Importance of a tRNA anticodon loop modification and a conserved, noncanonical anticodon stem pairing in tRNACGGPro for decoding

Modification of anticodon nucleotides allows tRNAs to decode multiple codons, expanding the genetic code. Additionally, modifications located in the anticodon loop, but outside the anticodon itself, stabilize tRNA–codon interactions, increasing decoding fidelity. Anticodon loop nucleotide 37 is 3′ to the anticodon and, in tRNACGGPro, is methylated at the N1 position in its nucleobase (m1G37). The m1G37 modification in tRNACGGPro stabilizes its interaction with the codon and maintains the mRNA frame. However, it is unclear how m1G37 affects binding at the decoding center to both cognate and +1 slippery codons. Here, we show that the tRNACGGPro m1G37 modification is important for the association step during binding to a cognate CCG codon. In contrast, m1G37 prevented association with a slippery CCC-U or +1 codon. Similar analyses of frameshift suppressor tRNASufA6, a tRNACGGPro derivative containing an extra nucleotide in its anticodon loop that undergoes +1 frameshifting, reveal that m1G37 destabilizes interactions with both the cognate CCG and slippery codons. One reason for this destabilization is the disruption of a conserved U32·A38 nucleotide pairing in the anticodon stem through insertion of G37.5. Restoring the tRNASufA6 U32·A37.5 pairing results in a high-affinity association on the slippery CCC-U codon. Further, an X-ray crystal structure of the 70S ribosome bound to tRNASufA6 U32·A37.5 at 3.6 Å resolution shows a reordering of the anticodon loop consistent with the findings from the high-affinity measurements. Our results reveal how the tRNA modification at nucleotide 37 stabilizes interactions with the mRNA codon to preserve the mRNA frame.

