Biochemical Journal

Review article

Two independent routes of de novo vitamin B6 biosynthesis: not that different after all

Teresa B. Fitzpatrick, Nikolaus Amrhein, Barbara Kappes, Peter Macheroux, Ivo Tews, Thomas Raschle


Vitamin B6 is well known in its biochemically active form as pyridoxal 5′-phosphate, an essential cofactor of numerous metabolic enzymes. The vitamin is also implicated in numerous human body functions ranging from modulation of hormone function to its recent discovery as a potent antioxidant. Its de novo biosynthesis occurs only in bacteria, fungi and plants, making it an essential nutrient in the human diet. Despite its paramount importance, its biosynthesis was predominantly investigated in Escherichia coli, where it is synthesized from the condensation of deoxyxylulose 5-phosphate and 4-phosphohydroxy-L-threonine catalysed by the concerted action of PdxA and PdxJ. However, it has now become clear that the majority of organisms capable of producing this vitamin do so via a different route, involving precursors from glycolysis and the pentose phosphate pathway. This alternative pathway is characterized by the presence of two genes, Pdx1 and Pdx2. Their discovery has sparked renewed interest in vitamin B6, and numerous studies have been conducted over the last few years to characterize the new biosynthesis pathway. Indeed, enormous progress has been made in defining the nature of the enzymes involved in both pathways, and important insights have been provided into their mechanisms of action. In the present review, we summarize the recent advances in our knowledge of the biosynthesis of this versatile molecule and compare the two independent routes to the biosynthesis of vitamin B6. Surprisingly, this comparison reveals that the key biosynthetic enzymes of both pathways are, in fact, very similar both structurally and mechanistically.

  • deoxyxylulose 5-phosphate
  • glutamine amidotransferase
  • Pdx
  • pyridoxal 5′-phosphate (PLP)
  • vitamin B6


The term vitamin B6 refers to pyridoxal, pyridoxine, pyridoxamine and their respective 5′-phosphorylated forms ( The chemical entity present at the 4′ position of the pyridine ring defines the different vitamers, which can be an aldehyde (pyridoxal), an alcohol (pyridoxol/pyridoxine) or an amine (pyridoxamine) (Scheme 1A). The vitamin is an essential metabolite in all organisms. It is well known in its enzyme cofactor forms PLP (pyridoxal 5′-phosphate) and PMP (pyridoxamine 5′-phosphate). PLP is necessary for over 100 enzymatic reactions, predominantly in amino acid metabolism [1,2]. It has been described as one of Nature's most versatile cofactors in that it participates in transamination, decarboxylation, racemization, Cα–Cβ bond cleavage and α,β-elimination reactions [3]. The unifying principle of all of these transformations is the function of the pyridine ring as an electron sink which stabilizes negative charges generated at the α-carbon of the substrate during the respective reactions [1,4,5]. PLP has also been implicated as a precursor of vitamin B1 (thiamine) biosynthesis in yeast [6,7]. On the other hand, PMP has been reported to play an important role in the biosynthesis of deoxysugars [8]. Thus vitamin B6 is implicated in more bodily functions than any other single nutrient (Scheme 1A), and, in more general terms, it is associated with, for example, nervous system function (e.g. serotonin synthesis) [9,10], red blood cell formation (haem biosynthesis) [11], vitamin B3 (niacin) formation [12,13] and one-carbon metabolism (e.g. nucleic acid synthesis) [14]. In addition, the vitamin may function as a regulator of a number of membrane ion transporters [1517], and it has been linked to modulation of hormone function because of its ability to bind to steroid receptors [18], as well as modulation of transcription factors [19]. There is growing evidence that the vitamin may also have a novel role as an anti-tumour agent [20]. Recently, a completely unprecedented function of vitamin B6 as an antioxidant was uncovered, with a potency equivalent to that of vitamins C and E [2123].

Scheme 1 Functions and biosynthesis of vitamin B6

(A) Diverse functionality of vitamin B6 and its involvement in bodily functions. The inner ring shows three of the vitamin B6 vitamers where the chemical entity at the 4′ position can be an aldehyde, an alcohol or an amine. R1 can either be a hydrogen or a phosphate group, thereby representing the vitamers shown or their phosphorylated derivatives respectively. The second and third rings indicate biochemical and physiological functions respectively in humans. (B) The three pathways of vitamin B6 biosynthesis. Pase*, the apparently unspecific phosphatases involved in dephosphorylating the phosphorylated B6 vitamers; Tase*, transaminase.

Only micro-organisms and plants have the ability to synthesize vitamin B6; other organisms must acquire it through their diet. Two independent pathways for the de novo biosynthesis of the vitamin are currently known. Considerable progress has been made in defining structural and functional relationships of both of these pathways in recent years. In the present review, we summarize these advancements and use the available information to provide a detailed structural and functional comparison of the key enzymes involved. A mechanistic evaluation provides unexpected insights into the evolutionary conundrum of two independent routes to the same molecule. We highlight remaining challenges and forecast what is to be expected in this emerging field.


The biosynthesis of the vitamin has been studied extensively in Escherichia coli and involves two branches with seven enzymatic steps (Scheme 1B, left-hand panel) [2431]. In one branch, the sequential action of the enzymes GapA, PdxB and PdxF results in the conversion of erythrose 4-phosphate into 4-phosphohydroxy-L-threonine [26,32]. The latter then undergoes oxidation and decarboxylation by PdxA to form 3-hydroxy-1-aminoacetone phosphate [33,34]. In the other branch, DXP (deoxyxylulose 5-phosphate) is derived from GAP (glyceraldehyde 3-phosphate) and pyruvate by the action of DXP synthase [34,35]. The products of the two branches, i.e. 3-hydroxy-1-aminoacetone phosphate and DXP, are then condensed by PdxJ to form PNP (pyridoxine 5′-phosphate) [36,37], which must undergo oxidation, catalysed by PdxH, to form the cofactor vitamer PLP [38]. For several years, it was tacitly assumed that this pathway is ubiquitous in all organisms that can synthesize vitamin B6. Yet, despite the importance of the vitamin, its biosynthesis had not undergone a thorough investigation in any other organism.

