1. INTRODUCTION

The synthesis of the great variety of molecules that are used by living organisms basically requires two things: a source of energy and a source of matter. Before the invention of photosynthesis a different energy source than the Sun was needed and the first life forms had to find ways to synthesize essential molecules using whatever was available in the, at the time, anoxic environment. An important source of energy could have been H₂, which was generated in the highly reducing environment of the earth crust and which may have been present at much higher amounts than today (Tian et al., 2005). Carbon, a central constituent of all biomolecules, was probably available as atmospheric CO₂ and as dissolved carbonate. For many years we have been interested in the study of structure/function relationships of metallo-enzymes catalyzing reactions processing gases such as H₂, CO and CO₂ (Fontecilla- Camps et al., 2009). In order to function, these enzymes require strict anaerobic conditions, like those existing at the onset of life. Two of them, carbon monoxide dehydrogenase (CODH) and acetyl-coenzyme A synthase (ACS), are involved in CO₂ fixation according to the Wood/Ljungdahl pathway (Ragsdale and Pierce 2008). In acetogenic bacteria, CODH and ACS form a tight complex with a long hydrophobic tunnel that allows the passage of CO between the two NiFeS-cluster-containing active sites (Fontecilla-Camps and Volbeda, 2008). Before the appearance of the enzymes, NiFeS-containing minerals may have catalyzed the same reactions (Wächtershäuser 1988; Russell 2007), allthough much less efficiently. A high structural similarity exists between the tunnel-forming domains of an archaeal (methanogenic) CODH, of its bacterial homologues and of the bacterial (acetogenic) ACS that forms a complex with CODH (Volbeda et al., 2009). The energy required for CO₂ reduction to CO, when this product is immediately used for acetyl-CoA formation in the CODH/ACS complex, may be provided by the oxidation of H₂. The latter reaction is typically catalyzed by a [NiFe] hydrogenase, using an active site with similar Ni-coordination as in the NiFeS clusters of CODH and ACS (Volbeda & Fontecilla-Camps, 2005). The small subunit of [NiFe]-hydrogenase contains a domain with a flavodoxin-like (βα)₅ fold. This domain binds a Fe₄S₄ cluster at the position where flavodoxin (fld) binds FMN (Volbeda et al., 1995).

The very common (βα)₅ fold is related to the nucleotide-binding Rossmann fold (Rossmann et al., 1974). In fld it is characterized by a central parallel five-stranded β-sheet with topology 21345 that is flanked by two α-helices on one side and three α-helices on the other (Sancho 2006). The fld-like domain in [NiFe]-hydrogenase contains a tunnel that allows H₂ diffusion between the enzyme environment and the buried active site (Montet et al., 1997). While preparing this paper we found that the tunnel-containing domains of CODH and ACS contain the same structural motif.

The most abundant protein fold on earth, the TIM-barrel, consists of an eight-stranded barrel of parallel β-sheets and eight surrounding α-helices. Following the crystal structure determination of two enzymes involved in the histidine biosynthesis pathway, HisA and HisF, it was shown that the (βα)₈ TIM-barrel fold probably originated by gene-duplication from a (βα)₄ half barrel (Lang et al., 2000; Höcker et al., 2001). The HisF half barrel shows a striking structural and amino acid sequence similarity with the (βα)₅ fld-like domain of Methylmalonyl CoA Mutase (MCM), suggesting a common evolutionary origin (Höcker et al., 2002). MCM also contains a TIM-barrel domain with (βα)₄ halves and it performs radical-based catalysis using adenosylcobalamin as a cofactor (Gruber & Kratky 2001). AdoMet Radical Proteins (ARPs) catalyze a similar reaction producing a 5'- deoxyadenosyl radical intermediate from S-adenosyl-L-methionine (SAM). ARPs also have a TIM-barrel-related structure and a conserved Fe₄S₄ cluster that initiates the reaction by donating an electron to SAM (Nicolet and Drennan, 2004). These enzymes catalyze key reactions in the biosynthetic pathways of numerous cofactors and vitamins, such as heme, lipoic acid and thiamin. One member, the ancient class III ribonucleotide reductase activating enzyme, is involved in the conversion of ribonucleotides into deoxyribonucleotides. Fld-like folds occur also in proteins binding tRNA and rRNA, like initiation and elongation factors. In this review, we will discuss the function of these folds in primordial biochemical processes.

