Allosteric Inhibition of NMDA Receptors by Low Dose Ketamine

Abstract

Ketamine, a general anesthetic, has rapid and sustained antidepressant effects when administered at lower doses. Anesthetic levels of ketamine reduce excitatory transmission by binding deep into the pore of NMDA receptors where it blocks current influx. In contrast, the molecular targets responsible for antidepressant levels of ketamine remain controversial. We used electrophysiology, structure-based mutagenesis, and molecular and kinetic modeling to investigate the effects of ketamine on NMDA receptors across an extended range of concentrations. We report functional and structural evidence that, at nanomolar concentrations, ketamine interacts with membrane-accessible hydrophobic sites on NMDA receptors, which are distinct from the established pore-blocking site. These interactions stabilize receptors in pre-open states and produce an incomplete, voltage- and pH-dependent reduction in receptor gating. Notably, this allosteric inhibitory mechanism spares brief synaptic-like receptor activations and preferentially reduces currents from receptors activated tonically by ambient levels of neurotransmitters. We propose that the hydrophobic sites we describe here account for clinical effects of ketamine not shared by other NMDA receptor open-channel blockers such as memantine and represent promising targets for developing safe and effective neuroactive therapeutics.

INTRODUCTION

Ketamine is a synthetic small molecule with a complex spectrum of clinical effects ¹. At micromolar plasma concentrations (5 – 10 μM), it induces general anesthesia by blocking excitatory synaptic transmission mediated by NMDA receptors ^{2, 3}. The mechanism involves ketamine binding at a centrally located binding site in the pore, where it reduces current flow with a use-dependent open-channel block mechanism. Ketamine shares this binding site and blocking mechanism with several other neuroactive drugs that cause sedation and loss of consciousness, including its parent compound PCP, the anticonvulsant MK-801, and the neuroprotective memantine ^4–6. However, at sub-micromolar plasma concentrations (0.35 – 0.85 μM), ketamine has anti-depressant, analgesic, and anti-inflammatory effects that are not shared by other NMDA receptor open-channel blockers ^{7, 8}.

The molecular mechanisms by which the clinical benefits of low-dose ketamine occur are unknown. The anti-depressant effect is of special interest because depression is one of the most prevalent mood disorders and a leading cause of disability globally. Moreover, existing therapies take weeks to work and remain ineffective for most patients ⁹. In contrast, the antidepressant effect of low-dose ketamine is apparent within hours, can last for up to a week, and can resolve treatment-resistant symptoms and suicidal ideation ^8–10. Multiple lines of evidence implicate NMDA receptors as the molecular target for the clinical effects of low-dose ketamine, however this hypothesis remains controversial ^{7, 11, 12}. One of the puzzling observations is that the effective antidepressant dose is much lower than its potency as an open channel blocker. Moreover, memantine, a drug that blocks NMDA receptor currents with potency, kinetics, and channel mechanism very similar to those of ketamine, has no anti-depressive effects ¹³. Lastly, the reported potencies of the two ketamine enantiomers, S and R, appear distinct, with S-ketamine (S-KET) being the more potent species as an NMDA receptor blocker, and R-ketamine (R-KET) the more effective anti-depressant ¹⁴.

We noticed that the available data on the effects of ketamine on NMDA receptor currents is surprisingly sparse in the sub-micromolar range of the dose-response relationship and the reported values for the half-maximal effective concentration of ketamine (EC₅₀) on NMDA receptor currents span a broad range (0.3 – 16 μM) ^{3, 15, 16}. This variability likely reflects the strong dependency of EC₅₀ values on experimental conditions such as the concentration and duration of agonist application, the cellular preparation and its transmembrane voltage, as well as the external H⁺, Mg²⁺, and Ca²⁺ concentrations, all of which differ considerably across studies ^15–18. To gain clarity on the effects of ketamine on NMDA receptor currents, we used molecularly defined preparations where we can control the many variables salient to the ketamine E₅₀ and we optimized the experimental conditions to reexamine the dose-response relationship over an extended range of concentrations.

Material and Methods

All electrophysiology reagents were of >99% purity and were from Sigma-Aldrich unless stated otherwise. Racemic (R, S)-ketamine (no: K2753–1; lot: SLCD3830) was from Sigma (CAS: 1867–66-9). (S)-ketamine (no: 9001961; batch numbers: 0599733–12, 0541513–21) and (R)-ketamine (no: 16519; batch numbers: 0595520–6; 0541512–9) were from Cayman Chemical. Polyethylenimine (linear, mol. wt. 25,000) was purchased from Polysciences (Polysciences, Inc.). All reagents for tissue culture were purchased from ThermoFisher Scientific and Corning.

Cell culture and transfections

Human embryonic kidney (HEK) 293 cells (ATCC CRL-1573) were maintained in Dulbecco’s Modified Eagle Medium with 10% Fetal Bovine Serum and 1% penicillin-streptomycin, at 37 °C in 5% CO₂ and 95% atmospheric air. Cells were transfected with plasmids expressing rat GluN1–1a (U08262.1), rat GluN2A (AF001423), rat GluN2B (M91562.1), rat GluN2C (NM_012575.3; GI:148747562), rat 2D (L31611.1; GI: 469066) or mutants as indicated, and GFP at a 1:1:1 ratio using polyethyleneimine. Transfected cells were grown in medium supplemented with 10 mM MgCl₂ and used for experiments 24 – 48 hours thereafter. All mutations were engineered with QuickChange and verified by full insert sequencing.

