Exploring the overlapping binding sites of ifenprodil and EVT‐101 in GluN2B‐containing NMDA receptors using novel chicken embryo forebrain cultures and molecular modeling

Abstract N‐methyl‐d‐aspartate receptors (NMDAR) are widely expressed in the brain. GluN2B subunit‐containing NMDARs has recently attracted significant attention as potential pharmacological targets, with emphasis on the functional properties of allosteric antagonists. We used primary cultures from chicken embryo forebrain (E10), expressing native GluN2B‐containing NMDA receptors as a novel model system. Comparing the inhibition of calcium influx by well‐known GluN2B subunit‐specific allosteric antagonists, the following rank order of potency was found: EVT‐101 (EC 50 22 ± 8 nmol/L) > Ro 25‐6981 (EC 50 60 ± 30 nmol/L) > ifenprodil (EC 50 100 ± 40 nmol/L) > eliprodil (EC 50 1300 ± 700 nmol/L), similar to previous observations in rat cortical cultures and cell lines overexpressing chimeric receptors. The less explored Ro 04‐5595 had an EC 50 of 186 ± 32 nmol/L. Venturing to explain the differences in potency, binding properties were further studied by in silico docking and molecular dynamics simulations using x‐ray crystal structures of GluN1/GluN2B amino terminal domain. We found that Ro 04‐5595 was predicted to bind the recently discovered EVT‐101 binding site, not the ifenprodil‐binding site. The EVT‐101 binding pocket appears to accommodate more structurally different ligands than the ifenprodil‐binding site, and contains residues essential in ligand interactions necessary for calcium influx inhibition. For the ifenprodil site, the less effective antagonist (eliprodil) fails to interact with key residues, while in the EVT‐101 pocket, difference in potency might be explained by differences in ligand‐receptor interaction patterns.


