Optimization of Replication, Transcription, and Translation in a Semi-Synthetic Organism

Abstract

Previously, we reported the creation of a semi-synthetic organism (SSO) that stores and retrieves increased information by virtue of stably maintaining an unnatural base pair (UBP) in its DNA, transcribing the corresponding unnatural nucleotides into the codons and anticodons of mRNAs and tRNAs, and then using them to produce proteins containing non-canonical amino acids (ncAAs). Here we report a systematic extension of the effort to optimize the SSO by exploring a variety of deoxy- and ribonucleotide analogs. Importantly, this includes the first in vivo structure-activity relationship (SAR) analysis of unnatural ribonucleoside triphosphates. Similarities and differences between how DNA and RNA polymerases recognize the unnatural nucleotides were observed, and remarkably, we found that a wide variety of unnatural ribonucleotides can be efficiently transcribed into RNA and then productively and selectively paired at the ribosome to mediate the synthesis of proteins with ncAAs. The results extend previous studies, demonstrating that nucleotides bearing no significant structural or functional homology to the natural nucleotides can be efficiently and selectively paired during replication, to include each step of the entire process of information storage and retrieval. From a practical perspective, the results identify the most optimal UBP for information storage, as well as the most optimal unnatural ribonucleoside triphosphates for its retrieval. The optimized SSO is now, for the first time, able to efficiently produce proteins containing multiple, proximal ncAAs.

Graphical Abstract

INTRODUCTION

In all natural organisms, information is encoded with the four-letter genetic alphabet, consisting of deoxyadenosine (dA), deoxyguanosine (dG), deoxycytidine (dC), and deoxythymidine (dT), with the storage and retrieval of this information made possible by the formation of two base pairs, (d)A-dT/U and (d)G-(d)C (where “(d)” indicates that the nucleotides may be either deoxyribo- or ribonucleotides). Over 100 years ago, the newly defined field of synthetic biology set its central goal as the creation of new forms and functions,¹ and perhaps the most general route to this goal is to increase the information that a cell can store and retrieve. Correspondingly, the last decade has seen significant effort and progress toward the identification of a fifth and sixth nucleotide that pair to form a third, unnatural base pair (UBP).^2–4

Our efforts have focused on the development of UBPs formed between synthetic nucleotide analogs with predominantly hydrophobic nucleobases that pair via hydrophobic and packing forces, as opposed to the complementary hydrogen bonding used by natural nucleobases. Through a medicinal chemistry-like approach, based on the elucidation of structure-activity relationships (SARs) from the evaluation of over 150 nucleotide analogs,⁵ we discovered a family of UBPs that, when incorporated into DNA, are well replicated by DNA polymerases in vitro. Within this family, the UBPs formed between the synthetic nucleotides dNaM and either d5SICS or dTPT3 (dNaM-d5SICS and dNaM-dTPT3, respectively; Figure 1) have received the most attention and are amongst the most promising. Towards the goal of creating new forms and functions, we have used these UBPs as the basis of a semi-synthetic organism (SSO). The SSO, a strain of Escherichia coli constitutively expressing a nucleoside triphosphate transporter from Phaedactylum tricornutum (PtNTT2),⁶ imports the requisite unnatural nucleoside triphosphates from the media and uses them to replicate DNA containing the UBP, transcribe mRNA and tRNA containing the unnatural nucleotides, and then use the resulting cognate unnatural codons and anticodons to translate proteins containing non-canonical amino acids (ncAAs).^7–12 In addition, the forces underlying the pairing of the unnatural nucleotides, as well as their physical properties have been explored by others.^13–19

Figure 1.

The dNaM-d5SICS, dNaM-dTPT3, dCNMO-dTPT3, and dPTMO-dTPT3 UBPs.

Although dNaM-d5SICS, and especially dNaM-dTPT3, are sufficiently well replicated in the SSO for the stable propagation of increased genetic information, they are still replicated less efficiently than a natural base pair, which has motivated continued chemical optimization. These efforts have identified several promising candidates, in particular dCNMO and dPTMO, which when paired with dTPT3, form UBPs that are better retained in the DNA of the SSO (Figure 1).^10,20 In addition, we have also explored the ability of the SSO to retrieve the information encoded with the dPTMO-dTPT3 UBP via transcription and translation, and interestingly, found that its use results in more efficient expression of unnatural proteins than dNaM-dTPT3.¹⁰ This result clearly motivates elucidation of the SARs governing the templating of transcription. In addition, only the NaMTP and TPT3TP ribonucleoside triphosphates have been used to retrieve the increased information in the SSO. Given that these analogs were designed based on replication SARs, which may or may not be the same as those governing efficient transcription and translation, it remains unclear if they are optimal.

