Phosphorylases are key enzymes of carbohydrate metabolism. Structural studies have provided explanations for almost all features of control and substrate recognition of phosphorylase but one question remains unanswered. How does phosphorylase recognize and cleave an oligosaccharide substrate? To answer this question we turned to the Escherichia coli maltodextrin phosphorylase (MalP), a non‐regulatory phosphorylase that shares similar kinetic and catalytic properties with the mammalian glycogen phosphorylase. The crystal structures of three MalP–oligosaccharide complexes are reported: the binary complex of MalP with the natural substrate, maltopentaose (G5); the binary complex with the thio‐oligosaccharide, 4‐S‐α‐D‐glucopyranosyl‐4‐thiomaltotetraose (GSG4), both at 2.9 Å resolution; and the 2.1 Å resolution ternary complex of MalP with thio‐oligosaccharide and phosphate (GSG4‐P). The results show a pentasaccharide bound across the catalytic site of MalP with sugars occupying sub‐sites −1 to +4. Binding of GSG4 is identical to the natural pentasaccharide, indicating that the inactive thio compound is a close mimic of the natural substrate. The ternary MalP–GSG4‐P complex shows the phosphate group poised to attack the glycosidic bond and promote phosphorolysis. In all three complexes the pentasaccharide exhibits an altered conformation across sub‐sites −1 and +1, the site of catalysis, from the preferred conformation for α(1–4)‐linked glucosyl polymers.
Structural, biochemical and time‐resolved studies on rabbit muscle glycogen phosphorylase have contributed to an understanding of the catalytic mechanism whereby the enzyme promotes the phosphorolysis of an α(1–4) glycosidic bond in glycogen or oligosaccharide substrates. The mechanisms of control by phosphorylation and by allosteric effectors are also understood from crystallographic studies on phosphorylated and unphosphorylated forms of the mammalian phosphorylase (Barford and Johnson, 1989; Barford et al., 1991; Sprang et al., 1991; Johnson, 1992). One question has remained outstanding. This relates to the recognition and conformation of the oligosaccharide component of the substrate when it is bound across the catalytic site. Numerous attempts to observe binding of oligosaccharide substrates with rabbit muscle glycogen phosphorylase (GP) in the active R state were not successful, probably because of the low affinity of the enzyme for linear oligosaccharides. In order to solve this problem we have turned to the Escherichia coli maltodextrin phosphorylase (MalP). As suggested by the nomenclature, GP and MalP exhibit distinct differences in their preference for substrates. GP has a high affinity for glycogen and branched polysaccharides and low affinity for small oligosaccharides (Hu and Gold, 1975; Kasvinsky et al., 1978). MalP has low affinity for glycogen but exhibits ∼100‐fold higher affinity for linear oligosaccharides than GP (Schwartz and Hofnung, 1967; Drueckes et al., 1996). This observation has been exploited to solve the major outstanding problem in phosphorylase catalysis, namely the recognition of the oligosaccharide component of the substrate at the catalytic site.
MalP (EC.18.104.22.168) is a 796 amino acid, non‐allosteric, dimeric enzyme that is 46% identical in sequence to GP (Palm et al., 1985, 1987). Unlike GP, MalP is not regulated by phosphorylation or by allosteric effectors but is regulated by control of gene expression (Raibaud and Schwartz, 1984). Phosphorylases are highly conserved from bacterial to mammalian enzymes, but share no homology with other carbohydrate degrading enzymes, with the exception of the noted similarity of the core structure around the cofactor pyridoxal phosphate binding site in GP with a DNA glucosyltransferase modifying enzyme (Artymuick et al., 1995; Holm and Sander, 1995). Phosphorylase isozymes exhibit nearly 100% identity in residues at the catalytic site (Watson et al., 1997) and all utilize the 5′‐phosphate of the essential cofactor pyridoxal phosphate (PLP) in catalysis (Palm et al., 1990). The bacterial and mammalian enzymes catalyse the phosphorolytic cleavage from the non‐reducing end of α‐1–4 linked maltodextrins (Glc)n to yield α‐d‐glucose‐1‐P (Glc‐1‐P) (Figure 1). The reaction proceeds with retention of configuration. For MalP, the optimal substrate maltodextrin has a minimum length of five glucosyl units for phosphorolysis and four glucosyl units for the reverse reaction of oligosaccharide synthesis (Becker et al., 1994).
Structure determinations of MalP (Watson et al., 1997) and GP have shown that the catalytic site is buried at the centre of the large subunit. It is accessible to the solvent through a 20 Å long channel that has been proposed as the substrate oligosaccharide binding site. Sub‐sites for oligosaccharide recognition are numbered −1 to +4 from the non‐reducing end to the reducing end, and the site of phosphorolysis is between sub‐sites −1 and +1 (Davies et al., 1995). In both the less active T state and the active R state conformations of GP, the catalytic site (sub‐site −1) has been defined through binding studies and observations of catalysis in the crystal with monomeric substrates and products such as Glc‐1‐P and heptulose‐2‐P, and by binding studies with inhibitors such as glucose, glucose analogues and transition state analogues (Hajdu et al., 1987; Johnson et al., 1989, 1990, 1992; Martin et al., 1990; Duke et al., 1994; Mitchell et al., 1996). The site is situated at the end of the channel, far removed from the bulk solvent and adjacent to the essential cofactor, pyridoxal phosphate. Co‐crystallization studies with MalP and maltohexaose (O'Reilly et al., 1997) showed a disaccharide, maltose, bound at adjacent sites to the catalytic site (numbered sub‐sites +1 and +2). Analysis of the oligosaccharide mixture from the co‐crystallization experiments indicated that both catalysis and hydrolysis of the oligosaccharide had occurred over the period of crystallization.