| INTRODUC TI ON
The N-methyld-aspartate (NMDA) receptors are found in all brain regions and are involved in synaptic plasticity, learning, and memory. 1 They belong to a subfamily of excitatory glutamate receptors that are ligand-and voltage-gated channels with permeability predominantly for Ca 2+ , but also for Na + and K + . 2 The NMDA receptors consist of heteromeric tetramers built up by the subunits GluN1, GluN2, and GluN3. Two GluN1 and two GluN2 or GluN3 subunits must be present to enable ligand binding. There are four variants of GluN2: GluN2A, GluN2B, GluN2C, and GluN2D, and the structure can be homo-or heterotetrameric with respect to the different GluN2 subunits. 3 The receptor distribution and composition are dynamic and change during development and in response to sensory input. 4 Neurodegenerative diseases such as Parkinson's disease, Alzheimer's disease, and epilepsy are associated with pathological changes in the assembly and location of NMDA receptors. [5][6][7] Changes in these receptors are also observed in psychiatric disorders such as schizophrenia and depression. 8,9 Memantine, a low affinity antagonist that alleviates symptoms of Alzheimer's disease, is one of the very few commercially available drugs targeting NMDA receptors. 10,11 The severe psychotomimetic side effects observed when administrating unspecific total channel blockers such as MK-801 and ketamine to humans and animal models, 12 indicate a need for new partial and/ or subtype-specific antagonists. To develop new drug candidates, it is essential to understand the receptor binding mechanisms and explore the conformational space of the receptor binding sites.
The amino terminal domain (ATD) of the subunits contains binding sites for allosteric compounds, such as the synthetic GluN2Bspecific antagonists eliprodil, Ro 25-6981, and ifenprodil. [13][14][15][16] Recently, an x-ray crystal structure of the GluN1 and GluN2B ATD dimer in complex with ifenprodil was resolved. 17 However, ifenprodil displays unspecific binding to sigma opioid receptors and monoamine receptors, [18][19][20] causes behavioral toxicity 21 and it is readily inactivated by hepatic metabolism. 22,23 Based on the structural features of ifenprodil, several new GluN2B-specific antagonists such as Ro 04-5595, which has been shown to displace Ro 25-6981, have been investigated. 18 X-ray crystal structure complexes showed that EVT-101 (another GluN2B antagonist) binds at the same GluN1/ GluN2B interface as ifenprodil, but occupies an overlapping and less explored binding site. 17 The GluN2B-specific allosteric antagonist HON0001 (structurally similar to Ro 04-5595) has also been shown to have a potent dose-dependent oral analgesic activity in rats, with less side effects and higher receptor specificity than ifenprodil 24 and it has been predicted to interact with the EVT-101 binding site. 17 In this paper, ligands predicted to bind the ifenprodil-binding site are denoted A-ligands, while those predicted to bind the EVT site are named B-ligands.
The NMDA receptor is evolutionarily conserved across species. 25 Many studies have therefore used overexpressed chimeric NMDA receptors with constituents from Rattus norvegicus (R. norvegicus), Xenopus laevis (X. laevis), or Homo sapiens (H. sapiens). [26][27][28] In vitro cultured neurons from the developing chicken brain was recently suggested as a suitable model for nonclinical drug testing. 29 Chicken forebrain culture expresses native, functional NMDA receptors with a high proportion of GluN2B subunits, features that make them suited for the present study.
In this paper, we utilized chicken embryo primary forebrain culture and a functional calcium influx assay to investigate the potency of GluN2B-specific allosteric antagonists. Their binding mode was investigated by docking studies and molecular dynamics simulations using experimental structures of GluN1/GluN2B ATD, and the predicted binding data were compared to functional results. We also investigated amino acids critical for ligand binding by in silico mutation studies and found that the residues that differentiate the EVTbinding site from the ifenprodil site are predicted to be located in the GluN2B subunit. EVT-101 was significantly more potent than Ro 04-5595 in terms of calcium influx inhibition, which may be explained by the interaction of EVT-101 with GluN2BMet134 and GluN2BAla135.
When comparing ligands that are predicted to bind to the ifenprodil site, it appeared that the less potent allosteric antagonist eliprodil failed to interact with residues GluN1Ser132, GluN2BTyr175 and GluN2BMet207, all of which display interaction with the stronger inhibitors Ro 25-6981 and ifenprodil. Among the compounds tested, the ligands proven (ifenprodil and Ro 25-6981) and predicted (eliprodil) to be located in the ifenprodil-binding pocket are structurally similar, while the EVT-101 binding site appears to accommodate more structurally diverse ligands and binding poses, which is supported by earlier work. 17
Animals were handled in accordance with the Norwegian Animal Welfare Act and the EU Directive 2010/63/EU. However, chicken embryos are not regarded as research animals before E14 (2010/63/EU; EU, 2010). They have a short incubation time, do not require animal housing and elicit fewer allergies than murine animal models. 30 It is also easier to predict the number of embryos obtained compared to rat or mice, and the hen is exempted from experiments. Thus, their use is in accordance with the 3Rs principles of animal research.

| Chicken embryo forebrain cultures
The eggs were submerged in crushed ice for 7 minutes to anesthetize the embryos before decapitation. The forebrain was surgically removed, and the meninges were discarded. The tissue was homogenized by chopping with a scalpel before trypsination in buffered solutions as previously described. 30,31 Cells were suspended in DMEM supplemented with 1% N-2, 100 U/mL penicillin and 0.1 mg/ mL streptomycin (Pen-Strep), 10% fetal bovine serum (FBS), and 0.75% GlutaMAX™. Cells were seeded (1.7 × 10 6 cells/mL) on 35 mm Petri dishes or in 96-well plates (Corning ® CellBIND ® 96 well plates; Merck) precoated with polyl-lysine, and incubated at 37°C, with 5% CO 2 . These cultures contain an abundance of functional GluN2B receptors on DIV1 (A. Ring, pers. commun.).
Isolated tissues from chicken forebrain (E7-18) and mouse cerebellum (postnatal day 21) were frozen in N 2 (−196°C) before longterm storage at −20°C. To prepare for western blotting analysis, tissue was homogenized as previously described. 32 In short, samples were kept on ice, added tris-EDTA (TE) buffer containing the same protease inhibitors as described above, and homogenized using a motorized pellet pestle. TE with SDS (final concentration 2%) was added to the sample before further homogenization by syringe (25 G) and heat inactivation of proteases (95°C, 5 min).
Protein concentration was determined with Pierce™ BCA Protein Assay Kit (ThermoFisher™, USA). Each sample (25 ug) was mixed with Laemmli buffer with 5% mercaptoethanol and then applied to a precast 10-well polyacrylamide Mini-Protean Tris-Glycineextended (TGX™) gel (BioRad, Germany). After electrophoresis, the proteins were transferred to a nitrocellulose membrane (TransBlot ® Turbo™; BioRad, Germany) which was blocked with 5% dry skimmed milk in 1% Tween-Tris-buffered saline solution (TBS-T) for 1 hour at room temperature (RT). The primary GluN2B antibody was diluted in 5% dry skimmed milk in TBS-T to a concentration of 1:1000 and added to the membranes which were then incubated for 24 hours at 4°C. The membranes were rinsed three times with TBS-T and incubated for 1 hour at RT with anti-rabbit secondary antibody (1:10 000 in TBS-T with 5% dry skimmed milk) before a further rinse cycle with TBS-T. Bands were detected using chemiluminescence with HRP substrates in the bio-imaging system Chemi Genius 2 with GeneSnap software (both by Syngene, UK). The amount of internal standard was assessed by immunostaining with β-actin antibody and anti-mouse secondary antibody. The data were analyzed using ImageJ software, 33 and the amount of GluN2B was normalized against the amount of β-actin protein.

