Allosteric discrimination at the NADH/ADP regulatory site of glutamate dehydrogenase

Abstract

Glutamate dehydrogenase (GDH) is a target for treating insulin‐related disorders, such as hyperinsulinism hyperammonemia syndrome. Modeling native ligand binding has shown promise in designing GDH inhibitors and activators. Our computational investigation of the nicotinamide adenine diphosphate hydride (NADH)/adenosine diphosphate (ADP) site presented in this paper provides insight into the opposite allosteric effects induced at a single site of binding inhibitor NADH versus activator ADP to GDH. The computed binding free‐energy difference between NADH and ADP using thermodynamic integration is −0.3 kcal/mol, which is within the −0.275 and −1.7 kcal/mol experimental binding free‐energy difference range. Our simulations show an interesting model of ADP with dissimilar binding conformations at each NADH/ADP site in the GDH trimer, which explains the poorly understood strong binding but weak activation shown in experimental studies. In contrast, NADH showed similar inhibitory binding conformations at each NADH/ADP site. The structural analysis of the important residues in the NADH/ADP binding site presented in this paper may provide potential targets for mutation studies for allosteric drug design.

1. INTRODUCTION

The role of glutamate dehydrogenase (GDH) in disease has been extensively studied and shown to be involved in some forms of the hyperinsulinism hyperammonemia syndrome (HHS).1, 2 Point mutations in a number of residues in and around the binding site of the major GDH inhibitor, guanosine triphosphate (GTP),3, 4, 5, 6, 7 lead to uncontrolled catabolism of glutamate. There are also a number of HHS mutations located outside the GTP binding site, particularly in the antenna region, that also affect allosteric regulation of GDH.2 This increased activity of mutant GDH leads to excess insulin secretion and accumulation of ammonium in HHS patients.8, 9

GDH catalyzes the reversible oxidative deamination of glutamate to 2‐oxoglutarate in the presence of coenzyme nicotinamide adenine dinucleotide (NAD(H)) or nicotinamide adenine dinucleotide phosphate (NADP(H)).10 In HHS, the increased production of 2‐oxoglutarate feeds into the Krebs cycle, stimulating adenosine triphosphate production, and increasing insulin secretion from β‐cells in the islets of Langerhans of the pancreas.8, 9 Thus, dysregulated GDH activity leads to excess insulin secretion and hypoglycemia in HHS patients. In addition, HHS depletes glutamate that is necessary to produce the important urea cycle intermediate N‐acetylglutamate, resulting in the accumulation of ammonium.8, 9 Mammalian GDH is allosterically regulated by several metabolites that are indicative of cellular energy levels including adenosine diphosphate (ADP), NADH, and GTP.7, 10, 11, 12 Therefore, under normal physiological conditions in mammals, GDH functions as an energy sensor of the pancreatic cells and correspondingly increases or decreases insulin secretion.8, 9, 13, 14

The major allosteric inhibitor of GDH is GTP that binds to an allosteric site above the catalytic mouth at the base of the antenna of mammalian GDH.7, 15 GTP binding stabilizes closing of the catalytic mouth and prevents product release (the rate limiting step).7, 15 In contrast to GTP, ADP allosterically activates GDH by binding to the NADH/ADP binding site, abrogating GTP binding.16, 17 The NADH/ADP binding site is beneath the pivot helix, behind the catalytic cleft.15, 16, 17 ADP binding facilitates the opening of the catalytic mouth, as seen in X‐ray crystal structures,16 thereby promoting product release as observed in binding and kinetic studies.10, 18 In addition to the active site, coenzyme NADH can bind to a secondary site, the NADH/ADP binding site, and enhances GTP inhibition of GDH activity by slowing product release.15, 17

GDH can function as a soft‐polymer of hexamers or as individual hexameric units in the mitochondrial matrix.19 The GDH hexamer is a dimer of homotrimers with approximately 500 amino acids per monomer.15, 16, 17, 20, 21, 22, 23 GDH structures show that each monomer has a GTP binding site as well as an NADH/ADP binding site, thus, each GDH hexamer can bind six GTP molecules, six ADP molecules, and six NADH molecules at their respective allosteric binding sites. The functional form of GDH is a hexamer,24 but for computational expedience, the GDH trimer was utilized in our previous17 and current studies. Figure 1 shows the structure of the GDH trimer and with NADH bound to the NADH/ADP binding site.

