Functional Class I and II Amino Acid-activating Enzymes Can Be Coded by Opposite Strands of the Same Gene

Background: Two distinct aminoacyl-tRNA synthetase classes may have descended from opposite strands of a single gene.

Results: Both products from a designed gene encoding ATP-binding sites from Class I and II synthetases catalyze amino acid activation.

Conclusion: The unique information in a gene can be interpreted two alternative ways, both of which accelerate similar chemistry.

Significance: Sense/antisense ancestry reshapes our assessment of the proteome.

Abstract

Aminoacyl-tRNA synthetases (aaRS) catalyze both chemical steps that translate the universal genetic code. Rodin and Ohno offered an explanation for the existence of two aaRS classes, observing that codons for the most highly conserved Class I active-site residues are anticodons for corresponding Class II active-site residues. They proposed that the two classes arose simultaneously, by translation of opposite strands from the same gene. We have characterized wild-type 46-residue peptides containing ATP-binding sites of Class I and II synthetases and those coded by a gene designed by Rosetta to encode the corresponding peptides on opposite strands. Catalysis by WT and designed peptides is saturable, and the designed peptides are sensitive to active-site residue mutation. All have comparable apparent second-order rate constants 2.9–7.0E-3 m⁻¹ s⁻¹ or ∼750,000–1,300,000 times the uncatalyzed rate. The activities of the two complementary peptides demonstrate that the unique information in a gene can have two functional interpretations, one from each complementary strand. The peptides contain phylogenetic signatures of longer, more sophisticated catalysts we call Urzymes and are short enough to bridge the gap between them and simpler uncoded peptides. Thus, they directly substantiate the sense/antisense coding ancestry of Class I and II aaRS. Furthermore, designed 46-mers achieve similar catalytic proficiency to wild-type 46-mers by significant increases in both k_cat and K_m values, supporting suggestions that the earliest peptide catalysts activated ATP for biosynthetic purposes.

Introduction

Previous studies, summarized in Fig. 1 (1, 2), established a strong posterior probability of the sense/antisense ancestry hypothesis (3, 4) by verifying biochemical predictions of the hypothesis. Parallel structural evolution of the two families is consistent with a similar hierarchy over a 10⁵-fold range in catalytic proficiency. We deconstructed members of both aaRS⁴ superfamilies to reveal structurally invariant cores that are relatively free of insertions and deletions that would rule out such coding (5, 6). These “Urzymes” (from Ur = primitive) contain the most highly conserved 120–130-amino acid fragments from the two superfamilies. They contain intact active sites, and they accelerate amino acid activation by 10⁸–10⁹-fold (7, 8) and tRNA aminoacylation by 10⁶-fold (9). Furthermore, they are consistent with the antiparallel sequence alignment implied by the Rodin-Ohno hypothesis. This is not true for the intermediate sized catalytic domains with 200–325 residues, which exhibit somewhat larger catalytic enhancements (1, 10), but in Class I they contain long and variable insertions.

FIGURE 1.

Deconstruction of Class I tryptophanyl (PDB code 1MAU)- and II histidyl (PDB code 2EL9)-tRNA synthetases into successively smaller fragments that retain catalytic activity (1, 7–10, 43). Graphics for smaller constructs are derived from coordinates of the full-length enzymes. Colored bars below each structure denote the modules contained within each structure; white segments are deleted. The number of amino acids (aa) in each construct is noted. Measured catalytic rate enhancements relative to the uncatalyzed second-order rate (k_cat/K_m)/k_non are plotted on vertical scales aligned in the center of the figure and are colored from blue (slower) to red (faster). The 46-mers are the subject of this work.

Bioinformatic evidence for sense/antisense ancestry is largely obscured by adaptive radiation of Class I and II genes to generate first a full repertoire of synthetases and then through the ∼3.5 billion years during which phyla speciated. We nonetheless identified statistically significant (34%) codon middle-base pairing between antiparallel sequence alignments excerpted from ∼200 contemporary Class I tryptophanyl- (TrpRS) and Class II histidyl (HisRS)-tRNA synthetases and smaller samples from the closely related tyrosyl- and prolyl-tRNA synthetase sequences. Those percentages increased to 42% as independently reconstructed ancestral sequences approached their respective roots (11).

Despite the strength of these two kinds of indirect evidence, it would be desirable to establish whether peptides translated from opposite strands of one gene can indeed accelerate the appropriate chemistry. We report such a direct experimental test here (Fig. 2), in which we extend the hierarchy in Fig. 1 to structures only 46 residues long.

FIGURE 2.

Sense/antisense gene coding for Class I and II ATP-binding sites from opposite strands. A, schematic of gene architecture. ATP and amino acid-binding sites of the two antisense gene products reflect across the nucleic acid backbone symmetry axis (center), so that ATP-binding sites are at the N terminus of Class I and C terminus of Class II aaRSs. Structural schematics drawn from PDB files output by Rosetta for Class I (left) Class II (right) 46-mers to suggest possible folded configurations are in similar but not identical orientations to those in Fig. 1. These configurations reflect structural constraints input to Rosetta. Spheres are activated aminoacyl (magenta)-5′-AMP (yellow). B, sequences designed by Rosetta to stabilize tertiary structures in the top panel, preserving coding sequence complementarity. The inset shows the Class I catalytic HIGH and Class II motif 2 signatures; nucleic acid sequence of the gene is shown in Fig. 6A. Red letters denote mutations made to catalytic residues.

Our earlier biochemical studies showed that Urzymes from both classes retain ∼60% of the free energies of activation associated with k_cat/K_m (7–9) and ∼25% of those associated with specific rejection by fully evolved synthetases of noncognate amino acids (1, 2). The catalytic power of peptides related by consensus phylogeny to contemporary aaRS is thus far greater than was anticipated from the gap between the uncatalyzed rates of amino acid activation and peptide bond formation. Uncatalyzed peptide bond formation (12, 13) is ∼10⁶-fold slower than the rate at which Urzymes produce activated amino acids. This million-fold excess catalytic proficiency argues that Urzymes closely resemble true ancestral forms, that they are highly evolved, and hence had simpler functional ancestors. We show here that such ancestors might now themselves be experimentally accessible, especially in context of the literature on uncatalyzed rates (12–16), which both motivated this study and helped to shape the methods used.

ATP binds strongly to specific locations in Class I and II aaRS (Fig. 2). The Class I-binding site (17) occurs in a 46-residue N-terminal fragment (46-mer) that has a catalytic signature containing the four amino acids, HIGH. The Class II-binding site contains an arginine-rich signature known as the motif 2 loop (18–21) near the C terminus of a comparable 46-mer. Both signatures have been implicated in catalysis of amino acid activation by full-length synthetases (7, 22–24). Moreover, they anchor sense/antisense alignments of corresponding 46-mers from the two aaRS classes, reducing ambiguity from insertions and deletions in designing a sense/antisense gene.

Earlier work by Mildvan and co-workers on P-loop peptides from F₁-ATPase (25, 26), DNA polymerase (27), and adenylate kinase (28, 29) showed that these three ∼50-residue nucleotide-binding peptides exhibit ATP-dependent conformational transitions and have ATP affinities similar to those of the full-length enzymes.