2. SELECTION AND SUPERPOSITION OF STRUCTURES

We preformed two DALI (Holm et al. 2008) searches of fld-like folds, using the [NiFe]- hydrogenase small subunit domain (hd1S) and the CODH-like tunnel-domain of ACS (ACS1b). Starting from hd1S, many of the highest scores were obtained for GTP-binding proteins such as elongation factor TU (EF-Tu) (Parmeggiani et al., 2006), whereas fld (Ludwig et al., 1997) and MCM (Mancia et al., 1999) had much lower scores. As already reported (Volbeda et al., 2009), using ACS1b we found the highest score for a CODH domain of the acetyl-CoA decarbonylase/synthase (ACDS) complex (Gong et al., 2008). High scores were also obtained for DNA-binding proteins such as the lac-repressor (Bell and Lewis, 2000). However, many fold similarities were not detected in this automatic way. For example, starting from a visual comparison of topologically equivalent β-strands in ACS1b and hd1S, 87 Cα positions from a minimum of 7 oligopeptide segments could be manually superimposed to a root mean square (rms) deviation of only 2.48 Å, using the BIOMOL (http://www.xray.chem.rug.nl/Links/Biomol1.htm) program suppos.

In an attempt to find more putative relationships dating back to the first stages of protein evolution we used the SCOP (Andreeva et al., 2007) hierarchy list in the SUPERFAMILY data base (http://supfam.org/SUPERFAMILY), looking for other proteins with similar parallel β-sheet topologies, i.e. with fld-like (21345) and Rossmann-like (321456, as in ACS1b) strand orders, using as additional criterion a function in RNA-, nucleotide- or metal-binding and favouring proteins involved in gas-processing reactions. At that stage [Fe]- hydrogenase (Hiromoto et al., 2009), pyruvate:ferredoxin oxidoreductase (PFOR) (Cavazza et al., 2006) and the human dual function laminin receptor/ribosomal binding protein (LamR) (Jamieson et al., 2008) were selected. Because fld-like (βα)₅ and TIM-like (βα)₄ half barrels are related (Höcker et al., 2002), proteins lacking one of the external β-strands of the fld-like β-sheet (2 or 5 in the 21345 topology) were also considered. This led to the selection of [FeFe]-hydrogenase (Nicolet et al., 1999), the MoFe subunit of nitrogenase (Einsle et al., 2002) and formate dehydrogenase H (Raaijmakers & Romao, 2006), besides the TIM-barrel enzymes HisF (Höcker et al., 2001) and HydE (Nicolet et al., 2008). A secondary structure based alignment of the selected models was performed with the program superpose (Krissinel & Henrick, 2004). The alignment was interactively improved, minimizing deletions, using suppos.

From the suppos structural alignments of the 15 selected structures, the fld-like domain of the α-subunit of MCM turned out to be the best reference structure, giving significant similarity scores with all the other structures. Defining the Structural Similarity Score as SSS = NCα / (Nseg^1/2 x Δrms), where NCα is the total number of superimposed Cα atoms, Nseg the minimal number of compared peptide segments in the two structures (choosing the one with the minimum number of insertions) and Δrms the rms deviation of the superimposed atoms (SSS), scores ranging from about 10 to 19 Å^-1 were obtained (Table 1). In the following sections, we will further discuss the structural alignments from a functional point of view.

