2.1. Crystallization and data collection

Recombinant B. halmapalus -amylase BHA was provided by Novozymes A/S. Briefly, the gene encoding BHA was isolated from the alkaliphilic Bacillus species B. halmapalus (Nielsen et al., 1995) and cloned into a B. subtilis expression vector derived from pUB110 (McKenzie et al., 1986). The resulting B. subtilis strain was grown in a complex soy-meal media from which the amylase was purified.

The hanging-drop vapour-diffusion method and in-house screens were used in a search for the crystallization conditions. Crystals in complex with acarbose were grown from enzyme solution at 30  mg  ml^-1 in 50  mM imidazole, 5  mM CaCl₂, 25  mM acarbose pH 7.5 mixed in a 1:1 ratio with well solution consisting of 0.1  M imidazole pH 7.5, 100  mM maltose, 5  mM CaCl₂ and with 18%(w/v) monomethylether polyethylene glycol 5K as precipitant. Data were collected at room temperature using an R-AXIS II image plate together with a Cu rotating anode and utilizing long focusing-mirror optics (Yale/Molecular Structure Corporation). 113° of data were collected with an oscillation range of 1° per image. Data were processed and reduced using the DENZO and SCALEPACK programs (Otwinowski & Minor, 1997). All further calculations used the CCP4 suite of programs (Collaborative Computational Project Number 4, 1994) unless otherwise stated.

2.2. Structure solution and refinement

The BHA-acarbose structure was solved by molecular replacement using AMoRe (Navaza & Saludjian, 1997) with the native BA2 amylase (PDB code 1e3x ) as the search model (Brzozowski et al., 2000). BA2 is a chimera consisting of residues 1-300 from a B. amyloliquefaciens -amylase and 301-483 from the corresponding B. licheniformis enzyme. 5% of the reflections were set aside for cross-validation (Brünger, 1992) and were used to monitor various refinement strategies such as geometric and temperature-factor restraint values, the insertion of solvent water and as the basis for maximum-likelihood refinement using the REFMAC program (Murshudov et al., 1997). As all observed data from 15  Å resolution were employed in the refinement, a low-resolution bulk-solvent mask correction was applied as implemented in REFMAC. Water molecules were added in an automated manner using ARP (Lamzin & Wilson, 1993) and verified by manual inspection. Coordinates and observed structure-factor amplitudes for the protein structure described in this paper have been deposited with the PDB. Model coordinates for the acarbose-derived nonasaccharide were generated in QUANTA (Accelrys, San Diego, USA).

3.1. Structure of the B. halmapalus BHA -amylase

Crystals of BHA belong to space group P2₁2₁2₁, with unit-cell parameters a = 48.2, b = 75.9, c = 155.2  Å, = = = 90°. There is a single molecule of BHA in the asymmetric unit. Structure solution with the known BA2 chimeric -amylase (Brzozowski et al., 2000), which displays 70% sequence identity to BHA, proceeded without problem. The final BHA model, at 2.1  Å resolution, has a crystallographic R value of 0.15, with a corresponding R_free of 0.20 using all observed data between 15 and 2.1  Å. A single residue, Tyr152, has main-chain / angles in a `forbidden' region of the Ramachandran plot, but the density (not shown) is unambiguous. Data and model-quality parameters are shown in Table 1.

Table 1 Data and structure-quality statistics for the B. halmapalus -amylase BHA at 2.1  Å resolution Values in parentheses are for the outer resolution shell. Data quality     Resolution (Å) 15-2.1 (2.2-2.1)   Rmerge 0.091 (0.337)   Mean I/(I) 13.6 (4.8)   Completeness (%) 99.7 (98.9)   Multiplicity 3.6 (3.5) Refinement     No. of protein atoms 4301   No. of metal ions 3 Ca2+, 1 Na+   No. of oligosaccharide atoms 389   No. of solvent waters 290   Resolution used in refinement (Å) 15-2.1   Rcryst 0.15   Rfree 0.20   R.m.s. deviation 1-2 bonds (Å) 0.017   R.m.s. deviation angles (°) 0.131   Average main-chain B (Å2) 18   Average side-chain B (Å2) 19   Average oligosaccharide B (Å2) 30   Average solvent B (Å2) 31

The three-dimensional structure of BHA (Fig. 2) is extremely similar to the closely related chimeric BA2 (Brzozowski et al., 2000), a feature that was utilized in the molecular-replacement structure determination. Both the overall structure and unusual Ca²⁺/Na⁺ binding sites are essentially identical to those of BA2. In both cases, Ca²⁺/Na⁺ identification was based upon electron-density levels (and consequent atomic B value; 8, 9, 13 and 8  Å² for the three Ca²⁺ and single Na⁺, respectively) and coordination geometry (heptavalent and hexavalent for the Ca²⁺ and tetravalent for Na⁺). In contrast to BA2, the initial 2F_obs - F_calc electron-density maps (and F_obs - F_calc difference density maps) clearly revealed a bound nonasaccharide species that must have been formed by a different transglycosylation route to that observed previously for related enzymes.

