|
The enzyme methylmalonyl-coenzyme A (CoA) mutase, an 
heterodimer of 150 kDa, is a member of a class of enzymes that uses coenzyme B12 (adenosylcobalamin) as a cofactor. The enzyme induces the formation of an adenosyl radical from the cofactor. This radical then initiates a free-radical rearrangement of its substrate, succinyl-CoA, to methylmalonyl-CoA.
Reported here is the crystal structure at 2 Å resolution of methylmalonyl-CoA mutase from Propionibacterium shermanii in complex with coenzyme B12 and with the partial substrate desulpho-CoA (lacking the succinyl group and the sulphur atom of the substrate). The coenzyme is bound by a domain which shares a similar fold to those of flavodoxin and the B12-binding domain of methylcobalamin-dependent methionine synthase. The cobalt atom is coordinated, via a long bond, to a histidine from the protein. The partial substrate is bound along the axis of a (/)8 TIM barrel domain.
The histidine–cobalt distance is very long (2.5 Å compared with 1.95–2.2 Å in free cobalamins), suggesting that the enzyme positions the histidine in order to weaken the metal–carbon bond of the cofactor and favour the formation of the initial radical species. The active site is deeply buried, and the only access to it is through a narrow tunnel along the axis of the TIM barrel domain.
Adenosylcobalamin, or coenzyme B12 ( Fig. 1a ), has fascinated chemists and biologists ever since its structure was determined by Lenhert and Hodgkin [1] over three decades ago. The principal interest in this molecule stems from the fact that it contains one of the very few naturally occurring metal–carbon bonds, a feature shared with the related molecule, methylcobalamin. The weakness of this bond between the cobalt atom and the 5'-carbon of the 5'-deoxyadenosyl group provides the special properties of this enzyme cofactor. The generally accepted mechanism of adenosylcobalamin-containing enzymes is that the CoIII–C bond splits homolytically to form the 5'-deoxyadenosyl radical, which initiates the enzyme reaction, leaving a pentacoordinated CoII atom in the cobalamin (cob(II)alamin or B12r). (For a review, see [2].)
| View Image |
|
Figure 1 Coenzyme B12 and reaction schemes. (a) 5'-Deoxyadenosyl-cobalamin (coenzyme B12), showing part of the atom nomenclature scheme. The dashed lines refer to the bonds between the central cobalt atom and two axial coordinating groups. In the free coenzyme, the 5'-C atom of 5'-deoxyadenosine provides upper axial coordination, while lower axial coordination comes from the N3B atom of a dimethylbenzimidazole linked to the porphyrin-like corrin ring. (b) Minimal mechanistic scheme for coenzyme B12-dependent rearrangements [2]. AdoCH2-B12 is 5'-deoxyadenosyl-cobalamin, Ado-CH2 and Ado-CH3 are respectively the 5'-deoxyadenosyl radical and 5'-deoxyadenosine. X is a generic migrating group. (c) The rearrangement reaction catalyzed by methylmalonyl-CoA mutase. The thioester group (O=C–CoA) is the group that migrates (CoA terminates with a sulphur). |
Ten adenosylcobalamin-dependent enzymes have been identified [3]. All but one of these, a ribonucleotide reductase, catalyze unusual 1,2 rearrangements. These are all reactions in which a carbon-bonded hydrogen is exchanged with a group on an adjacent carbon (H and X in Fig. 1b ). The reaction is thought to involve the adenosyl radical abstracting a hydrogen from the substrate to form a substrate radical [2] ( Fig. 1b ). This undergoes rearrangement to form the product radical, which then abstracts a hydrogen from the adenosine to form the product and to re-form the adenosyl radical. The evidence for this mechanism comes mainly from direct spectroscopic observation of free-radical intermediates by electron paramagnetic resonance (EPR) spectroscopy and from experiments with labelled substrates and cofactor [4] [5] [6] [7] [8] [9].