Preliminary experiments, described at the beginning of “Results,” show that the wild-type 46-residue ATP-binding peptides cloned from Geobacillus stearothermophilus Class I TrpRS and Escherichia coli HisRS both have high ATP affinity and both accelerate amino acid-dependent ³²PP_i exchange, encouraging us to carry out the protein design experiment reported here.

We adapted the Rosetta multistate protein design functionality (30–32) to constrain coding sequences for 46-residue Class I and II ATP-binding sites to be fully complementary. We describe here the isolation and functional characterization of 46-mers expressed from the designed sense/antisense gene, and we show that active-site mutations in both products eliminate catalytic activity of the designed peptides. Characterization of the wild-type 46-mers, together with the mutational inactivation and statistically significant dependences of the activities of the designed 46-mers on time, amino acid, and peptide concentrations, compose substantive evidence for the authentic catalytic activities of both peptides.

Materials and Methods

Rates of Potential Competing Reactions

We measured the [³²P]PP_i exchange reaction catalyzed by 46-mer peptides on an ∼14-day time scale, over which two competing reactions, uncatalyzed reaction of ATP with amino acid and hydrolysis of ATP itself, are potentially significant. We measured uncatalyzed amino acid-dependent synthesis of [³²P]ATP from unlabeled ATP and [³²P]PP_i in five independent experiments. The five rates range over about an order of magnitude; averaging their free energies of activation yields a second order rate constant of 4.2E-9 s⁻¹ m⁻¹, close to that (8.3E-9 s⁻¹ m⁻¹) estimated (5) from model compound reactions by Kirby and Younas (14). ATP hydrolyzes spontaneously over the same time frame (16) with a reported half-life of 212 days (0.0033 days⁻¹). We confirmed experimentally that a similar rate (0.0035 days⁻¹) applied to our assay conditions.

Computing Assay Blanks Reflecting the Uncatalyzed Rate

Uncertainty of the background (i.e. the “blanks” subtracted from the raw counts/min) compounds the noise associated with enzymatic assays of amino acid activation by 46-mer peptides. We experimented with different ways to track the changing background over the ∼2-week time course of the assay, measuring blanks separately at each time point in the assays of WT 46-mers.

For assays of the designed peptides we used the experimental determinations described above to compute time-dependent increase in disintegrations/min expected for uncatalyzed blanks as shown in Equation 1,

where B₀ is estimated from the mean counts/min of blanks counted at time = 0; k_uncat is the uncatalyzed ³²PP_i exchange rate constant; V is the reaction volume (200 μl); k_ATP is the ATP hydrolysis rate constant; t is the time in seconds; k_³²P is the ³²P decay rate constant; aa is the amino acid concentration; and 4 reflects the 0.25 efficiency of ATP recovery from charcoal, determined separately.

The [³²P]PP_i exchange reaction starts with unlabeled ATP; the only label in ATP is thus that arising from exchange of labeled PP_i into previously synthesized aminoacyl-5′AMP. The factor of 2 in the numerator of the exponential reflects the probability that ³²PP_i, labeled in only one phosphate, will insert the label in the γ-position of ATP. C₀, the factor converting disintegrations/min to micromolar ³²P at t = 0 was determined by fitting standards counted at successive time points to a single exponential and extrapolating to zero time. This procedure reduced noise from the background.

Design of the Sense/Antisense Gene

The algorithm was essentially a parallel implementation of a multistate design as described (30, 33). Rosetta was configured to constrain the choice of amino acids at permitted positions on the opposite coding strand to those stabilizing backbones of the Class I and II ATP-binding sites derived from crystal structures of full-length TrpRS (1MAU, residues 1–46) and HisRS (1EL9, residues 77–122) bound to cognate aminoacyl-adenylate ligands. Consensus active-site residues Pro-10, His-15, Gly-17, and His-18 in the Class I and Arg-113 in the Class II peptide were already coded by complementary codons and were not allowed to vary.

Scoring according to the Rosetta energy function (34) was applied to amino acids for both backbone configurations. An external loop filtered selected amino acids to permit only those joint mutations that had complementary codons. The resulting gene sequence had the lowest Rosetta energy score from an ensemble of 50 designs. The selected gene (PDB files output by Rosetta are appended in the supplement material) was commissioned from GeneScript in two orientations, so that each peptide could be expressed and purified separately.

Purification of Wild-Type 46-Residue Peptides

We cloned, expressed, and purified the 46-residue ATP-binding peptides (i.e. not fusion proteins) from TrpRS and HisRS with His₆ tags. Peptides were solubilized from inclusion bodies in 6 m guanidinium hydrochloride and purified by affinity chromatography on nickel-NTA. They were dialyzed into 10 mm ammonium acetate and concentrated by lyophilization. Their short length makes it unlikely that they have unique conformations. We therefore assumed that denaturation did not alter their structural ensembles in buffer, and we assayed them without a renaturation step.

Fluorescence Titration of ATP with WT Peptides

ATP fluorescence quenching was measured as described for the 50-residue Walker A sequence of F₁-ATPase (25), a putative homolog of the TrpRS 46-mer. Experiments were performed in 5 mm sodium acetate, pH 4.0, at 25 °C. TrpRS and HisRS 46-mer concentrations were increased from 0 to 510 μm and from 0 to 280 μm, respectively. Solutions were excited at 240 nm and recorded using a SPEX Fluorolog-3 spectrophotometer. The entire emission spectrum from 300 to 450 nm was recorded for buffer alone and for each peptide concentration in the presence and absence of 12.6 μm ATP (Trp-46-mer). ATP was increased to 120 μm for the His-46-mer to increase the signal.

Peptide/ATP interactions included both quenching of ATP by peptide and quenching of peptide fluorescence by ATP. The ATP concentration dependence of these changes was estimated by subtracting the integrated fluorescence of buffer alone from spectra with increasing concentrations of peptides. Difference curves were fitted to Equation 2,

where A is an amplitude of ATP-dependent quenching, and K_{D, ATP} is the dissociation constant for the peptide·ATP complex.

Expression of Designed Peptides

The 46-mer peptides were expressed as MBP fusion proteins with a His₆ tag and purified from BL21 E. coli cells. Cells were transformed by heat shock and left to recover for 1 h in LB growth media. They were grown overnight in media containing 100 μg/ml ampicillin, transferred to 1 liter of fresh media, and incubated until the A₆₀₀ = 0.6. Cells were harvested by centrifugation at 4,000 rpm at 4°C for 20 min and lysed in 50 mm Na₂HPO₄, pH 8.0, 300 mm NaCl, 10 mm β-mercaptoethanol, 0.5% v/v Nonidet P-40 by sonicating in three 45-s intervals with 15 s rest with cooling. Sonicated cells were centrifuged at 15,000 rpm for 20 min, and the supernatant was collected. Nickel-NTA beads were added to the supernatant and left overnight. The beads were washed once with lysis buffer and twice with 50 mm Na₂HPO₄, pH 7.0, 300 mm NaCl, 10 mm β-mercaptoethanol, 30 mm imidazole for 20 min each. Beads were transferred to a column, and protein was eluted using with 300 mm imidazole in five 2-ml fractions. Pooled fractions were dialyzed and stored at −20 °C in 50% glycerol and were assayed without tobacco etch virus protease cleavage. Concentrations of fusion peptide stocks were estimated from A₂₈₀ and confirmed by standardized Coomassie staining on polyacrylamide gels (Fig. 3).

