Essential structural elements in tRNAPro for EF-P-mediated alleviation of translation stalling

The ribosome stalls on translation of polyproline sequences due to inefficient peptide bond formation between consecutive prolines. The translation factor EF-P is able to alleviate this stalling by accelerating Pro-Pro formation. However, the mechanism by which EF-P recognizes the stalled complexes and accelerates peptide bond formation is not known. Here, we use genetic code reprogramming through a flexible in-vitro translation (FIT) system to investigate how mutations in tRNAPro affect EF-P function. We show that the 9-nt D-loop closed by the stable D-stem sequence in tRNAPro is a crucial recognition determinant for EF-P. Such D-arm structures are shared only among the tRNAPro isoacceptors and tRNAfMet in Escherichia coli, and the D-arm of tRNAfMet is essential for EF-P-induced acceleration of fMet–puromycin formation. Thus, the activity of EF-P is controlled by recognition elements in the tRNA D-arm.

T he EF-P bacterial translation factor was discovered in the 1970s and was reported to stimulate peptide bond formation between initiator fMet-tRNA fMet and puromycin (Pmn), an analogue of the 3 0 -end of aminoacyltransfer RNA 1,2 . Archaeal and eukaryotic EF-P homologues (aIF5A and eIF5A, respectively) were subsequently found 3 , suggesting that the EF-P-family proteins are evolutionarily conserved [4][5][6][7][8] . EF-P homologues in many organisms are posttranslationally modified. In EF-P from E. coli, Lys34 is modified to lysyl-hydroxylysine by three enzymes, EpmA, EpmB and EpmC (formerly known as YjeK, YjeA and YfcM, respectively) 5,6,[8][9][10] . Lysylation is required for the full activity of EF-P 11,12 . The knockout of genes coding for EF-P, eIF5A or their modifying enzymes affects a variety of biological functions or even causes lethality 13,14 . On the other hand, it is also known that E. coli reconstituted translation systems that lack EF-P, such as the PURE system 15 , can be used to express many different proteins, suggesting that EF-P is not essential for protein synthesis. This apparent paradox was resolved when two groups simultaneously demonstrated that EF-P is a specialized factor required for translation of an messenger RNA subset containing polyproline motifs 11,12 . When the ribosome encounters such motifs, it tends to stall, because peptide bond formation between three or more Pro residues, or two Pro residues in certain contexts, is very slow 11,12,[16][17][18][19] . EF-P rescues the stalled ribosomes to yield full-length proteins 11,12 . In E. coli, B270 proteins out of B4,000 proteins contain PPG or PPP motifs and their synthesis may be affected by the lack of functional EF-P 12 , explaining the pleiotropic or lethal phenotypes of deletion strains lacking EF-P/eIF5A or their modifying enzymes.
The structure of Thermus thermophilus 70S ribosomes in complex with EF-P and tRNA fMet revealed that EF-P binds between the exit (E) and the peptidyl (P) sites of the ribosome and interacts with the phosphate backbone of tRNA fMet at the acceptor stem, D-arm and the anticodon stem (Fig. 1a,b). Although these interactions may stabilize tRNA fMet in the P site 20 , the mechanism for the acceleration of peptide bond formation between fMet-tRNA fMet and Pmn remained unclear. Moreover, the structure of the ribosome in complex with EF-P and its cellular substrate Pro-tRNA Pro is currently not available, and the potential contribution of interactions between EF-P and the mRNA codon in the E site is uncertain 20,21 . Thus, it is not known whether interactions between EF-P and peptidyl-Pro-tRNA Pro accelerate Pro-Pro formation. Linear free energy relationships of peptide bond formation with a variety of Pro analogues on the ribosome and in solution suggest that the major determinant of the impaired reaction between consecutive Pro residues is the unfavourable steric arrangement of peptidyl-Pro-tRNA Pro in the P site 22 . The peptidyl moiety attached to the P-site tRNA Pro mediates ribosome stalling 12,16,17,19 . EF-P appears to act by entropic steering of Pro-tRNA towards a catalytically productive orientation in the peptidyl transferase centre of the ribosome 6 . However, how this is achieved remains unknown. This prompted us to address the following open questions: How important is tRNA Pro to elicit the stimulatory effect of EF-P on peptide bond formation? Does the codon-anticodon interaction of tRNA Pro contribute? Is the peptide bond formation with other secondary amino acids, such as N-methyl-amino acids, enhanced by EF-P? If the sequence of tRNA Pro is important, which parts of tRNA Pro are recognized by EF-P?
