Inhibition of GluN2B-containing N-methyl-d-aspartate receptors by radiprodil

Abstract

N-methyl-D-aspartate receptors mediate a slow, Ca2+ permeable component of excitatory synaptic transmission in the brain and participate in neuronal development and synaptic plasticity. Most NMDA receptors are tetrameric assemblies of two GluN1 and two GluN2 subunits encoded by five genes (GRIN1, GRIN2A-D), which produce GluN1 and GluN2A-D subunits. NMDA receptors that contain the GluN2B subunit have unique pharmacological properties, being inhibited by multiple structurally distinct series of biaryl compounds with high potency and selectivity. These agents are of considerable therapeutic interest, given the numerous roles that GluN2B-containing NMDA receptors play in normal brain function and pathological situations. Among GluN2B-selective negative allosteric modulators, radiprodil inhibits NMDA receptors that contain GluN2B with high potency and selectivity and appears to be safe in human. Here, we evaluate the structural determinants of radiprodil binding to the heterodimeric GluN1-GluN2B amino terminal domain by X-ray crystallography and explore the molecular mechanism of inhibition. A large number of de novo variants have been identified in the GRIN gene family in patients with various neurological and neuropsychiatric conditions, including autism, intellectual disability, epilepsy, language disorders, and movement disorders. We show that radiprodil is an effective antagonist at over 80% of human disease-associated GRIN1 and GRIN2B missense variants tested in vitro (22/27, equally or more effective as WT receptors), including variants in the pore-forming region, linker regions, and elsewhere that uniformly increase NMDA receptor-mediated charge transfer. We show radiprodil blocks synaptic GluN2B receptors in brain slices acutely isolated from a knock-in mouse line harboring the gain-of-function variant GluN2B-Ser810Arg associated with early-onset epileptic encephalopathy and intractable seizures in patients. In addition, radiprodil delays the onset of seizures (458 ± 90 sec vs 207 ± 23 sec of vehicle group) in response to the in vivo administration of the chemoconvulsant pentylenetetrazole. These data support the potential utility of GluN2B-selective antagonists like radiprodil for clinical treatments of neurological conditions, where clinical etiologies may involve increased current mediated by GluN2B-containing NMDA receptors.

Keywords: NR2B; PET scan; precision medicine; seizure susceptibility; structural biology; translational study.

Figure 1. Inhibition of diheteromeric and triheteromeric NMDA receptors by the GluN2B-selective compound radiprodil.

(A,B) Di-heteromeric rat GluN2B-containing NMDARs expressed with different N-terminal GluN1 splice variants (1a or 1b as indicated). (A) Representative normalized current responses recorded from Xenopus oocytes using two-electrode voltage clamp show the concentration-response relationship for radiprodil inhibition of rat GluN2B NMDARs expressed with either GluN1–1a (lacking exon5) or GluN1–1b (including exon5). Current responses were activated by saturated concentrations of glutamate (100 μM) and glycine (30 μM), applied as indicated by the white box; increasing concentrations of radiprodil are shown by grayscale boxes (0.01, 0.03, 0.1, 0.3, 1, 3 μM as indicated). (B) Representative current responses for radiprodil inhibition of rat triheteromeric NMDARs activated by saturated concentrations of glutamate (100 μM) and glycine (30 μM) are shown (white box) with increasing concentrations of radiprodil indicated by the greyscale boxes (0.003, 0.01, 0.03, 0.1, 0.3, 1 μM). (C) The concentration-response relationships for radiprodil inhibition of rat diheteromeric NMDARs containing two GluN1 splice variants are shown. (D) The concentration-response relationships for radiprodil inhibition of triheteromeric NMDAR-mediated current responses for GluN1/GluN2A_C1/GluN2B_C2, GluN1/GluN2B_C1/GluN2C_C2 and GluN1/GluN2B_C1/GluN2D_C2 are shown. All triheteromeric subunit combinations were expressed with the GluN1–1a splice variant. The fitted diheteromeric GluN1–1a/GluN2B trace from panel C is superimposed as a dotted line for comparison. (E) Representative current responses recorded from Xenopus oocytes show the concentration-response relationship for glutamate activation of rat GluN2B NMDARs expressed with GluN1–1a in the absence or presence of 1 μM radiprodil. (F) Concentration-response relationship for glutamate in the absence and presence of radiprodil (present in wash and agonist solutions). (G,H) The time course for the onset of radiprodil inhibition. (G) Upper panel: Concentration-dependent binding of radiprodil to rat GluN1/GluN2B NMDARs was assessed in HEK293 cells recorded under voltage clamp. The steady-state level of inhibition for all radiprodil concentrations approached that observed for saturating concentrations of antagonists (0.3 μM, 6.0% of control; 1 μM, 7.9% of control; 3 μM, 6.7% of control). Lower panel: Radiprodil washout proceeds very slowly, with a 3 min washout producing negligible recovery from inhibition (4.7% recovery from inhibition, which suggests a 62 min recovery tau value assuming an exponential time course). (H) A single exponential function could describe the relaxation during radiprodil application. The association and dissociation rates were determined from the linear relationship between radiprodil concentration and the reciprocal of the fitted deactivation tau, consistent with the law of mass action. The kinetically determined K_D was approximated to be 0.175 μM by the ratio of the fitted linear intercept (koff) and the slope (kon). Data was expressed as mean ± SEM,