α(1–4)‐O‐linked glucosyl oligosaccharides exhibit a preferred conformation in which there is an intramolecular hydrogen bond between the O2 hydroxyl and the O3 hydroxyl of contiguous residues, as observed in maltose (Takusagawa and Jacobson, 1978) and related compounds. Modelling studies from the MalP–maltose complex and from the GP–monosaccharide complexes indicated that an oligosaccharide bound across sub‐sites −1 and +1 could not be accommodated in the preferred conformation if the oligosaccharide were to follow the path of the catalytic channel (O'Reilly et al., 1997). Either there must be a significant conformational change in the enzyme structure, or a significant alteration of the conformation about the α(1,4) glycosidic bond between sub‐sites −1 and +1. It was therefore of interest to observe an oligosaccharide substrate bound across the phosphorylase catalytic site and to assess the significance of any conformational change for catalysis.
Acarbose is a pseudo‐tetrasaccharide inhibitor of α‐glucosidases that can adopt either the preferred conformation or an alternative conformation about the corresponding N‐linked glycosidic bond of its acarviosine moiety. Co‐crystallization of MalP with acarbose led to oligosaccharide binding in sub‐sites +1 to +4 and the identification of the further sub‐sites +3 and +4 (O'Reilly et al., 1999). However, in both the MalP–maltose and the MalP–acarbose binding experiments a glycerol molecule, used as a cryoprotectant for the low temperature data collection, blocked the crucial catalytic sub‐site −1. Thus the problem of observing an oligosaccharide bound across the phosphorolytic cleavage site remained.
In the present work, we developed a method of freezing crystals that avoided the use of glycerol. By carefully excluding phosphate from the purification and crystallization protocols we have been able to co‐crystallize the binary MalP complex with the natural substrate maltopentaose (G5). The X‐ray crystal structure of the MalP–G5 complex shows for the first time an oligosaccharide bound across the catalytic site in sub‐sites −1 to +4 spanning the point of enzymatic cleavage. In order to observe a ternary complex in the presence of phosphate and elucidate the catalytic mechanism, we have exploited the thio‐oligosaccharide, 4‐S‐α‐d‐glucopyranosyl‐4‐thiomaltotetraose (GSG4) which is resistant to enzymatic cleavage (Figure 1). Thio‐oligosaccharides (Driguez, 1997) have become useful tools for the study of carbohydrate–enzyme interactions (Sulzenbacher et al., 1996). Solution NMR studies (Bock et al., 1994) indicated that the thio analogues showed greater conformational flexibility than the corresponding O‐linked and N‐linked glucosyl polymers. Comparative studies with the thio and natural pentasaccharide binary complexes showed no significant differences in binding to MalP. Co‐crystallization of MalP with the thio‐oligosaccharide in the presence of phosphate has allowed us to determine the structure of a ternary enzyme substrate complex. The complex is consistent with previous site‐directed mutagenesis data for MalP and provides further understanding of the catalytic mechanism of phosphorylases.
The overall architecture of the native maltodextrin phosphorylase has been described, and details of the structure compared with the active and inactive forms of mammalian phosphorylase (Watson et al., 1997). In brief, MalP is an α/β protein consisting of two domains (N‐terminal domain 19–482; C‐terminal domain 483–829, using the rabbit muscle GP numbering) that form the monomeric subunit. The active site lies between these domains, at the centre of the molecule, and is accessible to the solvent via a channel ∼20 Å in length. There is a non‐crystallographic 2‐fold axis relating two subunits, giving rise to a dimer in the asymmetric unit. The overall quaternary and tertiary protein structures of the three oligosaccharide complexes reported here are similar to the native MalP structure. However there are significant local conformational changes from the native MalP structure in a loop of residues at the catalytic site upon binding the oligosaccharide.
The structures of the MalP–G5, MalP–GSG4 and MalP–GSG4‐P complexes
A summary of the data processing statistics and refinement parameters for each of the three MalP–oligosaccharide complex structures is given in Table I. Views of the final 2Fo−Fc weighted electron density maps for the three complexes are shown in Figure 2. In each of the complexes the five sugars are clearly defined in sub‐sites −1 to +4. Sub‐site −1 is the furthest buried in the catalytic site channel and is adjacent to the cofactor PLP. The oligosaccharide binds the length of the channel with sub‐site +4 most exposed to the bulk solvent.
Comparison of the isomorphous crystal structures of the MalP–G5 and MalP–GSG4 complexes at 2.9 Å resolution showed that the thio analogue bound in a similar manner to the natural pentasaccharide. The single crystal structure of thio‐maltoside (Perez and Vergelati, 1984) demonstrated that the thio analogue has similar structural parameters to those observed for the O‐linked maltose with an increase in the distance C1 to C4′ of 0.35 Å. In the MalP–G5 complex, the C1…C4′ distance is 2.4 Å, compared with 2.9 Å in the MalP–GSG4 complex showing a similar small increase. The binary structures are shown superimposed in Figure 4. The root mean square deviation (r.m.s.d.) in Cα coordinates of the protein atoms is 0.3 Å and the r.m.s.d. between the G5 and GSG4 oligosaccharide coordinates is 0.55 Å. The thio‐oligosaccharide (GSG4) is a good mimic of the natural substrate both in its conformation and its interactions with the enzyme.