Here, we explore the ability of the unnatural information encoded by UBPs formed by dTPT3 and either dNaM or a dNaM analog, to be retrieved in the form of proteins containing ncAAs. We also report the first systematic evaluation of the efficiency with which this information is retrieved using thirteen analogs of NaMTP or TPT3TP, eleven of which are novel. The results identify an improved system of unnatural deoxy- and ribonucleotides that form the basis of an SSO that more efficiently stores and retrieves increased information, and for the first time, permits the efficient production of protein bearing multiple, proximal ncAAs.

RESULTS

Replication and templating of transcription.

While we have examined the in vivo replication of a wide variety of our UBPs, retention in a tRNA gene or in any actively transcribed gene has to date only been examined with dNaM, dPTMO, dMTMO, and dTPT3.^9–10 To extend these studies, we first constructed plasmids with two dNaM-dTPT3 UBPs, such that the sequence AXC (here and throughout X refers to (d)NaM or a (d)NaM analog) was positioned to template codon 151 of sfGFP mRNA (sfGFP¹⁵¹(AXC)) and with the sequence GYT (here and throughout Y refers to (d)TPT3 or a (d)TPT3 analog) positioned to template the anticodon of the M. mazei Pyl tRNA (tRNA^Pyl(GYT)), which is selectively charged with the ncAA N⁶-(2-azidoethoxy)-carbonyl-l-lysine (AzK) by the M. barkeri pyrrolysyl-tRNA synthetase (Mb PylRS).^21–25 These plasmids were used to transform E. coli expressing the nucleoside triphosphate transporter PtNTT2 (strain YZ3⁸) and harboring a plasmid encoding Mb PylRS. After transformation, colonies were selected and grown to an OD₆₀₀ ~ 1.0 in liquid media supplemented with dTPT3TP (10 µM) and one of seven different dXTPs (Figure 2) added at varying concentrations (150 µM, 10 µM, or 5 µM). Cells were then diluted into fresh expression media containing the same unnatural deoxyribonucleotides as well as NaMTP (250 µM), TPT3TP (30 µM), and AzK (10 mM). After a brief incubation, T7 RNAP and tRNA^Pyl(GYT) expression was induced by the addition of isopropyl-β-d-thiogalactoside (IPTG, 1 mM). After an additional 1 h of incubation, the expression of sfGFP¹⁵¹(AXC) was initiated by the addition of anhydrotetracycline (aTc, 100 ng/mL).

Figure 2.

dXTP analogs. Ribose and phosphates omitted for clarity.

After 2.5 h, plasmids were isolated and then the genes of interest were independently PCR amplified using d5SICSTP and dMMO2^BIOTP, a biotinylated analog of dNaMTP.²⁶ The resulting PCR product was analyzed via a gel mobility shift assay using streptavidin to quantify the UBP retained as a percent shift of the total amplified product (hereafter referred to as the streptavidin gel shift assay) (Figure 3A and 3B). At the highest dXTP concentration examined (150 µM), retention in the sfGFP¹⁵¹(AXC) gene was similar for each dXTP analog, varying from 99% for dNaMTP to 92% for dMTMOTP. Retention within the tRNA^Pyl(GYT) gene was slightly lower, ranging from 82% for dMTMOTP to 74% for dMMO2TP. With the addition of 10 µM of each dXTP, retention in the sfGFP¹⁵¹(AXC) gene ranged from 96% for d5FMTP to 73% for dNaMTP. With the tRNA^Pyl(GYT) gene, retentions were again generally slightly lower, ranging from 82% for d5FMTP and dPTMOTP to 71% for dNaMTP. At the lowest concentration (5 µM), retention ranged from 94% for d5FMTP to 64% for dMTMOTP in the sfGFP¹⁵¹(AXC) gene, and from 83% for d5FMTP to 71% for dMTMOTP in the tRNA^Pyl(GYT) gene. When the same concentrations were employed, the results with dNaMTP, dPTMOTP, or dMTMOTP are indistinguishable from those previously reported, and also as reported previously, cells grown with 5 µM dNaMTP did not survive.¹⁰ Generally, at the highest concentration, all dXTPs performed well with nearly quantitative retention of the UBP within the sfGFP gene. However, at lower concentrations it became increasingly evident that dCNMO-dTPT3, d5FM-dTPT3, and dPTMO-dTPT3 were replicated with significantly higher retention. Interestingly, retention in the tRNA gene was significantly less dependent on dXTP concentration.

Figure 3.

Optimization of translation to incorporate AzK into sfGFP using various dXTPs. (A) UBP retention (%) in the sfGFP gene. (B) UBP retention (%) in the tRNA^Pyl gene. (C) Relative sfGFP fluorescence observed normalized to cell growth (Relative Fluorescence Units (RFU) per OD₆₀₀) in the presence or absence of AzK. (D) Protein shift (%) measured by western blot. Each bar represents the mean, with error bars indicating standard error (n = 3). Open circles represent data for each independent trial. Asterisk indicates that cells were unable to grow under the condition indicated.

