In this study, we examined whether excitatory synapses onto GABAergic interneurons in the human cortex undergo modulation mediated by metabotropic glutamate receptor (mGluR) signaling pathways. Human cortical tissue was obtained from neurosurgical resections performed for the treatment of deep-brain tumors, hydrocephalus, and aneurysms. To investigate synaptic interactions, we performed paired whole-cell patch-clamp recordings from synaptically connected pyramidal cells and interneurons using a biocytin-containing intracellular solution, enabling post hoc morphological identification of a subset of recorded neurons.

Human GABAergic interneurons comprise multiple subtypes defined by distinct morphological, electrophysiological, and transcriptional properties (Chartrand et al., 2023; Lee et al., 2023). To characterize the diversity of postsynaptic interneurons in our dataset, we extracted 55 active and passive electrophysiological parameters from voltage responses to step current injections using custom-written MATLAB scripts. The resulting feature matrix was analyzed using Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction, followed by K-means clustering. This analysis revealed two electrophysiologically distinct groups of interneurons (Figure 1A).

The primary differences between these groups were related to action potential kinetics. We classified them as fast-spiking (FS) and non-fast-spiking (non-FS) interneurons. FS interneurons exhibited significantly shorter membrane time constants (τ = 4.03 ± 2.15 ms vs. 6.65 ± 3.88 ms, p = 0.0205, Mann–Whitney test), narrower action potential half-widths (0.18 ± 0.02 ms vs. 0.43 ± 0.19 ms, p = 3.2 × 10⁻⁶), and faster maximum rates of action potential rise and decay (upstroke: 853.01 ± 163.16 mV/ms vs. 466.63 ± 154.90 mV/ms, p = 1.52 × 10⁻⁵; downstroke: −670.27 ± 196.37 mV/ms vs. -243.87 ± 132.07 mV/ms, p = 8.16 × 10⁻⁵; Mann–Whitney tests for all comparisons; Figure 1B).

FS interneurons also displayed significantly smaller voltage sag responses to hyperpolarizing current injections (sag ratio: 0.161 ± 0.09 vs. 0.337 ± 0.20, p = 0.030) and a more hyperpolarized resting membrane potential compared with non-FS interneurons (−63.12 ± 2.72 mV vs. −59.64 ± 2.43 mV, p = 0.002; Mann–Whitney tests). Although input resistance varied across cells, FS interneurons had significantly lower mean input resistance (76.63 ± 31.37 MΩ vs. 111.60 ± 38.44 MΩ, p = 0.035). When normalized to spike count, FS interneurons showed significantly less interspike interval (ISI) accommodation than non-FS cells (normalized ISI accommodation: 0.088 ± 0.078 vs. 0.172 ± 0.123, p = 0.049). Action potential amplitude adaptation was minimal in both groups, although a subset of non-FS interneurons exhibited mild adaptation (−0.0030 ± 0.005 vs. -0.0099 ± 0.0146, p = 0.524; Figure 1B). Additionally, FS interneurons had significantly lower rheobase values (84.62 ± 43.32 pA vs. 187.27 ± 114.29 pA, p = 0.013) and larger action potential amplitudes (96.51 ± 9.45 mV vs. 79.25 ± 15.00 mV, p = 0.045).

The electrophysiological profile of the FS interneuron group, short membrane time constant, rapid action potential kinetics, and narrow spike width, corresponds closely to previously described fast-spiking interneurons in rodent (Kawaguchi and Kubota, 1997; Ascoli et al., 2008; Lee et al., 2023) and human (Szegedi et al., 2016, 2024; Wilbers et al., 2023). These properties are characteristic of parvalbumin-expressing (PV+) interneurons (Somogyi, 1977; Somogyi et al., 1983; Gulyás et al., 1993) and are most consistent with the small basket cell subtype (Markram et al., 2004).

Each recorded neuron was subjected to post hoc morphological analysis after complete staining and recovery (see Methods), allowing detailed examination of dendritic, axonal, and somatic structures (full anatomical recovery: n = 13 of 24 cells). Among fast-spiking interneurons, seven of 11 cells were successfully reconstructed. Of these, six were identified as basket cells, characterized by densely branched axons surrounding pyramidal cell somata or innervating perisomatic dendrites and multipolar dendritic trees (Tremblay et al., 2016), and one was identified as an axo-axonic (chandelier) cell. The axo-axonic cell displayed characteristic axonal cartridges targeting the axon initial segments of pyramidal neurons (Somogyi, 1977; Szabadics et al., 2006) and exhibited a firing pattern typical of human axo-axonic cells (Szabadics et al., 2006; Molnár et al., 2008), consistent with our previous observations, where suprathreshold depolarizing current steps (0.8 s) evoked a single spike or brief burst at stimulus onset, followed by a 100–300 ms pause and subsequent high-frequency firing (Figure 1A, top left).

