Double mimicry evades tRNA synthetase editing by toxic vegetable-sourced non-proteinogenic amino acid

Hundreds of non-proteinogenic (np) amino acids (AA) are found in plants and can in principle enter human protein synthesis through foods. While aminoacyl-tRNA synthetase (AARS) editing potentially provides a mechanism to reject np AAs, some have pathological associations. Co-crystal structures show that vegetable-sourced azetidine-2-carboxylic acid (Aze), a dual mimic of proline and alanine, is activated by both human prolyl- and alanyl-tRNA synthetases. However, it inserts into proteins as proline, with toxic consequences in vivo. Thus, dual mimicry increases odds for mistranslation through evasion of one but not both tRNA synthetase editing systems. Non-proteinogenic (np) amino acids in the food chain present challenges for the human translation machinery. Here the authors show that, while AlaRS and ProRS activate toxic np azetidine-2-carboxylic acid (Aze) present in sugar beets and lilies, only the AlaRS editing system rejects Aze.

T he incorporation of amino acids into proteins is initiated by the 20 aminoacyl tRNA synthetases (AARSs) 1, 2 . These enzymes attach each amino acid to its cognate tRNA, which is transferred to the ribosome to be used as building blocks for protein synthesis. Similarities between some standard amino acids are so great that occasional errors in aminoacylation occur and generate mistranslated proteins. To attenuate these mistakes, synthetases evolved an editing activity that either removes misactivated amino acids or hydrolytically clears away the misattached amino acid from a mischarged tRNA. In this way, the accuracy of protein synthesis is tightly controlled 3,4 .
Non-proteinogenic (np) amino acids in the food chain present yet another challenge. Many have structures similar to the 20 standard amino acids and have the potential to be misincorporated into proteins. Little is known about how and if AARSs have evolved mechanisms to limit the incorporation of np amino acids into cellular proteins. Azetidine-2-carboxylic acid (Aze) is a np amino acid found in sugar beets and lilies 5,6 (Fig. 1a). Its consumption by gestating mothers is proposed to be connected with some forms of multiple sclerosis in direct offspring 7,8 . Based on its molecular and structural similarity to alanine (Ala) and proline (Pro) (Aze has a 4-and Pro a 5-member ring (Fig. 1a)), Aze is a dual mimic that might fit into the active site pocket of alanyl-tRNA synthetase (AlaRS) or prolyl-tRNA synthetase (ProRS) and be activated for protein synthesis. Indeed, earlier studies support the ability of Aze to be incorporated into proteins 9,10 .
Here we show that although Aze is activated by both human (Hs) AlaRS and Hs ProRS, it is rejected (>99%) by the AlaRS but not the ProRS editing system and therefore almost exclusively misincorporates into Pro positions of proteins.

Results
Aze fits in the active site of both Hs AlaRS and ProRS. To establish whether Hs AlaRS and ProRS confuse Aze for Ala or Pro or both, we co-crystalized a stable analog of the aminoacyl adenylate of Aze (5′-O-[N-(L-azetidinyl) sulfamoyl] adenosine, Aze-SA) with both synthetases. The co-crystal structure (2.1 Å resolution) of Hs AlaRS with Aze-SA (Supplementary Table 1) was compared with our previous structure with Ala-SA bound to Hs AlaRS (PDB code: 4XEM, 1.28 Å). Significantly, Aze-SA could be superimposed on Ala-SA in the active site pocket of Hs AlaRS (Fig. 1b, c), and interact with the same residues that interact with Ala-SA. Thus, Aze closely mimics Ala in the active site of Hs AlaRS.
Next, we obtained a 2.6 Å co-crystal structure of Aze-SA with Hs ProRS (Supplementary Table 1), which we compared with the previously solved structure of Pro-SA with Thermus thermophilus (Tt) ProRS (PDB code: 1H4S) 11 . Aze-SA fits into the active site of Hs ProRS in almost the same conformation as Pro-SA bound to Tt ProRS (Fig. 1d, e), showing that activated Aze is also a close mimic of Pro in the active site of Hs ProRS. The dual mimicry of Aze, and its ability to 'confuse' two human tRNA synthetases, is established by these crystal structures.
