Expanded use of sense codons is regulated by modified cytidines in tRNA

Abstract

Codon use among the three domains of life is not confined to the universal genetic code. With only 22 tRNA genes in mammalian mitochondria, exceptions from the universal code are necessary for proper translation. A particularly interesting deviation is the decoding of the isoleucine AUA codon as methionine by the one mitochondrial-encoded tRNA^Met. This tRNA decodes AUA and AUG in both the A- and P-sites of the metazoan mitochondrial ribosome. Enrichment of posttranscriptional modifications is a commonly appropriated mechanism for modulating decoding rules, enabling some tRNA functions while restraining others. In this case, a modification of cytidine, 5-formylcytidine (f⁵C), at the wobble position-34 of human mitochondrial () enables expanded decoding of AUA, resulting in a deviation in the genetic code. Visualization of the codon•anticodon interaction by X-ray crystallography revealed that recognition of both A and G at the third position of the codon occurs in the canonical Watson–Crick geometry. A modification-dependent shift in the tautomeric equilibrium toward the rare imino-oxo tautomer of cytidine stabilizes the f⁵C₃₄•A base pair geometry with two hydrogen bonds.

The genetic code was initially deemed to be universal and frozen in time (1). However, deviations in sense and nonsense codon use are found in bacteria, archaea, and both nuclear and organellar eukaryotic genomes (2, 3). Use of genetic codes that deviate from the universal code provides insight into its evolution (4) and possibilities for investigator-initiated manipulation (synthetic biology) (5). In many cases, however, the translation of the deviant sense codons is facilitated by posttranscriptional modification chemistries that are enzymatically added to nucleosides at the first anticodon position. The modification chemistries and their impact on anticodon conformation alter the decoding capacity of the modified tRNA (6). When first proposed, modification-dependent wobble decoding was limited to inosine as the first modified anticodon residue. Inosine, a deaminated adenosine residue, expands the ability of a single isoacceptor tRNA to read three codons by base pairing with either U, C, or A at the third position of the codon (7). Many other wobble position–modified residues, mostly pyrimidines, are now known to modulate use of specific codons (8). Although modified uridines constitute a great majority of these modifications, modified cytidines are prevalent in controlling a switch between the universal genetic code and a deviant code in the shared isoleucine/methionine codon box (9, 10). The modification, 4-acetylcytidine (ac⁴C), prevents from reading AUA through wobble geometry (11), and lysidine (k²C) (12) and agmatidine (agm²C) (13, 14) prevent from reading AUG in bacteria and archaea. Interestingly, a fourth cytidine modification, 5-formylcytidine (f⁵C), facilitates the reading of AUA and AUG as methionine by a single responding to both initiator and elongator codons in yeast and many metazoan mitochondrial genomes (6). Therefore, ac⁴C, k²C and agm²C are restrictive modifications that alter the physicochemical properties of the Watson–Crick edge, whereas f⁵C expands decoding using a modification on the C-H edge at the C5 position. It is evident that the many posttranscriptional modifications of the Watson–Crick edge alter base pairing abilities, but a clear mechanism of decoding expansion by C5 modifications remains poorly understood, especially modifications of cytidine.

The mitochondrion’s decoding of AUA as methionine is important for proper translation. In humans, this codon constitutes 20% of mRNA initiator methionines and 80% of internal methionines (15, 16). Using chemical synthesis and incorporation of the f⁵C₃₄ modification into the heptadecamer anticodon stem and loop domain (ASL) of human mitochondrial (17), we determined that f⁵C₃₄ destabilized the by increasing the motional dynamics of the loop residues (17). A further study detailing the codon-binding characteristics and solution structure of agreed with the f⁵C-dependent increase in residue dynamics (18). Proposed mechanisms for the decoding of AUA by depend on the f⁵C₃₄•A₃ base pair forming a specific geometry (18–20). Based on molecular dynamics simulations, we suggested that this base pair could be in a sheared geometry that is supported by a bridging water molecule (18). To test this hypothesis, the geometry of the f⁵C₃₄•A₃ base pair in the decoding center of the ribosomal A-site was observed in crystal structures of natively modified bound to AUA. Here, we show for the first time the modification-dependent f⁵C₃₄•A₃ base pair within the codon•anticodon interaction during ribosomal A-site decoding. Surprisingly, f⁵C₃₄ forms a canonical Watson–Crick base pair with both the G of AUG and the A of AUA, refuting the conformation predicted from molecular dynamics simulations (18). This geometry of the f⁵C₃₄•A₃ base pair requires a novel amino-imino tautomerism in f⁵C₃₄ similar to the keto-enol tautomerism seen for the two wobble position uridines cmo⁵U₃₄ and mcm⁵s²U₃₄ in Escherichia coli (21) and human (22), respectively.