Whole-cell current recordings

Whole-cell currents were recorded with the patch-clamp method using polished glass electrodes filled with (intracellular) solution containing (in mM): 135 CsCl, 10 mM BAPTA, and 10 HEPES, adjusted to pH 7.2 (CsOH) and clamped at −100 mV. After achieving a high-resistance seal in the whole-cell configuration, we perfused the cell with extracellular solutions containing (in mM): 150 NaCl, 2.5 KCl, 0.25 CaCl₂, 0.01 EDTA, 0.1 glycine, and 10 HEPES, adjusted at pH 7.2 (NaOH). Glutamate (1 mM) and ketamine were added to this solution as indicated. The pressurized perfusion system (BPS-8, ALA Scientific Instruments) exchanged solutions within 0.3 – 0.5 s. Sodium currents were amplified, and low-pass filtered at 2 kHz (Axopatch 200B; 4-pole Bessel), sampled at 5 kHz (Digidata, 1440A) and stored as digital files with pClamp 10.2 software (Molecular Devices, Sunnyvale, CA). Traces were inspected for stable baseline and absence of leak current and were used to measure levels of peak and steady-state current amplitudes in pClamp 10.2.

Dose-response relationships.

To construct dose-response relationships we represented the ratio of steady-state current measured with ketamine (I_KET) relative to the initial glutamate-evoked steady-state level observed without ketamine (I_CTR), as a function of ketamine concentration. For each receptor construct, data from all ketamine doses were pooled and used to optimize variables for a monophasic dose-response curve of form:

$$y=A_1+\frac{A_2-A_1}{1+{10}^{\left({\mathrm{l}\mathrm{o}\mathrm{g}}_{10}E{C}_{50}-x\right)p}}$$

Or a biphasic dose-response curve of form:

$$y=A_1+\left(A_2-A_1\right)\left[\frac{p}{1+{10}^{({\mathrm{l}\mathrm{o}\mathrm{g}}_{10}E{C}_{\mathrm{50,1}}-x)h1}}+\frac{1-p}{1+{10}^{({\mathrm{l}\mathrm{o}\mathrm{g}}_{10}E{C}_{\mathrm{50,2}}-x)h2}}\right]$$

Where ${\mathrm{A}}_1$ is the upper limit, ${\mathrm{A}}_2$ is the lower limit, $\mathrm{p}$ is the proportionality constant, and $\mathrm{h}$ is the slope; all parameters were free to vary. If a model did not converge during the fit, it was rejected in favor of the other model. If both models converged, the superior fit was determined by calculating the F-statistic according to:

$$F=\frac{\left(SS{E}_1-SS{E}_2\right)/\left(d{f}_1-d{f}_2\right)}{SS{E}_2/d{f}_2}$$

Where $\mathrm{S}\mathrm{S}\mathrm{E}$ is the residual sum of squared error and $\mathrm{d}\mathrm{f}$ is the degrees of freedom calculated as N – V where N is the number of data points and V is the number of model parameters. The resulting F-statistic was used to extract a p-value using the built-in fcdf function in MATLAB 2021b (MathWorks). Statistical significance was defined as p < 0.05.

Single-molecule current recordings.

Electrical signals from individual receptors were recorded with the cell-attached patch-clamp technique. Polished glass electrodes where filled with (extracellular) solutions containing (in mM): 150 NaCl, 2.5 KCl, 1 EDTA, 10 HEPBS adjusted to pH 8.0 (NaOH) and saturating concentrations of agonist glycine (0.1 mM) and glutamate (1 mM), as optimized previously for adequate resolution of NMDA receptor gating events ¹⁹.

Inward sodium currents were recorded after applying +100 mV through the recording pipette (to an estimated final membrane-patch potential ~−115 mV) ²⁰. Currents were amplified and low-pass filtered at 10 kHz (Axopatch200B; 4-pole Bessel), sampled at 40 kHz (PCI-6229, M Series card, National Instruments, Austin, TX), and written into digital files with QuB acquisition software.

Single-channel currents were preprocessed off-line to correct for baseline drifts and spurious signals. Idealization was done with the segmental-k-mean (SKM) method in QuB with a 0.025 ms deadtime and no digital filtering to maximize event detection ²¹.

Kinetic modeling and simulations.

Kinetic modeling was done for data obtained from individual receptors using the maximum-interval likelihood (MIL) method in QuB software, which uses this idealized data to optimize rate constants in user-defined state models, after imposing a 0.075 ms deadtime ²². Models of increasing complexity were fitted individually to each data file and the best fitting model was selected based on optimal visual agreement between the predicted probability density function and the data and the maximum of the log-likelihood (LL) function. The optimal number of states was determined according to Akaike criterion, which requires that the LL increase by at least 20 units for a state to be considered necessary in the model. The arrangement of states within the model was imposed a priori based on previously established models for GluN2A receptors ²³.

Microscopic rates for ketamine binding (k_on) and dissociation (k_off) were derived by progressive building of a tiered state model. First, gating rates were determined for the ketamine-free tier by fitting a 5-closed/1-open (5C1O) state model globally to pooled data files (n = 8) obtained in the absence of ketamine. Once determined, all rates in this model were fixed and served as the base ketamine-free tier. Next, we determined the ketamine binding rates for low doses (< 25 nM) by progressively adding states to the base model and globally fitting the resulting models simultaneously to data from multiple doses (n = 6, 2 recordings for each 5 nM, 10 nM, and 25 nM) to build an additional tier, representing ketamine-bound receptor states. Variants of the final two-tier model were made by changing the states that allowed transitions between tiers.

