Evolution of the genetic code

ABSTRACT

Diverse models have been advanced for the evolution of the genetic code. Here, models for tRNA, aminoacyl-tRNA synthetase (aaRS) and genetic code evolution were combined with an understanding of EF-Tu suppression of tRNA 3^rd anticodon position wobbling. The result is a highly detailed scheme that describes the placements of all amino acids in the standard genetic code. The model describes evolution of 6-, 4-, 3-, 2- and 1-codon sectors. Innovation in column 3 of the code is explained. Wobbling and code degeneracy are explained. Separate distribution of serine sectors between columns 2 and 4 of the code is described. We conclude that very little chaos contributed to evolution of the genetic code and that the pattern of evolution of aaRS enzymes describes a history of the evolution of the code. A model is proposed to describe the biological selection for the earliest evolution of the code and for protocell evolution.

1. Introduction

A model is presented for evolution of the genetic code based on analyses of tRNA and aminoacyl-tRNA synthetase (aaRS) evolution. The model is highly detailed and provides simple rules for filling code sectors. Strong selection rules are also apparent for the earliest evolution of the code.

A primordial tRNA^Pri was comprised of ordered sequences, GCG, CGC and UAGCC repeats and inverted repeats with 7-nt U-turn loops (homologous 17-nt anticodon and T stem-loop-stems) [1–6]. With the exception of a few anticodon loop and T loop bases, the tRNA^Pri sequence is completely known. Three tRNA evolution models are considered here [1,2,7–11]. Only the 3-31-nt minihelix model can be correct. The 3-31-nt minihelix model has been referred to as a theorem [2]. There are no theorems in evolutionary biology, but the 3-31-nt minihelix model for tRNA evolution is very close to being one.

AaRS enzymes attach amino acids to the 3ʹ-end of tRNAs [12]. Much has been published on evolution of aaRS (i.e. GlyRS-IIA; IIA indicates the class (I or II) and structural subclass (i.e. A-E)) [1,3,13–18]. We have simplified the understanding of aaRS evolution and brought it in line with the evolution of the genetic code [1,3].

Evolution of tRNA and the genetic code provides new models for evolution of life on Earth and the pre-life to life transition. In agreement with some others, we posit that Archaea are the oldest organisms, and Archaea are the most similar to the last universal common (cellular) ancestor (LUCA) (Figure 1) [1,3,19–21]. Our interest in this issue comes from studies of earliest evolution of transcription [22,23] and translation systems [1,3,24]. We find that tRNAomes (all the tRNAs of an organism) are simpler in archaeal systems relative to bacterial systems [2,4,24,25]. AaRS enzymes are closer to root sequences in archaeal systems, and aaRS evolution falls more in line with the simpler archaeal genetic code [1,3,24,26]. Archaeal TFB is a homolog of bacterial σ factors [23,27], indicating that evolution of bacterial transcription systems may have largely drove divergence of Archaea and Bacteria [28]. In this work, we concentrate on archaeal systems for these reasons.

Figure 1.

A working model for the pre-life to life transition and for divergence of Archaea and Bacteria. A major driving force for the divergence of Bacteria and Archaea is posited to have been divergence of transcription systems [28]. EF-Tu is highlighted (small green arrow) because EF-Tu evolution allowed expansion of the genetic code from an 8-aa bottleneck to the standard code. Abbreviations: LLPS) liquid–liquid phase separation; G) glycine; GADV) glycine, alanine, aspartic acid, valine; EF) translational elongation factor; IF) translational initiation factor; TBP) TATA-box binding protein; TF) transcription factor; DNAP (Pol)) DNA polymerase; 1 Ga) 1 billion years ago

2. Methods

Sequences of tRNAs were obtained from the tRNA database [29] and the genomic tRNA database [30,31]. Typical tRNA diagrams were generated using tRNAdb tools and modified as necessary. Longer V loops in type II tRNA^Leu and tRNA^Ser were analyzed using WebLogo 3.7.4 [32]. Molecular graphics was done using UCSF ChimeraX [33,34].

Evolution of aaRS enzymes was analyzed using the Phyre2 protein fold recognition server (see Figure 7c) [35]. Phyre2 identifies nearest and distant matches to a seed sequence, in the RCSB protein data bank. Phyre2 is a very useful tool for identifying close and distant protein family members that are related by both structure and sequence. Using only sequence-based tools, it is difficult to relate aaRS enzymes, which were driven to differentiate in order to establish and maintain translational accuracy of tRNA charging. Phyre2 was used to build a lineage of class I aaRS enzymes in which all class I aaRS were connected by both close and distant metrics (Phyre2 homology scores). Phyre2 could be used to build a model for a class II aaRS lineage. Some distantly related class II aaRS, however, could not be scored to one another. For instance, class IIA and class IID enzymes could not be scored as homologs without connecting intermediate class II aaRS. Because class I and class II aaRS have different folds, Phyre2 could not be used to identify homology of class I and class II aaRS enzymes. We used standard NCBI (National Center for Biotechnology Information) tools, such as Blast, to relate GlyRS-IIA to IleRS-IA and ValRS-IA (see Figure 7a-b) [1,3].

Figure 7.

Evolution of aaRS enzymes. A) Local alignment of a class I IleRS-IA and a class II GlyRS-IIA aaRS. Identities are shaded red and similarities are shaded yellow. Mbr) Methanobacterium bryantii; Mco) Methanobacterium congolense. B) Class II and class I aaRS with incompatible folds are simple sequence homologs. Red boxes indicate sequence homologies in the multiple alignment. Two Zn fingers are indicated. The position of the alignment in panel A is shown. C) Evolution of aaRS enzymes based on Phyre2 homology scores [1,3]. Distances in the map represent structural and genetic relatedness. In Archaea, related aaRS enzymes mostly cluster according to genetic code columns (boxes; numbers indicate genetic code columns; some rows are specified). Yellow highlighting indicates aaRS with editing active sites. Red arrow indicates homology of GlyRS-IIA and ValRS-IA. (A) Archaea-specific; (B) Bacteria-specific. (?) indicates that TyrRS-IC and TrpRS-IC may be derived from a primitive ArgRS-ID

3. Pre-life evolution of transcription, metabolism and translation

There are some shared concepts comparing the earliest evolution of transcription and translation systems. On Earth, complexity was often generated from repetition of a motif. We imagine a pre-life mechanism to duplicate and multimerize RNAs (i.e. by ribozyme-mediated ligation and replication), often resulting in duplication of a common or related sequence [22,23,28,36]. If the RNA was protein-encoding, dimerization would generate a protein motif duplication. In this way, a β-β-α-β unit was duplicated to create a β-β-α-β – β-β-α-β motif refolded into a 6-β-sheet barrel [37–40]. Double-Ψ-β-barrels were generated in this way. Cellular RNAPs are 2-double-Ψ-β-barrel type enzymes [27,28,41,42]. PolD is a 2-double-Ψ-β-barrel type replicative DNAP from Archaea, and may be the most ancient replicative DNA polymerase [42–45]. RNA template-dependent RNAPs can also be of the 2-double-Ψ-β-barrel type, and these appear to be the oldest form of the enzyme class [46]. In Archaea, TFB includes a duplication of a helix-turn-helix motif ((HTH)₂), also referred to as a cyclin-like repeat. In Bacteria, σA is a homolog of TFB that appears to be derived from a (HTH)₄ repeat, which probably arose as a TFB (HTH)₂ duplication [22,23,36]. TBP was generated by duplication of a motif encoding multiple β-sheets and may be coevolved with DNA [47]. We consider TFB and TBP to be founding general transcription factors. It appears that Bacteria evolved sigma factors from TFB and lost TBP and that these were defining events in the divergence of Archaea and Bacteria. In many ways, Bacteria seem to be more successful prokaryotes than Archaea. For instance, many Archaea appear to be pushed into extremophile environments on Earth. We posit that Bacteria evolved from Archaea, and that Archaea are most similar to LUCA (Figure 1) [28]. For many factors and functions, Bacteria are simplified relative to Archaea, presumably due to genetic loss.

Much of core metabolism, including the glycolysis pathway and the citric acid cycle, evolved around (β-α)₈ barrels (i.e. glycolysis; TIM (triosephosphate isomerase) barrels) and (β-α)₈ sheets (i.e. citric acid cycle; Rossmann folds). We posit that Rossmann fold (β-α)₈ sheets were refolded from (β-α)₈ barrels. We posit that (β-α)₈ barrels and sheets were initially generated from two serial duplications of β-α-β-α motifs [22,28].