In 1999, pioneering work from Margaret Daub and colleagues, during a screen for proteins involved in conferring singlet oxygen resistance in the phytopathogenic fungus Cercospora nicotianae, led to the serendipitous discovery that a gene they had previously named SOR1 (for singlet oxygen resistance 1) was involved in vitamin B6 biosynthesis [21,39]. An independent study by Stephen Osmani and colleagues identified a homologous gene in Aspergillus nidulans [40]. Owing to the involvement of this gene in vitamin B6 biosynthesis, it was renamed Pdx1. Subsequent studies identified an associated gene that was named Pdx2 [41]. Interestingly, Pdx1 and Pdx2 had previously been noted in yeast (named SNZ and SNO respectively), but their precise function could not be established [4244]. It then became clear from the increasing amount of information available from genomic studies that both of these genes are widely distributed and are in fact found in all archaea, fungi, plants and most bacteria [21,45]. Interestingly, Pdx1 is one of the most highly conserved genes found in these organisms to date. However, neither Pdx1 nor Pdx2 showed homology with any gene involved in the E. coli biosynthesis pathway. Moreover, a genomic analysis revealed that the two key genes of the E. coli pathway, PdxA and PdxJ, are present predominantly in only a small subset of the γ-division of proteobacteria [21,40,45]. Thus it became accepted that the majority of organisms must have a pathway of vitamin B6 biosynthesis completely different from the one established for E. coli.

This sparked renewed interest in the vitamin, and several studies appeared on defining the nature of this ‘alternative’ pathway. Many reports using genetic approaches confirmed the presence and involvement of Pdx1 and Pdx2 in vitamin B6 metabolism in various organisms [4650]. Yet, the precise function of the proteins remained to be resolved. Sequence comparison studies had predicted that Pdx2 is a member of the class I glutaminase family, characterized by a Cys-His-Glu catalytic triad [42]. Moreover, 15N-labelling studies in various eukaryotic and prokaryotic micro-organisms harbouring the Pdx1 and Pdx2 genes led to the conclusion that the nitrogen atom in vitamin B6 is derived from glutamine, rather than from glutamate as in E. coli [51,52]. The glutaminase activity of Pdx2 has indeed since been demonstrated in several organisms, i.e. Bacillus subtilis [53], Saccharomyces cerevisiae [54], Plasmodium falciparum [55,56] and, more recently, Arabidopsis thaliana [57]. In each case, the activity of Pdx2 was shown to be dependent on Pdx1, and it was assumed that the proteins function together as a glutamine amidotransferase. Glutamine amidotransferases are typically composed of two domains, a so-called synthase and a glutaminase domain. In the glutaminase domain, glutamine is hydrolysed yielding glutamate and ammonia, the latter being then transferred to the synthase domain and is utilized in the synthesis of the respective nitrogen-containing compound [58]. However, one crucial question still needed to be solved, i.e. the nature of the substrates of the Pdx1 protein. The answer was provided independently by two groups who demonstrated that, in B. subtilis, Pdx1 and Pdx2 directly produce the cofactor PLP in the presence of glutamine and either R5P (ribose 5-phosphate) or Ru5P (ribulose 5-phosphate) in combination with GAP or dihydroxyacetone phosphate [59,60] (Scheme 1B). Besides in bacteria, this pathway has now also been shown to be in operation in apicomplexan protists and plants [55,61,62]. The unequivocal identification of the Pdx1 substrates was derived from the numerous pioneering labelling studies carried out by Ian Spenser and colleagues [7,63,64]. Thus, in contrast with the seven enzymes required by the E. coli-type pathway, it is apparent that a single glutamine amidotransferase is capable of synthesizing PLP in most organisms. In order to distinguish the two routes, we have named the E. coli-type pathway ‘DXP-dependent’ and the alternative pathway ‘DXP-independent’, on the basis of their different substrate requirements (Scheme 1B) [62].

PdxA and PdxJ: key enzymes in the DXP-dependent pathway

As outlined above, PdxA and PdxJ catalyse consecutive reactions in the final steps of PNP biosynthesis in the DXP-dependent pathway. The biochemical characterization of these individual enzymes has been hampered by the fact that the product of the PdxA reaction, 3-amino-1-hydroxyacetone phosphate, is highly unstable. Therefore enzymatic analysis of PdxJ usually has to be coupled to that of PdxA, thus complicating the derivation of kinetic parameters. However, much information has been obtained from a structural analysis of the two proteins. The dehydrogenase, PdxA, forms tightly bound dimers, each monomer having an α/β/α-fold with a central 12-stranded mixed β-sheet flanked on both sides by α-helices [65]. The active site is located in a cleft at the interface of the dimers involving residues from both monomers. The protein has a strict requirement for a bivalent metal ion and can use either NAD+ or NADP+ as a redox cofactor [34]. PdxJ, on the other hand, folds as an (β/α)8 or TIM barrel in which the eight-membered cylindrical β-sheet is surrounded by eight α-helices (Figure 1A) [66,67]. Similarly to other structures of this kind, the loops between β-strands and α-helices at the C-terminal end of the barrel form the active site [67]. The overall architecture of the enzyme is a tetramer of dimers with intersubunit contacts being mediated by three helices (α1a, α6a and α8a) in addition to those of the classic (β/α)8 barrel. We will return to PdxJ later in this review, thus specific features are discussed in more detail below.