To further validate interneuron identity, we performed immunohistochemical labeling for parvalbumin in a subset of recorded cells (Somogyi, 1977; Somogyi et al., 1983; Markram et al., 2004). However, due to the prolonged recording protocol, intracellular components were substantially diluted, rendering immunolabeling unreliable (Pusch and Neher, 1988). Consequently, we could not systematically confirm parvalbumin expression in the recorded neurons. A subset of fully filled cells with anatomic recovery (n = 6) revealed multipolar interneurons with approximately spherical axonal arborizations and thin dendrites, lacking morphological features characteristic of basket and axo-axonic cells; these were classified as non-FS interneurons. Two of these cells displayed neurogliaform morphology.

Group I mGluRs are known to modulate multiple ion channel types, predominantly affecting K⁺ conductances (Chemin, 2003). Such modulation can alter membrane potential and intrinsic excitability, thereby influencing the induction and expression of synaptic plasticity, including spike-timing-dependent plasticity (STDP), LTP, and short-term plasticity (Fernández-Fernández and Lamas, 2021). To determine whether the synaptic effects observed in this study were associated with changes in intrinsic membrane properties, we analyzed holding current and input resistance before and after application of the group I mGluR agonist DHPG.

Group I metabotropic glutamate receptors, mGluR1 and mGluR5, are widely expressed throughout the brain, but exhibit pronounced regional and cell-type–specific expression patterns (Ferraguti and Shigemoto, 2006; Jian et al., 2010). GRM1, which encodes mGluR1, is most highly expressed in the cerebellum and olfactory bulb, whereas GRM5 expression is more prominent in the neocortex and hippocampus (Romano et al., 1995). Given the heterogeneity of synaptic responses observed in our electrophysiological experiments following group I mGluR activation, we investigated whether these functional differences might arise from interneuron subtype–specific expression profiles.

To this end, we analyzed publicly available single-cell transcriptomic data from the Allen Brain Institute (Tasic et al., 2018; Hodge et al., 2019; Bakken et al., 2021), focusing on layer 2/3 neurons from the human middle temporal gyrus. We compared gene expression in parvalbumin-positive (PV+), vasoactive intestinal peptide–positive (VIP+), and somatostatin-positive (SST+) interneuron populations. PV+ interneurons are typically fast-spiking, whereas VIP+ and SST+ populations include predominantly non-fast-spiking, irregular-spiking, or adapting interneurons (Markram et al., 2004; Tremblay et al., 2016).

RNA-seq analysis revealed marked differences in GRM1 and GRM5 expression across interneuron subclasses. Among the three major interneuron groups, PV+ (n = 289) cells were distinguished by high GRM5 expression with minimal or undetectable GRM1 expression. In contrast, SST+ interneurons (n = 501) expressed both GRM1 and GRM5 at comparable levels, while VIP+ neurons (n = 655) exhibited uniformly low expression of both transcripts (Figure 7A). These transcriptomic patterns are consistent with previous reports implicating mGluR5 in synaptic potentiation (Hagena and Manahan-Vaughan, 2022), and align with our electrophysiological finding that a substantial fraction of fast-spiking interneurons in both humans and rodents exhibit synaptic enhancement following group I mGluR activation. Together, these findings support the conclusion that interneuron subtype–specific expression of group I mGluRs contributes, at least in part, to the differential synaptic modulation observed across interneuron classes.

Because comparable single-cell transcriptomic data are not available for our selected rodent model, we next compared the human expression profiles with corresponding datasets from mouse cortex. This analysis revealed a less conserved pattern of group I mGluR expression across species. In mouse interneurons, VIP (n = 1,275) and SST (n = 187) populations expressed higher levels of GRM1 and GRM5 than observed in their human counterparts, whereas PV interneurons (n = 464) exhibited an even more pronounced GRM5 and GRM1 ratio (Figure 7B). These interspecies differences suggest that group I mGluR–mediated synaptic modulation may be supported by distinct transcriptional architectures in rodents and humans.

All procedures were performed according to the Declaration of Helsinki with the approval of the University of Szeged Ethical Committee and Regional Human Investigation Review Board (ref. 75/2014). For all human tissue material written consent was given by the patients prior to surgery. Written informed consent was obtained from all patients before surgery. For underage patients, consent was provided by a parent or legal guardian.

