Glutamate dehydrogenase: Structure of a hyperinsulinism mutant, corrections to the atomic model, and insights into a regulatory site

Abstract

Mammalian glutamate dehydrogenase (GDH) has complex allosteric regulation and the loss of GTP inhibition causes the hyperinsulinism/hyperammonemia syndrome (HHS) where insulin is hypersecreted upon consumption of protein. The archetypical HHS lesion is H454Y and lies in the GTP binding pocket. To better understand the mechanism of HHS, we determined the crystal structure of H454Y. When the bovine GDH crystal structures were minimized to prepare for further computational analysis, unusually large deviations were found at the allosteric NADH binding site due to chemical sequence errors. Notably, 387 lies in an allosteric where several activators and inhibitors bind and should be lysine rather than asparagine. All structures were re-refined and the consequence of this sequence error on NADH binding was calculated using free energy perturbation. The binding free energy penalty going from the correct to incorrect sequence found is +5 kcal/mol per site and therefore has a significant impact on drug development.

Broader Audience Abstract

Glutamate dehydrogenase is a key enzyme involved in amino acid catabolism. As such, it is heavily regulated in animals by a wide array of metabolites. The importance of this regulation is most apparent in a genetic disorder called hyperinsulinism/hyperammonemia (HHS) where patients hypersecrete insulin upon the consumption of protein. We determined the atomic structure of one of these HHS mutants to better understand the disease and also analyzed an allosteric regulatory site.

1 |. INTRODUCTION

Mammalian glutamate dehydrogenase (GDH) is a homohexameric enzyme that catalyzes the reversible conversion of glutamate to α-ketoglutarate in the mitochondrial matrix.¹ In animals, GDH is allosterically regulated by a wide range of small molecules with major activators being leucine and ADP and major inhibitors including GTP, NADH, and palmitoyl CoA (for a review, see²).

While the existence of this complex allosteric regulation has been known for decades, the importance of GDH allostery has only been recently identified from studies on the Hyperinsulinism/Hyperammonemia Syndrome (HHS).^3,4 HHS patients have point mutations that directly or indirectly abrogate GTP inhibition of GDH. Children with HHS have increased β-cell responsiveness to leucine and susceptibility to hypoglycemia following high protein meals.⁵ This is likely due to uncontrolled amino acid catabolism that stimulates insulin secretion and yields high serum ammonium levels. In the pancreas, dysregulated GDH causes an exaggerated insulin response to amino acid consumption by increasing the flux of glutamate to α-ketoglutarate that, in turn, boosts ATP formation that stimulates insulin degranulation.³ In the kidneys, high glutamate catabolism leads to increased ammonia output.⁶ In the CNS there is a high correlation between HHS and epilepsy, learning disabilities, and seizures that are independent of the high ammonium and hypoglycemia.⁷ This is likely because glutamate and its derivative, γ-aminobutyric acid, are important neurotransmitters.

The complexity of GDH allostery makes the thermodynamics and kinetics of GDH regulation in relation to its structure a long-standing problem.⁴ As the bovine GDH protein sequence is 98% identical to human GDH sequence with 100% similarity near substrate and regulatory binding sites, bovine GDH is an excellent surrogate for human GDH.

GDH is a homohexamer composed of a dimer of trimers (Figure 1).⁸ Each monomer contains approximately 500 residues.⁸ In the structure of the bovine GDH/NADH/Glu/GTP complex (Figure 1; pdb 3 MW9), NADH and glutamate (the abortive complex) are observed in the active site and cause the “closed” conformation where the NAD binding domain is closed tightly over the active site.⁸ The inhibitor, GTP, binds at the base of the antenna between the pivot helix and the top of the NAD binding domain. A second molecule per subunit of NADH binds beneath the pivot helix between adjacent subunits and is presumed to inhibit GDH.⁹ This allosteric site is particularly complex since it binds the inhibitor ECG⁹ as well as the activators ADP¹⁰ and 75-E10.¹¹

FIGURE 1.

