Domain Motions and Quaternary Packing of Phosphofructokinase-2 from Escherichia coli Studied by Small Angle X-ray Scattering and Homology Modeling*

The binding of MgATP and fructose-6-phosphate to phosphofructokinase-2 from Escherichia coli induces conformational changes that result in significant differences in the x-ray-scattering profiles compared with the unligated form of the enzyme. When fructose-6-phosphate binds to the active site of the enzyme, the pair distribution function exhibits lower values at higher distances, indicating a more compact structure. Upon binding of MgATP to the allosteric site of the enzyme, the intensity at lower angles increases as a consequence of tetramer formation, but differences along higher angles also suggest changes at the tertiary structure level. We have used homology modeling to build the native dimeric form of phosphofructo-kinase-2 and fitted the experimental scattering curves by using rigid body movements of the domains in the model, similar to those observed in known homologous structures. The best fit with the experimental data of the unbound protein was achieved with open conformations of the domains in the model, whereas domain closure improves the agreement with the scattering of the enzyme-fructose-6-phosphate complex. Using the same approach, we utilized the scattering curve modeling ( n m (cid:1) 1 for dimer modeling and n m (cid:1) 3 for tetramer modeling). F (cid:12) , as defined in Equation 5, is a measure of the relative improvement of (cid:12) when the simple CRYSOL fitting procedure with a single “fixed” molecular structure is substituted by the evalua- tion of several “variable” structures. A given increase in F (cid:12) (reduction in (cid:12) ) due to the introduction of n m additional parameters is statistically more than 95% significant if F (cid:12) (cid:12) 9.55 (for n m (cid:1) 1) or F (cid:12) (cid:12) 5.79 (for n m (cid:1) 3) (28). The parameter errors reported in this work are relative to the 95% confidence limit of the F -statistics.

Oligomeric enzymes frequently require conformational changes within or between subunits for their activity and regulation. Furthermore, allosteric ligands may affect enzymatic activity by means of conformational changes to the tertiary and/or quaternary structure, which may result in metabolic pathway regulation.
Phosphofructokinase activity, the ATP-dependent phosphorylation of fructose-6-P, 1 is an important step in the glycolytic pathway that is subject to strict regulation in a wide variety of organisms. In Escherichia coli this activity is accomplished by two isozymes that differ in their kinetic and structural properties. Phosphofructokinase-1 (Pfk-1) has been extensively characterized and belongs to the PfkA protein family that includes higher eukaryotic ATP-dependent phosphofructokinases ATPand pyrophosphate-dependent bacterial and plant phosphofructokinases (1). The atomic structure of Pfk-1 has been solved by x-ray crystallography (2); the enzyme is a homotetramer whose state of aggregation does not change upon ligand binding. The other isozyme, phosphofructokinase-2 (Pfk-2), presents inhibition by the substrate MgATP when the assay is performed at low fructose-6-P concentrations (3). Fluorescence studies demonstrated that MgATP inhibition occurs upon binding of MgATP to an allosteric site in Pfk-2 (4). Also, binding of MgATP promotes oligomerization of Pfk-2, changing from a dimer to a tetramer (5)(6)(7). Such features make Pfk-2 an excellent model to study allosteric regulation linked to protein aggregation and enzymatic inhibition.
Pfk-2 has no conserved patterns of sequence associated with the PfkA protein family. However, Pfk-2 is related to the PfkB superfamily of sugar kinases which includes ribokinases, adenosine kinases, fructokinases, and possibly, ADP-dependent glucokinases and phosphofructokinases (8,9). Several crystal structures have been solved for members of this family (9 -12) showing that the overall fold is strongly conserved. For example, the root mean square deviation for superimposed residues between E. coli ribokinase and human adenosine kinase is 2.4 Å even though the sequence identity among them is only 22% (11). The protein fold in this superfamily consists of two domains, a large ␣/␤ domain and a small domain. In ribokinase this small domain is a ␤-sheet that acts as a lid over the active site and also as the dimerization interface. Adenosine kinases have ␣-helical insertions in the small domain so that dimerization is precluded.
