The xylene ring of riboflavin (vitamin B2) is assembled from two molecules of 3,4-dihydroxy-2-butanone 4-phosphate by a mechanistically complex process that is jointly catalyzed by lumazine synthase and riboflavin synthase. In Bacillaceae, these enzymes form a structurally unique complex comprising an icosahedral shell of 60 lumazine synthase subunits and a core of three riboflavin synthase subunits, whereas many other bacteria have empty lumazine synthase capsids, fungi, Archaea and some eubacteria have pentameric lumazine synthases, and the riboflavin synthases of Archaea are paralogs of lumazine synthase. The structures of the molecular ensembles have been studied in considerable detail by X-ray crystallography, X-ray small-angle scattering and electron microscopy. However, certain mechanistic aspects remain unknown. Surprisingly, the quaternary structure of the icosahedral β subunit capsids undergoes drastic changes, resulting in formation of large, quasi-spherical capsids; this process is modulated by sequence mutations. The occurrence of large shells consisting of 180 or more lumazine synthase subunits has recently generated interest for protein engineering topics, particularly the construction of encapsulation systems.
All living cells are absolutely dependent on the flavocoenzymes FMN and FAD for a wide variety of redox reactions. More recently, flavocoenzymes have also been implicated in a variety of non-redox processes, including DNA repair, light sensing, circadian timekeeping and some aspects of non-redox catalysis . FMN and FAD are universally biosynthesized from riboflavin (vitamin B2) (6 in Scheme 1), whose discovery and synthesis were one of the highlights of organic chemistry and biochemistry in the 1930s. Whereas plants and many micro-organisms biosynthesize the vitamin de novo, animals depend on dietary sources.
Riboflavin is a bulk commodity that is required for vitamin supplementation, animal husbandry and food coloring. Its global annual sales are in the range of 3000 tonnes. Manufacture by chemical synthesis has been replaced by powerful fermentation technology .
Studies on the biosynthesis of riboflavin started in the 1950s . Biotechnological aspects were a significant driving force; early biosynthetic work was almost exclusively performed using naturally occurring flavinogenic organisms . A second driving force for the investigation of riboflavin biosynthesis was and continues to be the fact that the pathway is essential in many bacterial pathogens but absent in the human host; hence, inhibitors of riboflavin pathway enzymes may be developed as antibiotics that would be exempt from target-related toxicity, similar to the situation with cell-wall biosynthesis inhibitors.
The riboflavin pathway has been studied in considerable detail in a variety of micro-organisms, and, to a lesser degree, in plants [5, 6]. Starting from GTP as substrate, the initial steps of riboflavin biosynthesis involve hydrolytic release of inorganic pyrophosphate, opening of the imidazole ring with formation of formate, hydrolytic cleavage of the position 2 amino group, and reductive transformation of the ribosyl moiety into the ribityl side chain of the vitamin.
In all organisms studied, the early steps of the riboflavin biosynthesis pathway converge on 5-amino-6-ribitylamino-2,4(1H,3H)-pyrimidinedione 5′-phosphate as a common intermediate. This compound is dephosphorylated in an as yet unknown manner. The resulting 5-amino-6-ribitylamino-2,4(1H,3H)-pyrimidinedione (3 in Scheme 1) is condensed with l-3,4-dihydroxy-2-butanone 4-phosphate (4 in Scheme 1) with release of inorganic phosphate. The reaction is catalyzed by lumazine synthase (B in Scheme 1), and this is the main topic of this review. The reaction product, 6,7-dimethyl-8-ribityllumazine (5 in Scheme 1), is converted into a mixture of equivalent amounts of riboflavin (6 in Scheme 1) and 5-amino-6-ribitylamino-2,4(1H,3H)-pyrimidinedione (3) by riboflavin synthase (C in Scheme 1). The highly unusual reaction involves transfer of a four-carbon unit between two identical substrate molecules. The enzyme's second product, 5-amino-6-ribitylamino-2,4(1H,3H)-pyrimidinedione (3), may be recycled by lumazine synthase. Hence, the final steps in the biosynthesis of the vitamin are characterized by an unusual, cyclic reaction topology (Scheme 1).
In contrast to the variability in the early reaction steps of the pathway, the final reaction steps catalyzed by lumazine synthase and riboflavin synthase are universal in all organisms that produce riboflavin. However, as described in detail below, these enzymes show an astonishing degree of structural variation in different organisms, and their reaction mechanisms are characterized by extraordinary complexity.
Riboflavin biosynthesis has been repeatedly reviewed [5-8]. This review focuses on the structural features of lumazine synthase and riboflavin synthase assemblies, and on the main steps in the reactions catalyzed by lumazine synthase and riboflavin synthase, as far as there is a need for an understanding of their joint action in the lumazine synthase/riboflavin synthase complex. Both enzymatic reactions still present some intricate mechanistic problems, despite more than 50 years of research.
Homotrimeric riboflavin synthases of eubacteria, fungi and plants
Studies in the late 1950s on flavinogenic ascomycetes revealed an intensely green fluorescent material (initially designated G compound) that was subsequently identified as 6,7-dimethyl-8-ribityllumazine (5 in Scheme 1), the first known intermediate in the biosynthesis of riboflavin . This seminal observation opened the way for discovery of riboflavin synthase in the 1960s [10-12]. Surprisingly, the enzyme required no second substrate in addition to 6,7-dimethyl-8-ribityllumazine (5) and no cofactors. The apparent mystery was resolved by the discovery that 6,7-dimethyl-8-ribityllumazine (5) functions both as a donor and acceptor of a four-carbon moiety . Formally, the reaction must therefore be categorized as a dismutation, albeit a highly unusual one, as typical dismutation reactions involve exchange of simple particles such as hydride ions rather than multi-atomic moieties. Even more surprisingly, this extraordinary reaction proceeds without any catalyst under relatively mild conditions (boiling of neutral or acidic aqueous solutions of the enzyme substrate under anaerobic conditions) .
As shown in the 1960s, riboflavin synthase is characterized by rigid substrate specificity. The 5′-deoxy analog of the substrate may be converted at a low rate, but various stereoisomers of the substrate were unable to serve as substrates [15, 16].
In the 1970s, it was shown that transfer of the four-carbon unit between the two identical lumazine substrates of riboflavin synthase is regio-specific, with a head-to-tail orientation [17, 18]. Twenty years later, this pioneering finding prompted the hypothesis of a quasi-c2-symmetric arrangement of two identical substrate molecules at the active site .
Relatively recently, it was shown that the riboflavin synthase reaction proceeds via a pentacyclic adduct (22 in Scheme 2) of two substrate molecules ; this intermediate may be cleaved by the enzyme in two possible ways, which produce either one molecule each of riboflavin (6) and 5-amino-6-ribitylamino-2,4(1H,3H)-pyrimidinedione (3) (forward reaction) or two molecules of the enzyme substrate, 6,7-dimethyl-8-ribityllumazine (5) (reverse reaction) [21, 22]. Remarkably, the enzyme catalyzes the forward and reverse reactions with similar rates. Clearly, the mechanistic complexity of the enzyme-catalyzed reaction is astounding.
Structure and pseudosymmetry
By 1970, riboflavin synthase had been purified approximately 4000-fold from baker's yeast [23, 24]. Although only limited amounts of pure protein were prepared, the early investigators unraveled many features of the complex reaction, including a detailed study of isotope effects and stereo- and regio-chemical aspects.