Figure 2 Three-dimensional structure of the Bacillus sp. BHA -amylase with bound ligand in ball-and-stick representation and metal ions shown as spheres (drawn with MOLSCRIPT; Kraulis, 1991). The acarbose-derived nonasaccharide binds in the -6 to +3 subsites.

3.2. Identity of the bound ligand

In the initial electron-density maps, it was clear that the pseudotetrasaccharide inhibitor acarbose (Fig. 1b) was bound as an apparent nonasaccharide species (Figs. 1c and 3). Acarbose in complexes of -amylases is indeed frequently observed as longer species. These extended species are most likely to arise from transglycosylation. Transglycosylation involves the formation of a covalent glycosyl-enzyme intermediate and its subsequent interception, not by water but by a further sugar molecule, resulting in extension (Fig. 1a.) In this way, longer oligosaccharides build up whose formation, although thermodynamically disfavoured, is kinetically driven. In the case of acarbose, such ligands become even better inhibitors as they span a greater number of subsites, eventually inhibiting the enzyme and yielding stable complexes for X-ray crystallography.

Figure 3 Observed electron density for the acarbose-derived nonasaccharide bound to the Bacillus sp. BHA -amylase at 2.1  Å resolution. The maps shown are a maximum-likelihood (Murshudov et al., 1997) and A (Read, 1986) weighted 2Fobs - Fcalc synthesis contoured at 0.3  e  Å3 (approximately 1) in red and an Fobs - Fcalc difference map (at 3) in which the O6 atoms were not included in the refinement or phase calculation in blue. The difference map clearly reveals density for O6 atoms in the -6, -4, -3, -2, -1, +1, +2, and +3 sites (see Fig. 1c for interpretation and comparison).

In BHA-acarbose the nonasaccharide spans the subsites -6 to +3. As with BA2 (Brzozowski et al., 2000), particular care was taken during the refinement of the acarbose-derived nonasaccharide in BHA to identify potential half-chair rings or 6-deoxy sugars. All O6 hydroxyls were omitted from refinement, no torsion-angle restraints were initially applied to the pyranosides and 2F_obs - F_calc electron-density maps were calculated using phases generated prior to the incorporation of the ligand. This refinement and unbiased electron-density maps reveal that glucose units in the -4, -3, -2, -1, +1, +2 and +3 subsites all possess their O6 hydroxyl substituent. The -5 sugar may well be a deoxy sugar and may be indicative of the acarbose moiety at the reducing end (that is demanded by all schemes for transglycosylation), whilst the -6 subsite is partially disordered, making assignment ambiguous (Fig. 1c). Of the well ordered sugars, only the -1 subsite sugar refines to a half-chair conformation when torsion-angle restraints are lifted.

The presence of glucopyranosides (and not 6-deoxy sugars) in the -4 to -1 subsites suggests that the ligand observed in the BHA structure was not formed by the same mechanism as that observed in the comparable BA2 complex with an acarbose-derived decasaccharide (see Fig. 1d for comparison). In BA2, the acarbose-derived decasaccharide was clearly formed by successive transglycosylations of acarbose, yielding a species in which every third sugar was a deoxy moiety and each of these was adjacent to an unsaturated species in half-chair conformation (Brzozowski et al., 2000). The BHA complex features `acarbose' in the -1 to +3 subsites, as expected, yet sugars in the -4 to -2 subsites all appear to be -1,4-linked D-glucopyranosides, thus displaying both a hydroxymethyl substituent and possessing <sup>4</sup>C<sub>1</sub> chair conformation. The observed nonasaccharide thus cannot have been made by successive transglycosylations of acarbose. Indeed, this species could only arise following the initial formation of a covalent intermediate from acarbose, followed by a limited number of successive transglycosylations with maltose (or contaminating longer maltooligosaccharides) from the crystallization liquor and a final `capping' with a second molecule of acarbose. Given that both the initiation of extension and its completion involve acarbose, the newly formed ligand has an acarbose moiety at both its reducing and non-reducing ends. Whilst such a scheme is unusual, BHA in a crystal will necessarily select for the transglycosylation species that binds most tightly and its observation is thus in keeping with extended species seen in other systems.