Figure 1 Comparison of the overall architecture of PdxJ and Pdx1

(A) Ribbon representation of a PdxJ monomer (PDB code 1IXP). (B) Ribbon representation of a Pdx1 monomer (PDB code 2NV2). (C) The main-chain atoms of B. subtilis Pdx1 (PDB code 2NV2) and E. coli PdxJ (PDB code 1IXP) were superimposed and the root mean square deviation mapped on the three-dimensional structure of B. subtilis Pdx1. The increase in distance between aligned main-chain atoms is indicated by a shift in colour from blue to red. A total of 188 Cα atoms were aligned with an average root mean square deviation of 2.5 Å. (D) A ribbon representation of a monomer of B. subtilis Pdx1 from the ternary PLP synthase complex (PDB code 2NV2), coloured according to the B-factor values, which indicate an increase in thermal motion by a shift in colour from blue to red. An interactive three-dimensional version of this Figure is available at

Pdx1 and Pdx2: key enzymes in the DXP-independent pathway

Pdx1 and Pdx2 jointly catalyse the formation of PLP from a pent(ul)ose and a triose sugar in addition to glutamine, and the complex is now referred to as PLP synthase [59,60]. The first structure to be solved of an enzyme of the DXP-independent pathway was that of Pdx2 from B. subtilis (annotated YaaE) [68]. The structure at 2.5 Å (1 Å=0.1 nm) resolution revealed a classic mixed α/β three-layer sandwich fold with a seven-stranded twisted mixed parallel β-sheet, flanked by six α-helices on the N-terminal stretch of the sheet, i.e. a typical Rossmann topology. The architecture of the Pdx2 protein permitted the identification of the catalytic triad formed by Cys79, His170 and Glu172. A later study provided unequivocal evidence for the involvement of Cys79 in the B. subtilis Pdx2 through its modification with the glutamine analogue (2S-5S)-2-amino-3-chloro-4,5-dihydro-5-isoxazoleacetic acid, also known as acivicin [60]. This compound is a general inhibitor of glutaminase activity by virtue of its covalent attachment to the active-site cysteine residue [6972]. Subsequently, the elucidation of the Pdx2 structure from the apicomplexan protist Plasmodium falciparum provided further insight into the glutaminase function of this protein [55]. The latter organism is the causative agent of severe malaria and is of interest with regard to the potential of the pathway in providing a novel drug target owing to its absence from animals [55,56]. The structure of P. falciparum Pdx2, although very similar to that of the B. subtilis protein, displays some notable differences. These are predominantly the elongation of certain β-strands and an α-helix, in addition to a large insertion between β5 and β6, forming an extended loop and small helical segment. Although these features are interesting, their significance is currently not understood. Importantly, the higher-resolution structure of the P. falciparum Pdx2 protein (1.6 Å) compared with that of B. subtilis (2.5 Å) allowed a key feature of glutaminases to be observed. A prerequisite for glutamine hydrolysis by a glutaminase is the formation of a so-called oxyanion hole that serves to stabilize the negative charge carried by the glutamine amide oxygen during catalysis [73]. This feature is characterized by two backbone nitrogen atoms (Gly51 and Ala88 in P. falciparum Pdx2) [71,74]. Although the oxyanion strand was clearly resolved in P. falciparum Pdx2, the observed structure represents an inactive state, as the carbonyl oxygen of Gly51 points into the active site rather than the peptide nitrogen, thus obstructing the oxyanion hole [55].

Around the same time, the structure of the Pdx1 protein from the thermophilic organism Geobacillus stearothermophilus was published [75]. Despite the uniqueness of the amino acid sequence of this evolutionarily highly conserved enzyme [21] and the anticipation of a novel fold, it turned out to be a (β/α)8 barrel, the most common fold found in nature [76]. However, the quaternary structure of Pdx1 is indeed unique, consisting of 12 protein units assembled as two opposing hexameric rings. In solution, the protein was also shown to exist as a hexamer being in equilibrium with the dodecameric form (Kd 3.7 μM) [75,77]. The co-crystallization of two sulfate ions and a molecule of 2-methyl-2,4-pentanediol from the crystallization buffer led to the proposal that the active site is at the top of the barrel [75], as is the case for all (β/α)8 structures elucidated so far [78]. From this, certain amino acids were proposed to be involved in catalysis. Two such residues are Lys81 and Lys149; the latter was reported to be the site of imine formation with the pent(ul)ose phosphate substrate in an earlier study by Tadhg Begley and colleagues [59]. However, whereas Lys81 points into the proposed active site, Lys149 actually points away from it in the crystal structure. But to make it agree with the earlier study, a model was presented in which Lys149 was reoriented to face the active site [75]. Later, two independent studies, one employing substrate labelling and mass spectrometry [79], and the other elucidating the structure of Pdx1 from Thermotoga maritima with Ru5P bound [80], reported imine formation with the phosphorylated pentulose substrate at Lys81 rather than at Lys149 in Pdx1 (B. subtilis numbering) (Figure 2C). Whether imine formation occurs at both lysine residues is still controversial. A transfer of the pentulose-phosphate imine adduct from Lys81 to Lys149 in a so-called swinging arm mechanism has been suggested [80]. However, the recently detected chromophoric reaction intermediate formed subsequently to the imine mentioned above, was shown to be covalently attached to Lys81, and therefore would appear to rule out the swinging arm mechanism, at least before this step. In the latter study, an acid–base catalytic role was proposed for Lys149 [79]. Mutagenesis studies have indeed shown that Lys149 does play a role in PLP formation [59,79]. It is noteworthy that mutation of Lys149 with an arginine residue still permits both pentulose phosphate imine and chromophoric adduct formation, albeit at a significantly lower rate [79]. Thus the timing and capacity of the functionality of Lys149 remain to be established.