Figure 1.

Structure of half of the GDH hexamer with three NADH molecules bound at the NADH/ADP binding sites (PDB 6DHD). Top view is on the left and side view is on the right. Each monomer is colored magenta, green, or yellow. NADH molecules are colored in black and highlighted in light blue

While the role of GDH in several diseases has been elucidated, developing agonists and antagonists to target GDH remains a challenge because of its complex regulation.6, 7 There are several models and theories discussing the mechanism of GDH activation and inhibition, but they have yet to be validated experimentally.17 The X‐ray structures of wild type GDH in the open conformation and closed conformation as well as GDH mutant structures have been determined.15, 16, 17, 20, 21, 22, 23 In addition, many of the structures that initially were determined using the incorrect protein sequence have been updated with the correct sequence and showed improved electron density of the NADH/ADP/ECG binding site.17 This allosteric site is of particular importance, since it binds both activators (ADP16 and presumably the synthetic activator 75‐E107) and inhibitors (NADH⁵, and the green tea polyphenol EGCG/ECG15). It is not at all clear how these different ligands can cause opposite effects by binding to the same site.

To better understand this complex regulatory site, the computational free energy technique thermodynamic integration (TI)25 was used to compute the relative binding free energy difference of NADH versus ADP binding to GDH. Free energies calculated using free energy perturbation (FEP)26 are included in the supplemental information section. The conformational changes and free energy difference going from the GDH inhibited state (NADH bound) to the GDH activated state (ADP bound) were modeled. These simulations show that the NADH to ADP transition partially opens the catalytic mouth, which is consistent with previous models for ADP activation. In addition, there are differences among the ADP sites that could explain the limited activation caused by ADP binding alone. In contrast, the energy and interactions of NADH binding are equivalent in all of the subunits. Therefore, the differences between activators and inhibitors binding to the ADP/NADH allosteric site are more subtle than previously suggested and appear to be driven by the details of the ligand/GDH binding.

2. METHODS

2.1. Software

Nanoscale molecular dynamics (NAMD) version 2.1027 was used for running molecular dynamics simulations and visual molecular dynamics (VMD)28 version 1.9.3 was employed for analysis of the trajectories. PROPKA 3.129 in PlayMolecule30 was used to determine the protonation state of GDH at pH = 7 for each structure used and the psfgen, solvate, and auto‐ionize plugins in VMD were used for further preparation of the systems.

2.2. Ligand system (ΔG ₃ computation)

For the solute in solution leg of the cycle one NADH molecule was transformed into one ADP molecule using the dual topology paradigm.31, 32 This aqueous system consisted of the solute molecule (PDB ID: 6DHD) placed in a 39 Å × 39 Å × 39 Å cubic water box with 0.1 M NaCl. An constant number, pressure and temperature (NPT) ensemble using a Langevin barostat and Langevin thermostat was employed. The integration time step was 2 fs. Minimization of the system involved three stages utilizing the conjugant gradient method: (a) fixing the ligands and minimizing the solvent for 40,000 steps, (b) fixing the solvent and minimizing the ligand for 2,200 steps, and (c) minimizing the entire system for 40,000 steps. This process is used before equilibration of the system. Each of the 16 lambda intermediates was minimized (without fixing atoms) for 2,000 steps, equilibrated for 2 ns and run for 18 ns per intermediate. A total of 288 ns was run for the entire free energy path. Simulations were run in an NPT ensemble at 300 K and 1 atm with TIP3P explicit solvent model.33