The evolution of tRNA from ligation of 3–31-nt minihelices, two of which were identical, is described below. Because of our experience with evolution of transcription systems, we searched for repeating motifs in tRNAs and found them easily. In pre-life, mostly, ribozyme-dependent mechanisms must have existed to replicate 31-nt minihelices and tRNAs. We posit that genetic code evolution was mostly driven by replication of tRNAs, mutation of the tRNA anticodon and coevolution of tRNAs with aaRS. AaRS enzymes evolved by a chaotic pathway, described below. Class I aaRS have their active site mounted on a platform of parallel β-sheets [12]. For this reason, class I aaRS are often referred to as “Rossmann folds”, but this is improper, as described below. Remarkably, the lineages of aaRS enzymes in Archaea give the pattern of genetic code evolution. The somewhat more complex pattern of genetic code evolution in Bacteria can be derived from the older archaeal pattern [1,3].

4. Evolution of tRNA

4.1. Concepts

We posit that evolution of tRNA, from an RNA – and ribozyme-dominated world, laid the foundation for evolution of the genetic code. We posit that the genetic code sectored according to the tRNA anticodon. Initially, code columns were selected because the central anticodon base (2^nd position) was easiest to read on a primitive ribosome. Initially, both the 1^st and 3^rd anticodon positions were read as wobble positions with pyrimidine-purine discrimination. Evolution of EF-Tu suppressed wobbling at the 3^rd anticodon position allowing expansion of the code. Because of the pathway to evolution of tRNA, the anticodon is read in a register of 3-nt, so 2-nt code registers could not be supported using tRNAs. Wobbling and code degeneracy are described by the evolution of tRNA reading on a primitive ribosome and coevolution of EF-Tu. Sequence analyses of type II tRNAs with longer V loops in ancient Archaea provides reasonable models for serine jumping during evolution of the code. Sequences of tRNAs in Archaea show that tRNA evolved from repeat and inverted repeat sequences. Therefore, before evolution of tRNA and protein encoding, there must have been ribozyme-based mechanisms to generate RNA repeats and inverted repeats. In this way, tRNA is a central key to understand the pre-life to life transition and evolution of the genetic code. TRNA is uniquely suited as a genetic adapter to support evolution of a genetic code. To generate a genetic code with a different adapter than tRNA presents many problems that we would not know how to solve.

4.2. The 3-31-nt minihelix model

The pathway of evolution of tRNA has been controversial, but tRNA evolution is essential to grasp, in order to understand the pre-life to life transition and evolution of the genetic code. Here, three models are considered, but we focus on one model, the 3–31-nt minihelix model [1,2]. So far as we can discern, the 3–31-nt minihelix model is the only viable model, and alternate models are falsified. The alternate models can be described as the Uroboros model [7,48,49] and the 2-minihelix model [8,10,50]. Both of these models are accretion models, in which tRNA evolves in expanding and/or contracting segments. Because proposed expansion and contraction segments in tRNA would derive from highly ordered sequences (i.e. repeats and inverted repeats), no accretion model can reasonably describe early tRNA evolution. In an accretion model, expansions and contractions would need to result in ordered sequences [1–3]. Furthermore, analyses of archaeal tRNA sequences provides an irrefutable record of the 3–31-nt minihelix model.

The 3–31-nt minihelix model is summarized in Figure 2. 3–31-nt minihelices of mostly known sequence were ligated to form a 93-nt tRNA precursor, which was then processed by internal 9-nt deletion(s) into type I and type II tRNAs [2]. 31-nt minihelices were comprised of a 5ʹ-7-nt acceptor stem, a 17-nt core, and a 3ʹ-7-nt acceptor stem. The sequence of the 5ʹ-7-nt acceptor stem was originally GCGGCGG, which is a truncated GCG repeat. The sequence of the 3ʹ-7-nt acceptor stem was originally CCGCCGC, which is a complementary, truncated CGC repeat. For the D loop, the 17-nt core sequence was originally UAGCCUAGCCUAGCCUA, which is a truncated UAGCC repeat. Remarkably, the anticodon and T loop 17-nt core sequences were both originally close to CCGGGUU/AAAAACCCGG (/indicates a U-turn in a 7-nt loop). These 17-nt sequences form a stem-loop-stem with 5-nt 5ʹ-stems (CCGGG), a 7-nt U-turn loop (~UU/AAAAA) and 5-nt 3ʹ-stems (CCCGG). There is only slight sequence ambiguity in the primordial 7-nt U-turn loop, not in the stems.

Figure 2.

The 3–31-nt minihelix model for evolution of tRNA. Sequences are primordial but remain represented in archaeal tRNAs. A 93-nt tRNA precursor formed from ligation of 3–31-nt minihelices, as shown. Type I and type II tRNAs were processed by 9-nt internal deletion(s) from the 93-nt precursor. Minihelix world was preceded by polymer world. Colors: green) 5ʹ-acceptor stems (5ʹ-As) and derived 5ʹ-As*; magenta) D loop; cyan) 5ʹ-stem for the anticodon (Ac) and T stem-loop-stems (SLS); red) anticodon and T loops; purple) anticodon; cornflower blue) 3ʹ – stems for the anticodon and T stem-loop-stem; yellow) 3ʹ-As* and 3ʹ-As sequences. Arrow colors: red) internal deletion endpoints; blue) U-turns; cyan) discriminator (D); and gold) amino acid placements

Type II tRNAs have a longer V loop (V for variable) relative to type I tRNAs [4]. The model shown in Figure 2 describes the evolution of both type I and type II tRNAs. To generate a type II tRNA, the 93-nt tRNA precursor was processed by a single 9-nt internal deletion, as shown. The 9-nt deletion occurred within ligated 3ʹ – and 5ʹ-7-nt acceptor stems. The 5-nt segment that remains after deletion (originally GGCGG) became the last 5-nt of the D loop region just before the anticodon stem-loop-stem. The original type II tRNA, therefore, was 84-nt before addition of 3ʹ-ACCA (A is the primordial discriminator base), presumably via ligation. The type II tRNA V loop, therefore, was initially 14-nt (7-nt + 7-nt). 14-nt remains a dominant length of tRNA^Leu V loops in Archaea.

To generate type I tRNA from the 93-nt precursor, required two internal 9-nt deletions, as shown [2]. The 5ʹ-9-nt deletion was identical to the processing event in type II tRNAs. The 3ʹ-9-nt internal deletion in type I tRNA was also within ligated 3ʹ – and 5ʹ-7-nt acceptor stem segments. The 9-nt deletion generated the type I tRNA V loop, which was initially 5-nt (originally CCGCC). The primordial type I tRNA was 75-nt before addition of 3ʹ-ACCA.

Folding into the tRNA L-shaped structure brought the D loop into contact with the V loop and the D stem in contact with the V region, causing a small number of systematic changes in tRNA sequences [2]. The 5ʹ-acceptor stem remnant (5ʹ-As*) initially changed from GGCGG to GGGCG to form the D stem and to break base pairing contacts to the V loop (numbered V1-V5). The V loop (3ʹ-As*) changed with time from CCGCC to ~UGGUC. The V1 base is often U to form a wobble pair with G26. The V5 base tends to remain C to form the Levitt reverse Watson-Crick base pair to G15. Statistical tests support the homology of bases 3–7 of the 5ʹ-acceptor stem to the 5ʹ-As* sequence, with a p-value of 0.001 (highest indication of homology). Statistical tests support the homology of bases 1–5 of the 3ʹ-acceptor stem to the 3ʹ-As* sequence (the V loop), with a p-value of 0.001 (highest indication of homology) [5].

4.3. Type I tRNA

Figure 3a shows a type I tRNA colored according to the 3–31-nt minihelix model. Notice that the anticodon stem-loop-stem and the T stem-loop-stem are homologs (cyan-red-cornflower blue segments) with 7-nt U-turn loops, as indicated in the model (Figure 2). As in the model, the 5ʹ-acceptor stem is a homolog of the 5ʹ-As* sequence, and the 3ʹ-As* is a homolog of the 3ʹ-acceptor stem sequence (compare green and yellow segments).

Figure 3.

