Functional Autapses Form in Striatal Parvalbumin Interneurons but not Medium Spiny Projection Neurons

Abstract

Autapses selectively form in specific cell types in many brain regions. Previous studies have also found putative autapses in principal spiny projection neurons (SPNs) in the striatum. However, it remains unclear whether these neurons indeed form physiologically functional autapses. We applied whole-cell recording in striatal slices and identified autaptic cells by the occurrence of prolonged asynchronous release (AR) of neurotransmitters after bursts of high-frequency action potentials (APs). Surprisingly, we found no autaptic AR in SPNs, even in the presence of Sr²⁺. However, robust autaptic AR was recorded in parvalbumin (PV)-expressing neurons. The autaptic responses were mediated by GABA_A receptors and their strength was dependent on AP frequency and number. Further computer simulations suggest that autapses regulate spiking activity in PV cells by providing self-inhibition and thus shape network oscillations. Together, our results indicate that PV neurons, but not SPNs, form functional autapses, which may play important roles in striatal functions.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12264-022-00991-x.

Introduction

The striatum is the largest nucleus in the basal ganglia, receiving synaptic inputs from different cortical areas, thalamic nuclei, and limbic regions [1]. It plays important roles in various cognitive functions including motor control, behavioral adaptation, emotion processing, and learning and memory [2–4]. In the striatum, the most abundant cell type is the projecting GABAergic spiny projection neurons (SPNs), accounting for ~ 90% of total striatal neurons [5]. Their dendrites are densely covered by spines, a distinct morphological feature of this cell type [6]. SPNs receive information from different brain regions and send output signals to other basal ganglia nuclei, forming the well-known direct and indirect pathways associated with motor control and cognitive functions [7]. In addition, the striatum contains various types of interneurons with local axonal arborizations, regulating SPNs and striatal network activity [1]. Striatal interneurons have distinct morphological and electrophysiological characteristics [8]. Among them, parvalbumin (PV)-expressing interneurons are GABAergic cells with smooth dendrites and a non-adapting high-frequency firing pattern. Although much less abundant (<5% of the total) than SPNs, PV cells play key roles in striatal information processing by producing feedforward inhibition onto SPNs [9–12]. Abnormal spiking and synaptic activity in SPNs and PV cells causes malfunction of the whole basal ganglia network and contributes to the development of brain disorders, such as Parkinson’s disease and Huntington’s disease [13, 14].

Previous studies revealed that GABAergic interneurons in the cortex form massive autaptic connections (known as autapses), i.e. synaptic contacts between the axon of a neuron and its dendrites or soma [15]. Previous and recent findings have also shown that glutamatergic projection neurons form autaptic connections [16, 17]. Synaptic transmission mediated by autapses generates feedback signals after individual action potentials (APs), providing temporally precise self-control of neuronal spiking activity [17, 18]. In addition to the synchronous release (SR) of neurotransmitters tightly coupled with presynaptic AP generation, delayed asynchronous release (AR) also occurs at autapses [19–21]. Because of this feature of AR at autapses or the conversion of SR to AR with Sr²⁺ [22], it is relatively easy to determine whether a neuron forms autaptic connections [19].

Previous morphological findings suggest that striatal SPNs may form autaptic contacts [6, 23]. Moreover, in vivo recordings have provided indirect electrophysiological evidence that autapses may exist in SPNs [24]. It remains unclear, however, whether SPNs form physiologically functional autapses. Since PV interneurons in the striatum show morphological and electrophysiological properties similar to those in the cortex, it is of interest to know whether striatal PV cells provide feedback regulation of spiking activity via autapses [20, 21].

With the apparent occurrence of asynchronous neurotransmitter release under physiological conditions or in the presence of Sr²⁺ in the bath solution, we have shown abundant autaptic connections in cortical pyramidal cells [17]. In this study, with similar experimental protocols, we examined the occurrence of autaptic AR in both PV cells and SPNs. Surprisingly, we found that PV cells, but not SPNs, form functional autapses. Furthermore, our computational simulations of individual neurons and striatal networks suggest the functional roles of PV cell autapses in regulating neuronal activity in both self-activity and network oscillations.