Methylmalonyl-coenzyme A (CoA) mutase is one of the most studied enzymes of this class and is the only one which is present in both mammals and bacteria. In the Gram-positive bacterium Propionibacterium shermanii, methylmalonyl-CoA mutase is involved in the fermentation of pyruvate to propionate, whereas in mammalian liver mitochondria the enzyme is responsible for the conversion of odd-chain fatty acids and branched-chain amino acids via propionyl-CoA to succinyl-CoA for further degradation. The enzyme catalyzes the interconversion between (2R)-methylmalonyl-CoA and succinyl-CoA ( Fig. 1c ) by the migration of the O=C–CoA group [10] [11].
The P. shermanii enzyme is an heterodimer of 150 kDa [12] (80 kDa for 728 residues in the chain and 70 kDa for 638 residues in the chain [13]), with one binding site per dimer for acyl-CoA [14] and for B12 [15]. In contrast, the human enzyme is an 2 homodimer [16], and is highly homologous (60% sequence identity) to the active chain of its bacterial counterpart. The bacterial and chains show a lesser degree of homology, with 24% of the residues being identical.
There is one other B12 cofactor, methylcobalamin, in which the carbon of a methyl group serves as the upper axial ligand to the cobalt: this is the cofactor for methionine synthase. In this case the cobalt–carbon bond is cleaved heterolytically, and the cobalt is reduced from a CoIII oxidation state to a highly reactive CoI species, while the methyl group is transferred to the substrate homocysteine to form methionine [17]. The structure of the inactive B12-binding domain of this enzyme has been solved recently [18]. It showed for the first time how the dimethylbenzimidazole base that coordinates the cobalt in the free methylcobalamin is substituted on the enzyme by a histidine residue. Although sequence comparisons reveal relatively few conserved residues [19], extensive structural similarities exist between the B12-binding domains of methionine synthase and methylmalonyl-CoA mutase.
We present the crystal structure to 2 Å resolution of the ternary complex between recombinant methylmalonyl-CoA mutase, coenzyme B12 and the partial substrate desulpho-CoA (CoA with the terminal thiol group replaced by a hydrogen). The implications of the structure for the binding of substrate and cofactor, and for the reaction mechanism are discussed.
Crystals of methylmalonyl-CoA mutase with bound coenzyme B12 and in presence of an excess of desulpho-CoA grow under identical conditions in two morphologically indistinguishable and closely related forms, one monoclinic, the other orthorhombic. Each asymmetric unit contains two heterodimers of methylmalonyl-CoA mutase. The structure determination proved troublesome, in part because of the weak diffraction of these crystals, but mostly because of poor isomorphism between native crystals. All data were collected from crystals frozen at 95 K. Cryo-cooling was essential both to minimize the radiation damage caused by the high doses of X-ray radiation needed to record the high-resolution reflections, and to obtain complete data sets from single crystals thereby overcoming the problems of non-isomorphism. As one approach to phasing, multiwavelength anomalous diffraction (MAD) data were collected from a single orthorhombic crystal soaked in K2Pt(NO2)4. This experiment alone was not sufficient to produce an interpretable map, even after twofold averaging. This failure was partly due to the large molecular mass in the asymmetric unit (300 kDa), and partly due to a pseudo-centrosymmetric arrangement of the platinum atoms, which leads to ambiguity in the phasing. More useful phases were derived from isomorphous pairs of derivative and native crystals (obtained by cutting the monoclinic crystals into two halves and using one half for collection of native data and the other half for collection of derivative data). Averaging the four copies of the molecule from the maps in the two different space groups gave an interpretable map to 3 Å resolution (
Fig. 2
).
| View Image |
|
Figure 2 Part of the electron density map at 3 Å resolution that was used to build the model, showing the |
The current model, refined against 2 Å resolution data, consists of amino acid residues 2–728 from the active chain (A2–A728) plus cobalamin and desulpho-CoA, and residues 18–638 from the inactive chain (B18–B638).