Type I tRNA. A) structure of a type I tRNA colored according to the model (Figure 2). B) A typical type I tRNA diagram from Pyrococcus furiosis [29]. Typical tRNA diagrams from the tRNAdb website are exported as DNA sequences rather than RNA. Arrow colors are as in Figure 2

Figure 3b shows a typical tRNA diagram (as DNA sequence) from Pyrococcus furiosis [29], which is an ancient Archaeon. The typical tRNA sequence is almost identical to the tRNA^Pri sequence shown in the model (Figure 2), indicating that the model is correct. The D loop sequence is UAGCNUAGCC, indicating conservation of the UAGCCUAGCC repeat sequence of tRNA^Pri. The homology of the anticodon stem-loop-stem (CCGGNCU/NNNGANCCGG) and the T stem-loop-stem (CCGGGUU/CAAAUCCCGG) is obvious by inspection. Statistical tests support this homology, with a p-value of 0.001 (highest indication of homology) [5]. The typical 5ʹ-acceptor stem sequence is GCGGCGG, identical to the tRNA^Pri sequence. The 3ʹ-acceptor stem sequence is CCGCNNC, consistent with the tRNA^Pri sequence (CCGCCGC). Acceptor stem sequences vary among tRNAs because acceptor stems are determinants for amino acid placements by aaRS enzymes at 3ʹ-ACCA. The anticodon is highlighted (purple) because the genetic code evolved around the tRNA anticodon and its reading on the evolving ribosome.

4.4. TRNA^Gly was the first tRNA

The closest tRNA in Archaea to tRNA^Pri is tRNA^Gly (Figure 4) [1–3,24,25]. Figure 4a shows a typical tRNA^Gly from three Pyrococcus species. The typical sequence of tRNA^Gly is almost identical to tRNA^Pri. The sequence alignment is shown in Figure 4b. The tRNA^Pri, typical tRNA^Gly (Pyrococcus) and typical tRNA (Pyrococcus furiosis) are nearly identical sequences. This result indicates that tRNA^Gly was the first tRNA and that all archaeal tRNAs radiated from tRNA^Gly. This observation is relevant to the evolution of aaRS enzymes and the genetic code, as described below.

Figure 4.

TRNA^Gly was the first tRNA. A) A typical tRNA^Gly from three Pyrococcus species (P. furiosis, P. abyssi and P. horikoshii; 9 sequences). B) TRNA^Pri, tRNA^Gly and tRNA^Typical (P. furiosis) are close homologs. TRNA secondary structure is indicated. Colors and arrows are as in Figure 2

The typical D loop sequence of Pyrococcus tRNA^Gly is UAGUCUAGCCUGGUCUA (D1 to D17) versus UAGCCUAGCCUAGCCUA in tRNA^Pri. These sequences are nearly identical. The D12 A→G shift from the primordial sequence appears to be universal in Archaea. D12G (19 G in standard tRNA numbering) intercalates into the T loop between 57A and 58A ((54-UU/CAAAU-60); standard numbering) and forms a specific H-bond contact, explaining the A→G sequence change. Apparently, 19 G is a preferred intercalating base to the primordial 19A. Note that the lengths of the D loop are identical in tRNA^Gly and tRNA^Pri. Standard tRNA numbering can be somewhat confusing compared to tRNA^Pri because standard numbering is based on a 2-nt deletion in the D loop in eukaryotic tRNAs.

4.5. Type II tRNAs

In Archaea, tRNA^Leu and tRNA^Ser are type II tRNAs, with longer V loops [4]. As we have shown previously, tRNA^Leu is most similar in overall sequence to type II tRNA^Pri (Figure 5). Figure 5a shows a tRNA^Leu from Pyrococcus horikoshii (PDB 1WZ2). The typical V loop has the 14-nt sequence UCCCGUAGGGGUUC (V1-V14). The V loop sequence varies from the primordial 14-nt CCGCCGCGCGGCGG because the primordial sequence can pair along its entire length, which would be awkward for tRNA folding and for functional contacts (i.e., with aaRS enzymes). Instead, the tRNA^Leu V loop evolved to form a short stem (i.e., 3-nt) and loop (i.e., 4-nt). Also, V1C typically became V1U to form a wobble pair with 26 G, and V14G became V14C to form a Levitt reverse Watson-Crick base pair with 15 G. Statistical tests support the model that the V loop is derived from a 3ʹ-acceptor stem ligated to a 5ʹ-acceptor stem (Figure 2), with a p-value of 0.001 (the highest indication of homology) [4].

Figure 5.

Type II tRNA. a) Structure of tRNA^Leu from Pyrococcus horikoshii. The larger V loop of type II tRNA is the 7-nt yellow segment linked to the 7-nt green segment. b) Typical tRNA^Leu from three Pyrococcus species (P. furiosis, P. abyssi and P. horikoshii; 15 sequences). Light green arrows indicate unpaired bases (V12U and V13U) just before the Levitt base pair (V14C = 15 G) (thin red line). Parentheses indicate paired bases; * indicates loop. Arrows and colors are as in .Figure 2

TRNA^Leu has a 5ʹ-acceptor stem with the typical sequence GCGGGGG versus GCGGCGG in tRNA^Pri. As in tRNA^Gly (Figure 4), the typical Pyrococcus tRNA^Leu D loop (Figure 5b) includes no deleted bases, with the 17-nt typical sequence UUGCCGAGCCUGGUCAA versus UAGCCUAGCCUAGCCUA in tRNA^Pri. These sequences are very similar.

The tRNA^Ser V loop evolved from the tRNA^Leu and tRNA^Pri sequences, in order to be distinguished from the tRNA^Leu V loop. Notably, SerRS-IIA interacts with the tRNA^Ser V loop directly as a determinant in order to recognize and accurately charge tRNA^Ser [12]. By contrast, the tRNA^Leu V loop is an anti-determinant to reduce inaccurate aminoacylation of tRNA^Leu by SerRS-IIA. LeuRS-IA recognizes the opposite face of tRNA^Leu from the expanded V loop, so LeuRS-IA recognizes other determinants such as the acceptor stem and discriminator base. A comparison of tRNA^Ser and tRNA^Leu V loop sequences is shown in Figure 6. Figure 6a is a typical tRNA^Ser diagram from 3-Pyrococcus species. Figure 6b shows an alignment of typical tRNA^Ser and tRNA^Leu V loop sequences versus tRNA^Pri. Figure 6c shows the comparison of tRNA^Ser and tRNA^Leu V loop sequence logos. We could not find a suitable tRNA^Ser structure (i.e., from an ancient Archaeon) to compare to the tRNA^Leu structure in Figure 5a, for instance, to compare the expected different trajectories of the V loops.

Figure 6.

Comparison of tRNA^Leu and tRNA^Ser V loops. A) A typical tRNA^Ser from three Pyrococcus species (P. furiosis, P. abyssi and P. horikoshii; 12 sequences) (as DNA sequence). B) The alignment compares V loops of tRNA^Pri to tRNA^Leu and tRNA^Ser typical sequences. Parentheses indicate stems; * indicates loops. C) Sequence logos comparing tRNA^Leu and tRNA^Ser V loops (as DNA sequences). Light green arrows indicate unpaired bases after the V loop stem

Accurate recognition of tRNA^Ser by SerRS-IIA is important in understanding the evolution of the genetic code. Serine is the only amino acid that sectors within two code columns, as described below. We posit that differences in sequence, stem positions and unpaired bases were important to discriminate the tRNA^Ser and tRNA^Leu V loops. Specifically, in Pyrococcus, tRNA^Ser has a single unpaired base just 3ʹ of its short stem, while tRNA^Leu has two unpaired bases (light green arrows in Figures 5 and Figures 6). This difference should change the trajectory of the type II V loop stems. Also, the tRNA^Ser V loop is G-rich at its 5ʹ stem, while the tRNA^Leu V loop is C-rich at its 5ʹ stem. These and possibly other differences in type II V loops are expected to contribute to discrimination. The sequence logos in Figure 6c show that, in their major features, V loops are highly conserved in Pyrococcus. Solution of a structure of tRNA^Ser from a Pyrococcus species would contribute to this discussion.

