), consistent with the observation that a tRNA mini-helix functions as an efficient substrate for CCA addition18. The enzyme does not add CCA to other RNAs in the cell, presumably because they do not form a simple duplex structure of appropriate length containing a suitable 3′ nucleotide overhang.
The acceptor stems of the AC74 and ACC75 mimics, as well as the tRNA product, bind in the same location and orientation on the enzyme, in agreement with the conclusion derived from biochemical crosslinking studies that the tRNA does not translocate during CCA addition17. Comparison of the protein structures of these three complexes with that of the apo-AfCCA shows that there are no large conformational changes in the enzyme except for a small repositioning of the N-terminal head domain. Except for the 3′-terminal CCA bases, the enzyme interacts almost exclusively with the sugar-phosphate backbone of the tRNA through electrostatic, hydrogen-bonding and hydrophobic interactions, which is consistent with the enzyme's ability to bind any tRNA molecule in a sequence-independent manner. The only hydrogen bond to a base, which is between the O2 atom of C72 and Tyr 169, is nonspecific.
Snapshots of the CCA addition process
These three crystal structures represent snapshots of three steps that occur during CCA addition (Supplementary Movie 2). Throughout this process, the acceptor stem along with the discriminator base A73 remains fixed on the enzyme, whereas the growing 3′ terminus of the tRNA refolds to position the primer terminal nucleotide identically at each step, and allow the identical binding of the incoming CTP or ATP in a single pocket (Fig. 2). As anticipated in our earlier modelling study10, the base of the 3′-terminal nucleotide is not stacked on the acceptor stem bases. Rather, for both AC74 and ACC75, the base of the primer nucleotide flips about 90° from its stacking position on the acceptor stem and stacks onto the base of the incoming nucleoside triphosphate.
The structure of AfCCA in complex with AC74 and incoming CTP shows the base of the primer nucleotide C74 stacking on the base of the incoming CTP with its 3′ hydroxyl groups positioned to attack the α-phosphate of CTP (Fig. 2a). C74 is disordered without the incoming CTP in the crystal structure of the binary AfCCA–AC74 complex (data not shown). At this step in the process, the flat plane of the peptide connecting Ala 95 and Glu 96 and lying at the apex of a β-turn region in the AfCCA head domain stacks onto the base of A73.
After C75 has been incorporated, C74 moves from the primer position to a location closer to the acceptor stem (Fig. 2b); however, instead of stacking on the acceptor stem, the base of C74 forms four hydrogen bonds between its Watson–Crick base edge and its 2′-OH group and protein residues 95–99 located in the β-turn. The newly incorporated C75 shifts into the primer position, with its 3′ hydroxyl group pointed towards the α-phosphate of the newly arrived ATP, ready for the next round of nucleotide addition.
Upon the incorporation of A76, the CCA terminus adopts a continuously stacked helical conformation emanating from the acceptor stem, with the base surface of A76 packed against the β-sheet in the head domain of AfCCA (Fig. 2c).
Active site geometry and the nucleotide-binding pocket
The conformations of the catalytic site and the bound substrates observed in the current crystal structures (Fig. 3a) are nearly identical to the corresponding structure of the ternary complex of DNA Pol β19, which is also a class I nucleotidyltransferase. When the structure of the AC74 complex is aligned on that of the Pol β ternary complex by superimposing the catalytic carboxylates as well as the triphosphates of the incoming nucleotides, the primer terminal nucleotides of both complexes are also superimposed. Although the duplexes containing the primer nucleotide are bound to these two enzymes with orientations that differ by almost 90° (Fig. 3b), the primer termini are presented to the catalytic site in the same way. The class I AfCCA achieves this by unstacking the 3′-terminal nucleotide from the acceptor stem of the tRNA and rotating it by nearly 90°. In class II BstCCA, the modelled tRNA acceptor stem10 approaches the homologous catalytic domain from a direction that is perpendicular to both the duplex substrates bound to Pol β and AfCCA (Fig. 3b). As a consequence, the primer nucleotide must also unstack and flip about 90° to be positioned in the primer position observed in both the Pol β and AfCCA ternary complexes.
Two divalent metal ions (A and B) are observed bound to the catalytic carboxylates and the 3′ terminus of the primer strand of AC74 (Fig. 3a), in agreement with the proposal that all polymerases use a common two-metal-ion mechanism for their nucleotide incorporation12. Only one metal ion (B) is observed in the ACC75 complex, which has its 3′ terminus in a slightly different conformation (Supplementary Fig. 3). This is presumably due to the experimental conditions (see Methods).
The nucleotide-binding pocket for the incoming CTP or ATP is formed by residues from the head and neck domains of the enzyme as well as by backbone phosphates located near the 3′ terminus of the tRNA (Fig. 2). This observation that tRNA forms a component of the nucleotide-binding site explains why apo-AfCCA is unable to distinguish between correct and incorrect NTPs and why the base is not ordered in the binary complex10. The triphosphate moieties of both CTP and ATP interact extensively with the enzyme, as is the case in the binary complexes without tRNA. The 2′-OH group also forms a hydrogen bond to His 133 in both complexes.