Non-pathological human cortical tissue was obtained from surgically resected brain areas that were removed to treat deep-brain tumors, epilepsy, and hydrocephalus in male and female patients (ages 4–74 years) from the frontal, temporal, and parietal lobes (Supplementary Table 1). The observed mGluR-mediated modulation was not associated with age, sex, cortical region, or clinical indication (Supplementary Figure 1).

Anesthesia was induced with intravenous midazolam and fentanyl (0.03 mg/kg, 1–2 μg/kg, respectively). A bolus dose of propofol (1–2 mg/kg) was administered intravenously. The patients received 0.5 mg/kg rocuronium to facilitate endotracheal intubation. The trachea was intubated, and the patient was ventilated with O₂/N₂O mixture at a ratio of 1:2. Anesthesia was maintained with sevoflurane at care volume of 1.2–1.5.

Following surgical removal, the resected tissue blocks were immediately immersed in ice-cold slicing solution and transported from the operating room to the electrophysiology laboratory within 20 min. The solution contained (in mM): 75 sucrose, 84 NaCl, 2.5 KCl, 1 NaH₂PO₄, 25 NaHCO₃, 0.5 CaCl₂, 4 MgSO₄, 25 D(+)-glucose, saturated with 95% O₂ and 5% CO₂. The solution was maintained on ice in a thermally insulated transport container with continuous oxygenation (95% O₂ and 5% CO₂).

Experiments were performed on Wistar rats (postnatal day 18–30). Animals had ad libitum access to standard chow and water. Animals were anesthetized by inhalation of halothane prior to decapitation. Neocortical tissue from the somatosensory cortex was rapidly removed and placed in slicing solution.

From this point slice preparation followed the same protocol for both species. Slices of 320 μm thickness were prepared with a vibrating blade microtome (Microm HM 650 V). The slices were incubated at 36 °C for 1 h, while the slicing solution was gradually replaced by a peristaltic pump (6 mL/min) with a storing solution that contained (in mM) 130 NaCl, 3.5 KCl, 1 NaH₂PO₄, 24 NaHCO₃, 1 CaCl₂, 3 MgSO₄, 10 D(+)-glucose, and was saturated with 95% O₂ and 5% CO₂.

Whole-cell patch-clamp recordings were performed in a submerged recording chamber at 36 °C. Neurons were visualized with infrared differential interference contrast (IR-DIC) microscopy and were patched at depths of 60–130 μm from the slice surface. The recording solution was identical to the storage solution, except that CaCl₂ was increased to 3 mM and MgSO₄was reduced to 1.5 mM. To avoid EPSCs rundown during longer recordings (Biró et al., 2005; Molnár et al., 2016) micropipettes (5–7 MΩ) were filled with a low [Cl⁻]_i solution intracellular solution containing (in mM): 126 potassium-gluconate, 4 KCl, 4 ATP-Mg, 0.3 GaTP-Na₂, 10 HEPES, 10 phosphocreatine, 10 L-glutamate and 8 biocytin (pH 7.20; 300 mOsm).

Signals acquired with Patchmaster software were low-pass filtered at 8 kHz and digitized at 16 kHz. Presynaptic cells were stimulated using a paired-pulse protocol (PPP) consisting of brief (2–10 ms), paired suprathreshold current injections (800 pA) separated by 60 ms and delivered every 10 s. Baseline recordings were restricted to the first 5 min after the membrane rupture of the interneuron, before DHPG application. Postsynaptic cells were recorded in voltage-clamp mode; holding currents were adjusted slightly when necessary to maintain a stable holding potential. To visualize mean changes across cell groups, holding current values were baseline-subtracted for each cell and then averaged in 10 s time bins. Access resistance was monitored throughout the electrophysiological recordings (25.81 ± 9.38 MΩ), and cells showing a > 25% increase in access resistance were excluded from analysis.

The following pharmacological agents were used: 50 μM (S)-3,5-Dihydroxyphenylglycine (DHPG, Tocris), 25 μM 2-Methyl-6-(phenylethynyl)pyridine hydrochloride (MPEP, Tocris), 10 μM (S)-(+)-α-Amino-4-carboxy-2-methylbenzeneacetic acid (LY-367385, Tocris).

Electrophysiological sub- and suprathreshold features of recorded neurons were measured from voltage responses elicited by 800 ms current steps increasing by 20 pA from −100 pA. Custom-written MatLab (Mathworks) scripts were used to analyze the recorded electrophysiological signals. Resting membrane potential (RMP) was measured immediately after establishing whole-cell configuration with no injected current. The membrane time constant (tau) and input resistance (Rin) were calculated as the averages across all hyperpolarizing current steps, while the sag ratio was measured at the −100 pA step. Rheobase current was defined as the minimal current injection that elicited the first action potential. Action potential (AP) half-width was measured as the time between the rising and decaying phases at half-maximal spike amplitude. Maximum AP upstroke velocity was calculated as the average voltage at the maximum of AP dV/dt over all APs, and minimum AP downstroke velocity as the average voltage at the minimum (absolute maximum) of AP dV/dt over all APs. Normalized interspike interval (ISI) accommodation was computed as the average interspike interval from all sweeps containing at least three APs divided by the number of APs. AP amplitude adaptation was defined as the average change in AP amplitude between consecutive APs.