Crystals of the ternary complex MalP–GSG4‐P were obtained under crystallization conditions similar to those for the binary complex of MalP–GSG4, with the addition of 5 mM phosphate to the protein drop. The crystals belong to the monoclinic space group P21, a space group that is different from the orthorhombic P212121 space group of the MalP–G5 and MalP–GSG4 complexes. The crystals diffracted to 2 Å resolution but there was some disorder in the direction parallel to the thin edge of the plate‐like crystal. Using MOSFLM we were able to extract data to 2.0 Å resolution by using a dynamic resolution cut‐off applied to individual frames. All the data were used in the refinement but electron density maps were calculated to 2.2 Å resolution, where the data are more complete (Table I). The five sugars were immediately apparent in difference maps obtained after refinement of the protein atoms and each sugar could be fitted unambiguously, since all sugar hydroxyl groups could be clearly located. After refinement of the protein and GSG4 atoms, the position of the phosphate group was apparent as the highest peak in the difference Fourier synthesis. The final refinement gave an overall B factor of 60 Å2 for the phosphate group. This is higher than expected for a group that makes several contacts to the protein and to the sugars at sub‐sites −1 and +1. The position of the phosphate ion is nearly identical (overall shift of 0.8 Å) to the position observed for phosphate in the native MalP structure (Watson et al., 1997) where the B factors were 34 Å2. A water molecule modelled at the phosphate site in the MalP–GSG4‐P complex did not refine well. It is concluded that the phosphate is present but less well ordered than the other atoms in the structure, possibly as a result of its proximity to the thio linkage of the oligosaccharide (discussed below). The overall structure of the MalP–GSG4‐P complex was very similar to the MalP native structure (r.m.s.d. in Cα coordinates 0.8 Å), where the major difference in structure results from a movement in the 380s loop (discussed below) and, to the MalP–GSG4 complex (r.m.s.d. in Cα coordinates 0.3 Å). The only difference between the MalP–GSG4 and MalP–GSG4‐P complexes is the presence of a water molecule in the former complex that is replaced by the phosphate in the latter MalP–GSG4‐P complex, and an accompanying 4.5 Å shift in the position of Arg569. The binding of GSG4 is identical in the binary and ternary complexes.
A summary of the intermolecular contacts for the MalP–GSG4‐P complex is given in Table II and shown in Figure 5. The sugar in sub‐site −1, buried close to the cofactor, has the largest number of hydrogen bonds and van der Waals contacts with the enzyme. These contacts are similar to those observed for monomeric glucosyl residues binding to the mammalian GP at sub‐site −1. Each of the peripheral hydroxyl groups is involved in at least two hydrogen bonds and there are numerous van der Waals contacts, especially to His377. The other sites involve fewer hydrogen bonds but almost all of these involve hydrogen bonds between the sugars and charged residues (Glu88 and Asp339 at subsite +1; Arg292 and His341 at sub‐site +2; Arg292 at sub‐site +3; and Glu382 and His571 at sub‐site +4). The charged–polar hydrogen bonds might be expected to contribute more binding energy than polar–polar hydrogen bonds (Fersht et al., 1985; Street et al., 1986). The van der Waals contacts at sub‐sites +2 and +4 are dominated by glucosyl–tyrosyl interactions (to Tyr280 and Tyr613, respectively) as previously noted (O'Reilly et al., 1997, 1999). The contacts from Tyr280 to the sugar involve the C2, C4, O5 and C6 atoms of sub‐site +2. The contacts from Tyr613 to the sugar in sub‐site +4 also involve the C2, C4, O5, C6 and O3 atoms of the sugar. Both aromatic groups stack with the A side of the glucopyranosyl ring as expected for α(1,4)‐linked glucosyl polymers (Johnson et al., 1988). In addition, there are contacts from Tyr613 to C1, O5 and O6 of the sugar in sub‐site +3. Hence the contacts from Tyr613 span the two sub‐sites +3 and +4. Finally, we note that sub‐site +3 appears extraordinarily wet. There are eight water molecules in van der Waals contact with the sugar, although only four of these are hydrogen bonded to the sugar.
Due to the high B factors associated with the phosphate group, we are less certain of the phosphate oxygen positions than for the carbohydrate atoms. Nevertheless, the dynamic refinement has resulted in stereochemically plausible positions that are consistent with the X‐ray evidence. The MalP–phosphate contacts are dominated by contacts to an arginine residue, Arg569, where two phosphate oxygens contact the NE and NH2 atoms of the guanidine side chain in a classic binding mode (Johnson and O'Reilly, 1996) (Table II; Figure 5). One of these oxygen atoms also contacts Lys574. A third oxygen is hydrogen bonded to the main chain nitrogen of Gly135, a residue from a glycine‐rich loop at the start of an α‐helix (Acharya et al., 1991). There is a direct hydrogen bond between one of the oxygens (O2) of the inorganic phosphate and an oxygen (OP1) of the cofactor phosphate and a further phosphate–phosphate contact through water. The cofactor 5′‐phosphate is localized by hydrogen bonds from two of its oxygens to two lysine residues (Lys568 and Lys574), while the third oxygen is hydrogen bonded to the main chain nitrogen of a residue at the start of an α‐helix. Each of the 5′‐phosphate oxygens is also hydrogen bonded to water molecules. This arrangement is very similar to that observed with phosphate bound to the native MalP structure and also to the mammalian enzyme. Finally, in the ternary MalP–GSG4‐P complex there are contacts from the inorganic phosphate oxygens (O1 and O4) to the thio‐glycosidic linkage [distances 2.9 and 2.8 Å to the S atom, respectively (Table II)]. In the two binary complexes of MalP with G5 and GSG4, where there is no phosphate present, Arg569 adopts an alternative conformation in which it is turned away from the substrate phosphate binding site. The shift in the arginine residue seen on binding phosphate provides additional support for the presence of the phosphate in the ternary complex.
Mutagenesis experiments supported
Extensive site‐directed mutagenesis experiments have been carried out with E.coli MalP in which the target residues were identified by predicting the likely binding site for an oligosaccharide based on the rabbit muscle GP structure (Drueckes et al., 1996; O'Reilly et al., 1997). The mutagenesis studies have yielded results that are fully consistent with the present direct experimental observations of an oligosaccharide binding to MalP. The mutations Glu88Ala (sub‐site +1), Tyr280Ala (sub‐site +1, +2), Asp339Ala (sub‐site +1), His341Ala (sub‐site +1, +2), Thr378Gly (sub‐site −1, +1), Glu382Ala (sub‐site +2, +3, +4) and His571Leu (sub‐site +3, +4) result in mutant enzymes that exhibit significantly reduced activity. Each of these residues is involved in direct contact with the oligosaccharide (Table II). The results indicate the importance of each of the interactions in locating and directing the oligosaccharide for correct binding for catalysis. The mutant Tyr613Phe showed only slightly changed kinetic parameters, a result that can be rationalized from the structure since phenylalanine at position 613 could perform similar stacking interactions with the sugars in sub‐sites +3 and +4. The triple mutant Asn282Ala, Asp283Ala, Asn284Ala also had only slightly changed kinetic properties. The present results show that these residues from the 280s loop play only a minor role in binding oligosaccharide.
