tRNA acceptor-stem and anticodon bases embed separate features of amino acid chemistry

abstract

The universal genetic code is a translation table by which nucleic acid sequences can be interpreted as polypeptides with a wide range of biological functions. That information is used by aminoacyl-tRNA synthetases to translate the code. Moreover, amino acid properties dictate protein folding. We recently reported that digital correlation techniques could identify patterns in tRNA identity elements that govern recognition by synthetases. Our analysis, and the functionality of truncated synthetases that cannot recognize the tRNA anticodon, support the conclusion that the tRNA acceptor stem houses an independent code for the same 20 amino acids that likely functioned earlier in the emergence of genetics. The acceptor-stem code, related to amino acid size, is distinct from a code in the anticodon that is related to amino acid polarity. Details of the acceptor-stem code suggest that it was useful in preserving key properties of stereochemically-encoded peptides that had developed the capacity to interact catalytically with RNA. The quantitative embedding of the chemical properties of amino acids into tRNA bases has implications for the origins of molecular biology.

Introduction

Nucleic acids are the medium in which living systems store information, and RNA was probably the first polymer to record such information. Life as we know it would not have been possible without the universal genetic code, which furnishes a translation table by which the instructions in mRNA are read out during protein synthesis. Implementation of the code depends on the recognition of tRNA by aminoacyl-tRNA synthetases (aaRS). The genetic code probably represents some of the earliest information to be stored in RNA, and it has often been pointed out that similar amino acids seem to have similar codewords.^1–4 The lack of focus of that similarity led us to question whether there might be a more quantitative relationship between the measurable properties of the amino acids and their tRNA sequences than is required by Crick's adaptor hypothesis.⁵

The codons of the common amino acids and in particular their second codeletters are arranged in such a way that hydrophilic amino acids, which often occur at the surfaces of proteins and are rarely found in the interior, tend to arise by spontaneous mutation from other hydrophilic amino acids, and conversely for non polar amino acids. It seems reasonable to suppose that these tendencies may have been helpful in maintaining structural stability during the evolution of globular proteins.^4,6,7 To establish experimental metrics for how amino acid side chains behave, one of us^6,8,9 has measured partition coefficients of side chain mimics between water and cyclohexane (hence ΔG_w>c; an index of their polarities) and between the vapor phase and cyclohexane (ΔG_v>c; which correlates closely with their sizes). The vapor phase represents a reference state in which the molecules exist in isolation at ordinary temperatures and pressures, and are essentially free of interactions with a solvent or with each other.¹⁰ The use of partition coefficients, rather than solubilities, avoids anomalies arising from idiosyncratic properties of the solid state.^8,10

In two recent papers^11,12 we described a procedure for estimating the distribution of amino acids between the surfaces and cores of folded proteins, based on these experimental transfer free energies. Previous quantitative relationships between polarity and folding fell short of accounting for the surface-to-core distributions observed for different amino acids in folded proteins. Improved predictive power emerges when both polarity and size are included in the model.

Curiously, aaRS come in 2 flavors, distinguished by their structures.¹³ The term “Class” is used to distinguish both the enzymes and the amino acids that they activate.¹ Polarity and size also help to rationalize the distinction between the 2 Classes of amino acids.¹¹ Class II amino acids occur significantly more frequently at the surfaces of proteins, whereas Class I amino acids occur more frequently in their cores.

When we looked at the coding elements in tRNA, we were startled to find that tRNA anticodon and acceptor-stem bases, separately and independently, encode these same 2 physical properties. The details of how this information is read out by interactions between tRNAs and their cognate aminoacyl-tRNA synthetases aaRS;^16–18 have been recognized, at least in principle, for several decades, culminating in an encyclopedia of tRNA “identity elements” compiled by Giegé and colleagues.^19,20 Comparison of the domain structures of tRNAs with those of aminoacyl-tRNA synthetases suggested that an earlier code was built into the tRNA acceptor stem sequences.²¹ The identification of its translation table might furnish clues about the evolution of contemporary genetic coding.^11,12

The acceptor-stem code

The hypothesis that the tRNA acceptor stem incorporates an alternative code arose largely from truncation experiments indicating that the anticodon is not required for functional recognition of tRNA. First, aaRS can catalyze the aminoacylation of minihelices—smaller, stem-loop RNA fragments derived from tRNAs that contain the acceptor stem.^22,23 More recently, one of us²⁴ demonstrated that Urzymes, small fragments containing aaRS active sites but lacking anticodon-binding domains, can also catalyze acylation of tRNAs. How does information in the tRNA acceptor stem actually specify different amino acids? The surprising answer appears to be that the tRNA acceptor stem probably coded for their sizes and an additional subset of amino acid descriptors. Those descriptors¹¹ include whether or not the amino acid is branched at the β-carbon, whether it is aliphatic, and whether it is equipped with a carboxylate group (Fig. 1). The anticodon bases form a complementary code for amino acid polarity.