The comparison of the free and sugar-bound forms of ribokinase and adenosine kinase structures reveals an important conformational change that can be described as a hinge bending motion, with one domain rotating toward the other by an angle of 17-30° (12,13). The hinge-like movement helps bind-ing of the second substrate (MgATP) and further catalysis. In Pfk-2, the binding of fructose-6-P promotes a conformational change, as indicated by a 30% increment in the fluorescence emission of the single tryptophan (Trp-88) (4) and enables the subsequent binding of MgATP to the active site, as indicated by the ordered bi-bi reaction mechanism of Pfk-2 in which MgATP can bind the active site only after fructose-6-P binding (14). On the other hand, when MgATP is bound to the allosteric site, fluorescence quenching with a blue shift in the emission maximum is observed. It is not known whether this last effect can be accounted for by the observed quaternary structural transition only or if it also involves tertiary structural changes.
To determine the effects of fructose-6-P binding to the active site and MgATP binding to the allosteric site on the tertiary and quaternary structure of Pfk-2, we used homology modeling combined with small angle x-ray scattering (SAXS), a technique sensible to shape and oligomeric state changes. In this work, SAXS data have been used to detect shape changes upon fructose-6-P binding that have been interpreted in terms of quasi-rigid domain movements within the dimeric model of Pfk-2. Also, SAXS data have been used to detect changes in the aggregation state of the enzyme upon MgATP binding, and a model for the subunit arrangement in the tetramer is suggested. Based on our results, a mechanism for the inhibitory effect of MgATP is proposed.

EXPERIMENTAL PROCEDURES
Protein Purification-pET 21-d plasmid (Novagen) containing the cloned original gene (15) was transformed into E. coli strain JM109, and 2 liters of culture were grown in Luria Broth medium supplemented with ampicillin to a final concentration of 100 g/ml. Protein expression was induced at A 600 ϭ 0.6 by the addition of isopropyl-1-thio-␤-Dgalactopyranoside to a final concentration of 0.5 mM. After 4 h of induction, cells were collected by centrifugation. Purification was as described (16) using the above cell pellets as the starting point.
Homology Modeling-Homology models of Pfk-2 were constructed with the program MODELLER (Version 6) (17), which implements an automated approach to comparative protein structure modeling by satisfaction of spatial restraints. To find template structures, a positionspecific matrix was iteratively generated searching over a non-redundant protein sequence data base with the PSI-Blast program (18) using BLOSUM62 as the starting matrix and an expectation value threshold of 0.001 for inclusion in the next iteration. The convergence matrix was used to search the Protein Data Bank (PDB) amino acid sequence data base producing significant alignments with E. coli ribokinase (PDB code 1RK2; similarity 37%, expectation value 3 ϫ 10 Ϫ60 ), human adenosine kinase (PDB code 1BX4; similarity 29%, expectation value 2 ϫ 10 Ϫ51 ), and Toxoplasma gondii adenosine kinase (PDB code 1DH0; similarity 29%, expectation value 5 ϫ 10 Ϫ34 ). This alignment was further corrected using information derived from a structural alignment of these structures. The final alignment was used as input for MOD-ELLER with 1RK2 as template, since it is the only dimeric homologue. Ten models were built using the MODELLER default routine in which initial coordinates are randomized in Cartesian space by 4 Å before variable target function optimization and evaluated by pseudo-energy parameters. The probability of a good model structure, p(G), associated to the ratio between PROSA II Z-score (19) and the natural logarithm of its sequence length (20), was calculated on the pG server (guitar. rockefeller.edu/pg). The three-dimensional-one-dimensional score of the models was calculated using the Verify3D program (21).
