Engineering reaction and product specificity of cyclodextrin glycosyltransferase from Bacillus circulans strain 251

.


Introduction
Cyclodextrins are cyclic .(1Ú4) linked oligosaccharides consisting of mainly 6, 7 or 8 glucose residues, .-,orcyclodextrin, respectively. They find increasing use in industrial and research applications (Schmid, 1989;Allegre and Deratani, 1994;Pedersen et al. 1995), because of their ability to form inclusion complexes with many small hydrophobic molecules (Saenger, 1980). They are commonly produced from starch via an intramolecular transglycosylation reaction catalyzed by the enzyme cyclodextrin glycosyltransferase (CGTase,EC 2.4.1.19), in which a linear oligosaccharide (starch) chain is cleaved and the new reducing end sugar is transferred to the non-reducing end sugar of the same chain (cyclization). CGTase also catalyzes two intermolecular transglycosylation reactions; coupling, in which a cyclodextrin ring is cleaved and transferred to an acceptor maltooligosaccharide substrate, and disproportionation, in which a linear maltooligosaccharide is cleaved and the new reducing end sugar is transferred to an acceptor maltooligosaccharide substrate. In addition, the enzyme has a weak hydrolyzing activity (Penninga et al. 1995) (see Fig.1).
All known CGTases (Bender, 1986;Schmid, 1989) produce a mixture of cyclodextrins (and linear malto-oligosaccharides) when incubated with starch. The enzyme used in our studies, CGTase from Bacillus circulans strain 251, produces .-,andcyclodextrin in a ratio of 14 : 66 : 20. The isolation of pure cyclodextrins from this mixture requires a series of additional steps, including precipitation with organic solvents, which is a disadvantage for applications of cyclodextrins involving human consumption (Bender, 1986;Pedersen et al. 1995). A CGTase which only produces a single type of cyclodextrin is therefore of high industrial interest. In the past various site-directed mutations affecting the product specificity of CGTases have been made (Fujiwara et al. 1992a;Sin et al. 1994;Penninga et al. 1995). Most of these mutations were based upon residue 195 (B. circulans strain 251 CGTase numbering), an amino acid centrally located in the active site cleft (Tyr, or Phe, see Fig. 2). It was hypothesized that a starch chain entering the active site folds around it, and that thus the size of this residue is of prime importance for the enzyme's product specificity (Sin et al. 1994;Nakamura et al. 1994a). However, mutations of residue 195 only improved product specificity to a limited extent, if at all (Fujiwara et al. 1992a;Sin et al. 1994;Penninga et al. 1995). Even contradictory results, introduction of a smaller residue at position 195 resulting in larger cyclodextrin products, were obtained (Penninga et al. 1995).
In contrast, the X-ray structure of the CGTase from B. circulans strain 251 in complex with a maltononaose inhibitor (Strokopytov et al. 1996) suggested that sugar binding subsites further away from the catalytic site could be important for the enzyme's product specificity, and that it might be possible to change the ratio of the produced cyclodextrins by altering the affinities for glucose residues at these sugar binding subsites. As the amino acid residues contributing to the interactions with sugars bound at subsites -7 and -3 are variable among CGTases, these residues were chosen as targets for site directed mutagenesis. At subsite -7 residues 145, 146, and 147 all form hydrogen bonds with the inhibitor (Fig. 2). The S146P mutation was introduced to disturb this hydrogen-bonding network, aimed at reducing -cyclodextrin production. At subsite -3 the sugar residue is loosely positioned between the aromatic rings of Y195 and Y89. In CGTases from thermophilic organisms, which show an enhanced .-cyclodextrin production, the residue corresponding to Y89 is an aspartate (Knegtel et al. 1996;Wind et al. 1998). Therefore, mutation Y89D was constructed to specifically increase .-cyclodextrin production. Mutant Y89G was constructed to discriminate between the effects caused by removal of the tyrosine and those caused by the introduction of the aspartate.
Here we report the biochemical characterization of three single mutants (Y89G, Y89D, S146P) and one double mutant (Y89D/S146P) of B. circulans strain 251 CGTase, and the crystal structure of the Y89D/S146P mutant complexed with a maltohexaose inhibitor. The results confirm our hypothesis that interactions between enzyme and substrate at subsites remote from the catalytic site affect product specificity.

