Programming peptidomimetic syntheses by translating genetic codes designed de novo

Materials and Methods

Substrates. Synthetic genes were cloned to enable in vitro synthesis of tRNA^minusCA species (from FokI-cut templates) for ligation to an aa-pdCpA or synthesis of full-length tRNAs (from BstNI-cut templates). The tRNA sequences contained substitutions at their 5′ and 3′ termini to maintain the secondary structure of the aminoacyl stems while enabling efficient transcription initiation at the first nucleotide with GMP by T7 RNA polymerase. Nvoc-aa-pdCpA derivatives of eU (25), mS (26), bK (27), and yU§ (Fig. 3b) were prepared and ligated to tRNA^minusCA species by using general methods (11). The concentrations of Nvoc-aa-tRNA ligation products were estimated by urea polyacrylamide gel electrophoresis at pH 5 (ligation of certain aa-pdCAs required a concentration of T4 RNA ligase (New England Biolabs) severalfold higher than that recommended (11); only efficient ligations were used). Natural aatRNAs were prepared from pure isoacceptors (Subriden RNA, Rolling Bay, WA) as described (22) or with pure recombinant RSs (20). The specific activities of ³H-labeled amino acids were 21,400 (Fig. 2b), 14,600 (Fig. 2c), and 16,900 (Fig. 3) dpm/pmol.

Translations. mRNAs and translation mixes were prepared as described (22), except that translation components differed by omission of polyethylene glycol, addition of His-tagged EF-Ts (28), further purification of initiation factor (IF)2 by gelfiltration chromatography, and additional washing of ribosomes. Ribosome salt washes were as described (22), except that an additional high-speed spin of 1 min preceded the final pelleting of the four-times-washed ribosomes to remove residual insoluble material. Ribosomes and factors were not contaminated with RSs or proteases, as measured by charging of total tRNA (Sigma) with 15 ¹⁴C-labeled amino acids (New England Nuclear) and by stability of peptides. Macromolecular concentrations in translations were adjusted slightly to give 0.5 μM each of IF1, IF2, IF3, EF-G, and EF-Ts, 2.5 μM EF-Tu, four-times-washed ribosomes at 0.029 A₂₆₀ unit/μl [27 nM estimated to be active (22)], 1 μM mRNA, 0.2 μM , and 0.5 μM (unless otherwise indicated) each elongator aa-tRNA, and translations were typically performed at 37°C for 30 min without preincubation. Translations analyzed by cation-exchange (treatment with alkali, acidification, then minichromatography to separate anionic formylated peptides from unformylated amino acids) were all performed on a 1-pmol scale with respect to limiting input , whereas the scales varied for translations analyzed by HPLC. Peptide markers were synthesized on an Advanced Chemtech peptide synthesizer from commercial reagents.

Results

A Purified RS-Free Translation System with Modular tRNA Adaptors. Our system (Fig. 1a) was constructed from ribosomes purified exhaustively to remove measurable contaminating RS charging activities (see Materials and Methods), recombinant translation factors (22), in vitro-synthesized mRNAs, in vitro-charged native tRNA isoacceptors, and chemoenzymatically synthesized aatRNAs. In pilot studies, a tRNA^Asn-based chemoenzymatically synthesized elongator tRNA that was arbitrarily chosen (termed tRNA_GGU AsnB, where the subscript refers to the anticodon; Fig. 1b) was nonenzymatically charged with eU (structure shown in Fig. 3b) and assayed in the purified translation system for single incorporation directed by mRNA MTV (see Materials and Methods). It was comparable in specificity and efficiency to the natural , even at 0.5 μM (a concentration 20 to 40-fold lower than typically needed in crude translation systems; refs.11 and 31), alleviating concerns that altering the tRNA body (Fig. 1b) or charging with the unnatural amino acid eU might be problematic. Control translations with unacylated full-length in place of did not synthesize any full-length peptide, confirming that eU was indeed incorporated into products (data not shown).