Solution Scattering Experiments and Data Processing-SAXS data were collected at the SAS beam line of the National Synchrotron Light Laboratory (Campinas, Brazil) using a one-dimensional position-sensitive detector (22). Experiments in the absence of ligands and in the presence of MgATP were done at 2, 4.2, and 30 mg/ml of Pfk-2; experiments in the presence of fructose-6-P were performed at 5 and 25 mg/ml of Pfk-2. A wavelength ϭ 1.488 Å was used, with sampledetector distances of 631.2 and 1260 mm providing two scattering curves that, combined, covered the momentum transfer range 0.01 Å Ϫ1 Ͻ q Ͻ 0.55 Å Ϫ1 (q ϭ 4sin/, where is half the scattering angle).
The scattering curves of the protein solutions and the corresponding buffer were collected in several frames (15 s to 3 min) to monitor radiation damage and beam stability. The data were normalized to the intensity of the incident beam and corrected for non-homogeneous de-tector response. De-smearing for the 8-mm height detector entrance window was also performed. The scattering of the buffer was subtracted, and the resulting curves were scaled to equivalent concentrations. To provide curves free from concentration artifacts, we used samples with protein concentrations of 2-5 mg/ml to build the low resolution part of the SAXS curve (q Ͻ 0.15 Å Ϫ1 ) and 25-30 mg/ml for the high resolution part (0.15 Å Ϫ1 Ͻ q Ͻ 0.55 Å Ϫ1 ) of the curve shown in Fig. 2. The pair distribution functions p(r) were evaluated by indirect Fourier transformation using the program GNOM (23). The SAXS intensity produced by an isotropic, monodisperse dilute solution of proteins, in the limit for small q, obeys the Guinier approximation (24), where I (0) is proportional to the molecular weight provided I(q) is normalized to an equivalent protein concentration, and R g is the protein radius of gyration, defined by, where Z k is the atomic number of the atom located at a distance r k from the electronic center of mass. The radii of gyration of the Pfk-2 molecules in their different states were calculated from the slopes of the Guinier plots (ln I(q) versus q 2 ).
Ab Initio Shape Determination-The resolution of the resulting solution x-ray-scattering curve extended to 11.4 Å. The low resolution protein shape was restored using the ab initio procedure described by Svergun (25) as implemented in the program GASBOR. In this method, a dummy residue model is generated by a random-walk C␣ chain and is folded in a way to minimize the discrepancy between the calculated scattering curve from the model and the experimental data. The program simulates the protein internal structure, which makes it unnecessary to subtract a constant from the experimental data to ensure Porod's law (24). Several runs of ab initio shape determination with different starting conditions led to consistent results as judged by the structural similarity of the output models, yielding nearly identical scattering patterns and fitting statistics in a stable and self-consistent process. The final shape restoration for the Pfk-2 free protein and with fructose-6-P was performed using 618 dummy residues and 611 waters assuming 2-fold molecular symmetry. In the case of Pfk-2 with an excess of MgATP we used 1236 dummy residues and 988 waters assuming a 222 molecular symmetry.