The and chains exhibit similar folds. Each consists of two principal domains (
Fig. 3a
,
Fig. 3b
,
Fig. 4a
and
Fig. 4b
): an N-terminal eight-stranded / barrel, and a C-terminal / domain with five parallel -sheet strands (a fold resembling those of flavodoxin and the B12-binding domain of methionine synthase). However, the relative orientation of the domains differs, and the chain binds neither cobalamin nor desulpho-CoA.
| View Image |
|
Figure 3 Schematic views of the structure (drawn with MOLSCRIPT [52]). (a) The |
| View Image |
|
Figure 4 Stereoviews of C |
The N-terminal barrel is preceded by an extended segment which wraps around the other subunit. It is followed by a long linker region (about 160 residues) that encloses the barrel domain and connects it to the C-terminal domain. The linker region in the chain consists of two helices packed against one side of the / barrel, followed by a poorly ordered loop stretched across the surface of the barrel, then another group of four helices packed against the other side of the barrel domain before it leads into the C-terminal domain. The linker region in the chain is much shorter (95 residues). In the chain, the C-terminal domain packs on one end of the / barrel, sandwiching the corrin ring of the cobalamin and forming the active-site cavity. In the chain, the C-terminal domain is swung away from the / barrel domain, and no active site is formed (
Fig. 4c
). The / barrel domains of the and chains are related by a local twofold axis, with two helices from each subunit packed together around the dyad. The approximate dyad symmetry breaks down about halfway through the linker region, at residues A482 and B477.
The flavodoxin-like C-terminal domain of the chain is responsible for cofactor binding (
Fig. 3c
). This domain exhibits only slight similarity in amino acid sequence to the B12-binding domain of methylcobalamin-dependent methionine synthase, but has a similar tertiary structure. In fact, the C-terminal domain of the chain superimposes on the B12-binding domain of methionine synthase better (with a root mean square [rms] deviation of 1.7 Å for 110 C atoms) than it does on the corresponding domain of the chain (rms deviation of 2.1 Å for 83 C atoms). The cobalamin also binds in a very similar way to the two enzymes. In free coenzyme B12, the lower axial ligand of the cobalt atom comes from the N3B nitrogen of the dimethylbenzimidazole group, which is linked to the corrin ring via the pseudo-nucleotide tail (see
Fig. 1a
). In contrast, in methylmalonyl-CoA mutase, as in methionine synthase, the tail is completely displaced away from the cobalt and is bound in a deep pocket between the sheet and the C-terminal helix. The cobalt ligand is formed by the 
-nitrogen (N2) of histidine A610; the N
1 atom of this histidine starts a hydrogen-bonded chain to AspA608 and LysA604. This lysine is completely buried.
EPR spectroscopy has provided independent confirmation of a histidine ligand to the cobalt atom [18] [20]. The histidine and aspartate residues are conserved between methylmalonyl-CoA mutases from different species, and between methylmalonyl-CoA mutase and 2-methyleneglutarate mutase, glutamate mutase and methionine synthase [19]. The lysine, on the other hand, is only present in the methylmalonyl-CoA mutases. These three conserved residues form part of a B12-binding motif, which is not apparent in the sequences of ethanolamine ammonia-lyase and diol dehydrase, two of the other adenosylcobalamin-dependent enzymes.
On the upper side of the cobalamin we would expect to see the adenosyl group above the cobalt, but no such group is present in the current structure. Electron-density maps ( Fig. 5a ) show no evidence of any ligand in this position. This pentacoordinated state of the cobalt is consistent with a reduced CoII species (cob(II)alamin or B12r), rather than CoIII, which would be expected to be hexacoordinated. Absorption spectra from crystals grown under similar conditions ( Fig. 6 ) show that they contain substantial and variable amounts of CoII: the crystal used to collect the 2 Å resolution data was probably a mixture of CoII and CoIII species, with CoII predominating. The CoII species could be formed by diffusion of the adenosyl radical away from the enzyme.