Bacterial strains and plasmids
Escherichia coli MC1061 [hsdR mcrB araD139 û (araABC-leu)7679 ûlacX74 galU galK rpsL thi] (Meissner et al. 1987) was used for recombinant DNA manipulations. CGTase (mutant) proteins were produced with the .-amylase and protease negative Bacillus subtilis strain DB104A [amy his nprR2 nprE18 aprA3] (Smith et al. 1988). The mutations were constructed in the expression vector pDP66K containing the cgt gene from B. circulans strain 251 (Penninga et al. 1996). In this work the subsites will be numbered according to the general subsite labeling scheme recently proposed for all glycosyl hydrolases (Davies et al. 1997), in which the glycosidic bond between -1 and +1 is the scissile bond, and the substrate reducing end is at position +2.

Growth conditions
Plasmid carrying bacterial strains were grown on LB medium in the presence of the antibiotics kanamycin and erythromycin, at concentrations of 100 and 5 µg/ml for E. coli and B. subtilis, respectively (Sambrook et al. 1989). When appropriate, agar plates contained 1 % starch to screen for halo formation. For the production of (mutant) CGTase proteins B. subtilis strain DB104A, containing the pDP66K expression vector (Penninga et al. 1996), was grown in a 5 l flask with 1 l medium containing 2 % trypton, 0.5 % yeast extract, 1 % sodium chloride and 1 % casamino acids (pH 7.0) with 10 µg/ml erythromycin and 5 µg/ml kanamycin.

DNA manipulations
Restriction endonucleases were purchased from Pharmacia LKB Biotechnology, Sweden, and used according to the manufacturer's instructions. DNA manipulations and calcium chloride transformation of E. coli strains were as described (Sambrook et al. 1989). Transformation of B. subtilis was performed according to Bron (Bron, 1990).

Site-directed mutagenesis
For site-directed mutagenesis the method based upon PCR reactions using VENT-DNA polymerase described by Penninga et al. (Penninga et al. 1996) was used. The following oligonucleotides were used to produce the mutations: Y89G, 5'-AGC ATC ATC AAT GGA TCC GGC GTA AAC AAC-3'; Y89D, 5'-GC ATC ATC AAT GAT TCC GGA GTA AAC AAC ACG GC-3'; S146P: 5'-G CCC GCC TCT CCG GAC CAG CCT TC-3'. Successful mutagenesis resulted in the appearance of the underlined restriction sites (BamHI for Y89G, and BspEI for Y89D and S146P), allowing rapid screening of potential mutants. After mutagenesis the PCR products and plasmid pDP66K were cut with restriction endonucleases PvuII and SalI, after which the PCR fragments were exchanged with the corresponding fragment from the plasmid, ligated and transformed to E. coli MC1061 cells, yielding mutation frequencies close to 70 %.

DNA sequencing
All mutations were confirmed by restriction analysis and DNA sequencing. DNA sequence determination was performed on supercoiled plasmid DNA using the dideoxy-chain termination method (Sanger and Coulson, 1975) and the T7-sequencing kit from Pharmacia-LKB Biotechnology, Sweden.

Production and purification of CGTase (mutant) proteins
After positive characterization, pDP66K DNA was transformed to B. subtilis strain DB104A. Production strains were grown to an optical density at 600 nm of 4.5 (for approx. 24 h). Under these conditions high extracellular CGTase levels were produced. Cultures were centrifuged at 4 o C for 30 min x 16,000 g and (mutant) CGTases present in the supernatants were further purified to homogeneity by affinity chromatography, using a 30 ml .cyclodextrin-Sepharose-6FF column (Pharmacia, Sweden) (Sundberg and Porath, 1974) with a maximal capacity of 3.5 mg protein per ml. After washing with 10 mM sodium acetate buffer (pH 5.5), bound CGTase was eluted with the same buffer containing 10 mg/ml .-cyclodextrin.
This allowed purification to homogeneity of up to 30 mg of stable (mutant) CGTase proteins, with a 5 fold purification and a yield close to 90 %, from 1 l culture supernatants.