The most significant conformational change in the MalP structure on binding oligosaccharide involves residues 377–384, a region designated as the 380s loop. In the T state (less active) form of GP this loop helps close the catalytic site through hydrophobic and ion pair interactions. In the R state (active) form of GP this loop is displaced from the catalytic site. The movement of the 380s loop in the T–R transition is correlated with the movement of Arg569, the important substrate phosphate binding residue. It has been suggested that the changes in the 380s loop in the mammalian GP provide a means of communication between the catalytic site and the glycogen storage site (Barford and Johnson, 1992; O'Reilly et al., 1997).
MalP is constitutively active, and in the native MalP structure, Arg569 is already in its correct position to bind the phosphate substrate, since the crystals were grown from a phosphate‐buffered solution (Watson et al., 1997). There is no conserved glycogen storage site in MalP, which provides an explanation for the preference of linear maltodextrin substrates over large branched glycogen substrates. The simplicity of the bacterial enzyme in binding a linear oligosaccharide, allows direct observation of the role of the 380s loop. In MalP the conformational change of this loop results in closure of the catalytic site, which is important both for recognition and correct positioning of the oligosaccharide in the catalytic site (Figure 6). The changes in the pentasaccharide complexes are similar to those observed previously for the maltose complex (O'Reilly et al., 1997), where maltose bound to sub‐site +1 and +2. In contrast, in the MalP–acarbose complex where acarbose bound to sub‐sites +1 to +4, no closure of the 380s loop is observed. The lack of movement can be rationalized in terms of the presence of the cyclitol ring, and not a glucopyranosyl ring, in sub‐site +1 (O'Reilly et al., 1999). Acarbose, although a potent inhibitor of glucosidases, is a poor inhibitor of MalP with a Ki of 5 mM. The Ki value for GSG4 determined in the direction of glycogen degradation has been found to be 1.5 μM (R.Schinzel, unpublished results). Taken together these observations suggest that, although MalP is able to bind the pseudo‐oligosaccharide acarbose with low affinity, high affinity binding of substrates is associated with a conformational change in the 380s loop. The shift in the 380s loop appears to have three consequences for binding oligosaccharide. First, it results in a 2 Å shift in the carbonyl oxygen of His377 that may be significant for catalysis (discussed later). Secondly, it allows Thr378 to contribute van der Waals interactions to the sugar in sub‐site +1. Thirdly, it promotes closure of the protein domains through an ionic interaction between Glu382 and His571. Both these residues hydrogen bond to the sugar in sub‐site +4. The shift in the position of Glu382 is 13 Å. The mutant enzymes Glu382Ala and His571Leu exhibit nearly 100‐fold decrease in kcat and a significant increase in Km for substrates compared with the wild‐type enzyme (Drueckes et al., 1996), although both these residues are >15 Å from the site of catalysis.
Oligosaccharide conformation upon binding to phosphorylase
The torsion angles about the glycosidic bonds for the pentasaccharides bound to MalP are summarized in Table III. The torsion angles between the sugars in sub‐sites +1, +2, +3 and +4 are not too different from those observed in the crystal structure of maltose (Takusagawa and Jacobson, 1978), and related compounds, where φ = 116°, ψ = −118° (see Table III for φ and ψ definitions). For sub‐sites +1 to +4 of the MalP–GSG4‐P complex, there are intramolecular hydrogen bonds between the O2 hydroxyl of one sugar and the O3′ hydroxyl of the adjacent sugar although the distance for the hydrogen bond between sub‐sites +2 and +3 is long (3.5 Å) (Table III). The most striking observation is the significant difference in torsion angles for the link between sub‐sites −1 and +1 from the preferred values. The torsion angles of φ = 68°, ψ = −189° result in an ∼180° flip in the sugar in sub‐site +1 relative to that in sub‐site −1 (Figure 7).
The change from the preferred values for the torsion angles allows the oligosaccharide to bind along the catalytic site channel. With the first sugar fixed in sub‐site −1 via hydrogen bonds to each of the peripheral hydroxyl oxygens, the sugar in sub‐site +1 is required to adopt an alternative conformation from the preferred conformation in order to avoid clashes with the enzyme, as first recognized nearly 20 years ago (Johnson et al., 1980; Withers et al., 1982). Either the enzyme or the substrate must alter conformation and we can now see that it is the substrate that alters its conformation with a rotation of about −48° on φ (resulting in a change of φ from 116° to 68°) and a rotation of about −71° on ψ (resulting in a change of ψ from −118° to −189°).
The conformational preference observed for sub‐sites −1 to +1 in the MalP complexes is close to a second minimum energy conformer (Rees and Smith, 1975; Perez and Vergelati, 1984; Bock et al., 1994) (Figure 7). The conformation lies just outside the allowed contours but the calculations do not take into account interactions of the sugars with the enzyme. The energy required for the altered conformation may be provided by the short and strong hydrogen bonds that are made by the sugar hydroxyl in sub‐site +1 to acidic residues (i.e. hydrogen bonds between the O2 and O3 hydroxyls to Asp339 and the O6 hydroxyl to Glu88). Similar (but not identical) oligosaccharide conformations have been reported for other enzymes recognizing α(1–4)‐glucosyl polymers, as first noted in the pig pancreatic amylase–acarbose complex (Qian et al., 1994) and other glucosidase–acarbose complexes (summarized in O'Reilly et al., 1999), and also observed with glucosyl polymers bound to inactive glucosidase mutant enzymes (Fujimoto et al., 1998; Uitdehaag et al., 1999). The conformation is one that is available to the α(1–4)‐oligosaccharide and one that is preferred by phosphorylase.