To identify interneuron subtypes, we generated a 55-parameter feature matrix from the current-step response traces including subthreshold properties, action potential kinetics, and firing pattern adaptations. To account for different units and scales, all features were normalized using a range-scaling method ([0, 1]). Dimensionality reduction was performed using Uniform Manifold Approximation and Projection (UMAP). Based on the local density of our data, the UMAP hyperparameters were set to 4 neighbors and a minimum distance of 0.2, using a Euclidean distance metric. Classification was then achieved by applying K-means clustering (k = 2) directly to the resulting 2D UMAP coordinates. The cluster number was selected a priori to distinguish between the two primary physiological classes (Fast-Spiking and Non-Fast-Spiking), and the validity of this partition was subsequently confirmed by silhouette analysis and post hoc morphological and physiological verification.

Synaptic responses were analyzed using Fitmaster (HEKA), IgorPro (Wavemetrics) and Origin 7.5 (OriginLab). Synaptic strength and holding current analyses were performed on partially overlapping interneuron samples: EPSC-based synaptic analyses were restricted to recordings with a synaptic connection, whereas holding current changes were quantified in voltage-clamped interneurons regardless of whether a synaptic connection was detected. Evoked EPSC amplitudes were quantified as the difference between the holding current immediately preceding presynaptic stimulation and the peak of the evoked response recorded on the postsynaptic neuron. The significance of DHPG-induced changes in EPSC amplitudes was assessed by comparing ~3 min of baseline recordings with ~3 min of recordings obtained during DHPG application, starting 1 min after wash-in, using the Mann–Whitney test. Synaptic failures (EPSC = 0) were excluded from the analysis. The paired-pulse ratio (PPR) was defined as the second EPSC divided by the first EPSC amplitude evoked by the paired presynaptic action potentials. The latency was calculated as the time interval between the peak amplitude of presynaptic AP and the onset of the EPSP (detected by IgorPro).

All values are reported as mean ± standard deviation (s.d.), unless otherwise specified. Data normality was assessed using the Shapiro–Wilk test, and statistical comparisons were performed using parametric or non-parametric tests, as appropriate and defined for each paradigm. Differences with p < 0.05 were considered significant.

Expression of GRM1 and GRM5 was quantified using the publicly available single-cell RNA-seq datasets from the Allen Institute for Brain Science Cell Types Database (celltypes.brain-map.org). For human data, we extracted middle temporal gyrus (MTG), layer 2/3 VIP+, PV+ and SST+ interneuron transcriptomes. To make the same comparisons in the mouse, we selected layer 2/3 VIP+, PV+ and SST+ interneurons from the primary visual cortex (VISp) and the anterior lateral motor cortex (ALM).

Following electrophysiological recordings, slices were immersed in a fixative containing 4% paraformaldehyde, 15% (v/v) saturated picric acid, and 1.25% glutaraldehyde in 0.1 M phosphate buffer (PB; pH = 7.4) at 4 °C for at least 12 h. After several washes with 0.1 M PB, slices were frozen in liquid nitrogen and then thawed in 0.1 M PB, embedded in 10% gelatin, and further sectioned into 60 μm slices. Sections were incubated in a solution of conjugated avidin-biotin-horseradish peroxidase complex (ABC; 1:100; Vector Labs) in Tris-buffered saline (TBS, pH = 7.4) at 4 °C overnight. The enzyme reaction was revealed by the glucose oxidase-DAB-nickel method using 3′3-diaminobenzidine tetra-hydrochloride (0.05%) as chromogen and 0.01% H₂O₂ as oxidant. Sections were postfixed with 1% OsO₄ in 0.1 M PB. After several washes in distilled water, sections were stained in 1% uranyl acetate and dehydrated in an ascending series of ethanol. Sections were infiltrated with epoxy resin (Durcupan) overnight and embedded on glass slides. Perisomatic area of DAB-visualized cells was studied under light microscope for axonal and dendritic morphology. Three-dimensional light-microscopic reconstructions were carried out using a Neurolucida system (MicroBrightField) with a 100 × objective. Reconstructed neurons were quantitatively analyzed with NeuroExplorer software (MicroBrightField).