Materials and Methods

Ethical Statement

For each experiment, animals of similar ages were randomly assigned. The use and care of experimental animals were in line with the guidelines of the Animal Advisory Committee at Fudan University and the State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University. Experimental protocols were approved by the Institutional Ethics Committee at Beijing Normal University.

Slice Preparation

Wild-type C57BL/6J mice and PV-CRE::Ai9 mice (postnatal 45–70 days) were used to prepare striatal slices. Animals were housed with ad libitum access to water and food and with a 12-h light/12-h dark cycle. We anesthetized the mice with sodium pentobarbital (50 mg/kg, intraperitoneal injection), followed by decapitation when there was no sensorimotor reflex. We dissected out the brain and immersed it in 0 °C aerated (95% O₂ and 5% CO₂) sucrose-based artificial cerebrospinal fluid (ACSF; without NaCl, but with 126 mmol/L sucrose) slicing solution. Tissue blocks were cut coronally at 300 μm in this solution using a vibratome (Leica VT1200S). Slices were collected and then incubated in aerated ACSF at 35 °C. The ACSF contained (in mmol/L) 126 NaCl, 2.5 KCl, 2 MgSO₄, 2 CaCl₂, 26 NaHCO₃, 1.25 NaH₂PO₄, and 25 dextroses (315 mOsm, pH 7.4). After 40 min of incubation, slices were maintained in the same solution at room temperature until use.

Electrophysiological Recording

Slices were transferred to a recording chamber perfused with aerated ACSF (34–35 °C) at 1.2 mL/min. Dorsal striatal neurons were visualized using an upright infrared differential interference contrast microscope (BX51WI or BX61WI with the two-photon imaging system, Olympus). SPNs were identified by their medium-sized cell bodies and densely-distributed dendritic spines, together with their electrophysiological properties (see main text). PV cells were identified by the expression of tdTomato, together with their fast-spiking firing pattern and narrow AP waveforms. After recording, the recorded neurons were further identified using avidin staining.

Patch pipettes had an impedance of 5–8 MΩ when filled with a high Cl^– internal solution containing (in mmol/L) 71 KCl, 72 K-gluconate, 2 MgCl₂, 2 Na₂ATP, 10 HEPES 0.2 EGTA, and 0.2% biocytin (286 mOsm, pH 7.2). The reversal potential of Cl^– was approximately –15 mV. In order to better visualize the morphology of recorded neurons during recording, we added Alexa Fluor-488 (50 μmol/L) to the internal solution. The whole-cell recording was achieved using a MultiClamp 700B amplifier (Molecular Devices). Signals were filtered at 10 kHz and then sampled by Micro 1401 micro3 at 50 kHz using Spike2 software. Firing patterns and membrane properties were examined by step current injections (500 ms in duration, − 100 pA to 1,100 pA in amplitude) in current clamp mode. To examine the occurrence of autaptic AR, we stimulated the recorded cell with trains of brief step voltage pulses (10–60 pulses, 0.5–2 ms in pulse duration, 20–200 Hz) in voltage clamp mode. The amplitude of voltage pulses was carefully adjusted to ensure the successful generation of APs (i.e. action currents in voltage clamp mode) for each brief pulse.

Kynurenic acid (1.5 mmol/L) was added to the bath solution to block fast glutamatergic transmission (mediated by both NMDA and AMPA receptors). In the Sr²⁺ experiments, we added 5 mmol/L SrCl₂ to ACSF and reduced the concentration of both CaCl₂ and MgSO₄ to 1 mmol/L. In some experiments, we perfused the slices with 50 μmol/L picrotoxin (PTX, Tocris) to determine whether autaptic responses were mediated by GABA_A receptors.