The catalytic mechanism of phosphorylase
Previous structural and biochemical studies with small molecular weight substrates and inhibitors with the mammalian GP led to proposals for the catalytic mechanism (Helmreich and Klein, 1980; Klein et al., 1984; McLaughlin et al., 1984; Johnson et al., 1989, 1990). This mechanism can now be discussed in terms of an oligosaccharide substrate. The α‐retaining mechanism proceeds via a double displacement reaction with two steps (Figure 8). In the first step, the 5′‐phosphate of the cofactor pyridoxal phosphate promotes general acid attack by the inorganic phosphate on the glycosidic oxygen. The hydrogen bond between the 5′‐phosphate of the cofactor and the inorganic phosphate substrate, observed previously with the dead‐end product heptulose‐2‐phosphate in time‐resolved experiments (Duke et al., 1994), allows ready exchange of protons between the two phosphate groups, as observed in 31P NMR experiments (Klein et al., 1984). Protonation of the glycosidic oxygen, between subsites −1 and +1, results in cleavage of the glycosidic bond and formation of a carbonium–oxonium ion transition state that is favoured and stabilized by the now negatively charged phosphate. In the second step, the inorganic phosphate acting as a nucleophile, attacks the carbonium ion C1 carbon leading to formation of the product α‐d‐glucose‐1‐phosphate with retention of configuration.
The involvement of the substrate phosphate group as an obligatory part of the reaction mechanism explains why transfer is made preferentially to the phosphate rather than to water. In the binary MalP–G5 and MalP–GSG4 complexes, a water molecule is observed close to the position of the phosphorus of the substrate phosphate in the MalP–GSG4‐P complex. The water is within 3.1 Å of the glycosidic oxygen (or sulfur) between subsites −1 and +1 and also contacts another water. It is just too far (3.5 Å) to make a strong hydrogen bond to the cofactor 5′‐phosphate. Evidently, it would be difficult for the cofactor phosphate to act as an acid–base in promotion of attack by water because the distance is too large to promote a strong interaction and because of differences in the ionization properties of water and phosphate.
Implications for catalysis from the MalP–oligosaccharide complexes
In the present work we have been able to substantiate and expand the proposals for the mechanism with direct observation of the ternary enzyme–thio‐oligosaccharide–phosphate complex in the E.coli MalP enzyme. Comparison of the binary complexes with the natural substrate, G5, and that with the inactive thio‐analogue, GSG4, shows that the oligosaccharides bind in essentially identical conformations, and provides support for the notion that the thio analogue is a good mimic of the natural substrate. The bound oligosaccharides show alteration in the glycosidic torsion angles between sub‐sites −1 and +1 from the preferred conformation for α(1–4)‐linked glucosyl polymers.
In the ternary MalP–GSG4‐P complex, the oligosaccharide position is identical to that seen in the binary complexes. The phosphate position is also closely similar to the position observed for the phosphate moiety in the native MalP structure, indicating that the phosphate recognition site is already available in the native enzyme, and is not dependent on binding an oligosaccharide. The position of the inorganic phosphate is also similar to that observed in time‐resolved and inhibitor binding studies with mammalian GP (Duke et al., 1994; Mitchell et al., 1996), supporting the notion that this is the substrate phosphate binding position. The constellation of the cofactor 5′‐phosphate, the inorganic phosphate and the glycosidic linkage of the thio‐oligosaccharide are fully consistent with the mechanism shown in Figure 8, in which general acid attack on the glycosidic linkage is promoted by the cofactor phosphate via exchange of a proton between the two phosphate groups.
In an alternative mechanism, it has been proposed that the cofactor phosphate acts as an electrophile in order to withdraw electrons from the inorganic phosphate without exchange of protons between them (Withers et al., 1981a,b). This mechanism involves a constrained dianion pentavalent intermediate of the cofactor 5′‐phosphate in which one oxygen of the inorganic phosphate approaches the fifth co‐ordination position of the cofactor phosphorus atom, as in a transferase reaction. In the present crystal structure there is no evidence for distortion of the 5′‐phosphate from tetrahedral geometry, nor are there any additional interactions that could stabilize such geometry. This mechanism is not supported by the present evidence, but it cannot be excluded since the penta‐coordinate 5′‐phosphate of the cofactor is a transient and highly unstable intermediate that is unlikely to be captured by static X‐ray experiments.
Additional factors for catalysis
The present ternary complex suggests three additional factors that are likely to assist catalysis. First, the O2 hydroxyl group of the glucosyl residue in sub‐site −1 forms a hydrogen bond to the O1 oxygen of the inorganic phosphate, in addition to the hydrogen bonds with a tyrosyl phenolic group and two water molecules. The interaction with the negatively charged phosphate is likely to result in a reduction of electro‐negativity of the O2 hydroxyl which otherwise disfavours the positive charge of the carbonium–oxonium ion transition state intermediate. The hydrogen bond of the O2 hydroxyl with the O− of the phosphate negatively polarizes the O2 atom, and this partial negative charge may be stabilized by the hydrogen bonds to the tyrosyl and the waters. Although donation of a hydrogen bond to the phosphate might reduce the pKa of the substrate phosphate, the contribution of neighbouring charged groups are likely to influence the pKa more strongly. Glu672 is just too far (3.5 Å) to participate directly in these interactions. Such electronic effects of the O2 have been shown to be important for glycosidase catalysis (McCarter et al., 1992; Braun et al., 1995). A similar contribution to the stabilization of the transition state has been observed in cyclodextrin glucosyltransferase (CGTase) (Uitdehaag et al., 1999).