FIGURE 3.

Gel showing expression and purification of designed and active site mutants of the Class I (I-mer) and Class II (II-mer). MBP is maltose-binding protein; Lys is crude lysate; sol is soluble supernatant after lysis; and elute is the fraction eluted from nickel-NTA and assayed.

Pyrophosphate Exchange by WT Peptides

Pyrophosphate exchange, the standard assay for aminoacyl-adenylate (AA) synthesis, measures the incorporation of labeled inorganic pyrophosphate, ³²PP_i, into ATP by reversal of the amino acid activation Reaction 1.

The 2-week time courses equal the half-life of ³²P and hence approach but do not exceed the detection limit.

³²PP_i exchange assays were done at 37 °C and initiated with 10 μl of enzyme in 190 μl of assay mixture: 0.1 m Tris-HCl, pH 8.0, 0.01 m KF, 5 mm MgCl₂, 1 mm ATP, and 10 mm tryptophan for Trp-46-mer or 10 mm histidine for His-46-mer, 70 mm 2-mercaptoethanol plus 2 mm ³²PP_i at a specific radioactivity between 1 and 2 × 10⁵ cpm. Michaelis-Menten experiments were performed with 1 mm ATP and varying amino acid concentrations for amino acid-dependent kinetics, and 10 mm amino acid and varying concentrations of ATP for ATP-dependent kinetics. These conditions saturate the peptides in the fixed substrate.

All assays were processed by adding 200-μl samples to 100 μl of 7% perchloric acid on ice. A 15% activated charcoal suspension (50 μl) was added with mixing. The slurry was added to 0.2-mm centrifuge filter tubes and washed with 10 ml of water. Radiolabeled ATP was eluted with 100 μl of pyridine and counted using Ecoscint A. Peptide concentrations of 100 μm (Trp-46-mer) or 200 μm (His-46-mer) and incubation times of 168 h were necessary for a sufficient signal-to-noise ratio from the WT peptides.

High tryptophan concentrations result in the elution from charcoal of a yellow scintillant. Thus, blanks increased linearly with tryptophan concentration. Blanks for the WT experiments were therefore prepared at different tryptophan concentrations and were used to construct a nomograph, from which appropriate blank values could be determined. To avoid this problem in assays of the designed peptides, leucine was used as the Class I amino acid.

Pyrophosphate Exchange by Designed Peptides

In view of the observation that very high concentrations of WT 46-mer were required in Michaelis-Menten experiments and that K_m values in those experiments were smaller for ATP than those for amino acids, we decided to carry out Michaelis-Menten experiments only for amino acid dependence of the designed 46-mers. Triplicate reactions for the designed peptides were carried out in sealed tubes with 1.0-ml volumes at ∼35 °C in solutions of 50 mm Tris-HCl, pH 7.5, 10 mm KF, 2 mm ATP (>10× K_m); 5 mm MgCl₂, 70 mm β-mercaptoethanol, 2 mm sodium pyrophosphate, to which were added ³²PP_i to ∼6 × 10⁵ cpm/ml. Assays were done using two concentrations, 0.44 and 1.75 μm of the respective peptide, and two nonzero concentrations, 10 and 32 mm leucine (Class I 46-mer) or histidine (Class II 46-mer). At appropriate time points, 200-μl samples were withdrawn and treated as described for WT assays.

Designed peptides were assayed multiple times, including a single 8-day time point performed after the mutant peptides were assayed. A 16-day time point from that experiment exhibited similar WT sequence, peptide concentration, and amino acid concentration dependences but had slightly lower values than those from the 8-day time point and were discarded.

Estimating Steady-state Parameters

Catalytic proficiencies of WT and designed peptides can be compared with those of larger aaRS modules (Fig. 1) only by estimating their apparent second-order rate constants. The Michaelis-Menten formalism entails two unknown parameters that we wish to determine. The problem, especially with systems as noisy as those investigated here, is to estimate the two steady-state parameters with minimum uncertainty.

The 4-fold replication of assays for the WT peptides permitted the identification and elimination of ∼10 obvious outliers before the remaining data were fitted to the Michaelis-Menten equation using nonlinear regression in JMP (35). High noise levels observed especially with ATP dependence in these assays suggested that a different strategy might be useful for assaying the designed peptides.

Aspects of this problem have been discussed, for example, by Hardin and Sloane (36) in the general context of statistical design and inference. A key strategy is to use high replication to increase the accuracy of experimentally determined points with the greatest influence on the model parameters.

As an example, consider the slope and intercept of an assumed linear relationship. Points close to the minimum and maximum of interest have the largest influence on the slope and intercept. Their values can be determined most accurately if experimental points at corresponding x values are replicated. Points in between have far less influence on either parameter than those at the extremes.

To estimate the parameters of a rectangular hyperbola, one wants to know as precisely as possible the reaction velocities at saturation (i.e. for k_cat) and at half-saturation (i.e. for K_m). Amino acid concentrations in our design (10 and 32 mm) approximate those values of the independent variable and amino acid concentration, and multiple measurements enhance the precision of measured velocities at those points by √N for N experimental determinations. Use of multiple regression methods, which allow all experiments to influence inferences about all factors, helps to realize this enhancement of precision in the estimation of coefficients. We therefore believe that this procedure gives a more accurate estimate of the steady-state parameters than would the conventional practice of fitting to measurements with fewer replications over a range of increasing substrate concentrations.

Using Free Energies for Comparison

We compare rate constants that span 14 orders of magnitude. The comparison benefits from converting both rate and binding constants to free energies. The free energy of activation contains an additive constant from the Boltzmann factor, ln(kT/h = 6.2E¹² s⁻¹), which we omit, consistent with Ref. 37.

Multiple Regression Methods

Multivariate analysis is an extension of linear regression that makes the best use of appropriately balanced multivariate experimental designs (38, 39) necessary to characterize how phenomena depend on multiple independent variables.

³²PP_i exchange activity may depend on the following independent variables: the class of the synthetase; whether the catalyst is a 46-mer or an MBP control; whether the sequences are designed (WT) or mutant; incubation time; peptide concentration; and the amino acid concentration. A fully balanced experimental design would test each combination of the above variables an equal number of times. We approximate that ideal in the experiment designed to test the activity of the designed 46-mers.

Three dependent variables were evaluated as follows (i) for [³²P]ATP, in micromolars, the amount of labeled ATP generated per experiment was used for regression modeling of the combined effects of all independent variables, and their time dependence gives estimates for rates of product formation for Michaelis-Menten analysis; (ii) [³²P]ATP produced per peptide was used to assess distributions for WT, mutant, and MBP control experiments; and (iii) d[³²P]ATP/dt, s⁻¹ is the rate of [³²P]ATP synthesis for each data point. Each row of the experimental matrix associates these dependent variables for one observation of ³²PP_i exchange activity with its own time in seconds, and numerical values indicating whether the peptide sequence was WT or mutant, whether it was an MBP control, peptide, and amino acid concentration, and class. These independent variables are denoted x_i.

Multivariate regression models (Equation 3) express calculated values for the experimental observations as linear combinations of the values of the independent variables:

where Prod_calc is the reaction product; β₀ is the intercept where all x_i values are zero; β_i values are coefficients representing the slopes of the calculated function with respect to each independent variable, x_i; β_ij are coefficients for interactions between x_i and x_j; and ϵ is a residual. A two-way interaction represents the different response to one variable at high and low values of another variable. Higher order interactions can be invoked if there is sufficient statistical support.