The bases of the incoming nucleotides are recognized by both the enzyme and the tRNA by means of hydrogen-bonding interactions to their Watson–Crick edges. The exocyclic N4 atom of CTP forms two hydrogen bonds with the tRNA backbone phosphate of A73 and C74, whereas the N3 atom of CTP is recognized by the Arg 224 guanidinium group of AfCCA (Fig. 4a). Similarly, the N6 of ATP forms hydrogen bonds with phosphate oxygens of A73 and C74, and the N1 atom also interacts with Arg 224 (Fig. 4b). This arginine, conserved between class I CCA-adding enzymes, adopts different rotamers depending on whether or not the tRNA is present. Thus, the tRNA and the protein collaborate to form a binding pocket for the incoming nucleotide, consistent in some ways with an earlier proposal based on biochemical studies18.
Specificity for nucleotide selection and termination
It is clear from the hydrogen-bonding patterns between the incoming bases and the enzyme in the ternary complexes that discrimination against UTP and GTP arises owing to their lack of hydrogen-bonding complementarity to the binding site for the base (Fig. 4a, b). Because the backbone phosphate oxygens of the tRNA that form part of the base-recognition site are obligate hydrogen bond acceptors, they can only interact with hydrogen bond donors such as the N4 of CTP or the N6 of ATP. By the same token, base recognition by Arg 224 requires an acceptor such as the N3 of CTP or the N1 of ATP. As with the class II CCA-adding enzyme of B. stearothermophilus8, the Watson–Crick face of U nor G is not complementary to the base-binding site because their patterns of hydrogen donors and acceptors are the reverse of that for C and A. As a consequence, UTP and GTP are excluded from the binding site. The difference between the two classes of enzyme, however, seems to be whether the discrimination requires tRNA. With class II CCA-adding enzymes, the incoming NTPs can be recognized by protein side chains8, albeit in the absence of tRNA.
In contrast to its discrimination against UTP and GTP, the CCA-adding enzyme uses the more subtle strategy of exploiting the size difference between pyrimidine and purine bases to distinguish between CTP and ATP. The size of the nucleotide-binding pocket changes to accommodate the appropriate incoming NTP at different steps of nucleotide incorporation. For the tRNA analogue terminated at nucleotide C74, the binding pocket provides a snug fit for CTP (Fig. 4c), but is too small to allow the correct positioning of both the α-phosphate and base of an ATP. If an ATP molecule enters the catalytic site with its triphosphate bound to the location required for catalytic insertion, its Watson–Crick base edge clashes with both the enzyme and the tRNA (Fig. 4d). After the addition of C75, the binding pocket becomes enlarged sufficiently to accommodate an ATP (Fig. 4e). Because the binding pocket can still accommodate CTP, discrimination against CTP at this step may not be absolute. In this enlarged binding site the base of a bound CTP cannot reach and interact with Arg 224 and tRNA specificity determinants, and thus presumably cannot achieve an enzymatically productive binding (Fig. 4f). Indeed, biochemical studies in vitro show that in the absence of ATP both classes of CCA-adding enzyme can add a poly(C) tail to the 3′ terminus of immature tRNA15, with a twofold increase in Km and a 20-fold decrease in kcat for CTP compared with ATP at the step of adding nucleotide 76.
The interactions between the growing 3′ terminus of the tRNA and the enzyme allow the enzyme to detect the length of the 3′ terminus and change the size of the binding pocket. When C74 is presented to the active site as the primer nucleotide, the β-turn residues Ala 95 and Glu 96, which are part of the binding pocket opposite to the base-determinant side, stack onto A73 so that the enzyme is in a slightly closed conformation. This conformation produces a binding pocket that fits CTP. After C75 is added, C74 moves out of the primer position and makes extensive hydrogen-bonding interactions with protein residues 95–99. These interactions cause unstacking between the enzyme and A73, and drive open the nucleotide-binding site into a conformation that accommodates ATP. Because these interactions with C74 make use of all three Watson–Crick hydrogen-bonding sites of the base C, they are highly specific and might potentially provide a proofreading mechanism that detects an incorrectly incorporated nucleotide at position 74. Only the 3′ terminus with a correctly incorporated C74 would be able to alter the binding pocket and proceed to the addition of A76.
The termination of nucleotide addition seems to be associated with the formation of a stable, helically stacked tRNA CCA end, the 3′ terminus of which is not positioned in the primer-binding site. The stacked 3′-terminal nucleotides of the tRNA product are packed against a β-sheet in the catalytic site of the AfCCA complex (Fig. 2c), which may further stabilize the 3′-terminal nucleotides in the stacked conformation. The stability of this stacked base conformation may prevent A76 from adopting a catalytically active geometry for the next round of nucleotide addition, which would require that A76 rotate about 90° away from its observed position. Furthermore, the limited binding pocket size seems to be insufficient to accommodate the two-nucleotide bulge required to position A76 for further incorporation.