Figure 2 Active-site architecture of PdxJ and Pdx1

(A) View of the active site of the E. coli PdxJ structure in the absence of any ligand (PDB code 1HO1). The flexible loop 4 that includes Thr102 and Thr103 is shown in yellow. (B) View of E. coli PdxJ as in (A), but with 1-deoxy-D-xylulose-5-phosphate (pale blue) and inorganic phosphate (orange) bound (PDB code 1M5W). The PA and PB phosphate-binding sites are indicated as transparent orange spheres. The projection of Arg20 into the active site, which comes from a neighbouring monomer, is shown in pale green. (C) Pdx1 from the T. maritima PLP synthase complex with Ru5P bound (pale blue) via an imine bond and inorganic phosphate (orange) (PDB code 2ISS). The P1 and P2 phosphate-binding sites are indicated as transparent red spheres. In all cases, observed phosphate ions and bound substrates are shown in ball-and-stick representation. Residues involved in phosphate binding and potential catalytic residues are highlighted. Note that the numbering of Pdx1 is for T. maritima and differs from that of B. subtilis by +1.

The PLP synthase complex

As vitamin B6 biosynthesis requires assembly of the glutaminase (Pdx2) and synthase (Pdx1) subunits, insight into their mode of interaction is of great interest. A first model, based on the structure of Pdx2, proposed remote glutaminase and synthase active sites, which necessitates transfer of ammonia from the glutaminase domain through an internal tunnel to the Pdx1 active site [68]. Later, a second model, based on the structure of Pdx1 and the absence of an apparent ammonia tunnel, proposed adjacent glutaminase and synthase active sites [75]. The architecture of the glutamine amidotransferase was revealed upon successful elucidation of the structure of the PLP synthase complex by two independent groups within the last year [77,80]. In one case, stabilization of the complex from B. subtilis was accomplished by us upon substitution of asparagine for the catalytic histidine residue in Pdx2, rendering the protein unable to catalyse glutamine hydrolysis [77]. Addition of glutamine resulted in trapping of the substrate in a Michaelis complex in the Pdx2 active site and appeared to act as an adhesive holding the PLP synthase complex together. Subsequent studies employing isothermal titration calorimetry have shown that glutamine increases the affinity of the subunits for each other in the complex 23-fold [81]. Steven Ealick and colleagues, on the other hand, successfully obtained the PLP synthase complex from the thermophilic organism T. maritima [80]. For crystallization, the glutamine analogue acivicin had to be present in the buffer, but, oddly, it was not seen in the solved structure. In both cases, PLP synthase was observed to be a dodecamer of Pdx1 units, arranged in two hexameric rings to which 12 Pdx2 subunits attach, with synthase and glutaminase active sites remote from each other. The architecture of the complex has been likened to a cogwheel, with Pdx1 as the core of the wheel and the individual discrete Pdx2 units providing the cogs [77]. To date, this is a unique macromolecular assembly within the glutamine amidotransferase family. Moreover, a comparison of the structures of the subunits either individually or in the PLP synthase complex, revealed a novel mode of synthase and glutaminase interaction. A unique α-helix at the N-terminus of Pdx1, named αN (Figure 1B), proved to be instrumental for the interaction with the Pdx2 subunit. Mechanistic insight into the mode of glutaminase activation was also provided in these studies through the observation of an ordered oxyanion hole. In B. subtilis Pdx2, it is formed by the peptide nitrogens of Gly47 and Ala80 in the region between strand β3 and helix α3 and was found to be correctly poised for catalysis in the ternary complex with Pdx1 and glutamine [77]. However, in the structure of Pdx2 alone as described for the P. falciparum structure, the nitrogen of Gly47 is pointing away from the glutaminase active site, thereby impeding glutamine hydrolysis. The ternary structure revealed that the oxyanion conformation is stabilized by interaction with αN of Pdx1 and therefore explains the observed dependence of the glutaminase activity of Pdx2 on Pdx1.

Furthermore, the structure of the PLP synthase complex has provided insight into the mechanism of the transfer of labile ammonia between the remote glutaminase and synthase active sites [77,80]. The observation of cavities inside the β-barrel of Pdx1 led to the proposal of a methionine-rich tunnel for the ammonia transfer [77]. Although this hypothesis clearly requires validation, a recent study in plants has shown that the alteration of a residue at the proposed entrance to the tunnel disrupts the co-ordination between the glutaminase and PLP synthase activities [57].


Now that the key proteins of the two individual vitamin B6 de novo biosynthesis pathways have been reasonably well characterized, a structural and functional comparison of Pdx1 with PdxJ is possible and may reveal the extent to which these enzymes and their mechanisms are related.

Structural comparison of Pdx1 and PdxJ

A structural homology search with B. subtilis Pdx1 as the query using the DALI program [82] surprisingly reveals PdxJ as one of its closest homologues, despite the very low sequence identity of only 9% (Table 1). All of the listed proteins have (β/α)8 barrel structures and use phosphorylated substrates (apart from Ten1). Moreover, in addition to Pdx1 from G. stearothermophilus, four of the remaining nine catalyse the formation of heterocyclic ring compounds, i.e. TPS, PdxJ, ThiG, HisF.

View this table:
Table 1 Structural homology of Pdx1

The structural homology search was performed with DALI using B. subtilis Pdx1 (PDB code 2NV2, chain A) as a query. The matched structures are sorted by the Z-score (i.e. the strength of structural similarity in S.D.s above expected). RMSD, positional root mean square deviation of superimposed Cα atoms in Å; LALI, total number of equivalenced residues; LSEQ2, length of the entire chain of the equivalenced structure; IDE, percentage of sequence identity over equivalenced positions.