Figure 1.

Key elements of the acceptor-stem code. Size (blue) is encoded primarily by bases 1 (pyrimidine favors larger size), 70 (G or C favor larger size), and the Discriminator base 73 (G or C favor larger size).²⁶ Beta branching (yellow) is determined by positions 2, 3, and 4. Carboxylate side chains (red) are encoded by bases 2 and 71. Different aspects of bases 2 and 3 serve to code both for β-branched (2,3=pyrimidine) and for carboxylate (2 = G or A; 3 = C or U) side chains. Different interpretations are given to each constellation of bases, depending on which of the two grooves interacts with the synthetase.

Digital coding BY tRNA identity elements

To uncover the relationships between the tRNA bases recognized by synthetases and the physical properties of the amino acids, we built design matrices with one row per amino acid and with columns containing either amino acid physical properties—experimental transfer free energies (ΔG_w>c and ΔG_v>c)—or digital representations of the bases forming tRNA identity elements.²⁰ Each base, n, is identified by 2 entries. The first entry (i.e., nG/A) is a 1 or -1 depending on whether the base forms 3 hydrogen bonds or 2 in a canonical Watson-Crick base pair. The second entry (nY/R) is a 1 if the base is a pyrimidine and -1 if it is a purine. The null character, 0, was used to indicate that a particular bit of information was not used for a given amino acid.

The listing of entries can be illustrated for the example of isoleucine, whose anticodons RAU generate the 7-term vector {1 –1 −1 –1 1 0 –1}. This vector begins with the amino acid's Class (see below)] (1), followed by 2 2-bit elements for each base: first for base 2, then for base 3, and finally for base 1, these bases having been ranked in that sequence empirically according to their relative powers of discrimination¹ .² . ²⁵ Here, 0 represents ambiguity associated with the first base, which can be either G or A. Acceptor stem coding proceeds in analogous fashion for the 8 bases in the stem plus the unpaired “discriminator” base D (usually base number 73³ ).²⁶

The two aaRS Classes generally recognize opposite faces of tRNA interacting with either the major or minor groove.²⁷ Omission of the distinction made possible by these different modes of recognition seriously degraded the performance of both acceptor stem and anticodon models.

Results of regression analysis of tRNA identity element coding

Amino acid properties can be compared with linear combinations of the digital information in tRNA identity elements to uncover the coding relationships in each region of the molecule. As noted previously in other contexts^28,29 amino acid side chain size and hydrophobicity are both continuous variables related to the code. Their correlations with tRNA identity elements can be established by regression analysis. Values of either 1 or 0 were used to denote amino acids with β-branched, aromatic, positively and negatively charged, amide, and/or aliphatic side chains. These columns subsequently served as categorical dependent variables.

Statistics programs (such as JMP,³⁰) provide automated tools (e.g. stepwise regression) for comparing possible predictors and accumulating P-values for each of them. The column with the smallest P-value explains the largest fraction of the variance in the dependent variable and was therefore chosen as the first predictor. As noted previously,²⁹ the most significant predictor among the tRNA anticodon models for ΔG_w>c is whether the middle codon base is a pyrimidine (Y) or a purine (R) i. e., 2Y/R. That single factor predicts 48% of the variance in experimental ΔG_w>c values, leaving 52% unexplained. ΔG_w>c values for Class I and II amino acids have almost exactly the same slope when plotted against this single predictor, indicating that there is no interaction between 2Y/R and Class.

Values of ΔG_w>c (calculated from the slope and intercept of the regression against 2Y/R) are then subtracted from the experimental values, leaving a “residual” unexplained variance of 52% that is presumed to result from the fact that other significant predictors have not yet been included into the model. These residual variances can be predicted quite precisely by including 2 additional columns, 2G/A and 3G/A. At this Stage 2, in contrast to Stage 1, predictions for Class I amino acids differ markedly from those for Class II amino acids. This step therefore introduces 3 new main effects and 4 new interaction terms, denoted by asterisks: 2G/A, 3G/A, Class, 2G/A*3G/A, Class*2G/A, Class*3G/A, and Class*2G/A*3G/A.