Results

Codon Bound Structure.

The ASL domain of was chemically synthesized with the wobble modification f⁵C₃₄, the native C₃₃, and a pseudouridine, Ψ₂₇, at the 5′ terminus (Fig. 1A). C₃₃ is quite rare. The uridine at position 33 in tRNAs is considered invariant and recognized for its contribution to the “U-turn” structural motif. There are only 21 known instances of C₃₃, 13 of which are found in initiator tRNAs (23). The bound both AUG and AUA codons with significant affinity in the A-site of the bacterial ribosome (17) and within the A-site of the bovine mitochondrial 55S ribosome (18). In determining the crystallographic structure of the modified bound to the AUG or the AUA codons in the ribosomal A-site, native Thermus thermophilus 30S ribosomal subunit crystals were soaked with the and hexameric oligonucleotides, each containing either AUG or AUA codons. Under the conditions used, both crystals diffracted to 3.3-Å resolution (Table 1). Unbiased difference Fourier electron density maps were generated and used for building the and mRNA residues into the X-ray crystal structures. The structure of the 30S ribosomal subunit (including all RNA and protein) was nearly identical to those reported previously (21, 22). The took a conformation nearly identical to that of the ASL of a ribosome-bound tRNA-EF-Tu complex (24, 25), demonstrating the biological relevance of the present structures. More importantly, the conserved A-site residues G₅₃₀, A₁₄₉₂, and A₁₄₉₃ were in the correct orientation to constrain the codon•anticodon pair residues into the proper geometry for recognition (Fig. 1B) (26, 27). As such, differences in the characteristics of the codon•anticodon interaction can be attributed to the specific conformation of and properties of the f⁵C₃₄ modification.

Fig. 1.

secondary structure and A-site codon•anticodon interaction. (A) The ASL was synthesized as a heptadecamer containing two modifications: pseudouridine at position 27 and 5-formylcytidine at position 34. (B) The structure of bound to AUG (Left) showed strong electron density for ASL residues 31–39, whereas when bound to AUA (Right), electron density was strong for only residues 34–39. In both structures, the ASL is in green, the mRNA codon is in blue, and the A-site interacting residues (G₅₃₀, A₁₄₉₂, and A₁₄₉₃) are in black (2mF_O-dF_C contoured at 1.5 σ). The f⁵C₃₄ modification is colored red.

Table 1.

Data collection and refinement statistics

Dataset	;Ψ27-AUG	;Ψ27-AUA
Data collection
Space group	P41212	P41212
Cell dimensions
a, b, c (Å)	400.6, 400.6, 176.2	402.4, 402.4, 175.6
α, β, γ (°)	90, 90, 90	90, 90, 90
Resolution (Å)	80.1–3.3 (3.38–3.30)	97.6–3.3 (3.38–3.30)
Rmerge	22.5 (133.7)	19.8 (113.3)
I/σI	6.35 (1.19)*	5.50 (1.25)*
CC1/2	0.99 (0.41)	0.99 (0.49)
Completeness (%)	97.4 (95.0)	94.7 (93.5)
Redundancy	3.3 (3.2)	2.6 (2.5)
Refinement
No. reflections	207,954	230,910
Rwork/Rfree	18.8/22.2	19.5/23.6
No. atoms	52,227	52,187
RNA	32,766	32,726
Protein	19,231	19,231
Ion	188	188
Paromomycin	42	42
B-factor	80.1	87.4
r.m.s. deviations
Bond lengths (Å)	0.008	0.008
Bond angles (°)	1.36	1.35

Parentheses indicate highest-resolution shell.

^*I/σI = 2 at 3.48 Å for both structures.