To demonstrate the effect of autapses on the firing patterns of PV interneurons under relatively physiological conditions, we first obtained recordings from PV-positive cells with normal (low Cl^–) pipette solution containing (in mmol/L) 140 K-gluconate, 3 KCl, 2 MgCl₂, 2 Na₂ATP, 10 HEPES, 0.2 EGTA, and 0.2% biocytin (286 mOsm, pH 7.2), and applied step currents (100–2000 pA in amplitude, 3 s in duration) to examine neuronal input-output properties. Then, we carefully withdrew the patch pipettes and re-patched the same cells with high Cl^– internal solution to determine whether they had autapses. We also made whole-cell recordings using the low Cl^– pipette solution in the presence of PTX. With these experiments, we were able to compare the firing patterns of PV cells with and without autapses. An AP burst was identified when the inter-spike-intervals (ISIs) were <16 ms.

After recordings, slices were fixed with 4% paraformaldehyde for >12 h and stained with Alexa Fluor-488 conjugated avidin. The z-stack images (0.7 μm between successive images) of individual cells were acquired by a confocal microscope (A1 plus, Nikon) equipped with a 60× objective. Three-dimensional reconstruction of the cells was applied with ImageJ to form a continuous 3D representation of the entire cell structure. Individual sites where axons intersected dendritic branches were visually inspected and then imaged at a higher magnification (60× objective, z spacing of 0.2 μm).

Computational Models

The neuronal models of striatal PV interneurons and SPNs were similar to those previously reported [25]. Synaptic currents (I_GABA) mediated by synchronous and asynchronous GABA release were formulated as in our previous work [26] and that from Volman and colleagues [27]. The model parameters can be found in Supplementary Table S1.

For a single PV neuron, we reduced the number of PV neurons to 1 and focused only on its spiking activity and membrane potential V_m. We simulated the absence of autapses (No Aut) by adjusting the chance of autaptic connection to 0. For autaptic PV cells, we simulated conditions with different strengths of AR by adjusting the parameter ${\tau }_{\mathit{ar}}$. To assess the contribution of autapses to network oscillation, we normalized (z-score) the simulated local field potential (LFP) before plotting the power spectrum. All simulations were run on MatLab software (version R2021a). All differential equations were integrated using a fourth-order Runge-Kutta algorithm with a time step of 0.01 ms.

Data Analysis

Spike2, MiniAnalysis, and MatLab software were used for data analysis. All measurements were taken from different cells. Unless otherwise stated, data presented in the main text are the mean ± SEM. The error bars in the figures are also the SEM.

The frequency of spontaneous IPSC events was obtained from a 2-s baseline current just before the stimulus onset and considered as the baseline frequency. For a particular neuron, if the frequency of IPSC events within 300 ms after the train stimulation (60 pulses at 200 Hz) was 10 Hz higher than that at baseline, it was considered to be an autaptic neuron with asynchronous GABA release. The PT-AR instantaneous frequency was also calculated. The termination time of the post-train asynchronous release (PT-AR) IPSC barrage was the time of the last IPSC event before the AR frequency reached the baseline frequency. The PT-AR duration was the period between the cessation of the train stimulation and the end of the AR barrage.

For two independent observations with normal distributions (P > 0.05, Shapiro Wilk’s test), we used two-sample Student’s t-test. Non-normal data were compared with the Wilcoxon rank sum test. The Kruskal-Wallis test for analysis of variance (ANOVA), and Tukey’s test for post hoc analysis, were used for comparisons of multiple groups. Datasets were considered to be significantly different if P < 0.05.

Results

SPNs do not Appear to form Functional Autapses

To determine whether SPNs and PV cells form functional autapses, we selectively recorded from these cells in coronal slices containing the dorsal striatum of PV-CRE::Ai9 mice (P45–70). PV cells were identified by their expression of tdTomato, while SPNs in the dorsal striatum were identified by their characteristic morphological and electrophysiological features. As reported previously [28], SPNs recorded in our experiments had a medium-sized cell body, spiny dendrites (Fig. 1A, B), and a delayed firing pattern in response to current pulses just above the firing threshold (Fig. 1C, D). Meanwhile, consistent with the presence of an M-current in SPNs [29], a hyperpolarizing current pulse at a near-threshold V_m level caused rebound firing (Fig. 1C). Our recordings showed that the resting V_m of SPNs was − 75.7 ± 1.2 mV and the input resistance was 68.8 ± 9.7 MΩ (n = 22 cells). Consistent with previous studies [30, 31], the AP half-width of SPNs (0.83 ± 0.07 ms, n = 22) was much broader than that of PV cells (0.46 ± 0.02 ms, see below).