5. Evolution of aaRS enzymes

5.1. Concepts

Aminoacyl-tRNA synthetases (aaRS) place amino acids at the 3ʹ-end of tRNAs [12]. The idea behind this paper is that insight can be gained into the evolution of the genetic code based on coevolution of aaRS enzymes, tRNAomes and EF-Tu [1,3]. Using the Phyre2 protein fold recognition server [35], we were able to establish simplified pathways of aaRS evolution that appear to describe routes for genetic code evolution often within code columns (2^nd anticodon position). Also, we noted sequence homology between class I and class II aaRS enzymes with different folds [26], which initially was unexpected. Class I aaRS enzymes have been termed “Rossmann folds”, but this is a mischaracterization, as we describe.

5.2. Class II and class I aaRS enzymes

There are two structural classes of aaRS enzymes, termed class II and class I, and multiple structural subclasses (i.e., A-E) [12]. Some aaRS enzymes have a separate active site from the aminoacylating active site that removes non-cognate amino acids from their cognate tRNA. This process is referred to as “proofreading” or “editing”. Remarkably, in Archaea, aaRS enzymes that edit non-cognate amino acids are found in the left half of the genetic code, in columns 1 and 2. Column 1 encodes hydrophobic amino acids Val, Met, Ile, Leu and Phe. Column 2 encodes neutral amino acids Ala, Thr, Pro and Ser. Ser is found also in column 4. We posit that Ser jumped from column 2 to column 4 in establishment of the code. In Archaea, ProRS-IIA does not include an editing active site, but ProRS-IIA does edit in Bacteria. Because neutral and hydrophobic amino acids are limited in forming specific hydrogen bonds and ion pairs, editing may be necessary to reduce tRNA charging errors.

Evolution of aaRS enzymes is described in Figure 7. In Figure 7a, a detail of an alignment of IleRS-IA from Methanobacterium bryantii to GlyRS-IIA from Methanobacterium congolense is shown. The e-value for the alignment is 4 × 10⁻¹², so, qualitatively, about a 1 chance in 2.5 × 10¹¹ of the alignment being due to random chance rather than homology. Class I and class II aaRS enzymes, however, have incompatible folds, so these enzymes were not thought to be homologous. Class II aaRS appear to be the older fold, indicating that class I aaRS may be derived from class II aaRS [12]. In class II aaRS, the active site is mounted on a scaffold of antiparallel β-sheets. In class I aaRS, by contrast, the active site is mounted on a scaffold of parallel β-sheets. Commonly, class I aaRS are referred to as “Rossmann folds”, although there is no genetic relation of class I aaRS and Rossmann fold enzymes. Rossmann fold enzymes derive from (β-α)₈ sheets that appear to derive from refolding (β-α)₈ barrels (i.e. TIM barrels; TIM for triosephosphate isomerase) [22].

5.3. Falsification of the Carter-Rodin-Ohno hypothesis

Figure 7b is a schematic of a multiple sequence alignment comparing GlyRS-IIA, IleRS-IA and ValRS-IA in ancient Archaea [1,3,24,26]. Figure 7b also shows how the aligned genes encoding class II and class I aaRS compare. Archaeal GlyRS-IIA is a simple sequence homolog of IleRS-IA and ValRS-IA (i.e., Figure 7a) showing that class IA aaRS were derived from GlyRS-IIA by N-terminal extension (i.e., upstream transcription and in-frame translation start) and refolding. Probably, this sequence comparison can only be done successfully with aaRS enzymes from ancient Archaea. One reason this homology is relevant is that a model (referred to as the Carter-Rodin-Ohno model) has been proposed that class I and class II aaRS were derived from “molten globule” smaller “Urzymes” encoded on complementary DNA strands [13,51–54]. These primitive Urzymes were posited to have expanded to full-length aaRS enzymes. Molten globule Urzymes must expand to a more complex version to take on their eventual specificity and refined functions. The Carter-Rodin-Ohno model is certainly incorrect. GlyRS-IIA is the root of all aaRS evolution, including all class II enzymes and all class I enzymes.

ValRS-IA and IleRS-IA enzymes include an N-terminal extension relative to GlyRS-IIA (Figure 7b). The N-terminal extension includes part of the class I aaRS active site scaffold and, in ancient Archaea, also, a Zn-finger lacking in GlyRS-IIA [26]. Also, GlyRS-IIA, IleRS-IA and ValRS-IA can share a Zn finger, as indicated. It is not possible that class II and class I aaRS are simultaneously simple homologs, as we show here, and also that class II and class I aaRS were generated as molten globule Urzymes from an ancestral bi-directional gene. Rather, GlyRS-IIA was a large and complex protein, not a molten globule, at the time of its refolding to (probably) a primitive ValRS-IA. Furthermore, ValRS-IA was a large and complex protein and not a molten globule Urzyme, from its first formation. We speculate that GlyRS-IIA and ValRS-IA assumed their initial and incompatible folds based, in part, on Zn and tRNA binding. Class II and class I aaRS bind opposite faces of their cognate tRNAs [12].

5.4. Lineages of aaRS enzymes

Figure 7c summarizes the following: 1) lineage information of aaRS enzymes; 2) aaRS enzymes with editing active sites; and 3) relationships of the aaRS lineages to the pattern of the genetic code [1,3,24,26]. We used the Phyre2 protein fold recognition server [35] in order to determine close and distant structural and sequence homologs among class II and class I aaRS [1,3,26]. What Phyre2 does is to score nearest and more distant homologs using both structure and sequence. As seed sequences, we used aaRS enzymes mostly from Pyrococcus furiosis, an ancient Archaeon. The Phyre2 server searches all sequences in the Protein Data Bank to find matches. Based on the homology scores, the lineage in Figure 7c was drawn. Distances in the map represent evolutionary distances, so clustered nodes indicate closely related enzymes. The map represents both close and distant relationships in the placements of the nodes [1,3]. AaRS enzymes with editing active sites are highlighted in yellow. Remarkably, the map closely matches the evolution and structure of the genetic code, indicating that the analysis of aaRS enzymes is reliable. Other approaches have not been as informative for the structure and evolution of the code [16–18,55–57].

The genetic code evolved primarily in columns, which represent the 2^nd anticodon position. The anticodon central position is most important for translational accuracy. Closely related aaRS enzymes, therefore, tend to group within columns. This observation is explained in detail below. Strikingly, the pattern of aaRS evolution in Figure 7c describes a history of genetic code evolution.

6. EF-Tu and coding degeneracy

We posit that the translation functions of EF-Tu describe the evolution of coding degeneracy [24]. EF-Tu is a GTPase RNA chaperone that binds aminoacylated tRNA (aa-tRNA) and docks the aa-tRNA-EF-Tu complex on the ribosome. EF-Tu (translational elongation factor Tu) is a homolog of GTPases IF-2 (translational initiation factor 2) and EF-G (translational elongation factor G) [58]. These homologous GTPases occupy the same site on the ribosome during different phases of protein synthesis: initiation (IF-2), tRNA loading, clamping, accommodation (EF-Tu) and elongation (EF-G). EF-Tu is the major factor regulating translational fidelity on the ribosome [59–63]. With the incoming aa-tRNA-EF-Tu in the hybrid A/T ribosome docking site, EF-Tu hydrolyzes GTP and sets the aa-tRNA-mRNA “latch” or clamp. The latch tightens the tRNA anticodon-mRNA codon attachment. Specifically, the latch checks for Watson-Crick geometry at the 2^nd and 3^rd anticodon positions. The latch allows wobbling at the 1^st anticodon position, the wobble position. At a wobble position, without modification of the wobble tRNA base, only pyrimidine-purine discrimination is achieved. The aa-tRNA-mRNA connection is monitored (latched) by 50S subunit 23S rRNA A1913 and by 30S subunit S12 and 16S rRNA G530, A1492 and A1493 (Thermus thermophilus ribosome numbering). After setting the latch, EF-Tu dissociates, and the released aminoacylated 3ʹ-end of the aa-tRNA makes a long rotation into the peptidyl-transferase center (the A/A site), where peptide bond formation occurs. If a latched aa-tRNA-mRNA connection cannot be formed because of a base mismatch or inappropriate wobble pair, the inaccurately loaded aa-tRNA dissociates.