Protein determination
Protein concentrations were determined with the Bradford method (Bradford, 1976) using the Bio-Rad reagent and bovine serum albumin as a standard (Bio-Rad Laboratories, Richmond, CA, USA).

Enzyme assays
All assays were performed in a 10 mM sodium citrate buffer (pH 6.0) at 50 o C. Hydrolyzing activities were measured as described earlier (Penninga et al. 1995) using 1% soluble starch (Lamers & Pleuger, Belgium) as substrate. Dinitrosalicylic acid was used to determine the number of reducing ends.
Cyclization activities were measured as described earlier (Penninga et al. 1996) using 5% 2DE maltodextrins (Paselli SA2; partially hydrolyzed potato starch, with an average degree of polymerization of 50; AVEBE, Foxhol, The Netherlands) as substrate.and -Cyclodextrins were measured spectrophotometrically (Penninga et al. 1995) using phenolphthalein (Vikmon, 1982), and bromocresol green (Kato and Horikoshi, 1984), respectively. These compounds form specific inclusion complexes with the cyclodextrins. For the measurement of .-cyclodextrin formation the following method was used: After incubation of appropriately diluted enzyme with 2DE maltodextrins a 200 µl sample was taken, added to 40 µl 1.2 N HCl on ice, incubated for 10 min at 60 0 C to inactivate the enzyme, and neutralized with 40 µl 1.2 N NaOH. These samples were subjected to HPLC analysis (see below). Thecyclodextrin concentration was determined as described above to serve as an internal standard for the HPLC analysis in order to calculate the concentration of .-cyclodextrin accurately.
Cyclodextrin formation from starch was also measured under industrial production process conditions (Hokse et al. 1981). For this purpose 1 ml of a 10 % pregelatinized starch solution (Paselli WA4; pregelatinized drum-dried potato starch, with a high degree of polymerization (>100,000); AVEBE, Foxhol, The Netherlands) in a 10 mM sodium citrate buffer (pH 6.0) was incubated with (mutant) CGTase (0.1 U -cyclization activity) at 50 o C for 45-50 h. Samples were taken at regular time intervals, boiled for 5 min, and the products formed were analyzed by HPLC using a 25 cm Econosphere-NH 2 5 micron column (Alltech Associates Inc. USA) eluted with acetonitrile/water (70/30 to 60/40, v/v) at a flow rate of 1 ml per min.
Coupling activities were determined using the method described by Nakamura et al. Disproportionation activities were determined using the method described by Nakamura et al. (1994b) was used with the modifications described by van der Veen et al.

Crystallization and soaking procedures
The mutant CGTase Y89D/S146P was crystallized at room temperature from 60% (v/v) 2-methyl-2,4-pentanediol (MPD) and 100 mM Hepes buffer 7.1 with a 5% (w/v) maltose solution added to the hanging drop. Space group and cell dimensions (see Table 1) were similar to that of wild type CGTase. The crystals were soaked using a procedure analogous to that which resulted in a maltononaose inhibitor bound to CGTase (Strokopytov et al. 1996). First, the crystal was brought in a buffer of 100 mM 3-(cyclohexylamino)-1-propanesulfonic acid (CAPS) pH 9.8 and 60% (v/v) MPD (mother liquor), after which it was soaked overnight in a solution of 0.25% (w/v) acarbose in mother liquor. Subsequently it was transferred to 0.5% (w/v) maltohexaose in mother liquor for 7 days.