Domain Movement Modeling-The open and closed crystallographic structures of ribokinase (PDB codes 1RKA and 1RKD, respectively) were chosen as templates to model the domain movements in Pfk-2. Because these PDB entries contain coordinates for just one monomer of ribokinase, the second monomeric subunit of the dimeric ribokinase was obtained by application of the appropriate 2-fold crystallographic symmetry operations. The program DynDom (26) was then used to define the quasi-rigid domains, the flexible inter-domain connecting residues, and the screw axes that virtually describe the conformation transition in ribokinase. Using a sliding window of 5 residues, the program DynDom found three dynamic domains, corresponding to the two ␣/␤ domains of the dimer (residues 4 -13, 42-96, and 118 -309 from each monomer, following the 1RKD numbering scheme) and the central ␤ barrel to which both subunits contribute (residues 18 -38, 99 -113). Residues 14 -17, 39 -41, 97-98, and 114 -117 in each monomer were defined as composing a flexible hinge that rotates one ␣/␤ domain toward the central ␤ barrel. The directions of the two screw axes (one for each monomer) have been found as well as the associated rotation (ϳ17°) and translation (ϳ1 Å) parameters. Finally, the Pfk-2 model was superimposed on the ribokinase structure (using just the homologous residues in the central ␤ barrel domain as reference), and the residues corresponding to the two ␣/␤ domains were rigidly rotated along the screw axes found by DynDom for ribokinase. Several models were generated (program MOLEMAN2; x-ray.bmc.uu.se/usf/moleman2_man.html) rotating the ␣/␤ domains from Ϫ20°(most closed model) to ϩ40°(most open model) with an angular step of 2°. The theoretical SAXS curves of each rotated model were subsequently fitted to the experimental data as described under "Calculation of SAXS Intensity from Atomic Models." Tetramer Modeling-Two copies of each dimeric model (see above) were aligned putting their centers of mass (COMs) at the origin of an orthogonal reference system, with their 2-fold symmetry axes aligned along x and their major inertia axis aligned along y (program MOLE-MAN2). After rotation of one of the dimers by 180°along y, the dimers were translated in opposite directions along x, obtaining the two major configurations of the tetramer (tetramer-I and tetramer-II) depending on the direction of translation. Several tetrameric models were obtained applying different separations (along x) between the COMs of the dimers (31-39 Å for tetramer I, 26 -34 Å for tetramer II, 0.333-Å translation step). Furthermore, several shear angles (rotations in opposite directions along x) were applied to each dimer (0 -44°with 4°s tep). For each configuration, the fit against SAXS data from Pfk-2 in complex with MgATP was evaluated as described below.
Calculation of SAXS Intensity from Atomic Models and Estimation of the Uncertainty of the Fitted Parameters-The theoretical SAXS curves were determined from the atomic models using the program CRYSOL (27). The program calculates the SAXS intensity through the equation, where A a (q) is the amplitude scattered by the protein (calculated from the atomic structure factors), s A s (q) is the amplitude produced by the excluded volume (determined using dummy Gaussian spheres placed at the atomic positions), and s is the electron density of the solvent. The first solvation shell is modeled by a hydration shell that yields an where ␦ b is the electron density difference between the hydration shell and the solvent. The hydration shell is approximated by a 3-Å-thick uniform layer placed 2 Å away from the protein envelope. The symbol Ͻ Ͼ ⍀ indicates spherical averaging. Two parameters, the excluded volume of the particle (V) and the electron density in the hydration layer ( b ), are optimized by the program CRYSOL to minimize the discrepancy defined as, where I exp (q j ) and (q j ) denote the experimental SAXS intensity of the jth point and its standard deviation, respectively, and N is the number of experimental points. The excluded volume is varied around the value predicted from the molecular mass by changing the average displaced volume per dummy atom to account for the uncertainty in its partial specific volume.
To estimate the errors in the parameters fitted during dimer and tetramer modeling (i.e. the ␣/␤ domain rotation angle, the dimer shear angle, and the dimer COM distance), the F-statistics method was used (28). The F-statistics parameter, as applied to the present work, can be written as, where N is the number of experimental points used by the fitting procedure (n ϭ 89 in this work), n c ϭ 2 is the number of the CRYSOL fitted parameters (i.e. V and b ), and n m is the number of parameters varied during modeling (n m ϭ 1 for dimer modeling and n m ϭ 3 for tetramer modeling). F , as defined in Equation 5, is a measure of the relative improvement of when the simple CRYSOL fitting procedure with a single "fixed" molecular structure is substituted by the evaluation of several "variable" structures. A given increase in F (reduction in ) due to the introduction of n m additional parameters is statistically more than 95% significant if F Ͼ 9.55 (for n m ϭ 1) or F Ͼ 5.79 (for n m ϭ 3) (28). The parameter errors reported in this work are relative to the 95% confidence limit of the F-statistics.