Before evolution of EF-Tu, therefore, the tRNA 3^rd anticodon position could not have been read with 4-base accuracy. So, the 3^rd anticodon position must have been a wobble position, limiting the complexity of the genetic code to 2 × 4 = 8-aa complexity. We posit that, before EF-Tu evolved, only one wobble position (the 1^st or 3^rd anticodon position) could be efficiently read at a time. Also, in Archaea, A is not read in the anticodon wobble position, so A (i.e., row 1 of the genetic code is 3^rd position A), in a wobble position, formed an inefficiently utilized tRNA that functioned as a primitive translation stop signal. After evolution of EF-Tu, the 3^rd anticodon position was locked down by the latch, and the maximum complexity of the genetic code became 2×4×4 = 32-assignments. Because of translational fidelity mechanisms, the standard genetic code froze at a complexity of 21-assignments: 20-aa + stops. EF-Tu allowed the genetic code, therefore, to escape an 8-aa bottleneck and expand to the standard code, as described below. Significantly, we posit that coding degeneracy evolved as a natural consequence of how tRNA was read on the primitive ribosome. EF-Tu evolved to expand the genetic code beyond 8-aa. Of course, it is possible that protein EF-Tu evolved to replace a ribozyme with some shared properties in locking down the 3^rd anticodon position. At this time, it is difficult to know how complex the genetic code needed to become to encode functional enzymes. Here, we indicate that a code of 8-aa may have been sufficient to encode a primitive EF-Tu enzyme. Note that the 8-aa code we describe includes bending (G), bulky hydrophobics (A, V, L), hydrogen bonding (S), positive (R) and negative (D, E) amino acids. In Archaea, aa-tRNAs can be modified. Amination of D and E and addition of C for metal chelation could enrich an evolving code. C entered the code by a circuitous path described below.

7. Evolution of the genetic code

7.1. Overview

The genetic code evolved around the tRNA anticodon. In the wobble position of tRNA, only purine-pyrimidine resolution was typically achieved. Because of this limitation in reading tRNA on a ribosome, the genetic code evolved to have a maximum potential complexity of 32-assignments rather than 64-assignments, as in DNA and mRNA. In Figure 8, we show an annotated standard genetic code for Archaea with a maximum complexity of 32-assignments. The code is shown as a codon-anticodon table because the tRNA anticodon limits the complexity of the code. The amino acids are colored according to closely related aaRS enzymes (Figure 7c) to emphasize that most evolution occurred in genetic code columns, which represent the central position of the anticodon. AaRS enzymes that have editing active sites are highlighted in gray. Note that, in Archaea, aaRS that edit are limited to hydrophobic and neutral amino acids found in the leftmost two columns of the code. SerRS-IIA, which edits, is found in both column 2 and column 4. We posit that serine first invaded column 2 and then jumped to column 4 (see below).

Figure 8.

The standard genetic code in Archaea as a codon-anticodon table with a complexity of 32-assignments. Amino acids and aaRS (aa-aaRS) are colored according to closely related aaRS enzymes to emphasize evolution within code columns (Col) (Figure 7c). Anticodon (Ac) bases in red are rarely or never used in Archaea. Boxes highlighted in gray have aaRS enzymes with separate editing active sites. Codon 5ʹ→3ʹ positions are labeled 1^st, 2^nd and 3^rd

We posit an approximate order of addition for amino acids entering the genetic code (Figure 9). Glycine appears to be the first amino acid to enter the code [1,3,64,65]. Two reasons to consider glycine as the first encoded amino acid are: 1) tRNA^Gly appears to be the first tRNA (Figure 4); and Figures 2) GlyRS-IIA appears to be the first aaRS enzyme (Figure 7c). Glycine appears to occupy the most favored position in the code (anticodons GCC, UCC and CCC; 2^nd and 3^rd anticodon position C). Glycine, alanine, aspartic acid and valine (GADV) appear to be the first four amino acids to enter the code [66–70]. GADV are the four simplest amino acids chemically. These amino acids occupy row 4 of the code that appears to be the most favored row (3^rd anticodon position C). We posit that Arg, Glu, Ser and Leu enter the code next. Arg, Ser and Leu end up occupying 6-codon sectors in the code. This is described in more detail below.

Figure 9.

Proposed order of addition of amino acids into the genetic code. Amino acids appear to fill the code mostly by rows. 6-codon sectors for Leu, Ser and Arg were scored for their most favored anticodon. Ser jumps from column 2 to favored column 4, changing the Ser most-favored row assignment. Row 1 F, Y, C and W and row 3 M are posited to be the last additions to the code

As described above, to progress beyond an 8-aa code required evolution of EF-Tu to suppress wobbling at the 3^rd anticodon position [24]. After filling rows 4 and 2, row 3 and finally row 1 could be filled. Row 1 was disfavored because, initially, the 3^rd anticodon position was a wobble position, and A is not allowed in a wobble anticodon position in Archaea. Because row 1 was disfavored, stop codons located to row 1. Stop codons are read by protein release factors that bind to mRNA codons [71], so no tRNA is associated with stop codons, except in suppressor tRNA strains. We judge the order of amino acid additions that we propose to be consistent with the following rules for tRNA anticodon position preferences: 1) for the 2^nd and 3^rd anticodon positions, C > G > U⋙A; preferences are more extreme for the 3^rd anticodon position because the 2^nd anticodon position is central and easier to read; and 2) for the 1^st anticodon (wobble) position, G>(U ~ C).

The genetic code evolved mostly within columns (a working model is summarized in Figure 10). Please refer back to Figures 7–10 as reference figures for the details of amino acid placements in the code. Genetic code columns represent the 2^nd position in the tRNA anticodon, which is the most important position for translational accuracy. In the final steps, the genetic code filled row 1, which was initially disfavored. Row 1 was difficult to fill, because of wobbling of the 3^rd anticodon position. Wobbling at the 3^rd anticodon position was suppressed by evolution of EF-Tu [1,3,24,59,60]. We posit systematic rules for population of the code with amino acids (see above). These rules reflect preferences for the sequence of the tRNA anticodon. We posit that the genetic code evolved around the tRNA anticodon and the evolution of its reading on the ribosome. This mode of thinking describes the following features of the genetic code: 1) evolution in columns (Figures 7C and Figures 8); 2) evolution in rows (Figure 9); 3) the order of additions of amino acids (Figure 9); 4) late occupancy of row 1 (Figures 7C and Figures 9); 5) the complexity of the code; 6) evolution of 6-, 4-, 3-, 2- and 1-codon sectors; 7) evolution of stop codons; 8) coevolution of translation factors such as EF-Tu; 9) coevolution of aaRS enzymes with the code (Figures 7C and Figures 8); 10) complexity and structure in the 3^rd genetic code column; 11) selections of code structures and amino acid placements; 12) serine jumping during code evolution; 13) evolution of translational fidelity; and 14) freezing of the code. The model, therefore, is highly predictive and descriptive for the final structure and sectoring of the code. Figures 11–17 describe a proposed order of addition of amino acids into the code. We know of no comparable model for genetic code evolution.

Figure 10.

Summary of the proposed order of events in evolution of the genetic code, mostly by columns. In parentheses: (discriminator base (3ʹ-XCCA; X = the discriminator) from Pyrococcus, aaRS). Colors are used as a guide for closely related aaRS enzymes and tRNA discriminator sequences. Classic identifications of aaRS subclass (i.e., ArgRS-ID) are not necessarily reliable (see Figure 7c). Yellow highlighting indicates tRNA-mediated enzymatic reactions: A-t’ase (Asp-tRNA^Asn and Glu-tRNA^Gln amidotransferase); Cys S-thase) Sep-tRNA^Cys Cys synthase (Sep for o-phosphoserine). See the text for details. Arrows indicate the order of amino acid entries into the code, mostly within columns, and not necessarily the lineage of aaRS evolution. TRNAs and aaRS can be re-assigned in evolution

Figure 11.

Polyglycine world. A is inefficiently read in a wobble position (at this stage, both 1^st and 3^rd anticodon positions were wobble positions). Aa-aaRS) amino acid-aminoacyl-tRNA synthetase; RBZ) ribozyme; Ac) anticodon. Letters around the periphery indicate codon (mRNA) sequence. Colored shading for amino acids is maintained in Figures 11–Figures 17, so that placements of amino acids can be tracked. Red letters indicate anticodons that are not used or are used inefficiently

Figure 12.

GADV-world. Explaining evolution in code columns (Figures 7C and Figures 8)

Figure 13.

The 8-aa bottleneck. Columns 1, 2 and 4 sectored differently than column 3 because of the selection of the wobble position (3^rd anticodon position for columns 1, 2 and 4; 1^st anticodon position for column 3)

Figure 14.