Coefficients are then estimated by least squares minimization of the difference between observed and calculated [³²P]ATP. The regression model itself is an equation for a multidimensional surface that best fits the observed data. If there are no two-way or higher interactions, this surface is a plane whose slopes in each direction are given by the β_i. Interaction terms introduce deformations that improve the fit of the surface to the data.

Regression modeling is a robust way to extract meaningful signals from noisy data. The advantages of this statistical procedure and a balanced design are as follows: first, each data point can be used together with all of the others to assess average differences at different values of the independent variables; and second, it affords a sound statistical test, the Student's t value and its probability under the null hypothesis, for the significance of each predictor, x_i, x_i·x_j, etc.

Results

Absent a catalyst, amino acid activation is the slowest of the chemical steps necessary for protein synthesis and so is of central importance to the origin of biology. We describe catalysis of this key reaction by very short peptides. The experiments themselves and the conclusions that they support have little precedent. We highlight here the experimental observations supporting our conclusions, as well as the difficulties, the extent to which they have been overcome in the present experiments, and directions from which to seek further corroboration.

Wild-type TrpRS and HisRS 46-mers Retain Functionality

We first showed that peptides with wild-type sequences cloned by PCR from genes for the Bacillus stearothermophilus TrpRS and E. coli HisRS have ATP binding properties comparable with those found by Mildvan and co-workers (28–31) for peptides homologous to the Class I ATP-binding sites from several ATP-dependent enzymes, and we then documented their catalytic activities.

F₁-ATPase, adenylate kinase, and DNA polymerase I have P-loops and are distant homologs of the Class I ATP-binding site. Those peptides vary in length from 45 to 51 residues and were shown to bind ATP tightly (∼10 μm) by titrating ATP with peptide and to undergo ATP-dependent conformational transitions by two-dimensional NMR (25–29).

The wild-type sequences for the TrpRS and HisRS 46-mers bind ATP tightly (Fig. 4). Fluorescence changes due to ATP exceed the total fluorescence from ATP alone, so they include both quenching of ATP fluorescence by Trp-46-mer and His-46-mer and quenching of peptide fluorescence by ATP. We subtracted titrations of buffer with peptide from titrations of ATP with peptide (Fig. 4, A and B) to identify these ATP and peptide concentration effects (Fig. 4, C and D).

FIGURE 4.

Fluorescence changes due to the presence of ATP, derived from titrations of ATP with TrpRS (A and C) and HisRS (B and D) 46-mer peptides. Integrated (300–450 nm) spectra for peptides without ATP (A and B, black) were subtracted from those with ATP (A and B, gray) to estimate quenching of ATP fluorescence by the peptides and of the peptides by ATP. ATP-dependent fluorescence changes were fitted to Equation 1 for Trp-46-mer (C) and His-46-mer (D). Dashed lines indicate data points omitted from fitting because of development of turbidity developing at high [peptide]. Fitted dissociation constants are given in the text.

Gel filtration experiments (data not shown) showed ATP-dependent oligomerization when either peptide was at high concentrations. These experiments dilute the peptides by more than an order of magnitude, and therefore assaying the monomer and dimer fractions was unreliable. Data points at high peptide concentrations were therefore excluded from fitting (dashed line segments in Fig. 4, C and D).

ATP-dependent fluorescence changes were fitted to Equation 1, assuming one site (Fig. 4, C and D), to give dissociation constants of 145 ± 19 μm for the Trp-46-mer and 56 ± 8 μm for the His-46-mer. These dissociation constants are severalfold lower than ATP-dependent K_m values for the full-length enzymes (300 μm for full-length TrpRS (9) and 890 μm for full-length HisRS (40)), which also are thermodynamic dissociation constants by virtue of the exchange condition of the assay (Cleland et al. (56)). High ATP affinity was observed by Mildvan and co-workers (25) for the corresponding ATP synthase Walker A peptide.

Steady-state Kinetics Analysis of WT 46-mers

The WT 46-mers also accelerate cognate amino acid activation. We did 168-h time course assays of amino acid-dependent ³²PP_i exchange for both WT 46-mers. Catalytic activity was significant by these tests; regression of 18 time-dependent measurements with and without peptide showed that Student's t test p values were ∼10⁻⁴ for peptide catalysis and 0.01 for the effect of time in both cases.

Substrate-dependent assays (Fig. 5) show that both WT 46-mers exhibit saturation behavior with respect to both amino acid and ATP concentrations. The high peptide concentrations required for sufficient signal-to-noise in these measurements (shown by bold arrows in Fig. 5) place upper bounds on the Michaelis constants derived from fitting by nonlinear regression to rectangular hyperbolae. The apparent second-order rate constants for amino acids, k_cat/K_m = 3.5E-3 s⁻¹ m⁻¹ for Trp-46-mer and 8.3E-4 s⁻¹ m⁻¹ for His-46-mer, are therefore lower limits for 46-mer catalytic proficiency. These limitations do not compromise the conclusion that although turnover is weak (average k_cat ∼2.7 ± 2/8E-7 s⁻¹; Table 1), the 46-mers are at least 10⁻³–10⁻⁴ as proficient as the corresponding 120–130 residue Urzymes (9).

FIGURE 5.

Steady-state kinetic parameters for the WT TrpRS and HisRS 46-mer peptides. Assays were conducted as described under “Materials and Methods.” Data points to the left of vertical lines at 100 mm along the x axis cannot be interpreted unambiguously, because the substrate concentrations are below the catalyst concentration. Velocities are therefore underestimates.

TABLE 1.

Steady-state kinetic parameters for wild-type TrpRS and HisRS 46-mers, derived from the fits shown in Fig. 5

Standard errors for estimated parameters k_cat and K_m are calculated by the nonlinear least squares algorithm in JMP (35). The values for k_cat/K_m are derived from those standard errors by conventional propagation of errors for a ratio as follows: Std_error (k_cat/K_m) ≈ (k_cat/K_m · √((jk_cat/k_cat)²+(K_m/K_m)²).

Michaelis constants for ATP are significantly smaller than those of the wild-type enzymes, confirming conclusions drawn from ATP fluorescence in peptide titrations. Moreover, those binding isotherms (Fig. 4) cannot be attributed to a minor contaminant. This, together with the fact that the peptides were isolated without renaturation from inclusion bodies, and therefore should not contain active wild-type enzymes, argues that the catalysis is not due to tiny amounts of contaminating wild-type activity.

Designed Gene Encoding Class I and II 46-mers from Opposite Strands

It is a straightforward extension of multistate design (30, 33) to build sequences simultaneously to stabilize two different backbone scaffolds. We configured this feature of the protein design program Rosetta to enforce complementary coding sequences consistent simultaneously with structures of the Class I and II ATP-binding 46-mers (Figs. 2 and 6). Rosetta's semi-empirical energy function (30, 31) makes its choices an effective surrogate for sequence information inaccessible from multiple sequence alignments of related proteins (Fig. 6B). The selected gene sequence, Fig. 6A, was synthesized (GeneScript) and cloned in opposite directions to express each peptide as a maltose-binding protein fusion with an additional His₆ tag.

FIGURE 6.