Figure 2. Structure of GluN1–1b/GluN2B NTD in complex with radiprodil.

(A) The GluN1b/GluN2B NTD bound to radiprodil (green spheres) is superposed on the structure of the intact GluN1/GluN2B NMDAR in complex with glycine and glutamate (PDB code: 7SAA in gray). GluN1–1b-R1, GluN1–1b-R2, GluN2B-R1, and GluN2B-R2 are colored orange, light orange, dark cyan, and cyan, respectively. (B) GluN1–1b-GluN2B NTD viewed from the eye in (A). (C) The Fo-Fc omit map of the radiprodil density contoured at 4.5σ and the modeled radiprodil (green sticks). (D) Binding of radiprodil at the dimer interface between GluN1–1b (orange/light orange) and GluN2B (dark cyan and cyan). (E-F) Comparison of binding poses between radiprodil and ifenprodil (E) or EU93–31 (F). The GluN1–1b subunits of the ifenprodil-bound or EU93–31-bound structures were superimposed onto the radiprodil-bound structure. Ifenprodil and EU93–31 are shown as gray and magenta sticks, respectively. An arrow in panel F indicates the large difference in the binding mode between radiprodil and EU 93–31.

Figure 3. The missense variant GluN2B-S810R produces a likely gain-of-function and unchanged expression of synaptic proteins in GluN2B-S810R heterozygous mouse brain.

