The NMDA receptor subunit GluN2D is a potential target for rapid antidepressant action

Abstract

Ketamine is the first glutamatergic agent in clinical use for major depression, but its primary target remains unclear. Further research is needed to develop more specific interventions with fewer side effects. Ketamine is a noncompetitive antagonist of the glutamatergic N-methyl-D-aspartate (NMDA) receptor. Here, we show that ketamine preferentially targets GluN2D-containing NMDA receptors on interneurons, and that selective GluN2D antagonism is sufficient to produce rapid antidepressant-like effects. We use ketamine, the selective GluN2C/D inhibitor NAB-14, Grin2d-siRNA and chemogenetic approaches in hippocampal slices and in vivo mice. We find that GluN2D antagonism inhibits NMDAR currents in interneurons but not pyramidal cells, and that GluN2D-mediated recruitment of GABAergic interneurons controls inhibitory circuits regulating hippocampal activity and plasticity. In a mouse model of depression, GluN2D inhibition recovers excitation-inhibition balance, restores plasticity, and mimics antidepressant-like actions of ketamine with fewer side effects. These findings identify GluN2D as a highly specific target for novel antidepressant therapy.

Fig. 1. Differential inhibition of NMDA EPSCs in pyramidal cells and interneurons depends on the selectivity of antagonists for GluN2D.

a Schematic overview of patch-clamp measurements in hippocampal brain slices. Synaptic activity was evoked in CA1 pyramidal cells (PCs, pink) by Schaffer collateral (SC) stimulation and in interneurons (IN, green) by extracellular stimulation in the stratum oriens. Representative image of SOM-INs (green) in wild-type C57Bl6 (native, scale bar: 20 µm) and SOM-Cre (SST tm2.1(cre)Zjh/J) td-Tomato mice (fluorescent, scale bar: 100 µm). b Left: Cryo-EM structure of the heterotetrameric GluN1a/GluN2D NMDAR (PDB ID: 7YFF), which is composed of two GluN1a chains (cyan) and two GluN2D chains (orange). It consists of three domains: the N-terminal domain (NTD), the agonist-binding domain (ABD), and the transmembrane domain (TMD). Right: 2D structure and docking model of NAB-14 (green) bound to GluN2D-TMD. The blue and yellow dotted lines represent pi-stacking interactions and H-bonds, respectively. c Representative NMDAR-EPSCs before (black) and after 10 μM NAB-14 (green) in INs and PCs. d NAB-14 dose-dependently suppressed EPSCs more in INs (n = 12 [1/5/20 μM], n = 7 [10 μM]) than PCs (n = 11 [1/5/20 μM], n = 7 [10 μM]). Mixed-effects model (REML) with Geisser-Greenhouse correction found main effects of concentration (F = 15.82, p < 0.0001) and cell type (F = 8.33, p = 0.0088). Tukey’s post hoc tests showed significantly stronger inhibition in INs at 5 μM (p = 0.033), 10 μM (p = 0.036), and 20 μM (p = 0.010). Two-sided tests. e Ketamine inhibited EPSCs more in INs (n = 7) than PCs (n = 10 [1/5/20 μM], n = 8 [10 μM]). Mixed-effects model (REML) with Geisser-Greenhouse correction showed main effects of concentration (F = 20.23, p < 0.0001) and cell type (F = 11.40, p = 0.0042). Tukey’s post hoc tests indicated stronger IN inhibition at 1 μM (p = 0.029), 10 μM (p = 0.005), and 20 μM (p = 0.038). Two-sided tests. All data are shown as means ± SEM. Symbols: *p < 0.05, **p < 0.01. n = number of cells. Source data are provided as a Source Data file. Created in BioRender. Vestring, S. (2025) https://BioRender.com/dpm4zwz.

