Small Angle X-ray Studies Reveal That Aspergillus niger Glucoamylase Has a Defined Extended Conformation and Can Form Dimers in Solution*

The industrially important glucoamylase 1 is an exo-acting glycosidase with substrate preference for α-1,4 and α-1,6 linkages at non-reducing ends of starch. It consists of a starch binding and a catalytic domain interspersed by a highly glycosylated polypeptide linker. The linker function is poorly understood and structurally undescribed, and data regarding domain organization and intramolecular functional cooperativity are conflicting or non-comprehensive. Here, we report a combined small angle x-ray scattering and calorimetry study of Aspergillus niger glucoamylase 1, glucoamylase 2, which lacks a starch binding domain, and an engineered low-glycosylated variant of glucoamylase 1 with a short linker. Low resolution solution structures show that the linker adopts a compact structure rendering a well defined extended overall conformation to glucoamylase. We demonstrate that binding of a short heterobidentate inhibitor simultaneously directed toward the catalytic and starch binding domains causes dimerization of glucoamylase and not, as suggested previously, an intramolecular conformational rearrangement mediated by linker flexibility. Our results suggest that glucoamylase functions via transient dimer formation during hydrolysis of insoluble substrates and address the question of the cooperative effect of starch binding and hydrolysis.

dase that cleaves ␣-1,4 and, less efficiently, ␣-1,6 linkages at non-reducing ends of starch and related oligo-and polysaccharides (1,2). GA is industrially important in production of bioethanol, glucose, and fructose syrups. The glucoamylase 1 form (GA1) from Aspergillus niger has an N-terminal catalytic domain (CD) and a C-terminal starch binding domain (SBD) connected by a 69-amino acid-long linker that is decorated by short, predominantly mannose-containing O-glycosylations corresponding to a minimum content of 63 mol of hexose attached to about 32 serines and threonines (2-4) (see the schematic in Fig. 1). The functional role of the linker region is not fully understood. The isolated CD and the GA2 form ( Fig. 1), which includes the linker and lacks the SBD, are both able to hydrolyze soluble substrates, but not starch granules, in contrast to GA1 (4 -6). Moreover, it has been shown that the addition of free SBD to a mixture of GA2 and starch granules increases the rate of hydrolysis (7). The SBD is, therefore, suggested to enhance the substrate accessibility by disentangling ␣-glucan helices on the surface of the starch granule rather than to act as an intramolecular guide directing the substrate chain to the active site pocket of the CD (7,8). Conformational constraints on the linker are not required in this function and engineered linker variants, low-glycosylated or shortened ones inclusive, behave very similarly to wild-type GA1. Thus, the specific sequence of the linker seems to have a modest if any effect on the action of GA1 (9). This is much in contrast to cellulase, another glycoside hydrolase with comparable domain architecture, where the length of the linker region does affect the catalytic activity (10).
High resolution structures are available of the individual domains, the CD (11), and the SBD (12), but structural information on full-length GA1 or the linker region is very limited since the highly glycosylated linker (2,4,13) renders analysis of GA1 or either of the GA domains while attached to the linkerregion inherently difficult. Also, the dynamics of the functional cooperation between the domains are not well understood, and the overall tertiary structure of GA is a matter of debate. Scan-ning tunneling microscopy on the one hand indicates the presence of an extended linker conformation with a distance between the CD and SBD domain centers of about 9.5 nm (14), i.e. an elongated overall conformation. On the other hand, the crystal structure of the CD shows its C-terminal 31-amino acidlong segment, which constitutes the most N-terminal part of the linker region in GA1, to wrap around the globular domain in a well defined manner (11). If this conformation recurs in the solution structure of full-length GA1, a distance of 9.5 nm between the CD and the SBD domain centers would require the remaining part of the linker to be essentially fully extended.