Homology Modeling
The three-dimensional structure of dimeric Pfk-2 was constructed by homology modeling (program MODELLER-6) (17) using the crystal structure of E. coli ribokinase as template (37% similarity), as described under "Experimental Procedures." Ten models of Pfk-2 were built starting from different random initial atomic positions at the beginning of the optimization and evaluated by pseudo-energy parameters. The C␣chain trace of the 10 models of Pfk-2 are superposed in Fig. 1a, showing the positional variability between the models (only the monomer is shown for simplicity). The probability of good global folding, p(G) (20), was around 0.999 for each model, and the average three-dimensional-one-dimensional profile score (pro-gram Verify3D) (21) was Ͼ0.1 for 94% of all sequence positions using a 21-residue sliding window (Fig. 1b).
The Pfk-2 model presents the two-domain structure described for ribokinase (10). The larger domain forms an ␣/␤/␣ three-layer sandwich, which we call the "␣/␤ domain," whereas the smaller domain forms a ␤-structure responsible for interfacial contacts in the dimer. The active site lies in the cleft formed by both domains, and the single tryptophan, Trp-88, is located in the region between both domains. The lowest threedimensional-one-dimensional score region in the profile is located around position 120 (Fig. 1b). This region also has a low score in the experimental ribokinase structure (not shown) and corresponds to a distorted ␣-helix. The score worsens slightly in the Pfk-2 models, where there is a one-amino acid insertion in the primary structure.

Solution Small Angle X-ray Scattering Curves
Three sets of SAXS measurements were carried out with Pfk-2, without ligands, and in the presence of saturating concentrations of fructose-6-P or MgATP. All experimental curves are plotted in Fig. 2. The existence of induced conformational changes in the Pfk-2 structure is clearly observed when fructose-6-P or MgATP is bound to the enzyme. Differences in SAXS intensities can be observed along the whole q range. In the case of the Pfk-2-fructose-6-P complex differences are small but not negligible.
Differences in the scattering curve for q Ͻ 0.08 Å Ϫ1 upon MgATP binding indicate a change in the oligomeric state of the Pfk-2 with a 2-fold increment in scattering intensity at q ϭ 0, I(0), relative to the free and fructose-6-P-bound forms. When no ligands are bound to Pfk-2, its radius of gyration is 29.10 Ϯ 0.21 Å. Upon MgATP binding to the allosteric site, the radius increases up to 35.38 Ϯ 0.14 Å as a consequence of tetramer formation. The same difference in radius of gyration upon MgATP binding was observed previously in dynamic lightscattering experiments (7). On the other hand, when fructose-6-P is bound, the protein radius of gyration shrinks slightly to 28.67 Ϯ 0.19 Å.
Previous experimental results (5-7) indicate that binding of MgATP to the enzyme promotes a change in quaternary structure. The deep valley observed in the scattering intensity curve in the presence of MgATP, around q ϭ 0.2 Å Ϫ1 (Fig. 2), suggests that the ligand also affects Pfk-2 at the tertiary structure level. This feature is not observed in the SAXS curve of the unbound enzyme.
When fructose-6-P binds to Pfk-2, a conformational change associated to the tertiary structure of the protein occurs. This change is apparent in the pair distribution function of the protein, p(r), determined from the SAXS results, which is plotted in Fig. 3. This function exhibits lower values at distances between 30 and 50 Å in Pfk-2 complexed to fructose-6-P relative to the free form. It is therefore expected that Pfk-2 should be somewhat more compact when the sugar is bound to the active site. To obtain more detailed information about these changes, we have performed rigid body refinements of the domain orientations in the model, as described below.