Sense/antisense gene. A, DNA sequence, with codon/anticodon pairs translated. B, comparison of the sense antisense gene designed by Rosetta (30) (center) with Logos derived from multiple alignments for ∼200 sequences for each of six contemporary Class IA, -B, and -C and five Class IIA, -B, and -C sequences (Fig. prepared using Canvas 8.0 and WebLogo (55)). Rosetta was not allowed to vary the five blue amino acids in the designed sequence. Ovals are positions where, elsewhere, Rosetta chose consensus amino acids from the biological MSA. Percentages to the right indicate a wide variance in such selection between the Class I and II 46-mers as discussed in the text.

Class I and II Sense/Antisense MBP-46-mer Fusions Both Accelerate Amino Acid Activation

Amino acid activation by designed and mutant peptides was followed by the ³²PP_i exchange assay at different time points at different peptide and amino acid concentrations. This experimental matrix was used to assess the significance of answers to the following questions concerning the catalytic activity of the two peptides (Table 2). (i) Do levels of [³²P]ATP increase significantly with time? (ii) Are rate accelerations observed for the two peptides significantly greater than those observed in the presence of MBP controls? (iii) Are the rate accelerations sensitive to active-site mutations? (iv) Do the rates depend on the concentrations of peptide and of substrate?

TABLE 2.

Multivariate regression analysis of factors affecting amino acid activation by designed Class I and Class II 46-residue peptides containing ATP-binding sites

R² = 0.71 with 227 degrees of freedom; F = 61.0 for all 237 observations from two investigators on different occasions (P(F) <0.0001). Coefficients are sorted in descending order of statistical significance given by p value, excluding the constant term. p values given are those for the F ratio statistic from stepwise multiple regression to emphasize the relative importance of different coefficients. These are approximately equal to the p values of the Student's t test.

Term	Estimate	Standard error	t ratio	Probability >|F|
WT	7.7E-03	5.4E-04	14.25	2.20E-45
[Peptide], m	4.5E+03	4.2E+02	10.66	4.50E-35
(Time, s)·(WT)	1.5E-08	1.6E-09	9.78	1.70E-24
([Peptide], m)·(WT)	7.4E+03	8.0E+02	9.32	4.50E-19
Time, s	7.1E-09	7.8E-10	9.09	1.10E-17
([Peptide], m)·(class)	−5.9E+03	7.9E+02	−7.5	3.80E-15
MBP	−8.8E-03	1.4E-03	−6.43	1.50E-12
Class	−2.8E-03	5.2E-04	−5.42	7.40E-10
[Amino acid], m	4.2E-02	2.2E-02	1.94	5.39E-02

We made alanine substitutions to the second histidine of the HIGH signature of the Class I 46-mer (H18A) and arginine (R37A) of the Class II 46-mer (Fig. 2B) to authenticate these catalytic activities. The designed peptides were assayed initially and were re-purified and assayed again later, after the mutant peptides were assayed.

As the experimental data approach the level of detection, designed Class I and II 46-mers were assayed 96 times (six replicates, four times, two peptide concentrations, two amino acid concentrations), and the corresponding mutant peptides were assayed 48 times. This high replication increases the signal-to-noise ratio by √N ∼10- or 7-fold for the mutants, affording a more precise determination of the uncertainty in average values estimated over multiple measurements.

The redundancy and balance of the experimental design allowed us to verify that all dependences also are statistically significant (Table 2). Multiple regression methods average over the high noise level, to assess the relative statistical significance simultaneously for differences in the time-dependent [³²P]ATP synthesis that answer questions i–iv. To assess the joint statistical significance of our conclusions, columns of the design matrix were used to build regression models (see Equation 3) for the data measured by both investigators for WT and mutant peptides and MBP controls. All calculations were done using JMP (35).

Results of regression modeling in Table 2 convey in compact form many related conclusions that are not immediately apparent but that have bearing on the questions addressed by this work. The R² value, 0.71, tells us that the model in Equation 3 can explain more than 70% of the variation in observed [³²P]ATP values. The unexplained variance approaches the noise level. As the ratio of a coefficient to its standard error, the Student's t values measure the signal-to-noise of the regression coefficients. Their p values are the probability of a Student's t value as large as that of the model coefficient under the null hypothesis that the different experimental data do not arise from differences in the corresponding independent variable. The relative impact of the noise depends on the magnitude of the experimental signal. Small p values mean that, despite the high noise level, the range of measurements is sufficient to exhibit the dependences we hoped to demonstrate. The p values thus furnish quantitative estimates of the impact of noise and certify the significance of differences despite the noise in the data.

All main (i.e. intrinsic) effects have p > |t| values ≪0.0001. Their signs and magnitudes are physically reasonable. The most significant predictors of activity in Table 2 are the wild-type sequence (WT: t = 14; p > |F| <10⁻⁴⁵), the peptide concentration ([peptide], M; t = 10.7; p > |F| <10⁻³⁵), the difference between the time dependence of the WT and mutant sequences ((time, s)·(WT)); t = 9.8; p > |F| <10⁻²⁴) and the time (time, s; t = 9; p > |F| <10⁻¹⁷). All but one coefficient have p > |t| values ≪0.0001.

Figs. 7 and 8 unpack the two most important conclusions from Table 2 graphically. Fig. 7 shows that (i) designed peptides do accelerate amino acid activation whereas MBP, expressed and purified by the same procedure as that used for the peptides, does not; and (ii) active-site mutations substantially reduce catalysis, thereby implicating the peptides themselves as the authentic catalysts.

FIGURE 7.

Distributions of [³²P]ATP produced per peptide for all assays for designed 46-residue sense/antisense Class I and II peptide catalysts, MBP controls, and mutant peptides. Distributions of [³²P]ATP produced per peptide for all assays for the designed 46-residue sense/antisense Class I and II peptide catalysts, MBP controls, and mutant peptides are shown. All data were divided by concentration. White arrows denote averages of the distributions, and vertical cross-bars at the tails of the arrows give their standard errors of the mean. The y axis corresponds to the ³²PP_i exchange data, in units of mole of ATP/mol of peptide. Left, MBP controls. Center, distributions for designed peptides. Right, mutant peptides. Differences between peptide and MBP controls have Z-scores of 10.4 and 9.8. The differences for Class I and II peptides have Z-scores of 3.0 and 11.4. Prepared using JMP (38).

FIGURE 8.

Time, class, and peptide concentration (0. 44 μm; 1.75 μm) dependences of catalytic peptide activity. The y axis is in [³²P]ATP concentration. Variances of the time dependences are indicated by shading along either side of the fitted lines. Prepared using JMP (35).

Whereas the designed peptides are the dominant proteinaceous components of the solutions we assayed, there are doubtless other components, to which the observed rate accelerations might be attributed. These include possible contamination of the peptides by tiny amounts of wild-type aaRS not excluded during purification. As these concentrations also increase with the volume of catalyst added ([peptide]), the statistical significance of that predictor alone does not, by itself, address the question of contamination.