Further adding to the debate, in an attempt to estimate the distance between the CD and SBD sugar binding sites, synthetic heterobidentate GA inhibitors consisting of acarbose coupled either directly with ␤-cyclodextrin or via ethylene glycol linkers with varying lengths decreased the hydrodynamic radius of GA1 by complexation (15). The results suggest that even the shortest of the heterobidentate inhibitors (referred to as L0) induced a more compact GA1 molecule than the free GA1 by drawing the two domains closer to each other consistent with the linker region conferring flexibility. Isothermal titration calorimetry (ITC) studies further suggested a 1:1 complex of the heterobidentate inhibitor and GA1 (16). Although neither ITC nor dynamic light scattering data resolve the distance between the CD and SBD, it was suggested that the glycosylated linker is flexible and that the CD and SBD get in close spatial proximity when binding heterobidentate ligands.
Here we report the use of small-angle x-ray scattering (SAXS) to analyze different GA forms to unveil the solution architecture of these enzymes and, more specifically, to investigate the structural role of the glycosylated linker. SAXS is a well established method for studying the low resolution structures of particles on the length scale of ϳ1-100 nm. Recent advances in synchrotron facilities and data analysis (17,18) have greatly increased the use of SAXS in the more demanding analysis of weakly scattering biological systems in solution. We present joint SAXS and calorimetry analysis of GA1, GA2, a low-glycosylated GA1 (dgGA) as well as complexes with the heterobidentate inhibitor L0 of GA1 and dgGA, respectively. Our analysis shows that the natural linker region forms a compact structural entity in GA1 and GA2. This linker has previously been shown to confer structural stability to the SBD in GA1 (19), and this effect is even more pronounced in dgGA. The apparent rigid overall conformation of GA is incompatible with the notion that the CD and SBD form contacts within a single protein molecule in the absence or presence of heterobidentate ligands. Instead we show that heterobidentate ligand binding induces dimerization. Hence, these findings, motivate reconsideration of earlier interpretations of the highly O-glycosylated GA1 linker as being flexible (9,15,16) and suggest that GA molecules work in pairs on starch granules.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-The low-glycosylated linker-variant dgGA was subcloned from the GA1 A. niger construct using a single primer protocol for site-directed mutagenesis with the primer 5Ј-TCACAGCCACCGCGGTG-GGAGTGGTACAGGCAGCAGCGTACGGAGCCGGAGC-CGGGTACGGAGCCGGAGCCGGCACGTTCGGCCACG-AGGTGACAGTCA (from DNA technology A/S, Aarhus, DK), resulting in a linker region with the amino acid sequence TVTSWPNVPAPAPYPAPAPYAAACTTPTAVAV. GA1 and dgGA were expressed and purified essentially as previously described (9). GA2 was obtained as described (5). All proteins were concentrated using centrifugation concentrators (Amicon, cut off 10 kDa).
Small-angle X-ray Scattering-The synchrotron radiation SAXS data were collected on the X33 SAXS camera of the European Molecular Biology Laboratory on the storage ring DORIS III (DESY, Hamburg, Germany (20)). The detector was an image plate with online readout (MAR345), and the range of momentum transfer was s ⑀ {0.1 Ͻ s Ͻ 4.5 nm Ϫ1 } (s ϭ 4 sin()/, where is the scattering angle, and ϭ 0.1504 nm is the x-ray wavelength). Exposure time was 3 min, and all data were normalized to the intensity of the incident beam and corrected for the detector response. All initial data reduction, including subtraction of the scattering from the buffer, was made using the program PRIMUS (21). Data were collected at 2-8 mg/ml in 50 mM sodium acetate, pH 4.5, and the concentration-scaled data curves were merged and extrapolated to infinite dilution. The forward scattering I 0 and the radius of gyration R g were evaluated using Guinier approximations and compared with corresponding estimates using the indirect transform package GNOM (22,23) that estimates the pairwise distance distribution function P(r) of the particle by indirect Fourier transformations (IFT). The molecular masses (MMs) were calculated from the I 0 values, calibrated with a bovine serum albumin standard. The D max within the scatterer was computed via IFT using GNOM (22,23) and compared with values employing the program BIFT (24) using standard parameters, screening ␣-values from 10 Ϫ4 to 10 Ϫ1 . The BIFT program also calculates the evidence for P(r) functions with a given maximal distance within the scatterer (Fig. 4) and, thus, provides an objective measure for D max . , and dgGA. The linker variant was made by mutating part of the GA1 linker to the sequence, specified in the inset. CD, red; SBD, blue; the region of the linker, which is similar for all three constructs, is green, the mutated region is in pink on dgGA, and the corresponding region in wild-type (wt) GA1 and GA2 is in blue. The sequence is shown in corresponding colors, and asterisks denote glycosylation sites.