Modeling Ligand-induced Conformational Changes Using SAXS Data
Free and Fructose-6-P-bound Forms of Pfk-2-The SAXS data corresponding to the free and the fructose-6-P-bound forms of Pfk-2 indicate a probable substrate-induced conformational change of the enzyme structure. Substrate-induced conformational changes based on crystallographic results have already been reported for the related enzymes ribokinase and adenosine kinase (12,13). These enzymes exhibit a hinge-like movement of the ␣/␤ domain(s) toward the small domain (lid). This movement can be described as a quasi-rigid rotation around an axis passing through the connecting interdomain residues, although the rotation direction and angle are somewhat different for the two enzymes. To account for the different experimental SAXS curves obtained from Pfk-2 in the absence and in the presence of fructose-6-P, our Pfk-2 three-dimensional structure model has been modified by several rigid rotations of each of the two ␣/␤ domains around an axis equivalent to the domain rotation axis found in ribokinase, which is evolutionary more closely related to Pfk-2 than adenosine kinase. Fig. 4b shows the -fit to the SAXS data of the calculated scattering intensity for the proposed models at different ␣/␤ domain rotation angles. In this figure, the rotation angles 0°a nd ϩ17°correspond to the closed and open conformations of the original ribokinase structure, respectively. Also, a rotation of 0°corresponds to the original Pfk-2 homology model. It can be seen that, for the unbound form of Pfk-2, the lowest discrepancy is obtained using a more open conformation, with a clear minimum angle around 25°(Ϯ 7°). For the fructose-6-P-bound form, a better fit is obtained using a closed conformation with a minimum around 13°, although in this case the minimum of the function is less well defined, corresponding to a rather large uncertainty (Ϯ12°). This suggests that Pfk-2, upon fructose-6-P binding, adopts a more closed conformation than the free form, similarly to that observed for ribokinase with ribose. The rotation angles corresponding to the minima in Fig. 4b do not correspond exactly to the free and substrate-bound structures of ribokinase. This may reflect an actual different opening angle of the structure in the free and substrate-bound form. This is not totally unrealistic; a comparison of the substratebound forms of adenosine kinase and ribokinase reveals that the former is more open by a rotation angle of ϳ6 degrees, indicating that the closed conformations of this family of proteins may have different orientations of the ␣/␤ domains. Alternatively, the enzyme may present movements that are different from those proposed in our model.

MgATP-bound Form; Tetrameric Packing of Pfk-2-
It is known that MgATP binding induces tetramerization in Pfk-2, but the quaternary structure of the tetrameric enzyme in solution remains unknown. Using our SAXS data, we have modeled the tetrameric arrangement of the enzyme as being a "dimer of dimers." This implies a 222 point symmetry for the tetramer (D 2 , according to Schoenflies notation), where one of the 2-fold axes corresponds to the symmetry axis of both dimers (the x axis in Fig. 5). This axis passes through the two central ␤-barrels.
Two opposite orientations of the dimers are possible (Fig. 5); one (which we called tetramer-I) with the active sites looking outward from the tetramer and the other (tetramer-II) with the active sites looking inward. Fig. 6b shows the -fit to the experimental SAXS data for the tetrameric models of type I and II. The tetrameric models have been obtained using the dimeric models previously described in this article, i.e. using different opening angles of the monomers (rotation of the ␣/␤ domains, see Figs. 4a and 5). The -fit in Fig. 6b refers to optimized tetrameric models. The distances between the COMs of the dimers and the shear angle (Fig. 5) of the dimers along their symmetry axis have been refined to get the best fit to the SAXS data (Fig. 6a). These are the only rigid movements of the dimers that are allowed in order to maintain the 222 symmetry of the tetramer.