Simulations of open probability over time in response to pulses of glutamate were performed with the Q-matrices derived in QuB software for each model variant using the built-in expm function in MATLAB 2021b (MathWorks). Glutamate binding steps were added to the final model post-hoc using rate constants derived previously ^{24, 25}. Glutamate binding was restricted to the C3 state of the ketamine-free tier of the model. All simulations were performed assuming constant, saturating levels of glycine.

Molecular docking and MD simulations.

We used the crystal structure of S-KET⁺, or its mirror image for R-KET⁺. To model the protein, we used three structural models that capture distinct NMDA receptor conformations. These represent a closed-pore conformation obtained in the presence of inhibitors (inhibited) (PDB IDs: 6whs) ²⁶, a closed-pore semi-active conformation (pre-open) (PDB ID: 6wi1) ²⁶, and a modeled open-pore conformation (open) (https://zenodo.org/records/10970346) ²⁷. We used these structures as templates for the Swiss-Model homology modeling (https://swissmodel.expasy.org/) using sequences reported for GluN1 (P35439), and GluN2A (Q00959) subunits. We positioned their transmembrane domains in membranes with the PPM server (https://opm.phar.umich.edu/ppm_server) and used these as targets for docking and for MD simulations.

For global docking, we used S-KET⁺ and the transmembrane domains of the above three structures, and the (0,0,7) coordinate as the center of a grid box with dimensions: 30 Å × 30 Å × 25 Å.

For local docking, we used as starting points the results from global docking and used either S-KET⁺ or R-KET⁺, for the three putative sites of the three structures. We set the center of mass of ketamine as the center of a 10 Å cubic grid box, with the exhaustiveness at 1000. Molecular docking was performed using AutoDock Vina ²⁸.

For MD simulations, we set up a total of 12 systems, with S-KET⁺ or R-KET⁺ docked into site 1 or sites 2/3 of the three structures, from which we excluded N-terminal domains to save computation time. We performed three 200-ns long MD simulations for each system as follows. The forcefield for ketamine was prepared with the CHARMM General Force Field ²⁹ and the charge parameters adopted from previous work ³⁰. The simulation systems were prepared with the Membrane Builder function of the CHARMM-GUI webserver ^{31, 32} to embed the receptor in a bilayer of POPC lipids surrounded by a box of K⁺ and Cl⁻ ions in water (at a concentration of 0.15 M), which extended 20 Å from the protein in each direction. In the pre-open and open conformations, Cys mutations and disulfide bonds were introduced at GluN1 residues 499 and 686, and GluN2A residues 487 and 687, respectively, to mimic the binding of Gly and Glu ³³. Harmonic positional restraints (with force constant of 1 kcal/mol/nm²) were applied to the C-alpha atoms in the ligand-binding domain to prevent its distortion in the absence of the N-terminal domains. In total, the system has ~330,000 atoms. After energy minimization, we performed six steps of equilibration to gradually reduce harmonic restraints applied to the heavy atoms. We then conducted production MD runs in the NPT (isothermal-isobaric) ensemble using the Nosé-Hoover method ^{34, 35} at a temperature of 303 K. For pressure coupling, we used the Parrinello–Rahman method ³⁶. For non-bonded interactions, we used a 10-Å switching distance and a 12-Å cutoff distance. For electrostatics calculations, we used the particle mesh Ewald method ³⁷. We constrained the hydrogen-containing bond lengths with the LINCS algorithm ³⁸ to allow a 2-fs time step for the MD simulations. The MD simulations were performed by the GROMACS program ³⁹ version 2018-GPU using the CHARMM36 force field ^{40, 41} and TIP3P water model ⁴².

For binding free energy calculations, we used the last 50 ns of each trajectory and the Uni-GBSA program with default settings (https://github.com/dptech-corp/Uni-GBSA) ⁴³. For RMSD calculations, we aligned the backbone atoms of transmembrane helices and then computed RMSD for the heavy atoms of ketamine in the last 150 ns of each trajectory.

The steered MD simulation ⁴⁴ was done with the PLUMED2 program ^45–47 after equilibration. A total bias of 40 Å RMSD relative to the initial binding position was applied linearly in 2 ns as the collective variable to mimic an external unbinding force. The analysis and visualization of MD trajectories were carried out using the VMD program ⁴⁸ and Ligplot+ ⁴⁹.

RESULTS

Biphasic dose-response for ketamine inhibition of NMDA receptor currents

We measured the effects of racemic (R, S)-ketamine (KET) on whole-cell currents elicited from recombinant wild-type NMDA receptors (WT) containing the GluN1–1a and GluN2A subunits, with maximally effective agonist concentrations (1 mM Glu, 0.1 mM Gly), and strong hyperpolarization (−100 mV), in low Ca²⁺ (0.25 mM), and physiological pH (7.2). These conditions balance effects on KET potency, which increases with acidic pH and membrane potential ¹⁸, with effects on receptor gating, which decreases with acidic pH and Ca²⁺ concentrations ^{50, 51}. We applied KET at increasing concentrations (0.002 – 10 μM) onto the steady-state phase of the glutamate-elicited current (I_CTR) and measured the residual current in the presence of KET (I_KET) (Fig. 1A). Plotting the fraction of residual current (I_KET/I_CTR) as a function of KET concentration revealed a complex relationship, which has not been reported previously. Fitting a single-site dose response function to these data returned a potency of 0.37 ± 0.07 μM (R², 0.93; SSE, 2.28), which is at the low end of the previously reported range. However, a function with two sites described the data more closely (R², 0.98; SSE, 0.81) (Table S1). These results reveal a biphasic inhibitory effect of KET with a previously unrecognized inhibitory effect in the low nanomolar range (0.02 μM), which is separate from the one recognized previously (0.67 μM) (Fig. 1A).