3. FLD-LIKE FOLDS IN HYDROGENASES

The common function of flavodoxin and domain S1 of [NiFe]-hydrogenase (hdS1) is electron transfer. Although the folds are similar (Fig. 1), the electron carriers are very different, being a Fe₄S₄ cluster in hdS1 and a flavin mononucleotide (FMN) in flavodoxin. It may be speculated that, before the invention of oxygenic photosynthesis, hdS1 existed as an independent electron carrier, and afterwards its O2-labile cluster was replaced by a flavin. Alternatively, flavodoxin may have been selected at the expense of hdS1 after the great oxidation event that occurred about 2.4 Gyr ago (Konhauser 2009). It is remarkable (Fig. 1) that the same fold binds the Fe(CO)₂-guanylylpyridinol (FeGp) active site in [Fe]- hydrogenase, the Fe₄S₄-Fe₂(CN)₂(CO)₃S2CH₂NH active site of [FeFe]-hydrogenases and the MoFe₇S₉X cluster of nitrogenase, where X is probably N (Fontecilla et al., 2009 and references therein). All these metal sites are bound at the C-terminal side of a 4- to 5-stranded parallel β-sheet, as the nucleotide binding site in Rossmann-like folds (Rossmann et al., 1974), and have in common the capacity to metabolize H₂. As already noted by Douglas Rees, the active site in [FeFe]-hydrogenase and nitrogenase is bound between two such folds (Rees 2002). Since the reduction of N₂ to NH₃ is one of the most energy-consuming reactions in biochemistry, requiring 16 ATP molecules, it is difficult to imagine that this was the original function of nitrogenase. It seems more likely that it first catalyzed a simpler reaction, one possibility being the two-electron reduction of cyanide, produced by volcanic sources, to ammonia and formaldehyde (Li 1982), and that its contemporary function evolved only when more reduced sources of nitrogen became scarce and the atmosphere more oxidizing and N₂- rich (Fedonkin 2009).

4. FLD-LIKE FOLDS IN TIM-BARREL ENZYMES

A functionally important feature of the [FeFe]-hydrogenase maturase HydE is its binding of a Fe₄S₄ cluster. Like in hdS1 this has an electron transfer function and likewise it is located at the C-terminal side of a parallel β-sheet. Additional similarities may be observed in a structure-based sequence alignment (Fig. 2).

In HydE, like in other ARPs (Nicolet & Drennan 2004), the Fe₄S₄ cluster is bound by a Cx₃Cx₂C motif that is located in a loop between the first β-sheet and the first α-helix (labeled β2 and α1 in Fig. 2) of the (βα)₄ half-barrel that can be superimposed to the (βα)₅ fld fold. The same connecting loop in hdS1 contains a Cx₂C motif that binds two of the four iron atoms of the Fe₄S₄ cluster.

When the TIM-barrel structures of HydE and MCM are superimposed (Table 2), the Fe₄S₄ cluster of the former and the cobalamin (B12) cofactor of the latter occupy the same position (not shown). Taking into account the fold similarity between the HydE (βα)₄ half-barrel and the MCM fld-like domain (Table 1) and the fact that both enzymes catalyze 5’-deoxyadenosyl radical-based chemistry, a common evolutionary origin is very likely. The HydE half-barrel shows an even higher structural similarity (Table 2) with HisF, the other TIM-barrel enzyme selected here. HisF functions in the histidine biosynthesis pathway, catalyzing the reaction of N'-[(5'-phosphoribulosyl)-formimino]-5-aminoimidazol-4-carboxamid ribonucleotide with ammonia, followed by the cleavage of the condensation product into 5-aminoimidazole-4- carboxamide ribotide and imidazoleglycerol phosphate (Lang et al., 2000). The structure shows two phosphate binding sites (Fig. 1) that could be part of the ribonucleotide substrate. One of these corresponds to the Rossmann fold nucleotide binding site (Rossmann et al., 1974).

