The rapid progress in structural and molecular biology over the past fifteen years has allowed chemists to access the structures of enzymes, of their complexes and of mutants. This wealth of structural information has led to a surge in the interest in enzymes as elegant chemical catalysts. Enzymology is a distinguished field and has been making vital contributions to medicine and basic science long before structural biology. This review for the Colworth Medal Lecture discusses work from the author's laboratory. This work has been carried out in collaboration with many other laboratories. The work has mapped out the chemical mechanisms and structures of interesting novel enzymes. The review tries to highlight the interesting chemical aspects of the mechanisms involved and how structural analysis has provided a detailed insight. The review focuses on carbohydrate-processing pathways in bacteria, and includes some recent data on an integral membrane protein.
The importance of deoxy sugars
Carbohydrates remain a challenge for the synthetic chemist and the biochemist. In comparison with peptides, carbohydrates are intensely functionalized and stereochemically rich. Even a simple glucose molecule has four asymmetric centres (five in the hemiacetal form), five potential nucleophiles and an electrophilic carbon. The challenge for the chemist has been to isolate the reactivity of one hydroxy group from the other four. Nature has utilized carbohydrates extensively, not only in central metabolism, but as recognition molecules. Carbohydrates can polymerize with almost exponential diversity, which is in sharp contrast with amino acids, which can only be linked by the peptide bond. Varying only the linkage between two D-glucose molecules, one can make 20 different compounds. It is no surprise that carbohydrates are displayed on the surface of almost all cells in all domains of life. Bacteria are particularly adept at synthesizing and utilizing deoxy (hydroxyl-missing) sugars.
Deoxy sugars are found in lipopolysaccharides, glycoproteins, glycolipids on bacterial cell surfaces and as part of many secondary metabolites . As many are unique to bacteria, they have long been of interest as drug targets. dTDP–L-rhamnose is synthesized from glucose 1-phosphate and dTTP via a biosynthetic pathway utilizing four distinct enzymes: α-D-glucose-1-phosphate thymidylyltransferase (RmlA; EC 22.214.171.124), dTDP-D-glucose 4,6-dehydratase (RmlB; EC 126.96.36.199), dTDP-6-deoxy-D-xylo-4 hexulose 3,5-epimerase (RmlC; EC 188.8.131.52) and dTDP-6-deoxy-L-lyxo-4-hexulose-4-reductase (RmlD; EC 184.108.40.206).
L-Rhamnose is found as a common component of the bacteria cell wall, and evidence points to its essential nature. Deletion or disruption of the L-rhamnose biosynthetic pathway in Pseudomonas aeruginosa is effectively lethal. Analysis of Vibrio cholerae mutants with a deletion in the rmlB or rmlD genes have a severe colonization defect . Deletion of any one of the four rml genes in Streptococcus mutans inhibits cell wall polysaccharide synthesis, rendering the bacteria avirulent . In the uropathogenic strain Escherichia coli 075:K5, lack of a functional RmlD enzyme leads to loss of serum resistance . Disruption of the pathway in Enterococcus faecalis attenuates the pathogen . L-Rhamnose has been found to occupy important anchoring positions, e.g. in Mycobacterium tuberculosis, where it covalently links the arabinogalactan to the peptidoglycan layer . It has been demonstrated that the loss of arabinogalactan leads to loss of viability of this important pathogen [7–9]. In a recent study, inhibitors of L-rhamnose-synthesizing enzymes were shown to possess activity against whole M. tuberculosis cells . The fact that the L-rhamnose biosynthetic pathway is not found in humans makes all four enzymes potential targets for antibacterial agents. The structures of all four enzymes have been studied in our laboratory [10–15] (Figure 1).