Aze toxicity is only rescued by Pro in vivo. In order to investigate the toxic effects of Aze, we administered Aze to HeLa cell cultures and to zebrafish embryos and measured cell death after 24 h. Indeed, increased Aze concentrations in the cell media led to progressive cell death of HeLa cells (measured by trypan blue staining, Fig. 2a and Supplementary Fig. 1a), and 1 mM Aze injection into zebrafish embryos induced localized cell death detectable with increased fluorescence intensity (stained with acridine orange) in the injected embryos, when compared to uninjected controls (Fig. 2b). Because Aze is similar in structure to both Pro and Ala (Fig. 1a), we separately added each amino acid together with Aze to check whether they can compete with Aze and rescue cell death. Toxicity in HeLa cells from 5 mM Aze was rescued by addition of 1 mM Pro, while 1 mM Ala, valine (Val), and threonine (Thr) did not rescue (Fig. 2c, d, Supplementary Fig. 1b, c). Similarly, we injected 1 mM Aze into zebrafish embryos with either 0.25 mM Pro, Val, Thr, or Ala. As seen with mammalian cells, Pro rescued Aze toxicity, while Ala, Val, and Thr did not (Fig. 2e). Further, in HeLa cells, when we raised Aze to 40 mM, 5 mM Pro provided substantial rescue, while 5 mM Ala did not. When 5 mM Ala was added in addition to 5 mM Pro, a small additional protective effect was observed (Supplementary Fig. 1d).
Aze misincorporates into Pro positions. Next, we tested whether Aze misincorporated into mammalian cell proteins, specifically at codons for Pro or Ala. We transfected HEK 293 cells with a plasmid expressing maltose binding protein (MBP). The protein was expressed (in the presence or absence of Aze in the media), purified, and subjected to mass spectrometry (Fig. 3a) Supplementary  Fig. 2). Aze was also found at one Ala position at a frequency of 0.024%. For the Pro → Aze replacements, the substitutions were randomly scattered across the sequence. However, we noted that some positions were more susceptible to replacement than others (Table 1). For example, Pro 255 was replaced by Aze at a frequency of 6/125, while Pro 258 was replaced by Aze at a ratio of 35/147. Possibly, Aze fits into some places in the sequence better   than others, with the poorer fits resulting in protein instability and destruction.
Aze is activated by both Hs AlaRS and ProRS. Given the clear dual Ala, Pro mimicry of Aze, the above observations raised the question of why Aze was incorporated specifically at positions for Pro, and not also for Ala. For this purpose, we studied the activation of Aze by both enzymes. For all the experiments with Hs ProRS, recombinant protein corresponding to only the ProRS portion of the natural human glutamyl-prolyl-tRNA synthetase (GluProRS) fusion protein was used. Consistent with observations of the crystal structures, both enzymes activated Aze ( Fig. 4a-c; no contamination with other amino acids was detected above 0.01% in our Aze stock.) However, at the same concentration of amino acid (1 mM), activation of Aze by Hs ProRS (catalytic efficiency (k cat K m −1 )) was 71-fold slower than with Pro (Supplementary Table 2), and much more robust than with Ala, an amino acid known to be misactivated by ProRS 12 (Fig. 4a). For Hs AlaRS, the efficiency of activation of Aze was similar to its misactivation of serine (Ser) (Fig. 4c, and higher than the negative controls using Thr and Pro), which is known to be cleared away by editing 13 . In particular, Aze and Ser had 1229-and 950-fold lower relative catalytic efficiency (k cat K m −1 ), respectively, compared to Ala (Supplementary Table 2).