Secondly, as observed previously, the phosphorylase catalytic site contains no negatively charged protein groups in the correct position to promote the carbonium–oxonium ion intermediate, unlike the glycosyl hydrolases where this role is almost universally taken by a carboxylate group. However, we note that the main chain oxygen atom of His377 is directed towards the O5 and C1 atoms of the sugar in sub‐site −1 at a distance of 3.2 Å to both atoms, such that the partial negative charge on this oxygen could provide some stabilization energy. A similar interaction was noted in the complex of GP with nojirimycin tetrazole (Mitchell et al., 1996). Where the attacking group is a negatively charged phosphate group rather than a water molecule, it could be advantageous to have a neutral group acting in a stabilizing role (Wang and Withers, 1995).
Thirdly, the altered conformation about the glycosidic linkage also allows a contact between the O3 hydroxyl of the sugar in sub‐site +1 and the O5 ring oxygen of the glucosyl in sub‐site −1 (3.1 Å). The O3 atom is also 2.8 Å from the C1 atom of the sugar in sub‐site −1. In the MalP–GSG4‐P complex, the O3 hydroxyl is involved in a hydrogen bond to Asp339, presumably as a hydrogen donor at pH 6.7, the pH for optimal activity (Schinzel and Drueckes, 1991). This arrangement would allow the lone pair electrons on the O3 atom also to make some contribution to the stabilization of the positive charge on the developing carbonium–oxonium ion. The mutant Asp339Ala exhibits a 100‐fold decrease in kcat compared with wild type with no significant change in Km values for substrate (Drueckes et al., 1996).
In each of the MalP–oligosaccharide complexes, the sugar in sub‐site −1 can be accommodated in the usual chair conformation. In particular, the positions of the hydroxyl oxygens of the sugar, located in the 2.0 Å electron density map of the ternary complex, are consistent with a chair conformation and show no indication of a half‐chair or sofa conformation, although the possibility of subtle changes cannot be excluded at this resolution. This is in agreement with previous work with GP, which has shown that sub‐site −1 can accommodate a range of conformations from half chair to chair, for glucosyl and glucosyl‐related residues in sub‐site −1, and in contrast to recent work with CGTase where a partial distortion to half‐chair was observed (Uitdehaag et al., 1999).
Stereoelectronic theory suggests that in the grouping of atoms O‐C1‐O there is a possibility of mixing lone pair orbitals on one oxygen with the anti‐bonding σ* orbital of the adjacent C‐O bond. Such interactions are conformationally dependent and occur when the lone pair orbital on one oxygen is anti‐periplanar (180°) to the adjacent C‐O bond. The theory, reviewed by Gorenstein (1987), has received experimental and theoretical support but it is not without criticism, mostly concerning the magnitude of the effects and their likely contributions to catalysis. Restricting the discussion to the 4C1 chair conformation of pyranose rings, for α‐glycosides the axial C1‐O1(O4′) bond is anti‐periplanar to one of the O5 oxygen lone pair orbitals, resulting in strengthening (and shortening) of the O5‐C1 bond [the endo anomeric effect (Praly and Lemieux, 1987)]. Such interactions will tend to favour and stabilize a carbonium–oxonium ion. This suggests that α‐glycosides with the axial aglycon may be cleaved in the ground state conformation. We note that in the MalP–GSG4‐P complex there are no hydrogen bonds to the O5 atom of the glucosyl residue in sub‐site −1 that would tend to weaken the endo anomeric effect (Leu136 N is 4.2 Å away) and the glucosyl group is in the chair conformation.
The exo anomeric effect occurs when one of the lone pair orbitals on the O1(O4′) oxygen is anti‐periplanar to the C1‐O5 bond, and this will be conformationally dependent on the glycosidic torsion angle φ. Ab initio calculations suggest that the exo anomeric effect will be strongest for a gauche orientation, that is, when φ = 60° and weakest for an anti orientation, that is, when φ = 180° (Cramer et al., 1997). Because of the competition for electron deficiency at the C1 atom between the endo and exo effects in α‐glycosides, the exo anomeric effect is weaker for α‐ than for β‐glycosides. In β‐glycosides, where no endo anomeric effect is possible, ring distortion would facilitate glycoside hydrolysis. Such distortions have been observed in lysozyme (Ford et al., 1974; Strynadka and James, 1991; Hadfield et al., 1994), chitobiase (Tews et al., 1996) and endoglucanase (Sulzenbacher et al., 1996). For cleavage of the α‐C1‐O1(O4′) bond, catalysis will be assisted if the exo anomeric effect is weakened. However, we note that the alternative conformation adopted by the oligosaccharide when bound to phosphorylase (φ = 68°) is such that the exo effect is about the same as in the preferred conformation (φ = 116°). This apparent conflict with stereoelectronic theory can be resolved if we compare the conformation of the product, glucose‐1‐phosphate, at the catalytic site as deduced from studies with heptulose‐2‐phosphate (Johnson et al., 1990). The torsion angle in heptulose‐2‐phosphate φ = 220°, is such that would weaken the exo anomeric effect. In the reverse reaction the cofactor 5′‐phosphate acts as an acid to protonate the phosphate of glucose‐1‐phosphate and the C1‐O1 bond has to be cleaved without direct protonation of the glycosidic oxygen. In such circumstances weakening of the exo anomeric effect will be advantageous.
Oligosaccharide conformation and catalysis
What is the significance of the alteration from the preferred geometry about the glycosidic linkage between sub‐sites −1 and +1 for catalysis? Several factors are suggested.
(i) The alteration in torsion angles is demanded by the geometry of the catalytic site tunnel. Only in the altered conformation are the catalytic groups correctly aligned with respect to the susceptible bond. In particular, it allows in‐plane protonation (Heightman and Vasella, 1999) of the glycosidic oxygen by the phosphate.
(ii) The altered geometry allows the two substrates (either oligosaccharide and phosphate or oligosaccharide and glucose‐1‐phosphate) to be present at the same time. The reaction mechanism demands that the attacking group is present before the leaving group has left and it also demands that the reaction proceeds with retention of configuration.