The dTDP–L-rhamnose pathway
L-Epivancosamine is required to be attached to the vancomycin class antibiotic chloreomomycycin for full activity. On a commercial scale these complex glycopeptide antibiotics can only be produced by fermentation, and therefore rational redesign of the compounds requires a detailed understanding of their biosynthesis. Vancosamine (attached to vancomycin) and epivancosamine differ only in the chirality at the C4′ position (axial in vancosamine, equatorial in epivancosamine). It is therefore likely that both sugars are synthesized in the same manner, differing only in their final reductive step (Figure 2). dTDP–L-epivancosamine is synthesized from dTDP-6-deoxy-D-xylo-4 hexulose (product of RmlB) and five further steps (Figure 2) [16,17].
The dTDP–L-epivancosamine pathway
RmlB and RmlD: insights into hydride transfer
RmlB and RmlD are members of the SDR (short chain dehydrogenase) superfamily, which is one of Nature's most common enzyme families . This family of enzymes binds NAD(P), and is defined by the presence of a catalytic triad consisting of threonine, lysine and tyrosine. The chemistry these enzymes carry out is varied, but a key step to all of their reactions is the transfer of hydride and a proton. Some carry out an oxidation, others a reduction and some both oxidize and reduce during a complete cycle. Hydride transfer occurs concomitantly with proton transfer. The tyrosine residue, influenced by the lysine residue and the cofactor, has a pKa of ≈7. This allows it to function as the ultimate acid or base for proton transfer . The C6-deoxygenation catalysed by RmlB and other NDP-hexose 4,6-dehydratases is the first committed step in almost all deoxyhexose biosynthetic pathways. For this reason, the mechanism of RmlB has been the subject of extensive investigation [20–23]. Catalysis proceeds through three distinct chemical steps . The substrate dTDP–D-glucose is initially oxidized at glucosyl-C4′ by the tightly bound NAD+ coenzyme. This is followed by the concerted elimination of water from glucosyl-C5′ and C6′ to form the 4 keto-5,6-glucosene intermediate. The existence of the dTDP–4-keto-5,6-glucosene intermediate was confirmed by rapid mix-quench mass spectroscopy [20,23]. The final step is the reduction of this intermediate by NADH at glucose C6′, giving both the product dTDP–4-keto-6-deoxy-D-glucose and the regenerated NAD+ coenzyme (Figure 3). RmlD is a much simpler enzyme: it reduces the final keto sugar by hydride transfer (Figure 1).
The RmlB mechanism
We wished to identify what residues were responsible for each step, particularly the hydride transfer. Previous work, mainly on the SDR enzyme UDP–galactose glucose epimerase, had suggested a complex proton shuttle mechanism for the oxidation reaction [25–27]. This was necessitated by the fact that the tyrosine residue was too far from the substrate. We determined the structure of RmlB  from two different sources, Salmonella enterica and Streptococcus suis. The proteins are homologous, and consist of about 350 residues. We were able to obtain complexes of these proteins with dTDP, dTDP–D-glucose and dTDP–D-xylose , providing a reassuring consistency in our interpretation. We determined the structure of RmlD with dTDP–L-rhamnose bound . Comparison of native structures with the complexes show that the binding of the thymidine nucleotide triggers a rearrangement at the active site. These changes result in repositioning of the key residues (threonine, lysine and tyrosine) within the active site.
Transfer of hydride to NAD+ (RmlB) and from NAD+ (RmlD)
The structures indicate that, in contrast with studies on UDP–galactose epimerase [25–27], proton transfer is directly to the tyrosine rather than through a bridging threonine residue. The complexes that we have obtained appear to be very convincing. Not only do they show a direct proton transfer route, but they also have a good orbital alignment between the bonding orbital of the proton on the substrate and the anti-bonding molecular orbital of the NAD+ aromatic ring. From chemical first principles, such an alignment would be required for hydride transfer. The short hydrogen bond of approx. 2.5 Å (1 Å=0.1 nm) between tyrosine and the substrate would indicate that a low-barrier hydrogen bond  is formed between these groups in the transition state of the general SDR enzyme mechanism. In a low-barrier hydrogen bond, the two participating centres share a proton and little energy is required to shift the proton between groups [29,30]. Removing the high-energy barrier that normally prevents the protonation of a carbonyl group (pKa value approx. 7), or the deprotonation of a hydroxy group (pKa value approx. 16) would greatly accelerate the mechanism. We have suggested that the threonine residue fine-tunes the pKa of the substrate transition state to achieve an optimal match with tyrosine.