Protein synthesis is performed by the ribosome, a conserved protein-RNA macromolecular machine where mRNA, tRNAs, and translation factors read the genetic information as presented on mRNA into proteins. There are four defined stages of protein synthesis: initiation, elongation, termination, and recycling (reviewed in Ref. 1). During elongation, three nucleotides of the mRNA codon are read (or decoded) by three anticodon nucleotides of a tRNA in the ribosomal aminoacyl site (A site) 2 on the small 30S subunit. The three-nucleotide code on the mRNA defines a single amino acid delivered by the corresponding tRNA. The regulation of the mRNA frame is critically important to maintain the correct sequential addition of amino acids to the nascent chain (2). Despite the importance of accurate protein expression for cell viability, the molecular basis for how the ribosome maintains this three-nucleotide mRNA frame is not well-understood.
Because tRNAs decode mRNAs, these RNA molecules probably play a role in mRNA frame maintenance. tRNAs are ϳ76 -90 nucleotides in length and adopt an L-shaped tertiary structure allowing them to fit into ribosome-binding sites that span both subunits (Fig. 1). tRNAs undergo extensive posttranscriptional modifications important for the correct tertiary fold of the tRNA, including the conformation of the anticodon stem-loop (ASL) (3). RNA modifications that are located in the anticodon and neighboring nucleotides in the ASL contribute to the accuracy and speed of translation (4,5) by stabilizing the interactions between the anticodon and codon (3, 6 -8). After decoding, these tRNA modifications are also important during translocation of the mRNA-tRNA pairs (9) and have also been implicated in mRNA frame maintenance (6).
The selection of the correct tRNA for each mRNA codon relies on the formation of Watson-Crick base pairs between the first two nucleotides of the codon and nucleotides 36 and 35 of the anticodon (Fig. 1A). The interaction between the third nucleotide of the codon and anticodon nucleotide 34 is not required to be Watson-Crick. Instead, a G⅐U wobble pair or a modified anticodon nucleotide 34 -codon nucleotide pair can form. The modification of nucleotide 34 enables non-Watson-Crick interactions with the third nucleotide position of the mRNA that is accepted as cognate by the ribosome. The increased flexibility in codons that each tRNA can decode allows for the degeneracy of the genetic code where the 61 codons are decoded by fewer tRNAs (10). Therefore, tRNA modifications at nucleotide 34 have an important and essential role in the process of decoding.
Nucleotide 34 contains many diverse modifications that are typically required for accurate translation (11). Two examples include the uridine-5-oxyacetic acid (cmo 5 U34) in tRNA CGU Ala and the 5-methylaminomethyl-2-thiouridine (mnm 5 s 2 ) U34 modification in tRNA UUU Lys . The cmo 5 U34 modification in tRNA CGU Ala stabilizes its interaction with C6 at the wobble position (12). The mnm 5 s 2 U34 modification in tRNA UUU Lys allows for pairing with AAA or AAG codons (13,14). However, a 6-threonylcarbamoyladenosine (t 6 A) or a 2-methylthio derivative (ms 2 t 6 A) at nucleotide 37 is required for recognition of the AAG codon (4,7). Although all of these codon-anticodon pairings should be recognized by the ribosome, the instability of the anticodon loop of tRNA UUU Lys and, thus, its interactions with the AAA codon require both modifications at nucleotides 34 and 37. Nucleotide 37 of the tRNA is located 3Ј to the anticodon, adjacent to the first position of the Watson-Crick base pair between the codon and anticodon nucleotide 36 (Fig. 1A). The codon-anticodon pairing between tRNA UUU Lys and its codon is weak in the absence of the modifications because of the three A-U base pairs and poor stacking of the UUU anticodon (15). Both the t 6 A and ms 2 t 6 A modifications contain planar heterocycle moieties that promote cross-strand stacking interactions between A38 of the tRNA and the first base in the mRNA codon to stabilize the codon-anticodon pairing (16). Because decoding relies on both the high-affinity binding of cognate tRNAs to the decoding center and conformational changes of the 30S known as domain closure (17)(18)(19), these modifications influence both aspects of decoding.
Nucleotide 37 is modified in Ͼ70% of all tRNAs and is typically a purine (20). Among the modified nucleosides, the t 6 A and methylated guanosine (m 1 G) are the most common (11). In the absence of the modification at nucleotide 37, the anticodon loops of human tRNA Lys and yeast tRNA Asp lack structural rigidity compared with their modified forms (21,22). Likewise, the modification at nucleotide 37 in Escherichia coli tRNA Lys and tRNA Phe stabilizes the canonical uridine turn (U turn) in the ASL, which is required for high-affinity binding to the A site (23)(24)(25).
Approximately 75% of bacterial tRNAs containing the m 1 G37 modification decode CNN codons (where N indicates any nucleotide), including Leu, Pro, His, Gln, and Arg codons (11). The m 1 G37 modification is present in ϳ95% of all known sequences of proline tRNAs (26). In bacteria, the essential methyltransferase TrmD (Trm5 in eukaryotes and archaea) catalyzes the N1-methylation of G37 in tRNAs. Furthermore, the m 1 G37 modification stabilizes the anticodon of the tRNA to prevent ϩ1 ribosomal frameshift errors, whereas mutations in trmD also cause growth defects (6, 26 -28). tRNA CGG Pro , the major isoacceptor for proline, decodes the CCG codon and, in the absence of the m 1 G37 modification, causes ϩ1 frameshifting on CCC-N codons. It was previously thought that tRNA CGG Pro lacking the m 1 G37 modification would allow for a four-nucleotide interaction between the anticodon and the mRNA codon with G37 interacting with the mRNA codon (29). However, biochemical and structural studies of ASL CGG Pro lacking the modification revealed that this four-nucleotide interaction does not occur in the A site during decoding (30,31). Additionally, the mRNA is positioned in the unshifted or zero frame, indicating that the frameshift event occurs post-decoding, consistent with recent structures (32). Interestingly, the absence of the methylation at G37 causes a distortion of the tRNA on the opposite side of the anticodon loop at nucleotide U32 (31), leading to the disruption of interactions with A38. Collectively, these results suggest a previously unappreciated role of the stabilization of the 32⅐38 pairing in tRNA CGG Pro predicted to maintain the correct mRNA frame (31).
Frameshift suppressor tRNAs derived from tRNA Pro contain an insertion between anticodon loop nucleotides 37 and 38 (referred to as 37.5) and decode CCC-N codons as proline (26,(33)(34)(35)(36) (Fig. 1). These mutant tRNAs are genetic suppressors that perform noncanonical reading of the genetic code to restore the reading frame (37,38). In this case, frameshift suppressor tRNA SufA6 , isolated from Salmonella enterica serovar Typhimurium, contains an eight-nucleotide anticodon loop by the addition of G37.5 that causes ϩ1 frameshifting. The structure of 70S-tRNA SufA6 bound to CCC-A/U/C codons at the decoding center that undergo ϩ1 frameshifting revealed similarities to the structure of 70S-tRNA CGG Pro lacking the m 1 G37 bound to a near-cognate codon that also promotes ϩ1 frameshifting (31). Both tRNAs decode the mRNA in the unshifted or zero frame, indicating that the shift into the new frame occurred post-decoding. Moreover, the inserted 37.5 nucleotide and the lack of m 1 G37 both cause destabilization of nucleotides on the opposite side of the anticodon loop that ablates a conserved, non-Watson-Crick U32⅐A38 pairing. The 32⅐38 pairing was restored in both the tRNA Pro and tRNA SufA6 in the context of recognizing a cognate, three-nucleotide codon. The disruption of the 32⅐38 pairing is particularly notable due to its universal significance in tuning the ribosomal binding across tRNAs (39). These results provide insight into how tRNA modifications and the 32⅐38 pairing in the anticodon loop together lead to mRNA frame maintenance.
Here, we tested how the m 1 G37 modification in tRNA CGG Pro and tRNA SufA6 impacts binding at the decoding center to cognate and slippery ϩ1 codons. Further, we engineer tRNA SufA6 Figure 1. Frameshift suppressor tRNA SufA6 is a derivative of tRNA CGG Pro . A, tertiary structure of tRNA Pro with its anticodon depicted in cyan, anticodon loop nucleotide 37 in orange, and the mRNA in gray with the first P-site nucleotide denoted as ϩ1; P-site codon nucleotides listed as ϩ1, ϩ2, and ϩ3; and A-site codon nucleotides listed as ϩ4, ϩ5, and ϩ6. B, secondary structure of the anticodon stem loop of tRNA Pro bound to a cognate CCG codon containing three Watson-Crick base pairs. C, secondary structure of the anticodon stem loop of tRNA SufA6 bound to a four-nucleotide slippery CCC-U codon and containing a C6-C34 wobble interaction. tRNA SufA6 contains an extra guanosine between positions 37 and 38 (G37.5, blue).