Aze is edited by Hs AlaRS but not by Hs ProRS. Two types of editing prevent mistranslation. One is pre-transfer editing, where the misactivated aminoacyl adenylate is hydrolyzed to prevent the transfer of the amino acid to tRNA 14,15 (proposed to occur within the synthetic active site, distinct editing active site, or after release into solution 16 ). Another is post-transfer editing, where the mischarged amino acid is hydrolyzed to prevent its misincorporation into proteins 4 . Hs AlaRS showed pre-transfer activity against both Ser and Aze (0.09 s −1 ), with no activity against Ala (as expected) (Fig. 5a).
AlaRS has an editing domain that hydrolyzes mischarged Ser from Ser-tRNA Ala13 . We used a truncated version of Hs AlaRS (AlaRS AD ) in which only the aminoacylation domain (AD) was retained after ablation of the editing domain (Fig. 5b). We compared its activity to that of full-length (FL) Hs AlaRS (AlaRS FL ). Both AlaRS FL and AlaRS AD efficiently charged tRNA Ala with Ala, but Ser was only charged by AlaRS AD (Fig. 5b). In contrast to Ser, Aze was not efficiently charged by either AlaRS FL or AlaRS AD . Thus, pre-transfer editing activity is effective in removing misactivated Aze-AMP.
While Hs ProRS lacks a dedicated editing domain, such as found in most bacterial ProRSs, editing of mischarged tRNA Pro in human cells may be achieved by a separate genome-encoded editing factor, ProXp-ala. This factor clears Ala-tRNA Pro by hydrolytic editing 17 . To test whether Hs ProXp-ala editing activity acts on Aze, purified Hs ProXp-ala was incubated with tRNA Pro mischarged either with Ala or Aze. While Hs ProXp-ala hydrolyzes Ala-tRNA Pro , it is unable to clear Aze-tRNA Pro (Fig. 5c). Therefore, Hs ProXp-ala does not have editing activities against Aze.
In a competition assay, we observed that Aze could efficiently compete with Pro for charging onto tRNA Pro by Hs ProRS   Fig. 5d, Supplementary Fig. 3a). In contrast, Ser did not compete with Pro ( Supplementary Fig. 3b). Activation of Aze also resulted in significant production of Aze-tRNA Pro (Fig. 5e, Supplementary  Fig. 3c). Consistent with an earlier report, Aze clearly escaped the editing function of Hs ProRS 18 .

Discussion
More than 50 years ago, hundreds of non-protein amino acids were identified in plants 5 . Due to its 4-member ring structure, Aze has long been recognized as a mimic of Pro. Our work confirmed earlier studies that showed, mostly by indirect methods, Aze can replace Pro in proteins 10,19 . Importantly, our cocrystal of Hs ProRS with the adenylate analog of Aze-AMP was closely similar to that of Tt ProRS in complex with the Pro-AMP analog. However, the close similarity of the Aze side chain to the methyl group of Ala raised the question of why Ala also would not be susceptible to replacement by Aze. Our thought was that the active site of AlaRS might discriminate against Aze and reject it from being activated with ATP to form Aze-AMP. In this instance, we thought that a complex of AlaRS with Aze-AMP would either not occur or would be clearly distinct from that of AlaRS with Ala-AMP. Surprisingly, our co-crystal structure revealed a complex of Hs AlaRS with the Aze-AMP analog. This complex was closely similar at the atomic level to that of the corresponding complex with the cognate Ala-AMP analog. This robust fit of Aze-AMP into both active sites led us to speculate that Aze could be used by both tRNA synthetases and charged onto their respective tRNAs.
Surprisingly, while both Hs ProRS and Hs AlaRS activated Aze, only ProRS was successful in completing the charging event. On more closely inspecting the crystal structures, we could not visualize any differences, such as catalytic water molecules, which would make the AlaRS complex more labile and therefore subject to breakdown of the adenylate. While this may reflect a need for an even higher resolution of the structures to eliminate the possibility of an active water molecule, the data we obtained clearly pointed to the capacity of AlaRS, and not ProRS, to use its pretransfer editing function to clear away activated Aze.