| Immunocytochemistry
The cell culture was grown in polyl-lysine-coated petri dishes with glass bottom (MatTEK Corporation, USA). The cell medium was aspirated. Dishes were added 1 mL of PBS with 3.7% formaldehyde and left at RT for 10 minutes before washing twice with PBS (4°C).
The cell membranes were permeabilized with 0.1% Triton-X in PBS before blocking with 5% dry skimmed milk in 1%TBS/Tween for 30 minutes at RT. After washing twice with cold PBS, the neuronal marker antibody NeuN was diluted in PBS (1:100) and 100 μL was added to the dishes and incubated at 4°C for 12 hours. The dishes were washed three times with cold PBS before 400 μL of secondary FITC antibody diluted in 5% dry skimmed milk in 1%TBS/Tween was added at a concentration of 1:250 and left to incubate in the dark for 1 hour at RT. The cells were mounted with the nuclear marker DAPI and visualized with fluorescence microscopy (Eclipse TE300; Nikon, Japan).

| Calcium influx measurement
The procedure was similar to that previously described by Ring et al 34 Cells were plated in polyl-lysine coated 96-well black plates with clear glass bottom (Corning ® CellBIND ® ) and each well was incubated with 4 μmol/L fluorescent calcium (Ca 2+ ) indicator Fura-2 at 37°C, 5% CO 2 for 45 minutes. 34,35 The medium was then replaced with a standard buffer (140 mmol/L NaCl, 3.5 mmol/L KCl, 15 mmol/L Tris (pH 7.4), 1.2 mmol/L Na 2 HPO 4 -NaH 2 PO 4 (pH 7.4), 5 mmol/L glucose, and 2 mmol/L CaCl 2 in distilled water) with 1 mmol/L MgCl 2 (wash buffer) and further incubated for 15 minutes in the dark for de-esterification of Fura-2. Fura-2 fluorescence was measured using CLARIOstar ® plate reader (BMG Labtech, Germany). Intracellular  Prime. 40 The large missing loop (186-209 located in GluN1) was not modeled as it was far from the ligand binding pocket and was therefore not considered to have any impact on the binding pocket.
Crystal structure water molecules were retained, and the ionization state of the heteroatoms was handled with a pH of 7.4 ± 0.2.
The protonation state of the different residues and the optimization of the hydrogen bonds network were performed with PROPKA at pH = 7.4 ± 0.2 with sampling of the crystal water molecules before a final restrained minimization of heavy atoms.
The chicken GluN1 sequence was retrieved from UniProt (ID: Q4KXT1) 29 while the chicken GluN2B sequence was retrieved from the predicted target sequence with BLAST (Basic Local Alignment Search Tool, XP_015144845.2, NIH, USA). 30 The retrieved sequences were aligned with the sequences from chain A and B of the x-ray crystal structures, using the Multiple Sequence Viewer (MSV) tool. The chicken GluN1 ATD (1-400 residues) sequence is 91% similar to the GluN1 ATD sequence from X. laevis, while the chicken GluN2B ATD has a 95% sequence identity with the human and rat GluN2B ATD. A sequence alignment between human, rat, and chicken Glun2B subunits can be found in supplementary data ( Figure   S1), made with the Clustal Omega multiple sequence alignment program available at Uniprot's webpage. 41,42 The total rat and human

