Blocking serotonin transport and NMDA receptors together may cause ketamine’s fast antidepressant effects

We first used liquid chromatography-tandem mass spectrometry (LC-MS/MS) to detect the concentrations of ketamine in the PFC at different timepoints after intraperitoneal injection (Fig. 1A). mPFC concentrations of ketamine gradually dropped over 30 min after injection (Fig. 1B). To examine whether ketamine increases extracellular levels of 5-HT in freely behaving mice [49], we performed real-time in vivo microdialysis to measure the mPFC 5-HT concentration (Fig. 1C). We noted a rapid increase in the average relative-to-baseline 5-HT levels from ∼100% to ∼180% in the mPFC during the 15 min after ketamine injection. Still, memantine did not show such an effect (Fig. 1D). Notably, ketamine did not activate 5-HT neurons in the dorsal raphe nucleus (DRN) (Fig. S1A–F), which indicates that the increase in mPFC 5-HT levels is due to ketamine inhibition of SERT, rather than the activation of DRN 5-HT neurons or inhibition of NMDAR. To further assess ketamine inhibition of SERT in the mouse brain in vivo, we performed PET-computed tomography (PET-CT) imaging with N,N-dimethyl-2-(2-amino-4-[¹⁸F] fluorophenylthio) benzylamine (4-[¹⁸F]-ADAM), a highly selective radioligand for SERT imaging [54]. A dynamic, 90 min PET-CT scan was started after injection of 4-[¹⁸F]-ADAM (Fig. 1E). PET-CT imaging indicated that ketamine was able to inhibit 4-[¹⁸F]-ADAM binding to cortical SERT (Fig. 1F). The standard uptake value ratio data showed that ketamine significantly reduces the uptake level of 4-[¹⁸F]-ADAM in the mPFC (Fig. 1G). Furthermore, the SERT non-displaceable binding potential (BPnd) value of the ketamine group had an average 53.41% decrease compared to the control group (Fig. 1H), indicating that ketamine enhanced serotonergic transmission by inhibition of SERT activity.

To explore the function of SERT, we carried out saturation experiments using high-affinity [³H]-imipramine, for which we determined a dissociation constant (K_d) of 2.42 nM (Table S1). To characterize ketamine binding to SERT, we carried out competition experiments (Fig. 1I), with imipramine as the positive control in combination with [³H]-imipramine and measured a half-maximal inhibitory concentration (IC₅₀) of 4.49 μM for ketamine and 3.17 nM for imipramine (Fig. 1J), and an inhibition constant (K_i) of 2.45 μM for ketamine and 1.73 nM for imipramine (Table S1). Thus, both ketamine and imipramine compete for [³H]-imipramine binding.

To further assess whether ketamine affects the reuptake of 5-HT through the SERT, we performed a homogeneous assay for 5-HT transporter activity based on cellular uptake of a fluorescent dye coupled with a proprietary masking dye (Fig. 1K). We determined the Michaelis–Menten constant (K_m) for the fluorescent substrate mimetic for SERT to be 0.96 μM (Table S1), with citalopram as a positive control. Ketamine and citalopram significantly prevented the reuptake of 5-HT at an IC₅₀ of 20.63 μM (K_i = 6.69 μM) and 18.21 nM (K_i = 5.9 nM), respectively (Fig. 1L and Table S1). Both the in vivo imaging study and in vitro data indicate that ketamine directly inhibits SERT to increase cortical 5-HT levels.