2.3. Protein‐ligand complex system (ΔG₁ computation)

Using PROPKA 3.1 in PlayMolecule, GDH was protonated for a solution pH of 7. The apo form of GDH was used to determine the protonation state. It is important to note that the protonation state of the binding sites were compared between the crystal structures with ADP versus NADH bound. It was found that PROPKA 3.1 produced the same protonation state for all three structures. The protonation state of amino acids, particularly that of histidine, at the ligand binding site is not trivial and could influence the calculated binding free energy.34, 35, 36 The GDH homotrimer, specifically chains A, B, and C, was complexed with three ligand molecules: one ligand per NADH/ADP binding site. The system was placed in a 121 Å × 121 Å × 121 Å cubic water box with 0.1 M NaCl. The starting structure used was the GDH‐NADH bound complex (PDB ID: 6DHD) because the X‐ray structure was determined to higher resolution. The simulation began with NADH and ended with ADP. Thus, the simulation involved the gradual transformation of NADH to ADP at the NADH/ADP binding sites. The sampling methods used for the small system was also used for the large system, except that each lambda intermediate was equilibrated for ~5 ns and run for 23 ns instead of 18 ns to achieve better convergence. Thus, the total set of trajectories was 368 ns. Simulations were run in an NPT ensemble at 300 K and 1 atm with TIP3P explicit solvent model.33 It is important to note that the backwards transformation was not used because we wanted to reduce the expense of the computation which would be compounded by the modest resolution of the starting crystal structures. However, adding the backwards transformation can provide improved sampling thus precision of the free energy calculation under some circumstances.

2.4. Free energy calculations

Free energies were calculated using TI. The Hummer et al. correction37 was added for the finite size effects of changing the charge of the ligand (−2 to −3) using Particle Mesh Ewald (PME) for the long‐range electrostatic calculations. Equations 1, 2 show the formulas for calculating TI and the Hummer et al. correction, respectively, for charges in electrons and energies in kcal/mol. Here q₀ is the charge of the initial state, q ₁ is the charge of the final state, L is the length of a cubic box, and ζ_Ewald is the Ewald self‐term of a cubic lattice (ζ_Ewald = −2.837/L). Free energies computed using TI were integrated using the 3/8 Simpson's method. Also, it is important to note that the lengths used to compute the Hummer et al. correction (L = 39 Å and L = 121 Å) are approximate average lengths and that the length of the simulation fluctuated in the NPT system.

$$∆G_{\mathrm{i}}={\int }_{\mathrm{i}}^{\mathrm{i}+1}〈\frac{\partial U\left(\lambda \right)}{\partial \lambda }〉{\mathit{\lambda }}_{d\lambda }$$ (1)

$$\mathrm{\Delta }{G}_{\text{correction}}=+\frac{1}{2}{\mathrm{\zeta }}_{E\text{wald}}\left(q_1^2-q_0^2\right)\times 332$$ (2)

3. RESULTS AND DISCUSSION

3.1. Binding free energy calculation

Free energy changes presented in this paper were computed using TI (see supplemental information for computations calculated using FEP). The binding free energy ΔG₁ is equal to −50.90 kcal/mol (ΔG ₁ per binding site is −16.97 kcal/mol) and is comparable to ~ −20 kcal/mol association free energy of ADP‐protein binding studies.38 The alchemical free energy change ΔG₃ is equal to −14.64 kcal/mol and is comparable to the scale of free energy of the reported solvation free energies of adenine −12 kcal/mol39 and −16.3 kcal/mol.40 The binding free energy difference computed using TI (ΔΔG _TI = −0.3 ± 1.882 kcal/mol; see Table 1, Figure 2) is within the experimental binding free energy difference range (ΔΔG = −0.28 to −1.7 kcal/mol; see Table 1).11, 18 Figure 3 shows the free energies at each intermediate for ΔG₁ and ΔG₃.

Table 1.

Computed binding free‐energy differences compared with experimental binding free energy differences

Experimental binding constants		Experiment (kcal/Mol)	TI (kcal/Mol)
K D NADH	K D ADP	ΔΔG exp	ΔΔG TI
57 × 10−6 Ma	3 × 10−6 Ma	−1.7	−0.3
57 × 10−6 Ma	35 × 10−6 Mb	−0.275

ADP, adenosine diphosphate; TI, thermodynamic integration.

^aKoberstein, R. and Sund, H. (1973).18

^bFrieden, C. (1959).42

Figure 2.

Computed relative binding free‐energy difference, ΔΔG, of the alchemical conversion of NADH to ADP bound and unbound to GDH using TI is equal to −0.3 kcal/mol. Note that ΔΔG = ΔG ₁ − ΔG ₃ = ΔG ₂ – ΔG ₄

Figure 3.

Free energy calculations using thermodynamic integration (TI) per λ intermediate for ΔG ₁ (top) and ΔG ₃ (bottom)

The errors were calculated using the block standard error method41 with a block size of 1,000. The maximum errors for ΔG ₁ and ΔG ₃ are ±0.317 and ±0.471 kcal/mol, respectively. The standard square root of the sum of each window's variance for ΔG ₁ and ΔG ₃ are ±0.963 and ±1.882 kcal/mol, respectively. A more conservative propagation of errors taken as the sum of the 16 window errors is shown in Table 2.