Fig. 2. Inhibition of GluN2D decreases the activity of feed-forward and feedback loops.

a Schematic overview of hippocampal feedback- and feedforward loops. PC, pyramidal cell. SOM, somatostatin-expressing interneurons; PV, parvalbumin-expressing interneurons; SLM, stratum lacunosum moleculare; SR, stratum radiatum; SP, stratum pyramidale; SO, stratum oriens. b (Left) Time course and (right) group analysis of normalized maximum EPSP amplitudes following wash-in of 10 µM NAB-14 (n = 8), 20 µM NAB-14 (n = 9), or vector control (n = 8). One-way ANOVA (F = 5.59, p = 0.011) with Dunnett’s post hoc test showed a significant increase at 20 µM (p = 0.0056), not at 10 µM (p = 0.18), vs. baseline. c (Left) Time course and (right) group analysis of normalized maximum EPSPs after 10 µM (n = 17) or 100 µM (n = 8) KET, or vector control (n = 8). One-way ANOVA showed no effect (F = 0.318, p = 0.73); Dunnett’s post hoc: p = 0.99 (10 µM), p = 0.69 (100 µM). d (Left) Representative microcircuit-activating (MICA) protocol voltage traces at baseline (black) and after 10 µM NAB-14 (green); insets show magnified EPSPs. (Middle) Normalized EPSP amplitudes increased significantly after NAB-14 (p = 0.0019, two-tailed paired t-test, n = 9), with no change in control (p = 0.51, n = 5). (Right) EPSP decay constant (τ) increased in NAB-14-treated cells (n = 8) vs. control (n = 5; p = 0.0071, two-tailed unpaired t-test). e (Left) Representative voltage traces at baseline (black) and after 10 µM KET (blue); insets show magnified EPSPs. (Middle) Normalized maximum EPSP amplitudes increased significantly after KET (p = 0.0019, two-tailed paired t-test, n = 13), with no change in control (p = 0.3289, n = 7). (Right) Decay constant (τ) increased in KET-treated cells (n = 12) vs. control (n = 9; p = 0.0028, two-tailed unpaired t-test). f NAB-14 converts subthreshold EPSPs to APs. (Left) Representative traces before (black) and after NAB-14 (green). (Right) Heatmap showing mean AP occurrence per 2.5 min across neurons over 30 min. g KET converts subthreshold EPSPs to APs. (Left) Representative traces before (black) and after KET (blue). (Right) Heatmap showing mean AP occurrence per 2.5 min across neurons over 30 min. All data are shown as means ± SEM. *p < 0.05, **p < 0.01. n = number of cells. Source data are provided as a Source Data file.

Fig. 3. GluN2D inhibition increases event-related excitatory network activity in the mPFC.

a Experimental procedures for fiber photometry experiments. b Basal bulk activity remained unchanged after the injection of saline (red), 10 mg/kg NAB-14 (green) or 10 mg/kg KET (blue); averaged time traces (n = 4). n=number of animals. c (Left) Mean ΔF/F changes over time during video-tracked object exploration in mice after injection of saline (red), KET (blue), or NAB-14 (green). (Middle) Individual data points and group means of the area under the curve (AUC; ΔF/F×sec) during repeated exploratory events. (Right) Summary group analysis. One-way ANOVA revealed a significant treatment effect (F = 8.41, p = 0.0003). Tukey’s post-hoc tests showed significantly greater responses for NAB-14 vs. saline (p = 0.0006) and KET vs. saline (p = 0.0084), with no difference between NAB-14 and KET (p = 0.28). Data represent mean ± SEM from n = 29 (NAB-14), 75 (KET), and 61 (saline) event-related responses; collected from 12 mice total (4 per treatment group). **p < 0.01, ***p < 0.001. Source data are provided as a Source Data file. Created in BioRender. Vestring, S. (2025) https://BioRender.com/dpm4zwz.