The close structural similarity of the Pdx1 and PdxJ monomers is evident from the tight superposition of structurally aligned main-chain atoms and can be emphasized further by mapping the root mean square deviation of main-chain atoms on to the three-dimensional structure of Pdx1 (Figure 1C). The gradual change in colour from blue to red represents the most to least similar regions respectively. In this way, structural variations between Pdx1 and PdxJ can be summarized and are confined to (i) the N-terminus, (ii) α-helices α6′, α6″ and α8″ of Pdx1, and α1a and α6a of PdxJ, and (iii) the C-terminal face of the barrel, including helix α2′ of Pdx1 and various loop regions. Differences in the N-terminus mainly pertain to an extension of 15 residues in Pdx1 which form αN, necessary for interaction with Pdx2 and activation of glutaminase activity as stated above. The deviation can be accounted for by the different modes of nitrogen incorporation into the pyridine ring of the two vitamers between both systems. In Pdx1, the nitrogen is derived from ammonia, resulting from the glutaminase activity of Pdx2, which must be sequestered from the solvent to maintain reactivity, thus validating the necessity of αN for the interaction with Pdx2. In PdxJ, no such interaction is necessary as organically bound nitrogen is provided in the form of 3-amino-1-hydroxyacetone phosphate [33]. The deviations in the α-helices α6′, α6″ and α8″ of Pdx1 and α1a and α6a of PdxJ can be explained by the different quaternary organizations of Pdx1 and PdxJ. In Pdx1, the interdigitation of the two hexameric rings is accomplished by elongation of α6 and insertion of α6′ and α6″, whereas intersubunit contacts between adjacent (β/α)8 barrels within the hexameric rings are maintained by helix α8′' running parallel to α8. On the other hand, intersubunit contacts between the tetramer of dimers of PdxJ are maintained by helices α1a, α6a and α8a. The differences between Pdx1 and PdxJ in the C-terminal face of the barrel can be rationalized in the light of the different substrates and chemistry employed by these enzymes and is discussed in the next section.

Comparing PdxJ and Pdx1 active-site architecture

Both PdxJ and Pdx1 exclusively utilize phosphorylated substrates [36,59,60]. It is believed that phosphorylated substrates are anchored to an enzyme mainly via their respective phosphate group(s), making use of extensive hydrogen-bonding and ionic interactions, occasionally supported by the macrodipole of an α-helix [83]. A role of the phosphate group in anchoring the substrate in the active site of Pdx1 and PdxJ is apparent for the respective five-carbon substrates. The phosphate group of both R(u)5P and DXP is lost as inorganic phosphate during the course of the reaction (see below). Substrate binding has been extensively defined for PdxJ by the structural analysis of protein complexed with substrate, substrate analogue or product [66,67,84,85]. The phosphate of 3-phosphohydroxy-1-aminoacetone, which becomes the 5′ phosphate of PNP, is located between loops 7 and 8, similarly to other (β/α)8 barrel proteins [76]. Co-ordination involves the amide nitrogens of Gly194, Gly215, His216 and the guanidino group of Arg20 (the latter is part of a neighbouring monomer) (Figures 2A and 2B) [67]. This site is referred to as PA. In contrast, the phosphate moiety of DXP is co-ordinated predominantly by basic residues, Asp11, His12, Arg47, His52, Thr102, Thr103 and Arg20 of a neighbouring subunit [84,85], and is referred to as PB (Figures 2A and 2B); note that Thr102 and Thr103 are part of loop 4 shown in yellow.

On the other hand, although it had been assumed that the active site of Pdx1 is at the C-terminal face of the barrel [75], and several mutagenesis studies had already confirmed the involvement of certain residues [77], unequivocal evidence for its location came only from the elucidation of the structure of the PLP synthase complex from T. maritima with the pentulose substrate, Ru5P, bound [80]. In addition to the observation of the C-2 carbonyl of Ru5P in a Schiff base with Lys82 and the involvement of Asp25, Asp103 and Arg148, the amide nitrogens of Gly154, Gly215, Gly236 and Ser237, as well as the side chain of Ser237, form the pentulose phosphate-binding site (Figure 2C). Note that the numbering of the T. maritima residues differs from that of B. subtilis by +1. Interestingly, sulfate and chloride ions were detected in the vicinity of the phosphate of Ru5P in Pdx1 from G. stearothermophilus and B. subtilis, albeit displaced by 2.6 and 1.5 Å respectively [75,77]. A comparison of the Pdx1 subunits from the PLP synthase complexes of T. maritima and B. subtilis reveals that the binding of the phosphate of Ru5P results in a small inward movement of helix α8′ through additional hydrogen bonding with the amide nitrogens of Gly236 and Ser237 and the side chain of the latter (Figures 2C and 3). Alterations are also observed in α2′ upon binding of Ru5P (Figure 3). Intriguingly, a second phosphate ion was found in each Pdx1 monomer at the hexameric interface, co-ordinated by His116, Arg138, Arg139 and Lys188 (the latter resides in a monomer on the adjacent hexamer) (Figures 2C and 3). The significance of this anion-binding site is not yet known, but sulfate and chloride ions have been found in equivalent positions in Pdx1 from G. stearothermophilus and B. subtilis respectively [75,77]. The two phosphate-binding sites have been designated P1 (pentulose phosphate binding) and P2 (hexamer interface) (Figures 2C and 3).

Figure 3 Effect of substrate binding on Pdx1

Stereo representation of structurally superimposed T. maritima Pdx1 (PDB code 2ISS) with Pdx1 from B. subtilis (PDB code 2NV2). Secondary structures of Pdx1 from T. maritima and B. subtilis are coloured in orange and yellow respectively. Covalently bound Ru5P and bound phosphate in T. maritima Pdx1 are depicted in ball-and-stick representation, whereas the bound chloride ion of B. subtilis Pdx1 is shown as a green sphere.