dTDP–xylose reacts with RmlB in a stoichiometric reaction that produces dTDP–keto-xylose and NADH. In our 1.8 Å dTDP–xylose RmlB structure, we observed that electron density around NAD was consistent with a buckled NADH, which at this point had not previously been observed. A proper theoretical study of hydride transfer in SDR enzymes requires the conformation of NADH to be accurately known. A simplistic view is that NADH should not be flat, C4 having become a four-coordinate sp3 carbon. The situation for nitrogen is much less clear: if it remained in the plane, it would be part of the delocalized molecular orbital. However, if this is not the case, then the degree by which nitrogen is out of the plane would significantly perturb the molecular orbital of NADH. Since this molecular orbital becomes the HOMO (highest occupied molecular orbital) of NAD+ after hydride loss, its status is crucial in understanding the energetics of hydride transfer. We optimized our crystallization conditions to grow high-resolution crystals, and used microspectrophotometric analysis to ensure that NAD+ had been completely reduced to NADH. At that time, another report appeared on a high-resolution study of LADH (L-alcohol dehydrogenase) . This confirmed that C4 of the NADH ring had moved out of the plane, but showed that N1 remained in the plane. Modelling studies showed that in this case a hydroxide ion (essential for the alcohol dehydrogenase mechanism, but not involved in SDR catalysis) was crucial in controlling the conforming of NADH . We determined the RmlB dTDP–xylose complex structure to 1.5 Å, and the electron density clearly shows a buckled NADH . In our structures, both the C4 and N1 atoms of NADH have moved significantly out of the plane of the ring. The ring we observe is considerably more puckered than that seen for LADH. In direct contrast, we see that the N1 atom has become significantly pyramidal (sp3 hybridized) and has moved out of the plane (Figure 4). This distortion serves to make hydride a more potent reductant than it would be if the ring remained flat or the nitrogen remained in the plane of the ring. To assess if this is a general feature of SDR enzymes, we constructed simpler systems and performed a detailed calculation to see what factors favoured a flat ring versus a distorted ring. The calculations show that the internal hydrogen bond between the backbone phosphate and the C7 amide of NAD is the principal driving force for ring shape. A simple model reproduces almost all of the observed distortion of the ring, and a model without this hydrogen bond produces effectively a flat ring. The remainder of the observed distortion in the ring is caused by a polar contact between N1 and the catalytic tyrosine residue (Figure 4). The internal hydrogen bond between the amide and the phosphate in NADH is a feature of the way NAD(P) is bound by all SDR enzymes. This does suggest that this is indeed a general observation, and the structure of NAD(P)H will be buckled in all SDR enzymes with a concomitant increase in the enzymes reducing potential .
NADH in the SDR enzyme superfamily adopts a buckled conformation which activates hydride transfer
Dehydration step in RmlB
The keto sugar created by the oxidation step serves to acidify protons located α to the carbonyl. The proton on C5′ of dTDP–D-glucose is less than a distance of 3 Å from a conserved glutamate residue known from other studies to be essential [20,21,33]. This glutamate is therefore perfectly positioned to abstract the proton from C5′, with the resulting negative charge stabilized as an enolate. For the C6′–O6′ bond to cleave and water to be eliminated, the alcohol group must be protonated. The Oδ1 of a conserved aspartate (adjacent in sequence to the conserved glutamate) is 2.7 Å away from the hydroxy group of the glucose. A negatively charged aspartate would partly remove the hydroxyl proton, preventing C6′–O6′ cleavage. Logically, the aspartate must in fact not only be protonated, but transfers a proton to O6′, allowing the oxygen to be eliminated as water. Confirmation of this comes from a study of the substrate analogue dTDP–6-fluoro-6-deoxyglucose with both native and an Asp→Ala mutant enzyme . The analogue eliminates a fluoride cation from the enolate, and a proton is not required. The analogue turns over equally well in the native and mutant enzyme , whereas dTDP–D-glucose reacts much more slowly with the mutant .