To identify the elements of peptidyl-Pro-tRNA Pro steering the interactions with EF-P in the ribosome complex, we used a reconstituted in-vitro translation system coupled with flexizyme technology, referred to as the FIT (flexible in-vitro translation) system 23 . We analysed the EF-P effect for various tRNA mutants engineered to decode sequential Pro codons. Furthermore, time-resolved experiments by quench flow techniques attributed the global translation effects to the step of peptide bond formation. We conclude that the function of EF-P critically depends on two constraints: (1) the P and A sites of the peptidyl transferase centre must be occupied by a peptidyl-Pro-tRNA and a sufficiently unreactive substrate carrying Pro or other secondary amino acid, respectively; and (2) the P-site tRNA must bear the tRNA Pro D-loop closed by a stable stem (for example, tRNA Pro D-arm sequence) regardless of the sequence of the remaining part of the tRNA molecule; among E. coli tRNAs, only tRNA Pro and tRNA fMet share such sequences. Thus, the action of EF-P is promoted by the presence of a Pro and the recognition elements in the P-site tRNA, which explains the specificity of the factor for polyproline motifs. a EF-P tRNA fMet D-arm Acceptor stem

Anticodon stem
Thermus thermophilus tRNA fMet Acceptor stem Anticodon stem Figure 1 | Structure of T. thermophilus EF-P and P-site tRNA fMet . (a) T. thermophilus EF-P and P-site tRNA fMet bound to the 70S ribosome. The data were obtained from the Protein Data Bank (PDB) ID:4V6A. EF-P is shown in green and the tRNA in grey, except for the D-arm, which is shown in blue. The ribosome and mRNA are omitted from the figure for simplicity. (b) Secondary structure of T. thermophilus tRNA fMet . The D-arm region close to EF-P is indicated in blue. Cm, 2'-O-methylcytidine D, dihydrouridine; m 7 G, 7-methylguanosine; T, ribothymidine; c, pseudouridine.

Results
EF-P recognizes all isoacceptors of Pro-tRNA Pro . First, the translation enhancement by EF-P was reproduced in the FIT system using three different mRNA constructs (mR1, mR2 and mR3) coding for Pro-Gly-Gly, Pro-Pro-Gly and Pro-Pro-Pro, respectively ( Supplementary Fig. 1). With such short sequences, the proline-induced stalling results in a significant peptidyl-tRNA drop-off, which is proportional to the duration of stalling. EF-P addition reduces stalling and thus increases the final product level, which enabled us to assess the EF-P effect through quantifying the peptide yield of translation. Pro insertion into the respective peptides (P1, P2 and P3) depends on exogenous Pro-tRNA Pro1 prepared by flexizyme technology 23,24 . Translation was carried out by adding the Pro-tRNA Pro1 and the DNA template for the corresponding mRNA to the FIT system 23 containing ribosomes, translation factors, native tRNA mixture, five aminoacyl-tRNA synthetases (ARSs; MetRS, LysRS, GlyRS, TyrRS and AspRS) and five amino acids (Met, Lys, Gly, Tyr and [ 14 C]Asp) in the presence or absence of EF-P. In this translation assay optimized with respect to incubation time and EF-P concentration ( Supplementary Fig. 2), EF-P significantly improved the translation yields of peptides with consecutive Pro residues (3.5-and 14.4-fold increase for P2 and P3, respectively), whereas the yield of the peptide P1 with a single Pro was not affected, consistent with previous studies 11,12,16 . To investigate the impact of tRNA structural features on the function of EF-P, we prepared eight tRNA transcripts with the sequences of E. coli tRNA Ser4 , tRNA Trp , tRNA Leu2 , tRNA Ile1 , tRNA Ala2 , tRNA Pro1 , tRNA Pro2 and tRNA Pro3 , and charged them with Pro by means of flexizyme technology. The corresponding mRNA templates, referred to as mR2-XXX 2 , each contained two tandem XXX codons that were decoded by the respective Pro-tRNA bearing the complementary anticodon (Fig. 2a). EF-P improved translation yield only when tRNA Pro isoacceptors (tRNA Pro1 , tRNA Pro2 and tRNA Pro3 ) were used to decode the tandem Pro codons, whereas no enhancement was observed with other tRNAs (Fig. 2b and Supplementary Fig. 3a). This result shows that the EF-P alleviates translation stalling with tRNA Pro isoacceptors only.
To determine whether the tRNA Pro codon-anticodon interaction is responsible for the enhancement, we prepared chimeric tRNAs derived from the sequences of tRNA Pro1 or tRNA Ser4 , but with the anticodons swapped (CGG or GGA) ( Fig. 2c and Supplementary Fig. 3b). We then tested the effect of EF-P on translation of mR2-CCG 2 or mR2-UCC 2 . EF-P was able to enhance translation regardless of the anticodon, but only if Pro was charged onto an isoacceptor of tRNA Pro1 ; EF-P did not improve the yield of translation with Pro-tRNA Ser4 CGG or Pro-tRNA Ser4 GGA ( Fig. 2c and Supplementary Fig. 3b). This result suggests that codon-anticodon interactions do not serve as determinants for EF-P function, but rather some other part of the tRNA structure is crucial.