Data collection and refinement
Data were collected to 2.4 Å in house. Refinement was done in a standard procedure (Strokopytov et al. 1996;Knegtel et al. 1995). In the course of refinement, the free R-factor stabilized at 30.7%. New structure factors were subsequently calculated according to the formula F diff = F obs -(F obs,nat -F calc,nat ), where F obs indicates the observed structure factors, F obs,nat the observed structure factors of the 2.0 Å wild type CGTase structure (Lawson et al. 1994), and F calc,nat the calculated structure factor amplitudes from that same structure. For a few cycles the structure was refined using these modified structure factors, a procedure known as difference refinement (Terwilliger, 1995). Care was taken to retain the original test set of reflections, and the difference free R-factor decreased from 23.3% to 22.5%. When reverting to the original F obs , a new free R-factor of 29.4% was obtained (Table I).
Electron density appeared for sugars at maltose-binding sites (MBS) 1, 2, and 3 (Lawson et al. 1994). Furthermore, density was observed for 6 sugar units in the active site, at subsites +2 to -4 ( Figure 3). We started refinement with hexakis 6-deoxy maltohexaose as a carbohydrate model. As a result, omit F obs -F calc density showed up at all glucose 6-OH positions except at subsite +1. This convinced us, that in this structure the 6-deoxy glucose group is bound at subsite +1, connected to the valienamine moiety at subsite -1 through an Nglycosidic bond (see Fig. 5). This is in agreement with the revised structures of CGTase complexed with acarbose and the acarbose derived maltononaose inhibitor (PDB entries 2CXG (acarbose) and 2DIJ (maltononaose) (Mosi et al. 1998)). Subsites +2, -2, -3, and -4 were filled with glucose groups, so that finally a modified acarbose was modeled, with a glucose missing at its reducing end, and a maltotriose added to its non-reducing end.
After completion of the refinement (see Table 1 for final details), the model was analyzed with PROCHECK (Laskowski et al. 1993) and WHATCHECK (Hooft et al. 1996

The cyclodextrin product ratio of the mutant enzymes under industrial production process conditions
This was analyzed with pregelatinized starch (Fig. 4, Table 2), which functions identically to jet-cooked starch commonly used in industry. The S146P mutant produced more .-cyclodextrin at the expense of -cyclodextrin, while -cyclodextrin production was not affected significantly. For mutant Y89D .-cyclodextrin production had slightly increased while bothand -cyclodextrin production slightly decreased. The changes in specificity for the single mutants were combined in the double mutant Y89D/S146P: the production of .cyclodextrin had doubled, while production of -cyclodextrin had decreased to 80 % of that of the wild type, and no significant effect on -cyclodextrin production was observed.

87
Mutation Y89G resulted in a slight increase of -cyclodextrin production at the expense of both .and -cyclodextrin. For all mutants the total conversion of starch into cyclodextrins remained unaffected (39-40%), with no significant increase in the formation of linear maltooligosaccharides (< 1%). The data thus agree with our hypothesis that interactions between CGTase and substrate at subsites -3 and -7 contribute to the enzyme's product specificity. Under the conditions used above not only the formation of cyclodextrins via the cyclization reaction but also their degradation via the coupling reaction, and the formation of linear products via the hydrolyzing and disproportionation reactions, are important. Each of these reactions was analyzed in detail to elucidate its contribution to the changes in product ratio of the mutant proteins and to further delineate the mechanistic basis of these changes.

Cyclization activities
Cyclization activities of wild-type and mutant CGTase proteins were determined using 2DE maltodextrins, a starch derivative with a lower degree of polymerization than the above used pregelatinized starch. The k cat values of wild-type and mutant CGTase proteins are shown in Table 3. Substrate affinities could not be measured accurately, because the K M values were well below 0.05 % 2DE maltodextrins, and activity measurements at these low substrate concentrations are not reliable (van der Veen et al. 2000c). Mutations Y89G and Y89D caused increased k cat values for cyclization, except for .-cyclodextrin formation by Y89G, where a decrease was observed. The Y89D mutation, however, resulted in a doubling of the .cyclodextrin forming activity. Mutation S146P reduced the k cat value for -cyclization by a factor 2.5, with a concomitant increase of the k cat value for .-cyclization by a factor 2.3. The k cat value for -cyclodextrin production was slightly decreased by this mutation. In the double mutant Y89D/S146P the characteristics of the single mutants are combined, with most significant effects on k cat values for .and -cyclization (tripling and twofold reduction, respectively). The results show that changes in the cyclodextrin product ratio can be related to the altered k cat values for .-,andcyclization. The changes in the initial formation of cyclodextrins (k cat values), however, are more pronounced than those obtained during 45-50 h incubation periods (see Tables 2 and 3). The other CGTase catalyzed reactions thus interfere with the production of cyclodextrins, reaching an equilibrium situation over longer time periods. Table 3. Cyclization activities of wild-type and mutant cyclodextrin glycosyltransferases from B. circulans strain 251.