| Ligand preparation and docking studies
The docking procedures were performed in Schrödinger's Glide software. 43 Receptor grid maps were generated for both crystal structures in complex with ifenprodil and EVT-101 (PDB id: 5EWJ and 5EWM, respectively) using default settings 44 and co-crystallized ligands as the centroid of the map. Two overlapping allosteric binding sites have been described for the GluN1/GluN2B subunits: the ifenprodil and the EVT-101 binding pockets. 17,38 In order to study the ligand-protein interactions of the ligands used in vitro, the com-

| Molecular dynamics simulations
Molecular dynamics simulations were performed with the Desmond program. 45 The selected complexes were set up in an orthorhombic simulation box with periodic boundary condition, the OPLS3 force field TIP3 water model was employed for the solvation of the system before it was neutralized and 0.15 mol/L NaCl was added. The generated systems were relaxed using the Desmond default protocol and run for 100 ns on a GPU. The NPT ensemble was selected with a P = 1.01325 bar and T = 300 K using the Martynas-Tobias-Klein barostat method (relaxation time of 2 ps and isotropic coupling style) and the Nose-Hoover Chain thermostat method (relaxation time of 1 ps and one group for temperature), respectively. The RESPA integrator was selected and the bonded and close nonbonded interactions were handled with a timestep of 2 fs while for far nonbonded interactions the timestep was set to 6 fs. A cut-off of 9 Å was used for the short-range columbic interactions. The trajectories and energies were recorded every 10 ps giving a total of 10 000 frames. Root Mean Square Deviations of protein and ligand can be observed in Figure S3. The last 10 ns (90-100 ns corresponding to the last 1000 frames) were considered for analysis of the protein-ligand interactions, the generation of average ligand-receptor complexes, and alanine scanning calculation utilizing the residue scanning tool from BioLuminate. 46 All residues within 5 Å of the ligand in the averaged complexes were mutated into alanine and their contribution to the free energy of binding (ΔG) was analyzed by calculating the difference in ΔG before and after mutation for each residue. The averaged conformation of eliprodil and EVT-101's receptor-ligand complex required additional minimization before alanine mutation scanning could be performed. This was done by the minimization panel featured in the MacroModel software, 47 with OPLS3 force field, water as solvent and extended cut-off.

| Analysis and statistics
Outlier values were tested for by the built-in feature in GraphPad

| Cultures from chicken forebrain express GluN2B
Since chicken primary forebrain neuron cultures have not been

| Functional properties of chicken NMDA receptors resemble their human and rat counterparts
Functional properties of the NMDA receptor in chicken forebrain culture were tested with the calcium influx assay as described previously. 34 It was shown that the receptor was activated by standard  (Table 1 and Figure 2A) 17,48 and that the calcium influx was reduced by 90% by 10 μmol/L of the unspecific NMDA receptor inhibitor MK-801. 49,50 The less-explored antagonist Ro 04-5595 showed an IC 50 value of <200 nmol/L. The differences in IC 50 values between eliprodil and the other A-ligands were statistically significant: Eliprodil vs ifenprodil and eliprodil vs Ro 25-6981 (*P ≤ 0.05, and ***P ≤ 0.001, respectively, shown in Figure 2B). B-ligands EVT-101 and Ro 04-5595 gave significantly different IC 50 values when tested experimentally in the chicken forebrain primary culture calcium influx assay (****P < 0.0001, Figure 2C).