Requirement for Pro and tRNA Pro . It is not known what makes the ribosome stall during the synthesis of poly-Pro sequences. As Pro is the only secondary proteinogenic amino acid, one could hypothesize that this structural feature serves as a specific recognition element for EF-P. If so, then EF-P may also enhance peptide bond formation with other secondary amino acids such as N-methyl-amino acids. To test this notion, we chose two non-proteinogenic secondary amino acids, L-N-methylalanine ( Me Ala) and L-N-methylthreonine ( Me Thr), to study their tandem incorporation in the presence and absence of EF-P using mR2 (Fig. 3a,b and Supplementary Fig. 3c); as a positive control, we also examined consecutive insertion of Pro into the peptide. The unnatural amino acids were charged onto tRNA Pro1 , which decoded two sequential CCG codons in the mRNA. To our surprise, EF-P had almost no effect when two identical N-methyl-amino acids were inserted, compared with the Pro-Pro control (Fig. 3b). This result indicates that EF-P selectively enhances Pro incorporation.
We then investigated which of the two Pro residues, the first or the second, is needed to elicit EF-P activity. We designed four templates (mR2 and mR4-6) which encode Pro-Pro (CCG-CCG), Pro-Me Thr (CCG-UCC), Me Thr-Pro (UCC-CCG) or Me Thr-Me Thr (UCC-UCC) (Fig. 3a, mR2, mR4, mR5 and mR6, respectively). We note that EF-P enhances translation when Pro is encoded by a UCC codon read by tRNA Pro1 GGA (Fig. 2c); hence, tRNA Pro1 GGA should be functionally identical to tRNA Pro1 CGG . In this system, EF-P stimulated the synthesis of Pro-Pro and Pro-Me Thr but not of Me Thr-Pro and Me Thr-Me Thr ( Fig. 3c and Supplementary Fig. 3d). Interestingly, EF-P enhanced Pro-Me Thr insertion by fourfold (P4), which is more than the effect measured for Pro-Pro (P2) in this peptide sequence context. tRNA body AUG AAG AAG AAG XXX XXX GGU UAC AAG GAC AAG UAA fM To further test which part of the P-site tRNA Pro molecule is essential for the enhancement by EF-P, we designed a combinatorial permutation experiment where the body sequences of Pro-tRNA Pro1 CGG and Me Thr-tRNA Pro1 GGA were changed to those of tRNA Ser4 CGG and tRNA Ser4 GGA , respectively ( Fig. 3d and Supplementary Fig. 3e). EF-P enhanced Pro-Me Thr synthesis when Pro-tRNA Pro1 CGG was present in the P site regardless of the tRNA in the A site, although Me Thr-tRNA Pro1 GGA gave higher enhancement than Me Thr-tRNA Ser4 GGA in the A site. On the other hand, EF-P was unable to facilitate Pro-Me Thr synthesis when Pro-tRNA Ser4 GGA was present in the P site regardless of the tRNA body structure or the secondary amino acid in the A site ( Fig. 3d and Supplementary Fig. 3e). These results indicate that EF-P recognizes both the peptidyl-Pro moiety and the tRNA Pro body sequence in the P site to accelerate the slow elongation caused by Pro-Pro or Pro-Me Thr elongation 25 .
EF-P recognition elements are in the D-arm of tRNA Pro . In the X-ray crystal structure of the ribosomeEF-PfMet-tRNA complex, EF-P contacts the backbone of the D-arm, acceptor and anticodon stem regions of tRNA fMet . As we showed above, the action of EF-P depends on the presence of tRNA Pro in the P site.
To determine which part of tRNA Pro is essential, we designed six chimeric tRNA constructs (tRNA chim ) based on the tRNA Pro1 sequence, where the D-arm, acceptor stem or anticodon stem regions were independently substituted with those of tRNA Ser4 or tRNA Leu2 (Fig. 4a). Pro was then charged onto the respective tRNA chim by the flexizyme and used to decode tandem CCG codons. EF-P did not enhance peptide synthesis when the D-arm of tRNA Pro1 was substituted with that of tRNA Ser4 or tRNA Leu2 (tRNA chim 1 or 2) ( Fig. 4b and Supplementary Fig. 3f). In contrast, the effect of the substitutions of acceptor stem or anticodon stem was much less pronounced (tRNA chim 3-6). We note that the translation yield in the absence of EF-P was similar for all mutant and WT tRNAs, indicating that mutant tRNAs were functionally active. These results suggest that the D-arm of tRNA Pro contains the major determinants for EF-P recognition.
To further investigate the functional importance of the D-arm bases, we first prepared eight D-arm single-point mutants (C13G, G15A, C16U, C17aU, U17bG, G18U, U20A and A21G; Fig. 4c and Supplementary Fig. 3g), four double-point mutants (C13G/C16U, C13G/U20A, C16U/C17aU and C17aU/U20A) and three triple-point mutants (C13G/C16U/C17aU, C13G/ C17aU/U20A and C16U/C17aU/U20A; Supplementary Fig. 4a). These tRNA mutants were charged with Pro and used to synthesize the P2 peptide. Among the single-point mutants, only the C13G mutation at the edge of the D-stem significantly decreased the stimulation by EF-P, whereas other mutations in the D-loop had no effect ( Fig. 4c and Supplementary Fig. 3g). Similarly, the double-and triple-point mutants containing a mutation at position 13 decreased the sensitivity to EF-P, whereas all others were similar to the wild-type tRNA Pro ( Supplementary  Fig. 4a). The effect of the C13G mutation was robust under all conditions tested ( Supplementary Fig. 2a,b) and specific for EF-P-induced translation ( Supplementary Fig. 3g). These results AUG AAG AAG AAG CCG CCG GGU GAC UAC AAG GAC GAC GAC GAC AAG UAA fM indicate that among the D-loop bases the C13/G22 base pair plays a critical role in mediating the enhancement by EF-P. We further validated the importance of the D-stem in the interaction with EF-P on translation of the amino-terminal part of the flagellar protein FLK, which contains PPP and PPG sequences ( Supplementary Fig. 5). EF-P significantly improved the FLK    yield, whereas only slight improvement was observed for the C13G mutants. This result provides further support for the conclusion that the D-arm structure in tRNA Pro is crucial for EF-P activity.
