The concept of nucleic acid-assisted protein folding (nucleic acid chaperons) was first published in 2005 1 , when it was suggested that mRNAs participate in the formation of the tertiary structure of their coded proteins; this function is in addition to the well-recognized determination of the amino acid sequence of the proteins, as described in Nirenberg's Genetic Code. Messenger RNAs often have a tertiary structure that provides spatial information for protein folding, i.e. they are nucleic acid chaperons 2 . Nucleic acid-assisted protein folding requires that mRNA and coded peptides remain in contact with each other, even after the polymerization of the amino acids during translation. The original model described RNA-assisted protein folding, but the role of tRNA was not properly elucidated.

The role of the t-RNAs is to "translate" between mRNA and the encoded peptide, which means that tRNAs recognizes the codons in mRNA and arrange the amino acids in the order of the codons for polymerization by the protein synthetase. This function could be performed by a much smaller oligonucleotide containing only four nucleotides, three for the anticodon and one for binding the corresponding amino acid. However tRNAs are much larger, about 76 nucleotides in length.

The abundance of tRNA is another mystery. The 64 different codons require 64 different "mediators", but there are many more. There are, for example, 359 different cytoplasmic tRNAs in human cells, and many more when genomic variants are included. The generous size and lavish redundancy suggest that the function of tRNAs is far more complex than the passive connecting of codons with their amino acids that has been recognized to date.

The aim of this work is to identify additional functions for tRNAs in translation, and to explain their involvement and role in nucleic acid-mediated protein folding.

The existence of tRNA was hypothesized by Francis Crick in 1956 3 4 to explain the transfer of information (translation) from nucleic acids to proteins, which are two structurally very different molecules. The first tRNA sequences were described and structures were suggested in 1965 by Holley et al. 5 . Transfer RNA was required to "fill the gap" between amino acids and codons, or spatially "convert" the large codons (3 nucleotides, ~1 nm in length) to amino acids (single molecules, about 1/3 nm length when incorporated in proteins).

The tRNA seemingly solves the problem because of the 3:1 size difference between codons and amino acids (i.e. mRNA and the coded peptide). However, this creates a new spatial problem: the tRNA molecules are far too large to be able to align side-by-side (codon after codon) on the mRNA. Transfer RNAs are typically 73-93 nucleotides in length, and they have a highly conserved characteristic tertiary structure in which the amino acid acceptor CCA tail and the anticodon are located at opposite ends of the molecule, about 7.6 nm from each other. The molecule has three arms with loops (called D, T and anticodon loops), together with one additional, variable loop. The L-shaped structure fits into a 6 × 6 × 2 nm cube. The thickness of the tRNA (approximately 2 nm in a cloverleaf structure) is twice that of the distance between codons (1 nm). Consequently, only every 2^nd folded (cloverleaf) tRNA would be able to approach the mRNA from the same direction. Alternatively, the tRNA could unfold into a linear single-stranded form (with a diameter corresponding to the distance of 1 nm between codons). However, the unfolded, linear form would interact with its neighbor (Figure 1).

The structural complexity of tRNA is reminiscent of that of a protein, with 71 out of 76 bases participating in a stacking interaction (42 of the bases have a double helical stem structure). A series of specific 9 bp sequences crosslink the tertiary structure of the tRNA by interacting with bases from a different stem and loop region of the molecule. All of these 9-bp interactions are non-Watson-Crick associations and are highly conserved, which makes it likely that all tRNA molecules have similar structures, although only a few have been crystallized and their structures determined.

To provide a one-to-one correspondence between tRNA molecules and the codons that specify amino acids, 61 types of tRNA molecules would be required per cell. However, many cells contain fewer than the predicted 61 types of tRNAs because the wobble base is capable of binding to several, though not necessarily all, of the codons that specify a particular amino acid. A minimum of 31 tRNAs are required to translate, unambiguously, all 61 sense codons of the standard genetic code 6 7 .

However, the total number of tRNAs is much larger than the number of codons. In the human genome, which according to estimates has about 27,161 genes 8 , there are in total about 4,421 non-coding RNA genes, which include tRNA genes. There are 22 mitochondrial tRNA genes, 497 nuclear genes encoding cytoplasmic tRNA molecules, and 324 tRNA-derived putative pseudogenes 9 .

This means that there are, on average, eight slightly different cytoplasmic tRNAs for every possible codon. Therefore, the size and the redundancy of tRNAs are very large for the "tiny" function that is accepted today. Identifying new functions and explaining the size and redundancy of tRNA is one of the goals of our study.

The tRNA database that we used 10 11 lists 359 human tRNA sequences. There are tRNAs corresponding not only to the 20 amino acids, but also to the initiation (Ini, CAT, Met) and termination (Sec, TCA) sites. Many codons have no tRNA representations in this database, including AAA (Lys), ACA (Thr), ACC (Thr), ACT (Thr), ATC (Ile), ATG (Met), CTA (Leu), GAC (Asp), GAG (Glu), GCG (Ala), GGA (Gly), GGC (Gly), GGG (Gly), GGT (Gly), TTA (Leu). Each of the remaining 49 codons is represented by an average of 7-8 different tRNAs.

(The relatively large numbers of tRNAs may be readily explained by the fact, that many fundamentally important genes code isoforms of the same proteins (for example the enzymes of glycolysis), so that a mutation which result a loss of function will not be immediately lethal. In case of only one tRNA alternative, this would have been way too risky for life.)