SERT features a central site and an allosteric site for its substrate 5-HT binding, allowing antidepressants (e.g. paroxetine and citalopram) and psychostimulants (e.g. methamphetamine and cocaine) to occupy one or both of these sites and modulate its transport activity [55,56]. Previous crystal and cryo-EM structures of SERT have revealed that paroxetine, methamphetamine and cocaine bind to the central site, whereas 5-HT and citalopram bind to both sites [55,57–59]. We thus investigated whether ketamine interacts physically with SERT and, if so, which site it targets. We recombinantly expressed human full-length and wild-type SERT (hSERT) with a C-terminal 3C protease site, an enhanced green fluorescent protein (eGFP), and a Strep-tag II in HEK293F cells. Fluorescence-based size-exclusion chromatography (FSEC) analysis of hSERT solubilized in various detergents showed comparable biochemical behavior and molecular weight (Fig. S2A) compared to previously reported data for recombinant human SERT or the native porcine SERT, indicating that the hSERT used in this study is in a homogeneous and monomeric form. We then purified hSERT in the presence of ketamine in detergent micelles and used single-particle cryo-EM to elucidate the structure of the hSERT–ketamine complex (hSERT^ketamine) at an overall resolution of 3.2 Å (Fig. 2A and B, Fig. S2 and Table S2). A well-resolved 'Y'-shaped ketamine density in the central site was observed, allowing us to determine the pose and orientation of ketamine (Fig. 2D and E). Moreover, we also observed a detergent molecule of n-dodecyl-β-d-maltopyranoside bound to the allosteric site and cholesterols/lipids surrounding the transmembrane region (Fig. 2B), consistent with previous structural and computational studies [58,60].

The overall structure of hSERT^ketamine is similar to that of the 5-HT-bound recombinant human SERT and the (+)-methamphetamine-bound or cocaine-bound native porcine SERT, with a root-mean-square deviation (RMSD) value of 0.59, 0.64 and 0.73 Å, respectively (measured from Cα to Cα), indicating that hSERT^ketamine adopts an outward-open conformation (Fig. S3A–C). Ketamine lodges between the aromatic groups of Y95, Y176 and F341 (Fig. 2F). The hydrophobic chlorobenzene and cyclohexane groups of ketamine face the lower part of the central site (closer to the cytoplasm), where a hydrophobic barrier is formed by I172, F341, Y95 and V343 (Fig. 2F). The hydrophilic amine group, aryl chloride, and oxygen of ketamine, in contrast, face the upper part of the central site formed by hydrophilic residues D98, S438 and the main chain oxygen of F335 (Fig. 2G). The amine group of ketamine interacts with the carboxylic group of D98, a conserved residue in SERT, at a distance of 4.3 Å (Fig. 2F). Compared with previous structural and functional studies, ketamine occupies a similar but slightly larger volume of the central site than (+)-methamphetamine. In contrast, the larger drugs paroxetine and cocaine almost fill the entire volume (Fig. S3), which may explain, in part, the differences in their binding affinities [58]. Nevertheless, our structural and functional studies of hSERT^ketamine show direct binding of ketamine to the central site of SERT.

As ketamine is a non-competitive NMDAR antagonist and several NMDAR antagonists fail to produce rapid antidepressant effects similar to ketamine [17,26,30,34], this suggests that solely inhibiting NMDAR is insufficient to explain the rapid antidepressant effects of ketamine. We hypothesized that the rapid antidepressant action of ketamine may be achieved by simultaneously inhibiting SERT and NMDAR (Fig. 3A). To validate this hypothesis, we used a mouse model of depression that relies on chronic restraint stress [61]. We then intraperitoneally injected the NMDAR antagonist memantine in combination with SERT antagonists fluoxetine or imipramine into mice (Fig. 3B). At 1 h after drug injection, the combined antagonists (memantine in combination with fluoxetine or imipramine) increased the sucrose preference of mice with depression-like symptoms during the sucrose preference test (SPT) (Fig. 3C). During the tail suspension test (TST), the combined antagonists reduced the immobility time of mice with depression-like symptoms (Fig. 3D). Injecting either of the SERT antagonists alone did not, however, produce these effects. These effects were maintained at 24 h after injection of the drugs (Fig. S4A). Furthermore, during the open field test (OFT), the combined use of the NMDAR antagonist and SERT antagonists did not alter the locomotor activity of mice (Fig. 3E). These antagonists alone did not induce anxiety-like behavior in mice, and memantine combined with imipramine increased the activity time of mice in the central zone of the open field (Fig. 3F), indicating a reduction in anxiety-like behavior.