Fig. 4. Inhibition of GluN2D increases hippocampal LTP.

a Weak-aLTP was significantly enhanced by 10 µM NAB-14 (n = 11), while no effects were observed at 1 µM (n = 8) or 5 µM (n = 12) compared to control (n = 25). A Kruskal-Wallis test revealed a statistical trend toward a treatment effect (Kruskal-Wallis statistic=7.39, p = 0.061). Post hoc uncorrected Dunn’s test indicated a significant increase at 10 µM NAB-14 compared to baseline (p = 0.0066), whereas neither 1 µM nor 5 µM NAB-14 significantly altered weak-aLTP (p = 0.48 and p = 0.38, respectively). b The addition of NAB-14 to the bath solution substantially increased the weak-aLTP in a monoexponential pattern (left). At 10 µM NAB-14 (red line), both LTP induction and differences in NMDAR-EPSCs between the IN and PC reached a maximum (monoexponential fit, right). c Ketamine modulated weak-aLTP in a non-linear, inverted U-shaped manner: no enhancement was observed at the low dose (1 µM, n = 17), a significant increase occurred at 5 µM (n = 19), and this effect was lost again at 10 µM (n = 13) compared to control (n = 25). A Kruskal–Wallis test revealed a significant treatment effect across groups (Kruskal-Wallis statistic=7.893, p = 0.048). Post hoc uncorrected Dunn’s test confirmed a significant enhancement at 5 µM ketamine compared to baseline (p = 0.0054), while no significant changes were detected at 1 µM or 10 µM (p = 0.57 and p = 0.42, respectively). d The inducibility of weak-aLTP follows an inverted U-shaped relationship with the KET concentration added to the bathing solution (Gaussian fit). The maximum (red lines) weak-aLTP induction (135.9% of baseline) was calculated at 5.3 µM KET (left). Monoexponential fits to the decay of NMDAR-EPSCs in INs and PCs revealed that the difference in current inhibition between INs and PCs at 5.3 µM KET falls within the 5th percentile of the maximal observed difference (right). All data are presented as mean ± SEM. *p < 0.05, **p < 0.01. n = number of cells. Source data are provided as a Source Data file.

Fig. 5. Chemogenetic manipulation of SOM-INs bidirectionally modulates LTP.

a Schematic representation of DREADD (designer receptor exclusively activated by designer drugs) experiments in SOM-Cre mice. CNO, clozapine-N-oxide. b Fluorescence image of the CA1 region three weeks after DREADD virus injection in SOM-Cre mice showing the global expression of DREADD on SOM-INs (scale bar: 250 µm). This experiment was replicated once as proof of concept. Str., stratum. c Activation of inhibitory DREADD (Gi) in SOM-INs by CNO (2.5 µM) resulted in an increase in weak-aLTP (two-tailed paired t-test, p = 0.019, n = 12), whereas activation of excitatory DREADD (Gq) on SOM-INs blocked weak aLTP induction (two-tailed paired t-test, p = 0.093, n = 6). Group comparison: Two-tailed unpaired t-test (p = 0.0060). All data are presented as mean ± SEM. **p < 0.01. n = number of cells. Created in BioRender. Vestring, S. (2025) https://BioRender.com/dpm4zwz. Source data are provided as a Source Data file.

Fig. 6. Inhibition of GluN2D restores plasticity in an animal model of depression.