| View Image |
|
Figure 5 The active site and substrate binding. (a) The active-site pocket, with the SIGMAA-weighted 2Fo–Fc map [44] at 2 Å resolution after refinement. The upper axial ligand position on the cobalt atom is unoccupied. The position that would be occupied by the succinyl or methylmalonyl group of the substrate or product, is filled by a tightly bound water molecule (orange sphere, B=13.5 Å2), also hydrogen bonded to N |
| View Image |
|
Figure 6 Absorption spectra taken (a) from crystals and (b) in solution. In (a), the two solid lines represent data recorded from two crystals of the complex between methylmalonyl-CoA, coenzyme B12 and desulpho-CoA. The dashed line represents data recorded from a crystal grown in the absence of any CoA analogue. In (b), the spectrum typical of enzyme-bound CoIII-adenosylcobalamin (dashed line) was recorded from a 2.5 mg ml-1 solution of methylmalonyl-CoA mutase with bound cofactor. The spectrum typical of cob(II)alamin (solid line) was obtained by addition of the non-physiological substrate succinyl-(carbadethia)-CoA to the solution used for the CoIII spectrum [9]. |
Some details of the stereochemistry of the cobalt atom and the corrin are given in Table 1, both for methylmalonyl-CoA mutase and for various free cobalamin structures. The most notable feature in the mutase is the very long bond length, about 2.5 Å, between the cobalt atom and nitrogen atom of the lower axial ligand (N2 of HisA610). Comparison of the axial bond lengths in different cobalamins shows that a strongly bonded upper ligand such as CN or OH leads to a short Co–N bond to the lower axial ligand (1.93–2.01 Å), whereas the methyl and adenosyl groups, which have a Co–alkyl bond, have a longer lower axial bond (2.19–2.21 Å), similar in length to this bond in the pentacoordinated CoII of reduced B12r (2.16 Å).
| Table 1. Cobalamin stereochemistry. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Distances between the cobalt atom and axial ligands are given (in Å). *AdoCH2 is the 5'-deoxyadenosyl group. †Distance from the Co to the mean plane of the pyrrole nitrogens: a positive distance means a displacement towards the lower axial ligand. ‡The fold is along the Co–C10 line, defined as the angle between two linked conjugated planes in the corrin ring: C4 C5 C6 C7 C8 C9 C10 N21 N22; and C10 C11 C12 C13 C14 C15 C16 N23 N24. §The tilt is the angle between the C7–C8 bond and the plane of the pyrrole nitrogens (i.e. the out of plane twist of the C6 C7 C8 C9 N22 pyrrole ring). #Imidazolyl-B12 has an imidazole group replacing the dimethylbenzimidazole of vitamin B12. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
To examine the proposition that the Co–N2(HisA610) bond has a more conventional length, this bond was added as a restraint in the refinement, with a length of 2.20 Å and it was given the same weight as the other covalent bonds in the molecule (ensuring that there were no repulsion restraints between the histidine and the corrin). This reduced the bond length to 2.30 Å, but distorted the histidine ring. Very similar values for this bond length were obtained when restraints were applied within the programs REFMAC (which uses the same restraints as in PROLSQ, see the Materials and methods section) and X-PLOR [21] (which uses pseudo-energy restraints), showing that the bond length is insensitive to the type of stereochemical restraints used in the refinement. We interpret these results as validating the unusual bond length, and that the protein holds the corrin and the histidine further apart than their ideal positions.
The corrin ring in the mutase is flatter than in most cobalamins (Table 1): for instance, the fold angle of the conjugated systems about the Co–C10 line is similar to that in imidazolyl-B12 rather than to that in the normal cobalamins in which dimethylbenzimidazole is linked to the cobalt. The bulkier base of the free coenzyme bends the corrin upwards more than the smaller imidazole or histidine. In the mutase, pyrrole ring B is substantially less twisted from the corrin plane than in any of the free cobalamins. This can be seen from the angle between the C7–C8 bond and the plane of the pyrrole nitrogens ('tilt' in Table 1). It may be significant that the acetamide group attached to C7 is hydrogen bonded to TyrA89 ( Fig. 7a ), which is one of the putative active-site residues. Ring B is also less well ordered than the rest of the corrin, as indicated by higher temperature factors (data not shown).
| View Image |
|
Figure 7 Schematic representation of ligand binding. (a) The interactions between protein and coenzyme B12. The hydrophobic pocket for the dimethylbenzimidazole is lined by residues IleA617, TyrA705, GlyA685 and SerA655: this last forms a hydrogen bond to the N3B nitrogen atom of the base, which is the atom that coordinates the cobalt atom in the free coenzyme. LeuA657 stacks against HisA610, the residue that coordinates the cobalt, and forms hydrophobic interactions with the beginning of the pseudo-nucleotide tail (C56) and with the C20 methyl group of the corrin. (b) The interactions between protein and desulpho-CoA. TyrA75 stacks on the adenine ring. Note that ArgB45 is the only interaction between the |
Details of the B12-binding site are shown schematically in
Fig. 7a
. The pseudo-nucleotide tail runs through a cavity filled with well-ordered water molecules, then buries the dimethylbenzimidazole group in a tight hydrophobic pocket. The oxygens of the phosphate group interact only with two water molecules which are in turn bonded to other waters, to the nitrogen of GlyA613 and to the amide oxygen O63 of the corrin. The side-chain amides attached to the corrin interact with both the (/)8-barrel domain and the C-terminal domain.