Figure 1:

Extended dose-response relationships. (A) Two whole-cell current traces recorded from GluN1–1a/GluN2A receptors (24 cells, n >12 per concentration) expressed in HEK293 cells were elicited with glutamate (Glu, 1 mM) and −100 mV, allowed to equilibrate (ICTR), and racemic ketamine (KET) was applied at several concentrations (IKET). Panel at right shows pooled results (circles) overlayed with mono- (grey) and bi-phasic (red) dose-response functions (solid lines with associated 95% confidence intervals, shaded), and the calculated E50 values. (B) Summary of results obtained from cells co-expressing GluN1–1a with GluN2B (14 cells, n >3 cells per concentration), GluN2C (53 cells, n >10 cells per concentration), or GluN2D (37 cells, n >5 cells per concentration).

Ketamine has slightly different inhibitory potencies at the four principal NMDA receptor isoforms ^{15, 16, 52}. We used the assay described above to construct extended dose-response relationships for NMDA receptors containing GluN2B, GluN2C, or GluN2D subunits. These subunits are expressed differentially during development, and across brain regions, and have distinct biophysical and pharmacological, and physiological roles. We found that in all cases, the dose-response function was complex, and a biphasic function described the data more closely (Fig. 1B, Table S1). Therefore, all four receptor isoforms (GluN2A-D), this approach revealed two regions of KET inhibitory action: a high-potency region (EC₅₀, 0.02 – 0.2 μM), which accounted for 33 – 62% of the maximal effect, and a conventional-potency region (EC₅₀, 0.4 – 3 μM), which completed the inhibition.

Two types of ketamine-binding sites on NMDA receptors.

To search for potential ketamine-binding sites on NMDA receptors that are structurally distinct from the established central site, we used as a first pass approach global docking of protonated S-ketamine (S-KET⁺) on the transmembrane region of NMDA receptors. We used S-KET⁺ because it is a more potent blocker than unprotonated (S-KET) or its R enantiomer (R-KET⁺) ¹⁶. We focused on the transmembrane portion because ketamine can access its effector sites on NMDA receptors through membrane diffusion ⁵³ and because the level of inhibition varies with transmembrane voltage. Lastly, because open-channel block is strongly state dependent, with open receptors being more sensitive to inhibition ⁵⁴, we tested several structural models that capture receptors in discrete functional states. With these criteria, we selected three structures that also have adequate resolution within the transmembrane domain: a functionally inhibited, closed-pore conformation (inhibited), a semi-active closed-pore conformation (pre-open), and an active open-pore conformation (open) ^{26, 27}. The docking results suggested that S-KET⁺ may interact with the transmembrane domain of NMDA receptors at three sites: one located centrally in the pore corresponded to the standard blocking site, and two new lateral sites, located at the intersubunit interface adjacent to the pore.

Intrigued by these initial observations, we performed more extensive local docking of both S-KET⁺ and R-KET⁺ on the above three NMDA receptor structures as described in Methods. Consistent with previous reports, we identified GluN1 T648, V644, and N616, and GluN2A T646, L642, and N614 as principal contacts in the pore (site 1), as reported previously ^{4, 6}. In addition, S-KET⁺ engaged with GluN1 F558, W563, Y647, and L614 and GluN2A F636 as main contacts in the membrane-embedded lateral sites (sites 2/3), which were located within a lipid-filled tunnel described earlier ⁵ (Fig. 2A, S1).

Figure 2. Putative interactions between S-KET+ and NMDA receptors.

(A) Left, results from local docking of S-KET⁺ onto an inhibited NMDA receptor conformation (6whs) illustrates three putative binding sites (boxed in left panel): one located centrally in the pore (site 1, red) and two symmetry-related lateral sites: site 2 (dark blue) and site 3 (light blue). Right, detailed positioning of S-KET⁺ in site 1 (top) and in site 2 (bottom) and predicted key contacts with residues in GluN1 (blue) and GluN2A (green) subunits. (B) Trajectories predicted from MD simulation for S-KET⁺ within the lateral sites of an inhibited (left) and an open (right) conformations (blue). (C) Putative exit pathway for S-KET⁺ from site 2 of the inhibited conformation.

Next, to examine in more detail the dynamics and binding energies of ketamine at the central and lateral sites, we set up 12 MD simulations using the three structures with S-KET⁺ or R-KET⁺, docked in site 1 or sites 2/3, as described in Methods. At the central site, which we did not pursue further, our simulations captured multiple poses for ketamine, like what was observed in earlier studies ^{4, 6}. We focused next on the lateral sites, which have not been observed previously. For both S-KET⁺ and R-KET⁺, the dynamics in the lateral sites were clearly state dependent, with more stable binding in the inhibited conformation and increasing mobility in the pre-open and active conformations (Fig. 2B). This qualitative observation is supported by the calculated RMSD values, which increased from the inhibited, to the pre-open, and the open conformations, and the calculated binding free energies, which decreased in the same order (Tables S2,3). Such differential binding to distinct functional states is a strong indicator of an allosteric modifier of gating ⁵⁵. Based on these results, we hypothesized that ketamine binding to the lateral sites stabilizes receptors in closed-pore states to reduce NMDA receptor currents with an allosteric, rather than direct pore-block mechanism.