Together, these data indicate that the size of the D-loop in combination with the stable D-stem probably serve as critical EF-P recognition elements in the P site-bound peptidyl-Pro-tRNA Pro . The three naturally occurring isoacceptors of tRNA Pro have a 9-nt D-loop and a stable 4-bp D-stem with two G/C pairs at positions of 13/22 and 12/23. Among the 46 E.coli tRNAs, this structure is shared by only the three Pro isoacceptors and tRNA fMet , and would be a key factor for selective enhancement of translation by EF-P ( Fig. 4h and Supplementary Table 1). Robust EF-P activity was observed with quite a number of mutations in the D-loop, provided the length of the loop was maintained (Fig. 4c and Supplementary  Figs 3g and 4a). In addition, different base pairs in the D-stem were tolerated as long as the stability of the D-stem was maintained (Fig. 4d,e and Supplementary Fig. 3h,i). Even though a posttranscriptional dihydrouridine modification occurs at position 20 in the natural E. coli tRNA Pro ( Supplementary  Fig. 4b), the activity of an in-vitro transcript lacking this modification is very similar to that of the fully modified tRNA ( Supplementary Fig. 4c), including EF-P recruitment and catalysis 22 , suggesting that the dihydrouridine modification is not involved in EF-P recognition. These results allow us to hypothesize that EF-P does not recognize the D-loop of tRNA Pro in a sequence-specific manner, but rather senses the loop size closed by the stable D-stem via phosphate backbone interactions. EF-P recruitment by substituting D-arm of inactive tRNAs. To further verify the recognition determinants in the D-loop in the absence of possible interactions with other parts of tRNA Pro , we prepared chimeric tRNAs derived from tRNA Ser4 and tRNA Ala2 , which are normally not recognized by EF-P. The parental wild-type tRNA Ser4 has an 11-nt D-loop and a 3-bp D-stem, whereas the tRNA Ala2 has an 8-nt D-loop and a 4-bp D-stem. We replaced the native D-arm and anticodon sequences of these tRNAs with the D-arm of tRNA Pro1 and the CGG anticodon capable of decoding the CCG Pro codon (Fig. 5a,c). These chimeric tRNAs were charged with Pro and used for the translation of the mR2 mRNA.
Although translation was insensitive to EF-P when the single compensatory mutation of C12G/G23C in tRNA Ser4 (tRNA Ser4 -1) was used, stabilization of the D-stem by the G13C/A22G mutation (tRNA Ser4 -2), which changed the loop size to 9 nt, conveyed a small EF-P-dependent activation effect even though the D-loop sequence was quite different (Fig. 5b and Supplementary Fig. 3k). The EF-P-mediated translation enhancement was fully restored when the intact stable D-stem and D-loop derived from tRNA Pro1 were introduced into tRNA Ser4 (tRNA Ser4 -3). A similar trend was also seen in the chimeric tRNA Ala2 (tRNA Ala2 -3), in which the stable 4-bp D-stem and the 9-nt D-loop were engineered by insertion of a single base (17aU) even though the D-loop sequence was not identical to that of tRNA Pro (Fig. 5d and Supplementary Fig. 3l).
We further tested Pro incorporation using tRNA Val2A that comprised the same length of D-loop and D-stem as tRNA Pro (a 9-nt D-loop and a 4-bp D-stem) (Fig. 5e,f and Supplementary  Fig. 3m). EF-P slightly accelerated Pro incorporation with the WT tRNA Val2A , but the effect was smaller than that with tRNA Pro1 due to the weaker base pair (U/A) at the 12/23 than that of tRNA Pro1 (G/C). Given that U12/A23 and U13/A22 mutants of tRNA Pro1 also showed slight enhancement (Fig. 4d,e), U/A base pairs at these positions weakly contribute to EF-P function. The EF-P effect was fully restored by stabilization of the D-stem by the U12G/A22C mutation (tRNA Val2A -1), whereas the effect was abolished by destabilization of the D-stem by U12A mutation (tRNA Val2A -2).
These results suggested that alterations of the D-loop sequence can be tolerated, but the full EF-P effect can be only achieved when the geometry of the tRNA Pro D-loop is maintained, which presumably allows for EF-P recruitment. These results further support the conclusion that the major recognition determinant in tRNA Pro comprises the D-loop closed by the stable D-stem.