The 8-aa bottleneck: Leu, Ser and Arg invaded row 3, helping to describe evolution of 6-codon sectors

Figure 15.

The 8-aa bottleneck. Ser jumped from column 2 to column 4, invading the Arg sector. We posit evolution of a primitive EF-Tu or a ribozyme with EF-Tu-like activity suppressing wobbling at the 3^rd anticodon position

Figure 16.

An intermediate ~17-aa stage and evolution of high innovation in column 3. We speculate that TyrRS-IC may be derived from a primitive ArgRS-ID and that tyrosine may have jumped from column 4 to column 3

Figure 17.

The standard genetic code. Amino acids in charcoal and gray were the last to enter the code. Disfavored row 1 was filled. Stop codons evolved on disfavored row 1. Met invaded the Ile sector. The code now has stops and starts

7.2. Polyglycine world

We posit that glycine was the first encoded amino acid [64,65], and that the genetic code first evolved to synthesize polyglycine (Figure 11) [1,3,24,25]. Initially, this was an assumption, but it turned out to be such a useful assumption, it should not be rejected easily. Also, if one were to choose another amino acid (i.e., Ala or Pro) as the first encoded amino acid, we do not believe reasonable rules can be as easily established for filling the code. Selecting Gly, as the initial encoded amino acid, however, a reasonable mechanism and simple rules for populating the code became apparent. We posit that the entire primitive code, including all anticodons and all codons, evolved to encode glycine. Row 1 tRNAs (1^st anticodon position wobble A and 3^rd anticodon position (initially) wobble A) were utilized inefficiently, so these tended to function as stops. We posit that wobbling at the 3^rd anticodon position was suppressed by evolution of EF-Tu. In Archaea, A is not allowed in a wobble position [26]. Basically, we posit that a ribozyme mechanism existed to replicate templated tRNA (and other RNA) sequences. The anticodon is the easiest feature of tRNA to mutate without consequence for folding, so proliferation of tRNAs and mutations rapidly resulted in all possible anticodon sequences. We posit that a GlyRS ribozyme (GlyRS-RBZ) charged diverse tRNA^Gly with glycine because the code was not yet sufficiently evolved to generate protein catalysts. Hemolithin, recovered from meteorite samples, is a polyglycine peptide from outer space, indicating that a polyglycine world existed, even beyond an Earth environment [72].

One advantage of this mode of thinking is that it gives insight into the sectoring of the genetic code. If the entire primitive code encoded glycine, then invasion by other amino acids caused the glycine sector to contract. Invasion of the code by newly encoded amino acids, therefore, resulted in shrinking of previously occupied sectors. We then realized that clear rules could be stated for how incoming amino acids displaced previously added amino acids. Currently, glycine occupies the most favored anticodon positions in the code, which are GCC, UCC and CCC. If amino acids that entered the code at an early time protected the most advantageous sectors, then C was favored in the 2^nd and 3^rd anticodon positions. The rules for the 2^nd and 3^rd anticodon positions began to emerge as C > G > U⋙A [1,3]. Preferences were strongest for the 3^rd anticodon position, because the 2^nd anticodon position was easier to read.

7.3. GADV world

We posit that polyglycine world gave way to GADV world (GADV for glycine, alanine, aspartic acid and valine) (Figure 12) [66–68,73]. Positing GADV world explains why the genetic code sectored in columns (Figures 7C and Figures 8). Glycine, alanine, aspartic acid and valine are the simplest amino acids chemically, so it is reasonable that these might be the first four amino acids to enter the code [66,67,69,70,73–75]. Also, GADV are amino acids that end up on the 4^th row of the code, which corresponds to 3^rd anticodon position C, in keeping with the C > G > U⋙A rule for the 2^nd and 3^rd anticodon positions. At the GADV stage, we posit that tRNAs were charged by ribozymes.

7.4. The 8-aa bottleneck

Before evolution of EF-Tu, the genetic code froze at 8-aa (Figures 13–15). EF-Tu suppresses wobbling at the 3^rd anticodon position. Before EF-Tu, therefore, 3^rd anticodon position A could not easily be read on the primitive ribosome. We note here that 1^st anticodon position wobble A is not utilized in Archaea [1,3,26]. In Bacteria and Eukaryotes, wobble A is encoded only when it is modified to inosine, which broadens the set of recognized codons [76–79]. Interestingly, columns 1, 2 and 4 in the code sectored by one mechanism, and column 3, which became the most innovated column, encoding the most amino acids, sectored by a slightly different mechanism. We posit that the genetic code froze at 8-aa because both anticodon 1^st and 3^rd positions were wobble positions, and wobble positions are read with pyrimidine-purine (2-assignment) resolution. The 2^nd anticodon position was much easier to read because it is the middle position. We further posit that only a single wobble position could be read at a time. Because, at a wobble position, only purine versus pyrimidine discrimination was initially achieved, the maximum complexity of the code was 2 × 4 = 8-aa. In columns 1, 2 and 4, the 2^nd and 3^rd anticodon positions were primarily read. Interestingly, leucine, serine and arginine are the amino acids that maintained 6-codon sectors (3 boxes in the genetic code tables shown). By contrast, in column 3, the 1^st and 2^nd anticodon positions were primarily read, giving the striped pattern of the related amino acids Asp and Glu. These differences in wobble selection gave rise to 6-codon sectors, for leucine, serine and arginine, and high innovation in column 3 (encoding many amino acids). Because of the geometry of the 7-nt U-turn anticodon loop in tRNA, the reading register for the primitive ribosome was always 3-nt.

In Figure 14, we posit that leucine, serine and arginine invaded row 3 of the code, displacing valine, alanine and glycine into favored row 4 (3^rd anticodon position C). Positing this invasion of row 3 helps to describe the evolution of 6-codon sectors, the placement of threonine in the code and the jumping of serine from column 2 to column 4. Because valine, alanine and glycine retained the favored 4^th row (3^rd anticodon position C), these invasions were tolerated.

In Figure 15, we posit that serine jumped from column 2 into column 4. This event is described in more detail below. Leucine, serine and arginine began to invade disfavored row 1. This is considered in the model because leucine and serine end up in row 1 of the code. Also, a primitive ArgRS-ID may have evolved to TyrRS-IC, TrpRS-IC and CysRS-IB (Figure 7c) that end up in row 1. Evolution of a primitive EF-Tu at the 8-aa stage, or else evolution of a ribozyme to partly lock down the 3^rd anticodon wobble position, might have this effect. We do not propose a specific order of events. We note that, at this stage, row 1 probably included tRNAs that were charged but not efficiently utilized, until EF-Tu or a corresponding ribozyme evolved. Ser jumping from column 2 to column 4 required only a single base change in the tRNA anticodon at the 2^nd position (GGU→GCU). The jump was favorable for placement of serine in the code because column 4 (2^nd anticodon position C) was favored over column 2 (2^nd anticodon position G).

7.5. Evolution of EF-Tu suppressed wobbling at the 3^rd anticodon position and broke the 8-aa bottleneck

We posit that evolution of EF-Tu converted the 3^rd anticodon position from a wobble position with pyrimidine-purine (2-assignment) resolution to a position with 4-base (A, G, C, U) resolution. This advance expanded the genetic code from a maximum complexity of 2 × 4 = 8-aa to a maximum complexity of 2×4×4 = 32-assignments. It should be noted that, in mRNA, the maximum complexity of the code is 4×4×4 = 64-assignments. The complexity of the code, which froze at 20-aa + stops = 21-assignments, was limited to a large extent by reading of tRNA on the ribosome.

An intermediate state of the code with ~17-aa is posited in Figure 16. In column 1, Ile was added as a 4-codon sector. Because Leu entered the code before Ile, and because Ile displaced Leu, Ile occupied row 3 (3^rd anticodon position U), which was disfavored compared to row 2 (3^rd anticodon position G) (C > G > U⋙A). In column 2, Thr replaced Ser in row 3, and Pro replaced Ser in row 2. Because Thr and Ser are related amino acids, it is easy to see how SerRS-IIA could have duplicated and a copy could have diverged to evolve ThrRS-IIA (Figure 7c). ProRS-IIA may be derived from GlyRS-IIA. We note that, in Archaea, ProRS-IIA appears most closely related to GlyRS-IIA, and that neither enzyme evolved an editing active site (Figure 7c). Ser now occupied row 1. Column 3 filled to become the most innovated column encoding the most amino acids. The two founding amino acids in column 3, Asp and Glu, remained on favored row 4 (3^rd anticodon position C). We posit that Asp, which entered the code first, occupied row 4A and Glu occupied row 4B because tRNA wobble G was favored over wobble U/C. We posit that AspRS was originally AspRS-IIA and evolved to AspRS-IIB after evolution of HisRS-IIA. AsnRS-IIB was later derived from AspRS-IIB. This is a continuing evolution in some archaeal species. GluRS-IB gave rise to LysRS-IE and GlnRS-IB (Figure 7c). GluRS-IB→GlnRS-IB is ongoing in some archaeal species.