ASL elements that ensure accurate tRNA CGG Pro decoding
to contain a conserved U32⅐A38 pairing to attempt to restore high-affinity binding to the decoding center. Last, a 3.6 Å X-ray crystal structure of tRNA SufA6 containing this engineered 32⅐38 pairing bound to the 70S ribosomal A site reveals a reordering of the 32⅐38 pair required for decoding.

The m 1 G37 modification in tRNA CGG Pro stabilizes binding to the A site
To assess the importance of the m 1 G37 modification in tRNA CGG Pro in decoding, we used established filter binding assays to determine binding kinetics to the A site (39). E. coli 70S ribosomes were programmed with mRNA containing a peptidyl-site (P-site) AUG start codon, an A-site proline CCG codon, and P-site E. coli tRNA fMet . We used a chemically synthesized ASL containing 18 nucleotides of tRNA CGG Pro and a m 1 G37 modification to ensure that the RNA was completely modified (Table S1). ASL CGG Pro with the m 1 G37 modification binds to a cognate CCG codon in the A site with an equilibrium dissociation constant (K d ) of 284 nM ( Fig. 2A and Table S2). This affinity is within the range of reported values for ASLs binding to the A site (33-500 nM) (9,40). Removal of the m 1 G37 modification (⌬m 1 G37) significantly reduced binding, by ϳ6.5-fold (1.8 M; Fig. 2A and Table S2).
Although the data could be fit with reasonable confidence-( Fig. S1 and Table S2), the low maximum binding was concern-  ing, if not unprecedented (39 -41). Furthermore, impractical 70S concentrations required to reach maximum binding for weaker interactions prevented us from attempting to continue with this approach. Therefore, we instead performed competition binding assays, which allow for the calculation of the equilibrium dissociation constant (K d ) based on measured association (k on ) and dissociation (k off ) rates ( Fig. 2B) (42,43). Unlike measuring the experimental K d to observe equilibrium binding, the k on and k off measurements provide information regarding the influence of the m 1 G37 modification at each step of the binding event separately (i.e. the association with and dissociation from the ribosomal A site). Using this approach, we found a modest difference in k off between ASL CGG Pro and ASL CGG Pro ⌬m 1 G37, implying that the modification does not stabilize tRNA binding to ribosomes. Instead, the m 1 G37 modification is important for initial binding, as shown by the 2.6-fold slower association rate of ASL CGG Pro ⌬m 1 G37 (k on ϭ 0.012 M Ϫ1 min Ϫ1 ) as compared with ASL CGG Pro (k on ϭ 0.031 M Ϫ1 min Ϫ1 ) to the CCG A-site codon ( Fig. 2 (C and D) and Table 1). These data are consistent with 70S structures demonstrating conformational distortion of the anticodon loop in the absence of the m 1 G37 modification (31). Furthermore, we found that the calculated dissociation constant (K d ) of ASL CGG Pro m 1 G37 with the cognate CCG codon is 420 nM (Table 1), whereas ASL CGG Pro ⌬m 1 G37 binds to the A-site CCG codon with a calculated K d of 1.4 M, consistent with directly measured K d values ( Fig. 2A and Table S2).

tRNA CGG
Pro lacking the m 1 G37 modification undergoes high levels of ϩ1 frameshifting on CCC-N codons (26,28). Previous 70S structures of tRNA CGG Pro ⌬m 1 G37 bound to A-site CCC-N codons revealed that the tRNA decodes in the unshifted or zero frame (31). The three tRNA anticodon nucleotides C34-G35-G36 form three interactions with the C4-C5-C6 mRNA codon, respectively (where the first nucleotide of the P-site mRNA codon is denoted as ϩ1 and the A-site nucleotides are ϩ4, ϩ5, and ϩ6) (Fig. 1A). The interaction between the anticodon and the codon is near cognate as defined by a single mismatch between C34 and C6 (Fig. 3A). To test the impact of the m 1 G37 modification on the ability of tRNA CGG Pro to form a stable complex with the slippery CCC-U codon, we again measured binding kinetics. In the context of ASL CGG Pro binding to a slippery codon, the m 1 G37 modification influences binding but in the opposite manner to binding to the cognate CCG codon (Fig. 2). ASL CGG Pro lacking the m 1 G37 modification associates ϳ4-fold faster to the slippery codon (k on ϭ 0.017 versus 0.0039 M Ϫ1 min Ϫ1 ); however, lack of the modification only has a moderate impact on k off (0.009 versus 0.014 min Ϫ1 ) ( Fig. 3 (B and C) and Table 1). Calculated K d measurements of 3.62 and 0.41 M for ASL CGG Pro and ASL CGG Pro ⌬m 1 G37, respectively, indicate a 9-fold difference in binding affinity. In summary, the stabilizing effect observed in the cognate CCG context is reversed on a slippery CCC-U codon; the presence of the m 1 G37 modification actually impairs the association of ASL CGG Pro to the A site programmed with a slippery CCC-U codon.