Fig. 1.

Striatal SPNs do not form functional autapses. A Differential interference contrast (DIC; left) and fluorescent images (right) of an SPN loaded with Alexa Fluor-488. B A representative image of a recorded SPN with avidin staining. Note the densely distributed spines in the dendrites (inset). Arrowheads indicate some of the spines. C Representative traces showing voltage responses to positive step current injections (left and middle) and rebounds firing immediately after a hyperpolarizing pulse (right). Inset, expanded action potential. D Plot of spike frequency as a function of injected currents (F-I curve) in SPNs. E Representative current traces in voltage clamp mode. Note that there was no change in the IPSC event number before and after the train stimulation under two conditions (with and without Sr²⁺). F Group data showing changes in IPSC event frequency after the train stimulation. Note that none of the increments exceed 10 Hz. ns, not significantly different. Error bars represent the SEM.

In voltage clamp mode, the membrane potential (V_m) of SPNs was held at − 70 mV. With a high-Cl⁻ (~ 75 mmol/L) internal solution in the patch pipettes, inhibitory postsynaptic currents (IPSCs) would be inward at the holding potential (calculated reversal potential: − 15 mV). Since fast glutamatergic synaptic events were blocked by kynurenic acid (Kyn, 1.5 mmol/L), all inward synaptic currents would be IPSCs. We stimulated SPNs with trains of brief voltage pulses (50 mV–100 mV, 1 ms in duration, up to 60 pulses) at frequencies from 50 to 200 Hz. As reported previously in neocortical neurons [17, 19], if the recorded neuron forms functional autapses and shows a prolonged AR or autapses with SR only but in the presence of Sr²⁺, an increase in synaptic events immediately after the train stimulation would be detected. Surprisingly, we found no significant increase in IPSC event number after the train stimulation under both experimental conditions, with or without 5 mmol/L Sr²⁺ in the bath (Fig. 1E). With train stimulation of 60 pulses at 200 Hz in the absence of Sr²⁺ (normal ACSF), the increment of IPSC event frequency after the stimulation was − 0.30 ± 0.73 Hz (n = 11, Fig. 1F). Similar results were obtained in the presence of Sr²⁺ (1.26 ± 0.69 Hz, n = 12). Therefore, distinct from previous morphological observations [6, 23] and indirect electrophysiological evidence [24], our results indicate that striatal SPNs tend not to form functional autapses.

Autapses form in Striatal PV Cells

We did similar experiments in PV cells with tdTomato expression. Close examination of their morphology revealed that PV cells possessed smooth dendrites and dense axon collaterals (Fig. 2A, B), similar to previous studies [8, 10, 32]. PV cells had a resting potential of –73.3 ± 1.3 mV (mean ± SEM, n = 26 cells) and an input resistance of 87.3 ± 9.1 MΩ. They showed unique electrophysiological properties, including a non-adaptive fast-spiking pattern (up to 199 ± 18 Hz, n = 26 cells, Fig. 2C, D) and short duration of AP waveforms (half-width: 0.46 ± 0.02 ms). Consistent with previous findings, most of the recorded PV cells exhibited a stutter firing pattern in response to a series of current steps with increasing amplitudes (Fig. 2C) [32].

Fig. 2.