The widespread distribution of Aze at positions for Pro (observed in our mass spectrometry analysis of MBP) gives rise to an immense population of statistical proteins. Some of the micro species in this statistical population are undoubtedly antigenic. As highlighted by Rubenstein 8 , newborn lambs born from mothers fed with Aze-containing sugar beet silage were shown to have an autoimmune multiple-sclerosis-like phenotype 7 . Thus, while a causal link has not been established, this observation leads to speculation that Aze may be a contributing factor. This contribution might be especially penetrant at the vulnerable embryonic development stage, where mistranslated proteins may have more severe effects. Here we found that Aze is toxic to zebrafish embryos and that toxicity was rescued by administration of Pro. While our experiments used high concentrations of Aze, a reduced range of Aze dosage could lead to low levels of mistranslated proteins, with the potential to cause autoimmunity 8 .
Our results show that while dual mimic Aze was captured for removal by one tRNA synthetase (AlaRS), it evaded the other (ProRS 18   Aze induced cell death in zebrafish embryos. Zebrafish were maintained at 28.5°C under continuous water flow and filtration with automatic control for a 14-h/ 10-h light/dark cycle 23 . Single-cell stage embryos were injected with either 1 mM Aze, or 1 mM Aze with 0.25 mM Pro, Val, or Thr (n = 11-40, no randomization or blinding). At 24 h.p.f., zebrafish embryos were AO stained. Live zebrafish embryos were dechorionated in pronase (2.0 mg ml −1 in egg water, 5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl 2 , 0.33 mM magnesium sulfate) and rinsed with egg water. Embryos were then incubated in 10 μg ml −1 AO (Sigma A-6014) in egg water for 15 min, followed by five quick rinses 23  Protein purification. pET21a (Novagen) plasmid containing either his-tagged human ProRS (only the ProRS portion of the fusion protein GluProRS) or pET28a (Novagen) AlaRS (FL and AD (1-455aa)) was transformed into Escherichia coli BL21 (DE3) cells. Expression of proteins was induced with 0.2-0.5 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) at room temperature overnight. Overexpressed proteins were purified from E. coli with Ni-NTA resin (Qiagen), and buffer exchanged in either 50 mM HEPES pH 7.5, 150 mM NaCl buffer, or PBS buffer. Proteins were concentrated and stored at −80°C. Concentrations of proteins were determined by active site titration 24 .
Crystallization. AlaRS AD was digested with trypsin (to improve its crystallizability) overnight at 20°C, and further purified with a monoQ 10/100 GL column (GE Healthcare). The purified AlaRS was concentrated and stored in 2 mM Tris-HCl pH 8.0, 50 mM NaCl buffer at −80°C.
Diffraction data were collected with a single flash-cooled crystal at 100 K on an in-house Cu Kα X-ray system equipped with the Mar345 image plate (MAR research, Germany), and processed with HKL2000. ProRS-Aze-SA and AlaRS-Aze-SA complex structures were solved with molecular replacement using previously reported human ProRS structure (PDB code: 4HVC) and AlaRS structure (PDB code: 4XEM) as templates. The final models was refined to 2.59 Å with Rwork = 22.7% and Rfree = 27.2% for ProRS and 2.03 Å with Rwork = 18.0% and Rfree = 21.5%, and their stereochemical quality was validated with MolProbity.
ATP-PPi exchange assay. The ATP-PPi exchange assay was carried out in the following buffer: 100 mM Tris-HCl, pH 8.0, 10 mM MgCl 2 , 2 mM ATP, 2 mM DTT, 10 mM KF, 0.2 mg ml −1 BSA, 1 mM NaPPi, and [ 32 P]-NaPPi, and reaction was initiated with the addition of 0.05 µM Hs ProRS or 0.1 µM Hs AlaRS (carried out at 37°C) 26 . For the assays with ProRS, Pro concentrations ranged from 0.03 to 2 mM. Aze concentrations ranged from 1 to 30 mM. For the assays with Hs AlaRS, Ala concentrations ranged from 0.03 to 2 mM, Aze concentrations ranged from 2 to 300 mM. The reactions were quenched in quench solution (1 M HCl, 200 mM sodium pyrophosphate, 4% charcoal) at specific time points. Kinetic parameters were determined by fitting data (from three independent experiments) to the Michaelis-Menton equation with GraphPad Prism (± standard error). To compare the [ 32 P]-ATP formation of different amino acids, 1 mM Ala, Pro, Ser, Thr, or Aze was used with either 0.25 µM Hs ProRS or Hs AlaRS (0.5 µM Hs AlaRS was used for Fig. 4c). The graph shows the average values from two independent experiments (carried out in duplicates).