Nucleotide insertion in the ASL of tRNA SufA6 counteracts the stabilization exerted by m 1 G37
Frameshift suppressor tRNA SufA6 undergoes ϩ1 frameshifting similar to tRNA CGG Pro lacking the m 1 G37 modification (26,31). Both tRNA SufA6 and tRNA CGG Pro contain the m 1 G37 modification, and, in the case of tRNA SufA6 , this modification is located adjacent to the inserted nucleotide (30) (Fig. 4A). We next tested the importance of the m 1 G37 modification in the context of an eight-nucleotide anticodon loop in ASL SufA6 , using the same kinetic binding assays as previously described. In contrast to the stabilizing effect observed with the m 1 G37 modification in ASL CGG Pro on a cognate CCG codon, ASL SufA6 associates with the CCG codon 10-fold faster in the absence of the modification (0.053 versus 0.0053 M Ϫ1 min Ϫ1 ) ( Fig. 4B and Table 1). In contrast, the dissociation of ASL SufA6 was essentially unaffected by the absence or presence of the modification (k off ϭ 0.013 and 0.014 min Ϫ1 ; Fig. 4C). The 10-fold difference in the calculated K d between ASL SufA6 containing the m 1 G37 modification (2.5 M) and lacking the m 1 G37 modification (0.26 M) is thus reflective of the large changes in the tRNA association with the A site. One interpretation of these observations could be that, although the m 1 G37 modification imparts a stabilizing effect in anticodon loops of the canonical seven nucleotides (10,44), increasing the anticodon loop to eight nucleotides, as seen in ASL SufA6 , ablates any stabilization from the modification. Additionally, the overall trends of the ASL SufA6 association rates are similar to the rates seen with ASL CGG Pro on the slippery CCC-U codon.

ASL elements that ensure accurate tRNA CGG Pro decoding
Next, we tested the binding of ASL SufA6 to the slippery CCC-U codon. We found that the influence of the modification status of ASL SufA6 follows similar trends regardless of whether ASL SufA6 is recognizing a cognate CCG or a slippery CCC-U codon ( Table 1). The association rate of ASL SufA6 for a slippery CCC-U codon is 0.0041 M Ϫ1 min Ϫ1 as opposed to 0.0053 M Ϫ1 min Ϫ1 for binding to the CCG codon in the presence of m 1 G37 in the ASL. In the absence of the m 1 G37 modification, ASL SufA6 has a 10-fold greater association rate for both the CCG and slippery CCC-U codons (0.053 and 0.062 M Ϫ1 min Ϫ1 , respectively). The dissociation of ASL SufA6 from a cognate CCG or a slippery CCC-U codon are all very similar regardless of the G37 modification status (0.013-0.019 min Ϫ1 ). Together, these data indicate that the inserted G37.5 nucleotide in ASL SufA6 removes the dependence on the m 1 G37 modification required for tight association to the ribosome for the parent tRNA CGG Pro . Further, the G37.5 nucleotide also prevents the ribosome from distinguishing between cognate and near-cognate slippery codons, as evidenced by the similar calculated K d values in the absence of the m 1 G37 modification (0.26 and 0.31 M, respectively).