| Computer modeling reveals conserved tertiary structure of chicken GluN1/GluN2B ATD
The high percentage of amino acid sequence similarity between chicken GluN1/GluN2B ATD and X. laevis/H. sapiens GluN1/GluN2B F I G U R E 1 Embryonic chicken forebrain neurons can be grown in vitro and GluN2B is expressed in chicken forebrain tissue in the fetal period. Chicken forebrain was harvested at E10 and the cell culture was incubated overnight, before inspected at DIV1, using: A, Light microscopy and B, Immunostaining with neuronal marker NeuN and nuclear stain DAPI. B1: Brightfield image. B2: Staining with DNA marker DAPI, visualized by UV light. B3: Immunostaining with NeuN, visualized by fluorescence microscopy. B4: Composite image of DAPI stain and NeuN immunostaining. C: Western blot stained with anti-GluN2B antibody, concentration 1:1000 (ab65783, Abcam, UK) and antiβ-actin antibody. Lanes: 1-3: HEK cells transfected with GluN2A, GluN2B or CMV plasmid, respectively. Lane 4: Homogenized chicken forebrain tissue harvested at E10. Lane 5: Chicken embryo forebrain cell culture, harvested at E10 and analyzed at DIV1. Lane 6: Homogenized mouse cerebellum harvested at P21. D: Time series of GluN2B protein expression in homogenized chicken embryo forebrains, from E7 to E18. GluN2B protein expression relative to internal control protein β-actin expression. The values are normalized to expression level at E7 (n = 3), and statistical significance was investigated using the Kruskal-Wallis test. Variation is given as standard deviation and a representative example of western blot of GluN2B and β-actin is shown below the graph  Figure   S2). These similarities enabled the use of experimental x-ray structures instead of the chicken homology model for studying the dynamics of ligand interactions. X-ray structures are regarded as both structurally and energetically more stable than homology models and more reliable predictions are expected.

| Molecular dynamics simulations predict interactions between ligands and binding site residues
Our docking studies and molecular dynamic simulations supported that Ro 25-6981 and ifenprodil shared the ifenprodil-binding site and showed that eliprodil interacted with the ATD domain via the ifenprodil-binding site. This is supported by earlier findings. 17,38,51 Eliprodil gave a Molecular Mechanics/Generalized Born Surface Area

| The ifenprodil-binding site
The predicted common residues for the A-ligands Ro 25-6981, ifen- to the binding site was predicted to be severely decreased by mutating the residue to alanine, while the affinity of eliprodil was less affected, suggested to be caused by its bond type (ππ stack vs water bridge). According to the molecular dynamics simulations Ro 25-6981 interacted with two amino acid residues on its own: GluN1Leu131

TA B L E 2 Overview of predicted binding residues
A schematic overview of the shared binding residues, binding residues in each binding pocket and specified to ligands within each binding pocket. Prefix N1 denotes that the residue is located in the GluN1 subunit, while 2B indicates the GluN2B subunit. Residues shared between ligands are shown in black: 2BPro78 is shared between Ro 25-6981 and EVT-101, 2BPhe176 is common for ifenprodil, eliprodil, and EVT-101, while 2BAla107 is predicted to bind both Ro 25-6981 and eliprodil.
F I G U R E 3 continued F I G U R E 4 Observed interactions between the ligands and the receptor during molecular dynamics simulations and comparison of the free energy of binding ΔG (kcal/mole) before and after alanine mutation scanning. The percentage of frames (complexes) showing interactions between the ligands and their binding amino acids during the last 10 ns of the molecular dynamics simulations (1000 frames). The interactions comprise H-bonds, π-cation interaction, π-π stacking, other hydrophobic interactions, ionic bonds, and water bridges. One residue can have several interactions, which is why some values exceeded 100%. Prefix N1 denotes that the residue is located in the GluN1 subunit, while 2B indicates the GluN2B subunit. A, The interactions between Ro 25-6981, ifenprodil, eliprodil, EVT-101, Ro 04-5595, and the residues shared by all ligands: N1Tyr109, N1Phe113, N2BIle111, and N2BPhe114. N1Ile133 was shared by all except Ro 25-6981 and 2BPhe176 was shared by all except Ro 04-5595. B, Differences in the free energy of binding (ΔG) (kcal/mole) for Ro 25-6981, ifenprodil, eliprodil, EVT-101 and Ro 04-5595 when mutating residues shown in 4A into alanine. C, The interactions between Ro 25-6981, ifenprodil, eliprodil, and their shared residues located in the ifenprodil-binding site: N1Arg115, N1Ser132, N1Leu135, 2BGLN110 and 2BGlu236. D, Δ affinity (kcal/mole) for Ro 25-6981, ifenprodil and eliprodil when mutating residues shown in 4C into alanine. E, The interactions between EVT-101 and Ro 04-5595 and their shared residues located in the EVT-101-binding site: 2BAsp113, 2BAsp136, and 2BPro177.F, Δ affinity (kcal/mole) for EVT-101 and Ro 04-5595 when mutating residues shown in 4E into alanine and GluN2BLeu205. Both were weak hydrogen bond interactions.
Mutation of both residues into alanine was not predicted to have an extensive effect on the affinity of Ro 25-6981 to the binding site ( Figure 4D).