EF-P recognizes the D-arm structure of tRNA fMet . As discussed above, tRNA fMet and the three tRNA Pro isoacceptors share the same D-arm motif, that is, a 9-nt D-loop closed by a stable 4-bp D-stem with two G/C pairs at 12/23 and 13/22 (Fig. 4h and Supplementary Table 1). As EF-P accelerates not only the synthesis of poly-Pro motifs but also the fMet-Pmn reaction 1,12 , it is likely to be that EF-P also recognizes the D-loop of fMet-tRNA fMet . To test this hypothesis, we first studied Pro incorporation using tRNA fMet CGG derivatives (tRNA fMet CGG -1-4) containing D-arm substitutions along with the CGG anticodon (Fig. 6a) and charged with Pro. As expected, EF-P enhanced translation of P2 peptide when Pro-tRNA fMet CGG was used in the assay (Fig. 6b and Supplementary Fig. 3n, tRNA fMet CGG -1 versus tRNA Pro1 CGG ). Substitution of the D-arm of tRNA fMet with those of tRNA Ser4 and tRNA Leu2 or introduction of the C13G or DC17a mutations ( Fig. 6b and Supplementary Fig. 3n, 2-5) eliminated EF-P-mediated activation. These results indicated that Pro-tRNA fMet CGG , similar to Pro-tRNA Pro1 CGG , is able to recruit EF-P. The above result of the translation experiment using Pro-tRNA fMet CGG prompted us to test fMet-Pmn formation using fMet-tRNA fMet CAU or its mutants. We used four tRNA fMet CAU mutants in addition to the wild-type tRNA fMet CAU (Fig. 6c, 1-4). The initiation complex of ribosome containing f[ 35 S]Met-tRNA fMet CAU or its mutants was incubated with Pmn in the presence or absence of EF-P ( Fig. 6d and Supplementary  Fig. 3o, WT and 1-4). EF-P enhanced the formation of fMet-Pmn dipeptide over background when the wild-type fMet-tRNA fMet CAU was used, although its enhancement by 1.8-fold was less than the 4-fold effect observed in P2 translation. The Pmn reaction with tRNA fMet CAU mutants was insensitive to EF-P ( Fig. 6d and Supplementary Fig. 3o).
Kinetic effect of the D-arm mutants. Although the data described above were generated by the analysis of translation yields of 15-mer peptide containing consecutive Pro residues, they are expected to represent the rate-limiting step of translation, which is most likely to be the formation of the ProPro-Gly peptide bonds. We further tested the effect of the D-arm mutations by measuring the rate of Pro-Gly formation between fMet-Pro-tRNA and Gly-tRNA, which is known to be accelerated by EF-P 12,22 . We tested four representative D-arm variants derived from tRNA Pro1 CGG (Fig. 7a) and tRNA Ser4 CGG (Fig. 7b), and charged the tRNA constructs with Pro. The wild-type or mutant Pro-tRNA was incubated with EF-TuGTP and added to the ribosome initiation complex programmed with an mRNA coding for fMet-Pro-Gly tripeptide and containing f[ 14 C]Met-tRNA fMet in the P site, which resulted in the formation of fMet-Pro-tRNA Pro . Next, EF-G was added to form posttranslocation complexes that were subsequently purified by size-exclusion chromatography. Posttranslocation complexes (fMet-Pro) were rapidly mixed with [ 3 H]Gly-tRNA Gly3 EF-TuGTP to measure the rate of Pro-Gly formation using a quench-flow technique. The rate constant (k obs ) was determined by exponential fitting of time courses (Supplementary Fig. 6a-f).
With the wild type and all tRNA Pro mutants, the time courses of fMet-Pro-Gly formation in the absence of EF-P were similar, with a k obs in a range of 1-2 s À 1 (Supplementary Fig. 6a-f). This indicates that the variations of the D-arm sequence did not affect the intrinsically slow Pro-Gly formation when EF-P is absent. In the presence of EF-P, the rates of peptide bond formation with the native tRNA Pro1 CGG (WT) and the tRNA Ser4 CGG variant (4) bearing the D-arm derived from tRNA Pro1 increased (Fig. 7c,d) by about 18-and 8-fold, respectively. In contrast, the reactions with tRNA Pro1 CGG mutants and native tRNA Ser4 CGG were hardly affected by EF-P addition (Fig. 7c, 1-3 and Fig. 7d, WT, respectively). As expected, the kinetically resolved EF-P effects are much larger than those obtained from the translation yield analysis. These results fully   ARTICLE support the hypothesis that EF-P recognizes the D-arm structure of the peptidyl-Pro-tRNA to enhance the rate of Pro-Gly formation. Thus, we conclude that the EF-P enhancement observed in the translation of the 15-mer peptide should faithfully reflect the identity elements on tRNA required for the productive recruitment of EF-P.