The partial substrate, desulpho-CoA, is bound along the axis of the (/)8-barrel domain, with the ADP-3'-phosphate moiety at the N terminus of the strands, partly exposed to solvent, and the pantetheine chain completely buried between the strands (
Fig. 3d
). In the true substrate or product, the succinyl or methylmalonyl group would protrude into the active-site cavity (see below). The pantetheine chain of CoA is approximately the same length as the strands of the barrel.
The (/)8-
or TIM-barrel structure is one of the most frequently observed folds in proteins, with at least 45 examples in the Brookhaven Protein Database. In these structures, the centre of the barrel is typically filled by large, often branched, hydrophobic, side chains [22]. In the chain of methylmalonyl-CoA mutase, the barrel is lined by hydrophilic, mainly small, side chains (ThrA85, SerA114, SerA162, SerA164, SerA166, ThrA195, SerA239, SerA285 and SerA362), creating a hole through which the CoA can thread, and providing hydrogen bonds to the amide groups of CoA, some via a network of water molecules (
Fig. 5b
and
Fig. 7b
). In the chain, the hole is filled, partly by the replacement of short side chains by longer ones (e.g. LeuB137 replaces SerA114, GluB165 replaces SerA164 and AsnB281 replaces SerA285) and partly by movement of longer side chains, such as ArgB110, to obstruct the hole (
Fig. 5c
). The side chains are still largely hydrophilic. The ADP-3'-phosphate group is bound by five arginines and one lysine. One of the arginines, ArgB45, is contributed by the chain.
The structures of at least eight CoA-binding proteins are known, in addition to methylmalonyl-CoA mutase, but they are completely unrelated and have no common motif for binding CoA. Binding of a ligand along the axis of a (/)8-barrel fold has not, to our knowledge, been observed previously. It seems particularly appropriate in this case, not only because it provides a narrow access channel to a deeply buried active site, allowing the enzyme to protect its reactive intermediates from side reactions, but also because the dimensions of the barrel fit CoA derivatives well. The hole is created by the absence of large side chains: the shape of the barrel is not significantly different from that in other proteins with this fold.
The present structure lacks both the adenosyl group of the cofactor and the rearranged part of the substrate. A detailed description of the active site, and hypotheses regarding the mechanism of rearrangement will therefore have to await further structures of substrate and inhibitor complexes. However, some observations can be made. The active-site cavity is formed by the loops at the C termini of the strands of the (/)8-barrel domain, and it is closed off by the corrin ring of the B12 coenzyme and its associated amide side chains. The active site is completely inaccessible to solvent, except through the CoA channel along the barrel axis. Such a deeply buried active site would protect the reactive radical intermediates from undesired side reactions, and a similar inaccessibility has been seen in a number of other free-radical enzymes, such as galactose oxidase [23], copper amine oxidase [24], and prostaglandin H2 synthase [25].
Modelling succinyl-CoA or methylmalonyl-CoA into the active site places the cobalt atom too far from the substrate for its direct participation in the rearrangement reaction. The carbonyl oxygen of the thioester comes close to the position of a water molecule tightly bound to HisA244. This histidine is in a position to bind to the carbonyl oxygen in the rearrangement reaction, and is likely to be the major catalytic residue. Other groups which may be involved in substrate binding include ArgA207, TyrA89 and the C7-acetamide of the corrin.
chainThe common ancestor of the heterodimeric bacterial methylmalonyl-CoA mutases and the homodimeric mammalian enzymes was presumably a homodimer, as the subunits exhibit substantial sequence homology. The inactive chain has no obvious major function, although it contributes one residue, ArgB45, to the substrate-binding site. The chain is significantly less well ordered than the chain, as shown by the higher average temperature factor (38 Å2 and 37 Å2 for the two chains, compared with 43 Å2 and 53 Å2 for the two chains in the asymmetric unit), and its larger number of poorly ordered loops. Many of the residues which are important for cofactor and substrate binding have been lost in the chain. The chain seems to be an evolutionary relic, although we cannot exclude the possibility of it having other functions.