The protein became more accessible after identification of Bacillus subtilis mutants with enhanced riboflavin production, which produced pure protein after approximately 100-fold purification [25, 26]. At this point, analytical ultracentrifugation was used to show that the protein was a homotrimer of 25 kDa subunits. It was also shown that each subunit binds up to two substrate analog molecules but only one molecule of riboflavin . Moreover, 19F-NMR studies of enzyme/ligand complexes produced extremely complex datasets suggesting that the six substrate analog molecules bound to a given protein molecule may all be topologically non-equivalent [28-31]. In approximately 1990, the amino acid sequence of riboflavin synthase became available by peptide sequencing, and, in parallel, by sequencing of the riboflavin operon of Bacillus subtilis, which encodes all known riboflavin biosynthesis enzymes of the micro-organism [19, 32]. The amino acid sequence provided strong support for the concept of a pseudo-c2-symmetric active-site topology, as marked intramolecular sequence similarity suggested that each subunit folds into two closely similar folding domains, each with the capacity to bind one substrate molecule .
The two-domain concept was further supported by recombinant expression of the putative N-terminal domain, which was shown to form a c2-symmetric homodimer by analytical ultracentrifugation in conjunction with NMR analysis of protein/ligand complexes . Notably, each subunit of the homodimer binds one molecule of the riboflavin synthase substrate (5) or riboflavin (6). Not surprisingly, however, the artificial homodimer lacked catalytic activity. It had been taken almost for granted that the homotrimeric riboflavin synthase would have c3 symmetry, but X-ray structure analysis revealed a completely unexpected molecular topology that lacked trigonal symmetry but was replete with pseudo-c2-symmetry relationships . In line with the earlier arguments based on amino acid sequence, the X-ray structure of the enzyme from Escherichia coli confirmed that each monomer folds into two closely similar domains characterized by an RMSD of 2.0 Å. Each domain comprises two structurally similar six-stranded β-barrels [34-36]. The C-terminal segment, which does not participate in the intramolecular sequence similarity, forms a five-turn α-helix. The homotrimer arises by contacts between the C-terminal helical segments. The contact area is rather small. The two folding domains of each respective monomer are related by pseudo-c2 symmetry . Moreover, the N-terminal domain of one subunit and the C-terminal domain of a second subunit form an interface. The two domains are also related by pseudo-c2 symmetry. Due to the spatial relationship of the various pseudo-c2 symmetry axes, an early Patterson analysis incorrectly suggested a c3-symmetric molecule .
The interface of the special N-terminal/C-terminal domain pair encloses a cavity that is immediately suggestive of a catalytic site, although no ligand was present in the crystals under study. This hypothesis was confirmed by studies on riboflavin synthase from Schizosaccharomyces pombe and on the artificial N-terminal domain construct of riboflavin synthase from E. coli [33, 38]. The artificial homodimeric N-terminal domain is characterized by strict c2 symmetry, but the dimerization interface differs from the putative active-site interface of the catalytically active homotrimer. The artificial domain was crystallized in complex with riboflavin, thus enabling modeling of riboflavin into the N-terminal domains of the ligand-free X-ray structure of the native, homotrimeric E. coli riboflavin synthase.
Crystals of the S. pombe enzyme surprisingly failed to show any homotrimers, although the native enzyme in solution has been clearly shown to be a homotrimer . It is believed that the hydrophobic interaction of the C-terminal helices was broken by the presence of methylpentanediol in the crystallization solution. Both domains of the protein monomer carry a substrate analog in positions that are in perfect agreement with the position of riboflavin in the artificial N-terminal domain dimer.
Saturating mutagenesis of the active-site environment in the E. coli riboflavin synthase trimer did not identify any amino acid side chain that may be involved in covalent catalysis . However, it was shown that the catalytic activity of the enzyme depends on an intact N-terminus with a highly conserved sequence motif MFTG, which may be involved in positioning of the acceptor substrate. Modeling the substrate into the ligand-less X-ray structure of the native, homotrimeric E. coli riboflavin synthase suggests that the protein may function like a waffle iron or forging press to manufacture the pentacyclic dimer by covalently fusing two identical substrate molecules (Fig. 1).
More specifically, it is assumed that the protein undergoes major conformational changes during the catalytic cycle. Thus, the special domain pair may first open up (by breaking the contact between the N- and C-terminal domains of two adjacent monomers) in order to accommodate one substrate molecule in a shallow surface lacuna on each respective domain. Due to the pseudo-c2 symmetry of the domains in the ‘special pair’, subsequent closing of the ‘waffle iron’ brings the two bound substrate molecules into a pseudo-c2 arrangement where dimerization may take place, yielding the pentacyclic intermediate (Fig. 1). Based on the topological details of the protein substrate complex, there is no doubt that the N-terminal domain engaged in the special pair is the acceptor site and the C-terminal domain is the donor site for transmission of the four-carbon moiety. Re-opening of the special domain complex may be associated with fragmentation of the pentacyclic intermediate. Importantly, fragmentation of the pentacyclic intermediate by enzyme catalysis may yield either a pair of substrate molecules (reverse reaction) or a pair of product molecules (forward reaction). It appears likely that all six domains have the chance to participate in ‘special pair’ formation at different times, as a consequence of protein mobility. In fact, the junction of the three subunits via the C-terminal helices may confer a considerable degree of dynamic mobility to the protein, but details remain to be clarified. Fragmentation of the pentacyclic intermediate 22 (Scheme 2) in the forward direction (producing a pair of riboflavin and substrate 3 molecules) appears mechanistically straightforward. More specifically, two sequential eliminations are all that is required . Fragmentation may also be observed in the absence of a protein catalyst. On the other hand, formation of the pentacyclic intermediate from two identical substrate molecules (and the reverse reaction producing two substrate molecules from the pentacyclic intermediate) is a mechanistic enigma.
The first proposals for formation of a lumazine adduct were published in the 1960s, although, at that time, the pentacyclic intermediate was still unknown. However, these were revised in the 1970s after it was found that the two-four-carbon moieties from which the benzenoid ring is formed become fused in opposite orientations [17, 18, 23] (Scheme 2A). Later, this hypothetical mechanism was further modified to include the pentacyclic intermediate 22 that had been shown to fulfill the criteria for a kinetically competent intermediate  (Scheme 2B). However, the hypothetical reaction sequence comprises a daunting number of reaction steps.
Recently, a simpler, hypothetical reaction sequence has been proposed, with the central concept of a hydride transfer between the two substrate molecules at the active site (Scheme 2C) . A central tenet of both competing hypotheses relates to a unique property of the substrate 5, namely the extraordinary CH acidity of its position 7 methyl group, which is characterized by an apparent pKa of approximately 8. In aqueous solution, the carbanion species 11 resulting from deprotonation may engage in a number of ring closure reactions involving attacks of either the position 2 or position 3 hydroxy group of the ribityl side chain. However, in complex with the N-terminal riboflavin synthase domain, the substrate anion is exclusively bound in the exomethylene form 11, which results from deprotonation of the methyl group and is efficiently stabilized by the protein. The recently proposed reaction mechanism involves transfer of hydride ion from the position 6 methyl group of an exomethylene anion 11 bound to an N-terminal domain to an electroneutral substrate molecule 5b bound to a C-terminal domain. The resulting pair of dihydrolumazine/dehydrolumazine intermediates may then undergo a 4 + 2 cyclo-addition that results in tautomeric form 21 of the pentacyclic intermediate.
The cyclo-addition hypothesis is best analyzed in light of the fact that riboflavin synthase catalyzes conversion of the pentacyclic intermediate with similar rates in the forward and reverse direction. Viewed from this perspective, it appears that the tautomers 21 and 22 of the pentacyclic intermediate have different pre-determined fracture points that enable two fragmentation pathways, either by a reverse cyclo-addition or a sequence of two eliminations. Both processes must be visualized in the context of opening and closing the active site around the substrate or product molecules, which must involve major conformational motions. It should be noted that the novel mechanism involves a hidden redox process. This is relevant as dismutations are typically intermolecular redox reactions (e.g. the dismutations of hydroperoxide producing molecular oxygen and water by catalase, and of superoxide producing hydroperoxide and oxygen by superoxide dismutase). Database searches with the search term ‘dismutase’ retrieve over 100 000 references, but the vast majority of these papers concern superoxide dismutase. It is also important to note that hydride transfer is one of the most frequent elementary reactions in enzymology.