Translating mRNA into a Specific Peptidomimetic Polymer. We then used the chemically aminoacylated adaptor (Fig. 1b) to translate mRNA MT₅V (Fig. 2a) to test for site-specific incorporation of several adjacent unnatural amino acids. This combination of adaptor and template allowed us to compare the efficiency of synthesis of the unnatural fM(eU)₅V product to that of the natural fMT₅V product, synthesized by translation of the same template with the native (Fig. 2b, green). When was substituted for (Fig. 2b, blue), five adjacent eUs were incorporated into fM(eU)₅V product at ≈30% overall yield in comparison with fMT₅V, assuming similar recovery of fMT₅V and fM(eU)₅V during analysis. Control translations (Fig. 2b, red) without a cognate aa-tRNA for the T codon did not incorporate [³H]valine, confirming the specificity of decoding. HPLC analysis of the fM(eU)₅V translation revealed comigration of the product with authentic fM(eU)₅V marker prepared by chemical synthesis (Fig. 2c).

Creation of Unnatural Genetic Codes. To test the feasibility of creating rudimentary genetic codes, we constructed an mRNA containing three adjacent different test codons (AAC, ACC, and GUU) and also constructed the necessary two additional tRNA adaptors by using the tRNA^AsnB body (Fig. 3a). Importantly, the test codons had been chosen arbitrarily and are present in three different codon boxes (standard groupings) of the universal genetic code, whereas the adaptors for their translation had been created rationally, differing only by anticodon mutations guided by the rules of codon–anticodon base pairing. We charged each adaptor chemically with an unnatural amino acid according to our blue genetic code (Fig. 3b Upper) and used these substrates for translation of the five-codon template (Fig. 3a). The encoded fM-yU-mS-eU-E product is indeed synthesized, based on the incorporation of radioactivity into product (Fig. 3c, blue open triangle), omission experiments (see below), and the comigration on HPLC with authentic synthetic peptide (Fig. 3d). We further examined the modularity and specificity of the approach by translating the same mRNA (Fig. 3a) by using the purple genetic code (Fig. 3b Lower) to synthesize the closely related sequence fM-yU-eU-mS-E. This translation required the preparation of two new combinations of unnatural amino acids and tRNA^AsnB bodies and yielded the expected fM-yU-eU-mS-E product, as judged by incorporation of radioactivity into product (Fig. 3c, purple filled triangle) and comigration with authentic marker peptide on HPLC (Fig. 3e). The lack of read-through by noncognate aa-tRNAs in more stringent experiments, in which each chemically aminoacylated tRNA was individually omitted, further demonstrates the specificity of decoding in these translation reactions (Fig. 3c, red triangles).

The yield of fM-yU-mS-eU-E is ≈55% when compared with the amount of product translated from mRNA MVE with natural aa-tRNA substrates (Fig. 3c legend), and the amount of fM-yU-eU-mS-E is lower (Fig. 3c). When the adaptor is chemically aminoacylated with the bulky biotinyllysine derivative (Fig. 3b Right) and tested for a single incorporation, the yield is only 20% [measured by binding to avidin (22); data not shown]. In practice, therefore, it is advisable first to test each charged tRNA individually, as is currently done with crude systems. It is also apparent that even our tRNAs charged with smaller unnatural amino acids give somewhat lower yields of products in translations requiring three to five adjacent unnatural amino acid incorporations in comparison with control translations containing only natural aa-tRNAs.

Discussion

Prior engineering of translation has been limited to the specific reassignment of the amino acid identity of only a small subset of the 64 codons at two nonadjacent codon positions within an mRNA because of the inherent restrictions of suppressor tRNAs and crude translation systems. Here, our combination of a purified translation system free of RSs with chemoenzymatically synthesized nonsuppressor aa-tRNA substrates enabled several, adjacent, arbitrarily chosen codons to be completely reassigned to unnatural amino acids in a potentially generalizable manner. The plasticity of translation illustrated by our studies supports the notion that the universality of the genetic code is primarily due to intrinsic constraints imposed not by the core translation apparatus but rather by RSs and the rest of the proteome. This idea is consistent with the known greater divergence of mitochondrial genetic codes, which encode very few proteins (32).