We anticipated that a structural comparison of Pdx1 with liganded PdxJ structures might provide additional insight for understanding substrate co-ordination in Pdx1. Interestingly, the position of phosphate binding is almost identical at sites PA and P1 of PdxJ and Pdx1 respectively (Figure 4). Similarly to other (β/α)8 barrel enzymes, Pdx1 utilizes α8′ and loop 7 to co-ordinate the phosphate group, but is unique in also employing residues from loop 6. However, PdxJ uses PA for binding of the three-carbon substrate, whereas Pdx1 uses the equivalent P1 site for binding the five-carbon substrate. In contrast, the PB and P2 phosphate-binding sites of PdxJ and Pdx1 respectively differ substantially (Figures 2 and 4). The distance of 21 Å between P1 and P2 in Pdx1 further questions its functionality, but it is noteworthy that the previously mentioned Lys149 points towards P2. Regardless of the function of P2, the question still pertains as to the site of triose binding. A scenario presented by Janet Smith and colleagues envisages that binding of both triose and pentose phosphate might occur in the same site [75]. However, this was discounted owing to the resulting clash of the phosphate groups, but it could still be accommodated if movements were to occur in the loop regions at the top of the barrel, with helix α8′ in particular being implicated [75]. An alternative situation can also be envisaged with regard to the recently reported detection of a chromophoric reaction intermediate derived from the Ru5P imine adduct by loss of water and phosphate [79]. As these steps occur before binding of the triose phosphate substrate and, furthermore, as the phosphate from the pentose has been lost by then, the phosphate of the triose can be accommodated in the P1 site. Thus the clash of phosphate groups would be avoided.

Figure 4 Phosphate binding in PdxJ and Pdx1

Stereo representation of a structural superposition of the active site of T. maritima Pdx1 with Ru5P bound (gold) (PDB code 2ISS) on the active site of E. coli PdxJ with deoxyxylulose-5-phosphate bound (PDB code 1M5W) (green). Note the overlap between the phosphate-binding sites P1 and PA, whereas P2 and PB are disparate.

A striking difference between the structures of PdxJ and Pdx1 is the presence of a short helical segment, α2′, in the latter that caps the active site and, moreover, is observed only in the PLP synthase complex. This section is disordered and thus lacking in the structure of Pdx1 alone. As Pdx1 alone can bind the pentose phosphate substrate, and, as α2′ is formed upon interaction with Pdx2, it has been suggested that it primes the synthase domain for acceptance of the triose sugar [77,79]. This would imply that α2′ has an intrinsic flexibility, and this is indeed reflected in the above-average B-factor values for the respective residues in the structure of the ternary complex (Figure 1D). B-factors are empirical parameters that indicate the relative mobility of an atom [86]. It is of interest that rather high B-factors are also observed for α8′ and the loop region connecting β6 and α6 (Figure 1D). Both of these regions are involved in binding the pentose phosphate substrate [80].

In PdxJ, a high degree of structural deviation from Pdx1 in the active-site architecture results from an 11-amino-acid insertion in loop 4, between β4 and α4 (residues 96–106) (Figures 1A and 2, highlighted in yellow). This region has attracted interest because of two distinct conformations observed in various ligand-bound PdxJ structures [84,85,87]. Although loop 4 is folded away from the active site in the unliganded form, it reorients upon interaction with ligands and closes the active site like a lid (compare Figure 2A with Figure 2B). The resulting shielding is expected to suppress unwanted side reactions and stabilize chemically labile intermediates. It has been postulated that occupation of both the PA and PB phosphate-binding sites is a prerequisite for this rearrangement [85,87]. The equivalent loop in Pdx1 is shorter and lacks a corresponding inherent flexibility and therefore cannot fulfil a similar function (compare Figure 1A with Figure 1B). However, closing of the active site in Pdx1 could be adopted partially by α2′, which occupies a position similar to the extension of loop 4 of PdxJ in the closed form (Figure 4). It must also be mentioned that the C-terminus of Pdx1 extends 34 residues beyond that of PdxJ and has not been observed in its entirety in any available structure so far. In the structures of Pdx1 in the absence of ligand, the last 22 residues are absent, but, in the presence of Ru5P, an additional 12 residues were observed which surprisingly fold over to a neighbouring monomer [80]. Thus it is feasible that the Pdx1 C-terminus may fulfil a function equivalent to that of loop 4 of PdxJ, by acting as a lid to shield the active site upon substrate binding, albeit possibly to an adjacent monomer.

Mechanism of de novo synthesis of PNP compared with PLP

The structural analogy between PdxJ and Pdx1 can be extended to a comparison of their likely mechanisms of action. Both enzymes employ a diverse repertoire of chemical reactions to accomplish the complex synthesis of vitamin B6. Common mechanistic steps employed by both enzymes include Schiff base formation, elimination and addition of water, elimination of inorganic phosphate, ring closure and aromatization. Although both enzymes carry out de novo synthesis of vitamin B6, they differ with respect to the vitamer produced. Whereas PdxJ catalyses the formation of PNP [36,37,75], Pdx1 catalyses directly the synthesis of the cofactor vitamer PLP [59,60]. Inevitably, the question arises as to the underlying molecular rationale for this divergence in product specificity, as well as the question of a selective advantage of either route. It can be argued that this is either a consequence of completely different chemistries or different substrates, or a combination of both. Interestingly, the subtle differences in the reaction products of PdxJ and Pdx1 appear to be reflected in the nature of the five-carbon substrates employed (Scheme 2). Ru5P (2), utilized by Pdx1, carries a hydroxy group at the C-1 position, whereas DXP (a), utilized by PdxJ, carries a hydrogen at the equivalent position. Two possible routes for PLP formation by Pdx1 have recently been proposed [79]. The main difference between these routes (using the pyridine ring numbering) is the timing of proton abstraction from C-4′ and loss of the 2′-hydroxy group, both of which are derived from the pentulose substrate. It is hypothesized that both of these mechanistic steps can occur through an internal redox reaction and are thus coupled to each other. This proposal is derived from an elegant scheme originally put forward by Ian Spenser and colleagues [64] and involves proton abstraction from C-4′, with a concomitant reduction at C-2′, resulting in elimination of the hydroxy group. One of the routes proposed for PLP biosynthesis involves 2′-hydroxypyridoxine 5′-phosphate as an intermediate, and for convenience we have used this route for a comparison with the mechanism of PdxJ (Scheme 2). The dephosphorylated derivative of this compound was found to be incorporated into pyridoxine in yeast [7]. If one is to follow the fate of the C-1 hydroxy group of Ru5P by this route, it becomes evident that the hydroxy group is retained up to the formation of 2′-hydroxypyridoxine 5′-phosphate (17a). The 2′-hydroxy group is the only difference between the latter and the product of the PdxJ reaction, PNP, thus precisely reflecting the difference in the five-carbon substrates of Pdx1 and PdxJ. Moreover, the ability of Pdx1 to directly synthesize the cofactor form, PLP, is inherent to the substrate used, whereas PdxJ must employ the oxidase PdxH to catalyse its formation from PNP. In order for 2′-hydroxypyridoxine 5′-phosphate to be converted into PLP, the internal redox mechanism described above takes place, where a proton is abstracted from C-4′, the delocalization of the resulting negative charge into the pyridine ring facilitates a concomitant reduction at C-2′, resulting in elimination of the hydroxy group. An enamine–imine tautomerization yields PLP (Scheme 2). Thus the oxidation of C-4′, which establishes the characteristic functional aldehyde group of PLP, would critically depend on the 2′-hydroxy group as a partner for the redox reaction. This hypothesis would explain the use of the pentulose substrate by Pdx1 and provide the reason for the direct synthesis of the cofactor vitamer.