Taken by themselves, these latter predictors account for 64% of the variance in the original data. They also explain 86% of the residual unexplained variance remaining after Stage 1. Combining predictors from Stages 1 and 2 (Stage 3) accounts for 92% of the overall variance in ΔG_w>c. New predictors are added from the remaining information about the anticodon bases, until the ratio of a new coefficient to its standard error falls below a given significance level, indicating that additional predictors are fitting only noise.

Cross-validation

Numerous coefficients were used, as described above, to fine-tune the models, relative to the number of amino acids, leaving few degrees of freedom for estimating errors. To test for possible “overfitting” of that kind, we examined the noncanonical amino acids selenocysteine (Sec) and pyrrolysine (Pyl). Both amino acids are incorporated into proteins by codon dependent translation, co-opting either a stop codon (Pyl; UAG^31,32) or a redundant and rarely used serine codon (Sec; AGU^33-35) to enlarge the genetic code. These noncanonical amino acids involve changes in both the cognate tRNAs and the enzymes that recognize them. Thus, selenocysteine is synthesized from the aminoacylated substrate, seryl-tRNA^Sec by 2 different enzymes that are not aaRS but recognize aminoacylated tRNA^Sec. Pyrrolysine is synthesized from 2 molecules of lysine by a cassette of 4 biosynthetic enzymes³⁶ and has its own pyrrolysyl-tRNA synthetase. We used identity elements identified for O-phosphoseryl tRNA kinase (PSTK;³⁴) as the step most critical in tRNA selection during post-acylation synthesis.

ΔG_w>c values were estimated as follows for Sec and Pyl from their chemical similarity to canonical amino acids. The pK_a value of the selenol group is ∼5.³⁷ Thus, it is negatively charged at pH 7 and resembles Asp in that respect and so was given a value of +7 kcal/mole. Pyrrolysine is a lysine derivative in which the ε-nitrogen is joined by an amide linkage to 3-methylpyrroline-2-carboxylic acid. It seemed reasonable to assign to Pyl a ΔG_w>c value (5.7 kcal/mole) intermediate between that of glutamine and lysine. We estimated ΔG_v>c values for both non-canonical amino acids from their masses, using the linear correlation, described above, between mass and ΔG_v>c for the canonical 20 amino acids (R² = 0.89). We then compared predicted masses and transfer free energies from both classes of models using identity elements from tRNA^Pyl and tRNA^Sec.

Independent coding by the tRNA acceptor stem and anticodon

Identity determinants in both the acceptor stem and anticodon correlate with regression models for both ΔG_w>c and ΔG_v>c. This result is unsurprising, because the 2 sets of tRNA identity elements encode the same amino acids. What surprised us was that cross-validation results for the 2 coding sets revealed that the acceptor stem code predicts only the ΔG_v>c values for Pyl and Sec, whereas the anticodon code predicts only their ΔG_w>c values. Thus, the 2 regions of tRNA apparently derive their base compositions according to independent criteria.

Predictive models for categorical attributes of the amino acids showed that aliphatic vs non-aliphatic and carboxylate side chains are encoded by both acceptor stem and anticodon, side chain β-branching is encoded only in the acceptor stem; and the anticodon encodes basic, charged, aromatic, and amide side chains. Because they are categorical and qualitative, rather than quantitative, fewer predictors are necessary to specify them without ambiguity so these distinctions are less vulnerable to overfitting, and are more decisive statistically than the numerical terms based on size and hydrophobicity. They also highlight aspects of the code—side chain β-branching and carboxylates—that may have preceded expansion to the full universal code.

Stages in the acquisition of information by tRNA and mRNA

Evolution would seem more likely to have brought order out of chaos if it was able to build on pre-existing order. The new evidence from tRNA sequences suggests that coordinated control, first over amino acid sizes and later over polarities, may have given nature the tools to find a path to modern enzymes and other proteins. This stepwise development of the genetic code would have paralleled the hierarchy of structure in proteins proposed by Linderstrom-Lang.³⁸ One can envision mRNA sequence evolution as having progressed through successive stages that conferred selective advantages: first, peptides with stable secondary structures; then, molten globules with approximately-defined cores and surfaces capable of transiently forming binding sites;³⁹ and finally well-defined tertiary packing to stabilize binding sites with the binding discrimination needed for proficient catalysis and other functions.

The existence and distinctive properties of the acceptor-stem code¹¹ imply that the code, and hence the tRNAs and the synthetases, all passed through at least 2 successive stages. The acceptor-stem code (Fig. 1) incorporates precisely those characteristics—size and β-branching—that confer on peptides a greater or lesser tendency to form extended β-conformations. That relationship may be more than coincidental,¹¹ since extended peptide conformations tend to favor a stereochemical complementarity between peptides and double-stranded RNA that furnished an early model for templated cross-catalysis by rudimentary oligopeptides and oligonucleotides.^40,41

Selecting amino acids based primarily on size and β-branching would have introduced secondary structure preferences into the earliest gene sequences. If those preferences sustained the ability of extended polypeptides to bind RNA they would have strengthened the importance of indirect coding, before there was a clear separation between genotype and phenotype. Thus, the earliest gene sequences, and especially indirect coding by the acceptor stem sequences, may have been less tightly coupled to mRNA sequences than in modern organisms.