Ab initio Modeling-Low resolution models were built by the program DAMMIF, 4 a revised version of DAMMIN (25) that employs a simulated annealing protocol to a representation of the protein consisting of an assembly of dummy atoms inside a search volume, defined by a sphere of the diameter D max . Additional restraints on the resulting model are connectivity and compactness. The models were obtained by calculating 20 individual models fitted to the low angle half of the scattering curve (implying only low resolution features). These models were averaged using the programs DAMAVER and SUBCOMP (26,27). The resulting averaged and filtered model was submitted to a refinement round implying 100% of the data curves, again calculating 20 individual models, which were finally averaged to the represented model (see Figs. 3 and 5). Quasi-atomic modeling was performed using the program BUNCH (28), which can be used to model multidomain proteins. The program calculates the best fit to the experimental curve using a simulated annealing protocol for obtaining optimal orientations of individual domains. These domains are either described by calculated scattering from available high resolution structures of individual domains or by so-called dummy atom residues, where beads, each representing the average scattering from one amino acid, are allowed to move until the best fit is achieved. In particular cases such as the current where e.g. glucose moieties are present, these are also represented by dummy atom residues corresponding to the molecular weight. Amplitudes representing theoretical scattering from the CD and SBD were calculated using CRYSOL (29). PREBUNCH was used to generate a starting model. To simulate scattering from the linker, a total of 153 dummy alanines represented the 63-amino acid linker and 63 molar eq of hexose, corresponding to an MM of 84 additional amino acids. To select for a compact conformation, penalty weights were increased to 50 against extended loops and angles, and dummy residue form-factors were doubled to account for hydration.
Separation of Data Contribution from dgGA Dimers-The fraction of monomer and dimer in dgGA:L0 solutions was estimated from the extrapolated I 0 by comparison to the I 0 expected from different mixtures of monomer and dimer (supplemental data). The fraction of monomer was subtracted from the experimental data, and the resulting dimer curve correspondingly showed no presence of monomers using Bayesian IFT (24) (data not shown).
Differential Scanning Calorimetry-Enzyme samples (15 M) dialyzed against 50 mM sodium phosphate, pH 7.5, were degassed under vacuum before heating in a VP differential scanning calorimetry (MicroCal, Inc.) with a cell volume of 0.52061 ml and a scan rate of 60°C/h with 85°C set as the highest temperature. An external pressure of 3 bar was applied over the cells. Buffer scans were subtracted from sample scans, and the molar heat capacity was obtained by normalizing using the known protein concentration and cell volume of the calorimeter.

RESULTS
Solution Analysis of GA1 and GA2-SAXS data were collected from GA1 and GA2 at pH 4.5 ( Fig. 2A), the pH optimum of GA function (30). The SAXS data reveal the overall biophysical parameters from the scattering molecules in solution (see Table 1 for an overview). The MMs were estimated using the Guinier approximation of the extrapolated scattering intensities at zero angle (I 0 ). The resulting MMs were 10% lower (73 kDa) for GA1 than the expected MM of 82 kDa, whereas the MM of GA2 was 70 kDa, in excellent agreement with the theoretical value (assuming glycosylation with two sugar units per site (see Fig. 1)). IFTs provide the pairwise distance distribution functions (Fig. 2B). Estimated maximum distances (D max ) were 16.5 and 10.5 nm for GA1 and GA2, respectively, and the corresponding radii of gyrations (R g ) were 3.7 and 2.7 nm. The distribution function for GA1 indicates that this molecule has an elongated shape with two characteristic distances appearing with a higher probability, visible as two peaks (open triangles in Fig. 2B). The first distance corresponds to the average of the most common distances in each of the two domains, and the larger distance is the distance between centers of masses for the domains. This distribution is characteristic for two domains with a defined relative position, i.e. a rigid molecule, and not, as previously anticipated, a molecule composed of two domains separated by a flexible linker.