Engineering of the U32⅐A37.5 pairing in ASL SufA6 allows for tight association with the A site
Our affinity assays show that frameshift suppressor ASL SufA6 is unable to bind with high affinity to a cognate CCG codon despite containing the same GGC anticodon as tRNA CGG Pro (Fig. 1). Therefore, the m 1 G37 modification in ASL SufA6 has a very different role in stabilizing the interactions between the anticodon and codon in contrast to ASL CGG Pro . Although tRNA-SufA6 undergoes ϩ1 frameshifting, it does so with low efficiencies because of its poor association with the slippery CCC-U codon (Fig. 3B) (30). These data lead us to question whether it is the inserted anticodon loop nucleotide alone that causes reduced binding affinity as seen with other frameshift suppressor ASL-binding studies (40). In the case of both ASL CGG Pro and ASL SufA6 that bind poorly to the A site, 70S structures of these same tRNA-mRNA pairs bound have been solved (Fig. 5, A-C) (31). In the case of ASL CGG Pro ⌬m 1 G37 bound to a cognate CCG codon, electron density is missing for nucleotide U32, which is located on the opposite side of the anticodon loop from A38 (Fig. 5B). The destabilization of the ASL is likely due to the apparent flexibility of the 5Ј stem, which, in turn, disrupts the conserved U32⅐A38 interaction located at the base of the RNA stem. The 32⅐38 disruption is noteworthy because the identity of these nucleotides is universally important in fine-tuning tRNA affinity and therefore translation fidelity (39, 45). The same structural phenomenon is also observed in 70S structures containing ASL SufA6 ; ASL SufA6 binding to a slippery CCC-U codon results in local distortion of the 5Ј stem disrupting the U32⅐A38 pairing (Fig. 5C). In both cases, the tRNA-mRNA pair undergoes ϩ1 frameshifting. Therefore, we postulated that the frameshift event was directly influenced by the destabilization of the 32⅐38 pairing after tRNA selection but before movement to the P site.
The G37.5 insertion in tRNA SufA6 changes the potential base pairing interaction of U32⅐A38 to U32⅐G37.5 (Fig. 5C). In this context, the U32⅐G37.5 pairing should render the ribosome unable to distinguish a cognate from noncognate interaction as the 32-38 nucleotide identity is directly correlated to the anticodon sequence (39, 45). Indeed, ASL SufA6 binds to cognate CCG and near-cognate (i.e. slippery) CCC-U codon with similar affinities (calculated K d of 0.31 and 0.26 M, respectively, in the absence of the m 1 G37 modification; Table 1). We next tested whether changing G37.5 to A37.5 could restore highaffinity A-site binding due to the possible formation of a new U32⅐A37.5 pair. We found that potentially restoring the U32⅐A37.5 base pair does not result in high-affinity binding to a cognate CCG codon in the absence or presence of the m 1 G37 modification (calculated K d of 7.4 and 15 M, respectively; Fig. 5 (D and E) and Table 1). Notably, in contrast to ASL CGG Pro , ASL SufA6 A37.5 displays similar association rates both in the presence (0.0057 M Ϫ1 min Ϫ1 ) and absence (0.0046 M Ϫ1 min Ϫ1 ) of m 1 G37, but k off is reduced ϳ3-fold (0.09 and 0.034 min Ϫ1 , respectively) (Fig. 5, D and E).
In binding to the slippery CCC-U codon, ASL SufA6 A37.5 has a ϳ2-fold higher affinity (calculated K d ϭ 1.8 M) than ASL SufA6 containing G37.5 (calculated K d ϭ 3.7 M) in the presence of the m 1 G37 modification ( Table 1). Removal of m 1 G37 results in ASL SufA6 A37.5 binding with high affinity, similar to ASL SufA6 G37.5 (calculated K d ϭ 0.31 M for ASL SufA6 G37.5 and 0.45 M for ASL SufA6 A37.5). The K d for ASL SufA6 A37.5 ⌬m 1 G37 binding to a slippery CCC-U codon is comparable with modified ASL CGG Pro binding to a cognate CCG codon (calculated K d ϭ 0.42 M). For the ASL SufA6 A37.5 ASL elements that ensure accurate tRNA CGG Pro decoding ⌬m 1 G37 binding to a slippery CCC-U codon, both the k on and k off rates are higher than those of other ASLs (k on ϭ 0.146 M Ϫ1 min Ϫ1 , k off ϭ 0.06 min Ϫ1 ), implying that the recognition mechanism is altered. The increase in affinity implies that in the case of ASL SufA6 with the restored U32-A37.5, the lack of m 1 G37 enables high-affinity binding and recognition of the ASL when the slippery CCC-U codon is presented in the A site. Most importantly, we demonstrate that by changing the identity of the base insertion and controlling the modification at position 37 in the anticodon stem loop, we can tune the affinity of the ASL to the ribosomal A site. This has significant implications for understanding how the ribosome interacts with rationally engineered tRNAs.

Engineering the 32⅐38 pairing in ASL SufA6 to U32⅐A37.5 reorders the 5 stem of the anticodon loop
To determine whether the engineered ASL SufA6 A37.5 ⌬m 1 G37 does indeed reorder the ASL as suggested from the binding kinetics, we solved a 3.6 Å resolution X-ray crystal structure of ASL SufA6 A37.5 ⌬m 1 G37 bound to the Thermus thermophilus 70S ribosome (Fig. 5F and Table 2). The ASL and mRNA density are well-ordered and unambiguously demon-strate a change in the anticodon loop (Fig. 5F) as compared with other ASL SufA6 structures bound to the ribosome (Fig. 5, A-C) (31). The ASL SufA6 A37.5 ⌬m 1 G37 has good density for U32 in contrast to the previous structures that showed distortion of the 5Ј stem of the ASL (Fig. 5, B and C). The phosphate backbone of nucleotide A37.5 shifts by 2.8 Å as compared with WT ASL CGG Pro bound to its cognate CCG codon and by 6.2 Å when compared with ASL SufA6 with the G37.5 (Fig. 5F, S3A). This movement places A37.5 across from U32, allowing the possible formation of a single hydrogen bond similar to the 32⅐38 orientation observed in other tRNAs (Fig. 5A, S2) (31,46,47). Overall, the A37.5 insertion in ASL SufA6 seems to orient the ASL to a conformation more similar to that of ASL CGG Pro than that of WT ASL SufA6 (Fig. S3, B and C). 16S rRNA nucleotides A1492 and A1493 flip from their internal position in helix 44, and G530 is positioned close to A1492, demonstrating recognition by the ribosome (Fig. S4).