Scheme 2 Mechanisms of vitamin B6 biosynthesis

Mechanisms proposed for the DXP-independent and DXP-dependent pathways of vitamin B6 biosynthesis. On the left is shown one of the two recently proposed mechanisms for the Pdx1-catalysed reaction involving 2′-hydroxypyridoxine 5′-phosphate as an intermediate. The panel on the right shows the proposed mechanism for the PdxJ-catalysed reaction. The green inset depicts the mechanism proposed by Ian Spenser and colleagues for the conversion of 2′-hydroxypyridoxine 5′-phosphate into PLP [64].

In the light of this hypothesis, one might wonder whether PdxJ could actually evolve into a ‘PLP synthase’. Surprisingly, a literature review reveals that the answer may have already been provided in 1996 by Malcolm Winkler and colleagues [88]. In that study, a PdxJ G194S variant was isolated in a suppressor mutant screen that bypassed the requirement for PdxH in PLP biosynthesis by E. coli. It was speculated that this PdxJ mutant could utilize a different substrate, employ a different reaction mechanism or activate what was called a ‘cryptic pathway’ for the biosynthesis of PLP. Unfortunately, in vitro experiments to prove that PdxJ G194S can actually synthesize PLP were not conducted at that time. However, in the light of the recent knowledge of the DXP-independent pathway of vitamin B6 biosynthesis, and owing to the in vivo nature of the experiment, a plethora of substrates would be available to the PdxJ mutant, including pent(ul)ose 5-phosphate. It is therefore plausible that the G194S substitution permits PdxJ to accept Ru5P as a substrate, thereby yielding 2′-hydroxypyridoxine 5′-phosphate as an intermediate, which could then be either enzymatically or spontaneously transformed into PLP as described above. Clearly, the PdxJ G194S mutant must be carefully reinvestigated.

Likewise, a comparison of the reaction mechanisms for the synthesis of PNP and PLP by PdxJ and Pdx1 respectively reveals a high degree of similarity. Pdx1 and PdxJ both employ iminium formation as a means of eliminating the C-4 hydroxy group by acidification of the C-3 proton (compare 4 with e in Scheme 2). The protonated Schiff base functions as an electron sink, providing stabilization of the developing carbanionic intermediate. Whereas Pdx1 employs an active-site lysine residue for formation of the Schiff base (3a), PdxJ uses its substrate 3-amino-1-hydroxyacetone phosphate (b) for this purpose. An identical elimination/addition mechanism has been postulated for Pdx1 and PdxJ for the subsequent elimination of inorganic phosphate (5 and f respectively in Scheme 2) [79,89]. In addition, both enzymes form the covalent C-4–C-5 bond in an aldol-type condensation reaction by combining C5 and C3 building blocks (compare 11a with h in Scheme 2). However, differences exist in the proposed timing of the formation of the N-1–C-6 bond, which is initiated by nucleophilic attack of the primary amine of either 15a or 3-amino-1-hydroxyacetone phosphate (b) on the carbonyl of 15a or DXP (a) respectively (Scheme 2). Whereas this step is proposed to occur in an early phase in the DXP-dependent pathway [36,37], a later time point has been suggested for the Pdx1 mechanism [59,79].

The structural characterization of ligand-complexed PdxJ permitted the potential assignment of residues to some of these roles [85]. First, the pyridine ring of bound PNP was observed to be sandwiched between Glu72 and His193. The positioning of the interaction of the γ-carboxy group of Glu72 with the C-2 carbonyl of DXP would be suitable to catalyse its protonation (Figure 2B). This would enable the nucleophilic attack of the primary amine of 3-amino-1-hydroxyacetone phosphate (b) yielding the Schiff base (c), in addition to removal of the C-3 proton from the DXP:3-amino-1-hydroxyacetone phosphate imine (c). Alternatively, Asp11, which is located in close proximity to the phosphate group of DXP, could act as the base to remove the C-3 proton of (c) (Figure 2B). The protonated imidazolium form of His193 in PdxJ is positioned on the opposite face of bound PNP/DXP and could serve as the Lewis acid to promote elimination of the C-4 hydroxy group (d). The γ-carboxy group of Glu153 in PdxJ is also of importance as it stays in close contact with the aldimine to ensure stabilization of the positively charged nitrogen of the pyridine ring moiety. Unfortunately, the proposed participation of these residues has never been corroborated by biochemical data on mutant PdxJ proteins. In Pdx1, on the other hand, with the exception of the lysine residues discussed above, the definitive assignment of catalytic residues is more difficult because of the unavailability as yet of appropriate ligand-bound structures. However, Asp25 has been observed to interact with the C-3 hydroxy group of Ru5P in the recently determined T. maritima structure (Figure 2C) and has been proposed to play a proton shuffling role [75,80], probably corresponding to that of Glu72 in PdxJ. Asp103 and Arg148 have been suggested to be involved in a late dehydration step after condensation of the C-5 and C-3 carbon scaffolds. The involvement of these residues has been substantiated ([77], and T. Raschle and T.B. Fitzpatrick, unpublished work), but the unambiguous designation of amino acids requires further efforts to solve the structures of additional substrate/product–enzyme complexes in combination with biochemical studies.