The lumazine synthase/riboflavin synthase complex
In the 1970s, it was found that B. subtilis produces two proteins with riboflavin synthase activity [25, 26, 41]. A species sedimenting at 4.1 S appeared homologous to the homotrimeric yeast protein that had been the object of previous studies. A second species sedimenting at 26 S, initially designated ‘heavy riboflavin synthase’, was shown to comprise a homotrimer that was identical to the 4.1 S species, together with approximately 60 copies of a 15 kDa subunit that was designated the β subunit.
The function of the β subunits remained unknown for almost two decades, but was eventually assigned as the penultimate reaction step in the biosynthesis of riboflavin. More specifically, the β subunits catalyze the condensation of 5-amino-6-ribitylamino-2,4(1H,3H)-pyrimidinedione (3) with 3,4-dihydroxy-2-butanone 4-phosphate (4) to give 6,7-dimethyl-8-ribityllumazine (5) and inorganic phosphate . The substrate of lumazine synthase is also the second product of riboflavin synthase. This apparent paradox implies that substrate 3 produced by riboflavin synthase may be recycled by lumazine synthase. Generation of one molecule of riboflavin requires two molecules of 6,7-dimethyl-8-ribityllumazine (5), which are biosynthetically equivalent to two molecules of 5-amino-6-ribitylamino-2,4(1H,3H)-pyrimidinedione (3). As two molecules of 3,4-dihydroxy-2-butanone 4-phosphate must pass through the 6,7-dimethyl-8-ribityllumazine (5) stage in order to yield one molecule of riboflavin, 5-amino-6-ribitylamino-2,4(1H,3H)-pyrimidinedione (3) serves both as a building block of riboflavin and the carrier of a four-carbon unit; hence, a proportion of 3 must be processed more than once by lumazine synthase.
Historically, the riboflavin synthase subunits of the complex were designated α, and the lumazine synthase subunits were designated β at a time when the enzymatic function of the latter was unknown. The stoichiometry of the complex appeared staggering, with three α subunits and 60 β subunits as shown by hydrodynamic studies. Electron microscopy indicated that the β subunits are arranged to form a spherical capsid, with icosahedral 532 symmetry strongly suggested by shape and stoichiometry. Last but not least, immunochemical studies confirmed that the α subunit trimer must be enclosed in the central cavity of the β subunit cavity .
A more detailed investigation of these surprising results became possible by timely expansions of crystallographic methodology and computational power. Most notably, introduction of powerful electron density averaging techniques made it possible to improve the apparent resolution of electron density maps by exploitation of structural redundancy in cases of molecular ensembles with high intrinsic symmetry [44, 45]. These techniques were applied when the lumazine synthase/riboflavin synthase complex was crystallized in 1983 . Based on the results, a multi-national group including participants from Sweden, the USA, China, Russia, Bulgaria, South Korea, Argentina and Germany started a long-term cooperation with the aim of further elucidating biochemical, structural and medical aspects of the unusual, bi-functional lumazine synthase/riboflavin synthase, which was designated as ‘heavy riboflavin synthase’ during the early phase of the investigation.
Detailed hydrodynamic studies had already shown that the ‘heavy riboflavin synthase’ comprises 60 β subunits. As electron micrographs of negatively stained molecules indicated an essentially spherical shape (Fig. 2), it appeared likely that the large protein particle would obey icosahedral symmetry [43, 47]. In the absence of specific ligands (analogs to 3 or 5), the stability of the enzyme complex was dependent on the presence of phosphate ions in a narrow pH range close to neutrality. Dissociation occurred between pH 7 and 8, and led to formation of polydisperse mixtures of large oligomers with the shape of hollow or massive spheres and a wide range of molecular weights, as determined by sedimentation analysis in an analytical ultracentrifuge [41, 43, 48]. Whereas the native particles sediment at a rate of 26 sedimentation rate (S), the larger particles resulting from re-aggregation showed complex sedimentation patterns with rates up to 48 S. However, ligand-mediated renaturation in the absence of α subunit trimers produced monodisperse spherical particles with a sedimentation coefficient of 26 S.
Small-angle X-ray scattering using a synchrotron beamline enabled a detailed, independent comparison of particle shapes and dimensions (Fig. 3) . Spherical shapes were confirmed for the small and large particles with sedimentation rates of 26 and 48 S, respectively. The diameters were 160 and 330 Å, in excellent agreement with the electron microscope observations. The ratios of the inner and outer diameters of the hollow capsids were 0.3 and 0.7 for the small and large particles, respectively, again in excellent agreement with negatively stained electron micrographs.
Crystallization of the lumazine synthase/riboflavin synthase complex with the unusual α3β60 stoichiometry was reported in 1983 . The space group P6322 appeared compatible with two hexagonal crystal packing arrangements, i.e. hexagonally most dense packing or hexagonal layer packing. As a successful X-ray structure determination is dependent on unambiguous determination of the translation components of the particles in the crystal cell, an attempt was made to discriminate between the two packing models that appeared compatible with the crystal symmetry by freeze-etching electron microscopy of microcrystals . The lattice dimensions observed in the electron micrographs were in excellent agreement with the crystallographically observed lattice constants of 156.4 × 156.4 × 312.6 Å. The observed particle size was also in excellent agreement with the data from small-angle X-ray scattering and from electron micrographs of negatively stained single molecules. Most notably, the arrangement of the protein molecules in a hexagonal layer packing was unequivocally established (Fig. 4A,B). Moreover, particle orientations in the crystal were determined by heavy metal decoration of freeze-etched crystals (Fig. 4C), due to preferential formation of heavy metal crystallites at specific surface sites of the protein molecules.
The initial X-ray dataset measured at the European Molecular Biology Laboratory outstation at the Deutsches Elektronen Synchrotron (Hamburg) showed reflections extending to a resolution of 3.2 Å. Patterson self-rotation analysis indicated the presence of all c2, c3 and c5 elements of point group I, thus confirming beyond doubt that the 60 β subunits form an icosahedral shell with triangulation number T =1 , as hypothesized previously on basis of quaternary structure arguments and the appearance of the particles in negatively stained electron micrographs [49-51].
The structure of the icosahedral capsid of the lumazine synthase/riboflavin synthase complex (initially designated ‘heavy riboflavin synthase’; subunit composition α3β60) was solved by a multiple isomorphous replacement in conjunction with symmetry averaging over the 10 subunits of the asymmetric unit, and by subsequent phase averaging to a resolution of 3.3 Å . Although such technology was already being used for analysis of spherical viruses [54, 55], the challenges with regard to computing time and storage capacity were still considerable. With the improvements of crystallographic computing, symmetry averaging, cyclic phase extension and structure solution would now be accomplished using one of the crystallographic program systems, such as the CCP4 suite . However, even the most advanced crystallographic techniques do not allow unraveling of the spatial orientation and structure of the α subunits, due to the static disorder of the α subunit trimers in the cavity of the icosahedral particles. The α trimer may in principle sit on one of the 10 local threefold symmetry axes of the capsid. However, from more recent studies of the riboflavin synthase structure (α trimer) we know that the trigonal symmetry of this particle, depending on its functional state, may be extremely poor (see section on riboflavin synthase above) . Thus, an orientation of the α trimers in the capsid core may be postulated that does not follow any symmetry at all. Encapsulation studies appear to support this assumption [57, 58]. Structure analysis of the lumazine synthase/riboflavin synthase complex has therefore only provided the atomic structure of the β subunits, which form the icosahedral T =1 capsid .