a Schematic overview of the Repeated Stress Model (RS). b After RS, no significant aLTP was induced (magenta; two-tailed paired t-test, p = 0.4354, n = 11), but a single NAB-14 injection (10 mg/kg, green; two-tailed paired t-test, p = 0.0001, n = 16) fully restored LTP. Exemplary EPSP traces (left), time course (middle) and group analysis (right) of the effect of NAB-14 on aLTP inducibility. Two-tailed unpaired t test control/vector vs. NAB-14 group, p = 0.0028. c A single KET injection (10 mg/kg, blue; two-tailed paired t-test, p = 0.0045, n = 6) resulted in a full rescue of RS-induced impaired aLTP (magenta; two-tailed paired t-test, p = 0.9272, n = 7). Exemplary EPSP traces (inset), time course (left) and group analysis (right). Two-tailed unpaired t-test control vs. KET group, p = 0.0013. d Treatment with intrathecal Grin2d siRNA (orange; two-tailed paired t-test, p = 0.0002, n = 21) but not with scrambled siRNA (magenta; two-tailed paired t-test, p = 0.2858, n = 19) reversed the RS-induced impairment of LTP. Exemplary EPSP traces (inset), time course (left) and group analysis (right). Two-tailed Mann-Whitney test, p = 0.0234. e RS reduced spontaneous excitatory postsynaptic potentials (EPSPs), which were significantly restored by both KET and NAB-14. The experimental timeline is shown on the left, representative traces for each condition are presented in the center, and quantification of spontaneous EPSP frequency is displayed in the bar graph on the right. A Kruskal–Wallis test revealed a significant overall effect of treatment (Kruskal-Wallis statistic = 10.03, p = 0.018). Two-tailed post hoc Dunn’s test indicated that RS (n = 20) significantly reduced EPSP frequency compared to naive animals (p = 0.039, n = 16), and both KET (10 mg/kg, n = 14, p = 0.026) and NAB-14 (10 mg/kg, n = 16, p = 0.0031) reversed this effect. Data are shown as mean ± SEM. n = number of cells. *p < 0.05, **p < 0.01, ***p < 0.001. Created in BioRender. Vestring, S. (2025) https://BioRender.com/dpm4zwz. Source data are provided as a Source Data file.

Fig. 7. GluN2D inhibition rescues stress-induced synaptic deficits.

a Workflow (left) and quantification (right) of biocytin-filled spine analysis. One-way ANOVA showed a treatment effect on spine density (F = 3.36, p = 0.030). Dunnett’s post-hoc test revealed increased density in NAB-14 vs. RS (p = 0.021; n = 7 vs. 12), with no differences for KET (p = 0.73, n = 7) or naive controls (p > 0.99, n = 11), n=number of cells. b Representative images of biocytin-filled (scale bar: 200 µM) and high-magnification dendritic segments (scale bar: 10 µM). c Immunohistochemistry procedures. d Representative vGlut1-immunostained boutons (stratum pyramidale, scale bars: 5 µm) under naive, RS, RS + NAB-14, and RS + KET conditions. Quantification of vGlut1 puncta revealed a significant treatment effect (Kruskal-Wallis, p = 0.0054). Dunn’s multiple comparisons test showed a significant increase in puncta density in the NAB-14-treated group (n = 10) compared to RS (p = 0.0248, n = 10), no differences were observed for KET-treated mice (p > 0.99, n = 9) or naive controls (p = 0.2599, n = 6). n = number of animals. e Representative GAD65-immunostained GABAergic boutons (stratum moleculare, scale bars: 5 µm) under naïve, RS, RS + NAB-14, and RS + KET. Quantification of puncta revealed a significant effect of treatment (one-way ANOVA, F = 8.02, p = 0.0017). Dunnett’s post-hoc test showed an increase in puncta density in the NAB-14-treated group (n = 6) compared to RS (p = 0.0005, n = 4), while no differences were observed for the KET-treated group (p = 0.1582, n = 5) or naive controls (p = 0.0813, n = 5). n=number of animals. f Western Blot (WB) procedures. g Group analyses of WBs. NAB-14 and KET treatment increased PSD95 in RS mice (two tailed-one sample t-test, normalized to RS group). Naive: p = 0.441, KET: p = 0.022, NAB-14: p = 0.048, all n = 8. n = number of animals. GAPDH (36 kDa, loading control) derived from the same experiments (gels and blots processed in parallel). h Group analyses of WBs. KET and NAB-14 treatment increased GluR1A in RS mice (two tailed-one sample t-test, normalized to RS group). Naive: p = 0.058, KET: p = 0.003, NAB-14: p = 0.039, all n = 8. n = number of animals. GAPDH (36 kDa, loading control) was blotted on the same membrane. Data are shown as mean ± SEM. p < 0.05, **p < 0.01, ***p < 0.001. Created in BioRender. Vestring, S. (2025) https://BioRender.com/dpm4zwz. Source data are provided as a Source Data file.