Discussion
Modification of tRNAs adds an important layer of regulation during translation. These modifications are so functionally important that more genes are devoted to tRNA modification  Fig. 1); B, 70S ribosome with ASL CGG Pro ⌬m 1 G37 decoding a cognate CCG codon (PDB code 4P70); C, 70S ribosome with ASL SufA6 decoding a slippery CCC-U codon (PDB code 4L47). These structures demonstrate that the m 1 G37 modification stabilizes the U32⅐A38 interaction in ASL CGG Pro on a cognate CCG codon (A) whereas the lack of the m 1 G37 modification results in disorder of the 3Ј region of ASL CGG Pro (B). A similar disordering is seen when ASL SufA6 containing an inserted nucleotide in its anticodon loop (G37.5) decodes a slippery CCC-U codon (C). D and E, k on (D) and k off (E) rates of ASL SufA6 with a mutated A37.5 with (ϩm 1 G37) or without (⌬m 1 G37) the modification bound to either the cognate CCG (black) or slippery CCC-U (green) codon. Data are the mean Ϯ S.E. (error bars) of at least five independent experiments. F, 2F o Ϫ F c electron density maps from a structure containing the 70S ribosome with ASL SufA6 A37.5 bound to an A-site slippery CCC-U codon. Mutation of G37.5 to A37.5 reorders the 3Ј stem of the ASL, specifically nucleotides 31 and 32. 2F o Ϫ F c electron density maps are contoured at 1.5.