The study of translation using custom-designed substrates with our purified system (a purified “polypeptide polymerase”) has advantages over crude systems. Although inefficiencies with unnatural aa-tRNAs in crude systems are at least partly due to competing activities, this cannot be the case in our system because competitors have been deliberately excluded. Thus, the interesting observation here that unnatural substrates are still less efficient than natural ones directly reveals substrate determinants necessary for efficient translation (Figs. 1b and 3a) and suggests the existence of other competing reactions [one possibility is peptidyl-tRNA drop-off from the ribosome (23, 33)]. Nevertheless, the findings that our incorporations were more efficient than generally observed in crude systems, despite using lower concentrations of unnatural aa-tRNAs and the omission of all termination factors (33), and that adjacent unnatural amino acids can be incorporated broadens the range of experimental possibilities. For example, appropriately chosen pairs of adjacent unnatural amino acids might be used to probe the chemical mechanism of ribosomal peptide bond formation.

What recognition elements have been altered by the introduction of unnatural features into our substrates? Given that the affinities of EF-Tu for native aa-tRNAs are similar, whereas its affinities for mismatched combinations of amino acids and tRNA bodies are very different, one hypothesis is that efficient delivery by EF-Tu to the ribosome of each amino acid may require matching with a tRNA body of appropriate compensatory affinity for EF-Tu (34). Alternatively, tRNA nucleoside modifications can be important for efficient translation in vivo (35), whereas interpretation of in vitro studies with unmodified tRNAs by using crude charging or translation systems is complicated by the presence of endogenous modification activities (24). The loss of anticodon loop nucleoside modifications in many tRNAs is hypothesized to decrease anticodon–codon stability on the ribosome (36), which could lead to increased dissociation of cognate aa-tRNAs from the ribosome at both the initial selection and proofreading steps (37). Perhaps unnatural amino acid incorporation from our substrates could be improved by altering EF-Tu or the ribosome.

Our system should facilitate unambiguous definition of substrate elements that affect translational activity, including the enigmatic nucleoside modifications. Synthetic aa-tRNAs could be constructed to more closely resemble readily available natural aa-tRNAs for comparative studies, e.g., by using tRNA sequences other than that of tRNA^Asn and unmutated tRNA sequences created by cleavage of in vitro synthesized precursors (38, 39). These findings may also be helpful in extending the synthetic scope of our initial system and existing suppressor systems.

Finally, we propose a potential application of our system for ligand discovery. Although the scalability of our system is not designed to provide an alternative to solid-phase peptide synthesis for preparation of individual peptides or to suppressor technology for preparation of proteins containing one or two nonadjacent unnatural amino acids, it is designed to produce the largest screenable libraries of small peptidomimetics. Theoretically, translation with 20 different aa-tRNAs of mRNA templates containing 10 random codons gives a library of 10¹³ different peptidomimetics (we synthesized 10¹² peptidomimetics in a 70-μl translation, a fraction of which was analyzed in the experiment depicted in Fig. 2c) when ≈0.01 nmol of each aa-tRNA is incorporated (the typical research scale for aa-tRNA preparation is 1 nmol). Recycling of aa-tRNAs is also plausible by extrapolation of generalizable methods for engineering new synthetase specificities (8, 40, 41). Our approach therefore provides another route to create libraries for the discovery of small-molecule ligands or stereospecific catalysts (42), complementing other potential approaches for generating genetically encoded degradation-resistant lead compounds, such as mirror-image ligand display (43, 44) and nonribosomal peptide synthesis (45). The potential attraction of such approaches is that directed Darwinian evolution in vitro (29) is much faster than the many person-years of random chemical syntheses typically needed in industry for lead optimization, and their unrivalled library sizes may produce ligands with higher affinities. It is even conceivable that such selections could yield drug candidates directly (compare the orally available, 11-residue cyclic peptide cyclosporin A) when building blocks such as N-methyl amino acids are chosen to encode pharmacologically desirable properties such as protease resistance and membrane permeability.