Reduction of the C–C double bond by RmlB
The argument for hydride transfer should hold true whether it is a C–C or a C–O double bond which is being reduced. Applying this logic to the complex structures we had obtained presented a problem. The hydride is known to be transferred to C6′ of the glucosene, but this atom in our complexes is not in the correct position for facile hydride transfer. Although dTDP–glucose and dTDP–xylose are not perfect mimics for the glucosene, even correcting for difference in bond lengths, C6′ remains in the wrong position to accept a hydride from NADH. In the dTDP–xylose (which lacks C6′) complex, a water molecule is indeed found at the same position as O6′, bound to the protein. Formerly the oxygen and the carbon would be covalently bonded (distance 1.4 Å), but once the bond is broken they will repel each other until a separation of about 3.0 Å is established. As the water cannot move, it seems likely that the carbohydrate moves to relieve the van der Waals repulsion. A rotation about the Pβ-O bond connecting dTDP with the anomeric oxygen atom relieves the repulsion, and places the C6′ atom in the correct position to accept a hydride. A model of the rotated carbohydrate reveals no clashes. This rotated position allows the tyrosine residue to transfer a proton to the carbon, completing the reduction of the double bond (Figure 3).
Epimerization by RmlC and EvaD
Nature employs a variety of chemical reactions to epimerize the stereogenic carbon centres in carbohydrate skeletons . A common method is to activate a proton for removal by placing a keto group α to the C-H bond. The result is to lower substantially the pKa of the proton. Despite this, the pKa is still beyond the reach of any base found in biology. Nature has therefore had to evolve a means to stabilize the negatively charged enolate and thus further reduce the pKa. RmlC, the third enzyme in the rhamnose pathway, was first identified over 30 years ago , and is the archetypical carbohydrate epimerase. RmlC works by abstracting the acidified proton from one face of the molecule and replacing it on the other. Epimerization of a single position requires a base and an acid, for a double epimerization two bases and two acids are required (Figure 1). Initial structural studies confirmed that RmlC has a novel structure, and does indeed represent a new enzyme class. However, these studies could not identify all the key catalytic residues; the nucleotide-binding site was located, but not the crucial carbohydrate-binding site. We have made extensive efforts to bind D-configured sugars to a group of highly conserved RmlCs (S. enterica, P. aeruginosa and M. tuberculosis). All the complexes we obtained have had dTDP bound, but the carbohydrate was either disordered or cleaved. RmlC from S. suis is significantly different in sequence from other RmlCs (identity 25%), yet it does have normal RmlC activity . With this protein we were able to obtain complexes with the substrate analogues dTDP–glucose and dTDP–xylose in which the carbohydrate moieties were visible. The sugar rings are located in the entrance to the β sandwich of RmlC, and are almost completely enveloped by protein. There is a subtle difference between the orientations of the two rings: xylose is offset by 20° compared with glucose. Looking at the carbohydrates, the O2 atom of the sugar is part of a hydrogen-bond network involving two water molecules. The O3 atom is hydrogen-bonded to the NZ atom of the absolutely conserved lysine. A glutamic acid (found only in Streptococci RmlCs) makes a bidentate hydrogen bond to O2 and O3. This type of interaction is commonly observed in protein carbohydrate complexes, and correlates with binding affinity . The O4 atom, where the enolate is expected to form hydrogen bonds to a non-conserved asparagine residue, is within 4 Å of an absolutely conserved lysine. dTDP–D-glucose contains an additional hydroxy group relative to the true substrate; this hydrogen-bonds to a conserved serine in the active site.