The discrepancy from the experimental SAXS data obtained for the best fitted model is small ( Ϸ 1) both for tetramer-I and for tetramer-II. It is evident that, no matter which orientation (tetramer-I or -II) is used, the best fits with SAXS data are achieved with closed conformations of the monomers, similar to what is observed for the complex with fructose-6-P. Although the curve for tetramer-II shows lower values than that for tetramer-I, inspection of the three-dimensional models indicates that these lower values are reached at the expense of unrealistic clashes between the two dimers in tetramer-II (not shown). Furthermore, tetrameric models of type II, built with a reasonable distance between the dimers to prevent steric hindrance, poorly fit the SAXS data ( Ͼ 3). On the other hand, the minimum for tetramer-I reflects a configuration that could reasonably represent (within the limits of our rigid-body modeling of the domains configuration) the real structure. Fig. 6a shows the -fit as a function of the COM distance and the shear angle between the dimers in tetramer-I (using a domain opening of 7°). Considering the three variables of our rigid-body modeling for the tetramer (COM distance between the dimers, shear angle of the dimers, monomer aperture), the region around the minimum appears to be relatively flat. As a matter of fact, the 95% confidence limits, according to the F-statistics, correspond to COM distances between 34.7 and 36.3 Å, shear angles between 4 and 16°, and monomer opening angles between 1 and 9°around the minimum (maximum variation from 1.19 to 1.23). The parameter that mostly affects is the distance between the dimers. The opening and the shear angle, within the ranges mentioned above, do not significantly modify the value. However, the discrepancy increases quickly outside these ranges.
Inspection of the proposed tetrameric model indicates that the dimer-dimer interface should be formed by contacts between the ␣/␤ domains from opposite monomers, although the fine details of this interaction cannot be deduced from our simple, low resolution model.

Ab Initio Low Resolution Structures
The SAXS data from the free, fructose-6-P, and MgATPbound forms of Pfk-2 were used to restore the low resolution shape of the protein using the dummy residue model method of Svergun et al. (25), as described under "Experimental Procedures." The restored shape for each conformation of Pfk-2 yields a good fit to the experimental data (see the legend of Fig.  2), indicating that imposed symmetry restrictions actually reflect prevalent features in the structure. The molecular envelope of each ab initio model is shown superimposed to the corresponding refined homology model in Fig. 7, showing good agreement between both kinds of structures. DISCUSSION Domain closure induced by sugar binding has been demonstrated in crystallographic high resolution studies on ribokinase and adenosine kinase (12,13). These domain motions are characterized by large changes in main chain torsion angles of a small number of residues that comprise the hinge that separates the small and large domains of these enzymes. In adenosine kinase, comparison of dihedral angles of the free and sugar-bound structures reveals that two glycines (Gly-68 and Gly-69) undergo the large torsional changes necessary for hinge bending. When sugar binds to its site, displacement of these glycines occurs to avoid steric hindrance. The nearly absolute conservation of these glycine residues throughout the PfkB superfamily indicates that they probably play the same critical role in hinge bending. In addition, sugar binding affects the nearby nucleotide binding site in ribokinase, shifting it toward a conformation that is observed when nucleotide is bound, possibly increasing the affinity for this substrate. The main feature of domain closure is the complete occlusion of the sugar site from bulk solvent.
These observations suggest a general mechanism for the reaction catalyzed by kinases in this superfamily; sugar binds initially in an open active site, favoring domain closure, affecting the nucleotide binding site, and therefore, increasing its affinity for ATP. Three independent lines of evidence corroborate these predictions about structural changes of Pfk-2 upon sugar binding. First, previous kinetic studies show that a compulsory ordered kinetic mechanism occurs in Pfk-2, in which MgATP binds the active site only after fructose-6-P binding (14); therefore, sugar binding increases the affinity for MgATP in the active site. Second, and supporting domain closure, the structure of Pfk-2 in solution appears to be more compact when it is bound to fructose-6-P, as suggested by limited proteolysis experiments (7), where binding of fructose-6-P increases the resistance to cleavage by several proteases. And third, rigid body ␣/␤ domain closure of the homology model improves the agreement with the experimental SAXS data of the fructose-6-P-bound form.