Fig. 8 shows the increased rates of [³²P]ATP synthesis by different concentrations of the two designed peptides at the amino acid concentration = 0.032 m. The increased velocities at the higher peptide concentrations are also evidenced by the mean velocities (Table 3), which have been normalized for peptide concentration and are within experimental error. Increased peptide concentrations do not increase velocities either for the mutated peptides or for MBP only controls. That observation, which is confirmed quantitatively by the significance of the [peptide]·WT coefficient in Table 2, is evidence that the observed rate accelerations in Fig. 8 do not arise from contaminating activities, which would also increase activities in those controls. Table 2 thus simultaneously documents that solutions containing the designed peptides accelerate amino acid activation by different amounts in proportion to time, peptide, and amino acid concentrations and, especially, only with WT active-site residues.

TABLE 3.

Mean values and standard errors of rate data used to evaluate steady-state kinetic parameters for the designed peptides

Average rates were estimated by fitting the time-dependent values for each [amino acid] and each [peptide] for each class over individual data points after correcting for peptide concentration and are plotted in Figs. 8, A–D. S.E. = standard error; n = number of replicates.

[Amino acid]	n	[32P]ATP_Leu	S.E._Leu	[32P]ATP_His	S.E._His
m		s−1		s−1
0	36	8.5E-07	1.3E-06	3.4E-07	9.5E-07
0.44 μm Peptide
0.01	36	3.4E-05	2.0E-05	2.0E-05	6.4E-06
0.032	36	5.0E-05	1.4E-5	2.8E-05	5.9E-06
1.75 μm Peptide
0.01	36	4.6E-05	6.5E-06	0.00001	6.0E-06
0.032	36	7.4E-05	1.2E-05	2.6E-05	4.0E-06

The time dependences of the distinct MBP and mutant controls are both slightly negative (−6 ± 2.2 E-11 s⁻¹). However, their absolute values are ∼2% of that from WT sequences (3.5 E-9 ± 1.2 E-9 s⁻¹) and hence are essentially independent of time. The negative time dependences are quite similar, and we conclude that our estimate of the uncatalyzed rate is a good approximation. For these reasons, the data in Table 2 argue significantly in favor of unique catalysis by the peptides themselves.

Estimation of Apparent Second-order Rate Constants, k_cat/K_m

The Michaelis-Menten plots in Fig. 5 suggest that the WT 46-mer peptides saturate at high substrate concentrations. It is unlikely that such simple peptides can exhibit cooperative behavior with respect to substrate concentrations. Cooperative catalytic behavior can be difficult to establish even for full-length enzymes. Thus, Michaelis-Menten behavior is the simplest postulated behavior, and we assume it here for the purpose of comparing proficiencies with those of other previously characterized aaRS constructs (7–9).

Activities for both designed peptides also appear to saturate. In view of the intrinsically noisier plots of the [ATP] dependence of WT activities, we decided to fit to more highly replicated rates at three amino acid concentrations (0, 10, and 32 mm), shown with standard errors of the means in Table 3, rather than dividing them between amino acid and ATP experiments.

We estimated the steady-state parameters from the full data set by calculating specific velocities, d[³²P-ATP]/dt s⁻¹, for both peptide concentrations, and fitting to the Michaelis-Menten equation to obtain minimum variance estimates for the apparent second-order rate constants (Fig. 9). Data for each peptide concentration constitute independent experiments; agreement between these four experiments, each of two concentrations of the Class I (Fig. 9, A and C) and II (Fig. 9, B and D) peptides, provides estimates of consistency for each peptide, which is expected, and between estimates for the two Classes, which is unexpected.

FIGURE 9.

Estimation of apparent second-order rate constants. Amino acid concentration dependences plotted in Michaelis-Menten format. A–D, data for all time points for each unique combination of peptide and substrate concentrations were fitted against time in seconds by linear least squares. The mean values of rates for both amino acid concentrations were plotted against amino acid concentration for experiments with 0.44 μm (A and B) and 1.75 μm peptide (C and D), giving two independent experimental estimates of steady-state kinetic parameters for Class I (A and C) and Class II (B and D) peptides, which are given in Table 4.

The fits are in reasonably good agreement (Table 4). Michaelis-Menten parameters afford average estimates, 7.7 ± 0.9E-3 s⁻¹ m⁻¹ for Class I and 3.3 ± 1E-3 s⁻¹ m⁻¹ for Class II apparent second-order rate constants, k_cat/K_m, for amino acid activation. Thus, the biochemical behavior of the Class I and II 46-residue peptides encoded by opposite strands of the designed gene (Fig. 2B) exhibits all appropriate characteristics of enzymes; activities of both peptides are greatly reduced in active-site mutants, and K_m values for amino acids are significantly different from those of full-length enzymes. We conclude that the observed activities represent the authentic catalytic activities of gene products coded by opposite strands of the same gene.

TABLE 4.

Apparent second-order rate constants, k_cat/K_m for Class I and II designed 46-mers

Standard errors (S.E.) for estimated parameters k_cat and K_m are calculated by the nonlinear least squares algorithm in JMP (35). Those for k_cat/K_m are derived from those standard errors by conventional propagation of errors for a ratio: Std_error (k_cat/K_m) ∼ (k_cat/K_m)·√(](σ _kcat/k_cat)² + (σ_KM/K_m)²). The estimates of k_cat and k_cat/K_m for each peptide are approximately the same for peptide dilutions that differ by 4-fold.

Parameter	Class I	StdErr_I	Class II	StdErr_II
0.44 μm
kcat, s−1	6.4E-05	2.4E-06	3.6E-05	8.6E-07
Km, m	0.009	0.001	0.008	0.0006
kcat/Km, m−1 s−1	7.1E-03	8.3E-04	4.5E-03	3.5E-04
1.75 μm
kcat, s−1	1.0E-04	1.2E-06	4.30E-05	1.6E-06
Km, m	0.012	0.0004	0.02	0.002
kcat/Km, m−1 s−1	8.3E-03	3.0E-04	2.2E-03	2.3E-04

Design Altered the Enzymatic Properties of the Sense-Antisense Class I and II 46-mers

WT and designed peptides have approximately the same catalytic proficiencies, as given by their activation free energies for k_cat/K_m (∼3.5 kcal/mol; Table 5). Qualitatively, however, the WT peptides required ∼100-fold higher peptide concentrations than the designed peptide assays to achieve adequate signal-to-noise ratios. That qualitative discrepancy implies significant differences in individual Michaelis-Menten parameters.

TABLE 5.

Comparison of averaged steady-state kinetic parameters of the WT and designed 46-mers

Free energies were calculated as ΔG = −RT ln(κ), where κ is either k_cat, K_m, or k_cat/K_m. K_mvalues are for amino acid dependence only. Standard deviations (Stdev) of the averages are given in rows below the mean free energies. A fuller comparison of individual free energies is shown in Fig. 10.

	〈ΔG kcat〉	ΔG Km	ΔG kcat/Km
	kcal/mol	kcal/mol	kcal/mol
WT	9.2	5.4	3.8
Stdev	0.7	0.6	1.0
Designed	5.8	2.7	3.2
Stdev	0.3	0.2	0.4

Free energies estimated from the steady-state parameters from six different determinations (TrpRS and HisRS WT 46-mers with ATP and amino acid, plus the Class I and II designed 46-mers with amino acid; Table 5) cluster significantly according to WT versus designed classes and substrates (Fig. 10). The two WT 46-mers bind substrates ∼100-fold more tightly but have ∼100-fold smaller turnover numbers than do the designed Class I and II 46-mers (Tables 1 and 4).

FIGURE 10.