We begin to model the filling of disfavored row 1. We posit that phenylalanine and tyrosine may have entered the code at about this stage. PheRS-IIC may be derived from a primitive AspRS-IIB in several steps (Figure 7c). TRNA^Phe, in Pyrococcus, appears to be derived from tRNA^Lys [25]. These identifications do not provide a simple model for placement of phenylalanine in the code. TyrRS-IC may be derived from a primitive ArgRS-ID, filling column 4, row 1, and jumping to column 3 by a single base change in the anticodon. We posit that pressure is building to evolve modern stop codons at this stage of evolution.

7.6. Filling disfavored row 1, evolution of stop codons and Met invasion of an Ile sector

To evolve the standard genetic code (Figure 17), then required filling in disfavored row 1 (3^rd anticodon position A). We posit that Phe invaded column 1, row 1A, displacing Leu. Tyr invaded column 3, row 1A, perhaps as described above. Stop codons located to columns 3 and 4, row 1B. Cys invaded column 4, row 1A. We posit that CysRS-IB was derived from duplication and repurposing of a primitive ArgRS-ID. The classic naming of these aaRS enzymes is deceptive. CysRS-IB and ArgRS-ID are closely related enzymes, despite their IB and ID structural subclass designations (Figure 7c). Trp invaded column 4, row 1B. There are few 1-codon sectors in the genetic code, but Trp shares a sector with a stop codon, which is read in mRNA by a protein release factor [71], so there is no competing tRNA occupying the Trp sector (column 4, row 1B).

Evolution of row 1 appears somewhat chaotic. Some chaos might be expected in filling the last available positions in the code. It appears from Figure 7c that TyrRS-IC and TrpRS-IC might have been derived from a primitive ArgRS-ID. Tyrosine and tryptophan may have been two of the last amino acids added to the code [80]. CysRS-IB appears to be derived from a primitive ArgRS-ID. PheRS-IIC appears to have evolved in steps from a primitive AspRS-IIB. In Pyrococcus, tRNA^Phe (discriminator A) appears to be derived from tRNA^Lys (discriminator G). TRNA^Tyr (discriminator A) appears to be derived from tRNA^Asn (discriminator G). TRNA^Trp (discriminator A) appears to be derived from tRNA^Pro (discriminator A). TRNA^Cys (discriminator U) appears to be derived from tRNA^Thr (discriminator U) [25]. We guess that all of these tRNAs were assigned and then reassigned, and, therefore, their apparent derivations do not indicate the precise steps in adding these amino acids to the code. Reassignments of aaRS enzymes and tRNAs in evolution enhances translational accuracy by suppressing mischarging of tRNAs.

Cys appears to have entered the genetic code via a circuitous path [81–83]. Notably, some Archaea generate Cys from Sep-tRNA^Cys charged by SepRS-IIC (Sep for o-phosphoserine). Modification of amino acids bound to tRNAs is a repeated theme in ancient Archaea that may reflect chemistry from a pre-life world [84,85]. This is also how Asn and Gln entered the code. Asp-tRNA^Asn and Glu-tRNA^Gln were aminated by Asp-tRNA^Asn and Glu-tRNA^Gln amidotransferase [86–93]. TRNA-linked and RNA-linked chemistry must have been common in the pre-life world before evolution to cellular life [84,85].

In column 1, row 3B, Met invaded a 4-codon Ile sector. MetRS-IA and IleRS-IA are closely related enzymes (Figure 7c). Furthermore, tRNA^Met and tRNA^Ile are closely related tRNAs, in ancient Archaea such as Pyrococcus [25]. In keeping with our contention that 1-codon sectors were difficult to form and maintain, tRNA^Ile(CAU) and tRNA^Met(CAU) are both utilized in Archaea. The UAU anticodon, however, is rarely used. In tRNA^Ile(CAU), wobble C is converted to agmatidine to recognize Ile codon AUA but not Met codon AUG [94–100]. In tRNA^Met(CAU), wobble C is lightly modified and recognizes Met codon AUG but not Ile codon AUA. Wobble anticodon modification, therefore, resolves the ambiguity of tRNA^Ile(CAU) and tRNA^Met(CAU) but anticodon UAU was generally lost in the process. Met provides translation starts, in addition to bringing another amino acid into the code.

7.7. Code punctuation

We posit that genetic code starts and stops were late additions to the code. Translation initiation in Archaea has recently been reviewed [101]. Archaea utilize Met-tRNA^iMet (iMet for initiator methionine) and a set of translation initiation factors. Bacteria utilize fMet-tRNA^iMet (fMet for N-formyl-methionine) and a simplified initiation mechanism. In Pyrococcus, tRNA^iMet(CAU), tRNA^Met(CAU) and tRNA^Ile(CAU) are discriminated mostly based on acceptor stems. For tRNA^iMet, the 5ʹ-acceptor stem sequence is AGCGGG(G), with the 3ʹ-G uncharacteristically unpaired (opposite 3ʹ-acceptor stem (G)CCCGCU). For tRNA^Met, the 5ʹ-acceptor stem sequence is GCCGGGG, with all bases paired. For tRNA^Ile (CAU), the 5ʹ-acceptor stem sequence is GGGCCCG, with all bases paired. It appears that selection of methionine as the initiating amino acid in Archaea was a complex coevolution of Met-tRNA^iMet and initiation factors. We guess that the need for a translation initiation start signal was a powerful driving force to evolve this system. For instance, to initiate translation at internal mRNA sites for gene expression regulation may have required evolution of the Met-tRNA^iMet translation initiation system. We guess that Bacteria simplified the archaeal system in evolution, adopting fMet-tRNA^iMet and shedding initiation factors during divergence from Archaea.

Evolution of translation stops appears to have been a complex process with multiple stages [71]. We posit that initially the problem was generating longer peptides to form more complex proteins. The system appears to have started with inefficient tRNAs, i.e., 1^st or 3^rd anticodon A, as primitive stop signals. The system suppressed 3^rd anticodon position wobbling by evolving EF-Tu. Finally, protein translation release factors evolved to recognize stop codons in mRNA. Suppression of wobbling at the 3^rd anticodon position by EF-Tu expanded the genetic code and may have driven evolution of protein release factors and stop codons.

7.8. Serine jumping from column 2 to column 4

We posit that jumping across columns was rare in establishment of the code. The advantage for Ser to jump from column 2 to column 4 was that serine obtained a favored anticodon. GCU was favored over GGU, because 2^nd anticodon position C was favored over G. Serine could invade the arginine sector because ArgRS-ID reads type I tRNA^Arg and cannot read type II tRNA^Ser. Also, SerRS-IIA must bind the type II tRNA^Ser V loop to add Ser. There is no advantage to serine invading the leucine sector, and such an invasion would be problematic, partly because both tRNA^Leu and tRNA^Ser are type II tRNAs. Invasion of row 3 would not be advantageous for serine and would also eliminate an amino acid from the code.

7.9. Summary

In summary, a highly detailed working model is possible for evolution of the genetic code. The model is mostly based on tRNA and aaRS sequence analyses. The genetic code evolved around the tRNA anticodon. The model tracks evolution of aaRS enzymes, indicating that both the genetic code model and the model for aaRS evolution are substantially correct. When we started this work, we did not think such a detailed and predictive model was possible or reasonable. Now, we consider this a very strong model for further analysis of the code.