Conformational Characteristics of

Nearly the entire exhibited strong electron density when bound to the cognate AUG codon (Fig. 1B). In the structure of bound to the wobble AUA codon, the codon•anticodon interaction and the 3′ side are clearly resolved. However, there was poor electron density from the 5′-end to the anticodon similar to the poor electron density observed in crystal structures of the mitochondrial (28) and 8-nt loop containing ASLs (29), both in the T. thermophilus ribosomal A-site. Therefore, the overall conformational properties of the are best described from the perspective of the cognate codon-bound structure, whereas conclusions about the AUA-bound structure will be addressed from comparisons of its highly resolved codon•anticodon interaction and 3′-stacked nucleosides. In the canonical U-turn motif of tRNA’s anticodon loop, a sharp backbone curvature occurs between U₃₃ and the wobble nucleoside at position 34 (30). Despite the presence of a C₃₃, the had a U-turn–like architecture. This result confirms NMR (18) and biophysical (17) characterizations. At physiological pH, protonation of U₃₃N3 facilitates hydrogen bond formation between U₃₃N3 and the phosphate of nucleoside 36, stabilizing the characteristic U-turn (30, 31). Although C₃₃N3 is not protonated, the solution structure of by NMR suggested a weak C₃₃•U₃₆ interaction (18). In the crystal structures, the backbone at C₃₃ was found to be distorted in such a way that the position of C₃₃ was shifted from that expected of a U₃₃. A superimposition of our structure with that of the E. coli elongator (32) and the human elongator (22), both of which contain a U₃₃ and a U₃₆ (Fig. 2B), confirmed that the position of C₃₃ was slightly altered from the geometry of a U₃₃. This positioning allows for a hydrogen bond between C₃₃N4 and the phosphate of U₃₆ (Fig. 2A), thus stabilizing the backbone turn and the base pairing of the wobble position f⁵C₃₄.

Fig. 2.

Unusual features of the structure. (A) A hydrogen bond is able to form between C₃₃N4 and the phosphate of U₃₆. ASL carbons are colored green, and mRNA carbons are blue (m2F_O-dF_C contoured at 1.5 σ). (B) (green) is superimposed with E. coli elongator (blue) and human elongator (black), showing a slightly distorted conformation. (C) In E. coli , the cross-loop interaction involves the familiar C₃₂•A₃₈ noncanonical base pair with A₃₇ displaced from the loop, where it cannot participate in base stacking. (D) A cross-loop interaction in involves a unique interaction between C₃₂ and A₃₇ that consists of a base pair between the Watson–Crick edge of C₃₂ and the Hoogsteen edge of A₃₇.

Two other properties of the canonical U-turn were investigated: the positioning of A₃₇ over the first codon•anticodon base pair (22, 32, 33) and the formation of a noncanonical base pair between positions 32 and 38 (32–34). In ASLs responding to codons beginning with A, the conserved A₃₇ is modified and positioned directly above the U₃₆•A₁ base pair (Fig. 1B) (22, 32, 33). In most organisms, the cytoplasmic methionyl-tRNAs contain an N6-threonylcarbamoyladenosine-37 (t⁶A₃₇) (23) stacked with, and thus stabilizing, the U₃₆•A₁ base pair. However, A₃₇ of is unmodified, yet appears to satisfactorily promote a stable U₃₆•A₁ base pair. The geometry of the noncanonical base pair between residues 32 and 38 was also examined because of its importance in wobble decoding (34). Although the E. coli initiator tRNA is closed by a common C₃₂•A₃₈ noncanonical base pair (Fig. 2C), the primary sequence and secondary structure folding of would allow for a C₃₂•C₃₈ cross-loop interaction. Interestingly, A₃₇ and C₃₈ adopt positions within the anticodon loop where C₃₂ is planar with A₃₇ rather than C₃₈. C₃₂ is base paired with A₃₇ (Fig. 2D).

f⁵C₃₄•A₃ Base Pair Adopts a Watson–Crick-Like Geometry.