The coupling reaction
Degradation of cyclodextrins via the coupling reaction may have a significant effect on cyclodextrin yields under industrial cyclodextrin production conditions. The coupling activities of wild-type and mutant CGTase proteins were analyzed using cyclodextrins as donor substrates and methyl-.-D-glucopyranoside as an acceptor substrate. The mutations Y89G and Y89D at subsite -3 both resulted in 3-4 fold decreased affinities for cyclodextrins (Table 4), indicating that some of the interactions with these cyclic substrates were lost, without being compensated by other interactions. Mutation Y89D resulted in a general increase in the k cat of the coupling reaction, whereas mutation Y89G resulted in a general decrease in the coupling k cat . Mutation S146P had no significant effect on the affinity for cyclodextrins, which indicates that subsite -7 is not involved in binding of cyclodextrins. Mutation S146P, however, clearly had a negative effect on the k cat values, especially for -cyclodextrin coupling.

Hydrolysis
The CGTase hydrolyzing activity may result in formation of linear products from starch under industrial cyclodextrin production conditions. The low hydrolyzing activity of wild-type B. circulans strain 251 CGTase was even further decreased for mutant Y89G, while mutant Y89D showed a slight increase (Table 5). Mutants affected in subsite -7 (S146P and the double mutant Y89D/S146P) showed a doubling in the hydrolyzing activity (Table 5).

The disproportionation reaction
In the assay for the disproportionation reaction 4-nitrophenyl-.-D-maltoheptaoside-4-6-O-ethylidene (EPS) was used as a donor substrate and maltose as an acceptor. The k cat values for disproportionation (Table 5) were at least a factor three higher than the values for cyclization and coupling (Tables 3 and 4). Mutations Y89G and Y89D at subsite -3 both resulted in 3 fold decreased affinities for EPS (3 fold increased K M values, Table 5), indicating that interactions with the substrate were lost and that replacement of Tyr89 with Asp or Gly apparently did not result in new interactions. Both mutations resulted in increased k cat values, the most pronounced being the one for mutant Y89D. The S146P mutation had very little effect on the disproportionation activity (k cat and K M ), whether it was made in the wild type (mutant S146P) or in the Y89D mutant (double mutant Y89D/S146P). This indicates that Y89, but not S146, is important for the reaction of the enzyme with the EPS substrate. Evidently, this maltoheptaose substrate does not reach beyond subsite -6. The crystal structure of the Y89D/S146P mutant with a bound maltohexaose inhibitor In .-amylases and CGTases it has frequently been observed that acarbose molecules undergo transglycosylation reactions resulting in oligosaccharide inhibitors (Strokopytov et al. 1996;Wind et al. 1998;Brzozowski and Davies, 1997). Presumably the chain with highest affinity subsequently dominates the electron density map. Whereas the soaking experiment of Strokopytov et al. (Strokopytov et al. 1996) with wild type CGTase resulted in the binding of a "straight" maltononaose inhibitor, repeating the same procedure with crystals of the Y89D/S146P mutant resulted in a shorter "bent" maltohexaose inhibitor complex, bound from subsite +2 to -4. In both structures the sugar conformations at subsites +2 to -2 are very similar, with the glucose residues at subsites -1 and +1 linked by an N-glycosidic bond, which is elongated (1.75 Å, while we refined against an ideal value of 1.45 Å). This is not necessarily functionally significant, but might be a result of multiple sugar conformations contributing to the observed electron density (Brzozowski and Davies, 1997); this was not further investigated due to the limited resolution (2.4 Å) of our structure. Sugar binding at subsites -3 and -4 is, however, totally different when comparing the wild type (Fig. 5(a)) with the mutant enzyme ( Fig. 5(b)). At subsite -3 the glucose of the maltohexaose inhibitor (overall B factor 44 Å 2 ) occupies a position more bent towards Tyr195 (Fig. 5(b)). In this orientation the glucose O2 atom binds to the Asp371 O/2 atom and to the glucose O3 atom at subsite -2. The glucose O6 atom binds to Asp196 O/1 via a water mediated contact (Fig. 5(b), Table 6), whereas in the maltononaose conformation the glucose O6 atom directly binds to Asp196 O/1 (Fig. 5(a)), with the glucose O6 atom occupying the position of the water molecule that mediates the hydrogen bond in the maltohexaose structure.
In both conformations, the glucose at subsite -4 is only weakly bound. In the maltohexaose inhibitor the sugar is pointing towards the acceptor site +2, (Fig. 5(b)). The glucose at the maltohexaose subsite -4 is not involved in hydrogen bonding interactions, but stabilization might come from hydrophobic interactions with Tyr195, which is illustrated by the fact that the sugar replaces at least 3 ordered waters, present in the CGTase maltononaose inhibitor structure (Strokopytov et al. 1996). The bent conformation in CGTase mutant Y89D/S146P; a water molecule is now mediating the glucose O6 and Asp196 interaction. The figure was constructed using MOLSCRIPT (Kraulis, 1991).