View of bovine GDH trimer in the inhibited state bound to NADH. NADH bound at the second site is shown in mauve. The trimer is composed of monomer a (blue), monomer B (red) and monomer C (green). The overlap of each NADH binding site between two adjacent monomers is shown

The goal of this work was to examine the properties of the GTP inhibitory site with regard to HHS. We determined the structure of mutant human GDH with the HHS mutation, H454Y. Even though this mutation causes >400-fold decrease in sensitivity to GTP, it causes insignificant changes to the GTP binding site other than sterically interfering with GTP binding and eliminating a potential salt bridge with the β-phosphate. Conversely, the conformation of the GTP binding site remains the same, but the presence of tyrosine eliminates favorable interactions GTP has with histidine. In preparation for in silico studies on the GTP binding site, the structure of bovine GDH complexed with NADH/Glu/GTP was energy minimized. Abnormally high deviations from the starting structure were observed in the second, allosteric, NADH binding site. While trying to understand the cause for these deviations, we found several inconsistencies in the bovine GDH sequence that affects the X-ray crystallographic structures 1HWZ, 3MW9, 3MVQ, 1HWY, 1NQT, 1NR7, 3ETE, 3JCZ, 3MVO, 3QMU, 3ETD, 3ETG, and the cryo-electron microscopy structures 3JCZ, 3JD0, 3JD1, 3JD2, 3JD3, 3JD4, 5 K12.^{8–11,13–16} An interesting consequence of the errors involves the NADH/ADP/ECG allosteric binding site geometry and interactions. Further these sequence errors have nontrivial consequences on the computed thermodynamics of NADH binding and therefore other compounds at the same site. From the corrected and refined structures, it is also clear that the previously proposed model for ADP activation is likely too simplistic.

2 |. MATERIALS AND METHODS

2.1 |. Structure determination of the H454Y-huGDH mutant

The H454Y mutant of human GDH was cloned, expressed, and purified as previously described.¹⁷ H454Y GDH crystals were prepared using the vapor diffusion method and sitting drop apparatus. The enzyme was dialyzed against 0.1 M sodium phosphate, pH 6.8 and adjusted to a concentration of 3.75 mg/mL. The reservoir solution contained 10% PEG 8000 (w/v), 0.1 M sodium chloride, 1.3% octyl-ß-glucopyranoside (w/v), 7.5% methyl pentanediol (v/v), 0.1 M sodium phosphate, pH 6.8, and 1 mM sodium azide. The drop was composed of 8 μL reservoir, 1.35 μL enzyme solution, and 0.65 μL water. The crystals were prepared for freezing in the Oxford system cryostream by incubation in a synthetic mother liquor solution containing increasing concentrations of glycerol. This solution contained 60 mM sodium phosphate, 1% octyl-ß-glucopyranoside (w/v), 8% methylpentanediol (v/v), 50 mM sodium chloride, 14% PEG 8000 (w/v), and 1 mM sodium azide. The crystals were incubated for 30 minutes in this solution containing 0%, 2%, 5%, and 10% glycerol (v/v). At the end of these incubations, the concentration of PEG 8000 in the synthetic mother liquor was increased to 22%, and the crystals were incubated for 30 minutes in each of the solutions containing 12.5%, 15%, 17.5%, and 20% glycerol. Data were collected from a single crystal with dimensions of 0.22 mm × 0.24 mm × 0.11 mm on a R-axis IV imaging plate system attached to a Rigaku generator. The data set is composed of 260 diffraction images with oscillation angles of 0.4° and exposure times of 45 minutes. The crystal belonged to the space group P1 and had unit cell parameters of a = 96.78 Å, b = 98.44 Å, c = 124.28 Å, α = 85.95, β = 69.34, and γ = 60.98. The data collection statistics are shown in Table 1.

TABLE 1.

Data collection and refinement statistics

Wavelength
Resolution range	29.77–2.7 (2.797–2.7)
Space group	P1
Unit cell (a,b,c,α,β,γ)	96.803, 98.38, 124.3, 85.94, 69.35, 61.0
Unique reflections	90 899 (7177)
Multiplicity	2.2(2.0)
Completeness (%)	88.82 (70.29)
Mean I/sigma(I)	12.9 (2.0)
Wilson B-factor	45.77
R-merge	0.06 (0.24)
Reflections used in refinement	90 875 (7176)
Reflections used for R-free	4607 (362)
R-work	0.2406 (0.2898)
R-free	0.2786 (0.3737)
Number of non-hydrogen atoms	23 591
Macromolecules	23 256
Ligands	75
Solvent	260
Protein residues	2976
RMS(bonds)	0.008
RMS(angles)	0.95
Ramachandran favored (%)	91.36
Ramachandran allowed (%)	6.88
Ramachandran outliers (%)	1.75
Rotamer outliers (%)	5.73
Clashscore	10.71
Average B-factor	61.78
Macromolecules	61.95
Ligands	66.92
Solvent	45.24

Statistics for the highest-resolution shell are shown in parentheses.