The folding topology of the β subunit resembles flavodoxin
The β subunit fold comprises a β sheet of four parallel strands that is flanked on both sides by pairs of helices, with some similarity to proteins of the large flavodoxin superfamily (Fig. 5), although there is no detectable sequence similarity between β subunits and flavodoxins. On the other hand, there is no structural similarity whatsoever between β subunits and capsid proteins of icosahedral plant or animal viruses.
The capsid of the lumazine synthase/riboflavin synthase is best described as a dodecamer of pentons, in light of the close interaction in the pentamer assemblies as compared to the weaker interactions in the trimer and dimer ensembles . Thus, the capsid may best be described as a dodecahedral assembly (dodecahedra and icosahedra belong to the same point group), but the details were not known in the early phase of structure determination, hence the historic designation as ‘icosahedral’. The N-terminus of each subunit serves as an additional, parallel β strand for the central β sheet of the adjacent subunit in the pentamer, thus adding additional stability to the pentamer assembly (Fig. 6B). The α3 helices of each subunit in the penton jointly form the walls of channels along the c5 axes of the penton. Due to the twist of the helices with respect to the c5 axes, the channels with a length of 30 Å and a minimum diameter of 9 Å have superhelical shape (Fig. 6A).
Cyclic protein oligomers are quite common in cells. However, the importance of the residues that line the central pore of the protein rings is poorly understood. In yeast lumazine synthase, there is a considerable tolerance to mutations of the β subunit surface that forms the channel lumen of the pentameric ring, leading to plasticity of the pore surface and quaternary structural integrity of the pentamers . This property may have consequences for protein engineering.
A solvent-accessible surface representation of an entire β60 capsid is shown in Fig. 7A. The 60 topologically equivalent active sites of the dodecahedral/icosahedral β60 capsids are all located at interfaces between adjacent penton subunits, close to the inner capsid surface . Binding of appropriate ligands, such as substrate 3 and its analogs, but also of inorganic ions such as phosphate, may substantially enhance the stability of the capsids (Fig. 6C).
A lumazine synthase with exceptional thermostability
The pioneering elucidation of the Haemophilus influenzae genome in 1995 was followed by an avalanche of hundreds of genome sequences. Subsequently, lumazine synthases, even of exotic origin, could be easily obtained by recombinant expression of the cognate genes. In the context of our ongoing work on factors relating to protein thermostability, we decided to study the lumazine synthase of the hyperthermophilic eubacterium Aquifex aeolicus . At approximately the same time, the structures of various recombinant lumazine synthases of bacterial, yeast and plant origin were determined [62-67].
The autotrophic A. aeolicus grows preferentially at temperatures of approximately 85 °C, but its lumazine synthase may be conveniently expressed in recombinant E. coli host strains. The structure of an icosahedral empty capsid of 60 β subunits and a relative mass of 960 kDa was solved by molecular replacement using the B. subtilis protein as a template, and was refined to a resolution of 1.6 Å with Rfree = 23.6% . By comparison, crystallographic analysis of the α3β60 complex of B. subtilis was limited to a resolution of 3.3 Å. The similarity with B. subtilis was very high, with an RMSD of 0.8 Å for the Cα carbon atoms.
With an apparent melting temperature of 120 °C, the lumazine synthase of A. aeolicus is one of the most thermostable proteins known. However, the thermally denatured protein cannot spontaneously re-fold at lower temperatures. The extraordinary heat stability is believed to be due to (a) an increase in surface-exposed, charged amino acid residues, and (b) a concomitant decrease in hydrophobic surface residues (Fig. 7A), in close agreement with observations for other thermotolerant proteins . In this context, it is important to note that the dielectric constant of water decreases at higher temperature, leading to an increased strength of ionic interactions [69, 70]. Incidentally, as the lumazine synthase of A. aeolicus has only been studied as a recombinant protein, it remains unknown whether a lumazine synthase/riboflavin synthase may be formed in the hyperthermophilic bacterium.
Reaction steps catalyzed by lumazine synthase
Lumazine synthase has 60 topologically equivalent active sites. Each of these is located at the interface of two adjacent subunits in close proximity to the inner surface of the icosahedral capsid (Fig. 8) [53, 61]. Kinetic and spectroscopic studies  have revealed a number of mechanistic details of the reaction catalyzed by lumazine synthase.
It has been shown that both enantiomers of the dihydroxybutanone phosphate (2) may serve as substrates for lumazine synthase; however, the reaction rate for the natural (S)-enantiomer is approximately sixfold higher than that for the (R)-enantiomer . The KM value for substrate 2 exceeds the KM value for the pyrimidine substrate 3 by more than an order of magnitude, i.e. 130 μm versus 5 μm, respectively, which suggests an ordered kinetic mechanism. Furthermore, strict regio-specificity was detected for the enzyme-catalyzed condensation of the carbohydrate phosphate with the pyrimidinedione 3. The methyl protons of substrate 4 exchange spontaneously with the solvent, as shown by NMR spectroscopy . In the absence of the pyrimidine substrate 3, the enzyme did not react with the carbohydrate substrate 4, i.e. the enzyme complex did not catalyze the exchange of the proton at C3 with water, nor did it act as a racemase. Thus, it was concluded that the initial step of the enzymatic reaction requires the presence of the pyrimidine substrate 3 at the active site, as well as a phosphate ion bound to the side chain of Arg127, which has a crucial influence on the stability of the β60 shell (see discussion below on assembly-controlled catalysis). The ribityl side chain of the pyrimidine substrate 3 is bound via several main-chain and side-chain hydrogen bonds in an extended conformation, which is closely similar in all available X-ray structures. Binding of substrate 4 competitively displaces the phosphate ion at Arg127; however, its manner of leaving the active site is still unclear. The enzymatic reaction is then initiated by formation of a Schiff base upon reaction of the 5-amino group of the pyrimidine substrate 3 with the carbonyl group of substrate 4 (Scheme 3). It has been assumed that this step is followed by proton abstraction, resulting in a leaving water molecule, and cleavage of the phosphate group, which is assumed to remain bound to Arg127 will be removed during the next reaction cycle. The exact timing of phosphate cleavage remains unknown. The resulting double bond is in favorable conjugation with the pyrimidine system. The enolate intermediate 8 then tautomerizes, with formation of a carbonyl group, which is then attacked by the 6-amino group of 9, resulting in ring closure. The reaction cycle is terminated by release of a water molecule, resulting in energetically favorable conjugation in the hetero-aromatic bicyclic chromophore of the product 6,7-dimethyl-8-ribityllumazine (5). A hypothetical structural model of the catalytic process describing and illustrating the binding of substrates, enantiomer specificity, proton abstraction/donation, phosphate elimination, Schiff base formation and ring closure has been presented .
On the basis of structural proximity and activity measurements on mutants of B. subtilis lumazine synthase, a critical involvement of the active-site residues Phe22, His88 and Arg127 in substrate binding and catalysis has been suggested . The phenyl ring of Phe22 moves into an orientation that is almost parallel to the hetero-aromatic system upon binding of the pyrimidine substrate 3. The observed offset-stacked geometry (inter-ring distance 3.4 Å) suggests a gain in stabilization energy by π–π interaction. Phe22 acts like a gate, controlling the path between the bulk solvent and the active-site cavity. Position 22 is occupied either by a Phe or a Trp residue, with only one exception (Ser22) among 59 compared lumazine synthase amino acid sequences . Amino acid sequence alignment has shown that position 88 is occupied exclusively by a positively charged residue, i.e. His (93%) or Arg (7%). His88 forms a hydrogen bond with the amino group of substrate 1. His88 probably plays a role in the proton donation step to the hydroxy group of compound 10, which is followed by elimination of a water molecule. Moreover, keto/enol tautomerization, cyclization and the dehydration of in termediate 8 are all candidates for proton transfer mediated by the enzyme. Arg127, which forms the ionic contact with the phosphate ion, is highly conserved in all lumazine synhase orthologs under comparison. Arg127 forms a salt bridge with Glu/Asp126, a highly conserved negatively charged side chain. Both of these charged residues form an ionic tetrad together with Lys131 and His 132, which extends over the subunit interface (Fig. 7C).