Cyclodextrin product specificity of CGTase
The products formed by CGTase over longer periods of incubation (the net effect of all four CGTase catalyzed reactions with starch, after reaching equilibrium) provides relevant information about the performance of the wild type and mutant CGTase proteins under conditions resembling those of industrial production process conditions (see Fig. 4 and Table  2). Although initially the ratio of the produced cyclodextrins will reflect the preference of the enzyme for the formation of the specific cyclodextrins (see Table 3), any CGTase capable of forming (cyclization) and degrading (coupling) .-, -, and -cyclodextrin will eventually produce a fixed ratio (27:58:15 (Tewari et al. 1997)) of these compounds determined by their thermodynamic equilibrium. This is clearly visible when comparing the ratios expected from the initial cyclization reaction rates (Table 3) with those obtained after prolonged incubation (Table 2). CGTase product specificity must therefore be considered as a kinetic feature, in which initial substrate binding and the rate of circularization (the conformational change from a linear substrate to a cyclic product) have been proposed to be important factors (van der Veen et al. 2000c). Figure 6. An overview of the interactional changes near sugar subsite -7. The maltononaose sugar residues at subsites -6 and -7 are indicated. Enzyme residues in light grey: the conformation in wild-type CGTase, in dark grey: CGTase mutant Y89D/S146P. The figure was constructed using MOLSCRIPT (Kraulis, 1991).

Binding subsites distant from the catalytic site affect product specificity
Since the same substrate is used for the simultaneous production of the various cyclodextrins, .-, -, and -cyclization must be regarded as competing activities (van der Veen et al. 2000c). Conceivably, initial substrate binding (up to subsite -6, -7, or -8) determines which cyclodextrin will be formed. From the structure of the CGTase from B. circulans strain 93 251 complexed with a linear maltononaose inhibitor in the active site the loop region containing residues 145-151 was found to be intimately involved in hydrogen bonding interactions with the glucose residue bound at subsite -7 (see Fig. 2 and Fig. 6). This region was therefore suggested to be involved in the preference of this CGTase for the formation of -cyclodextrin (Strokopytov et al. 1996). Indeed, a deletion mutant (û(145-151)D) of the -CGTase from B. circulans strain 8 in which residues 145-151 are replaced by a single aspartate, resulted in a shift towards -cyclodextrin production (Parsiegla et al. 1998). The S146P mutant described here was aimed at disturbing the hydrogen bonding network at subsite -7 without damaging the substrate binding cleft. The largely unaffected disproportionation kinetics indicate that in this mutant the catalytic machinery was not affected. The mutation did specifically affect the cyclization and coupling reactions involving -cyclodextrin (containing 7 glucose residues). The increased .-cyclization activity can be explained by the fact that the introduced proline destabilizes the maltononaose binding mode at subsite -7 by stereochemical clashes (the distance of glucose O2 to Pro146 C/ would be 1.6 Å, see Fig. 6), promoting substrate binding upto subsite -6. These findings confirm the importance of the loop region containing residues 145-151 in initial substrate binding, and thus in determining which cyclodextrin will be formed. Subtle modifications in this loop enable us to tailor CGTase product specificity.