Fig. 8. NAB-14 crosses the blood–brain barrier and GluN2D inhibition reverses stress-induced behavioral effects.

a Schematic representation of pharmacokinetic and brain penetration workflow after intravenous administration of 1 or 2 mg/kg NAB-14 in male rats. b (Left): Plasma concentration-time profiles following NAB-14 i.v. administration at 1 mg/kg (light green) and 2 mg/kg (dark green). (Middle): Corresponding NAB-14 concentrations in brain homogenates at the same time points. All n = 3 per dose and time point. N = number of animals. (Right): Brain-to-plasma concentration ratios at pooled time points 5/15/30/60 min. for 1 mg/kg (n = 10) and 2 mg/kg (n = 12) NAB-14 administration. N = number of animals. c (Top): Workflow for assessing escape behavior after repeated stress (RS)(Bottom): Short-term effects were tested 2 hours after i.p. administration of NAB-14 (5 or 10 mg/kg), KET (10 mg/kg), or saline. Two-tailed paired t-tests showed no effect for saline compared to values at day 5 of the RS protocol (p = 0.091, n = 10), but significant reductions in immobility time for NAB-14 (5 mg/kg: p = 0.042, n = 10; 10 mg/kg: p = 0.030, n = 10) and KET (p = 0.032, n = 7). Long-lasting effects were assessed 48 h after NAB-14 or KET administration and 72 h after Grin2d siRNA injection. Grin2d siRNA reduced immobility at day 5 after RS (two-tailed paired t-test, p = 0.019, n = 8). One-way ANOVA across NAB-14, KET, and saline revealed a treatment effect (F = 5.95, p = 0.0104, all n = 7), with Dunnett’s post hoc showing reduced immobility for KET (p = 0.024) and NAB-14 (p = 0.010) vs. saline. d (Top): Schematic overview of the procedures in the behavioral RS experiments assessing reward behavior. (Bottom): Treatment with NAB-14 (10 mg/kg), KET (10 mg/kg) and Grin2d siRNA reversed the stress-induced impairment of sucrose preference at 24 (POST 1) and 48 (POST 2) hours. Treatment with the vector used to dissolve NAB-14 and scrambled siRNA did not normalize sucrose preference. Repeated measures ANOVA, vector injection: F = 5.564, p = 0.006, n = 6; NAB-14: F = 13.02, p < 0.0001, n = 7; KET: F = 18.94, p < 0.0001; n = 5; scrambled siRNA F = 3.618, p = 0.031, n = 6; Grin2d siRNA F = 11.09, p = 0.0002, n = 6. Data are shown as mean ± SEM. N = number of animals. p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Created in BioRender. Vestring, S. (2025) https://BioRender.com/dpm4zwz. Source data are provided as a Source Data file.

Fig. 9. No effect of NAB-14 on spatial memory, motor coordination, locomotion or anxiety-like behavior.