An important role of the enzyme is to favour the breaking of the Co–C bond when the substrate is bound, in order to initiate the reaction. In the free coenzyme, the dissociation energy for homolysis of this bond is about 100–125 kJ mol-1 [26] [27], with an estimated dissociation rate of about 10-7 s-1 at 30°C [2] [3]. This rate is much too slow to explain the reaction rates of the order of 102 s-1that have been reported for several of these reactions, including the one catalyzed by methylmalonyl-CoA mutase [28]. Thus, one of the roles of the enzyme is to favour the formation of the initial cofactor-derived radical, both thermodynamically and kinetically, and to do so in response to substrate binding, which weakens the Co–C bond beyond that in the holoenzyme [4] [7] [8] [9]. The strength of the Co–C bond with respect to homolysis could be modulated by the lower axial ligand (trans effect), either by steric or electronic effects. Alternatively, the crowding of the adenosyl group by protein groups or by the upward folding of the corrin ring (cis effect) could influence the Co–C bond. We propose that the major effect in this enzyme is trans and steric, in that the protein holds the HisA610 side chain too far from the corrin, putting the cobalt into a strained or 'entatic' state [29]. The 2.5 Å long Co–N bond would stabilize the CoII species relative to CoIII, thus favouring the formation of the adenosyl radical. At the same time, the formation of hydroxycobalamin (which would lead to irreversible inactivation) is disfavoured. In order to hold the histidine in this position, the enzyme provides a strong hydrogen bond to the negatively charged AspA608. However, the aspartate will induce a negative charge on the histidine, which will favour CoIII with respect to CoII, and strengthen the Co–C bond. This undesired side effect of the structural anchor on the histidine is compensated for by the positively charged LysA604. On binding of the substrate, the position of the histidine relative to the corrin may be altered by conformational changes, such as relative movements of the domains. The charge state of the histidine is less likely to alter on substrate binding, because the buried LysA604 will lock the His–Asp–Lys system into a state with net zero charge. The proposed modulation of the Co–C bond strength by movement of the Co-linked histidine is reminiscent of the regulation of oxygen binding in haemoglobin by positioning the proximal histidine, which is linked to the R–T state allosteric transition [30].
Coenzyme B12 (5'-deoxyadenosylcobalamin) consists of a central cobalt atom in a porphyrin-like corrin ring, bonded to the 5'-C of 5'-deoxyadenosine on one side and to a base on the other side. This molecule is the cofactor for a series of enzymes, the mutases, which catalyze unusual rearrangement reactions. The reactions take place via radical intermediates, formed indirectly following breakage of the cobalt–carbon bond. It has been a long-standing puzzle how the enzymes catalyze the breaking of this bond.
Methylmalonyl-CoA mutase is the only member of this class found in mammals as well as in bacteria. The enzyme catalyzes the interconversion between succinyl-CoA and methylmalonyl-CoA. In bacteria such as Propionibacterium shermanii, it is involved in the fermentation of pyruvate to propionate, a pathway which helps balance the redox state in anaerobic growth. In mammals, the enzyme is required for the degradation of odd-chain fatty acids, to convert propionyl-CoA to succinyl-CoA.
The crystal structure of methylmalonyl-CoA mutase in complex with coenzyme B12 and the partial substrate desulpho-CoA (lacking the succinyl group and the terminal sulphur atom of the substrate succinyl-CoA) shows that the base which coordinates the cobalt in the free coenzyme B12 is replaced by a histidine side chain from the protein. Both the mode of binding and the fold of the B12-binding domain are very similar to those in the B12-binding fragment of methionine synthase [18], which binds the other B12 cofactor, methylcobalamin. Methylcobalamin provides a methyl group for methylation reactions as a positively charged group rather than as a radical, and it remains unclear why the cobalt–carbon bonds in the two cofactors are split in different ways (i.e. homolytically versus heterolytically). In the structure of the complex presented here, the coenzyme is largely in the reduced CoIIform, having lost the adenosyl radical. The bond between the cobalt and the histidine is significantly longer than in model compounds, and this stretched bond explains how the enzyme stabilizes the reduced state, favouring the formation of the adenosyl radical.