The f⁵C₃₄•A₃ and f⁵C₃₄•A₃ base pairs are clearly defined in our crystal structures, and both are in Watson–Crick geometry (Fig. 3A). A superposition of the two crystal structures revealed nearly identical codon•anticodon base pair conformations (Fig. 3B). More importantly, this geometry indicates that N4 of the f⁵C₃₄ must be in the imino-oxo form rather than the common amino-oxo form (Fig. 4). If N4 of f⁵C₃₄ was in the universal amino form, the close proximity of f⁵C₃₄N4 and A₃N6 (2.8 Å) would cause a very high electronic and steric repulsion. We hypothesized that the formyl group at the C5 position may reduce the energy difference between the tautomeric forms, thus shifting the equilibrium toward the imino-oxo tautomer. A quantum mechanical ground-state free energy calculation was performed on the amino-oxo and imino-oxo forms of the bases cytosine and 5-formylcytosine. A free energy difference of 6.3 kcal/mol between the two tautomeric forms of cytosine favored the amino-oxo form and confirmed previous estimations between 5.4 and 6.7 kcal/mol (35). The tautomers of 5-formylcytosine differed in free energy by 7.8 kcal/mol. Therefore, the f⁵C-dependent shift in tautomeric equilibrium is not caused by a simple reduction in the energetic properties of the free base.

Fig. 3.

Geometry of the wobble base pair and codon•anticodon interaction. (A) The electron density shows that both the f⁵C₃₄•G₃ and f⁵C₃₄•A₃ base pairs are in Watson–Crick geometry. ASL carbons are colored green, and mRNA carbons are blue (m2F_O-dF_C contoured at 1.5 σ). (B) Superposition of the AUG complex with the AUA structure aligned with respect to the mRNA residues. The overlay reveals a nearly identical orientation of the codon•anticodon interaction between the AUA-bound and AUG-bound structures. The ASL and mRNA residues for the AUA-bound structures are green and blue, respectively. The ASL and mRNA residues of the AUG-bound structure are both gray.

Fig. 4.

Prototropic tautomerism alleviates steric repulsion and allows a Watson–Crick f⁵C₃₄•A₃ base pair. (A) Three possible tautomeric forms of 5-formylcytidine are denoted by the chemical nature of the N4 and O2 positions. (B) In the common amino-oxo form (Upper) of f⁵C₃₄, a Watson–Crick base pair with adenosine results in steric repulsion due to the proximity of f⁵C₃₄N4H and A₃N6H (marked in red); however, the imino-oxo form (Lower) allows for a favorable U•A-like hydrogen bonding.

Discussion

Cytoplasmic translation initiation at the AUG codon is distinct from elongation in which AUG is decoded in the A-site. There are two different cytoplasmic isoacceptors: one for initiating and one for elongating protein synthesis (17). In contrast in the human mitochondrion, a single initiates and elongates at either AUG or the universal isoleucine codon AUA. Eighty percent of mitochondrial mRNAs initiate translation with AUG; whereas 80% of all methionine codons within the mRNA are AUA (15, 16). The f⁵C₃₄-modified decodes the isoleucine AUA codon as methionine in vivo and in vitro (20). The crystal structures presented here offer a structural and chemical rationale for how decodes both AUA and AUG and alters the anticodon domain architecture to act as both an initiator and elongator tRNA.

Three possible f⁵C₃₄•A₃ base pairing strategies have been proposed. The base pair could be in a sheared geometry in which A₃N6 acts as a proton donor in a bifurcated hydrogen bond with f⁵C₃₄N1 and O2. A bridging water molecule acts as a dual proton donor for hydrogen bonds with both f⁵C₃₄O2 and A₃N1 (18). In another possible geometry, a single hydrogen bond is formed between f⁵C₃₄N4 and A₃N1 (20). Finally, it has been proposed that A₃ could become protonated and form an f⁵C₃₄•A₃⁺ base pair. The resulting noncononical pair could be in either of two different geometries: one in which f⁵C₃₄ is shifted such that f⁵C₃₄O2 and N3 act as hydrogen bond acceptors with A₃N1 and N6, respectively (19), and one in which the bases are in the Watson–Crick orientation (20) with f⁵C₃₄N3 bound to A₃N1⁺.