(iii) The altered conformation breaks the intramolecular hydrogen bond between the O2 of one sugar and the O3′ of the adjacent sugar and the polymer is opened up for attack, whereas in the preferred conformation the O2…O3′ hydrogen bond partially prevents access for protonation of the glycosidic bond. The presentation of the glycosidic oxygen to the attacking phosphate is optimized and the lone pair electron on the O3′ oxygen may assist in stabilization of the oxonium‐carbonium ion.
These considerations provide some rationalization for the alternative conformation adopted by the oligosaccharide when bound across the catalytic site in MalP. It is striking that in each of the α‐glucosidases (amylases and cyclodextrin glucosyl transferases), where the reaction is known to proceed via a two‐step mechanism with retention of configuration, a similar but not identical alteration of the conformation from the preferred conformation is observed (summarized in O'Reilly et al., 1999). In contrast, in the single structural example of an α‐glucosidase, glucoamylase, in which the reaction proceeds via a single step mechanism with inversion of configuration, the oligosaccharide binds in the preferred conformation for α(1–4)‐linked glucosyl polymers (Aleshin et al., 1996). The significance of the bound oligosaccharide conformation for the presentation of the susceptible glycosidic bond to the catalytic residues merits further investigation.
Materials and methods
Expression and purification of E.coli maltodextrin phosphorylase
Maltodextrin phosphorylase was purified with minor modifications as described previously (Schinzel and Palm, 1990; Schinzel et al., 1992). Briefly, four 2 l flasks each containing 700 ml of Luria–Bertani broth were inoculated from an overnight culture of the expression strain (E.coli ΔmalA518 harbouring the plasmid pMAP101). Maltodextrin phosphorylase was induced with 2 g/l maltose, after ∼4 h at 37°C (OD595 = 0.8–1.0). A further 16 h of growth was allowed, after which time the cells were harvested by centrifugation. The pellet was resuspended in buffer A (10 mM Tris pH 7.3, 1 mM EDTA), and disrupted by 10 mM MgCl2, 1:1000 dilutions of 10 mg/ml RNase A, 10 mg/ml DNase I and 50 mg/ml lysozyme for 30 min at 0°C, followed by sonication (5 × 30 s), and then centrifugation to remove the cell debris. The supernatant was fractionated by ammonium sulphate precipitation (35% then 75%) stirring for 30 min at 0°C, and centrifuged. The pellet was resuspended in 5 ml buffer A and dialysed overnight against buffer A. The filtered dialysate was applied to a Q‐Sepharose anion exchange column (Pharmacia), equilibrated with buffer A, and eluted using a gradient of 0–0.5 M NaCl in buffer A. MalP elutes as the second peak (∼0.2–0.25 M NaCl). These fractions were dialysed against buffer B (20 mM Tris pH 6.9, 2 mM EDTA). The resulting dialysate was loaded onto a glycogen‐glycin–Sepharose affinity column (prepared by coupling glycogen to Sepharose through CNBr), equilibrated with buffer B, and eluted with a gradient of 0–0.5 M NaCl in buffer B. The resulting dialysate was applied to a Superdex 200 gel filtration column (Pharmacia) equilibrated against 0.15 M NaCl in buffer B. Maltodextrin phosphorylase eluted as the second peak. The purified enzyme was concentrated to ∼10 mg/ml in buffer B and used immediately for crystallization experiments. The final protein concentration was determined from the absorbance at 280 nm, using E1cm0.1% = 1.36.
The stock solutions used for the crystallization experiments were 1 M Tris pH 8.5; 50% (w/v) 4K PEG; 2 M LiCl; 50 mM Na2HPO4 (Pi). A maltopentaose (G5) solution (100 mM) was prepared in 1 M Tris pH 8.5. GSG4 was made up as a saturated solution (∼20 mM) in water. The synthesis of GSG4 will be described elsewhere (N.Payre, S.Cottaz and H.Driguez, manuscript in preparation).
All the crystallization experiments were performed using the hanging drop method and micro‐seeding (1:100 dilution) in order to obtain large diffraction quality crystals (typically 100 × 50 × 10 μm3). For all three complexes, crystals were obtained from a screen of 26–32% 4K PEG, 0.1–0.6 M LiCl and 1 M Tris pH 8.5. The drops for the co‐crystallization experiments were set up as follows. For the MalP–G5 and MalP–GSG4 binary complexes, a 4 μl drop consisted of 2 μl of concentrated protein solution, 0.4 μl of G5 or GSG4 stock solution and 1.6 μl of well solution. For the MalP–GSG4‐P ternary complex, a 4 μl drop consisted of 2 μl of a 10 mg/ml protein solution, 0.4 μl of GSG4 solution, 0.4 μl of 50 mM Na2HPO4 and 1.2 μl of well solution. Two different crystal forms were obtained under these conditions. The GSG4‐P ternary complex crystals had a square, plate‐like morphology that correspond to space group P21 with cell dimensions a = 107.4 Å, b = 61.7 Å, c = 126.7 Å, β = 104.3°, one dimer/asymmetric unit. The MalP–G5 and MalP–GSG4 complexes both crystallized in the orthorhombic space group P212121 exhibiting a rectangular plate‐like morphology, one dimer/asymmetric unit, and cell dimensions a = 74.45 Å, b = 105.69 Å, c = 214.91 and a = 74.74 Å, b = 105.02 Å, c = 217.96 Å, respectively. This crystal form is similar to that obtained for the MalP–acarbose complex crystals grown under similar conditions (O'Reilly et al., 1999). A further crystal form could be obtained for the binary complexes in the presence of high concentrations of LiCl and 4K PEG (typically 0.6–0.8 M LiCl and 30–32% 4K PEG). This second crystal form had a prismatic morphology and was found to correspond to a C2 cell with dimensions a = 170.1 Å, b = 111.7 Å, c = 120.8 Å, β = 119.3°. This is similar to the previously reported MalP–maltose complex (O'Reilly et al., 1997).