${\mathrm{R}}_{\mathrm{merge}}={\sum }_{\text{hkl}}\left({\sum }_{\mathrm{i}}\left[\left|{\text{I}}_{\text{hkl},{\mathrm{i}}^{-}}<{\mathrm{I}}_{\text{hkl}}>\right|\right]\right)/{\sum }_{\text{hkl}}{\sum }_{\mathrm{i}}{\text{I}}_{\text{hkl,i*}}$

As this unit cell was nearly identical to that of the previously characterized native huGDH crystals,¹⁸ molecular replacement was used to determine this structure. Phenix and COOT were used for structure determination and refinement, respectively. Initially, the model was subjected to 30 steps of rigid body refinement without noncrystallographic (NCS) restraints and then energy minimization with NCS restraints to make sure the coenzyme binding domain was free to move if necessary. Simulated annealing was then applied with NCS restraints, followed by energy minimization and individual B value refinement. The histidine at position 454 (mutation site) was initially modeled as an alanine. After two rounds of crystallographic refinement, a tyrosine residue was modeled with respect to the electron density observed in a 2F_o-F_c map. This was further corroborated by analysis of a F_o-F_c map with both positive and negative contours. The final refinement statistics are summarized in Table 1.

2.2 |. Sequence analysis

Bovine and human GDH share 100% sequence identity in the allosteric binding sites. Thus, when residue 387 was found by modeling comparison to be identified as asparagine in bovine but lysine in human, the bovine GDH sequence was reinvestigated.^19,20 The bovine GDH sequence originally used in all bovine GDH structures came from a protein sequence determined by chemical modification published in 1972.²¹ Five residues were misidentified: N387 K, G47S, A248V, V271I, and A272T. Of the five, only N387 K was located in a binding site and was determined to be the most deleterious to previous interpretations of function.

2.3 |. Model refinement

Crystal structure 3 MW9 containing the incorrect sequence was minimized by conjugate gradient for 4000 steps using the NAMD software. The root-mean-square deviation was calculated for each atom using the incorrect sequence crystal structure as the reference state. A number of atoms located near the allosteric ligand binding pocket moved greater than 3 A which is unusual for a 2.7 Å structure. We compared the NADH binding pocket (Figure 1) of 3MW9 (incorrect sequence) to a crystal structure of H454Y mutant human GDH. When comparing both structures, it was evident that the sequences near the NADH allosteric site were not identical when considering the free phosphate molecules located near the NADH binding pocket in the mutant human GDH structure, which should be in a similar location as the NADH β-phosphate group in 3 MW9 (incorrect sequence). Thus, residues located near the NADH binding pocket, particularly those near the NADH phosphate group, were further analyzed via sequence alignment.

2.4 |. Free energy simulation

As we are interested in ligand binding to the allosteric sites, we calculated the consequences of the sequence/structure issue on binding free energy differences in the presence and absence of NADH. The GDH model used for simulation was the homotrimer (see Figure 1). We considered the difference in unbound versus NADH bound to pdb 3MW9 in the previously published form and with the correct sequence. Each pair of monomers contained a NADH molecule bound to the NADH binding site (3 NADH molecules bound total) and each monomer initially contained the correct residue (Lys 387). This structure was placed in 0.1 M NaCl solution, minimized for 6000 steps and equilibrated for 1 ns in an NPT ensemble with a 2 fs time step. The CHARMM36 force field was used for atomic topology and parameters. Particle-mesh Ewald with tinfoil boundary conditions was used for the long range electrostatic calculations. The free energy was then computed for changing Lys to Asn at residue 387. The dual topology technique was used to calculate the binding free energy, where λ = 0 state is Lys 387 and λ =1 state is Asn 387. The binding free energy simulations were run for over 100 ns per λ. The calculation was divided into 16 windows and the free energy was calculated using free energy perturbation techniques (Equation (1)), where k_B is the Boltzmann constant, T is the temperature at 300 K and < > represents the ensemble average.²²

$$\Delta G=-{\text{k}}_{\text{B}}\text{T}\ln <e^{-\beta \Delta U}>$$ (1)

The binding free energy difference (ΔΔG) was calculated for the thermodynamic cycle shown in Figure 4. ΔΔG may be computed as the difference of the vertical legs, which is equal to the difference of the horizontal legs. The horizontal legs of the thermodynamic cycle (ΔG₁ and ΔG₃) represent the free energy of changing lysine to asparagine with and without NADH. ΔG₁ and ΔG₃ are the sums of their electrostatic terms and van der Waals terms.