The effects of Tyr89 mutations on oligosaccharide processing
Removal of the hydrophobicity of residue 89 generally leads to increased activities as shown by mutants Y89D and Y89G in this paper (Tables 3-5), and also by mutant Y89S of the CGTase from alkalophilic Bacillus sp. I-5 (Kim et al. 1997). The increased K M values and decreased k cat /K M values in the coupling and disproportionation reactions indicate that the energy levels of both bound substrate (related to K M ) and transition state (inversely related to k cat /K M ) have increased. The increased k cat values show that the activation energy has decreased compared to the wild type, indicating that the transition state is less affected by these mutations than the bound substrate ground state. Furthermore, Y89D results in a higher increase in reaction rates when shorter oligosaccharides are processed (.-cyclization, disproportionation, hydrolysis) than Y89G, indicating a stabilization of the transition state of these reactions in mutant Y89D when compared to mutant Y89G. Interestingly, a soaking experiment with crystals of mutant Y89G identical to that with the double mutant Y89D/S146P resulted in a bound maltopentaose inhibitor in a conformation similar to the maltohexaose, however, with bad electron density for the sugar bound at subsite -3 (J. C. M. Uitdehaag, unpublished result). The maltohexaose bound in mutant Y89D/S146P shows much better density at subsite -3, which indicates that the "bent" conformation of the bound oligosaccharide is stabilized by Asp89 at subsite -3. Probably the water mediated hydrogen bond between Asp196 and the bound sugar (see Fig. 5(b)) is promoted by adding hydrophilicity at subsite -3. The biochemical and structural data shows that a (more) increased catalytic activity is combined with (stabilization of) a "bent" intermediate. This suggests that in the transition state a "bent" conformation of the oligosaccharide is preferred. The observation that the Y89D mutation specifically stimulates processing of shorter oligosaccharides (.-cyclization, disproportionation) can be explained by the fact that longer oligosaccharides have a natural tendency to adapt a (bent) helical conformation. This also provides an explanation for the sometimes contradictory results obtained with mutations in residue 195. This aromatic residue (Tyr or Phe) is important for cyclization (Fujiwara et al. 1992a;Sin et al. 1994;Penninga et al. 1995;Parsiegla et al. 1998;Wind et al. 1998) and has recently been shown to stabilize an intermediary (bent) stage in circularization (Uitdehaag et al. 1999a). Mutations in residue 195 will probably destabilize the intermediary stage, and consequently lead to increased production of larger cyclodextrins, irrespective of the size of the introduced residue (e.g. Trp or Leu both lead to enhanced -cyclodextrin production (Penninga et al. 1995)). The results obtained with the Y89 mutants thus show that besides initial substrate binding also the consecutive circularization rate of the intermediate determines product specificity of the CGTase catalyzed cyclization reaction. Furthermore, the high hydrolysis and .-cyclization activity of the CGTase from Thermoanaerobacterium thermosulfurigenes EM1 can at least partially be related to the presence of Asp89 in this enzyme (Wind et al. 1995).
The kinetic behaviour of mutant Y89D/S146P (Tables 3-5) is largely the sum of the single mutants. This additive behaviour can be explained by the fact that: i) both residues are well separated from each other, as can be seen in the structure; ii) different aspects of product specificity are affected by the mutations, as shown by the biochemical characteristics. As expected this double mutant shows the highest increase in .-cyclodextrin production, with a decrease in especially -cyclodextrin production.

Conclusions
The ratio of the products formed in the cyclization reaction largely depends on hydrogen bonding interactions at specific subsites in the CGTase active site, as shown here for subsite -7 (S146), determining how far the linear substrate will enter the active site. However, also the subsequent conformational change from linear substrate to cyclic product affects this ratio (Y89). Cyclodextrin product specificity of CGTases is thus largely determined by initial substrate binding and the consecutive circularization rate of the intermediate. Our combination of biochemical and biophysical techniques has provided new insights into the cyclization mechanism of CGTase, which pave the way for further improvement of CGTase product specificity by rational protein engineering. Mutational, kinetic and crystallographic studies which take advantage of these new insights, are currently being undertaken.