(A) Linear schematic of a cDNA encoding different NMDAR domains, with the pre-M4 linker expanded below. Ser810 is conserved across species and within the GluN2 gene family. (B) The substitution of Arg for Ser at position 810 produces steric and electronic changes to the protein at a site that is critically involved in gating . (C,D) Concentration-response curves show an increase in potency of GluN2B-S810R for glutamate (C, variant EC₅₀ 0.023 μM, WT EC₅₀ 1.4 μM, n = 13–14) and glycine (D, variant EC₅₀ 0.0064 μM, WT EC₅₀ 0.32 μM) compared to WT controls recorded on the same day. (E) There was no detectable effect on the potency of Mg²⁺ inhibition evaluated at −60 mV (variant IC₅₀ 20 μM, WT IC₅₀ 22 μM, n = 12–19). (F) There was a clear reduction in proton sensitivity, with the ratio of current recorded at pH 6.8 to pH 7.6 for variant being 29 ± 2.8% (n = 15) and for WT being 15 ± 0.60% (n = 10). (G,H) The weighted mean time constant (Tau_WEIGHTED) describing the time course for deactivation following rapid removal of glutamate was prolonged from 708 ± 33 ms (n = 17) to 8,370 ± 745 ms (n = 5) for variant. (I) The open probability was increased for the variant (0.082 ± 0.0046, n = 13) compared to WT (0.043 ± 0.0076, n = 13). (J) There was no detectable change in the surface expression relative to WT controls (100 ± 12%, n = 9 experiments). (K-N) Using adult WT and GluN2B-S810R heterozygous mouse whole brain, a subcellular fractionation assay was performed as described in Materials and Methods, and homogenates (Total) and PSD lysates were immunoblotted with indicated antibodies. (M,N) Quantification of blots divided by β-actin, then normalized to WT (n = 4 independent experiments). Each band intensity in homogenate (Total) using GluN2A (P = 0.0955), GluN2B (P = 0.2017), GluN1 (P = 0.2544), GluA1 (P = 0.3796), GluA2 (p = 0.0581), and PSD-95 (p = 0.4787), as well as GluN2A (p = 0.1368), GluN2B (P = 0.3667), GluN1 (P = 0.1865), GluA1 (P = 0.2895), GluA2 (P = 0.2615), and PSD-95 (P = 0.3550) in the PSD fraction was measured using ImageJ software (NIH). Error bars represent ± SEM, * P < 0.05, unpaired t-test.

Figure 4. Effects of GluN2B-S810R on NMDAR-mediated EPSCs, dendritic spine density, cerebral glucose metabolism, behaviors, and seizure threshold.

(A) Representative CA1 pyramidal cells that were filled with dye following patch clamp recording. (B) Representative evoked NMDAR-mediated EPSCs were recorded in CA1 pyramidal cells in response to Schaffer collateral stimulation in the presence of the AMPA receptor antagonist NBQX and reduced extracellular Mg²⁺ (0.2 mM). (C) Input-output curve for stimulus intensity vs the amplitude of the evoked NMDAR-mediated component of the EPSC (eEPSC) in CA1 pyramidal cells recorded from WT and GluN2B-S810R hippocampal slices. (D) Evoked NMDAR-mediated EPSC amplitude for WT and GluN2B-S810R CA1 pyramidal cells. (E,F) Dendritic spines in CA1 pyramidal cell by Golgi staining showed a significantly reduced total spine density and mature type mushroom spine in S810R (n = 4 independent experiments). (G,H) Effect of the S810R variant on cerebral glucose metabolism evaluated by dynamic PET scans (FDG). (I,J,K,L,M) Effect of the S810R variant on motor coordination (rotarod; I), locomotion (J), recognition memory (novel object recognition, NOR; K), spatial learning (Y maze; L), and anxiety level (EZM, test; M). (N) Effect of the S810R variant on the seizure threshold (PTZ). MJ: myoclonic jerk, GTCS: generalized tonic colonic seizures; unpaired t-test. * P < 0.05.

Figure 5. Effects of radiprodil and memantine on agonist-evoked GluN2B NMDAR currents, evoked NMDAR-mediated EPSCs, and seizure threshold.

(A,C) Composite concentration-response curves for radiprodil (A) and memantine (C) at a holding potential of −40 mV by the two-electrode voltage clamp current recordings from Xenopus oocytes expressing GluN1/GluN2B and GluN1/GluN2B-S810R. (B,D) Effects of radiprodil (B, 3 μM) and memantine (D, 30 μM) on evoked NMDAR-mediated EPSCs for WT and GluN2B-S810R were recorded from CA1 pyramidal cells in response to Schaffer collateral stimulation in the presence of the AMPA receptor antagonist NBQX (10 μM for GluN2B-S810R) and reduced extracellular Mg²⁺ (0.2 mM). (E,F) Effect of radiprodil (3.0 mg/kg, i.p.) and memantine (20 mg/kg, i.p.) on the latency to myoclonic jerk or generalized tonic-clonic seizures (GTCS) induced by PTZ (40 mg/kg, i.p.).