The requirement for a movable head domain
The ability of the flexibly attached catalytic head domain to reposition in response to the length of the growing 3′ end of the tRNA is responsible for the alterations of the binding pocket that render it specific for CTP, ATP or termination. The changes in the size and shape of the binding site are not generated by local structural rearrangements, but rather they are achieved through a progressive repositioning of the head domain (Fig. 5). The flexibility of the head domain seems to be an intrinsic property of the AfCCA enzyme, as it has a high thermal B-factor and its orientation relative to the rest of the enzyme differs slightly in all of the complex structures. In fact, the head domain was largely disordered in one of the apo-AfCCA crystal forms10. The growing 3′ terminus of tRNA—presumably with its 3′-terminal nucleotide in an equilibrium between stacked and flipped-out conformations—along with the incoming NTP locks down a conformation of the flexible head domain that produces a pocket size appropriate for the correct incoming nucleotide at a given addition step, or for the final stacked structure required for the termination of the reaction. On the basis of our modelling studies, it seems that the ability of the head domain to reposition may also be required for the addition of the first C in the CCA triplet.
The mechanism by which the class I CCA-adding enzymes are able to specifically add CCA to the end of immature tRNA without a nucleic acid template contains aspects of previous proposals plus numerous unanticipated features. The growing primer terminus is scrunched in the active site as it elongates, and both the tRNA and protein cooperate to provide a template for the incoming base. The enzyme's specificity for adding a C or A, or terminating, however, arises from changes in the size of the nucleotide-binding site, which is altered by the elongating tRNA (Supplementary Movie 2).
Note added in proof: The crystal structure of a class II CCA-adding enzyme complex with tRNA presented in the accompanying paper27 agrees with our modelling of the class II acceptor stem duplex and priming terminal nucleotide orientations shown in Fig. 3.
Crystallization and data collection
The AfCCA molecule containing a his-tag at its C terminus was expressed and purified as described previously16. The tRNAphe from baker's yeast was purchased from Sigma and the synthetic oligonucleotides AC74 and ACC75 were purchased from Dharmacon. AfCCA (5 mg ml-1) and tRNAphe were mixed at a molar ratio of 1:1.5 in a sample buffer containing 20 mM Tris (pH 7.5), 5 mM MgCl2 and 150 mM NaCl. Crystals grew by the hanging-drop vapour diffusion method after six months at 4 °C from a 1:1 mixture of sample (2 µl) with a solution containing 1.5 M (NH4)2SO4, 100 mM Tris (pH 8.5) and 12% glycerol. The crystals diffracted to 6.5 Å resolution at synchrotron beamlines.
The duplex RNA oligonucleotides were mixed with AfCCA (5 mg ml-1) at a molar ratio of 3:1 in the same sample buffer used for the tRNA complex. Crystals of AC74 and ACC75 were obtained after two weeks at room temperature in 70 mM sodium tartrate and 7% PEG 3350. Crystals complexed with NTPs were obtained by soaking the binary crystals in mother liquor containing 4 mM ATP or CTP. To trap the complex in a catalytically active conformation but avoid nucleotide incorporation in the crystal, all soaking experiments were carried out at 4 °C for about 2 h. Furthermore, the reaction was inhibited by the tartrate buffer, which chelated the Mg2+ ions and prevented their binding to the metal position A, as determined by the crystal structures. To observe the binding of metal ion A, crystals of the AC74 ternary complex were soaked in a buffer containing chloride and Mn2+, instead of tartrate and Mg2+, for 30 min at 4 °C. Our attempt to repeat this procedure with the ACC75 complex was unsuccessful due to the limited number of crystals. The ACC75 ternary complex was also obtained by co-crystallization with a non-reactive αβ-methylene ATP. Data for the ternary complex crystals were collected at resolutions from 2.2 Å to 3.4 Å and were processed using the HKL package20. The statistics are summarized in Table 1.
Structure determination and refinement
The structure of AfCCA complexed with tRNA was solved by molecular replacement with the program AMoRe21 using the apo-AfCCA dimer structure as a search model. The program RESOLVE22 was used for density modification and two-fold non-crystallographic symmetry (NCS) averaging. Rigid-body refinement of the model was carried out using the program Refmac23. The ternary complexes were also solved by molecular replacement with AMoRe using the same search model. The crystallographic asymmetric unit contains two AfCCA dimers bound to three duplex RNAs. One of the duplex RNA molecules is shared by two AfCCA proteins, which presumably is a crystallization feature. The electron density maps calculated using model-derived phases were improved by cycles of multi-domain cross crystal averaging using the program Dmmulti24 and real-space NCS averaging within the crystal using the program RAVE25. The final model was rebuilt with the graphics program O26 and refined with Refmac.
We thank the beamline staff at X25 and X12C of the National Synchrotron Light Source, 8BM at the Advanced Photon Source, A1 at the Cornell High Energy Synchrotron Source, and 8.2.1 and 8.2.2 at the Advanced Light Source for assistance in data collection. We thank A. M. Weiner for suggestions on the manuscript; J. Wang for discussions; members of the Steitz laboratory for assistance at various stages of the project; and R. Evans for work with the ACC75 complex. This work was supported by an NIH grant.
About this article
The EMBO Journal (2008)