To characterize the amount of protein produced, bulk culture fluorescence normalized to cell growth was measured 2.5 h after the induction of protein expression (Figure 3C). In the absence of AzK, fluorescence was generally low, with the exceptions of dPTMOTP and dMTMOTP at the lower concentrations, which appeared slightly higher. When AzK was added to the media, cells grown with each dXTP at either 150 or 10 µM generally showed significant and similar levels of fluorescence, with the exceptions of dNaMTP, which showed significantly less fluorescence at 10 µM than at 150 µM. At the lowest concentration (5 µM), the addition of AzK resulted in less fluorescence observed with dMTMOTP while it remained the same with dCNMOTP, d5FMTP, dClMOTP, dMMO2TP, or dPTMOTP. Again, as mentioned above, cells were unable to grow when provided with only 5 µM dNaMTP. Remarkably, the data indicate that while similar at high concentrations, at lower concentrations the use of each dXTP analog results in a greater AzK-dependent increase in fluorescence than does the use of dNaMTP. In particular, dCNMOTP, d5FMTP, and dClMOTP showed the greatest AzK-dependent increase in fluorescence, and were thus considered the most promising.

To directly assess the fidelity of unnatural protein production, cells were harvested 2.5 h after the induction of protein expression and the sfGFP produced was purified and subjected to a strain-promoted azide-alkyne cycloaddition reaction²⁷ with dibenzocyclooctyne (DBCO) linked to a TAMRA dye by four PEG units. In addition to tagging the proteins containing the ncAA with a detectable fluorophore, conjugation produces a shift in electrophoretic migration, allowing quantification of protein containing AzK as a percentage of the total protein produced (i.e. fidelity of ncAA incorporation; Figure 3D).¹⁰ As with dNaM, use of each dXTP at the highest concentration (150 µM) resulted in virtually complete shifts of the purified protein, reflecting high fidelity incorporation of the ncAA. When grown in the presence of 10 µM of the dXTP, the fidelity of ncAA incorporation remained high for dCNMOTP, d5FMTP, dClMOTP, dMMO2TP, dPTMOTP, and dMTMOTP, but dropped precipitously for dNaMTP. Finally, at a concentration of 5 µM dXTP, the fidelity of ncAA incorporation remained similar to that observed at 10 µM for dCNMOTP, d5FMTP, dClMOTP, dMMO2TP, and dPTMOTP, but dropped for dMTMOTP (again due to viability, fidelity with dNaMTP could not be measured at this concentration).

The most promising members of the family that produced the greatest quantity of pure unnatural protein, especially at lower concentrations, are dCNMOTP and d5FMTP. However, relative to d5FMTP, the use dCNMOTP has been previously shown to result in higher UBP retention in more difficult to replicate sequences²⁰. Thus, we conclude that the dCNMO-dTPT3 UBP is the most optimized for the storage and retrieval of increased information.

Synthesis and first SAR analysis of ribonucleotide candidates.

Retrieval of the information made available by the UBP has only previously been explored using NaMTP and TPT3TP. To begin to elucidate the SARs governing efficient transcription and translation in the SSO, we designed and synthesized nine novel NaMTP analogs (Figure 4A) and four novel TPT3TP analogs (Figure 4B). These analogs were designed to explore the role of nucleobase shape, aromatic surface area, and heteroatom derivatization. Generally, the synthesis of the XTP analogs proceeded via lithiation of the corresponding aryl halide, followed by coupling of the lithiated species to either the benzyl- or TBS-protected ribolactone. Reduction of the resulting hemi-acetal intermediate in the presence of boron trifluoride diethyl etherate and triethylsilane afforded the desired protected nucleoside in each case. Following deprotection, the resulting X nucleoside analogs were converted to triphosphates using standard Ludwig phosphorylation conditions (see Supporting Information).²⁸ NaMTP and MMO2TP were synthesized as reported previously.²⁹ The synthesis of the YTP analogs generally proceeded via intramolecular Curtius rearrangement of the corresponding acyl azide, followed by Lewis-acid mediated coupling to 1-O-acetyl-2,3,5-tri-O-benzoyl-β-d-ribofuranose, resulting in pure β-anomer of the desired protected nucleoside. Following conversion of the pyridine to the corresponding thio-pyridone and subsequent benzoyl deprotection, the corresponding free nucleosides were converted to triphosphates using standard Ludwig phosphorylation conditions (see Supporting Information).²⁸ 5SICSTP was synthesized as reported previously.²⁹

Figure 4.

Ribonucleotide analogs. (A) XTP analogs. (B) YTP analogs. Ribose and phosphates omitted for clarity.