Table 2.

Total error and errors per lambda window for ΔG ₁ and ΔG ₃ calculated using the block standard error method

λ window	ΔG 1(λ) (kcal/Mol)	ΔG 3(λ) (kcal/Mol)
Total	0.963	1.882
0 to 0.0625	0.317	0.471
0.0625 to 0.125	0.153	0.214
0.125 to 0.1875	0.142	0.121
0.1875 to 0.25	0.127	0.175
0.25 to 0.3125	0.110	0.121
0.3125 to 0.375	0.102	0.127
0.375 to 0.4375	0.093	0.138
0.4375 to 0.5	0.063	0.107
0.5 to 0.5625	0.142	0.167
0.5625 to 0.625	0.189	0.217
0.625 to 0.6875	0.200	0.267
0.6875 to 0.75	0.240	0.181
0.75 to 0.8125	0.183	0.219
0.8125 to 0.875	0.213	0.254
0.875 to 0.9375	0.197	0.344
0.9375 to 1	0.311	0.418

3.2. Structural analysis of conformational changes

Figure 4 shows the catalytic motion of GDH. The catalytic mouth in the closed position is shown in light blue while the open position is shown in orange. The mouth rotates approximately 18° between the open and closed state. The core region is colored in red and is typically bound to a core region of a second trimer (the GDH hexameric form is a dimer of trimers). However, the GDH trimer used is solvent exposed and lacks some protein–protein interactions found in the hexameric form.

Figure 4.

RMSD comparison of crystallographic GDH open (orange) conformation (PDB ID: 6DHK) relative to GDH closed (cyan) conformation (PDB ID: 6DHD). In purple is the GDH trimer in the closed conformation and in red is the core region in the open conformation, which is solvent exposed in our simulations

The root‐mean‐squared‐deviation (RMSD) of the open conformation (PDB ID: 6DHK) versus closed conformation (PDB ID: 6DHD) was compared to the GDH‐Ligands trajectories and the crystallographic structures.5, 16 Opening of the catalytic mouth was observed in the GDH trimer/3ADP complex at the end state (λ = 1; see Table 3). The conformation of the end GDH state diverges from the closed conformation by ~3 Å and approaches the open conformation by ~1.5 Å. To further investigate the conformational changes of GDH, the RMSD was determined for the regions of the protein that have been identified as catalytically important and specified in Table 3.

Table 3.

Root‐mean‐squared‐deviation (RMSD) comparison of crystallographic glutamate dehydrogenase (GDH) open conformation (PDB ID: 6DHK) and simulation relative to GDH closed conformation (PDB ID: 6DHD) at different regions of GDH

			Antenna region
	Core	Catalytic mouth	Antenna	Flexible loop	Pig tail	Pivot helix
Residues	50–175	250–350	400–418	418–425	425–448	448–475
RMSD (Å) crystals	2.04	6.70	1.90	4.72	3.41	1.70
Average RMSD (Å) trajectory	1.84	4.06	0.986	2.28	1.98	1.47
Maximum RMSD (Å) trajectory	3.63	6.82	1.77	4.66	2.90	2.187

Note: RMSD crystals refers to the difference between the open and closed crystal forms.5, 16 Average refers to the difference between the experimental ADP structure and that averaged over the λ = 1 trajectory. Maximum refers to the difference between the experimental ADP structure and that of the maximum deviation over the λ = 1 trajectory.

Table 3 shows the RMSD of the crystallographic open conformation as well as the conformations from the simulations relative to the closed conformation. The regions that deviate greater than 3 Å are the catalytic mouth (residues 250 to 350) and the antenna region (residues 440 to 448), specifically the flexible loop (residues 418 to 425) and pig tail regions (425–448). As expected, the core region has increased flexibility due to increased protein‐solvent interactions that are typically restricted by protein–protein interactions in the hexameric form. In addition, the catalytic mouth opens significantly, but rarely completely based on the 3.5 Å resolution X‐ray structure (PDB ID: 6DHK).16

This is an exciting result since it appears to agree with the proposed models for GDH activation and inhibition. During each catalytic cycle, the NAD binding domain of the catalytic mouth closes down upon the substrate and coenzyme. After catalysis, this domain opens to release the products and begin the catalytic cycle again. Therefore, activators such as ADP cannot “lock” the enzyme into a particular conformational state but rather facilitates the slowest step of the reaction, NAD domain movement and product release. The corollary is that inhibitors, such as GTP and NADH, act in the opposite manner and make the NAD binding domain more difficult to move. These simulations suggest that ADP facilitates the opening of the catalytic cleft while the NADH bound complex favors the closed state.