ATP hydrolysis assay. ATP hydrolysis assay was set up as follows: 100 mM HEPES pH 7.5, 20 mM KCl, 10 mM MgCl 2 , 1 mM ATP, 2 U ml −1 inorganic pyrophosphatase, 2 mM DTT, and γ-[ 32 P]-ATP with either 20 mM Ala, 100 mM Ser, or 100 mM Aze. Reactions were initiated by the addition of 1 µM Hs AlaRS, and aliquots of the reaction were quenched in 1.4% HClO 4 , 8% charcoal, 10 mM NaPPi buffer at specific time points. Plate format was used 26 , and reactions were carried out at room temperature.
TLC-based aminoacylation assays. Human tRNA Pro was prepared with the in vitro run off transcription method 27 . Briefly, pUC57 vector containing human tRNA Pro was purified from E. coli with a Giga-prep kit (Qiagen). The purified plasmid was digested with BstNI, and used as a template in a T7 polymerase transcription reaction. The transcribed tRNA Pro was purified in a Mono Q column, and refolded by heating to 80°C and adding 2 mM MgCl 2 while cooling. Human tRNA Ala was prepared using a DNA template containing a T7 promoter and human tRNA Ala gene that was synthesized by PCR of overlapping oligonucleotides. The transcription reaction was performed using MEGAshortscript T7 kit (Ther-moFisher) according to the manufacturer's instructions. The DNA template was then digested with DNase and the tRNA transcript was purified by phenol-chloroform extraction, ethanol precipitated and refolded.
tRNA was [3′-32 P] labeled by incubating tRNA with 50 µM sodium pyrophosphate, 0.2 µM tRNA nucleotidyltransferase, 0.3 µM [α-32 P] ATP, 10 mM MgCl 2 , and 50 mM glycine (pH 9.0) at 37°C for 5 min followed by addition of 1 µM CTP and 10 U ml −1 PPiase and an additional 2 min incubation 28 . For Hs AlaRS, aminoacylation reactions were carried out at room temperature in a mixture of 500 nM human AlaRS FL or AlaRS AD with 20 mM amino acids (Ala, Ser, or Aze), and 1 μM [3′-32 P] labeled tRNA Ala . For Hs ProRS, the reaction was performed with 100 nM ProRS at room temperature in the presence of 2.3 µM [3′-32 P] labeled tRNA Pro , and 2.5 mM amino acids (Pro, Aze, Ala, or histidine (His)). Aliquots of the reaction were quenched with sodium acetate (pH 5.0) followed by phenol-chloroform extraction, ethanol precipitation, and digestion with either P1 nuclease (Sigma-Aldrich) or S1 nuclease (ThermoFisher). After digestion, aliquots were spotted on PEI cellulose plates (EMD Millipore), and run in glacial acetic acid/1 M NH 4  cold Pro. Increasing concentrations of Aze (or Ser as a negative control) were added (0-100 mM) to the reaction to test whether Aze could compete with Pro for charging; 100 nM ProRS was added to initiate the reaction at room temperature. Reactions were quenched in 0.5 mg ml −1 DNA, 100 mM EDTA, 300 mM NaOAc pH 3.0 buffer, with addition of 20% TCA for tRNA precipitation (in a filter plate (Millipore)). The wells were washed in ice-cold 5% TCA, 100 mM Pro solution, dried, and tRNA was hydrolyzed by addition of 0.1 M NaOH. The plate was centrifuged, and the flow through was mixed with supermix cocktail, and the amount of [ 3 H]-Pro-tRNA Pro formed was quantified on the MicroBeta plate reader (PerkinElmer).