Interestingly, this tetrad constructs a pocket with the c2 symmetry-related tetrad from another pentamer. Active sites related by a twofold icosahedral symmetry axis are located adjacent to those tetrads and are accessible from the solvent space. These pockets have been considered as alternative paths for substrate entry into the lumazine synthase active sites, in contrast to the pentamer channels. From X-ray structure analysis of lumazine synthase, useful guesses (see above) were made about active-site residues that may take part in the catalytic process. In order to study the catalytic role of individual active-site residues, all amino acid residues lining the active-site cavity were modified by PCR-assisted mutagenesis . Overall, it was demonstrated that the catalytic activity of lumazine synthase was surprisingly resilient to exchange of active-site residues . The involvement of His88 in deprotonation of C7 of intermediate 5 had been proposed on the basis of X-ray structure studies . Based on its proximity to the phosphoester side chain of the Schiff base 7, His88 appears to be in an ideal position for abstraction of a proton from that side chain to initiate phosphate elimination. Furthermore, His88 may play a role in the subsequent enolization of intermediate 8 by proton donation to the methylene group and/or proton abstraction from the enolic hydroxy group. The replacement of His88 by various amino acids resulted in 5–30-fold reduced kcat values, indicating a certain influence of His88 in the proton donation/abstraction steps proposed above. Specifically, the kcat of the His88Ala mutant was 10% of the wild-type activity . Replacement of Phe22 by aliphatic or polar side chains reduced kcat by less than a factor of 10. However, the polar replacement of Phe22 by Asp reduced the KM value for the pyrimidine substrate 1 by a factor of 80, showing that the π–π interaction contributes favorably to the binding affinity of substrate 3. Substitution of Ala for Arg127, which is involved in an ionic interaction with orthophosphate, the phosphonate group of substrate-analogous inhibitors or the phosphate group of substrate 4, caused a reduction of the catalytic activity to 2% of the wild-type activity (Arg127Ala mutant). However, the Arg127His mutation retained 62% of wild-type activity. The replacements of Arg127, together with the results of structure analysis by X-ray crystallography and electron microscopy, have indicated the importance of this ionic interaction for binding of substrate 3 and the stability of the β60 capsid (see below for further discussion of capsid stability and assembly state).
It came as a surprise that the free activation energy ΔG‡ of the uncatalyzed reaction was characterized by a large entropic term TΔS‡ of −37.8 kJ·mol−1, whereas the activation entropy (TΔS‡) of the enzyme-catalyzed reaction was −6.7 kJ·mol−1 . Thus, the rate enhancement by the enzyme is predominantly achieved by arranging both substrates in a favorable topology, according to the well-known proximity effect of enzyme catalysis. Acid/base catalysis appears to be of secondary importance. In addition, it has been shown that formation of 6,7-dimethyl-8-ribityllumazine (5) from substrates 3 and 4 proceeds in neutral aqueous solution at room temperature with non-negligible velocity. However, the enzyme-catalyzed reaction is strictly regio-specific, whereas the uncatalyzed reaction appears to proceed by partitioning via at least two pathways .
Structural modeling of the bi-functional lumazine synthase/riboflavin synthase complex
Attempts to study the position of α subunit trimers inside the icosahedral β subunits by X-ray crystallography have been unsuccessful, possibly because the α subunits inside the capsids are disordered in relation to the crystal lattice. During the early phase of the work on the structures of lumazine synthases and riboflavin synthases, it was assumed that the α subunit trimer obeys strict c3 symmetry, and that its axis coincides with one of the c3 axes inside the icosahedral capsid. However, whereas the domains of the α subunits are involved in several pseudo-c2 relations, the homotrimer does not show any strict symmetry whatsoever. Despite these caveats, a computer modeling approach showed that the α3 riboflavin synthase of E. coli could be placed remarkably well inside the core space of the icosahedral β60 capsid of B. subtilis (Fig. 8), with the pseudo-c3 axis of the trimer coinciding with a c3 axis of the capsid. The model implies the presence of solvent lacunae inside the capsid, but the mean packing density of approximately 2.7 Å3·Da−1 is well within the confines expected for a multi-subunit complex.
However, a problem with this hypothetical model arises because the X-ray structure data indicate that the riboflavin synthase trimer requires significant conformational mobility for catalysis. On the other hand, it seems possible that the α subunit core and the β subunit capsid of the enzyme complex may jointly undergo structural fluctuations that present an answer to that question.
We know that both reactions, formation of the lumazine intermediate and formation of riboflavin via a pentacyclic intermediate, may proceed under relatively mild conditions in solution. However, during the reaction inside the complex, the reactants are separated from the bulk solvent and brought into optimal proximity by binding to the catalytic surfaces. This results in enhancement of the reaction velocity through a substrate-channeling mechanism , compared to the uncatalyzed reaction. Furthermore, both inward flow of substrates and outward flow of the product riboflavin influence the kinetics of the overall reaction catalyzed by the complex. Because the lumazine synthase/riboflavin synthase complex is known to be an extremely slow enzyme (turnover number = 2 min−1), speed enhancement may not be its only purpose. Other relevant factors include protection of intermediates from undesirable side reactions by compartmentalization. Inside the capsid, the reaction intermediates may move between the active sites of the α and β subunits (approximately 20 Å; indicated by yellow circles in Fig. 8) by means of three-dimensional or two-dimensional diffusion, in association with the protein surface. Per subunit, the catalytic rate of riboflavin synthase is approximately tenfold higher than that of lumazine synthase . Given the ratio of 60 β subunits and three α subunits in the complex, the overall catalytic capacity appears to be well-tuned for efficient cooperation between the lumazine synthase and riboflavin synthase modules.
The exchange of substrate and product molecules between the solvent and the capsid presents substantial problems. First, the α subunits are enclosed inside the tight-walled capsid. Second, the active sites of the β subunits are located close to the inner capsid surface. Hence, substrates must penetrate the wall of the icosahedral capsids in order to reach the active sites, and products must pass through the capsid wall in order to exit to the general cytoplasmic compartment. The capsid wall has a mean thickness of 37 Å (Fig. 7B).
The channels along the fivefold icosahedral axes have been suggested as potential ports for substrate entry and product exit. However, based on their diameter, with a bottleneck width of approximately 9 Å, the channels are wide enough to allow passage of substrates 3 and 4 but not of the more bulky product, riboflavin, which measures approximately 14 × 16 Å (Fig. 7B). Also of note, a c5-symmetric tungsten cluster was shown previously to bind tightly to the entrance of the fivefold channels, without affecting the rate of catalysis . Hence, a search for other potential ports of entry was necessary (Fig. 7C). The structure of lumazine synthase from A. aeolicus suggested channels along the twofold icosahedral axes as alternative ports for substrate entry and product export (Fig. 7C) . These channels are lined by polar amino acid side chains (Arg127, His132, Glu126 and Lys131). The inner channel entrances are located close to the β subunits. The channels appear large enough for entry of substrates 3 and 4, and may be large enough for exit of the biosynthesis intermediate 5. The lumazine synthase/riboflavin synthase complex is a peculiarity of Bacillaceae and has not been found with other eubacteria such as E. coli , which form only empty β subunit capsids that catalyze formation of 5 but not of the more bulky riboflavin. The question of complex formation has not been investigated for hyperthermophilic eubacteriaceae or plants.