To examine how ketamine diffuses into and out of the lateral sites, we started from the S-KET⁺-docked inhibited structure and then applied an external unbinding force on the ligand to run steered MD simulations as described in Methods. We observed S-KET⁺ diffuse into the membrane along the lipid-filled tunnel described previously ⁵ (Fig. 2C and Movie S1). This result suggests that S-KET⁺ can access the lateral site of inhibited receptors by diffusion through the membrane. Repeating the simulation with S-KET⁺-docked in the lateral site of the open structure, we observed diffusion in both directions: along the hydrophobic tunnel into the membranes as seen in the inhibited structure and directly into the water-filled pore (Movie S2). Together these results suggest that ketamine can diffuse through the lipophilic tunnel in and out of the lateral sites regardless of receptor state, however as receptors transition towards active states, S-KET⁺ can also escape into the pore. This dual access route is consistent with previous reports that both ketamine and memantine can inhibit NMDA receptor currents through a membrane route ^{53, 56}, and may explain the previous observation that HNK, which is a ketamine metabolite, reduces subsequent currents even when pre-applied onto resting receptors ⁵⁷. Therefore, the lateral sites may represent bona-fide allosteric inhibitory sites, or they may serve as a reservoir of ligand, ready to slide into the pore and block currents once the receptor transitions into active conformations, as was demonstrated for memantine ⁵⁶.

Taken together, the docking and MD simulation results suggest that S-KET⁺ interacts with a previously unrecognized lateral site within the transmembrane domain of NMDA receptors, which differs from the centrally located site in three key aspects: it relies on contacts with distinct, mostly aromatic residues (GluN1 Y647 and GluN2A F636); it may be accessed from the membrane through a lipophilic tunnel; and it may inhibit NMDA receptor currents through an indirect allosteric mechanism rather than a direct open-channel blocking mechanism. Next, we set up to test experimentally these specific predictions with functional assays of intact full-length receptors in live cells.

Aromatic residues in the lateral sites modulate the nanomolar region of the extended dose-response relationship.

To examine whether the residues predicted by structural models to interact with ketamine contribute to its inhibitory effect on NMDA receptor currents, we constructed extended dose-response relationships, as described above, for a series of GluN1/GluN2A receptors that had single-residue substitutions at sites predicted to interact with site 1, site 2/3, or with all three sites (Fig. 3A). Results showed that modifying the side chains of residues that are specific to sites 2/3, such as GluN1 Y647 and GluN2A F636 changed the dose-response relationship in the low nanomolar range, as expected when perturbing a higher-affinity site. In contrast, modifying side chains of residues that are specific to site 1, such as GluN2A L642 or that are common to all three sites, such as GluN1 N616 strongly shifted the dose-response function in the micromolar range (Fig. 3B, C and Table S1).

Figure 3. Probing the predicted interactions between KET and NMDA receptors.

(A) Diagram highlights predicted contacts in the central and lateral sites. (B) Whole-cell traces illustrate Na⁺ currents elicited with Glu (1 mM) at pH 7.2 and −100 mV from cells expressing WT or mutated NMDA receptors, with the indicated series of KET concentrations. (C) Pooled data for mutated receptors (grey circles) (12 – 31 cells per construct, 6 – 15 cells per concentration) with overlayed best-fitting dose-response functions (solid lines with 95% confidence intervals as shaded area) and EC₅₀ values calculated with the best fitting function relative to results from WT receptors (dashed grey line).

R-KET⁺ forms additional contacts with the lateral sites

Existing reports demonstrate consistently stronger effects (2 – 5-fold) for S-KET relative to R-KET as a binding ligand or pore blocker of NMDA receptors ^{16, 58, 59}. However, the R enantiomer appears to be a more potent anti-depressant. This discrepancy has been used as an argument against a role for NMDA receptors in the antidepressant effect of ketamine ^{59, 60}. Presently, experimentally derived structural models are only available for the interaction of S-KET⁺ with residues in a closed-pore NMDA receptor conformation ^{4, 6}.Results from our structural modeling showed largely similar contacts for R-KET⁺ and S-KET⁺ in the lateral sites, except that the aromatic ring of R-KET⁺ afforded additional stabilizing interactions with GluN1 W563 and Y647 (Fig. 4A). These additional contacts allowed R-KET⁺ to remain more stable along the MD simulation relative to S-KET⁺, as reflected in both smaller RMSD values and larger binding energies (Fig. 4B, Tables S2,3).

Figure 4. Predicted contacts and effects of KET enantiomers.

(A) Detailed positioning of R-KET⁺ in site 2 of inhibited receptors illustrates its principal contacts with the receptor. (B) MD simulated trajectories for R-KET⁺ (blue) in site 2 of inhibited receptors. (C) Pooled whole-cell currents (circles) recorded from WT receptors and S-KET (left, 22 cells, n >11 per concentration) or R-KET (right, 16 cells, n >8 per concentration), overlayed with the respective best fitting dose-response functions, relative to KET (dashed line).