We initiated our SAR analysis with NaMTP and the nine XTP analogs. Based on performance in the dXTP screen described above, and to eliminate variable loss at the DNA level as a complicating factor, we encoded the unnatural information with the dCNMO-dTPT3 UBP. Additionally, we used our recently reported E. coli strain ML2,¹¹ which expresses the nucleoside triphosphate transporter PtNTT2, but has also been genetically engineered for higher fidelity replication of the UBP by deletion of the gene encoding RecA (as is common in cloning strains) and overexpression of DNA Pol II. The same plasmid described above, harboring sfGFP¹⁵¹(AXC) and tRNA^Pyl(GYT), was used, and to focus the screen to a single unnatural ribonucleotide, the M. mazei Pyl tRNA was transcribed in the presence of 30 µM TPT3TP.

Cells were grown and induced to produce protein as described above, except that the various XTPs were provided at either high (250 µM) or low (25 µM) concentration in the expression media (Figure 5). Expressed sfGFP was purified 3 h after induction and ncAA content analyzed using the DBCO-mediated gel shift assay described above (Figure 5B). Remarkably, the use of each XTP at 250 µM resulted in sfGFP gel shifts of at least 63%. Along with the lack of observable shift in the absence of an XTP, this demonstrates that each XTP is imported by PtNTT2 and participates in transcription and translation at the ribosome with at least reasonable efficiency. While CNMOTP and 5F2OMeTP each performed well, resulting in gel shifts of 92% and 94%, respectively, NaMTP, MMO2TP, and 5FMTP performed the best, with protein gel shifts of 98%, 97%, and 98%, respectively. With similarly high levels of protein purity, we were able to compare the relative fluorescence produced using these three XTPs in the absence of any complications resulting from significant levels of natural sfGFP contaminant. Cells grown with 250 µM of either MMO2TP or 5FMTP produced 63% and 90% bulk fluorescence, respectively, compared to cells grown with NaMTP at the same concentration (Figure 5A).

Figure 5.

SAR analysis of translation using various unnatural ribonucleotides to incorporate AzK into sfGFP. (A) Total sfGFP fluorescence (RFU) observed in the presence of AzK for XTP analogs, where (−) represents control samples provided with only TPT3TP (XTP withheld). (B) Protein shift (%) measured by western blot for XTP analogs, where (−) represents control samples provided with only TPT3TP (XTP withheld). (C) Total sfGFP fluorescence (RFU) observed in the presence of AzK for YTP analogs, where (−) represents control samples provided with only NaMTP (YTP withheld). (D) Protein shift (%) measured by western blot for YTP analogs, where (−) represents control samples provided with only NaMTP (YTP withheld). Each bar represents the mean, with error bars indicating standard error (n = 4). Open circles represent data for each independent trial.

At the lower concentration tested (25 µM), the use of NaMTP resulted in a lower fidelity of ncAA incorporation, with a protein shift that dropped to 86%. While 7 of the 9 NaMTP analogs also showed significant decreases in fidelity, the use of MMO2TP and 5FMTP each yielded protein shifts of 94%. Comparing the relative fluorescence of cells grown with these ribonucleotides (again possible due to the similar and high fidelity of unnatural protein produced), 5FMTP produces 34% more fluorescence than MMO2TP at this lower concentration.

We next performed similar experiments, except we supplied a constant amount of NaMTP (250 µM) and either a high (250 µM) or low (25 µM) concentration of a YTP analog. Consistent with previous reports⁹, the addition of TPT3TP at the higher concentration resulted in significantly reduced cell growth and little fluorescence relative to the control sample that did not receive a YTP (Figure 5C). In contrast, each of the other YTP analogs produced fluorescence above background, and protein shifts of at least 51% (Figure 5D). In particular, TAT1TP performed the best at this concentration, with its use resulting in at least 2.6-fold more fluorescence than any other YTP, while maintaining a protein shift of 96%. However, the addition of TAT1TP at this concentration did result in a modest level of reduced cell growth (Figure S1).

Consistent with previous reports,^9–10 TPT3TP is somewhat less toxic when provided at lower concentrations, and correspondingly, when provided at 25 µM, cells produce significant quantities of pure protein. Under these conditions, TPT3TP is more optimal than SICSTP, FSICSTP, and 5SICSTP, producing 2-, 5-, and 6-fold greater fluorescence, respectively. Interestingly, the toxicity observed with TAT1TP at the higher concentrations was almost completely eliminated at lower concentrations (Figure S1), and its use resulted in even greater fluorescence than at the higher concentration. When provided at 25 µM, TAT1TP produces 41% more fluorescence than when provided at 250 µM, and interestingly, it produced 57% more fluorescence than when TPT3TP is provided at 25 µM. Most importantly, the use of TAT1TP at these concentrations resulted in the production of protein with 98% ncAA incorporation.

Optimization of unnatural protein production.