| The EVT-101 binding site
The amino acid residues Asp113, Asp136, and Pro177 (GluN2B) were predicted to interact with both B-ligands Ro 04-5595 and EVT-101. GluN2BAsp113 presumably displays a stable interaction to Ro 04-5595 and a weaker connection to EVT-101. in Figure 4E and the corresponding alanine mutation scanning results are shown in Figure 4F.
Some of the residues that did not interact with the ligands still affected their affinities when running an alanine mutation scan, probably due to local conformational changes within the binding cavities or indirect effects. The residues are summarized in Figure   S4 in supplementary data, and we found that Ro 25-6981, ifenprodil, and eliprodil shared some of them. The only residue shared by all ligands is GluN1Thr110, for which an alanine mutation is predicted to be especially critical for the affinity of EVT-101, but enhances the affinity of Ro 04-5595. Overviews of all predicted interactions and the ΔG differences (kcal/mole) for all residues and ligands predicted by alanine mutation scanning are included in the supplementary data ( Figure S5).

| D ISCUSS I ON
In the present study, we have used primary cultures from chicken embryo forebrain as a model to study potencies of different GluN2B polyamine site antagonists to reduce calcium influx. To support the experimental data, computational methods were applied to predict binding to amino acids in the two overlapping ifenprodil and EVT-101 sites.
The chicken embryo forebrain cell culture expresses GluN2B, established by using a specific antibody raised against a rat GluN2B epitope. Compared to human and rat, the expression pattern of GluN2B in developing chicken forebrain follows a similar trajectory. However, the decline in GluN2B protein expression appears to take place prenatally in chicken, as opposed to postnatally in human and rat. This may reflect a higher degree of relative maturity of the cortex in newly hatched chickens compared to new-born rats or humans. This is an advantage when considering chicken embryos as an animal model for NMDA receptor development, as it enables easy access to study developmental processes occurring postnatally in other research animals, while the chicken is still contained within the egg.
Since expression studies confirmed the presence of GluN2Bcontaining receptors, we wanted to confirm that these were functional in vitro. This was done by studying NMDA-and glycine-induced calcium influx. However, it is important to note that the chicken fore- which predicted that HON0001 would bind to the EVT-101 site as well. 17 The analgesic effect of orally administrated HON0001 encourages further investigations of Ro 04-5595 as a potential re- GluN2BPro177 to cysteine did indeed increase the IC 50 value of EVT-101. 17 Stroebel et al 17 have analyzed three in vitro alanine mutations. They observed that in vitro mutation of GluN1Ile133 led to a lower IC 50 value for ifenprodil, and a higher value for EVT-101.
This corresponded with our in silico observation of higher loss of affinity for EVT-101 than for ifenprodil. However, mutation of GluN1Leu135 to alanine in vitro, which gave small changes in IC 50 values, did not correspond with our in silico alanine mutation scan results, which predicted a reduction in both ifenprodil and EVT-101 affinities. Also, in vitro alanine mutation of GluN2BPhe176 increased the IC 50 values of both ligands drastically. Interestingly, only the least efficient antagonist, eliprodil, was predicted to have a stable interaction with this residue, and ifenprodil and EVT-101 affinities were predicted to be less affected by this mutation.
These discrepancies underscore the importance of comparing in silico data with experimental data.
In conclusion, we have established the chicken primary fore-

D I SCLOS U R E S
None declared.