5. FLD-LIKE FOLDS IN PROTEINS INTERACTING WITH RNA AND DNA

Elongation factors such as EF-Tu, along with GTP, play a central role in protein synthesis, controlling the growth of the protein polypeptide upon aminoacyl-tRNA binding to the ribosome. This is a function that probably evolved at the beginning of protein synthesis. Some of the EF-Tu interactions with ribosomal RNA have been recently characterized at 3.6 Å resolution (Schmeing 2009). Elongation of the polypeptide chain depends on the GTPase activity of EF-Tu, which is performed by its fld-like N-terminal domain, assisted by the ribosome. The GDP product binds at the C-terminal end of the β-sheet (Fig. 1), close to the position of the Fe₄S₄ cluster in the superimposed hd1S and HydE structures. As a ribosomal binding protein, human LamR, which likewise contains a fld-like domain, is also involved in protein synthesis, whereas in its second, probably more recently evolved function as laminin receptor, it is a cell surface protein (Jamieson 2008). An fld-like domain has also been found in proteins interacting with DNA, such as the lac repressor. In this case it may function mainly as a support for the N-terminal helical region that is responsible for DNA-binding function (Bell and Lewis, 2000).

6. USE OF FLD-LIKE FOLDS IN ANAEROBIC CO₂ FIXATION

In the primordial Wood-Ljungdahl carbon fixation pathway, two CO₂ molecules are converted to a Coenzyme A (CoA) bound acetyl group (Ragsdale & Pierce 2008). In the socalled methyl branch one CO₂ is stepwise reduced to a cobalt-bound CH₃ in a Corrinoid FeS Protein (CFeSP). In bacteria this requires the successive involvement of a formate dehydrogenase (FDH) and four tetrahydrofolate (H₄F)-dependent enzymes: 10-formyl-H₄Fsynthetase (FH₄FS), 5,10-methenyl-H₄F-cyclohydrolase/dehydrogenase (MH₄FCD), 5,10- methylene-H₄F-reductase (MH₄FR) and methyl-H₄F-dehydrogenase:CFeSP-methyltransferase (MetR). Archaea use methanopterin (MPT)-dependent enzymes (Shima et al., 2002) that are not further discussed here, although it is interesting to note that the common pterin part of H₄F and MPT originates from GTP (Maden 2000). Upon reduction of the second CO₂ molecule of the Wood-Ljungdahl pathway by the NiFe₄S₄ cluster of CODH, the CO product is combined by the ACS Ni₂Fe₄S₄ cluster with the CFeSP-carried CH₃ group and CoA. As shown in Fig. 1, the CODH and ACS domains containing the tunnel used for CO transfer have a fld-like structure similar to the [NiFe]-hydrogenase hdS1 domain, except that the tunnel is located on the other side of the β-sheet. The NiFe₄S₄-containing active site of CODH is located between two such domains (for simplicity only one is shown in Fig. 1), at a position not far from the Fe₄S₄ cluster of the superimposed hdS1 domain.

The first enzyme of the methyl branch, FDH, contains a tungstopterin cofactor between two fld-related folds, at the C-terminal side of the parallel β-sheets (Raaijmakers 2002). It is homologous to molybdopterin-containing FDH (Raaijmakers and Romao, 2006) (Fig. 1) and forms a complex with a [NiFe]-hydrogenase in formate-hydrogen-lyase. The transition from W- to Mo-containing enzymes was probably provoked by the progressive oxygenation of the atmosphere and the oceans: W sulfides are more soluble than W oxides whereas the opposite is true for Mo (Fedonkin 2009). It is remarkable that the H₄F-dependent enzymes of the methyl branch also contain fld-like or TIM-barrel folds (Table 2). A significant fraction of FH₄FS (Radfar et al., 2000) resembles domain 1 of EF-TU, whereas the two domains of bifunctional MH₄FCD (Shen et al., 1999) are most similar to the C-terminal domain of MCM and the fld-like domain of [Fe]-hydrogenase, respectively. Moreoever, the superposition of the latter with [Fe]-hydrogenase results in a sequence identity of 27.8%. The TIM-barrels of MH₄FR (Guenther et al., 1999) and MetR (Doukov et al., 2000) closely resemble their counterpart in HisF. A next step in CO₂ fixation: the reductive carboxylation of acetyl-CoA to form pyruvate may be catalyzed by pyruvate:ferredoxin oxidoreductase (PFOR) (Chabrière et al., 1999). This modular enzyme contains 3 domains with fld-like folds and the Mg-thiamin pyrophosphate (TPP) active site cofactor binds between two of them (Fig. 1 shows only one), again at the C-terminal end of the β-sheets. All these similarities suggest that the same basic fold was sufficient to bind the nucleotide-based and metal-cluster cofactors required to catalyze the initial steps of carbon fixation.