We conducted additional experiments utilizing other NMDAR antagonists, such as MK-801, which failed to replicate ketamine's effects 24 h post-injection in antidepressant-relevant behavioral mouse models [32,62]. These experiments involved administering varying doses of MK-801, both alone and in combination with fluoxetine. Our findings reveal that a single injection of either MK801 or fluoxetine alone does not result in antidepressant effects after 24 h (Fig. S4B and C). In contrast, both 0.03 and 0.1 mg/kg of MK-801 combined with fluoxetine results in lasting antidepressant effects, with 0.03 mg/kg being more effective than 0.1 mg/kg (Fig. S4B and C). However, a combination of 0.01 mg/kg of MK-801 with fluoxetine does not yield a lasting effect. Furthermore, these results showed no significant impact on mouse locomotion and anxiety-like behaviors (Fig. S4D and E).

To further validate whether inhibition of NMDAR in the mPFC combined with SERT inhibition is sufficient to induce rapid antidepressant effects, we again used chronic restraint stress. We injected the NMDAR antagonist AP5 into the mPFC in combination with an intraperitoneal injection of the SERT antagonist fluoxetine in mice (Fig. 3G and H). At 1 h after drug injection, the combined use of both antagonists increased the sucrose preference of mice with depression-like symptoms during the SPT (Fig. 3I). The combined use of both antagonists reduced the immobility time of mice with depression-like symptoms during the TST (Fig. 3J). The combination of AP5 and fluoxetine did not alter locomotor activity or induce anxiety-like behavior in mice during the OFT (Fig. 3K and L). These behavioral results demonstrate that synergic inhibition of SERT and NMDAR induces rapid antidepressant effects similar to those of ketamine.

VIP, parvalbumin (PV) and somatostatin (SST) neurons are the three main types of gamma-aminobutyric acid (GABA)ergic interneurons in the mPFC [63]. Inhibition of PV and SST neurons is involved in the rapid antidepressant effects of ketamine [51–53], but the role of VIP neurons in the effects of ketamine has not been studied. To investigate the effects of ketamine on these interneuron types in the mPFC, we labeled VIP, PV and SST neurons by crossing Ai47 (Cre-dependent GFP reporter) mice with VIP-Cre, PV-Cre and SST-Cre mice [64]. Immunolabeling for c-Fos in mice expressing GFP in specific neuronal populations showed that ketamine specifically activated VIP neurons at the antidepressant dose (38.8% of neurons activated on average) in the mPFC (Fig. 4A–C), but PV and SST neurons were not activated (Fig. S5A–F).

To investigate whether ketamine changes the electrophysiological characteristics of VIP neurons in the mPFC, we then carried out patch-clamp recordings of the identified mPFC VIP neurons in mice 1 h after injection of ketamine (Fig. 4D). The percentage of firing VIP neurons increased significantly from 11.5% (3 of 26 cells firing) in controls to 75% (21 of 28 cells firing; 1 h after injection) in the ketamine group (Fig. 4E and F). In addition, the resting membrane potential (RMP) of VIP neurons was, on average, more depolarized (−44.2 mV) after ketamine injection compared with that of saline controls (−58.8 mV) (Fig. 4G). These neurons also exhibited a remarkable increase in the number of spikes in response to depolarizing current pulses (Fig. 4H and I). Interestingly, these firing neurons showed an increase in neuronal membrane resistance with no change in membrane capacitance (Fig. S6A and B). In contrast, neither PV nor SST neurons showed any changes in electrophysiological characteristics of excitability (Fig. S6C–L). Thus, ketamine specifically elevates the excitability of mPFC VIP neurons.

Traditional antidepressants that target the serotonergic system to increase brain levels of 5-HT have long been used to treat psychiatric disorders [4,5]. To examine whether 5-HT is necessary for the rapid antidepressant effects of ketamine, we intraperitoneally injected the 5-HT hydroxylase inhibitor 4-chloro-dl-phenylalanine (p-CPA) three times, with one injection per day, to decrease the levels of 5-HT (Fig. S7A). p-CPA attenuated the ketamine-related decrease in the duration of immobility during the TST and forced swimming test (FST) (Fig. S7B, C) but did not affect locomotion or cause anxiety-like behaviors during the OFT at 1 h after injection (Fig. S7D and E). Meanwhile, activation of mPFC VIP neurons by ketamine was blocked after the p-CPA injection (Fig. 4E and F) and the RMP of VIP neurons was, on average, more hyperpolarized (−69.7 mV) after p-CPA relative to saline controls (−58.8 mV) (Fig. 4G). The number of spikes in response to depolarizing current pulses also decreased after the p-CPA injection relative to slices treated with ketamine alone (Fig. 4H and I). These results indicate that ketamine specifically activates mPFC VIP neurons and that this activation requires 5-HT.