Data collection and processing
The crystals were quickly passed through a solution of mother liquor and 20% MPD for flash freezing to 100 K prior to data collection. Glycerol as cryoprotectant was avoided, since previous experiments with MalP had shown that glycerol bound at sub‐site −1 and blocked oligosaccharide binding at this site (O'Reilly et al., 1997, 1999). Data to 2.9 Å resolution for the orthorhombic forms of the binary complexes MalP–G5 and MalP–GSG4 were collected using a 300 mm Mar Imaging Plate at the Daresbury Synchrotron Radiation Source station PX7.2 (λ = 1.488 Å). For the MalP–GSG4‐P ternary complex, data to 2.0 Å resolution were collected on PX9.6 (λ = 0.87 Å) using the ADSC CCD detector.
All three data sets were processed using MOSFLM (CCP4, 1994). Although data to 2.9 Å resolution were collected for the G5 complex, the Rmerge, and the refinement suggested that all the data could be used for the refinement, but for the map calculation only data to 3.4 Å were used (81.0% complete, Rmerge = 0.15, for outer shell 3.7–3.4 Å). The GSG4‐P data were anisotropic resulting in diffraction to 2.0 Å resolution, when the X‐rays were normal to the plate‐like form of the crystal, but for some images, only to 2.5 Å resolution when the X‐rays were parallel to the plate of the crystal. For this reason, a dynamic high resolution cut‐off limit in MOSFLM that depends on mean I > σ(I) was applied to single frames. Consequently, all the data to 2.0 Å resolution for the MalP–GSG4‐P complex were used in the refinement but the map calculation used only data to 2.2 Å (81.3% complete, Rmerge = 0.101, for outer shell 2.4–2.2 Å).
Phasing and refinement
Both binary structures were solved by molecular replacement in AMoRe (Navaza, 1994) using the 2.4 Å resolution native MalP structure as the starting model. For the MalP–G5 structure, the rigid body refinement in AMoRe gave an R factor of 0.411 and a correlation coefficient (Cc) of 0.643. For the MalP–GSG4 structure, the R factor was 0.390 and the Cc was 0.615. The oligosaccharides (G5 and GSG4) were initially modelled using the biopolymer module of the program SYBYL (Tripos, 1992) and were easily located in the initial difference electron density maps using O (Jones et al., 1991).
The MalP–GSG4‐P ternary complex structure was also solved by molecular replacement in AMoRe, using the protein atoms only of the MalP–GSG4 binary complex as the starting model since there were significant differences observed in the region of the catalytic site between the native enzyme and the binary complex structures. The rigid body refinement between 8 and 4 Å resolution of the ternary complex gave an R factor of 0.316 and a Cc of 0.702. The GSG4 substrate was then easily located in an initial difference Fourier electron density map. Once the position of the oligosaccharide had been partially refined, the substrate phosphate anion in the ternary complex was located first by positioning the phosphorous atom from the location of the highest peak in a second difference Fourier calculation, followed by positional and B factor refinement. A subsequent difference Fourier map clearly revealed the position of two of the oxygen atoms. Positional and B factor refinement with these additional atoms, and a final difference Fourier map revealed the location of the last two oxygen atoms. The final ternary complex model was then subjected to simulated annealing, as described below.
All three complexes were refined using X‐PLOR (Brünger, 1992) following an iterative process of manual model building using O, positional refinement, isotropic B factor refinement, and simulated annealing (starting temperature 2500 K, 5 fs steps for 50 cycles with harmonic restraints applied to the water molecules) and energy minimization. Throughout the initial refinements during model building, 5% of each data set was flagged for calculation of Rfree. Final X‐PLOR refinements for all three complexes were performed by simulated annealing, followed by restrained anisotropic B factor refinement using a free R flag of between 2 and 5% (equivalent to ∼1200 reflections in each complex). For all three complexes, strict NCS constraints were applied to the dimer molecule. The final refinements gave R/Rfree = 0.245/0.306, R/Rfree = 0.214/0.277 and R/Rfree = 0.228/0.261, for MalP–G5, MalP–GSG4 and MalP–GSG4‐P, respectively. In each case, electron density maps were calculated using shell‐scaling in which the amplitudes were matched and scaled in resolution shells using Rayment weighting (Rayment, 1983; D.I.Stuart, unpublished programs), and 2‐fold averaging using RAVE (Kleywegt and Jones, 1994). The additional phasing information to 2.0 Å for the ternary MalP–GSG4‐P complex, allowed a different assignment of eight amino acids in the primary sequence that could be clearly traced in the 2.2 Å resolution electron density map. These sequence changes were not observed in the previous 2.4 Å native structure, but are in agreement with the primary sequence reported by F.R.Blattner in the DDBJ/EMBL/GenBank data bank. The assignments using the E.coli numbering scheme (with the previous assignment in brackets) are Lys294 (Glu), Glu488(Val), Phe490(Leu), Gln499(Leu), Glu502(Val), Glu522(Asp), His548(Arg), Glu682(Lys) (Palm et al., 1987; Blattner et al., 1997).
The program PROCHECK was used to assess the geometry of the final structures. The co‐ordinates for all three complexes have been submitted to the Brookhaven Protein Data Bank.
The authors are grateful to Professor Dieter Palm and Dr Reinhard Schinzel for providing us with the MalP expression strain. We thank Dr Stephen Withers for discussions. We wish to thank the staff at SRS Daresbury on Stations PX7.2 and PX9.6 for providing excellent facilities for data collection. Dr Richard Bryan and Dr Yuguang Huang are acknowledged for maintaining excellent computing facilities within the laboratory. This work is supported by the British Diabetic Association, the Medical Research Council and by the European Union Biotechnology Programme (contract BIO4‐CT98‐0022).
- Copyright © 1999 European Molecular Biology Organization