FIGURE 4.

Thermodynamic cycle used to calculate the binding free energy difference of asparagine versus lysine as residue 387

3 |. RESULTS

3.1 |. Structure of the H454Y mutant

Several mutations in the GTP binding site and antenna desensitize GDH to GTP inhibition.^23,24 While the HHS H454Y mutant has been extensively studied and is used as an archetypal example of HHS,^1,25 the structural effects of the mutation had not been determined. The structure of H454Y was determined to 2.7 Å (Table 1, Figure 2). Using the alignment routine in Pymol,²¹ the structure of apo human GDH was compared to apo H454Y human GDH and the root-mean-square difference between the two structures was 0.35 Å(using 2855/3872 atoms). As with all apo GDH structures, the outermost region of the NAD⁺ binding domain has weak density because of its inherent flexibility in the absence of bound substrates. However, the electron density around the GTP site was well-resolved and clearly showed the H454Y mutation (Figure 2A). As shown in Figure 2B, the H454Y mutation does not affect any of the neighboring residues. Therefore, it is most likely that the H454Y mutation directly interferes with GTP binding rather than by deforming the binding site. To better examine this, the H454Y mutation was modeled in the bovine GDH structure (H450Y in bovine GDH), where we have the structure of GTP bound. As the huGDH H454Y structure demonstrated that the mutation does not affect other residues in the GTP binding site, we made the H450Y mutation in bovine GDH using the VMD plug-in, Mutator. The complex placed in 0.1 M NaCl and GDH was held fixed while the solvent was minimized for 2000 steps. The solvent was then held fixed and GDH and its ligands were minimized for 200 steps. The entire system was then minimized for 2000 steps. As shown in Figure 2C, the effects of the tyrosine at 450 appear to be modest. The hydroxyl group is very close to the β-phosphate of the GTP and may sterically interfere with the phosphate interactions. Perhaps more importantly, the His to Tyr mutation would likely eliminate the charge interaction.

FIGURE 2.

Structure of the HHS H454Y mutant. (A) Shown here is the electron density of the region around the human GDH H454Y mutation that causes HHS. (B) This figure shows the structures of the sidechains that contact GTP before (green) and after the mutation. (C) Shown here are the energy minimized structures of bovine GDH complexed with GTP/NADH/Glu before (green) and after the H450Y (H454Y in human) mutation [Color figure can be viewed at wileyonlinelibrary.com]

3.2 |. Electron density versus protein sequence

In preparation for examining GDH allostery, we compared the wild type crystal structure to its minimized structure and searched for residues that moved more than 3 Å away from their initial position. This was done to see if there were any significant deviations in the GDH allosteric ligand binding pockets. The atoms in the GTP binding site remained well within 3 Å of their respective crystal position while atoms in the NADH binding site deviated greater than 3 Å. This led to the further investigation of residues in the NADH binding site.

To ensure consistency of the NADH binding pocket between the H454Y mutant GDH and the wild type GDH, both the residue composition and conformation were compared. It was found that residue 387 was a lysine in the H454Y mutant structure and asparagine in the bovine GDH structure. As mentioned earlier, there should be a 100% identity between the bovine and human GDH sequences near the NADH binding site. However, the older published bovine GDH sequence had residue 387 incorrectly identified as asparagine instead of lysine, which was used in the crystal structure refinements. Residue 387 is located at the interface of the allosteric NADH binding site. Table 2 shows the distances between residue 387 and NADH in the incorrect, correct, and mutant GDH crystal structures.

TABLE 2.