To assess whether the elevation of 5-HT alone could activate mPFC VIP neurons, we intraperitoneally injected the SERT inhibitor fluoxetine into the VIP-Cre mice to increase the 5-HT level. Fluoxetine alone did not increase the excitability of mPFC VIP neurons relative to saline controls (Fig. 4J–L). Consistently, the application of 5-HT (5 μM) to brain slices also did not activate VIP neurons in normal mice (Fig. 4M). To further assess whether inhibition of NMDAR directly activates mPFC VIP neurons, we carried out an identical set of electrophysiological experiments and made whole-cell recordings of mPFC VIP neurons in brain slices from normal mice after application of the NMDAR inhibitor AP5 (100 μM). Notably, the direct application of AP5 did not activate VIP neurons (Fig. 4N). Thus, inhibition of NMDARs alone is not sufficient to activate VIP neurons.

To further assess the roles of SERT and NMDAR in the activation of mPFC VIP neurons, AP5 (40 nmol) was injected into the mPFC in conjunction with an injection of fluoxetine. This increased the firing of some VIP neurons (58.6%, Fig. 4J and K), and the RMPs of these VIP neurons were, on average, more depolarized (−49.8 mV) than those of neurons after treatment with AP5 and saline (–65.1 mV) (Fig. 4L). Importantly, bath application of 5-HT (5 μM) and AP5 (100 μM) together induced the firing of some VIP neurons (Fig. 4O) and significantly increased the RMP of these VIP neurons (Fig. 4P). These results indicate that activation of VIP neurons requires the synergistic effect of elevated 5-HT and NMDAR inhibition together.

To investigate whether there is a specific connection between mPFC VIP neurons and 5-HT neurons, we applied a rabies virus (RV)-mediated trans-synaptic retrograde tracing strategy to map the afferent inputs of the mPFC VIP neurons (). After 3 weeks, the mice were perfused, and their brains were processed for subsequent data analysis (). Initially infected cells (expressing both GFP and DsRed) and retrogradely labeled RV neurons (expressing DsRed alone) were identified (). RV-tracing results showed that DRN 5-HT neurons project onto VIP neurons (). These results identify that serotonergic signaling is involved in the ketamine-induced increase in the excitability of mPFC VIP neurons. Fig. S8A Fig. S8B Fig. S8C Fig. S8D

The data above indicate that serotonergic signaling is important for the activation of mPFC VIP neurons. However, the pathways through which VIP neurons mediate the effects of 5-HT signaling are not clear. Morphological and electrophysiological results have indicated that VIP neurons are the 5-HT3A receptor (5-HT3AR) interneurons in the neocortex [63], and ketamine can potentiate 5-HT3AR-mediated currents in ganglion neurons [65]. We thus determined whether the 5-HT3AR antagonist palonosetron, when injected into the mPFC, can prevent the enhanced excitability of mPFC VIP neurons induced by ketamine. To do so, we performed whole-cell recordings of mCherry-labeled VIP neurons in the mPFC (Fig. S9A). The percentage of firing VIP neurons was ∼10%, which was almost equal to the controls (∼10.5%) (Fig. S9B). Consistently, blocking 5-HT3AR also blocked the effect of 5-HT combined with AP5 in activating VIP neurons (Fig. 4Q and R). Furthermore, there was no change in the RMP (Fig. S9C). There was no change in membrane resistance (Fig. S9D) and a moderate increase in membrane capacitance (Fig. S9E) among VIP neurons. Next, we tested whether the rapid antidepressant effects of ketamine on mPFC VIP neuron activation require 5-HT3AR. We injected palonosetron or saline bilaterally into the mPFC 0.5 h before the injection of ketamine (Fig. S9A and F). Mice that did not receive palonosetron had a shorter duration of immobility during the TST and FST at 1 h and 24 h after injection of ketamine compared with control mice, but this effect was blocked by palonosetron (Fig. S9G–J). In addition, injection of palonosetron into the mPFC did not affect locomotion or cause anxiety-like behaviors during the OFT at 1 h after injection (Fig. S9K and L). To further verify whether directly interfering with 5-HT3AR expression in VIP neurons can block the effects of ketamine, we used RNA-specific knockdown of 5-HT3AR expression in the mPFC VIP neurons of VIP-Cre mice (Fig. S10A). Consistent with the results of the blocker, there was no difference in the immobility time in the TST of ketamine-treated mice compared to the control group (Fig. S10C), and the SPT did not show a significant increase (Fig. S10B). Moreover, the mice's motor abilities and anxiety-like behaviors were not affected (Fig. S10D and E).