The average RMSD indicated some conformational changes within the trimer. Also, there were differences in structure and fluctuations between the monomers. For example, each monomer had the catalytic mouth open to different degrees and with different motions. Chain A, which showed the opening of the catalytic mouth but lacked ADP interactions with R459 (see next section for further discussion), had the largest opening of the catalytic mouth with an RMSD of ~7 Å. Chain B deviated ~5 Å, however, the mouth compressed inward towards the antenna and slightly along it. In contrast to chain A, interactions between R459 and ADP were observed in chain B. Chain C deviated the least, ~4 Å, and moved in a direction similar to that of chain A.

3.3. ADP binding at the NADH/ADP site

Our simulations agree with crystallographic studies17 in that residues K387, R396, H209 (HSE), R459, R491, D119, V120, R86 and H85 (HSE) are involved in ADP binding. However, the interaction of these residues varied at each site, with the exception of V120 and H85 that always interacted with the adenine purine ring. At one ADP site, R491 interacts with the α and β phosphate groups while D119 interacts with the 2′ hydroxyl group of the ribose and R86 interacts with the ribose ether oxygen. This is different from previously suggested crystal contacts found in bovine GDH‐ADP complexes16 in which R491 appeared to interact with the ribose hydroxyl groups, D119 appeared to interact with the γ phosphate and R86 appeared to interact with 2′ hydroxyl group. The second NADH/ADP site in the trimer shows a different ADP‐GDH interaction, one that better resembles the suggested interactions. Residues R396, R459, K387, and H209 interact with the β phosphate and R86 interacts with the ribose ether oxygen. R459 moves toward the β phosphate as the catalytic mouth opens, thereby possibly stabilizing the open conformation. The importance of R459 interacting with ADP phosphates for GDH activation was experimentally delineated by the R463A mutation (which is equivalent to the R459A in bovine GDH) that prevents ADP activation, possibly due to the elimination of the arginine‐phosphate electrostatic interaction when residue 459 is mutated to alanine.16, 43 The 2′ hydroxyl group on the ribose interacts with D119 for 14 ns then moves to interact with R491. The third NADH/ADP site shows the same β phosphate interactions with residues H209, R459, K387, and R396; however, D119 and R491 both interact with the 2′ hydroxyl group on the ribose. It is important to note that both crystal observations and our molecular dynamics models show each monomer varying in NADH/ADP binding site conformation as well as varying in the degree the catalytic mouth opens. Figure 5 shows each ADP binding site with their respecting ligand‐protein interactions.

Figure 5.

ADP interactions at each NADH/ADP binding site

The differences in ADP binding at each site can explain the complexity of ADP activation of GDH shown in binding.19, 44 Figure 5 shows that although ADP can bind at all 3 NADH/ADP sites, only two sites interact with R459 via the phosphate moiety and induce ADP activation (see sites 2 and 3 in Figure 5). R491 is another viable option for the ADP phosphate group to interact with and thus prevents interactions with R459 (see site 1 in Figure 5). Thus, R491 shows ADP binding but not activation as seen in the other sites. Since the sites are chemically equivalent, the statistical distribution of binding structures at equilibrium which have different secondary structural effects provides an unexpected picture complicating simpler mechanistic interpretations.