Distances (Å) between residue 387 and ligand in NADH binding pocket in the incorrect bovine GDH, correct bovine GDH, and H454Y mutant human GDH

GDH complex	Ligand	387 to ligand distance (Å)
Incorrect bovine seq (N387)	NADH (βPO43− group)	6.2
Correct bovine seq (K387)	NADH (βPO43− group)	3.9
H454Y mutant human GDH	PO43−	3.2

As shown in Table 2, the distance between NADH and residue 387 is reduced by 63% when correcting residue 387 from an asparagine to a lysine. This moves the NADH β-phosphate one shell (water diameter) closer to the NADH binding site, which is similar to the free PO₄³⁻ ligand in the H454Y mutant NADH binding site. This geometric difference affects the binding free energy of NADH interacting with residue 387. Thus, we calculated the binding free energy difference of lysine vs asparagine in the presence and absence of NADH to compute the significance of the N387 K sequence error for future drug or ligand design.

3.3 |. Evidence for the sequence error in the electron density

Shown in Figure 3 is the difference map (F_o-F_c) when 387 is an asparagine and the 2F_o-F_c map after it was substituted by a lysine and rerefined. In the difference map, the red density represents negative density where atoms should not be in the model while the positive blue density shows areas lacking atoms in the model. After refinement, the 2F_o-F_c density well-describes the lysine sidechain. Thus, the crystallographic data had pointed to, and been consistent with, the error in the published sequence.

FIGURE 3.

Comparison of asparagine and lysine in 3 MW9 electron density at residue 387. Asparagine (left) in the electron density is unfavorable as indicated by the negative and positive electron density, (red and blue, respectively) in the Fo-fc maps. However, these differences disappear when lysine is used instead. It is important to note that the position of asparagine and lysine are different in the figure because the rerefinement of 3 MW9 resulted in a conformational change when using lysine instead of asparagine in the protein sequence [Color figure can be viewed at wileyonlinelibrary.com]

3.4 |. Computed free energy penalty

The total computed binding free energy penalty is +5 kcal/mol per NADH between the incorrect and correct sequence forms. This was an average from all three binding sites (Figure 4). The errors were calculated using the block SE (BSE) method²⁶ with a block size of 1000. The maximum errors in any window (see Equation (1)) for ΔG₁ electrostatics, ΔG₁ van der Waals, ΔG₃ electrostatics, and ΔG₃ van der Waals were ± 0.086 kcal/mol, ±0.043 kcal/mol, ±0.047 kcal/mol, and ± 0.046 kcal/mol, respectively. The standard square root of the sum of each window’s variance for ΔG₁ electrostatics, ΔG₁ van der Waals, ΔG₃ electrostatics, and ΔG₃ van der Waals were ± 0.20 kcal/mol, ±0.088 kcal/mol, ±0.13 kcal/mol, and ± 0.11 kcal/mol, respectively. We can use a more conservative propagation of errors taken as the sum of the 16 window errors, shown in Table 3. The size of the computed sequence error effect on the free energy of binding for the native ligand, NADH, would preclude any productive sort of computational screening of other compounds based on free energy calculations with such a structure.

TABLE 3.

Binding free energy differences in kcal/Mol for the trimer with summed window errors using BSE n = 1000

ΔG1 = 107 ± 1.1	ΔG3 = 92 ± 0.94
ΔG1 electrostatics = 99 ± 0.77	ΔG3 electrostatics = 83 ± 0.52
ΔG1 van der Waals = 8 ± 0.36	ΔG3 van der Waals = 9 ± 0.42

3.5 |. Structural analysis of the NADH/ADP/ECG site

As mentioned above, the NADH allosteric site is particularly puzzling since it binds both activators (ADP) and inhibitors (NADH and ECG/EGCG). While we have not yet determined the structure of GDH complexed with the compound 75-E10, our previous binding studies demonstrated that it likely binds to this same site as well.¹¹ Shown in Figure 5 is the electron density of bound NADH, ADP, and ECG bound to this site. In all cases, the quality of the electron density improved upon rerefining these complexes with the corrected sequence. With the change of N387 to K387, it is necessary to reexamine the interactions between GDH and the various allosteric regulators that bind to this site to better understand how the various ligands can cause such disparate effects (Figure 6). The view in this figure is approximately parallel to the “pivot helix,” looking into the interface between two subunits within one of the trimers. To emphasize that this allosteric site is at the interface between two adjacent subunits, the carbon atoms in the stick models are shown in light blue or white. Each of the figures is labeled as either being “open” or “closed” to denote the conformation of the catalytic cleft. It is immediately evident that the newly introduced K387 interacts with all three ligands in both the open and closed conformations while the previous erroneous N387 did not.