8. Discussion

8.1. tRNA evolution

Remarkably, tRNA evolution was determined to the last nucleotide (Figure 2) [2]. The solution to tRNA evolution was possible because tRNA^Pri sequences were highly ordered and these repeats and inverted repeats can still be detected in ancient Archaea. No accretion model can describe tRNA evolution, because of conservation of highly regular tRNA^Pri sequences. For an accretion model to have credence, tRNAs would need to expand, inserting ordered and preordained sequences, which seems unlikely if not impossible. Only the 3–31-nt minihelix model describes tRNA evolution. Accretion models, such as the Uroboros [7] and 2-minihelix [8,10] models are, therefore, falsified. Also, the genetic code evolved around the tRNA anticodon [1–3,24]. The genetic code complexity is determined by the way the tRNA was read on a ribosome, explaining why the genetic code went from being frozen at 8-aa complexity (Figures 13–15), before attaining the standard code (Figure 17) that has 21-assignments (20-aa + stops).

8.2. aaRS evolution

Analysis of aaRS evolution tracks evolution of the genetic code (Figure 7c) [1,3,24]. This analysis is easiest to do in archaeal systems because Archaea are the oldest organisms, and Bacteria are more derived (Figure 1). Because GlyRS-IIA, ValRS-IA and IleRS-IA are simple sequence homologs (Figure 7a and Figure 7b), the Carter-Ohno-Rodin hypothesis for aaRS evolution [13,51–53] is falsified. Class I and class II aaRS were not generated from opposite DNA strands of a primordial, bi-directional gene encoding molten globule Urzymes.

8.3. genetic code evolution

A highly detailed model has been generated that describes standard genetic code evolution in Archaea (Figures 10–17). Every aspect of code evolution is described by this model. Simple rules were developed to describe sectoring of the code. The model can be modified to generate the more derived genetic codes of Bacteria and Eukarya.

8.4. Freezing the code

We posit that new amino acids were introduced through tRNA charging errors and through aa-tRNA linked chemistry, and that translational fidelity mechanisms froze the code [1,3]. Based on archaeal systems, Asn, Gln and Cys appear to have initially entered the code through enzymatic mechanisms in which aa-tRNAs were modified. Subsequently, the tRNA-linked reactions were replaced by evolution of AsnRS-IIB, GlnRS-IB and CysRS-IB. Other amino acids may also have entered the code via tRNA-linked reactions. For instance, Arg may have replaced ornithine early in code evolution. Ornithine can be converted enzymatically to Arg in two steps [102]. Similarly, Leu may have been synthesized from tRNA-linked Val in 5 enzymatic steps. Because of initial wobbling in the 1^st and 3^rd anticodon positions, EF-Tu evolution was necessary to expand the code beyond 8-aa. Subsequently, EF-Tu contributed to freezing of the code by enforcing translational accuracy. Some aaRS have proofreading (editing) to remove inappropriately added amino acids from their cognate tRNAs [12]. Remarkably, the aaRS that edit are limited in Archaea to amino acids located in the left half of the code (columns 1 and 2; Figures 7C and Figures 8). Hydrophobic and neutral amino acids locate to columns 1 and 2 of the code, so editing helped with translational accuracy for amino acids with limited chemical character, such that these amino acids could not be as easily specified in the aaRS active site. Editing helped to freeze the code by protecting 6- and 4-codon sectors in the left half of the code. To add additional amino acids required splitting larger sectors of the code. 6-codon sectors encoding Leu, Ser and Arg resulted from the history of evolution, as described. Splitting a 2-codon sector into two 1-codon sectors was problematic because of tRNA wobble ambiguity (generally, in Archaea, tRNA wobble U ~ C and only G is allowed, not A). High innovation in column 3 of the code resulted from the history of column 3 sectoring and the initial selection of the wobble base (1^st anticodon position). In the case of Ile and Met, co-occupancy of the CAU anticodon through wobble modifications resulted in suppression of the Ile UAU anticodon, as described.

8.5. A model for evolution of protocells

Figure 18 shows a model for evolution of the first cells. We posit that a number of ribozymes must have been present in order for tRNA to evolve prior to evolution of complex proteins [1,3]. TRNA^Pri was comprised completely of ordered sequences, notably, repeats and inverted repeats, so ribozymes must have existed to generate repeats, inverted repeats, 31-nt minihelices and tRNA [2] (Figure 2). Furthermore, polyglycine and GADV polymers appear to have been the first products of the evolving genetic code. Therefore, a selection for polyglycine and GADV polymers would describe the selective pressure for evolution of the code before complex proteins became possible. We posit that polyglycine and GADV polymers formed essential structures in protocells. Structures included cell walls, internal structures, amyloid plaques and LLPS (liquid–liquid phase separation) compartments. We note that (Gly)₅ is a component of bacterial cell walls [103,104] and may be a relic of a pre-life world.

Figure 18.

A detailed model for evolution of protocells, polyglycine- and GADV-world. PTC) peptidyl transferase center

Amyloid plaques form from assemblies of long, mis-associated β-sheets. In eukaryotic cells, amyloid plaques are a symptom of neurological diseases [105–107]. We posit that amyloid plaques may be generated from misregulation of LLPS membraneless compartments, which are a normal feature of eukaryotic cells. In the ancient world, amyloid plaques would have regulated hydration in protocells to enhance diverse chemistries such as polymerization reactions. LLPS compartments stimulate diverse chemistries in eukaryotic cells and may function similarly in prokaryotic cells [108–110]. We posit that polyglycine and GADV LLPS compartments were essential features of early protocells. Amyloids and LLPS are posited to have been selected because they supported novel and essential protocell chemistries, partly by regulating hydration and dehydration. We posit that amyloids and LLPS provided the selective driving force for the early evolution of the genetic code before complex proteins could be encoded.

8.6. Other models

Here, we briefly contrast our genetic code evolution model with some other models that have been advanced. Koonin and Novozhilov review a number of genetic code models [14]. Much emphasis has been given to three models: 1) stereochemical; 2) error-minimization; and 3) coevolution. We do not consider these models to be highly predictive [1]. A new approach has been proposed based on codon energetics [111]. Computational approaches may be of interest [112–114], although the concept of late evolution of degeneracy seems unlikely compared to our model that degeneracy is a natural result of evolution of anticodon reading on the ribosome and evolution of EF-Tu. Kunnev and Gospodinov have proposed models that include RNA-aa linked reactions in pre-life, as we also support [84,85]. Rogers has put forth a model in some ways similar to ours [115]. Another somewhat similar model to ours has been advanced by Chatterjee and Yadav [116]. A different view of serine sectoring than that we propose was recently advanced [117]. Simply stated, our models are more detailed than others and make many more specific predictions. We provide a clear selection strategy and a set of rules for the placements of all amino acids in the standard code. We enrich the discussion of Kunnev and Gospodinov on tRNA – and RNA-linked reactions in the ancient world before the first true cells. Our approach is centered on the tRNA anticodon, and others should adopt the anticodon-centered view. For instance, evolving directly to codons is a mistake, because the genetic code was limited in complexity by the tRNA anticodon [1,3]. Emphasis on the tRNA anticodon and reading of the anticodon on the primitive and modern ribosomes (i.e., +/– EF-Tu) also describes degeneracy of the code. Furthermore, we consider filling the code piecemeal to be a mistake. We take a more orderly approach to code-filling that is based on clear rules for anticodon sequence preference. We provide strong selections for the initial steps of code evolution before complex proteins can be encoded. Our model for tRNA evolution reaches far back into the pre-life world with many predictions for pre-life ribozymes and, surprisingly, ordered pre-life RNA chemistry. Remarkably, existing tRNA sequence provides a record of chemistry in the pre-life world.

Funding Statement

This research received no external funding.

Abbreviations

aaRS	aminoacyl-tRNA synthetase (i.e. GlyRS)
Ac	Anticodon
As	Acceptor stem
EF	Translation Elongation Factor
G	Glycine
Ga	1 Ga=1 billion years
GADV	Glycine, Alanine, Aspartic Acid, Valine
IF	Translation Initiation Factor
LLPS	liquid–liquid phase separation
LUCA	Last Universal Common (cellular) Ancestor
PDB	Protein Data Bank
Pfu	Pyrococcus furiosis
PTC	Peptidyl Transferase Center
RBZ	ribozyme
Sep	o-phosphoserine
SLS	stem-loop-stem
TBP	TATA-box binding protein
TFB	Transcription Factor B (homolog of TFIIB in Eukaryotes)
TFE	Transcription Factor E (homolog of TFIIE in Eukaryotes)
tRNAPri	Primordial tRNA
V loop	Variable loop

Author Contributions

The authors wrote the paper, made the figures and did the research.

Disclosure of potential conflicts of interest

No potential conflict of interest was reported by the authors.