Ligplot

The docking simulation was performed using AutoDock Vina 1.1.2. The active site of the protein structure was kept rigid, and inflexible docking was performed. The parameters were set to exhaustiveness 8, num_modes 10, and energy_range 3 kcal/mol. The Vina code predicts the adopted form with a binding affinity (kcal/mol). The best docking pose was analyzed according to the binding affinity obtained from 50 independent runs to create the final docked pose. The binding energy of a cluster is the average binding energy of all the conformations in it. The clusters with the lowest binding energy and the highest number of conformational structures were chosen as representative binding modes for the ligand. The protein-ligand complexes were visualized using PyMol Version 2.4.1 (Schrödinger, New York, NY, USA, https://pymol.org/, accessed on 20 November 2020) and LigPlot+ version 2.2.4.

For normal molecular docking, we used Dock version 6.5 (28 (link)). For TNF, the homodimer (Chains A and B) was used for molecular docking, and the native inhibitor bound within the binding site formed by both monomers (Chains A and B) was used as a probe for the TNF homodimer binding site. The protein and ligand preparation required for molecular docking and visualization at different stages of docking was performed using Chimera version 1.6.2 (29 (link)). The illustrations for binding poses were generated using Pymol version 2.4.0 (Schrödinger, LLC) and the protein-ligand interaction plots were prepared using Ligplot+ version 2.1 (30 (link)). The binding energy and dissociation constant scores were predicted using XScore version 1.2.11 (31 (link)).

Structures of compounds were downloaded from the chemical library collection of Vitas-M laboratory (Causeway Bay, Hong Kong) (https://vitasmlab.biz/). Virtual screening was performed using Lead Finder software package (MolTech LLC, Moscow, Russia^{42 (link)}, the dock_filter protocol^{43 (link)} was applied to filter the subset of ligands interacting simultaneously with both protein “arms” and the beta-saddle so as to screen for compounds with the best abilities to bind to HUSpm’s DNA-binding cleft. Ligands were processed with OpenBabel 2.3.2 software^{44 (link)} to remove counterions, to generate physiological charge states and tautomers wherever applicable. Energy-minimized 3D structures for all compounds were generated and subjected to the molecular docking.
To prepare the target protein (PDB ID 5L8Z^{9 (link)}) for docking, unresolved residues were restored using Modeller softwar^{45 (link)}. Hydrogen atoms were added to the HUSpm spatial structure using Build Model unit of the Lead Finder software package. Visualization of protein-inhibitor complexes and intermolecular interaction was performed using the VMD^{46 (link)}, LigPlot^{47 (link)} and the PyMOL Molecular Graphics System, Version 1.9.0.0 (Schrödinger, LLC, New York, NY, USA).

X-ray diffraction measurements were performed at beamline P13 at PETRA III, DESY, Hamburg (Germany), beamline X06DA at the Swiss Light Source, Paul Scherrer Institute, Villigen (Switzerland), or beamline ID30A-3 at the European Synchrotron Radiation Facility (ESRF), Grenoble (France). The human full-length GFAT-1 structure was determined by molecular replacement with phenix.phaser^{50 (link),51 (link)} using the model of the human GFAT-1 isomerase domain (PDB 2ZJ3) as search model. After a first round of autobuilding using the ARP/wARP Web Service^{52 (link)} GFAT-1 was further manually built using COOT^{53 (link)} and iterative refinement rounds were performed using phenix.refine^{51 (link)}. One of the glutaminase domains (chain B) was not well defined in the structure. After placing initial strands, the domain was completed by superposition with the glutaminase domain of chain A. Structures of GFAT-1 variants and UDP-GlcNAc/UDP-GalNAc soaked crystals were solved by molecular replacement using the full-length GFAT-1 structure as a search model. Geometry restraints for ligands were generated with phenix.elbow software^{51 (link)} or the Grade Web Server. Structures were visualized using PyMOL (Schrödinger) and 2D ligand–protein interaction diagrams were generated using LigPlot+^{54 (link)}.

Enzyme-ligand docking simulations were conducted using the AutoDock Tools 1.5.6 platform (ADT) [22 (link)]. A grid box of size 42 × 30 × 70 points was applied to all XOS. The box encompassed the entire active-site cleft with +1 and −1 subsites at the center. The receptor molecule was treated as rigid, and the ligands as flexible. The maximum number of energy evaluations was set to 25,000,000 (10 times the default). The other parameters were set as default. The resulting docking structures were ranked according to their binding energy scored by the function of ADT (based on the United Atom version of the AMBER force field) [30 (link)]. The lowest energy conformation was used for positional binding analysis. The criterion for hydrogen bond judgment was that the maximum distance between donor and acceptor atoms should be less than 3.4 Å. Each docking simulation was performed three times and the same results were obtained. The complex structures were visualized and analyzed using PyMol (Schrödinger) and LigPlot [31 (link)].