The structures identify that a single, absolutely conserved histidine residue is positioned below the plane of the ring, and is in the correct location to abstract a proton from C3′ or C5′. This histidine residue is hydrogen-bonded to an absolutely conserved aspartate residue, consistent with its role as the base. There is no other conserved residue in an appropriate position, or even within 6 Å, suggesting that histidine alone acts as the base for both proton abstractions. The resulting enolate would be stabilized by the absolutely conserved lysine residue. The identification of the acid is less obvious: an absolutely conserved tyrosine residue sits above the C5 position of the ring and on the opposite face to the conserved histidine . This would suggest that tyrosine supplies the proton to this position; whether or not it supplies the proton to C3′ is less clear (this would require tyrosine to be reprotonated during turnover). It is interesting to note that dCDP–6-deoxy-D-glucose epimerase, which is almost indistinguishable from the RmlC family in sequence terms, catalyses only the C3′ epimerization . This enzyme lacks this tyrosine residue and has phenylalanine instead . In the RmlC structures, there is a water molecule close to the C3 position that could also act as the acid. Mutation of the conserved tyrosine, lysine or histidine essentially inactivates the enzyme . On close inspection, although these complexes are illuminating, there are some details which remain unresolved. The S. suis complexes are not ideal mimics for the true substrate, as neither are keto sugars, and one is bigger and one is smaller than the substrate. Furthermore, a non-conserved glutamate residue found in Streptococci RmlCs makes an important binding contact.
We investigated the role of tyrosine more fully with a study on EvaD. This protein is required to epimerize only the C5′ position of dTDP–3-amino-2,3,6-trideoxy-3C-methyl-D-erythro-hexopyranosyl-4-ulose. This is a rather different substrate from RmlC (dTDP–6-deoxy-4-keto-D-glucose) in that the C3 position has an amino group. In addition, if EvaD was in fact a competent RmlC enzyme, then it would compete for the substrate of EvaA. Whereas kinetic analysis showed that EvaD possessed a much lower ability to catalyse a double epimerization, the structure of EvaD seemed identical with RmlC in almost every aspect. As with other RmlCs we were only able to obtain a dTDP complex. Based on this complex and structures from S. suis, we constructed models of the substrate in EvaD. Careful examination revealed that the conserved tyrosine has a different conformation in EvaD than in all RmlCs, and would clash with our modelled substrates. Interestingly, this conformation is determined by other residues at the active site, which themselves seem unimportant but lock the tyrosine in this conformation. We predicted that the true substrate is rotated in the active site, relative to dTDP–glucose/dTDP–xylose, to correctly position C5 relative to the tyrosine. In order to make EvaD more RmlC-like, we reasoned that if we could force tyrosine to adopt the RmlC conformation by mutating an adjacent methionine residue to phenylalanine. The native and mutated enzymes were characterized by kinetic analysis, and by monitoring their ability to exchange deuterium at the C3′ and C5′ positions. The mutant enzyme is more active under both criteria than native EvaD, and is thus more like RmlC. The deuterium-labelling studies indicate that C5′ epimerization occurs first and is followed by C3′ (A.B. Merke and J.H. Naismith, unpublished work).
More recently, we have investigated complexes of RmlC with dDTP–L-rhamnose (which is a very good analogue of the product); in contrast with the D-configured sugars, we have been able to obtain identical dTDP–L-rhamnose complexes with S. suis, M. tuberculosis and P. aeruginosa. A major difference between dTDP–L-rhamnose and dTDP–D-glucose/xylose is that the nucleotide is equatorial to the ring rather than axial. The consequence is that the carbohydrate ring sits in a different orientation in the active site than was seen for dTDP–glucose or dTDP–xylose. The dTDP–L-rhamnose is positioned in the active site such that O4 is now directly in hydrogen-bonding contact with the lysine. In addition, the ring sits above the conserved histidine, the conserved tyrosine above the C5′ atom and a water above C3′. Although these complexes do not change our overall interpretation of the mechanism, the orientation of dTDP–L-rhamnose at the active site is more consistent with the positioning of the key acid and base residues. This has raised the possibility that the substrate is not the normal ‘chair’ form of dTDP–6-deoxy-4-keto-D-glucose, but rather a ‘ring-flipped’ form, which would explain the difficulty in obtaining dTDP–D-glucose complexes. In other words, it is possible that the ring flip shown in Figure 1 occurs before and not after catalysis. However, even this cannot be entirely correct: in a completely ring-flipped form, the C3′ and C5′ protons would be equatorial to the carbonyl group. In this position, their pKa values would be too high to be abstracted. An alternative mechanism would therefore involve a ‘half-chair’ conformation of substrate (Figure 5).