a Following training with two objects, mice received saline, 10 mg/kg NAB-14, or 10 mg/kg KET. Six hours later, one object was relocated (object location memory task, OLT). All groups showed increased exploration of the moved object, with no group differences. Two-way ANOVA: p = 0.0003 for time, 0.715 for treatment, p = 0.805 for time x treatment. Tukey’s multiple comparison posttests, baseline vs. test session: saline p = 0.044, n = 7; NAB-14: p = 0.033, n = 8; KET: p = 0.004, n = 9. n=number of animals. b The latency to fall in the rotarod test was reduced by KET (10 mg/kg, blue, n = 6) compared to saline (black, n = 6) or NAB-14 (10 mg/kg, green, n = 7) injection. (Left): Two-way ANOVA: p < 0.0001 for time, p = 0.002 for treatment, 0.492 for trial × treatment. Post-hoc Tukey’s test: vehicle vs. KET: p = 0.0073, vehicle vs. NAB: p = 0.8571, KET vs. NAB: p = 0.0018. n=number of animals (Right): Group analysis of the means of all time points per group (n = 4, n represent the four time points). One-way ANOVA F = 10.25, p = 0.0048. Post-hoc Tukey’s test: vehicle vs. KET: p = 0.0137, vehicle vs. NAB-14: p = 0.8608, KET vs. NAB-14: p = 0.0062. c Lorazepam treatment (LOR, red, 0.25 mg/kg, n = 7) decreased the distance traveled in the open field test, while KET (10 mg/kg, n = 7) and NAB-14 (10 mg/kg, n = 9) hat no effect vs. saline (n = 8). One-way ANOVA F = 6.163, p = 0.0025. n=number of animals. d The time spent in the center of the open field (as a percentage of the total time) was affected by KET (10 mg/kg, n = 7) treatment but not by NAB-14 (10 mg/kg, n = 9) treatment compared to saline (n = 9). One-way ANOVA F = 7.096, p = 0.004. Post-hoc Tukey’s test: control vs. NAB-14: p = 0.632, control vs. KET p = 0.027, NAB-14 vs. KET: p = 0.004. n = number of animals. Data are shown as mean ± SEM. *p < 0.05, **p < 0.01. Created in BioRender. Vestring, S. (2025) https://BioRender.com/dpm4zwz. Source data are provided as a Source Data file.

Fig. 10. Schematic representation of the mechanistic cascade linking GluN2D to the development and treatment of depression.

Left panel (rose background): Repeated stress induces a depressive-like state characterized by a shift in the excitation–inhibition (E/I) balance toward inhibition, impaired synaptogenesis, reduced AMPAR-mediated currents, and loss of associative long-term potentiation (LTP) in the hippocampus. Behaviorally, this is reflected by reduced escape responses and blunted reward sensitivity. Right panel (green background): Inhibition of GluN2D-containing NMDARs by ketamine, NAB-14, or Grin2d siRNA restores network function. In this recovered state, the E/I balance is normalized, synaptogenesis is enhanced, AMPAR currents are enhanced, and Schaffer collateral LTP is reinstated - paralleled by normalization of escape- and reward-related behaviors. Central upper panel: The cryo-EM structure of the heterotetrameric GluN1a/GluN2D NMDAR highlights the putative NAB-14 binding site within the receptor complex (see Fig. 1b for details). Central middle panel: A simplified diagram of the hippocampal microcircuit illustrates glutamatergic excitatory inputs onto pyramidal neurons and interneurons within feedforward and feedback loops (cf. Figure 2a). Magnified insets show that synapses onto interneurons are enriched in GluN2D-containing NMDARs, rendering them selectively responsive to ketamine, NAB-14, or Grin2d siRNA, whereas GluN2D-negative synapses are comparatively unaffected. Representative NMDAR current traces illustrate that NAB-14 reduces NMDAR-mediated EPSCs at GluN2D-positive synapses (green vs. black), but not at GluN2D-negative sites (cf. Fig. 1c). Central lower panel: Representative images show reduced synapse density following repeated stress and its reversal after GluN2D-targeted interventions (cf. Fig. 7d). Correspondingly, hippocampal recordings indicate that the stress-induced blockade of LTP is rescued by in vivo NAB-14 treatment (cf. Fig. 6b). The bottom line illustrates the normalization of escape and reward behavior following GluN2D modulation (cf. Fig. 8c, d), highlighting a multilevel restoration of structure, function, and behavior. Created in BioRender. Vestring, S. (2025). https://BioRender.com/dpm4zwz.