ASL elements that ensure accurate tRNA CGG Pro decoding
pathways than to the expression of tRNAs themselves (48).
Modifications in the ASLs of tRNAs are critical, given that only 7 of the 61 sense codons are decoded by tRNAs that lack modifications at nucleotide 34 or 37 in E. coli (49,50). In this work, we determine that the m 1 G37 modification in ASL CGG Pro is required for high-affinity binding to a cognate CCG codon in the decoding center (Fig. 2). The absence of the modification results in low-affinity binding, and specifically, the association (k on ) is reduced almost 3-fold, whereas k off is unaffected (Fig. 2). These results indicate that the m 1 G37 modification in tRNA CGG Pro provides stability in association with the decoding center rather than causing A-site drop-off. Consistent with these data are our previous structural studies that showed destabilization of the anticodon loop when ASL CGG Pro lacks the m 1 G37 modification and interacts with a cognate CCG codon (31) (Fig. 5B).
Both tRNA CGG Pro and tRNA GGG Pro isoacceptors lacking the m 1 G37 modification undergo ϩ1 frameshifting on CCC-N codons (26,29,51). Although ASL CGG Pro containing the m 1 G37 modification significantly impairs binding to a slippery CCC-U codon (3.62 M; Fig. 3A), its removal causes a ϳ4-fold enhancement in ASL CGG Pro association with the slippery codon (Fig. 3B). This association results in tighter binding (calculated K d of 0.41 M) that is comparable with binding of WT ASL CGG Pro to a cognate CCG codon (0.42 M). These data suggest that the additional stability that the m 1 G37 modification imparts in binding to a cognate CCG codon is lost in the context of a non-Watson-Crick C34-C4 pair at the third or wobble position. Collectively, these results show that the m 1 G37 modification in ASL CGG Pro stabilizes high-affinity interactions in the cognate case but prevents recognition of slippery codons that would result in ϩ1 frameshifting.
tRNAs containing expanded anticodon stem loops can cause "slipping" on mRNA codons, resulting in frameshifts (37). An extra nucleotide insertion in the anticodon loop of tRNA CGG Pro was identified in a frameshift suppressor tRNA (named tRNA SufA6 ) that reverts a ϩ1 frameshift. Primer extension anal-yses revealed that tRNA SufA6 was also modified at nucleotide 37, similar to all three tRNA Pro isoacceptors (30,52), but the extent of modification was not determined. It is unclear what role, if any, the m 1 G37 modification has in tRNA SufA6 -mediated ϩ1 frameshifting. We find that the presence of the m 1 G37 modification renders ASL SufA6 unable to bind to both cognate CCG and slippery CCC-U codons (Fig. 4B). In contrast, ASL SufA6 lacking m 1 G37 binds with high affinity to either a cognate CCG or a slippery CCC-U codon. These data support the notion that the m 1 G37 modification and the inserted G37.5 nucleotide likely stabilize the anticodon loop in similar ways. In support of the functional similarities of m 1 G37 and G37.5, structures of tRNA CGG Pro ⌬m 1 G37 or tRNA SufA6 decoding codons that allow for ϩ1 frameshifting reveal structural similarities. The 3Ј stem of the ASL, in particular nucleotides 30 -32 on the opposite side of the G37.5/G37, is conformationally dynamic in both structures, strongly suggesting that ϩ1 frameshifts induced by these two tRNAs occur by a similar mechanism (Fig. 5, B and C) (31).
In tRNA CGG Pro , nucleotide U32 normally forms a single hydrogen bond with A38 and thus is not a Watson-Crick base pair ( Fig. 5A and Fig. S2). The nucleotide identity of the 32⅐38 pairing in all tRNAs is inversely correlated to the strength of the codon-anticodon interaction (39, 41,45). For example, the anticodon of E. coli tRNA GGC Ala is considered strong because of the three GC pairs between the codon and the anticodon. Therefore, in this strong case, the 32⅐38 pairing needs to be correspondingly weak to counterbalance the strength of the codon-anticodon. Changing the 32⅐38 pairing in tRNA GGC Ala from a weak, conserved U32⅐A38 pair to a strong pair, such as C32⅐A38, prevents the ribosome from being able to distinguishing correct from incorrect tRNA-mRNA pairs (41,53). In the context of tRNA SufA6 , the inserted G37.5 displaces A38, preventing a U32⅐A38 pairing (31). Binding of ASL SufA6 to a cognate CCG or a ϩ1 slippery CCC-U codon is extremely weak as indicated by both the k on and k off rates (Fig. 4). We attempted to restore the WT U32⅐A38 found in tRNA CGG Pro by changing G37.5 in tRNA SufA6 to an adenosine. ASL SufA6 A37.5 binds poorly to a CCG codon regardless of the m 1 G37 modification status (Fig.  5D). Interestingly, the A37.5 mutant bound tightly to the slippery CCC-U codon, but only in the absence of the m 1 G37 modification, similar to ASL SufA6 . An X-ray crystal structure of the ribosome with an A-site ASL SufA6 A37.5 (lacking the m 1 G37 modification) bound to a slippery CCC-U codon reveals a reordering of the 3Ј stem such that U32 regains rigidity as assessed by its electron density (Fig. 5F). We predict that this engineered tRNA SufA6 does not undergo ϩ1 frameshifting because of ordering of the ASL, but further studies are required to test this.
The studies here demonstrate that the m 1 G37 modification of tRNA Pro influences recognition of both cognate and nearcognate, slippery codons. tRNA CGG Pro lacking the m 1 G37 modification undergoes ϩ1 frameshifting, but our previous structures, along with other structures of extended ASLs that frameshift, show that the shift into the new frame does not occur in the decoding center (31,54,55). At what stage of elongation does tRNA CGG Pro lacking the m 1 G37 modification cause a ϩ1 frameshift? Kinetic analyses of tRNA GGG Pro movement through the ribosome reveal that the shift can occur at two distinct stages: a fast mechanism during translocation of the tRNA-mRNA pairs to the P site and a slower mechanism when tRNA GGG Pro is stalled in the P site while waiting for A-site tRNA delivery (28). A recent structure of ASL SufA6 bound to a ϩ1 codon in the P site demonstrates the ASL alone is sufficient for the ϩ1 mRNA frameshift consistent with the slow mechanism presented above (32). Another possibility is what is also observed in the structure of ASL CGG Pro lacking the m 1 G37 modification in the A site (31): the 3Ј anticodon stem is destabilized, which, in turn, may influence how elongation factor G (EF-G) recognizes the tRNA in the A site to initiate translocation. This second possibility is consistent with the fast mechanism as observed in the kinetic analyses.
Although the data presented focus on the impact of the m 1 G37 modification on decoding of single proline codons, polyproline sequences in protein-coding genes are prone to ϩ1 frameshifts (56 -60). The unique nature of proline where it is both a poor donor and acceptor during the peptidyltransferase reaction results in a slow rate of peptide bond formation. Therefore, consecutive prolines cause ribosome stalling (60,61). Additionally, the nucleotide repeats in the proline codons present the same codon-anticodon interactions regardless of whether the tRNA binds in the zero or ϩ1 frame (62). These events can collectively lead to the shifty nature of tRNA CGG Pro but are counterbalanced by the action of elongation factor P (EF-P) that helps to stabilize peptidyl-tRNA Pro located in the P site of the ribosome (28,60,61,63). EF-P binds in the exit site of the ribosome on both the small and large ribosomal subunits and abuts against P-site peptidyl-tRNA Pro (63). A modified Lys residue of EF-P protrudes into the 50S P site and stabilizes peptidyl-tRNA Pro to help resume protein synthesis stalled at a stretch of polyprolines (60,64,65). Cryo-EM structures of ribosomes bound to peptidyl-tRNA Pro reveal the flexibility of the peptidyl-tRNA Pro in the absence of EF-P. EF-P binding orders peptidyl-tRNA Pro to facilitate efficient peptide bond formation of the cyclic proline moiety (63). In addition to this function, EF-P also can suppress ϩ1 frameshifts, suggesting a previously unappreciated role in helping to maintain the mRNA frame (28). Conserved EF-P residues that interact with anticodon stem nucleotides 41 and 42 are essential for function (63). In the absence of the m 1 G37 modification, the interaction between EF-P and tRNA Pro may be destabilized due to the flexibility of the 3Ј stem (31). The interplay between the m 1 G37 modification in tRNA Pro and EF-P suggests that this elongation factor has synergistic roles in translational fidelity dependent on tRNA metabolism.