The RmlC mechanism
Carbohydrate export through the membrane
Bacteria have polymers of carbohydrates on their cell surface or capsules where they are important virulence determinants. Capsules enable pathogenic bacteria to establish infection and may be non-, or only weakly, immunogenic . The capsular K antigens of E. coli are classified into four categories (denoted group 1 to group 4) . K antigens are synthesized at the inner membrane and need to be translocated to the cell surface, passing through the bacterial outer membrane. The oligosaccharide repeat units of the group 1 K30 antigen from E. coli (O9a:K30) are assembled on an undecaprenylphosphate carrier by a series of glycosyltransferases . The resulting repeat-unit polymers are transferred across the inner membrane by an unknown process involving the protein Wzx, before a final polymerization at the periplasmic face of the inner membrane by Wzy. These proteins are the characteristic components of a Wzy-dependent assembly pathway . The final polymerization of the repeat-unit oligomers requires phosphorylation of an inner membrane tyrosine autokinase, Wzc . Wzc is dephosphorylated by Wzb, which is also required for capsule assembly [44,45]. The Wzi protein is located in the outer membrane and acts late in the assembly process, although its exact role is unknown . The key protein is Wza, a lipoprotein that forms a multimeric outer-membrane complex that allows the translocation of the carbohydrate polymer across the outer membrane [42,46]. Single-particle analysis of Wza has identified a putative C8 rotational symmetry, suggesting that the complex is octameric . Negatively stained two-dimensional crystals revealed ring-like multimers with an average outer diameter of ≈9–10 nm and a central stain-excluding region of ≈2–3 nm , consistent with an octameric arrangement. We have crystallized the Wza protein , and in the asymmetric unit there are eight monomers.
Using negative cryostaining electron microscopy and three-dimensional reconstruction, we have determined a 16 Å structure of Wza (Figure 6) . The protein appears to be made from two stacked rings. One ring is larger than the other, and the overall effect is a molecule shaped like a mushroom. We suggested that the larger ring of the Wza complex is embedded in the membrane. This is because the thickness of the large ring matches the thickness of the membrane. In addition, there are large holes in the smaller ring which would seem incompatible with it being membrane-embedded. The Wza multimer is closed to the external environment at both ends, but access to the complex cavity can be achieved via the ‘side’ holes from the periplasm. This may be the entry point for the capsular polysaccharide, with the exit point being in the centre of the large ring. The centre of both large and small rings are closed, and it seems likely this is the closed form of Wza. For the 100 kDa substrate to exit Wza, the ring embedded in the membrane would have to open up in some way. Experimental data have suggested some flexibility in the larger (but not smaller) ring, consistent with our hypothesis that this is embedded in the membrane. The rings seem to have 4-fold rotational symmetry, rather than 8-fold, as suspected previously.
Two orthogonal views of the electron microscopic structure of Wza, the carbohydrate-exporting enzyme found in the outer membrane of bacteria
Enzymes have fascinated scientists for many years. Structural biology now plays a key role in modern investigations into enzyme mechanism. It provides the three-dimensional information on how catalysis is accomplished, which is of interest to chemists. Work in my laboratory has focused on understanding the enzymic transformations used by bacteria in sugar biosynthesis. The work is not only of basic science interest, but provides the rational platform on which to design novel inhibitors or contemplate protein engineering.
Colworth Medal Lecture Delivered at the SECC, Glasgow, on 21 July 2004 James Naismith
I wish to express my sincere thanks to all my co-workers, not just those whose work is presented here. I also wish to thank my long-term collaborator Professor Chris Whitfield, University of Guelph, Canada. I am a Biotechnology and Biological Sciences Research Council (BBSRC) Career Development Fellow. The work has been supported financially by the University of St Andrews, The Wellcome Trust and the BBSRC.