Discussion
Our experiments using Pro-tRNA derivatives identify the structural elements in tRNA Pro that are required for the EF-P function: the D-loop structure closed by the stable base pairs in the D-stem of peptidyl-Pro-tRNA Pro (Fig. 4a-e). Many point mutations in the D-loop are tolerated ( Fig. 4c and Supplementary  Figs 3g and 4a) but alteration of the D-loop size inhibits EF-P function (Fig. 4f). These observations suggest that the overall loop structure of the D-loop is the essential recognition element for EF-P. In contrast, alterations in the acceptor stem, anticodon stem and anticodon of tRNA Pro have only modest effects. In fact, the chimeric derivatives of tRNA Ser , tRNA Leu (Figs 4a,b and 5a,b, and Supplementary Fig. 3f,k), tRNA Ala (Fig. 5c,d and Supplementary  Fig. 3l) and tRNA Val2A (Fig. 5e,f and Supplementary Fig. 3m) in which the D-arm sequence of tRNA Pro is implanted have a gain-of-function phenotype. The structure of Pro in the P-site peptidyl-Pro-tRNA Pro also plays a critical role in EF-P function, that is, peptide bond formation with other secondary amino acids such as N-methyl-amino acids is not enhanced by EF-P ( Fig. 3 and Supplementary Fig. 3c). Importantly, the kinetic measurement of EF-P-stimulated Pro-Gly formation using various 'active' and 'inactive' derivatives of Pro-tRNA Pro CGG and Pro-tRNA Ser CGG supports the results observed for 15-mer peptide expression (Fig. 7). This detailed analysis also showed that all tested tRNA constructs formed stable posttranslocation complexes and showed similar rates of peptide bond formation in the absence of EF-P, suggesting that the alterations in these tRNAs did not affect their activity in translation. These results also validate the loss-of-function experiments in which the absence of an EF-P effect may have otherwise been explained by a strongly impaired translation reaction. EF-P was originally identified as a protein factor that stimulates peptide bond formation between fMet and Pmn, and the D-arm of tRNA fMet is virtually identical to that of tRNA Pro (Fig. 4h); consistent with this observation, the effect of EF-P on translation in the presence of Pro-tRNA fMet CGG is dependent on the presence of the correct D-arm (Fig. 6a,b and Supplementary Fig. 3n). Furthermore, this was also the case for fMet-Pmn formation with fMet-tRNA fMet CAU , although the observed degree of fMet-Pmn acceleration is less than that for translation. In addition, for the fMet-Pmn reaction, tRNA fMet variants with the D-loop sequences derived from tRNA Ser4 and tRNA Leu2 do not support EF-P function (Fig. 6c,d). Based on these results we propose that tRNA fMet Relative yield of fMet-puromycin, EF-P(+)/ EF-P(-) EF-P recognizes the shape of the phosphate backbone of the D-loop in the P-site peptidyl-Pro-tRNA Pro , to accelerate the formation of Pro-Xaa peptide bonds, in which Xaa is a poor A-site substrate such as Gly, Pro or other secondary amino acids. Comparison of the D-arm sequence of E. coli tRNA Pro with that of other prokaryotic species shows that the 9-nt loop and stable 4-bp stem structure is well conserved among them (Supplementary Table 2). Although U12/A23 and U13/G22 pairs are found in several species, these base pairs would also be tolerated considering that the E. coli tRNA Pro1 mutant and tRNA Val2A with these sequences elicited a modest EF-P effect (Fig. 4d,e and Fig. 5e,f). Therefore, D-arm recognition by EF-P may be similar in diverse prokaryotic organisms. Eukaryotic translation factor eIF5A is also able to stimulate incorporation of consecutive Pro amino acids 26 and peptide bond formation between Met-tRNA i Met and Pmn [27][28][29] . Although the D-stem of eukaryotic tRNA Pro is generally shorter by 1 bp than in the prokaryotic counterparts, a 9-nt D-loop closed by a 3-bp stem is well conserved among eukaryotic species (Supplementary  Table 2). Thus, it is likely to be that eIF5A recognizes the phosphate backbone of the D-loop in peptidyl-Pro-tRNA Pro in a similar way as EF-P. Hydroxyl radical probing data of the P-site tRNA region proximal to eIF5A broadly suggested potential contacts of eIF5A with the acceptor stem, D-loop, anticodon stem and T-loop of the P-site-bound tRNA in 80S ribosome 26,30 .
However, such analyses did not reveal the critical contacts of the D-arm essential for the EF-P enhancement, probably because the ground-state mapping of the EF-P-tRNA contacts might not reflect the kinetic importance of the D-arm contributing to the EF-P enhancement. Therefore, appropriate biochemical studies similar to the experiments reported in this work would be necessary to confirm the importance of the D-arm in eukaryotes.
On the other hand, the length of the D-loop and D-stem of archaeal tRNA Pro differs among species (Supplementary Table 2), suggesting that aIF5A, the archaeal counterpart of EF-P, may not recognize the D-arm of tRNA Pro in the same manner. If so, archaeal aIF5A might be evolutionarily distant from prokaryotic EF-P and eukaryotic eIF5A in terms of its tRNA recognition mechanism; this question should be addressed in future experiments using the flexizyme technology and archaeal translation components. Notably, it was reported that a prokaryotic translation factor, EttA, binds to the E site of the ribosome and bridges the ribosomal L1 stalk and the P-site tRNA 31 , involving the EttA recognition of C17a and U17b in the D-arm of P-site tRNA fMet . Although these bases are also shared by tRNA Pro1 and tRNA Pro2 , point mutations at these positions in tRNA Pro1 did not affect the function of EF-P (Fig. 4c), indicating that the binding mode of EF-P is different from that of EttA. Apart from EF-P and EttA, there are a growing number of translation factors such as RbbA and HflX that bind to the tRNA Pro1  E site 32,33 . The cellular functions of these factors are different: EttA senses the energetic status of the cell and gates the entry into the elongation cycle of translation, HflX splits ribosomes under stress and RbbA ejects the E-site tRNA from the ribosome. The use of recognition motifs such as presented here for EF-P might be a common motif to orchestrate these diverse functions.