The screens described above identified dCNMO-dTPT3 as the most promising UBP for the storage of information, and TAT1TP and NaMTP or 5FMTP as the most promising ribonucleoside triphosphates for its retrieval. Thus, we next turned to exploring the concentrations of each ribonucleotide used to optimize the yield and fidelity of protein expression in the SSO. Upon transformation with the same plasmids used above, 10 µM dTPT3TP and 25 µM dCNMOTP were provided in the growth media, which was then also supplemented with TAT1TP at concentrations ranging from 100 µM to 12.5 µM, and either NaMTP or 5FMTP at concentrations ranging from 200 µM to 12.5 µM, all in series of 2-fold dilutions, and after the addition of 1 mM AzK, the cells were induced to express sfGFP.

Total sfGFP fluorescence observed in cells provided with TAT1TP and NaMTP was generally higher than that observed in cells provided with TAT1TP and 5FMTP (Figure 6A and 6B). In both cases, fluorescence was higher at lower concentrations of NaMTP or 5FMTP (due to increased production of contaminating natural sfGFP, see below). Additionally, cells generally produced higher fluorescence at higher concentrations of TAT1TP. However, due to a slight reduction in growth, cells provided with 100 µM TAT1TP produced less fluorescence than those provided with 50 µM TAT1TP.

Figure 6.

Optimization of unnatural ribonucleotide triphosphate concentrations. (A) Total sfGFP fluorescence (RFU) as a function of the concentrations of NaMTP and TAT1TP (µM). (B) Total sfGFP fluorescence (RFU) as a function of the concentrations of 5FMTP and TAT1TP (µM). (C) Protein shift (%) as a function of the concentrations of NaMTP and TAT1TP (µM). (D) Protein shift (%) as a function of the concentrations of 5FMTP and TAT1TP (µM). Error bars indicate standard error of each value (n = 3).

Protein production was again quantified via the gel shift assay (Figure 6C and 6D). Generally, as the concentration of NaMTP decreased below 200 µM, incrementally lower fidelity of AzK incorporation into sfGFP was observed, while with use of 5FMTP this reduction in fidelity was only observed below a concentration of 50 µM. Clearly lower concentrations of 5FMTP can be used without compromising fidelity. Cells provided with a high concentration of 5FMTP (≥ 50 µM) produced high protein shifts at all concentrations of TAT1TP explored (100 µM, 50 µM, 25 µM, or 12.5 µM). However, when the concentration of 5FMTP was 25 µM or less, decreasing the concentration of TAT1TP resulted in a reduced protein shift. When NaMTP was provided at 200 µM, all concentrations of TAT1TP explored resulted in the production of protein with high fidelity ncAA incorporation, but with lower concentrations of NaMTP, decreasing the concentration of TAT1TP again resulted in a decreased protein shift.

In all, these studies revealed that the combined optimization of protein purity and yield is achieved with NaMTP and TAT1TP provided at concentrations of 200 µM and 50 µM, respectively, or with 5FMTP and TAT1TP both provided at a concentration of 50 µM. In terms of protein production alone, the use of NaMTP and TAT1TP is optimal, whereas the use of 5FMTP and TAT1TP results in slightly lower yields of pure ncAA-labeled protein, but requires significantly lower concentrations of the XTP.

Storage and retrieval of higher density unnatural information.

With optimized unnatural nucleotides and conditions, we next sought to challenge the SSO by examining the storage and retrieval of information from a gene containing a higher density of UBPs. Towards this goal, we first validated the ability of the SSO to replicate DNA containing the sfGFP gene with the UBP positioned to encode codons 149 or 153, which are each separated from the codon described above (codon 151) by a single natural codon. Accordingly, expression plasmids were constructed, as described above, but in which the sequence AXC was positioned to encode either codon 149 or 153 (sfGFP¹⁴⁹(AXC) or sfGFP¹⁵³(AXC), respectively). Upon transformation of ML2, cells were grown in the presence of unnatural nucleoside triphosphates, corresponding to either our previously reported system (the deoxyribonucleotides dNaMTP and dTPT3TP and the ribonucleotides NaMTP and TPT3TP, denoted dNaM-dTPT3/NaMTP,TPT3TP),^8–9 or the optimized system discovered in the current study (dCNMO-dTPT3/NaMTP,TAT1TP). UBP retention was then characterized using the streptavidin gel shift assay, as described above. High retention of the corresponding UBP in both sfGFP¹⁴⁹(AXC) and sfGFP¹⁵³(AXC) genes (≥95%) as well as in the tRNA gene (≥91%) was observed under both conditions (Table S1).