Mutation studies have identified His88 and Arg127, which are part of the channel components, as important residues for lumazine synthase catalysis. Specifically, Arg127 is involved in binding of orthophosphate and of the phosphate moiety of substrate 4. In addition, the stability of icosahedral capsids with triangulation number T =1 is substantially enhanced by binding of phosphate or other multivalent ions to the phosphate binding site at Arg127.
All Bacillus and Clostridium strains that have been analyzed were found to express the lumazine synthase/riboflavin synthase complex . However, the amount of expressed β subunits is not sufficient to envelop all riboflavin synthase molecules; hence, Bacillus and Clostridium strains contain free homotrimeric riboflavin synthase together with the icosahedral lumazine synthase/riboflavin synthase complex.
The lumazine synthase of E. coli forms icosahedral capsids, as shown by crystallographic analysis, but they do not enclose riboflavin synthase . Recombinant, pseudo-mature lumazine synthase of spinach (Spinacia oleracea) forms icosahedral capsids when expressed in E. coli host strains. It is presently not known whether a lumazine synthase/riboflavin synthase complex is formed in plant cells. It should be noted in this context that the riboflavin biosynthesis genes of plants specify proteins with N-terminal plastid-targeting sequences that are believed to be removed during import of the proteins into plastids.
The pathogenic bacterium Brucella abortus, in contrast to other eubacteria, has a decameric lumazine synthase . The d5-symmetric protein is a dimer of pentamers. Interestingly, this biosynthetic protein is a strong Brucella antigen.
Archaea and yeasts, in contrast to plants and many eubacteria, have pentameric lumazine synthases that do not form icosahedral capsids or to associate with lumazine; X-ray structures have been determined for the pentameric lumazine synthase of baker's yeast Saccharomyces cerevisiae , fission yeast Schizosaccharomyces pombe [38, 78] and and the rice pathogen Magnaporthe grisea . The structure of the S. cerevisiae enzyme in complex with a structural analog of the heterocyclic substrate 3 has been solved by molecular replacement using the B. subtilis protein as template, and was refined to a resolution of 1.85 Å (Fig. 9A).
The icosahedral respective pentameric structure of lumazine synthase in various taxonomic groups has been attributed to structural features of the N-terminus and of the loop connecting helices α4 and α5. In the case of the icosahedral enzymes, the N-terminus of each respective subunit associates with the central β sheet of an adjacent subunit, where it functions as a fifth strand. However, the N-termini of the pentameric lumazine synthases appear less ordered. In fact, the lumazine synthase of S. cerevisiae tolerates shortening of its N-terminus by up to 17 residues, albeit with a reduction of catalytic activity . On the other hand, by comparison with the B. subtilis enzyme, the yeast enzyme has an insertion of four amino acid residues (sequence motif IDEA, see below) in the loop connecting helices α4 and α5 (Fig. 9B). Computer modeling suggested that the increased spatial requirement of the loop may interfere with further assembly of the pentamers into larger structures. Specifically, clashes at the trimer interfaces that arise during attempted formation of icosahedral capsids may make it impossible for the yeast subunits to assemble beyond the pentamer level (Fig. 9C).
A combined analysis, including studies of sequence, structure and evolution, was performed with the aim of detecting determinants of quaternary structure in icosahedral lumazine synthases . Eight sequence sites were found that showed a larger increase in constraints than the rest. One site appears to be particularly prominent: the five-residue kink in the C-terminal α-helix of the subunit type forming icosahedra. The most remarkable feature is that folding of the kink in the three known structures serves to position the highly conserved Lys182 such that it is pointing outwards toward a neighboring pentamer. Lys182 is 100% conserved among icosahedral lumazine synthases, and is practically absent in the non-icosahedral enzymes.
Unexpected results from a sequence insertion into icosahedral lumazine synthase
The potential role of an amino acid insertion into the loop between helix α4 and α5 as a hindrance to icosahedral assembly has been tested experimentally and gave unexpected results. Specifically, the IDEA sequence motif of the pentameric lumazine synthase from yeast was inserted homotopically into the sequence of the enzyme from A. aeolicus, with the expectation that it would allow formation of stable, non-associating pentamers. However, the modified subunits were found to associate as hollow, quasispherical assemblies with a diameter of approximately 290 Å (Fig. 10A), sedimenting at a rate of approximately 48 S. Thus, the insertion did result in modulation of the assembly behavior, but in a different way to that which was expected. The data were compatible with icosahedral capsids of 180 subunits with triangulation number T =3. Capsids of similar size had been observed previously  when β subunits from B. subtilis were exposed to slightly acidic pH conditions.
Whereas icosahedral proteins with triangulation number T =1 may be assembled entirely from pentamers, larger modules with triangulation numbers T >1 require pentamers and hexamers as building blocks . Based on this concept, we have attempted to model an icosahedron with T =3 using lumazine synthase subunits of A. aeolicus. The diameter of the modeled icosahedron was in line with the experimentally observed diameter of the large quasispherical shells (approximately 290 Å).
Ligand binding, capsid stability and assembly
Inside wild-type B. subtilis cells, the β subunits of lumazine synthase associate with α subunit trimers to form the bi-functional α3β60 lumazine synthase/riboflavin synthase complex, but empty icosahedral shells of β subunits may be assembled in vitro (Fig. 12). Both capsid types are only stable within a relatively narrow pH range around neutrality. Their stability may be enhanced by binding structural analogs of the lumazine synthase substrate 3 and by binding of divalent anions such as phosphate (Fig. 11) . Outside the narrow pH range where capsids with triangulation number T = 1 are stable, larger particles are formed. As observed by electron microscopy, they may be quasispherical or distorted. The electron microscopic observation of large filled particles may result from enclosure of a β60 particle inside a larger β180 particle (Fig. 10B).
Cryo-electron microscopy of the large particles resulting from insertion of the IDEA motif into the sequence of A. aeolicus lumazine synthase yielded unexpected results , and provided clear evidence that quasi-equivalent contacts  were not formed in these particles. Most importantly, the central channel of the pentameric building blocks appeared significantly widened, indicating that the mode of interaction between the pentamer units and the topology of the subunit interfaces must have undergone significant changes. Specifically, the subunits of the pentamer units are rotated, and their distance from the c5 axis appears larger compared to the wild-type protein. However, if we assume that even the wild-type protein is able to temporarily undergo similar modifications by way of dynamic fluctuations, the alternative interfacing of the pentamer units may hold the key to the insufficiently explained transit of enzyme substrates and products through the wall of the capsids that is required for catalysis (Figs 13A,B and 14).
The role of electrostatic interactions for the stability of β60 particles has been studied by calculations of the electrostatic part of the stabilization energy, which indicated that pentamers of β subunits have maximal stability at a pH of approximately 8 and are more stable than dimers or trimers (Fig. 15). Although the presence of appropriate ligands has been shown to enhance the stability of β60 capsids, the calculations suggest that the experimentally observed stabilization is not due to charge effects . Modulation of the charge state of the amino acid residues at the channels along the twofold axes by changes of pH or complexation with ligands may be expected to modulate the interaction between pentamers with consecutive quaternary structure modulation, i.e. formation of large capsids. There is no experimental evidence for the existence of either isolated pentamers, trimers or dimers at appreciable concentration in aqueous solution; hydrodynamic studies indicated a strong preference of β subunits for the formation of larger assemblies encompassing multiple subunits. The electrostatic model calculations support the concept that lumazine synthase is best described as a dodecamer of pentamers.