There is as yet an open question as to what is the mechanism of EF-P acceleration of Pro-Pro bond formation. A plausible hypothesis is that via the interaction of EF-P with the D-arm of tRNA Pro in the P site, the ester's carbonyl group Pro-tRNA Pro is appropriately positioned in the peptidyl transfer centre and this facilitates the nucleophilic attack of the secondary amino group on Pro or an N-methyl-amino acid, which are intrinsically poorer nucleophiles compared with the primary amino groups of other proteinogenic amino acids. Although this hypothesis does not tell us the mechanism at the molecular level, structural studies with a higher resolution may reveal this in more detail.

Methods
Preparation of aminoacyl-tRNAs. All tRNA constructs used for Pro and N-methyl-amino acid incorporation were prepared by in vitro transcription, except for the experiment in Supplementary Fig. 4c. Template DNAs for T7 RNA polymerase transcription were prepared using Taq DNA polymerase by extension of forward and reverse primers pairs (Supplementary Table 3), followed by PCR using forward and reverse PCR primers (Supplementary Table 3). The DNA products were purified by phenol extraction and ethanol precipitation, and then transcribed at 37°C for 12 h in a 200-ml reaction mixture consisting of 40 mM Tris-HCl (pH 8.0), 22.5 mM MgCl 2 , 1 mM dithiothreitol (DTT), 1 mM spermidine, 0.01% Triton X-100, 0.04 U ml À 1 RNasin RNase inhibitor (Promega), 3.75 mM NTP mix and 5 mM CMP or GMP, or AMP, depending on the 5 0 -end nucleotide of the tRNA. The resulting RNAs were treated with RQ1 DNase (Promega) for 30 min at 37°C and then purified by 8% denaturing PAGE containing 6 M urea.
Aminoacylation of the tRNAs was carried out on ice in reaction mixtures containing 50 mM HEPES-KOH (pH 7.5), 600 mM MgCl 2 , 20% dimethyl sulfoxide, 25 mM dFx, 25 mM tRNA and 5 mM proline dinitrobenzyl ester (DBE), N-methylalanine DBE or N-methylthreonine DBE. Reactions with N-methylalanine and N-methylthreonine were performed for 6 h, those with proline for 2 h. Aminoacyl-tRNA was precipitated by ethanol and dissolved in 1 mM sodium acetate (pH 5.2) to a concentration of 250 mM.
The [ 14 C]Pro-labelled native tRNA Pro1/Pro2/Pro3 mixture used in Supplementary  Fig. 4c was prepared as follows. Total tRNA was purchased from Roche. For aminoacylation, 400 mM total tRNA, 5% (v/v) S100 fraction as source of ARSs, 3 mM ATP and 25 mM [ 14 C]Pro were incubated for 45 min in aminoacylation buffer (50 mM HEPES-KOH (pH 7.8), 70 mM NH 4 Cl, 30 mM KCl and 20 mM MgCl 2 ). [ 14 C]Pro-tRNA Pro was phenolized and precipitated with ethanol. Aminoacyl-tRNA was dissolved in water and added to EF-TuGTP to form ternary complex [ 14 C] Pro-tRNA Pro EF-TuGTP and separated from uncharged total tRNA by sizeexclusion chromatography on two tandem Superdex 75 columns (GE Healthcare). Isolated ternary complexes were phenolized and [ 14 C]Pro-tRNA Pro was precipitated with ethanol.
Preparation and analysis of lysylated EF-P. E. coli EF-P gene was cloned into a modified pET28a vector that has PreScission protease cleavage site instead of thrombin cleavage cite. E. coli EpmA and EpmB genes were cloned into pETDuet vector. These proteins were expressed in E. coli Rosetta2 (DE3) cells with 0.5 mM isopropyl-b-D-thiogalactoside, followed by cell lysis by sonication and loading onto a Ni-NTA column to purify the histidine-tagged EF-P. The column was washed with Buffer A containing 20 mM Tris-HCl (pH 8.0), 200 mM NaCl, 2 mM imidazole and 1 mM 2-mercaptoethanol, and then the protein was eluted by Buffer A with 300 mM imidazole. The eluate was dialysed against the Buffer A and the histidine tag of EF-P was cleaved using Turbo3C protease at 4°C overnight. The protein was reloaded on a Ni-NTA column and the flowthrough and wash fractions were pooled. Finally, the protein was further purified on a Resource Q column (GE Healthcare).