In methylmalonyl-CoA mutase, the substrate is bound through a narrow tunnel along the axis of a (/)8- or TIM-barrel domain. This tunnel provides the only direct access to an active-site cavity, which, in common with other enzymes that produce reactive radical intermediates, is protected from unwanted side reactions. The mode of substrate binding is completely different from that in the other 45 known structures with this common fold.
Recombinant P. shermanii methylmalonyl-CoA mutase was prepared from an overexpressing clone of Escherichia coli as described elsewhere [31]. Crystals of the complex between methylmalonyl-CoA mutase, coenzyme B12, and desulpho-CoA were grown by vapour diffusion at 23°C, by equilibrating against a reservoir solution containing 14% (w/v) polyethylene glycol (PEG) 4000 and 20% (v/v) glycerol in 100 mM Tris-HCl buffer at pH 7.5 a 10 
l drop consisting of equal volumes of the reservoir solution and of a solution of 20 mg ml-1 protein, 2 mM coenzyme B12 and 12 mM desulpho-CoA in Tris-HCl buffer at pH 7.5. Crystals grow within three weeks to a size of up to 1.0×0.4×0.3 mm3. The crystals were stabilized in a solution containing 16% PEG 4000 and 20% glycerol in 100 mM Tris-HCl buffer at pH 7.5. The crystals grow in two closely related and morphologically indistinguishable forms, one monoclinic (crystal form 1 [32]) the other orthorhombic (crystal form 2). Each form contains two molecules in the asymmetric unit, with a solvent content of 48%.
Data collection statistics are given in Table 2. The crystals were suspended in a thin liquid film of stabilizing solution on a rayon loop [33] and frozen at 95 K directly in the cold nitrogen gas stream with a 600 Series Cryostream Cooler (Oxford Cryosystems, Oxford, UK) [34]. In order to overcome severe problems of non-isomorphism of derivatives in crystal form 1, each crystal was cut in two halves with a thin glass blade. The two halves were then transferred into two separate stabilizing solutions, one containing a heavy-atom compound. Native A and derivative A1, and Native B and derivative B, are isomorphous pairs made in such a way. Derivative A2 was found to merge as well with native A as with its own matching half, and so it was paired to native A. Native A and derivative A1 data sets were collected at a wavelength of 1.000 Å on station 9.5 at the Synchrotron Radiation Source (SRS), Daresbury, UK. Derivative A2, native B and derivative B data sets were collected using CuK radiation from an Elliot GX13 rotating anode generator with a 100 m focal spot, collimated by a Supper double-mirror system (Natick, MA). Native D data set was collected at a wavelength of 0.900 Å on Beamline 4 at the European Synchrotron Radiation Facility (ESRF), Grenoble, France. MAD data were collected from a single orthorhombic crystal (crystal form 2), pre-reacted by soaking in 2 mM K2Pt(NO2)4 for 48 h, on station 9.5 at the SRS, Daresbury, UK. The crystal was aligned with the b* axis approximately along the rotation axis to allow the near simultaneous recording of Bijvoet pairs of reflections and six data sets were collected on and around the platinum LIII absorption edge. Native C data were collected at a wavelength of 0.898 Å on station 9.6 at the SRS, Daresbury, UK. All the data were recorded on either 180 mm (derivative A2, native B, derivative B, MAD data) or 300 mm diameter MAR Research image plates (Hamburg, Germany) and integrated with MOSFLM [35]. Scaling and processing were performed using the CCP4 suite of programs [36].