Low resolution models calculated using the ab initio modeling program DAMMIF, 4 a modified version of DAMMIN (25), emphasize the existence of such a shape of GA1 in solution (Fig. 3). Models of GA1 and GA2, thus, comprise a rather large globular domain of an overall size and shape in accordance with the isolated CD. GA2 displays a distinct extension from this domain that is ascribed to the highly O-glycosylated linker in a defined conformation. A similar extension is apparent for GA1 in addition to a distal globular domain similar in shape to that of the SBD.
A quasi-atomic model of GA1 has been obtained using BUNCH (28); see Fig. 3. BUNCH applies a simulated annealing protocol to optimize the orientations and positions of individual domains of multidomain proteins guided by SAXS data. Available high resolution models for individual domains were applied, whereas modeling of the linker region was performed using beads as dummy atom residues. Each dummy atom residue is represented by average scattering amplitudes from average amino acids, in this case alanines. As input, the atomic coordinates from the crystal structure of the CD (11) and NMR structures of the SBD (12) have been given, and an MM of the linker was calculated from 69 amino acids and 63 molar eq of hexose, i.e. a total of 153 alanines. In Fig. 3 the linker region is shown as spheres, whereas atomic resolution representations are shown of the CD and SBD. It should be stressed that the resolution of the SAXS data does not allow resolving the relative orientation of these domains. However, the extended overall conformation is confirmed by this model as well as the presence of a compact linker region. In conclusion, SAXS-based solution analysis of GA1 and GA2 suggests that the glycosylated linker region forms a defined domain, interspersing the CD and SBD, resulting in an overall rigid elongated conformation of GA1 in solution.

The Heterobidentate Inhibitor Evokes Dimerization of GA1-
To further investigate the structural rigidity of the glycosylated linker, the heterobidentate inhibitor L0 was added to GA1. It has been suggested that L0 binds simultaneously to the CD and SBD in one GA1 molecule, forming a 1:1 complex, thereby bringing the two domains spatially close together (9,16). This domain repositioning requires that the glycosylated linker is flexible. Remarkably, analysis of SAXS data from the mixture reveals that GA1 in the presence of L0 has maximum internal distances of 21 nm, i.e. substantially larger than that of the isolated GA1. This is in contradiction to an intramolecular binding of L0 to both domains simultaneously. Dynamic light scattering data show the same tendency (supplemental data). The calculated average R g increases slightly from 3.7 to 3.9 nm ( Table 1). The MMs calculated from the extrapolated forward scattering (I 0 ) of the solute indicated a clear increase in the average MMs in solution to 94 kDa. These results can only be interpreted in terms of partial oligomerization in solution. The observed maximum distance and R g are compatible with the formation of compact dimers. Based on the calculated average MM, a mixture of 81% monomers and 19% dimers is estimated (supplemental data). The coexistence of two species in solution is also supported by using Bayesian IFT (24,31), which further suggests the existence of two species in solution, the smaller with a maximal distance approximately corresponding to GA1 monomers and the larger in accordance with the 21-nm distance (Fig. 4A (31)). This is seen in particular from the calculated evidence (Fig. 4), where two populations of different size are indicated. It is concluded that addition of L0 induces partial dimerization of GA1 which may arise by intermolecular binding of inhibitor molecules to CD and SBD in different GA1 molecules in a compact complex. There is no observation of shorter molecular species, which would be expected if binding of L0 occurred to CD and SBD in the same GA1 molecule (16). The Low-glycosylated Linker Variant Is Less Thermostable Than Native GA1-For further investigation of the structural importance of the glycosylated linker, we analyzed the solution properties of the low-glycosylated linker variant dgGA (Fig. 1). Analysis of the substrate binding abilities of the linker variant by ITC confirmed glucoamylase-like behavior of dgGA (supplemental data). The stability of GA1 and dgGA was compared using differential scanning calorimetry, which showed a 4°C lower denaturation temperature for dgGA than for GA1 (T m ϭ 63.3°C; Table 2). Differential scanning calorimetry analysis of GA1 previously indicated that the CD unfolds irreversibly, whereas the SBD unfolds reversibly (19,32,33). Successive scans revealed a denaturation temperature T m2 of the SBD of GA1 of 57.7°C, whereas the SBD from dgGA gave a T m2 of 50.3°C, clearly reflecting a substantially less stable SBD in dgGA, which furthermore is evident in the reduced unfolding enthalpy of the SBD in dgGA compared with GA1 (Table 2). Reversible unfolding can be characterized by the calorimetric enthalpy change of unfolding (⌬H) determined as the area of the transition peak in the differential scanning calorimetry thermogram, and the van't Hoff enthalpy change (⌬H v ), determined from the peak shapes under the assumption of a twostate unfolding mechanism. If the unfolding follows a two-state mechanism, then ⌬H v ϭ ⌬H (34), but because ⌬H v Ͼ ⌬H for the SBD, the domains unfold according to a multistate mechanism (Table 2). In conclusion, the SBD in dgGA is less thermostable than the SBD in GA1, suggesting that the glycosylated linker in native GA1 stabilizes the fold of the SBD, assumingly via direct contacts.