Deacylation assays. Human tRNA Pro was [3′-32 P] labeled 28  Mammalian MBP purification. MBP from pMAL-c4X (NEB) was cloned into a pCDNA6 vector. HEK 293 cells (ATCC, mycoplasma tested) were transfected with pCDNA6-MBP, and 4 h later media was changed to 2% FBS DMEM media, with or without 1.2 mM Aze; 47 h after transfection, cells were harvested and lysed through sonication in 20 mM Tris-HCl pH 7.5, 250 mM NaCl, 10 mM imidazole, 5 mM β-Me buffer (lysis buffer). Lysed cells were then centrifuged at 16k rpm, 15 min at 4°C . MBP was purified from the supernatant with Ni-NTA resin (Qiagen) and eluted in lysis buffer containing 100 mM imidazole. Purified MBP was concentrated and buffer exchanged to reduce the imidazole concentration to 10 mM.
Mass spectrometry analysis. MBP samples were reduced with tris (2-carboxyethyl) phosphine hydrochloride (TCEP, Sigma-Aldrich) and alkylated with chloroacetamide (Sigma-Aldrich). Proteins were digested overnight at 37°C in 2 M urea/100 mM TEAB, pH 8.5, with trypsin (Promega). The digested samples were analyzed on a Q Exactive mass spectrometer (ThermoFisher). The digest was injected directly onto a 15 cm, 100 μm ID column packed with Aqua 3 μm C18 resin (Phenomenex). Samples were separated at a flow rate of 300 nl min −1 on an Easy nLCII (Thermo). Buffers A and B were 0.1% formic acid in water and acetonitrile, respectively. A gradient of 1-35% B over 180 min, an increase to 80% B over 40 min, and held at 80% B for a final 5 min of washing before returning to 1% B was used for 240 min total run time. Column was re-equilibrated with buffer A prior to the injection of sample. Peptides were eluted directly from the tip of the column and nanosprayed directly into the mass spectrometer by application of 2. 5 kV voltage at the back of the column. The Q Exactive was operated in a datadependent mode. Full MS 1 scans were collected in the Orbitrap at 70 K resolution with a mass range of 400-1800 m/z. The 10 most abundant ions per cycle were selected for MS/MS and dynamic exclusion was used with exclusion duration of 15 s.
Protein and peptide identification were done with Integrated Proteomics Pipeline-IP2 (Integrated Proteomics Applications). Tandem mass spectra were extracted from raw files using RawConverter 29 and searched with ProLuCID 30 against human UniProt database appended with the transfected protein sequences. The search space included all fully tryptic and half-tryptic peptide candidates. Carbamidomethylation on cysteine was considered as a static modification. Aze incorporation was considered at proline with a mass difference of −14.01565 and at alanine with a mass difference of +12.0. Data were searched with 50 ppm precursor ion tolerance and 600 ppm fragment ion tolerance and filtered to 10 ppm at the precursor level. Identified spectra were filtered using DTASelect 31 and utilizing a target-decoy database search strategy to control the false discovery rate to 1% at the spectrum level.
Aze stock mass spectrometry analysis. Aze stock from Sigma or Santa Cruz were analyzed on an Agilent 6495 triple quadrupole mass spectrometer in MRM mode coupled to an Agilent 1290 UPLC stack. Column was from Imtakt (Amino Statistical analysis. Statistical analyses were carried out using the GraphPad Prism software. Data are presented as mean ± s.d., unless stated otherwise. Data availability. The coordinates and structure factor files of the ProRS:Aze-SA complex (PDB code: 5V58) and AlaRS:Aze-SA complex (PDB code: 5V59) have been deposited in the Protein Data Bank (PDB). All relevant data are available from the corresponding authors upon reasonable request.