FIGURE 5.

Electron density for the rerefined ligands bound to the NADH/ADP/ECG site [Color figure can be viewed at wileyonlinelibrary.com]

FIGURE 6.

Rerefined binding environments for (A) NADH, (B) ADP, and (C) ECG. Shown here are the potential contacts between GDH and the various ligands. As this binding site is comprised of two adjacent subunits within the trimer, the carbon atoms are colored white or light blue to designate the difference. Also noted is whether the catalytic cleft is in the open or closed mouth conformation. Note that the K387 residue is now in contact with all three ligands whereas the incorrect N387 did not in the previous model [Color figure can be viewed at wileyonlinelibrary.com]

Also noted on the figures are the results of our previous site-directed mutagenesis studies.¹⁰ In the case of ADP, R86S, and S393I mutations had deleterious effects on activation. The R86S mutation effects are not entirely surprising since the guanidine group on the arginine stacks up against the purine ring in addition to interacting with the ribose hydroxyl groups. The effects of the S393I mutation are more complex in that S393 interacts with all of the bound ligands but the mutation also abrogates GTP inhibition, even though it does not contact GTP. Therefore, it is not clear whether the S393I effects are entirely due to changes in the binding interface or whether it may also be related to subunit/subunit communication as implied by the loss of ADP and GTP regulation when the antenna is genetically removed.¹ R459 interacts extensively with the β-phosphate of ADP. When mutated to an alanine, ADP activation is completely abrogated. However, using TNP-ADP as a fluorescent ADP analog,¹ it was clear that the R459A mutation did not block ADP binding. This was somewhat unexpected and suggested that the mutation did not affecting binding but rather the conformational changes necessary for ADP activation. The existence of this unusual phenomenon was also suggested by the fact that the R459A mutation increased the ED50 for diethylstilbestrol (DES) and GTP by nearly 5-fold and palmitoyl-CoA by approximate 3-fold. From the structure of DES, it is likely to bind to the GW5074/bithionol site deep within the enzyme rather than the ADP site.¹⁴ Therefore, the effects of the R459A mutation is not simply removing a charge on the pivot helix that interacts with ADP. The effects of the D119A mutation is more simply understood. As shown in this figure, the carboxylic acid group of D119 is close to the β-phosphate of ADP and the charge on this acid is at least partially negated by the basic residues in the area. Unlike the other mutations, the D119A mutant enhances the activation of ADP by approximately 2-fold. It seems most likely that this mutation enhances binding and/or action by increasing the positive charge in that area.

The effects of NADH bound to this site are less clear. NADH inhibition is only seen at very high NADH concentrations, well above the measured binding of NADH to this site.² Nevertheless, it is well documented that GTP and NADH bind in a synergistic manner²⁷ and this structure is in the “closed” conformation where the catalytic mouth is closed tightly on the abortive complex of NADH+Glutamate.⁸ Two of the more notable changes are the positions of H209 and R459. In the open conformations in the ECG and ADP structures, both of these residues are rotated down onto the bound ligand. Here, H209 rotates and interacts with the bound GTP while R459 is lifted up and away from the NADH because of the closed conformation. The adenosine/ribose rings of the bound NADH closely mimic that of ADP. However, the nicotinamide/ribose moiety hooks back down into the protein near the base of the antenna. It is not clear whether this nicotinamide/ribose moiety binding here causes inhibition and why NADH inhibition requires such a high concentration relative to its Kd.

The structure of ECG is similar to ADP in that it is bound to the open conformation of GDH in spite of the fact that NADPH was added to the crystallization mixture.¹⁰ In this conformation, H209 swings over to interact with the gallate ring (ring D) but does not interact to the extent that the α-phosphate does in the ADP structure. Also, similar to ADP, R459 swings down from the pivot helix to interact with the gallate ring. While the D119A mutation enhances ADP activation, the same mutation abrogates ECG inhibition. The difference is likely due to the fact that the carboxyl side chain forms hydrogen bonds with the C ring of the flavanol moiety that are lost with the D119A mutation while the negation of these charges eliminates the charge conflict around the β-phosphate in ADP. As with ADP, the R86S mutation blocks ECG inhibition presumably by eliminating a hydrogen bond to the OAR atom of the C ring of the flavanol group.