It is important to keep in mind that almost all of these tRNAs have been identified using bioinformatics tools that were looking for sequence characteristics assigned to real tRNAs. However, their existence and biological functions have yet to be confirmed by biochemists. Even the Protein and Nucleic Acid Databases (PDB, NDB, 12 13 ) are surprisingly empty of real tRNA structures, although many tRNA-associated proteins have been the subject of structural studies.

When speaking about tRNA the reader has to keep in mind that translation is still a very "hot" subject for many scientists, "the dark side of molecular biology" 14 driven by a powerful paradigm. This 50 year old paradigm suggests that translation is a "tape reading" where there is a "tape" (mRNA) and step-by-step reading it (tRNA) provides proteins on the surface of ribosomes. The ribosome has a door "in" (A site) and a door "out" (P site) 15 16 . This "cartoon guide to translation" 14 very strictly separates the "RNA World" from the "Protein World", where the only connection permitted is the tRNA (the "adaptor"). However the connection between codons and amino acids (provided by tRNA) is not logical or the result of evolution, but accidental, a so called "frozen accident" 17 and codons and amino acids never interact with each other. Consequently, in every published structure, the distance between the amino acid and the anticodon is well over 50 Å

Although this mechanical model is supported by structural studies 18 19 20 21 22 23 24 the reality is that crystal structures present only snapshots of thermodynamically stable conformations. Solution and computational methods provide evidence for the inherent flexibility of tRNA structure under a variety of conditions and for differing tRNA species. Transfer RNAs perform a wide range of motion and conformational changes and allosteric transition during the process of translation. 25 26 27 Lapointe, Alexander, Caulfield]. Molecular Dynamic methods, like simulations of Cryo-EM Microscopy and X-Ray Data to explore intermediate conformational space, often provide previously unidentified structures which are very different from the canonical cloverleaf-like manifestations. 27 28 29 30 . Interaction with other macromolecules is known to cause structural alteration in tRNAs. 31 32 33 . The tRNA structure is sensitive for changes in the molecular environment, post-transcriptional modifications. A single base mutation is sufficient to change the "cloverleaf" configuration into a "hairpin"-like folding 29 34 .

An estimation of the folding energy (Quickfold program 35 ) indicates that all human tRNAs have negative folding energies (-26.1 ± 4.42 kcal/mole, SD, n = 359). However, even randomized tRNA sequences possess significant folding energies (-21.1 ± 4.79 kcal/mole, SD, n = 369). The difference between these two values (-4.95 ± 5.66 kcal/mole, SD, n = 369), the Free Folding Energy, FFE, is rather small, only approximately 19% of the total folding energy, and there are even numerous positive values. This indicates that the majority of the total folding energy of tRNAs is provided by the nucleotide composition, and not by the nucleotide sequence itself, i.e. tRNAs may fold into numerous different structures in addition to the typical cloverleaf configuration.

The total and free folding energies show wide variation among different tRNAs, even when these different tRNAs code for the same amino acids. It is not possible to correlate this variation to the codons either, because tRNAs carrying the same anticodon vary widely in the size of their folding energies (Figures 2 and 3).

The common feature of all tRNAs is that their primary structures contain a large number of complementary residues and these Watson-Crick-type interactions, within the same sequence, provide a characteristic tertiary structure. This property suggests that complementary interactions do exist not only within but also between tRNAs, as they have a high degree of similarity. The question of interactions between tRNAs was studied here using in silico and hybridization (melting) studies.

Interaction between tRNAs is thermodynamically favored. Not unexpectedly, the tRNA hybridizes with itself (dG values for intact and randomized sequences are -43.7 ± 8.22 kcal/mole, SD, n = 359 and -33.5 ± 6.76, SD, n = 359, respectively, and the difference, the Free Hybridization Energy, FHE, is -10.2 ± 7.18 kcal/mole, SD, n = 359). The codon- and amino acid-related distribution of dG values for hybridization is similar to the distribution of dG values for folding in that there is a wide range, and even some positive values; however, the correlation between FFE and FHE is not significant (data not shown). The mean dG of hybridization is much higher than the mean dG of folding, indicating that the potential for tRNA interactions (Watson-Crick pairs) is not fully utilized during folding compared to hybridization.

Hybridization between different tRNAs (359 vs. 359) in any combination is possible and is thermodynamically favored (Figure 4).

The idea that tRNAs interact with each other during translation is not new. It was first suggested by Carl Woese in 1970 36 in his "reciprocating ratchet mechanism" for protein synthesis which depends upon conformational changes in tRNA and allosteric transitions in place of translocation.

The cloverleaf model, the paradigm of tRNA structure, emphasizes the high conservation of the base pairing pattern or secondary structure among tRNAs encoded in the genome of all organisms and eukaryotic organelles. However, some tRNAs are not able to fold into this patter 37 . Other tRNAs exists in more than one thermodynamically stable conformations (conformers) and these unconventional structures are in dynamic equilibrium with each other 38 and even interact with each other 39 . It is well recognized that sequence variations and post-transcriptional modifications can case large variations in the folding energy, half-life and activity of tRNAs 40 41 . There are many different tRNAs with the same anticodon, therefore it is reasonable to suppose that tRNAs are heterogeneous regarding their structure and thermodynamic properties 42 43 . Dimer formation by tRNAs under physiological conditions have been suggested 44 and modeled 39 . We suggest that the hairpin-like conformers of tRNAs are able to form complementary base interactions not only within bot even between identical or similar tRNAs and form homo- and heterodimers (Figure 5 ).

We suggest that tRNAs have a far more complex role in translation than functioning only as passive adaptors. They are able to change form and interact with each other, and this change seems to be necessary for mRNA-assisted protein folding to be carried out. These new roles explain the large size and generous redundancy of tRNAs.