Significant clustering of steady-state parameters for wild-type (amber) and designed (blue) Class I and II 46-mer peptides. Plots show observed values for the free energies of activation derived from k_cat and binding, K_m, versus calculated values from linear regression models using the coefficients for the binary descriptors: Designed {0,1}, Class {1(▴),2(●} (k_cat only), and substrate {ATP (*) = 1, amino acid (●) = 2} (K_m only). A, two designed 46-mers have substantially higher turnover numbers (〈ΔGk_cat〉 = 5.8 kcal/mol) than the wild-type 46-mers (〈ΔGk_cat〉 = 9.2 kcal/mol), and the TrpRS (Class I) wild-type 46-mer has significantly higher turnover numbers than the HisRS (Class II) 46-mer with both amino acid and ATP substrates. B, both wild-type 46-mers bind amino acid substrates with higher affinities (〈ΔGK_M〉 = 5.9 kcal/mol) than does either designed 46-mer (〈ΔGK_M〉 = 2.7 kcal/mol). Similarly, wild-type 46-mers have higher affinity for ATP than they do for amino acid substrates. Student's t test probabilities for these effects under the null hypothesis are <0.0001 for WT sequences in both models, 0.01 for Class in the model for ΔGk_cat, and 0.01 for substrate in the model for ΔGK_M, so these differences are statistically significant.

Because these differences are so large, their statistical significance is high (Tables 5 and 6). The design process consistently increased turnover number and reduced amino acid affinity. It remains to be tested whether or not the striking differences observed here result from imposing the sense/antisense constraint. An interesting implication is that changes required for sense/antisense coding may be inconsistent with high amino acid specificity.

TABLE 6.

Coefficients for regression models relating ΔG k_cat and ΔG K_m to independent design variables

Designed is 1 for designed and 0 for WT 46-mers. Class is 1 for Class I and 2 for Class II. Substrate is 1 for ATP and 2 for amino acid. R² values for the regression models are ∼0.99 and are given below the x axes of Fig. 10. The intercept is the constant term in the regression model and is equal to the average over all values of the dependent variable. Std. Error is standard error.

Term	Coefficient	Std. Error	t ratio	Probability >|t|
	kcal/mol
ΔG kcat
Intercept	8.1	0.2	34.8	<0.0001
Designed	−3.4	0.1	−24.5	<0.0001
Class	0.8	0.1	5.5	0.01
(Designed)*(Class)	−0.7	0.3	−2.5	0.07
ΔG Km
Intercept	5.9	0.1	47.6	<0.0001
Designed	−2.8	0.2	−13.0	<0.0001
Substrate	−0.5	0.1	−3.7	0.01

This unexpected result is, however, consistent with the possibility that sense/antisense-encoded 46-mers functioned initially to bind ATP and to facilitate its reaction with low molecular weight substrates during the origin of life and that their specialization as amino acid-activating enzymes developed later (41). The surprisingly balanced catalytic behavior of the products of the designed gene also indirectly validates the novel use of Rosetta in the manner described here. The designed 46-mers behave much more like generalized ATP activators. The smaller turnover numbers of the WT 46-mers, in contrast, likely reflect the fact that much of the contemporary aaRS catalytic machinery involves interactions between the ATP-binding sites and other modular components (42, 43), which have been deleted.

The explanation that differences in the active fraction of molecules purified from inclusion bodies (wild type) and as fusion proteins (designed) account for these kinetic differences is rendered unlikely by the ATP quenching behavior of the wild-type peptides, which are inconsistent with activity vested in 1% of active molecules, as that would imply 100-fold tighter dissociation constants than those in Table 1.

Comparison of Biological and Rosetta Multiple Sense/Antisense Sequence Alignments with the Assayed Sense/Antisense Sequences

Many layers of amino acid substitutions were probably necessary to embed the putative primordial 46-residue catalysts represented here by the designed sequences into the contexts of the 10 contemporary aminoacyl-tRNA sequences that we propose descended from each strand. The gene sequence in Fig. 6A resulted from an initial design involving 50 trials. The resulting sequence is compared in Fig. 6B to sequences from Fig. 1 and to those from a multiple alignment of ∼200 biological sequences from six (LeuRS, IleRS, GluRS, GlnRS, TrpRS, and TyrRS) Class I and five (HisRS, ProRS, AspRS, AsnRS, and PheRS) Class II sequences as described for TrpRS, TyrRS, HisRS, and ProRS (11).

Together with functional activities of the wild-type sequences described above, the differences between all three sets of sequences suggest that a relatively large number of sequences may be consistent with comparable catalytic activities for both types of peptides. It may be noteworthy that Rosetta selected consensus amino acids from the biological MSAs shown in the logos in Fig. 6B at 32% of the positions for the Class I, but at only 7% for the Class II peptide, suggesting different tolerances to mutation of the 2-folds.

Discussion

The validity of our conclusions depends on establishing the authenticity of catalysis by wild-type and the designed 46-mer peptides described here. It is therefore important to outline arguments that the results may be artifacts. These include the following: (i) the data are very surprising; (ii) the rate accelerations are based on experimental data scarcely above the level of detection and may result from contamination by wild-type enzymes; (iii) turnover numbers for the designed 46-mers, which were purified as fusion proteins, are much higher than those of the wild-type 46-mers, which were purified from inclusion bodies, consistent with higher concentrations of contaminating full-length enzymes in the former; (iv) high concentrations of wild-type 46-mers place upper bounds on K_m values.

The following points summarize reasons we feel that the data represent authentic catalysis by both sets of 46-mers. (i) Comparable levels of full-length enzymes should exist in both 46-mer and MBP preparations; yet MBP controls show no activity. (ii) Activities of the designed 46-mers depend on the presence of wild-type residues in positions known from studies of the full-length enzymes to be essential for full catalytic activity. (iii) Increased activity results from higher concentrations only of designed 46-mers; neither MBP nor mutant controls exhibited this behavior. (iv) Quenching of ATP fluorescence by WT 46-mers cannot be explained by minute amounts of contaminating full-length enzymes. (v) There is qualitative agreement between the relative ATP affinities measured by fluorescence quenching and those determined by steady-state kinetics for the wild-type 46-mers. (vi) Upper bounds for K_m values for ATP substrates of the wild-type 46-mers are severalfold smaller than the corresponding values for full-length enzymes. (vii) K_m values for amino acid substrates of the designed peptides are substantially increased, relative to those of the full-length enzymes. (viii) The comparison between the WT and designed 46-mers is consistent with expectations that enforcing genetic complementarity by protein design might reduce the specificities of the WT peptides by increasing K_m values for amino acid substrates.

The data reported here furnish strong, complementary sources of evidence for the following conclusions. However, the results are important enough to merit validation by further experiments. Such experiments include characterizing multiple designed sense/antisense genes from Rosetta runs sampling more of sequence space, as well as combinatorial mutations, coded by complementary codons, of active-site residues that were fixed in this design.

Direct Experimental Validation of the Sense/Antisense Coding Hypothesis

Catalysis by both designed 46-mers directly substantiates the sense/antisense coding ancestry of Class I and II aaRS. Together with biochemical (1, 5, 7, 8, 10) and bioinformatic (11) evidence and the repeated failure to falsify the Rodin-Ohno hypothesis (1), they are an exacting, direct experimental test of the duality proposed (4) to underlie the ancestral aaRS classes and represent an important unification in biology.