Indirect Proof for the Structural Integrity of the Glycosylated Linker-SAXS data from free dgGA corroborates the above findings (Fig. 2). The mutant dgGA has a calculated MM of 62 kDa, which is 11% lower than the theoretical MM of 70 kDa. Compared with native GA1, dgGA has a significantly decreased maximal size, as the estimated R g and D max values of dgGA are 3.0 and 12.5 nm, respectively. The pairwise distance distribution function shows that the larger of the two typically occurring distances is around 8 nm in dgGA, whereas it is around 10 nm for native GA1 (Fig. 2B). This suggests that the CD and SBD are closer together in solution in the dgGA molecule. Low resolution models calculated using DAMMIF (22) emphasize this feature (Fig. 3). Volumes corresponding in size and shape to the CD and SBD are visible. Compared with the DAMMIF and BUNCH (28) models of GA1, the region dispersing these domains in dgGA is significantly shorter in accordance with the non-native linker having 24 amino acids fewer and much less glycosylation. Because the most evident difference between GA1 and dgGA is the reduced volume interspersing the CD and SBD, the solution structure of dgGA serves as indirect proof of the presence of a structurally well defined glycosylated linker in GA1.
Low Resolution ab initio Models of dgGA Dimers Suggest a Head-to-tail Dimerization-A mixture of dgGA and L0 was analyzed using SAXS, for comparison with the data from GA1:L0.
Upon the addition of L0 to dgGA, the average MM increased from 62 to 90 kDa, reflecting a more pronounced degree of dimerization than observed for native GA1 ( Table 1). The cor- Comparison of (from top to bottom) quasi-atomic model of GA1 (gold) and ab initio models of GA1 (gold), dgGA (rose), and GA2 (red). The quasi-atomic model depicts the CD and SBD in schematic representation, and the linker is modeled with spheres. Surface representations of ab initio models are used. Models are made using the programs BUNCH (28) and DAMMIF (25), respectively, and depicted using PYMOL (41).

TABLE 1 Biophysical parameters estimated by SAXS
GA1, wild-type GA1; dgGA, low-glycosylated linker variant of GA1; GA1:L0 and dgGA:L0, wild-type and low-glycosylated linker variant, respectively, of glucoamylase 1 with heterobidentate inhibitor L0. MM Calc , theoretical molecular mass calculated from amino acid sequence, assuming each glycosylation site carries two sugar moieties and that complex formation takes place between one monomer of glucoamylase with one heterobidentate inhibitor (1:1 complex). MM Exp , experimentally based molecular mass calculated from the extrapolated intensity of scattering at zero angles (I 0 ) using IFT/Guinier plots, respectively. D max represents the maximal pairwise distance within the solutes, estimated using the programs GNOM (22,23) and BIFT (24); two values are interpreted as a mixture of monomers and dimers. R g , estimated from the Guinier approximation using the program AutoR g . V, volume of solutes, calculated from ab initio modeling. responding calculated distribution of dgGA is 71% monomer and 29% dimer (supplemental data). The average R g increased from 3.0 to 4.0 nm, whereas the D max is estimated to 17.0 nm. Bayesian IFT (24,31) analysis of dgGA:L0 solutes suggests the existence of two species in solution, the smaller with an estimated D max of a size comparable with monomeric dgGA (Fig. 4B). Thus, a significant proportion of dimers is present in solution, and the partial scattering from the dimers could be isolated by subtracting scattering from monomers, according to the calculated volume fraction. The resulting scattering curve was used for ab initio modeling, and the resulting overall shape of the dgGA:L0 dimers indicates a head-to-tail dimerization (Fig. 5).