Lysylation of EF-P was confirmed by mass spectrometric analysis according to the following procedure ( Supplementary Fig. 7). Fifty to 100 pmol EF-P was separated on a 4-20% gradient SDS-PAGE gel, stained with Coomassie Blue and excised for in-gel proteolysis. Cysteines were reduced with 10 mM DTT for 30 min at 56°C and alkylated with 55 mM iodoacetamide for 60 min at room temperature in the dark. EF-P was proteolysed with 12.5 ng ml À 1 trypsin for 16 h at 37°C. Extracted peptides were dried and dissolved in 5% acetonitrile/0.1% formic acid. Next, proteolyszed EF-P was analysed by reverse-phase HPLC-electrospray ionization-tandem mass spectrometry using a Dionex Ultimate 3000 HPLC system connected to a QExcative Plus mass spectrometer. To create a spectral library that comprises peptides that correspond to the unmodified, lysylated and lysylated/ hydroxylated form of EF-P initial data acquisition was performed in the data-dependent mode. The tryptic peptides of interest were included in a global inclusion list with their charge states z ¼ 2-5. Peptide identification was achieved by database searching with the MaxQuant software (version 1.5.2.8) (ref. 34) against the UniProt E. coli (K12) proteome. Lysylation ( þ 128) and lysylation/ hydroxylation ( þ 144) were used as variable modifications. To build a spectral library, the results were further analysed using Skyline software (version 3.5) (ref. 35). Prominent charge states (unmodified peptide z ¼ 3, modified peptides z ¼ 4) were chosen for relative quantification by scheduled parallel reaction monitoring 36 . The resolution of tandem mass spectrometry scans was 35,000, the maximum ion time was 50 ms using an isolation window of 1 m/z and the collision energy of HCD fragmentation was 28 eV. The tryptic peptides of interest have a methionine close to the N terminus, oxidation of which leads to a þ 16 mass shift (identical to hydroxylation). As the degree of interfering oxidation may vary between different preparations, only y-ions that did not have this interference were used for quantification. The oxidation status was independent of the degree of modification. Parallel reaction monitoring transitions were extracted and integrated in Skyline at a resolution of 35,000. For quantification, the sum of peak areas of the seven most intense product ions was considered.
Translation of model peptides. Translation was carried out in a cell-free coupled transcription-translation FIT system 23 . The reactions contained only five ARSs (MetRS, LysRS, GlyRS, AspRS and TyrRS) and five amino acids (Met, Lys, Gly, Asp and Tyr) together with E. coli total tRNA; under these conditions, free Pro is not charged onto the natural tRNA Pro isoacceptors by ProRS. Transcription of the DNA templates into mRNAs was carried out by T7 RNA-polymerase present in the FIT system. Reactions were carried out at 37°C for 20 min, except for the time-course analysis in 2.5 ml reaction mixture consisting of the following reagents: 50 mM HEPES-KOH (pH 7.6), 100 mM potassium acetate, 12.3 mM magnesium acetate, 2 mM ATP, 2 mM GTP, 1 mM CTP, 1 mM UTP, 20 mM creatine phosphate, 0.1 mM 10-formyl-5,6,7,8-tetrahydrofolic acid, 2 mM spermidine, 1 mM DTT, 1.5 mg ml À 1 E. coli total tRNA, 1. Assay validation. To screen for EF-P interaction determinants in tRNA Pro , we established an assay using the FIT system described above. The reaction was performed in the absence or presence of EF-P. To measure the EF-P effect by point measurements we selected a strong stalling motif (Pro-Pro-Gly) 16,37,38 ( Supplementary Fig. 1b). When such strong stalling occurs on synthesis of short peptides, ribosomes release some of the peptidyl-tRNA. The extent of the peptidyl-tRNA drop-off is proportional to the duration of stalling. Therefore, the proline-induced stalling-and the EF-P effect-are not only reflected in the differences in the translation rates but also in the final product level, enabling us to assess the EF-P effect through quantifying the peptide yield of translation. To optimize the assay, we first measured the time courses of reactions and the optimum EF-P concentration ( Supplementary Fig. 2a,b); in all subsequent experiments, the incubation time was set for 20 min and EF-P concentration 3 mM. Next, we validated the EF-P effect and established the dynamic range of the assay by using three different mRNA constructs (mR1, mR2 and mR3) coding for Pro-Gly-Gly, Pro-Pro-Gly and Pro-Pro-Pro, respectively ( Supplementary Fig. 1). EF-P significantly improved translation yields of peptides with consecutive Pro residues (3.5-and 14.4-fold increase for P2 and P3, respectively), consistent with previous studies 11,12,16 . In contrast, the yield of the peptide P1 with a single Pro residue was significantly higher and not affected by the presence of EF-P, presumably due to a lack of stalling, establishing the dynamic range of our assay. As we use tRNA transcripts that lack all posttranscriptional modifications ( Supplementary Fig. 4b), we compared the EF-P effect in our assay using native and transcribed tRNA Pro (Supplementary Fig. 4c). The modifications did not affect the translation yield, in agreement with previous reports suggesting that the tRNA Pro transcript is fully active with respect to aminoacylation, ternary complex formation and translation 22 .