| Table 2. X-ray structure determination. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
*Rmerge= i|I(h)–I(h)i|/iI(h)i, where I(h) is the mean intensity after rejections. †Riso=||FPH|–|FP||/|FP|, where FPH and FP are the structure factors for the derivative and native, respectively. For the MAD analysis, Riso is calculated between data sets at two different wavelengths: wavelength  5 is treated as the reference in this calculation. ‡Phasing power is rms (|FH|/E), where |FH| is the heavy-atom structure factor amplitude and E is the residual lack of closure (|FPH–FP|–|FH|). §RCullis=|E|/||FPH|–|FP||. The summation is over centric reflections only. #Crystal form 1 is monoclinic, spacegroup P21 (a=119.8 Å, b=161.3 Å, c=88.4 Å, =105.07°). Crystal form 2 is orthorhombic, spacegroup P212121 (a=123.4, Å b=162.0 Å, c=166.2 Å). Derivative A1 (two heavy-atom sites) was soaked for 20 h in 0.2 mM ethylmercury thiosalicylate. Derivative A2 (four heavy-atom sites) was soaked for 24 h in 0.4 mM methylmercury. Derivative B (eight heavy-atom sites) was soaked for 24 h in 15 mM KAu(CN)2. The crystal used for MAD data collection (eight heavy-atom sites) was soaked in 2 mM K2Pt(NO2)4 for 48 h.. |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
The isomorphous difference and anomalous difference Patterson maps were solved with the SHELXS-90 program [37]. Minor sites were subsequently found in difference Fourier maps. Heavy-atom parameter refinement and phase calculations were carried out for both crystal forms using the program MLPHARE [38]. In crystal form 1 two phase sets were derived: one for native A to 3 Å resolution, with the contributions from two derivatives, derivative A1 and A2; the other for native B to 3.2 Å resolution, with one derivative, derivative B. Phase probability coefficients were then combined and applied to the structure-factor amplitudes of native A. The derivative phase set (FPH) derived from the MAD analysis was transferred to a native set of amplitudes (native C) by vectorial subtraction of the calculated heavy-atom contribution. An initial molecular envelope was determined in crystal form 1 by performing a local correlation search with the DEMON package [39]. The density inside the molecular mask was then used as a molecular replacement model for crystal form 2 with a combination of MAVE [40], AMoRe [41] and TFFC [42]. The mask was subsequently improved by iterative averaging. The initial electron-density maps were improved within each crystal form by real-space solvent flattening and twofold averaging using SOLOMON [43], and between the forms by reciprocal-space phase combination using SIGMAA [44]. The procedure was carried out at 3.0 Å resolution for crystal form 1 and at 3.2 Å resolution for crystal form 2 and resulted in an interpretable 3.0 Å resolution map.
The atomic model was built using the graphics program O [45]. The model was fitted to the high-resolution native D data by rigid-body refinement with the program TNT [46], and the resolution extended in four steps from 3 Å to 2 Å. Refinement has been carried out with TNT and REFMAC (G Murshudov, unpublished program), which refines a maximum-likelihood target, using stereochemical restraints as in PROLSQ [47]. Initially, a tight non-crystallographic symmetry restraint was applied to keep the two molecules in the same conformation. At a later stage of refinement, this restraint was released to allow some small but significant differences between the molecules.
At the current stage of refinement the model has an R-factor of 21.9% for 95% of the data between 20 Å and 2 Å, including 1532 water molecules and 9 glycerol molecules, a total of 22683 atoms. The free R-factor for the remaining 5% of the data within this resolution range is 27.4%. The rms deviation from standard values of bond lengths is 0.017 Å.
The coordinates and structure factors have been deposited in the Protein Data Bank, ID code 1REQ.
We dedicate this paper to the memory of Dorothy Hodgkin. We are extremely grateful to P Williams and J Hajdu for recording the spectra from the crystals. We thank S Knight for his contributions earlier in the project, C Cardin and A Murzin for discussions, T Woollard, R Read and I Fearnley for their assistance, and C Kratky for cobalamin coordinates. We thank the EU for funds to travel to ESRF Grenoble, the staff of EMBL Grenoble, especially A Åberg, for assistance in experiments performed at the ESRF on BL4, and the staff of the Synchrotron Radiation Facility at Daresbury for their help. FM was supported by the Italian Consiglio Nazionale delle Ricerche (CNR), and by Sigma-Tau industrie farmaceutiche riunite s.p.a., Pomezia, Italy. NHK was supported by a Wellcome Trust Studentship and a Denman-Baynes Studentship from Clare College, Cambridge. AN was supported by the Japan Society for the Promotion of Science.