Based on the present results, we also propose a model for the effect of MgATP on the enzymatic activity of Pfk-2. As can be seen from Fig. 5b, a closed structure is locked when MgATP is bound to the allosteric site no matter what configuration (tetramer-I or -II) is used to fit experimental data. By analogy with the ribokinase and adenosine kinase closed forms, it might be expected that occlusion of the sugar site from solvent hinders either fructose-6-P binding or product release, thus giving the first structural support to a mechanism for MgATP enzymatic inhibition in this enzyme.
It should be noticed that if domain closure exposes surface determinants needed for tetramerization, oligomerization would be expected also as a consequence of fructose-6-P binding, but this is not the case. Unless MgATP itself takes part in the interaction surface, this ligand must induce another conformational change different from the ␣/␤ domain closure to induce tetramerization. In this regard, the calculated scattering curves from tetrameric models (not shown) does not fit very well the valley around q ϭ 0.2 Å Ϫ1 in the experimental data from the Pfk-2-MgATP complex (Fig. 2). In ribokinase, the dimer interface bears striking similarity to domains of two ligand binding and transport proteins that are built up from orthogonal ␤ sheets (29), but ribokinase lacks an internal space large enough to hold a small molecule ligand. Unfortunately, the existence of such appropriate internal space for MgATP binding in the Pfk-2 dimer interface cannot be established by the present work.
Chemical modification studies suggest that cysteine 295 is involved in the dimer-dimer interface. 2 Mapping this amino acid onto the three-dimensional structure of the proposed tetramer-I model indicates that, although this residue is not located at the interface, it occupies a nearby location.
In previous fluorescence studies of the single tryptophan in Pfk-2 (Trp-88), three states have been reported (4), one with a high quantum yield (Pfk-2 saturated with fructose-6-P), another with an intermediate emission (Pfk-2 without ligands), and a third one with a low quantum yield state (Pfk-2 in presence of MgATP). Acrylamide quenching of protein fluorescence demonstrates that the tryptophan solvent accessibility is reduced when Pfk-2 is bound to MgATP as compared with the free enzyme and the Pfk-2-fructose-6-P complex, whose accessibilities are similar. Our SAXS results fully agree with these previous observations; in the model proposed here, Trp-88 is located in the ␣/␤ domain near the hinge; thereby this intrinsic probe could directly detect different openness of domains in monomers. In the tetramer model, Trp-88 is located near the dimer-dimer interface; thus, solvent accessibility might be reduced as a consequence of tetramerization.
Our model refinement of dimeric Pfk-2 in its free and fructose-6-P-bound states using SAXS results indicates that the sugar promotes a domain closure with ϳ12 degrees of rotation. Results of our modeling indicate that the tetrameric structure of Pfk-2 complexed with MgATP is composed by two parallel or slightly misaligned dimers located at a distance of around 34 Å between each COM, with the active sites looking outward from the tetramer and the monomeric subunits in an almost closed conformation. This tetrameric model provides satisfactory agreement with previous studies on intrinsic fluorescence and chemical modification of the enzyme. It represents the result of a combination of theoretical homology modeling and experimental low resolution structure determination, which we consider to be valuable in the absence of a crystal structure.
Most hinge-bending proteins appear to display a dynamic equilibrium between their open and closed states, the latter stabilized by ligand binding. Because the present SAXS measurements of E. coli Pfk-2 were taken under equilibrium conditions, our observations could be interpreted as follows. In solution, with no ligands added, equilibrium is displaced toward an open conformation. When fructose-6-P binds the active site, closed structures become populated, helping the subsequent binding of MgATP to the active site, the first step toward catalysis. However, binding of MgATP to the allosteric site promotes, along with tetramerization, a domain closure that occludes the active site, and as a consequence, this impedes the entrance of fructose-6-P to the active site (or the release of products), thus producing the observed enzymatic inhibition.