Therefore, ketamine directly inhibits SERT to increase the levels of 5-HT in combination with inhibition of the NMDAR to activate mPFC VIP neurons. Furthermore, the blockade of 5-HT3AR prevents the ketamine-induced increase in the excitability of VIP neurons in the mPFC, which ultimately blocks the antidepressant effects of ketamine.

To determine whether VIP neurons in the mPFC are necessary for the rapid antidepressant effects of ketamine, we performed chemogenetic inhibition and genetic ablation experiments. In the chemogenetic inhibition assay, we injected the AAV-DIO-hM4Di-mCherry vector into the bilateral mPFC of VIP-Cre mice. Subsequent expression of the hM4Di receptor allowed inhibition of VIP neurons (Fig. 5A and B, Fig. S11A–F). Specifically, the application of clozapine N-oxide (CNO, 10 μM), a specific ligand of the hM4Di receptor, significantly inhibited mPFC VIP neuron activity in brain slices (Fig. 5C). Behavioral experiments were performed after mice had been allowed to recover for 3 weeks after viral injection (Fig. 5A). Ketamine injected 1 h after the initial saline treatment decreased the time that the mice were immobile during the TST and FST (Fig. 5D and E). However, chemogenetic inhibition of mPFC VIP neurons by CNO injection increased the duration of immobility of the mice during the TST and FST (Fig. 5H and I). The effects of ketamine were blocked after inhibition of mPFC VIP neurons (Fig. 5H and I). These findings are consistent with results at 24 h after injection of ketamine (Fig. S12A–D). Furthermore, chemogenetic inhibition of mPFC VIP neurons did not affect locomotion or anxiety-related behaviors during the OFT (Fig. 5F, G, J and K).

To further examine whether ablation of mPFC VIP neurons would alter the effects of ketamine, we injected the vector AAV-DIO-taCasp3-TEVp or the control vector AAV-DIO-GFP bilaterally into the mPFC of VIP-Cre mice. Behavioral experiments were performed after mice had been allowed to recover for 4 weeks after viral injection. Ablation of mPFC VIP neurons decreased sucrose preference during the SPT and increased the duration of immobility during the TST and FST (). Similarly, the ablation of mPFC VIP neurons blocked the antidepressant effect of ketamine (). Ablation of mPFC VIP neurons did not change the locomotor activity or anxiety-like behaviors of these mice during the OFT (), which indicated that the change in immobility was not likely due to motor defects. Meanwhile, no obvious changes were observed in measurements of body weight or water consumption () or in a social interaction test (). Therefore, the activity of mPFC VIP neurons signals mood states and is important in mediating the rapid antidepressant effects of ketamine. Fig. S12E–G Fig. S12E–G Fig. S12H, I Fig. S12J, K Fig. S12L, M