Total sfGFP fluorescence observed 3 h after induction revealed significant protein production from both constructs in the presence of AzK under both sets of conditions (Figure 7A). However, fluorescence from cells expressing the sfGFP¹⁴⁹(AXC) construct provided with dCNMO-dTPT3/NaM,TAT1 was 58% higher than cells provided with dNaM-dTPT3/NaM,TPT3. In the case of the sfGFP¹⁵³(AXC) gene, 43% more fluorescence was observed with dCNMO-dTPT3/NaMTP,TAT1TP than with dNaM-dTPT3/NaMTP,TPT3TP. Under both sets of conditions, approximately 2-fold more fluorescence was observed with sfGFP¹⁵³(AXC) than with sfGFP¹⁴⁹(AXC). Protein was purified and AzK incorporation was analyzed as described above (Figure 7B). With dNaM-dTPT3/NaMTP,TPT3TP, protein shifts of 86% and 94% were observed with sfGFP¹⁴⁹(AXC) and sfGFP¹⁵³(AXC), respectively. With dCNMO-dTPT3/NaMTP,TAT1TP, however, a 96% shift was observed with protein produced from either construct. These results clearly demonstrate that the two additional codon positions are both transcribed and translated efficiently, but again they are transcribed and translated better with the newly identified dCNMO-dTPT3/NaMTP,TAT1TP system.

Figure 7.

Storage and retrieval of higher density unnatural information with either dNaM-dTPT3/NaMTP,TPT3TP or dCNMO-dTPT3/NaMTP,TAT1TP. (A) Total sfGFP fluorescence (RFU) observed in the presence of AzK. (B) Protein shift (%) measured by western blot. For strip charts, each bar represents the mean, with error bars indicating standard error (n = 4), and open circles represent data for each independent trial. (C) Representative spectrum of quantitative HRMS analysis of triple labeled protein produced using the dCNMO-dTPT3/NaMTP,TAT1TP. Peak labels show deconvoluted molecular weight of intact protein, with amino acid residues at positions 149, 151, and 153 shown and quantification of each peak (%, n = 3) shown below. See Supporting Information for assignment of unlabeled peaks.

We next constructed expression plasmids with an unnatural codon simultaneously encoded at two or all three of the positions examined (sfGFP^149,151(AXC,AXC), sfGFP^151,153(AXC,AXC), sfGFP^149,153(AXC,AXC), and sfGFP^149,151,153(AXC,AXC,AXC), respectively). ML2 cells were transformed, grown in the presence of either dNaM-dTPT3/NaMTP,TPT3TP or dCNMO-dTPT3/NaMTP,TAT1TP, and protein expression was induced as described above. While UBP retention in the tRNA^Pyl(GYT) gene remained high (≥88%) in all cases (Table S1), the biotin shift assay with the mRNA genes produced complex and uninterpretable band patterns, likely due to, at least in part, the formation of a mixture of complexes with single PCR products bound to multiple streptavidins. Thus, we proceeded to analyze the protein produced via conjugation to DBCO-TAMRA as described above (Figure 7B). Gratifyingly, relative to the shift observed with a single ncAA, a significantly further shifted band was observed for proteins expressed from the sfGFP^149,151(AXC,AXC), sfGFP^151,153(AXC,AXC), and sfGFP^149,153(AXC,AXC) constructs, indicating the conjugation of two DBCO-TAMRA molecules to sfGFP bearing two AzK residues. When analyzing purified proteins expressed with dNaM-dTPT3/NaMTP,TPT3TP, quantification of these double shifted bands relative to total sfGFP revealed that 80%, 87%, and 83% of the protein, respectfully, had two AzK residues, and 20%, 13%, or 9% , respectfully, had a single AzK when using the sfGFP^149,151(AXC,AXC), sfGFP^151,153(AXC,AXC), or sfGFP^149,153(AXC,AXC) constructs, respectively. With dCNMO-dTPT3/NaMTP,TAT1TP, 81%, 89%, and 93% of the protein had two ncAAs with sfGFP^149,151(AXC,AXC), sfGFP^151,153(AXC,AXC), and sfGFP^149,153(AXC,AXC), respectively, while 19%, 11%, and 6% had a single ncAA. Cells transformed with the sfGFP^149,151,153(AXC,AXC,AXC) construct expressed protein that produced an even further shifted band, clearly indicating the incorporation of three AzK residues. Quantification of each band relative to total sfGFP revealed that the use of dNaM-dTPT3/NaMTP,TPT3TP resulted in 39%, 24%, and 33% of the protein having three, two and one ncAAs, respectively, and with fluorescence and protein shifts that were highly variable (Figure 7A and 7B). In contrast, use of the dCNMO-dTPT3/NaMTP,TAT1TP system resulted in 90% of the produced protein having all three ncAAs, with the remainder having two.

To further verify the successful incorporation of all three ncAAs with the dCNMO-dTPT3/NaMTP,TAT1TP system, we analyzed the isolated sfGFP by quantitative intact protein mass spectrometry. Briefly, purified proteins were desalted using centrifugal filter devices (Amicon® Ultra-0.5 – Millipore), and analyzed by HRMS (ESI-TOF). The mass spectra acquired were subsequently deconvoluted using the Waters MaxEnt 1 software, which proved to be quantitative upon peak integration (Figure S2). In agreement with the gel shift assay, this analysis revealed that that 88% of the isolated protein contained the expected three AzK residues, while the remaining 12% contained two AzK residues and a single Ile or Leu residue (Figure 7C).