The occurrence of two distinct and mutually exclusive pathways for the de novo biosynthesis of vitamin B6 poses an attractive challenge regarding the rationale for the evolution of two independent pathways for the same molecule. The presence of a complete DXP-dependent pathway is, however, restricted to the γ-divison of proteobacteria [90], with the exception of three species of the family Pasteurellaceae, which carry Pdx1/Pdx2 homologues [45]. Interestingly, putative PdxA orthologues have been found in the genomes of the Gram-positive bacteria Bacillus halodurans and G. stearothermophilus [45,91], which also have Pdx1/Pdx2 homologues. It has been suggested that, because the γ-divison of proteobacteria is thought to represent the most recent lineage of prokaryotic evolution [92], the genes of the ancestral Pdx1/Pdx2 pathway were lost during evolution of the proteobacteria and the PdxA/PdxJ pathway evolved in this lineage [45]. The reason for the loss of the Pdx1 and Pdx2 genes is unknown. If the conditions of these organisms during the evolutionary process were such that the vitamins could be taken up from the environment (perhaps even by diffusion), then the Pdx1/Pdx2 genes might have been lost without deleterious consequences. Emergence to an environment again necessitating de novo synthesis of the vitamin resulted in the recruitment of available enzymes (some of which are involved in other pathways), permitting survival.

In this context, it must be mentioned that the different B6 vitamers can be interconverted by the so-called salvage pathway (Scheme 1). In this pathway, pyridoxine, pyridoxal and pyridoxamine can be reversibly phosphorylated and dephosphorylated through the action of specific kinases (PdxK and PdxY) [93,94] and apparently unspecific phosphatases respectively. However, a specific phosphatase has been reported in humans [95,96]. Furthermore, PNP or PMP can be converted into PLP by the action of the oxidase PdxH [30]. Note that the latter enzyme is included in the de novo pathway of vitamin B6 biosynthesis by the DXP-dependent route. In addition, a pyridoxal reductase has been identified in yeast [97,98]. It is assumed that the majority of organisms have one or the other of these enzymes, but the complete pathway has been studied extensively only in E. coli [25,30,93,94] and humans [99102]. As the conversion of 3-hydroxy-4-phosphohydroxy-α-oxobutyrate into 4-phosphohydroxy-L-threonine during de novo vitamin B6 biosynthesis in E. coli by the transaminase PdxF requires a PLP-dependent enzyme, the salvage pathway (and perhaps also the DXP-independent pathway) of vitamin B6 biosynthesis must have evolved before its own de novo biosynthesis pathway. This would lend support to the notion that there may have been a time during the evolution of the γ-division of proteobacteria where uptake of the vitamin from the environment was selected for and the Pdx1 and Pdx2 genes became dispensable. Later, it became necessary to produce the vitamin de novo leading to ‘reinvention’ of the synthesis of the pyridine ring.


Clearly, the Pdx1/Pdx2 route represents a much more direct way for the formation of vitamin B6 than the PdxA/PdxJ route. The remarkable complexity of vitamin B6 biosynthesis in E. coli, mainly manifested in the synthesis of 3-hydroxy-1-aminoacetone phosphate, has already been pointed out [103]. The reason for this complexity remains elusive. Surprisingly, a comparative analysis of the key enzymes of the two pathways, i.e. Pdx1 and PdxJ, has revealed that they are in fact highly similar in structural and mechanistic terms. It is perhaps noteworthy in this context that Pdx1 from C. nicotianae has been shown to functionally complement an E. coli pdxJ mutant, although the chemical nature of this complementation is not clear [104]. The utilization of the universal TIM-barrel fold as a scaffold may provide at least a partial answer for the apparent convergent evolution of both Pdx1 and PdxJ. Stable folds such as the TIM barrel and structural features of enzyme active sites have been frequently reused in evolution and adapted for new catalytic purposes [105]. In the case of Pdx1 and PdxJ, this statement is supported by the employment of a common phosphate-binding site (P1 and PA respectively). For the foreseeable future, the mechanism of Pdx1 still poses many challenges with regard to the function of the P2 site, the location of the triose-binding site and identification of reaction intermediates to validate mechanistic proposals, in addition to the nature of the transfer of ammonia between glutaminase and synthase active sites.

In any case, the discovery of the DXP-independent pathway of biosynthesis has led to examination of vitamin B6 in an entirely new light. It is now clear that its functions go beyond that of a cofactor, and this opens up entirely new and exciting areas of research with regard to its role in stress responses in particular. Moreover, the absence of the pathway from animals means that vitamin B6 biosynthesis may provide a novel drug target, and can now be examined with fervour in organisms of interest. The progress on the elucidation of the biosynthesis of the vitamin has been remarkable over the last decade. New questions now pertain to the transport of the vitamin, which is necessary to complete its fundamental characterization, to the definition of its roles as an antioxidant and in the modulation of hormone and transcription factor function.


The support of the Swiss National Science Foundation Grant 3100A0-107975/1 (to N. A. and T. B. F.), and the European Commission Grant VITBIOMAL-012158 is gratefully acknowledged. N. A., T. B. F. and T. R. extend special thanks to Professor Duilio Arigoni (ETH Zurich) for his input and numerous stimulating discussions with regard to the biosynthesis of this molecule.

Abbreviations: DXP, deoxyxylulose 5-phosphate; GAP, glyceraldehyde 3-phosphate; PLP, pyridoxal 5′-phosphate; PMP, pyridoxamine 5′-phosphate; PNP, pyridoxine 5′-phosphate; R5P, ribose 5-phosphate; Ru5P, ribulose 5-phosphate


View Abstract