The pentamers in this structure adopt an expanded conformation, including a widened central channel, compared to that of the native T =1 icosahedral lumazine synthase. As shown by computer modeling , expansion of the pentamer structure is the result of a rotation/translation movement of each monomer away from the fivefold axis, which also forces the inserted IDEA fragment to point toward the center of the capsid. Most strikingly, the expansion increases the diameter of the central pentamer channel and disrupts the charge network involved in ligand binding at the active site of the enzyme (Fig. 16A,B). The expanded pentameric structure provides a suitable model for an alternative conformation of the lumazine synthase pentamer, as it may also be formed during catalysis in the enzymatically active T =1 form of lumazine synthase.
A model for assembly-controlled catalysis, including capsid breathing motions
The only point mutation in the lumazine synthase active site that results in considerable loss of enzymatic activity is exchange of Arg127, the binding site for the phosphate group of the substrate 1, for an uncharged residue, e.g. Arg127Thr. The mutants Arg127His and Arg127Lys retained enzymatic activity of 62% and 9%, respectively, compared to the wild-type enzyme . As already discussed above, Arg127 and Arg127′ were identified as parts of the symmetry-related ionic tetrads at the local twofold axes, comprising favorable electrostatic contacts between pentamers (Fig. 8C). It was found that mutations at the phosphate binding site Arg127 perturb enzymatic activity and, surprisingly, also capsid assembly.
Electron micrographs obtained by negative staining and native acrylamide gel electrophoresis showed that the Arg127His mutant protein consisted of a mixture of T =1 and T =3 capsids, whereas in the mutant proteins Arg127Thr and Arg127Ala, almost exclusively large T =3 capsids (diameter approximately 290 Å) were visible. It is thus tempting to speculate that mutations of the residues comprising the ionic tetrads close to the icosahedral twofold axes cause an impaired capsid assembly and possibly also changes in enzymatic activity. As a logical consequence, enzymatic activity, capsid stability and assembly of the icosahedral lumazine synthase complex must be coupled. Based on this obvious correlation between enzymatic activity and capsid assembly, and the findings regarding pentamer expansion described in the previous section, a model for assembly-controlled catalysis by lumazine synthase may be suggested. We assume that the expanded pentamer structure as found by electron microscopy in the T =3 assembly may also form during the catalytic process in the enzymatically active T =1 form of lumazine synthase, because the change in the overall diameter of the pentamer is only in the range of approximately 8 Å, which causes a slight widening of the T =1 icosahedral capsid. The T =1 form is characterized by a sedimentation constant of 26 S. Ligand-driven renaturation experiments of isolated β subunits (Fig. 12) resulted in a capsid species sedimenting at 26 S, with a shoulder sedimenting at approximately 29 S , with the 29 S peak having approximately one-third of the area of the 26 S peak. These components appeared as separate bands in gel electrophoresis, whereas the 29 S component migrated more slowly. According to the model below and the animation of capsid breathing (see Movie S1), it is tempting to interpret the 29 S component as a slightly widened T =1 capsid, in which only a fraction of the pentamers are in the closed state (binding sites occupied by ligand molecules) and the remaining fraction are in the open state (with unoccupied ligand binding sites).
Via breathing motions of the capsid, which may occur as random or concerted changes of the 12 pentamer structures in the capsid (open versus closed form), the flow of substrates and products to/from the active sites may be controlled (Movie S2). In a computational analysis of the low-frequency fluctuation dynamics of the T =1 lumazine synthase capsid by a coarse-grained elastic network model , collective motions (ensembles of normal modes) were detected that are in agreement with radial dimension changes of the capsid. Certain deformations of the capsid described by vector spherical harmonics preserved icosahedral symmetry and corresponded to radially directed displacements, i.e. shrinking or swelling of the entire structure, which is referred to as ‘breathing’.
With reference to Fig. 17, a catalytic cycle may be described: the substrates S1 and S2 (1 and 2 in Scheme 1) bind to the active sites in the open form of the pentamer; upon binding, the pentamer structure changes to the closed form (the favorable binding energy is used to support this conformational change); the bound phosphate ion is released at the start of each reaction cycle when substrate 4 competitively occupies the phosphate binding site (Arg127); the reaction proceeds to formation of the product lumazine, leaving an ortho-phosphate ion bound at Arg127; the open pentamer form is restored and the product lumazine is released.
Nano-biochemistry: the icosahedral lumazine synthase capsid as a ‘container’
In cellular systems, well-defined barriers and compartmental boundaries enable the distinction of ‘self’ from the rest of the molecular world. Although many biological barriers consist of lipid membranes, there is growing evidence for protein-based compartments that act as isolated environments within a cell [82, 83]. Protein-based compartments are usually made of highly symmetric structures assembled from a certain number of protein subunits. These structures are reminiscent of icosahedral viruses (as is the case for the lumazine synthase capsid), and may serve as a molecular basis for development of nano-materials that incorporate catalytic activities, sequester reactions or control transport of small molecules across the capsid wall.
The encapsulation of cargo molecules by ‘containers’ provided by protein capsids appears to be a promising strategy for the design of novel catalysts [84, 85]. A number of natural virus proteins self-assemble to form spherical capsids, to which the icosahedral lumazine synthase complex is similar. Seebeck et al.  have constructed an encapsulation system that uses a tagging scheme based on charge complementarity. Lumazine synthase from A. aeolicus was used as the container component, its capsids having been shown previously to encapsulate nanoparticles  or riboflavin synthase .
The inner surface of the capsid was engineered in such a way that four residues per monomer (Arg 83, Thr86, Thr120 and Gln123) that point into the central core space were mutated to Glu residues (construct AaLS-neg). These mutations thus produced 240 or 720 additional extra negative charges at the inner capsid surface. In addition, a His tag was attached to the solvent-exposed C-terminus of the subunits to facilitate purification of the AaLS-neg construct. An Arg10 tag was fused to the C-terminus of the chosen cargo protein, the easily detectable green fluorescent protein (GFP). Both proteins were co-expressed and purified by affinity chromatography. A significant amount of GFP–Arg10 co-purified with the tagged AaLS-neg construct. By using size-exclusion chromatography 6 h after initial elution of the sample, a fluorescence trace was detected that indicated slow dissociation of the guest molecule from the host AaLS-neg. X-ray structure analysis has shown a 60-subunit T =1 icosahedral structure for A. aeolicus lumazine synthase . However, sedimentation equilibrium studies of the GFP–Arg10/AaLS-neg complex indicated a single 3.1 MDa protein, which is consistent with a T =3 icosahedron built from 180 subunits. Comparing this result with our own work on the yeast lumazine synthase IDEA mutant (see above), a common theme appears: alteration of certain parts of the lumazine synthase β subunit surface appears to drive formation of large T =3 capsids instead of the small form. The number of GFP–Arg10 cargo molecules in the large capsids was estimated to be 3.8 ± 0.8, corresponding to a protein concentration of 22 mg·mL−1.
The simple design strategy based on charge complementarity will be suitable for encapsulation of diverse tagged cargo proteins and the design of multi-functional enzymatic complexes. The potential to artificially incorporate multiple catalytic centers, perhaps part of an enzymatic pathway, has not been realized to date, but the large protein micro-compartments, such as the T =3 lumazine synthase capsid or large virus capsids, certainly have the capacity for this level of molecular engineering.
Riboflavin synthase from Archaea: close structural resemblance to lumazine synthase
An archaeal riboflavin synthase was first purified from Methanothermobacter thermoautotrophicus . The enzyme showed no internal sequence similarity, which is a general feature of the previously known riboflavin synthases. However, it was shown by sequence analysis that its primary structure is related to that of lumazine synthases . It was later shown that the homologous enzyme from Methanococcus jannaschii forms pentamers in solution and in crystals . By crystal structure analysis, local fivefold symmetry of the oligomer was found. The folding pattern of the monomer closely resembled that of the known lumazine synthases . The active sites are located at the interfaces of adjacent subunits, and their topology resembles that of lumazine synthases. Two substrate analogs are bound per active site.