DISCUSSION

The previously reported SSOs stored information with the UBPs dNaM-dTPT3, dPTMO-dTPT3, or dMTMO-dTPT3, and retrieved that information using NaMTP and TPT3TP. To explore the optimization of the SSO, we examined retention of the UBP, transcription into sfGFP mRNA and tRNA^Pyl, and decoding at the ribosome, using a collection of previously and newly reported deoxy- and ribonucleotide triphosphate analogs. We first examined the ability to store information with seven different dX-dTPT3 UBPs. In each case, the strand context of the UBP was the same, with dTPT3 and dX positioned in the corresponding antisense (template) strands of the sfGFP and tRNA^Pyl genes, respectively. With high concentrations of each dXTP provided, each dX-dTPT3 UBP is retained at a high level in the mRNA gene, with variation between 92% for dMTMOTP and 96% to 99% for dCNMOTP, dPTMOTP, and dNaMTP. Retentions in the tRNA gene were somewhat reduced, varying between 74% to 82%. As the concentrations of dXTP decreased, retentions remained roughly constant in the tRNA gene, but decreased in a dXTP specific manner in the mRNA gene, decreasing to 64% for dMTMOTP, but remaining relatively high, at ~94%, for dCNMOTP and d5FMTP. The different concentration dependencies for retention in the tRNA and mRNA genes likely result from sequence context effects causing nucleotide insertion to be rate limiting in the mRNA and continued extension to be rate limiting in the tRNA gene. Exceptions were observed with dNaMTP, where at 10 μM retention decreased to 73%, and as reported previously¹⁰ the cells did not survive when dNaMTP was provided at 5 μM. In addition, with dCNMOTP and d5FMTP, retentions remained high in the mRNA (~93%) at even the lowest concentration examined. Loss of retention in the tRNA gene could result in less unnatural protein production, and perhaps more problematically, reduced fidelity of ncAA incorporation due to increased competition for decoding of the unnatural codon by “near-cognate” natural tRNAs. However, the fidelity of ncAA incorporation is clearly correlated with retention in the mRNA gene (Figure S3). Thus, the data demonstrates that each dX templates transcription of tRNA with sufficient efficiency and fidelity to not limit the fidelity of unnatural protein production. Based on these results, as well as those previously published²⁰, d5FM-dTPT3, and especially dCNMO-dTPT3 are the most optimized UBPs, with their utility relative to dNaM-dTPT3 deriving principally from their higher retention and protein production at lower unnatural triphosphate concentrations.

We next examined the ability of ten different XTPs to mediate the retrieval of information stored by the dCNMO-dTPT3 UBP. Remarkably, all ten XTPs explored at high concentration are capable of mediating the production of proteins with at least moderate ncAA incorporation fidelity (98% to 63%). Generally, as XTP concentrations are decreased, the fidelity of ncAA incorporation decreased, indicating a reduction in the fidelity with which the unnatural mRNA is transcribed. This is consistent with the significant fluorescence observed when XTP was withheld.

The current study provides the first SARs for the transcription and translation of predominantly hydrophobic ribonucleotides. With the XTPs, when considering NaMTP, PTMOTP, and MTMOTP, it is clear that ring contraction and/or heteroatom derivatization is deleterious. However, with the monocyclic nucleobase XTPs, with the exception of ClMOTP, higher fidelity protein production is observed, relative to MTMOTP and PTMOTP, suggesting that the effects are more complicated than just aromatic surface area or heteroatom derivatization. It is likely that specific interactions between the unnatural nucleobases or with the polymerase are critical. Substitution at both the 4- and 5-positions of the monocyclic nucleobases has significant effects. Compared to 2OMeTP, a Cl or Br substituent at the 4-position (ClMOTP and BrMOTP) is modestly deleterious, while a methyl group at the 4-position (MMO2TP), reduces overall fluorescence, but with a significant increase in protein fidelity. A nitrile substituent at the 4-position (CNMOTP) results in the production of the greatest amount of unnatural protein, and also modestly increases the fidelity with which the ncAA is incorporated. Addition of a fluoro substituent at the 5-position (5F2OMeTP) also increases both protein production and fidelity, relative to 2OMeTP. The effects of substitution at the 4- and 5- positions appear at least approximately additive, as combining the 5-fluoro and 4-methyl substituents (5FMTP) allows for the high yield production of pure unnatural protein at lower concentrations, relative to the other unnatural triphosphates. However, while requiring higher concentrations, NaMTP provides the most optimal combination of yield and purity of the XTP analogs examined.

It is interesting to note that the SARs derived for the XTP analogs are distinctly different from those derived from the replication of dXTP analogs. For example, while dPTMOTP, dClMOTP, and especially dCNMO are more optimal than dNaMTP, ClMOTP and PTMOTP are modestly and significantly less optimal than NaMTP. Moreover, while dCNMO is the most optimal partner for dTPT3 discovered to date, the use of CNMOTP results in slightly reduced fidelity of ncAA incorporation, relative to NaMTP, although its use does result in the most unnatural protein production.