3.4. NADH binding at NADH/ADP site

We can analyze the first and last windows of the TIs as unconstrained molecular dynamics of the respective systems. The residues involved in NADH binding shown in figure 6 include K387, H209 (HSE), D119, R86, H85 (HSE), V120, H195 (HSE), I192, H391 (HSE), T87, N388, and Y382 (all make contact with NADH except for Y382). The residues that interact with the nicotinamide group are H195, I192, T87, H391, and N388. The delta amine on H195 interacts with the ribose hydroxyl; this interaction lasts for 20 ns. The carbonyl oxygens on I192 and T87 interact with the nicotinamide amine group. The hydroxyl group on T87 also interacts with the nicotinamide amine. During the last 3 ns of the simulation, the amine on the nicotinamide moves to interact with the delta amine on H391. The amine on N388 interacts with O2' hydroxyl group (2‐3 Å distance) and O3' hydroxyl group (3‐4 Å distance) for ~20 ns. Although Y382 does not directly interact with NADH, it appears that the Y382 hydroxyl stabilizes the position of H391 by interacting with the H391 carbonyl for the entire 23 ns trajectory. See Figure 6 below for the residue contacts with NADH at the NADH/ADP site.

Figure 6.

NADH binding at the NADH/ADP site

The residues that contact the adenosine moiety of NADH include K387, H209, D119, R86, H85, and V120. Residue D119 carboxyl side chain maintains a close interaction (1‐2 Å) with O2' hydroxyl group throughout the trajectory while H85 and V120 maintain contact with the adenine group similar to ADP binding. K387 moves 2 Å to maintain a close contact with the nicotinamide phosphates for the entire trajectory. In contrast, H209 remains distant for ~20 ns then moves to interact with NADH phosphates for the last 3 ns. R86 begins by making contact and interacting with the NADH phosphates for ~9 ns then moves ~7‐9 Å away. Residues R459, R491, and R396 remain distant from NADH, which contrasts with ADP binding. It appears that the increase in charge from −2 on NADH to −3 on ADP allows for increased duration and number of interactions of the phosphate moiety with GDH, specifically with residues R459, R491, H209, and R396.

The description above describes the protein‐NADH interactions shown in all three binding sites with the following exceptions. The R86 on chain B loses its interaction with NADH halfway through the simulation. H209 on chain A remains distant from NADH. H195 on chain A interacted with NADH while H195 on other chains did not. The same applies to I192 in that the interaction is only observed for I192 on chain A. The delta amine on H391 on chain B interacts with the nicotinamide ribose, specifically O3'.

Contrary to previous interpretations of the X‐ray crystal structures,17 S393 does not appear to favor ADP interactions in our simulations. Instead, it appears that the S393 hydroxyl group interacts with the E445 carboxyl group. This phenomenon is observed for both ligands in both of our NADH‐GDH and ADP‐GDH simulations. E445 is located on the pivot helix, connecting the pivot helix to the pigtail region and S393 is located on a disordered region directly connected to the end of the antenna. The interaction between E445 and S393 limits the flexibility of the disordered region where S393 is located and is present in both the closed and open conformation. Thus, this interaction may be important for both the closed and open conformation possibly explaining why an S393I mutation results in loss of both GTP and ADP regulation. The position of E445 in close proximity to S393 is preserved by the ionic interaction between the carboxyl group on E445 and the amine group on K387.

4. CONCLUSION

Our computational analyses show intriguing results for ADP binding at the NADH/ADP binding site in which ADP can exhibit multiple conformations that allow binding but not always activation. Indeed, this is entirely consistent with previous results showing that an R459A mutation affected ADP activation but not binding.16, 43 This model helps to address the complex nature of ADP regulation observed in past kinetic experiments. These results are encouraging and the detailed analysis of the conformational changes leads to insight into the cooperative motions of GDH upon ligand binding. In addition, our models show comparable results to the crystallographic results with regard to the open and closed conformations in the presence of ADP and NADH, respectively, as well as reproduction of the binding free energy difference. The ΔΔG produced by TI of −0.3 kcal/mol is close to the experimental ΔΔG of −0.275 to −1.7 kcal/mol, while ΔG ₁ and ΔG ₃ values were comparable to previous adenine binding and solvation free energies. Future directions will include computing ΔΔG utilizing a potential of mean force to guide the conformation from the inhibited state (NADH bound) to the activated state (ADP bound). Also, if a better resolution structure were available a backward transformation from the ADP bound to NADH bound state could improve the sampling and precision of the calculation.

CONFLICT OF INTEREST

The authors declare no conflicts of interest with this work.

Supporting information

Appendix S1: Supplementary Material

Nassar OM, Wong K‐Y, Lynch GC, Smith TJ, Pettitt BM. Allosteric discrimination at the NADH/ADP regulatory site of glutamate dehydrogenase. Protein Science. 2019;28:2080–2088. 10.1002/pro.3748