The earlier appearance of an acceptor-stem code, before the emergence of the universal genetic code,²¹ is supported experimentally by (i) the reciprocal biochemistry of minihelix acylation by full length synthetases,^22,42–44 and (ii) the acylation of full-length tRNAs by truncated synthetases called Urzymes.²⁴ Di Giulio^45,46 and Rodin & Ohno⁴⁷ have described models in which the anticodon loop arose from pre-existing, acceptor-stem RNA hairpins. The acceptor stem code may have helped mediate an early transition—from stereochemical coding of peptides directly by RNA to a rudimentary indirect coding mechanism^40,41,48—paving the way for subsequent development of the universal genetic code and the interpretation of mRNAs by a ribosome.

Revisiting the central dogma

Within a Darwinian context (Fig. 2), these findings imply the existence of a tetrahedral network relating biology to chemistry. The 4 nodes (bold face) are: amino acid properties, protein folding (the proteome), a programming language (the code), and a set of programs (the genome). We showed recently that amino acid physical chemistry drives both protein folding and the base sequences of tRNA. By implication, it also dictates how tRNA bases are recognized by aminoacyl-tRNA synthetases, and which mRNA sequences will encode foldable, functional proteins (the 3 unidirectional arrows). The RNA population (red-tint) incorporates 2 distinct types of information, as follows. The genetic code furnishes a translation table relating tRNA bases to amino acid physical properties, and serves as a programming language that is read by aminoacyl-tRNA synthetases. Gene sequences, on the other hand, are programs that have evolved to exploit amino acid chemical behavior. Functions in the proteome result from physical properties of amino acids that therefore function like the elements in a periodic table, except that they dictate protein folding (blue-tint).

Figure 2.

Network analysis of the Central Dogma of Molecular Biology. 1. Protein folding depends on both amino acid polarity and size.¹² These two properties play a role analogous to those of the elements in chemistry, in that the amino acids form a kind of “periodic table” on which the folding of proteins is based. 2. tRNA bases encode amino acid size and polarity separately.¹¹ The acceptor stem codes for amino acid sizes; the bases of the anticodon code for amino acid polarities. Amino acid properties therefore dictate how tRNA bases are recognized by aminoacyl-tRNA synthetases. 3. tRNA codes are related to protein folding. Together, statements (1) and (2) imply that for the aminoacyl-tRNA synthetases, the code (and mRNA sequences) must define folded structures capable of binding specifically to particular tRNAs in order to read the language of genetics. The bi-directionality of this arrow illustrates a somewhat deeper self-referential element than those identified in molecular biology by Hofstadter as generators of complexity according to Gödel's incompleteness theorem.⁴⁹ 4. Gene sequences (mRNA) are analogous to computer programs. Genetic instructions assemble amino acids according to their physical properties in ways that, when translated according to codes in tRNA (i. e., the programming language), yield functional proteins (enzymes, switches, regulators). 5. mRNA sequences (i.e., the genotype) determine amino acid sequences in proteins, and hence how amino acid sizes and polarities are exploited to produce different folded proteins. The spontaneous folding of amino acid sequences gives rise to functions (i.e., the phenotype) that ultimately determine whether or not a particular sequence survives natural selection. Changes accumulated in gene sequences result from selection acting on the phenotype. (4) and (5) localize how selection incorporates information about amino acid behavior into gene sequences, hence depict the evolutionary dimension in biology. 6. The evolution of mRNA sequences only makes sense in the context of the translation table (or programming language) established by the genetic code.

Each node of the schematic tetrahedron in Fig. 2 is connected directly to the others. Each triangle can be interpreted differently. The upper triangle relates to the Central Dogma envisioned by Crick. The new findings anchor that triangle to the physical chemistry of amino acids. The amber triangle (1,2,3) represents invariant chemical properties. The green triangle, (1,4,5) reflects the dynamic element introduced into biology by selection. As the 2 distinct types of information in RNA (language and programs) are both derived from amino acid properties, it seems likely that they developed concurrently.

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

Acknowledgments

This work was supported by National Institutes of Health Grants GM78227 (to C.W.C.) and GM18325 (to R.W.).