To test the prediction that R-KET⁺ interacts with the lateral sites more strongly than S-KET⁺, we constructed extended dose-response relationships for S-KET and R-KET, as described above for racemic ketamine (KET). Both enantiomers displayed complex dose-response behaviors (Fig. 4C). Fitting a monophasic dose response function to these data returned a 2-fold higher potency for S-KET relative to R-KET (EC₅₀, 0.24 μM vs 0.42 μM), as reported previously with a similar approach ¹⁶. However, a biphasic dose-response function described our extended data more closely and predicted a 2-fold higher potency for R-KET relative to S-KET in the nanomolar range (EC₅₀, 0.06 μM vs 0.13 μM), and similar potencies in the micromolar range. (Fig. 4C, S2 and Table S1).

These results suggest that although there are small structural differences in how the two enantiomers interact with the lateral sites, which may be leveraged to develop more specific ligands, the differences in potency are too small to explain enantiomer-specific clinical effects. Moreover, the approach of fitting dose response data with mathematical function is notoriously imprecise, and multiple factors (pH, Ca²⁺, Mg²⁺, voltage, etc.), which vary in situ, can influence simultaneously both the distribution of ketamine across pharmacologically active species (protonated enantiomers) and the distribution of NMDA receptors across its multiple states. Therefore, the reported differences in the clinical spectrum of the enantiomer are multifactorial, and likely unrelated to differences in contacts with NMDA receptor residues.

Ketamine can access its effector sites from the membrane through a lipophilic tunnel.

Next, we tested the prediction that ketamine, in its protonated form, can reach its effector sites on NMDA receptors through a membrane-accessible pathway. To test this hypothesis, we leveraged the cell-attached patch-clamp recording approach, where receptors are enclosed within the small tip of a recording pipette and thus are inaccessible to the solute bathing the cell (Fig. 5A). In this arrangement, stationary currents can be recorded for extended periods of time from receptors experiencing a constant external milieu, while the contents of the solution bathing the rest of the cell can be manipulated separately. We recorded on-cell Na⁺-only currents from NMDA receptors exposed to saturating concentrations of agonists (1mM Glu and 0.1 mM Gly), and in the absence of blocking divalent cations (1 mM EDTA) and low proton concentrations (pH 8) ¹⁹. In patches with a single active receptor, ascertained by the absence of overlapping openings, we recorded basal receptor activity for 5 min (pre) and then applied KET in the bath to record activity for another 30 min (post).

Figure 5: Probing membrane-access and mechanism of receptor inhibition by low dose KET.

(A) Cartoon illustrates the setup used to record on-cell stationary Na⁺ currents from receptors isolated within the recording electrode and exposed to agonists (Glu and Gly) at pH 8, with applied +100 mV (blue). After recording basal activity for 5 min (pre), we changed conditions in the bath, and recorded activity for another 30 min (post). (B) Example traces of unitary currents recorded from WT before (pre) and after (post) equilibration in KET at the indicated concentrations (top), and the associated time-dependent change in P_o obtained after adding 0 KET (grey) and the indicated KET concentrations (red). (C) Currents recorded after adding KET (1 μM) in a pH 8 (blue) or a pH 7.2 bath (grey). (D) Activity with 10 μM KET in the bath (pH 7.2) from F636A (yellow) and WT (grey) receptors. (E) Activity from WT (left) or F636A (right) with 1 μM MEM added to a pH 7.2 (grey, and yellow) or a pH 8 bath (blue). (F) Cartoon (left) illustrates setup used to record the unitary currents (right) with KET in the recording pipette (pH 7.2). (G) Left, NMDA receptor gating scheme consisting of a top KET-free tier and a bottom KET-bound tier, with C and O representing closed-pore and open-pore states, respectively. Right, simulated current responses obtained with brief (1 ms) or sustained (5 s) Glu (1 mM) application.

For WT receptors in HEK cells, we observed a time- and concentration-dependent reduction in receptor open probability (P_o) after adding KET into a pH 7.2 bath at concentrations of 1 μM (cellular resting membrane potential, ~−15 mV) ²⁰ (Fig. 5B). However, when added to a pH 8 bath, 1 μM KET had no effect on receptor activity (Fig. 5C). KET is a weak base with a reported dissociation constant (pKa) of 7.5. Our result that increasing the pH by 0.8 units reduced the effectiveness of bath-applied KET, indicates that the active inhibitory species is protonated even when KET can only access its effector site by membrane diffusion. Although KET is a strongly lipophilic molecule, with reported partition coefficient (logP) of 4.4, the partition coefficient for protonated ketamine in biological membranes is more difficult to ascertain and is likely voltage- and membrane-composition dependent. Therefore, the results from on-cell recordings can be compared only qualitatively with those obtained with whole-cell recordings, due to differences in membrane distribution for protonated ketamine, caused by (at least) 6-fold differences in the voltage differential driving its diffusion (approx. −15 mV for cell-attached, and −100 mV for whole-cell).

Therefore, our molecular modeling identified GluN1 Y647 and GluN2A F636 as two key residues specific for binding KET in the lateral sites (Fig. 2A), and whole cell measurements demonstrated strong and specific reductions in the potency of low-dose KET by substitutions at GluN2A F636 (Fig. 3C). Moreover, MD simulations with the GluN2A F636L substitution showed decreased stability for S-KET⁺ in the lateral sites as illustrated by an increased RMSD value, lower van der Waals interaction energy, and a higher tendency to diffuse into the membrane, relative to WT (Fig. S2 and Tables 2, 3). We tested these predictions with cell-attached patch-clamp recordings and found that substituting GluN2A F636 produced a drastic resistance of NMDA receptor currents to inhibition by bath-applied KET (Fig. 5D). Together these results demonstrate a critical role for GluN2A F636 in the inhibitory potency of ketamine through a membrane pathway.