The enzyme-catalyzed reaction proceeds with the same regio-chemistry as the uncatalyzed reaction and the reaction catalyzed by riboflavin synthase from eubacteria; the two-four-carbon units for assembly of the xylene are oriented in opposite directions . The spectroscopic findings are in perfect agreement with the orientation of substrate molecules at the active site as determined by X-ray structure analysis. Pre-steady-state kinetics showed the formation of a pentacyclic intermediate resulting from connection of both lumazine moieties with cis stereochemistry . However, it was also found that the pentacyclic compounds Q (22) and Q′ produced by the eubacterial and archaeal enzymes, respectively, are in fact diastereomers (Fig. 17). The chiral centers that are generated by formation of the pentacyclic intermediate are subsequently destroyed during its fragmentation. Hence, the stereochemical differences that are inherent in catalysis by homotrimeric riboflavin synthases of plants and bacteria and the homopentameric riboflavin synthases of Archaea are not reflected in the final products (riboflavin and 3).
The known facts of riboflavin synthase catalysis suggest that one of the main features of enzyme catalysis, i.e. the proximity effect noted by Jencks , including correct positional arrangement of the substrates, is the main driving force for rate enhancement.
Lumazine synthase and riboflavin synthase as potential targets for anti-infective drugs
Flavocoenzymes are indispensable mediators of catalysis in all cells. Whereas animals and humans depend on dietary sources, most bacterial pathogens rely on endogenous biosynthesis of the vitamins. Recent work has shown that vitamin biosynthesis genes are virulence factors and/or essential genes in several Gram-negative pathogens [91-96]. This is no surprise, as it was known that Gram-negative pathogens are unable to acquire flavins efficiently from the environment. They are therefore strictly dependent on endogenous biosynthesis, and should be susceptible to suitable inhibitors of riboflavin biosynthesis. Evidence for the essentiality of riboflavin biosynthesis genes has also been reported for the major Gram-positive pathogen, Staphylococcus aureus , mycobacteria including Mycobacterium tuberculosis [98-100] and the pathogenic yeast, Candida albicans [101, 102]. As these enzymes of riboflavin biosynthesis are absent in the mammalian host, their potential inhibitors should be exempt from target-related toxicity in the mammalian host.
X-ray structures of lumazine synthases from several major pathogens have been reported (M. tuberculosis, E. coli, C. albicans, Brucellaabortus and Bacillus anthracis), some of them in complex with substrates or inhibitors [63, 67, 68, 103-106]. Notably, the recombinant M. tuberculosis enzyme has produced structures with near-atomic resolution (up to 1.6 Å). These structures provide important background information for inhibitor development.
Work on the synthesis and evaluation of inhibitors of riboflavin synthase extends over five decades. An early highlight was the discovery that introduction of oxo groups at the 6 and/or 7 position of the substrate produces inhibitors for riboflavin synthase with Ki values in the low nanomolar range . These derivatives also act as inhibitors of lumazine synthase [38, 68, 108]. Their inhibitor efficacy may be due to their structural similarity to proposed intermediates in the pathways of both enzymes that are characterized by position 7 exomethylene anion structures.
Whereas lumazine synthase was described at the protein level in the 1970s, elucidation of its enzymatic function was delayed by almost two decades . That discovery was rapidly followed by the synthesis of substrate and intermediate analogs, which have contributed substantially to our current understanding of the enzyme mechanism [28, 29, 103, 109-120]. Typically, however, these inhibitors had no antibacterial activity; this is probably due to the fact that they are not taken up efficiently, as a consequence of their predominantly hydrophilic character.
Recently, screening of random libraries [121-123] as well as virtual screening techniques , have been used for the identification of new lead compounds, and some activity against M. tuberculosis has been observed for one class of compounds identified by this approach . The large body of protein structure information that is now available provides a favorable basis for work on inhibitors of lumazine synthase and riboflavin synthase with therapeutic potential. In light of the rapid emergence of pathogens with multiple drug resistance, there is an urgent requirement for new approaches to the discovery of antibiotic compounds, and the riboflavin pathway enzymes provide an opportunity that should be explored further.
A riboflavin synthase homolog serving as an optical transponder in bacterial bioluminescence
A small family of highly fluorescent proteins with sequence similarity to trimeric riboflavin synthase have been found to serve as optical transponders in bacterial bioluminescence. They bind 6,7-dimethyl-8-ribityllumazine, FMN or 6-methyl-7-oxo-8-ribityllumazine, and have been designated lumazine protein, yellow fluorescent protein and blue fluorescent protein, respectively . The bound chromophores may be excited to the S1 state by radiation-less energy transfer from bacterial luciferase, for which they act as optical transponders that shift the wavelength maximum and intensity of emission. Although these optical transponders share considerable sequence similarity with trimeric riboflavin synthase, they lack the N-terminal trimerization helix and are monomeric in solution. In close parallel to trimeric riboflavin synthase, lumazine protein folds into two closely similar domains [127, 128]. However, only the N-terminal domain binds the fluorescent ligand 6,7-dimethyl-8-ribityllumazine . Their evolution is likely to have diverged from the riboflavin biosynthesis pathway.
After half a century of research on riboflavin biosynthesis, numerous open questions remain. Although formation of 6,7-dimethyl-8-ribityllumazine from its biosynthetic precursors and its conversion into riboflavin may proceed without catalysis under remarkably mild conditions, the enzymes that catalyze these reactions show enormous structural complexity but do not achieve much in terms of rate enhancement, in comparison with many other enzymes. Although the rates may be assumed to have been adjusted under selective pressure aimed at optimization of the amount of product required, this is not helpful in explaining either the structural complexity or the intricacies of intermediate acceleration by channeling observed for the lumazine synthase/riboflavin synthase complex. Despite our rather detailed structural knowledge of the subunits, we still do not understand the architecture of the α3β60 complex in detail. The lumazine synthase fold shows an astonishing potential to self-assemble and form molecular aggregates of different sizes and symmetries. The enzyme appears as at least four kinds of assemblies: pentamers, decamers (stacking pentamers), T =1 icosahedral capsids consisting of 60 monomers, and larger capsids with 180 (T =3) and 240 (T =4) monomers. The large aggregates consisting of 180 and 240 monomers have to date only been observed in in vitro experiments. Electron micrographs have suggested that even double capsids may be formed, as for virus structures (Fig. 10B). However, we have no evidence as to whether the large icosahedral assemblies are functional in a cell-biological context. We are also puzzled by the mechanistic pliability of the lumazine synthase fold, which has been adapted, in the case of Archaea, to serve as riboflavin synthase. We still have no detailed understanding of the complex molecular choreography of the domains of trimeric riboflavin synthase in the context of catalysis, but we are aware that the reaction may be catalyzed with similar efficacy inside the much more rigid active-site cavity of the Archaea riboflavin synthase. We have no fully conclusive explanation for the exit pathway of the enzyme product from the lumazine synthase/riboflavin synthase complex, and it is unknown how the molecular choreography believed to be involved in catalysis by trimeric riboflavin synthase may occur inside the spatially limited cavity of the icosahedral enzyme complex. The hypothesis that enzymatic formation of riboflavin from its lumazine precursor may involve a Diels–Alder-type cyclo-addition reaction deserves further scrutiny .
Last but not least, it has been reported that 6,7-dimethyl-8-ribityllumazine, but not riboflavin, activates mucosal-associated invariant T cells (MAIT cells), which may use the microbial metabolite to detect microbial infection . This discovery may suggest an entirely new outlook on riboflavin biosynthesis. This review will hopefully stimulate further work to address these questions.