At the highest concentration, all five YTPs explored were effectively incorporated into the anticodon of tRNA^Pyl and capable of mediating the production of proteins with at least moderate ncAA incorporation fidelity (98% to 51%). However, the UBP retention data suggests that the fidelity of protein production is not sensitive to modest loss of unnatural tRNA, implying that for the YTPs that resulted in lower protein gel shifts, transcription of the tRNA was either very inefficient or low fidelity. TPT3TP, SICSTP, FSICSTP, and TAT1TP were all, at least slightly toxic at the highest concentration (Table S2), and bulk cell fluorescence increased with decreasing concentrations. Unlike with XTPs and transcription of the mRNA, fidelity of unnatural protein did not decrease, suggesting that the increased protein production was simply the result of increased cell growth. In contrast, 5SICSTP is not toxic (Table S2), and both unnatural protein production and fidelity decreased with decreasing concentrations, again presumably due to significantly compromised unnatural tRNA production.

When considering this YTP SAR, and using SICSTP as a reference, it is clear that a 7-fluoro substituent (FSICSTP) is quite deleterious, virtually ablating protein production, either due to effects on tRNA transcription or on translation, and only a small amount of protein is produced and with low fidelity. Addition of a methyl group at the 5-position (5SICSTP) reduces toxicity, but also appears to significantly reduce unnatural tRNA production. Ring contraction and heteroatom derivatization (TPT3TP) greatly improves both protein production and fidelity, but at high concentration it is the most toxic of the YTP analogs. Further heteroatom derivatization of the thiophene ring of TPT3TP, to produce the thioazole of TAT1TP, results in the production of even more pure unnatural protein than does TPT3TP, and importantly, with a substantial reduction in toxicity. Interestingly, unlike the case with the XTP SARs discussed above, the YTP SARs are relatively similar to those characterized with dYTP analogs.⁵ Given both the amount of protein produced and the fidelity of ncAA incorporation, TAT1TP is the most promising YTP identified to date.

Overall, the SARs identify the dCNMO-dTPT3/NaMTP,TAT1TP system as the most optimized for protein production, producing high amounts of protein with high fidelity incorporation of the ncAA. The use of dCNMO-dTPT3/5FMTP,TAT1TP produces protein with the same high fidelity, and while it produces slightly less protein, it requires the use of significantly less of the unnatural ribonucleotides. The utility of the dCNMO-dTPT3/NaMTP,TAT1TP system, relative to the previously reported dNaM-dTPT3/NaMTP,TPT3TP system, is particularly apparent with the encoding and retrieval of higher density unnatural information. As would be expected based on the results of the single labeled proteins, both systems produced protein with two ncAAs with high fidelity, but the dCNMO-dTPT3/NaMTP,TAT1TP system generally produced the desired protein in greater quantities. Moreover, when encoding three ncAAs, the dNaM-dTPT3/NaMTP, TPT3TP system produced triply labeled protein with significantly reduced and more variable fidelities and yields, while the fidelity and yield with the dCNMO-dTPT3/NaMTP,TAT1TP system remained reproducibly high. The contaminant, where an Ile or Leu replaced a single ncAA, is unlikely to result from unnatural tRNA production, as UBP retention in the tRNA gene was high and similar for the both systems, and even with small differences, the single ncAA-incorporation data suggest that they should not cause significant reductions in the fidelity of unnatural protein production. It is also unlikely to result from mRNA transcription, which should be identical for the two systems (in both cases the dTPT3 directs the incorporation of NaM into the mRNA). Thus, the origin of the Leu/Ile contaminant is likely to be loss of the UBP in the mRNA gene during replication (which, as mentioned above, we were unable to directly measure). This is also consistent with the most common mutation expected, (dX to dT), which would produce an Ile codon.

CONCLUSION

This study reports the first SARs underlying efficient transcription by T7 RNAP and translation at the ribosome in the E. coli based SSO. It is particularly interesting that the X and dX SAR appear unrelated, while the Y and dY SAR appear more similar; it is possible the replicative polymerase(s), likely Pol III and to a lesser extent Pol II,¹¹ and T7 RNAP recognize distinct aspects of the (d)X nucleobases, but similar aspects of the (d)Y nucleobases. While the origins of this recognition will be pursued in future studies, the results have already identified a more optimal SSO, specifically an SSO that stores information with dCNMO-dTPT3, and retrieves the information it makes available with TAT1TP and NaMTP or 5FMTP. The optimization of the SSO is particularly apparent in its ability to produce protein with a higher density of ncAAs. The optimized SSO further attests to the ability of hydrophobic and packing forces to replace the complementary hydrogen bonds that underlie the storage and retrieval of natural information and represents significant progress towards the creation of a SSO with a fully unrestricted expansion of its genetic alphabet and code.