GluN2A F636 is required for ketamine, but not memantine, membrane access.

In contrast to the results obtained with bath-applied KET, experiments with bath-applied memantine (MEM) showed that the receptor sensitivity to bath-applied MEM is largely independent of the bath pH (Fig. 5E). This result likely reflects the stronger base character of MEM (pKa = 10.3). However, the observation that receptor sensitivity to bath-applied MEM is independent of GluN2A F636 demonstrates that although both KET and MEM can access their inhibitory sites through a membrane route, only KET requires GluN2A F636. This result demonstrates that GluN2A F636 is not required simply for normal gating behavior of NMDA receptors, which would impair similarly sensitivity to both KET and MEM, but rather it plays a key role in the interaction of KET with NMDA receptors that is not shared by MEM. Therefore, the lateral sites we describe here, in which GluN2A F636 provides critical interactions, are specific for KET and may be responsible for the unique clinical effects of low-dose ketamine.

Low-dose ketamine reduces currents from tonically activated NMDA receptors.

Upon visual inspection of single channel current traces, it was apparent that the mechanism by which high concentrations (10 μM) of bath-applied KET (at pH 7.2) reduced the activity of GluN2A F636A receptors (monitored at pH 8) was phenotypically different from the mechanism by which KET (1 μM) produced a similar level of inhibition on WT receptors. Traces recorded from GluN2A F636A receptors had short interruptions within bursts of activity, as expected for open-channel block, whereas traces from WT had longer silent periods between bursts of activity, as expected from a mechanism that increased desensitization. This observation is consistent with the location of the two binding sites, with the pore site mediating a direct blocking effect on current, and the lateral site mediating an allosteric effect on gating.

To investigate the hypothesis that at sub-micromolar concentrations ketamine inhibits NMDA receptor currents by reducing receptor kinetics, rather than by blocking the pore, we obtained unitary currents from cell-attached patches containing a single active receptor (WT), with low concentrations of KET included in the recording pipette (0.025 – 0.25 μM), at physiologic pH (7.2) (Fig. 5F). In this approach, ketamine can access its effector sites on the monitored receptor through the membrane independent of receptor activation (patch potential ~−115 mV), and through the pore, when receptors occupy open states. In these records (n = 6), we observed a substantial concentration-dependent reduction in P_o from 0.3 ± 0.1, in the absence of KET, to 0.07 ± 0.02 in the presence of 0.25 μM KET, with no change in unitary current amplitude (Table S4). The reduction in P_o reflected entirely a lengthening of mean closed durations from 8 ± 3 ms to 38 ± 4 ms, with no change in the mean duration of openings (from 3.4 ± 0.8 ms to 2.9 ± 0.9 ms).

Fitting kinetic models to current traces obtained from individual receptors can identify the principal transitions that make the NMDA receptor gating sequence and can predict how ketamine will affect macroscopic currents produced by NMDA receptors in response to various patterns of stimulation. To this end, we used the traces obtained from individual receptors with low doses of KET to develop kinetic models that describe the pattern of receptor openings and closures in each trace. In the absence of KET, NMDA receptors activate by transitioning from resting (unliganded) states (C₀, C₀₀) into current-passing open states (O), by passing sequentially through three agonist-bound pre-open closed-pore states (C₁-C₃). In addition, this linear activation path is interrupted occasionally by slipping off-path into long-lived closed-pore states (C₄, C₅), which represent desensitized receptors ^{61, 62}. To model KET association and dissociation of KET, we developed a tiered model in which the top tier represents receptors in KET-free states, and the bottom tier represents receptors in KET-bound states (see Methods). Further, given the prediction from our molecular modeling that KET binding is state-dependent, we tested models where receptors transitions between tiers via one of six possible pathways (Fig. 5G; Table S5).

We fitted each of these six models globally to the pooled single-channel data obtained with low doses of KET and optimized rate constants for each model (Table S5). The KET dissociation constants predicted with each of the six models were close to the macroscopic potency we determined in this range; therefore, we could not exclude any of the six models on this basis. However, using a log likelihood criterion, the model in which the two tiers were connected through the closed-pore pre-open state C₂, described the data best, followed by models that connected the tiers through C₃ or C₁. Models that connected tiers through O, C₅, or C₄, returned substantially poorer fits (Table S5). These results support the hypothesis that when present in sub-micromolar concentrations, KET reduces NMDA receptor currents by binding preferentially to, and stabilizing a closed-pore pre-open receptor state, therefore slowing its gating kinetics through an allosteric mechanism rather than simply blocking the pore. The C₂ model predicted for KET an association rate constant of 3.4×10⁷ M⁻¹s⁻¹ and a dissociation rate constant of 1.8 s⁻¹ (calculated K_D,5.3 nM).