Glucoamylase Has a Rigid Extended Conformation in Solution-
The present SAXS data provide experimental evidence that GA1 adopts a defined extended conformation in solution rather than being represented by two domains connected by a flexible linker (Fig. 3). The glycosylated linker region is well defined in both ab initio models of GA1 and GA2 and in the quasiatomic model of GA1. The glycosylated linker folds into a stable structure, even without the SBD. The approximate distance of 10 nm between the centers of masses for the CD and SBD, which is also revealed by the pairwise distance distribution function of GA1 (Fig.  2B), corresponds well with scanning tunneling microscopy analysis (14). Kramer and co-workers (14), however, do not exclude the existence of a flexible linker in solution, because scanning tunneling microscopy requires dehydration of the protein, which possibly results in changes conferring non-natural facets to protein structure. One of the great advantages of SAXS analysis is that there is no need for specific experimental conditions, allowing the data to be collected e.g. in the same solution as used for other experi- mental analyses. In the present work, conditions were chosen as used for determination of the highest enzymatic kinetic activity in solution (30). The extended linker consists of 69 amino acids and glycosylations corresponding to 63 molar eq of hexose (3,4). In the crystal structure of the CD, the 31-residue C-terminal segment of CD constitutes the N-terminal part of the linker in GA1 and wraps around the CD (11). If this conformation is conserved in solution, the remaining 38 residues of the linker must cover the distance observed between the CD and SBD and, thus, be in a highly extended conformation. Analysis of dgGA in solution corroborates this observation. Here, the linker is 20 amino acids shorter, yet the observed distance between the center of masses of the CD and SBD is ϳ8.0 nm (Fig. 2B).
The Linker Stabilizes the Starch Binding Domain-Differential scanning calorimetry reveals that the glycosylated linker exerts a stabilizing effect, most pronounced on the SBD. This is illustrated in the large difference in T m of 7°C upon reversible domain unfolding in GA1 and dgGA. This is in excellent agreement with findings on other linker variants of GA1 (9). Previous data from isolated SBD in solution reveal an even lower unfolding temperature (35). Taken together, these findings suggest that the non-native linkers present in the glucoamylase 1 linker variants cannot be ascribed a destabilizing effect but, rather, that the presence of the correctly folded, full-length glycosylated linker confers thermostability to the SBD. This further supports the intuitive interpretation of the ab initio SAXS model that the glycosylated linker domain and the SBD are in close proximity. Unresolved questions remain regarding the role of the glycosylated linker and substrate binding and degradation. Kinetics analysis on selected maltooligosaccharide substrates demonstrate that the GA2 and the GA1 forms have essentially the same properties (19,36). The linker may only play a stabilizing role in solution as previously suggested (19), although the effect was never assigned to a specific part of the linker (9). Because the dgGA variant contains a less stable SBD, it has been speculated that close contact with the glycosylated linker in GA1 supports correct refolding of SBD (19).