Of the three possible base pairing schemes, only the Watson–Crick geometry with the protonated adenosine on the codon fits with our refined structure. The importance of Watson–Crick base pairing for codon recognition on the ribosome was first proposed by Crick (7), and more recently reiterated with the support of extensive crystallographic data (21, 22). However, a Watson–Crick geometry for an f⁵C₃₄•A₃⁺ base pair should be highly unstable, if not energetically impossible, due to the close proximity of the amine protons of f⁵C₃₄N4 and A₃N6 (Fig. 4B). The most likely scenario for resolving the clear geometric restraints observed in the electron density with the physicochemical properties of the two residues involves a novel shift in the tautomeric equilibrium of modified f⁵C₃₄. Of the three possible prototropic tautomers of cytidine (Fig. 4A), equilibrium must exist between the most common amino-oxo form for base pairing with G₃ and the less common imino-oxo tautomer for interacting with A₃. Although theoretical calculations coupled with empirical photoemission data have identified a rarely populated existence of the imino-oxo tautomer (36), the C5-formyl modification may perturb the electronic structure of the heterocycle, reducing the energy of this tautomer. Prototropic tautomerism is not unprecedented at the wobble position. C5-modified uridines at position 34 in E. coli (21) and human (22) also make use of tautomeric shifts to enable wobble decoding in the ribosomal A-site. Although the different tautomeric forms of uridine have been shown experimentally, this property has not been suggested for cytidine due to significant differences in the electronic structures of the two pyrimidines.

The apparent commonality of using prototropic tautomerization in tRNA’s decoding of bacterial, metazoan, and mitochondrial mRNA compelled us to explore how the C5 position modifications facilitate tautomeric shifts. It has been estimated that common tautomers of nucleobases outnumber their rarer counterparts by ∼10⁴–10⁵ (35), constituting an energy difference between the two states of a mere 5.4–6.7 kcal/mol. Such a relatively small energetic difference resulting in a large shift in the populations of the different tautomeric states indicates that these modifications must be finely tuned for their function in protein synthesis. We hypothesized that the addition of the formyl group at C5 may contribute to the preference for a proton shift from the N4 amine to N3. However, our computational investigation of the effects of the weak electron withdrawing formyl group shows that it increases the energy difference between the tautomers of the free base from 6.3 kcal/mol for cytosine to 7.8 kcal/mol for 5-formylcytosine. However, the tautomeric equilibrium may be affected by interactions of the modification with the environment. This proposition is justified by a study of the interaction between glycine and uridine using density functional theory and showing that the enol form of uridine can be stabilized by up to five orders of magnitude through intermolecular interactions (37). Also, the potential for a pseudobicyclic ring system resulting from a hydrogen bond between the formyl oxygen and the N4 imine proton may contribute energy to the system in the form of improved base stacking interactions and ring current effects.

In addition to providing clear evidence of mechanism by which decodes AUA, the crystal structures provide insight into the ability of this single isoaccepting species to act as both an initiator and an elongator tRNA. Many other characteristics of the tRNA may be responsible for this dual role; however, architectural features of the ASL may also contribute. Because the mitochondrial translation system is more similar to prokaryotic translation (38, 39), we compared with prokaryotic initiator tRNAs. Although many examples of insect, plant, and vertebrate initiator tRNAs contain C₃₃ (40–45), all known prokaryotic initiator tRNAs contain the universal U₃₃. The ASL of the E. coli initiator has a peculiar characteristic in which U₃₃ is positioned outside of the anticodon loop rather than interacting with the phosphate of residue 36 in the canonical U-turn (46). Although this is at odds with a solution structure indicating that U₃₃ is positioned in the canonical U-turn geometry (47), it raises the possibility that the positioning of residue 33 is important for discriminating A-site and P-site binding. In comparison, the E. coli and archaeal would decode AUG was it not for the restrictive recognition of A₃ by the modifications lysidine and agmatidine, respectively (12, 13). The contrast with the is dramatic. The in participating in both A-site and P-site decoding should have the flexibility to adopt different conformations depending on the present function. Indeed, a solution structure of the modified illustrates very little interaction between C₃₃ and U₃₆ (18), and the present crystal structure shows a clear hydrogen bonding between the two residues, albeit in an uncommon geometry. Therefore, our results suggest an induced-fit model that results in the A-site–bound conformation. The resulting conformation favors the canonical U-turn in which C₃₃ interacts with U₃₆ and may also fit into a slightly different conformation for the P-site. Additionally, the unique C₃₂•A₃₇ base pair may further highlight the dual character of this tRNA. In elongator tRNAs, especially where a U₃₆•A₁ base pair is present, A₃₇ is typically modified to promote stacking that stabilizes this first base pair of the codon•anticodon interaction. Here the lack of an A₃₇ modification may allow the to adopt the more “initiator-like” conformation for P-site binding, whereas the unusual C₃₂•A₃₇ base pair promotes architecture more similar to the canonical elongator conformation for A-site codon recognition. The unusual C₃₂•A₃₇ base pairing does not preclude the possibility of a change in the C₃₂ base-pairing partner between A₃₇ and C₃₈ (Fig. 2C). Such a change in base pairing could correspond to an architectural switch of the elongator to that of the initiator function. Thus, we showed that tRNA’s codon recognition is expanded through a modification-dependent f⁵C₃₄ amine-imine tautomerization of cytidine. Overall, the architecture afforded by the natural modifications and sequence change to C₃₃ allow for a significantly altered mode of codon recognition, one in which a single tRNA acts as both an initiator and elongator. The single tRNA^Met contributes to the mitochondrion’s challenge of maintaining a minimal genome.