The N387 K fixed and refined structures all yield more accurate details of the ligand/GDH contacts. However, it is still puzzling why some ligands binding to this site activate while others inhibit the enzyme. As we noted previously, GDH is a highly flexible enzyme that undergoes large conformational changes during each catalytic cycle.² To better quantify these structural changes, several measurements were made comparing the open (ADP/GDH) and closed (GTP/NADH/GLU/GDH) structures (Figure 7). To better illustrate the movements described below, the ascending antenna helices (red helix) of the open (lighter hues) and closed (darker hues) conformations were aligned in this figure (noted as #1). In the first measurement, the center of masses of the pivot helices were calculated and found to move ~2.5 Å away from each other as the catalytic site closes (#3 and #4). Similarly, when the bottom quarter (N-terminal end) of the ascending helices were analyzed, they also moved ~2.5 Å away from each other as the catalytic site closes. However, the distances between the top quarters of the ascending helices remained unchanged as the catalytic site closed. Essentially, that portion of the three helices (#5) is the pivot point about which the ascending helices rotate about each other as the catalytic site opens and closes. The descending helix of the antenna (#6) is caught right in the middle of this conformational change with the top (N-terminus) being fixed near the pivot axis and the bottom (C-terminus) being carried along for the 2.5 Å expansion as the catalytic site closes. For this reason, the helix itself is distended in the closed conformation.

FIGURE 7.

(A) Stereo pair: Expansion and contraction of the NADH/ADP/ECG binding site. GDH trimers from the open mouth (lighter hues) and closed mouth (darker hues) are aligned along one of the ascending antenna helices (#1). As the catalytic mouth closes, the ascending antenna helices (#2) and the pivot helices move ~2.5 Å away from each other (#3, #4) causing an opening of the NADH/ADP/ECG binding site. This rotation occurs about the upper quarter of the ascending antenna helices (#5), causing a distension of the descending antenna pivot helix (#6). (B) Schematic of the overall movement in GDH as the catalytic mouth closes. GDH binds (#1) causing the NAD binding domain to rotate and close upon the substrate (#2). The base of the antenna rotates counter clockwise (#4) and the NAD binding domain rotates away from the center of the hexamer. These rotations are twisting motions that cause the core of the hexamer to compress (#5) [Color figure can be viewed at wileyonlinelibrary.com]

From these results, the effects of the various activators and inhibitors binding to this same site may be due to more than particular GDH/ligand interactions but rather due to the large conformational changes occurring in this subunit interface. For example, one of the HHS mutations that abrogate GTP inhibition lies in this helix¹⁸ and the deletion of the antenna region eliminates both GTP and ADP regulation.¹ The S393I mutation not only affects ADP activation but GTP inhibition as well. From these and several other observations, it is apparent that the widely varying effects of ADP, NAD⁺, ECG, and 75-E10 binding to this site may not be due to interactions with particular residues but rather their larger effects on this complex movement between adjacent subunits within the trimers.

The schematic in Figure 7B summarizes these movements in entire hexamer. When the substrate binds (#1), the NAD binding domain rotates by ~20° about the pivot helix (#2) and tightly closes the catalytic cleft (mouth). The base of the antenna rotates away from that subunit (#3) and pushes into the NAD binding domain of the adjacent subunit. In this side view of the adjacent subunit, closure of the catalytic cleft from #2 rotates the NAD binding domain away from the center of the hexamer (#4). This, in turn, opens up the ADP binding site (#5). All of these rotations and twisting motions compresses the core of the hexamer (#6).