Ribosome purification
E. coli 70S ribosomes were purified as described previously (66). E. coli MRE600 cells were grown to an A 600 of ϳ0.7 in Luria broth (LB) medium at 37°C and then cooled on ice for 20 min. All centrifugation steps were performed at 4°C. Cells were pelleted by centrifugation and washed with buffer 1 (10 mM HEPES-KOH, pH 7.6, 10 mM MgCl 2 , 1 M NH 4 Cl, 6 mM ␤-mercaptoethanol (␤-Me)) twice and then resuspended in buffer 2 (10 mM HEPES-KOH, pH 7.6, 10 mM MgCl 2 , 100 mM NH 4 Cl, 6 mM ␤-Me). Cells were lysed using an EmulsiFlex-C5 high-pres-sure homogenizer (Avestin), and cell debris was removed by centrifuging at 13,000 ϫ g for 15 min. The lysate was further centrifuged at 27,000 ϫ g for 30 min to obtain the S30 fraction. Ribosomes were pelleted by centrifuging at 42,000 ϫ g for 17 h. The pellets were resuspended in buffer 2, and ribosomes were further purified over a 10 -40% sucrose gradient in buffer 2 at 70,000 ϫ g for 12 h. 70S ribosomes were separated from polysomes and subunits using a Brandel gradient fractionator. The 70S fractions were pooled, pelleted, resuspended in buffer 2, and stored at Ϫ80°C.

70S complex formation
ASLs and mRNAs were chemically synthesized (Integrated DNA Technologies), and purified E. coli tRNA fMet was purchased from Chemical Block (Table S1). mRNAs contained either an CCG or CCC-U in the A site. The E. coli 70S ribosome complex was formed by incubation with a 2-fold molar excess of mRNA for 5 min followed by a 2-fold molar excess of tRNA fMet for 30 min at 37°C. A-site ASLs were 5Ј-labeled with [␥-32 P]ATP (PerkinElmer Life Sciences) using T4 PNK enzyme (New England Biolabs).

Kinetic binding assays
A modified 96-well Bio-Rad dot-blot apparatus with two membranes was used to study binding kinetics of tRNAs to ribosomes (67). An upper nitrocellulose membrane and a lower nylon membrane (Amersham Biosciences Hybond-Nϩ, GE Healthcare) were pre-equilibrated in cold buffer 3 (5 mM HEPES-KOH, pH 7.5, 50 mM KCl, 10 mM NH 4 Cl, 10 mM Mg(CH 3 COO) 2 , 6 mM ␤-Me). All binding experiments were performed in buffer 3. Nonspecific binding was controlled for by having a [ 32 P]ASL-only sample with each experiment. Reactions were filtered by vacuum and immediately washed using 100 l of cold buffer 3. After incubations, membranes were dried and exposed to a PhosphorImager screen (GE Healthcare) and imaged on a Typhoon FLA 7000. Quantification was performed using ImageQuantTL software and analyzed using GraphPad Prism. The fraction of A-site ASL bound was calculated as the ratio between nitrocellulose counts and the total counts on both membranes after correcting for nonspecific binding.

Measuring K d values
Serial dilutions of the ribosome complex (70S, mRNA, P-site tRNA fMet ) were performed to generate a range of 70S concentrations from 0.98 nM to 1 M. [ 32 P]ASL was added and incubated for 3 h at 25°C. Reaction volumes of 10 l were applied to the filters, washed, and then quantified as described above. The ASL fraction bound was fit using a one-site-specific binding equation in GraphPad Prism as described previously (fraction bound ϭ (B max ϫ [70S])/(K d ϩ [70S]) (39, 68).

Measuring k on values
Association with the 70S A site was measured as described previously (42,43). Briefly, 15 l of 4.5 nM 32 P-labeled ASL was added to 15 l of increasing concentrations of 70S complex programmed with mRNA and tRNA fMet . Three-l aliquots were removed at different times (0.5, 1, 2, 3, 4, 5, 6, and 7 min), immediately filtered, and washed with 100 l of cold buffer 3.

ASL elements that ensure accurate tRNA CGG Pro decoding
Initial association rates were obtained using different concentrations of the 70S complex (final concentrations of 12.5, 25, 50, 100, and 200 nM). k on was derived as the slope of the linear regression performed on the initial rates versus [70S] plot.

Measuring k off values
Dissociation of the ASL from the 70S A site was initiated by a 1:100 dilution of the equilibrium binding reaction (1 M 70S, 3 M mRNA, 5 M tRNA fMet , 0.05 M [ 32 P]ASL) in buffer 3 containing 0.3 M unlabeled ASL. At 5-min intervals, 10-l aliquots of the reaction were removed, filtered, and washed. The ASL fraction bound was normalized to t ϭ 0. The natural log of the normalized fraction bound was fitted with a linear regression versus time, and k off was derived as the negative of the slope.

Crystallization, X-ray data collection, and structural determination
Purification of T. thermophilus 70S ribosomes, formation of complexes with mRNA and tRNAs, and initial screening conditions followed previously established protocols (46,68). Two l of the ribosome complex  (Table 2). Data sets were integrated and scaled using XDS (69), and the structure was solved by molecular replacement using 70S coordinates from PDB entry 4Y4O. Crystallographic refinements were performed with PHENIX (70) followed by manual model building in Coot (71). Figures were generated in PyMOL (72).