Next, we used this best fitting model to envision the effects of ketamine on synaptic and extrasynaptic NMDA receptor currents. We simulated macroscopic currents with brief (1 ms) or sustained (5 s) stimuli (1 mM Glu), in the presence of increasing concentrations of KET (10 nM – 50 nM) as described previously ⁶¹. Because KET had no effect on the amplitude of the unitary current, we set the conductance of the open states in both tiers to be equal (Table S4). Results show that in response to brief pulses of glutamate, NMDA receptor responses remain largely unaffected even at concentrations of KET that are predicted to saturate the allosteric site (50 nM). However, similar levels of KET reduced steady-state currents from NMDA receptors activated with long pulses of glutamate by about 40% (Fig. 5G and Table S5). These results are consistent with the observation that in a brain slice preparation, where KET accumulates in tissue at 2-fold higher concentration than in the surrounding fluid, perfusion with KET up to 2 μM had no effect on NMDA receptor mediated synaptic currents ⁶³, and support the hypothesis that low-dose ketamine modulates excitability of brain tissue at a circuit level by preferentially reducing the activity of tonically activated NMDA extrasynaptic receptors ⁶⁴.

Together the results we report here demonstrate that low dose KET binds to previously unrecognized membrane embedded hydrophobic sites and increases the stability of a pre-open receptor state, which results in increased occupancy of closed-pore states at the expense of open states. This allosteric mechanism is dependent on the type of stimulation experienced by receptors and specifically reduces currents from receptors activated tonically and spares those activated transiently.

DISCUSSION

We report here evidence that ketamine inhibits NMDA receptor-mediated currents with a dual mechanism: allosteric in the low nanomolar concentration range, and open-channel block at micromolar concentrations. The allosteric mechanism went unnoticed in previous studies likely due to insufficient experimental sampling of a system where both ketamine and NMDA receptors exist in dynamic equilibria of multiple molecular species.

Ketamine is a small organic molecule, which has a chiral center, weak base behavior in aqueous solutions, and amphipathic character. It is usually supplied commercially as a racemic mixture of S and R enantiomers, which exist in solution as a dynamic equilibrium of protonated and neutral forms. Each of these four molecular forms distributes in biological membranes in proportions that fluctuate with external pH, membrane composition, and electric potential. Our results show that, in this mixture, only protonated ketamine is effective at reducing NMDA receptor currents. However, the dose-response relationship is complex due to the existence of two separate effector sites, with distinct sensitivities, mechanisms, and accessibility. These may be each responsible for a different region of the dose-specific clinical effects of ketamine.

Tour results are consistent with a large body of literature demonstrating that the anesthetic effects of micromolar ketamine occur due to its ability to bind in the water-filled pore of NMDA receptors an occlude current flow through open NMDA receptors. This open pore-block mechanism is voltage- and use-dependent; it is slightly more sensitive to the S-KET⁺ enantiomer; and it relies heavily on interactions with GluN2A L642, which can be rapidly accessed directly from solution, when receptors are open. In contrast, we demonstrate here that at sub-micromolar concentrations, ketamine interacts with membrane-embedded aromatic residues, such as GluN1 Y647 and GluN2A F636, where it stabilizes receptors in agonist-bound closed-pore states and decrease the channel open probability. Therefore, this allosteric mechanism is also voltage- and use-dependent. However, it is accessible from the membrane through a lipophilic passage, even when the pore is closed, and it may be slightly more sensitive to the R-KET⁺ enantiomer. We propose that this site may be the molecular target for the anti-depressive effects of ketamine.

Consistent with this hypothesis, the hydrophobic site we describe here is insensitive to MEM (1 μM), which like KET is a NMDA receptor pore blocker with dual access to its inhibitory site ⁵⁶, but has no anti-depressive effect. Notably, the mechanism we propose spares phasically activated receptors, as was demonstrated for NMDA receptor-mediated excitatory post-synaptic currents in brain slices exposed to KET concentrations of up to 2 μM ⁶³. In contrast, KET concentrations as low as 0.1 μM reduce substantially the equilibrium gating of NMDA receptors (P_o, from 0.3 to 0.09) in the presence of saturating levels of agonists when applied directly on the cell-attached patch (Table S4).Therefore, depending on cell type (membrane composition, resting membrane potential), the subunit composition and the number of NMDA receptor expressed, and the levels of ambient glutamate and glycine (or D-serine), sub micromolar doses of ketamine can produce a notable reduction intrinsic excitability of selective neurons ⁶⁴, and further on the circuits controlled by these neurons.

Importantly, given that our model proposes that the accessibility of ketamine to its lipophilic inhibitory site is maximal when receptors occupy the pre-open state C₂, any perturbation that changes the fractional occupancy of state C₂ will affect the apparent potency of KET at this site ²⁵. Among such factors are the levels of H⁺, Ca²⁺, Zn²⁺, and the concentration and the duration of exposure to agonist ^{24, 51, 65, 66}. Given that the levels of these endogenous modulators vary naturally with brain region, developmental stage, metabolic state, mental arousal, and so on, the variability in reported values for ketamine potency is not surprising and may explain patient-specific reactions. Lastly, the occupancy of state C₂ is affected by a variety of mutations that have been observed in patients with naturally occurring NMDA receptor variants ⁶⁷. Therefore, these patients may present further variability in their response to ketamine.

In summary, we describe here a new lipophilic site, with specific membrane accessible pathway, and allosteric inhibitory mechanism for low-dose ketamine on NMDA receptors. These results open new directions of research on the roles of NMDA receptors in health and disease and may support the development of more effective interventions to correct neuropsychiatric disorders.

Data and materials availability

Structures and trajectories were deposited in Zenodo https://zenodo.org/records/10711149. Requests for further information, resources, or reagents should be directed to and will be fulfilled by the corresponding author (GKP).