To better visualize the effects of these rotations and movements on the ADP binding site, Figure 8 shows a surface rendering of this binding site in the open and closed conformations. As with Figure 7A, the open conformation is represented by lighter hues and the two adjacent subunits are colored in green and blue. The view here is approximately parallel to the antenna, looking down into the allosteric site. In general, as the catalytic site closes, the gap between the two subunits expands, enlarging the binding pocket. In the case of the NADH/GLU/GTP/GDH complex, this allows for the nicotinamide/ribose moiety of NADH to penetrate further into the subunit interface. Interestingly, this could explain the binding synergy between GTP and NADH.²⁸ Our previous structural analysis showed that the triphosphate binding site for GTP only opens up when the catalytic site is closed. Here, it seems that NADH would also prefer the closed conformation by virtue of the enlarged binding pocket. Therefore, it seems likely that the positive binding cooperativity between these two ligands²⁸ is due to this selection of the closed conformation. As the catalytic site opens, this binding cavity compresses and the pivot helix rotates back towards it. This places R459 directly above it and makes a smaller binding pocket that is better suited to the ADP structure. An overly simplistic interpretation of this effect in relationship to ADP activation is that the binding of ADP helps to compress this allosteric site that in turn makes it easier to open the catalytic cleft and release the product. However, the ECG complex appears to argue against this simple model. ECG was found to bind to the open conformation in spite of the addition of NADPH to the crystallization mixture. Therefore, ligands that prefer the smaller pocket found in the open conformation are not necessarily all activators.

FIGURE 8.

Surface details of the NADH/ADP/ECG binding site as a stereo pair. In the closed mouth conformation, the subunits (green and blue) spread apart by 2.5 Å, making more room for the bound NADH in the top figure. In contrast, the subunits compress by 2.5 Å in the open mouth conformation, making for better interactions with the smaller allosteric ligands [Color figure can be viewed at wileyonlinelibrary.com]

4 |. DISCUSSION AND CONCLUSIONS

The original published sequence of bovine GDH has been used for a large number of studies^{8–11,13–16} but unfortunately contained a few mistakes including that at residue 387.¹⁹ Those studies included 18 X-ray diffraction and cryo-electron microscopy structures.^{8–11,13–16} Lysine is the correct amino acid identity of residue 387 in the allosteric NADH binding site, not asparagine. The thermodynamic impact of this mistake is shown to be +5 kcal/mol per NADH binding site, which would disrupt most modeling and virtual screening studies for allosteric compounds. In addition to residue 387, four other residues were corrected in the bovine GDH sequence, specifically G47S, A248V, V271I, and A272T. However, unlike N387 K, these residues were located far from allosteric and catalytic binding sites and are unlikely to require significant changes in interpretations.

The adjusted structures were rerefined with particular attention being paid to the ligands bound to this allosteric site. Notably, the quality of the density for NADH, ADP, and ECG all improved with N387 K adjustment. However, as before, the density for the outermost regions of the NAD binding domain were very poor in the open conformation compared to that of the closed structure (GTP/NADH/GLU/GDH) even when the positions of all 12 NAD binding domains were refined individually. This strongly suggests that the NAD binding domain has marked flexibility when not clamped down upon substrate.

Together, these results suggest a modification of our previous model for GDH allostery.¹¹ Whereas emphasis was placed on the rotation of the pivot helix and the interactions between R459 and the β-phosphate of ADP, it now seems more likely that allostery at this site depends upon the expansion and contraction between subunits within the trimer as the catalytic site closes and opens, respectively. As this motion involves a wide area of the enzyme, including the entire antenna, there are numerous sites where allosteric ligands and mutations could affect catalysis. Further, it becomes less of a paradox as to how different ligands can bind to the same location and cause opposite effects as this expansion/compression process is such a large system of conformational changes.

Previous biochemical data on the H454Y mutant also suggests that allosteric regulation of GDH is more complex than just ligands binding to their corresponding sites.¹⁷ The Ki for GTP in wild type human GDH is ~40 nM and this increases by ~400-fold (16 μM) with the H454Y mutation. Interestingly, when the GDH from lymphoblasts was examined from patients with the H454Y lesion, the inhibition pattern was monophasic with a Ki for GTP of ~250 nM. The monophasic nature of the inhibition curve suggests that the GDH in the patients is a chimeric mix of the two types of subunits. If both types of subunits are approximately equally expressed in the patients, then it follows that the effects of the H454Y mutant are greatly dampened by having even a fraction of wild type subunits. This would be consistent with the implications outlined above where subunit communication plays a major role in the effects of allosteric regulators. In this case, the apparent cooperativity among the wild type subunits is sufficient to overcome the lack of GTP binding to the H454Y subunits. Importantly, this is in agreement with our previous finding that eliminating the antenna that does not bind any of the allosteric effectors, eliminates ADP, GTP, and ECG regulation.¹ Clearly, further studies are necessary to fully understand the structural details of the structural changes that occur in the hexamer after allosteric ligand binding.