We show above that the combined use of SERT inhibitors (fluoxetine or imipramine) and NMDAR inhibitors (memantine or AP5) can produce rapid antidepressant effects similar to those of ketamine. To validate whether inhibiting VIP neurons can also block these effects, we performed chemogenetic inhibition experiments (Fig. 5M). The combined drug administration reduced the baseline immobility time in control mice. However, inhibiting mPFC VIP neurons substantially blocked the alterations in immobility time caused by the combined drug administration during the TST and FST (Fig. 5M and N). These findings are consistent with results at 24 h after the combined drug administration (Fig. S4B). Similarly, chemogenetic inhibition of mPFC VIP neurons did not affect locomotion or anxiety-related behaviors during the OFT (Fig. 5O and P). These results indicate that activation of mPFC VIP neurons by dual inhibition of both NMDAR and SERT is key for the rapid antidepressant effects of ketamine.

The mice were group-housed (four to five mice per cage) under a 12 h light/dark cycle (light on from 7:00 A.M. to 7:00 P.M.), with ad libitum free access to water and food at 22°C–25°C with 40%–60% relative humidity. VIP-Cre (JAX), PV-Cre (JAX), SST-Cre (JAX), ePet-Cre (JAX), Ai47 (JAX) and C57BL/6 mice (Shanghai Jihui Laboratory Animal Breeding Co.) were used. Transgenic mice were bred in Shanghai Model Organisms Center, Inc. VIP-Cre::Ai47 mice were derived from crosses of the VIP-Cre and Ai47 genotypes. PV-Cre::Ai47 mice were derived from crosses of the PV-Cre and Ai47 genotypes. SST-Cre::Ai47 mice were derived from crosses of the SST-Cre and Ai47 genotypes. Male adult (8–16 weeks of age) mice were used for the experiment and were randomly selected. All animals were habituated for 1 week before the experiments. All experimental procedures were approved by the Institutional Animal Care and Use Committee at Fourth Military Medical University and performed according to National Institutes of Health guidelines.

Adult (10–16 weeks old) VIP-Cre-expressing mCherry mice with or without a cannula were firstly anesthetized by being injected intraperitoneally with tribromoethanol (250 mg/kg) and then perfused transcardially with 20 mL of ice-cold N-methyl-d-glucamine (NMDG) artificial cerebrospinal fluid (ACSF) solution (oxygenated with 5% CO₂ + 95% O₂) containing (mM): 93 NMDG, 93 HCl, 2.5 KCl, 10 MgSO₄·7H₂O, 1.2 NaH₂PO₄·2H₂O, 30 NaHCO₃, 25 glucose, 20 HEPES, 5 sodium ascorbate, 3 sodium pyruvate and 2 thioureas, with 0.5 CaCl₂·2H₂O added (pH: 7.35, 303 mOsm). The brain was removed quickly after decapitation and immediately put into ice-cold oxygenated NMDG ACSF. A few minutes later, coronal slices containing the mPFC (300 μm thickness for mice) were sectioned in cold ACSF using a vibratome (VT1200 S, Leica) and then transferred into oxygenated NMDG ACSF at 32°C for incubation and recovery for 10–15 min. The brain slices were transferred to a normal oxygenated ACSF solution (126 mM NaCl, 2.5 mM KCl, 2 mM MgSO₄·7H₂O, 1.25 mM NaH₂PO₄·2H₂O, 25 mM NaHCO₃, 10 mM glucose, 2 mM CaCl₂) at room temperature for 1 h. All chemicals were purchased from Sigma–Aldrich (St. Louis, MO, USA).

The images of brain sections were processed by ImageJ (NIH, USA) software and QuPath (University of Edinburgh, Scotland). The statistical analyses were conducted using GraphPad Prism software v10.0 (GraphPad, USA) and MATLAB R2023b (MathWorks, USA) software. All statistical tests were two-tailed, and significance was considered at *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. When sample groups had normality and equal variance, data were analyzed by paired or unpaired t-test and one-way or two-way analysis of variance (ANOVA) (followed by Bonferroni's multiple comparisons test). The difference in the proportion of firing was analyzed using the chi-squared test. The experimental data retain two significant figures, expressed as mean ± standard error of the mean (mean ± SEM). The details of sample sizes, statistical tests used, P value and treatment effect of our statistical analyses for each experiment are described in Table S3.