Materials and Methods

Crystallization.

T. thermophilus 30S ribosomal subunits were purified, crystallized, and cryoprotected in 26% (vol/vol) 2-methyl-2,4-pentanediol (MPD), 100 mM 2-(N-morpholino)ethanesulfonic acid (MES) (pH 6.5), 200 mM KCl, 75 mM NH₄Cl, and 15 mM MgCl₂ (48). Hexameric mRNA oligonucleotides were chemically synthesized and gel-purified using preparative PAGE (Thermo Fisher) with the sequences 5′-AU(A/G)AAA-3′ (codons underlined). After cryoprotection, the empty 30S ribosome crystals were soaked in cryoprotection solution supplemented with 80 µM paromomycin, 300 µM ASL, and 300 µM mRNA (>48 h) (26, 27). Crystals were flash-frozen using liquid nitrogen and stored for data collection. Paromomycin was used for its ability to enhance ASL density and resolution (26) by inducing a closed form of the 30S ribosomal subunit without altering the ASL conformation (27). A recent study has shown that lack of covalent attachment between P- and A-site codons in 30S ribosome structures negates some restraint naturally imposed on the first codon•anticodon base pair (corresponding to U₃₆•A₁ in the current structures), allowing them to adopt a wobble base pairing geometry (49). However, currently known structures of cognate codon recognition in the 30S ribosomal subunit show this base pair in the canonical Watson–Crick geometry (21, 22), thereby imposing the same geometric restraints on the remaining codon residues as that seen in the 70S structures. Indeed, structures of cognate codon recognition on the 30S ribosomal subunit superimpose almost exactly with those of the 70S structures [all atom rmsd of codon•anticodon residues from Protein Data Bank entries 1IBL (30S) (26) and 2J00 (70S) (50) = 0.345 Å].

Data Collection and Refinement.

Data were collected at The Northeastern Collaborative Access Team (NE-CAT) beamlines 24-ID-C and 24-ID-E of the Advanced Photon Source. Processing was performed using XDS (51), PHENIX (52) for format manipulation and refinement, Coot (53) for visualization and model building, and PyMOL (54) for figure production. Geometry restraints and dictionary files for the nonstandard residues (paromomycin and 5-formylcytidine) were generated with the eLBOW (55) module within PHENIX using the semiempirical quantum mechanical AM1 method. The two different structures were generated from data sets that were collected at different times; therefore, resolution differences in the structures were likely caused by differences in the X-ray beam rather than in the crystals themselves. The CC1/2 values are calculated using the newest version of XDS (51).

Relative Ground-State Energy Calculations.

Structures corresponding to the amino-oxo tautomers of cytidine and 5-formylcytidine base were prepared using GaussView03 (56) by replacing the ribose moiety with a methyl group. All subsequent calculations were performed using Gaussian-03 (57). Additionally, the imino-oxo tautomers of each were built with the imino proton in the trans-orientation (facing away from the Watson–Crick face) to account for the geometry that would allow for base pairing without steric repulsion. The ground-state geometries were first optimized at the semiempirical level using the AM1 method. Further geometry optimization was achieved at the HF/6–31G(d,p) level. Single point energy calculations were then performed at the B3LYP/6–311++G(d,p) level. All HF and B3LYP calculations were performed with a Polarizable Continuum Model using the integral equation formalism variant (IEFPCM) to simulate water solvation (58).