7. THE ORIGIN OF FLD-LIKE FOLDS IN THE RNA WORLD

The discovery of ribozymes prompted W. Gilbert in 1986 to propose the existence of an "RNA World", preceding the emergence of DNA and proteins (Gilbert, 1986). The most attractive aspect of this proposition is the plausible unification of informational capability and catalysis in the same molecule. How did "RNA World" molecules eventually become associated with proteins and what was the driving force? In extant organisms, protein synthesis relies deeply on RNA molecules: mRNAs carrying the genetic message transcribed from DNA, tRNAs binding amino acids and rRNAs catalyzing peptide synthesis in the ribosome (Nissen et al., 2000). Since evolution has no foresight, the translation system defined by the genetic code could not have evolved as the result of selection for protein synthesis, but should have been a by-product of evolution drive by selection for another function via a so-called exaptation route (Wolf and Koonin, 2007). Abiotically generated amino acids and oligopeptides may have stabilized RNA structures (Poole et al. 1998; Noller et al., 2004), thereby improving their replicase function and lifetime. Originally such peptides probably lacked well-defined three-dimensional structures and may have resembled contemporary intrinsically unfolded proteins (IUPs). Interestingly, there are many examples of IUP-nucleic acid interactions (Tompa and Csermely 2004). For instance, several 50-S ribosomal proteins appear only partially globular in the crystal structure, with their extended unfolded regions interacting with the RNA catalytic core (Ban et al., 2000).

A discussion of the development of the genetic code (see e.g. Davis, 1999; Yarus et al., 2009) is outside the scope of this review, but it is attractive to speculate that it was linked to the selection of increasingly longer oligopeptides to stabilize RNAs that eventually evolved into ribosome-like structures with improved capacity to catalyze peptide bond formation. A small fraction of the polypeptides that were eventually produced that way may have adopted fld-like structures, assuming that these fold easily. Indeed, the structure-based sequence alignment of the selected enzymes of this review (Fig. 2) shows little amino acid conservation, suggesting a large tolerance of the fld-like fold for sequence variations. Some sequence conservation is visible at the level of side chain hydrophobicity, especially for the more buried β-strands (labeled β2, β3 and β4) and there appears to be a concentration of active site residues after the β-sheets. In this respect, a recent database study suggests many distant relationships between various fold superfamilies arising from 20 to 40 residue long peptide fragments that are supposed to descend from an ancestral pool originating in the RNA world; the first folded proteins would have resulted from the combination and amplification of such fragments (Alva et al., 2010). However, our data do not corroborate the speculation that the fld-fold arose from the repetitious duplication of a Lys-Gly/Glu-Ala-Asp-Val motif (Kobayashi & Fox, 1978). The production of fld-like structures capable to bind RNA-derived nucleotides and mineral-derived metal clusters with much improved catalytic properties might have conferred an evolutionary advantage, giving rise to the development of life forms more and more resembling those we know today.

A recent structural genomics study of metabolic pathways in Thermotoga maritima, which may represent the deepest lineage of eubacteria, showed the central pathways to be dominated by a small number of folds, with significantly higher frequencies for TIM-barrel and fld-like ones than observed for the ensemble of structures in the Protein Data Bank (Zhang et al., 2009). The presence of fld-like domains and their derived TIM-barrels in a wide variety of enzymes with important functions, ranging from electron transfer, energy-generating H₂- cleavage, carbon and nitrogen fixation to the radical-based generation of special cofactors and the elongation of growing polypeptide chains witnesses their great utility for living organisms and their great antiquity in evolution.

Acknowledgements:

We thank the Commissariat à l’Energie Atomique and the Centre National de la Recherche Scientifique for institutional support, and the Agence Nationale de la Recherche and the BIOTEC programme of the European Union for funding.