Extended Conformations May Be a Common Trait in Glycoside Hydrolase Families-Enzymes from another glycoside hydrolase family have also been studied using SAXS analysis. The cellulase Cel45 (EC 3.2.1.4) from Humicola insolens contains a catalytic module and a cellulose binding module separated by a glycosylated linker peptide and, thus, exhibits a domain organiza-  tion comparable with that of GA1. SAXS analysis of wild-type cellulase, and linker mutants showed that, as in GA1, the linker is in an extended conformation (37). However, this analysis indicated that the linker region was flexible, albeit this flexibility was limited with a high preference for extended conformations, arguably due to the high level of glycosylation of the linker. The extended conformation is even more pronounced in the present solution analysis of the A. niger GA1, which has a longer and even more heavily glycosylated linker region than the H. insolens cellulase (40 molar eq of hexose on a 36-residue linker (38)). We conclude in the present study that if any flexibility is present in the linker region of A. niger GA1, this flexibility is very limited and does not allow for interdomain contacts between the CD and SBD. Thus, there are seemingly strong similarities between the overall conformation of A. niger GA1 and H. insulens cellulase, although the latter may be slightly more flexible in solution. Similar studies of cellulosome-active cellulase from Clostridium cellulolyticum and Clostridium thermocellum revealed a pronounced flexibility of the free enzymes (39,40). These enzymes are, however, significantly differently organized than both cellulases from H. insolens and the A. niger glucoamylase in that they lack a carbohydrate binding domain. Instead they consist of a catalytic domain, a short 10 -12-residue non-glycosylated linker, and a dockerin domain. The latter domain attaches to cohesin domains on large scaffolding proteins in the cellulosome. In the context of the present data, it is noteworthy that SAXS studies of complexes between cohesin domains and cellulases revealed that the cellulase linker significantly compacted and rigidified the structure, and the complex state had clear structural analogies to the solution structure of the A. niger glucoamylase forms presented (40).

Glucoamylase Forms Dimers in Solution-
The SAXS analysis of glucoamylase solutions containing the inhibitor L0 shows an elevated MM accompanied by moderate increases in the average R g (from 3.7 to 3.9 nm for GA1 and from 3.0 to 4.0 nm for dgGA) and the D max within the molecules (from 16.5 to 21 nm for GA1 and from 14 to 17 nm for dgGA). We interpret this as a partial dimerization of enzymes arranged head-to-tail and mediated via binding of L0 to two protein molecules as corroborated by the modeling of dgGA dimers (Fig. 5). Isothermal titration calorimetric measurements of acarbose, L0, and ␤-cyclodextrin binding to dgGA are in good accordance with previous results of binding to wild-type GA1 (16).
The present L0 complexes of GA1 and dgGA provide important novel insight to the function of GA and the role of the glycosylated linker region, which apparently does not allow for intramolecular domain:domain (CD:SBD) contacts mediated via L0. Rather, the limited flexibility of the linker region renders an extended conformation that is suitable for dimerization. Previous interpretation of ITC data of the L0 binding by GA1 showed a 1:1 interaction (16). However, ITC cannot distinguish between 1:1 and 2:2 interactions (or higher multiples).
Albeit the synthetic heterobidentate inhibitor L0 lacks physiological significance, it seems likely that the SBD of one glucoamylase molecule can expose glucans on the surface of starch granules and that the CD of a second enzyme molecule can bind and hydrolyze the substrate. The interaction of two molecules of glucoamylase would in that case resemble the visualized head-to-tail dimer observed in this study. The present analyses, thus, do not support the hypothesis that the SBD can act as an intramolecular guide, directing substrate to the CD via a linker that was hitherto assumed to be flexible. On the contrary, the SBD may expose insoluble substrate to the CD of a second protein molecule. This is in accordance with the observation that the addition of isolated SBD to reaction mixtures of starch granules and GA2 increases the catalytic activity. The presence of intermolecular contacts also agrees with the reported indifference to linker length and glycosylation levels for glucoamylase catalytic activity (9). Although a similar solution conformation of monomers has been reported in the case of cellulases (37,39,40), it remains to be investigated whether other enzymes of the glycoside hydrolase families can form active dimers or perhaps higher oligomers in solution.
In conclusion, the combined biophysical and structural analysis shows that the highly glycosylated linker region in glucoamylase adapts to a defined structure in solution and that the linker region plays a stabilizing role for the SBD. It has been demonstrated that complex formation with a heterobidentate ligand induces dimerization. In light of these results, a previous hypothesis on high flexibility of the linker facilitating a direct CD and SBD interaction has been revisited, and we suggest that glucoamylase instead may form dimers during catalytic activity on starch granules.