Unique Information in a Gene Can Have Two Alternative Functional Interpretations

Comparable activities for products of both strands of the designed 46-mer gene directly confirm that two distinct interpretations exist for the information contained in some genes. Here, remarkably, the two gene products catalyze essentially the same reaction and with approximately the same rate accelerations. Yet, the two distinct catalysts are tightly linked through strict genetic complementarity. Complementary gene sequences impose such a strong constraint that one would expect such coding to be rapidly superseded, leaving computational modeling such as we have reported here as perhaps the only access to such extinct information.

Catalysts Like These 46-mers Were Essential for the Origin of Translation

We argued previously (5, 8) from model reactions (14) that the uncatalyzed rate of amino acid activation is ∼8.3E-9 s⁻¹ m⁻¹. However, the essential time independence of MBP and mutant peptide controls imply that we have approximated the uncatalyzed rate correctly with a slightly slower rate of ∼4.2E-9 s⁻¹ m⁻¹. Apparent second-order rate constants for the Class I and II peptides ∼6E-3 s⁻¹ m⁻¹ (Table 4) suggest that they accelerate amino acid activation 750,000–1,300,000-fold.

Uncatalyzed peptide bond formation from activated amino acids, 3 × 10⁻⁵ s⁻¹ m⁻¹ (12), is 2 orders of magnitude slower than the 46-mer catalyzed rates. Thus, the existence of similar catalytic peptides would eliminate amino acid activation as the kinetic barrier to peptide bond synthesis, even without a ribosome. Such catalysis could therefore have had a decisive and pleiotropic selective advantage, perhaps helping to select early evolutionary development of ribosomes, whose emergence might therefore have co-evolved later, with more sophisticated peptide-based aminoacyl-tRNA synthetases (e.g. synthetase Urzymes).

Sense/Antisense 46-mers May Represent a Transition between Uncoded and Genetically Coded Peptides

The small size of the 46-mers and their widespread conservation throughout the proteome (44) underscore the significance of their catalytic activities. They contain the Class I (PXXXXHXGH; a P-loop variant) and Class II (VTDVXXXXXR) phylogenetic signatures of longer, more sophisticated aaRS Urzymes; yet they are short enough to bridge the gap between Urzymes and simpler uncoded peptides. They might therefore be described as “Protozymes.”

Do Amino Acid Specificities of the Two Peptides Reflect Contemporary Synthetase Class Preferences?

Neither peptide offers a readily identifiable amino acid-binding site. However, different patterns of side chains and hydrogen bonding in parallel (Class I) versus antiparallel (Class II) β-structures provide a potential basis for differently sized pockets that might discriminate between sizes of otherwise structurally related amino acids, as proposed by Belrhali et al. (20, 45).

Catalysis Probably Arises from Transient Structures

The PDB files of the designed peptides output by Rosetta (supplemental material) closely resemble the native conformations that constrained them. The work of Mildvan and co-workers (28–31) suggests, however, that the predominant structures of homologous peptides in solution, and which can be determined by NMR methods, do not closely resemble the structures that the same peptides adopt in the native protein crystal structures, except in the presence of ligands. Thus, the peptide catalysts described here probably adopt transient structures in the presence of substrates that enable them to stabilize the transition states for amino acid activation. Such transient structures may actually differ markedly from what is observed in crystal structures (25, 26). That transient structures may be able to form tighter bonds to transition states has begun to be discussed in detail from both experimental (57) and theoretical (46) standpoints. In turn, this suggests that the peptides have significant normal modes allowing them to sample catalytically active conformations. Access to such structures by computational biology is thus an important and intriguing goal.

Experimental Recapitulation of Possible Assembly of Other Modular Components of Both Synthetase Families

Measurable 46-mer catalysis makes it possible to ask how the two additional modules of the Urzymes, related roughly to specific amino acid and tRNA binding, became assimilated by testing their effects in cis and in trans. Furthermore, complementation can be characterized in greater detail by measuring the energetic coupling between them using modular thermodynamic cycles (7, 43). Characterization of additional pairs of WT 46-mers and designed sense/antisense peptides also may help resolve the question of whether the constraint of genetic complementarity is inconsistent with high amino acid affinity.

Class I and II 46-mers May Have Spawned an Extensive Portion of the Contemporary Proteome

Unification of two entirely different aaRS superfamilies through their complementary genetic ancestry may signal more substantial relationships yet to be verified.

The corner between what in full-length crystal structures is the α-helix and the second β-strand of the Class I 46-mer occurs with minimal variation of amino acid, spacing, and three-dimensional packing in ∼125 different protein families from the Rossmannoid superfamily (44), the largest in the proteome (47–49). P-loop peptides studied by Mildvan (28–31) also arguably have structural homology to parts of the Class I aaRS 46-mers that bind ATP; the segment N-terminal to the first helix shares glycines in homologous positions before, within, and following the Class I HIGH signature. The root mean square deviation = 3.6 Å for 25 residues, including the P-loop, preceding β-strand, and subsequent α-helix, and superimposes two conserved Gly residues; see also Fig. 11A in Ref. 1.

These observations suggest that the Class I 46-mer has widespread distribution elsewhere in the proteome (44) and may thus have been ancestral not only to the Rossmannoid superfamily but also to other protein superfamilies. These include the enzymes characterized by Mildvan and co-workers (25–29) and a wider class of energy-transducing enzymes, including myosins (58), AAA⁺ motors (50, 51), STAND regulators of programmed cell death (52), and signaling GTPases (59, 60), all of which use P-loops to activate ATP and hence are arguably directly homologous to the Class I 46-mer. Further afield, the superfamily of metabolic enzymes in nucleotide biosynthetic pathways (53, 54) may also have its origins in peptides related to the Class I 46-mer.

One also expects that the Class II 46-mer would have driven a comparable adaptive radiation, generating homologies with a comparable range of superfamilies. The comparative anatomy of antiparallel β proteins, however, suggests a more plastic structure than the readily recognizable parallel β-α-β topology. We have noted previously that the motif 2 loop in the Class II 46-mer may be homologous to the nucleotide-binding sites in the actin/HSP70 superfamily (10). We note here as well that nucleotide-binding sites in the majority of nucleic acid polymerases, containing conserved aspartate Mg²⁺ ligands, have a similar β-α-β topology to that of the Class II motif 2 signature, albeit with a generally much longer α-segment.

Catalytic properties of the 46-mers also support previous suggestions (41) that the earliest catalytic peptides mobilized ATP and other NTPs for biosynthetic purposes. Thus, the peptides described here catalyze a synthetic reaction central to biology, and phylogenetic analysis suggests they may be among the most broadly conserved and hence perhaps the oldest surviving functional peptide motifs in the proteome.

Author contributions

C. W. C. Jr., L. L., and O. E. conceived the study. O. E., X. A., and B. K. designed the gene. L. L. isolated and V. W. performed fluorescence titrations and assayed the wild-type peptides. L. L. performed the mutagenesis of the active sites. M. C. purified the MBP used as a control. L. M.-R., K. G.-R., M. J.-R., and T. W. coordinated the purification and assay of the designed peptides. S. N. C. performed the bioinformatic analysis shown in Fig. 6. C. W. C. Jr., V. W., L. L., and T. W. wrote the paper. All authors reviewed and interpreted the results and approved the final version of the paper.