Electrophysiological properties and autaptic responses in PV cells. A DIC (left) and tdTomato fluorescent images (right) of a recorded PV cell. B A representative image of a recorded PV cell with post hoc staining. C Example voltage responses evoked by current pulses. Note the generation of a single AP (left) and the stutter firing pattern (right) evoked by the indicated current steps. Inset, expanded action potential. D F–I curve of PV cells. E Barrages of autaptic events (arrowheads) occur immediately after the cessation of an AP burst evoked by a strong current pulse. Error bars represent the SEM.

We frequently recorded voltage fluctuations after AP bursts evoked by positive current steps (Fig. 2E). These fluctuations contained barrages of depolarizing postsynaptic potentials (PSPs). Since fast glutamatergic synaptic events were blocked by 1.5 mmol/L Kyn, these PSPs would be inverted inhibitory PSPs at the resting V_m. Since PV cells release GABA at their axon terminals and usually hyperpolarize the postsynaptic neurons under physiological conditions, the discharge of PV cells was unlikely to drive other neurons to generate APs [20, 21]. Therefore, the PSP barrages occurring immediately after a PV cell burst were unlikely to be caused by polysynaptic events through the network. They reflected asynchronous GABA release at autapses of the recorded PV cell, similar to those of fast-spiking cells in mouse and human neocortex [20]. From three-dimensional reconstruction, we could easily distinguish the dendrites and axons of PV cells. The dendrites were relatively smooth and much thicker than the axons. The axons were thin and long, often beaded. We often saw axons intersecting dendritic branches in single-plane images, indicating that those close contacts could be autaptic connections (Fig. S1A).

Autapses are Abundant in PV Cells and Mediated by GABA_A Receptors

Similarly, in voltage clamp mode, we recorded barrages of autaptic currents immediately after train stimulation in PV cells (Fig. 3A). In response to a train of stimuli with 60 APs at 200 Hz, autaptic events reflecting PT-AR of GABA lasted for 329 ± 35 ms with a total number of 11.0 ± 1.3 events (n = 55 cells, Fig. 3A). These autaptic events were completely blocked by bath application of the GABA_A receptor antagonist, picrotoxin (PTX, 50 μmol/L, n = 10, Fig. 3B, C). In the current clamp mode, PTX also diminished the depolarizing autaptic AR events immediately after the burst firing evoked by current steps (n = 4, Fig. S1B). These results indicate that autaptic transmission is mediated by GABA_A receptors.

Fig. 3.

Autaptic transmission is mediated by GABA_A receptors and autaptic PV cells are abundant. A Upper: An example current trace showing the barrages of post-train asynchronous release (PT-AR) events (arrowheads). The PV cell is stimulated with voltage pulses to evoke 60 APs at 200 Hz (holding potential: –70 mV). Lower: Plot of synaptic event frequency versus time (bin size: 50 ms). The dashed line indicates the onset of the AP burst. B Example traces before and after the application of 50 μmol/L PTX. C Group data showing the effect of PTX on PT-AR duration and event number. D An example current trace showing the occurrence of PT-AR in a recorded PV cell bathed with Sr²⁺-containing ACSF. E Group data comparing the PT-AR duration and event number under two conditions (Ctrl vs Sr²⁺). F The percentages of autaptic cells in all recorded PV-positive neurons. **P < 0.01.

Previous studies have shown that asynchronous neurotransmitter release occurs selectively in certain types of synapses. For example, AR is much stronger in the output synapses of the neocortical pyramidal cell onto somatostatin-containing neurons as compared to those in PV neurons [33]; hippocampal granule cells receive greater AR from cholecystokinin cells than that from PV neurons [34]. To exclude the possibility that some autaptic PV cells might have only SR, but no detectable AR, we added SrCl₂ (5 mmol/L) to the bath solution to desynchronize autaptic GABA release [22] (Fig. 3D). In our experiments, the presence of Sr²⁺ significantly increased the PT-AR duration from 387 ± 43 to 586 ± 63 ms (control, n = 27; Sr²⁺, n = 21, P = 9.97×10⁻³, two-sample Student’s t-test). The number of AR events also slightly increased from 13.1 ± 1.7 to 15.8 ± 2.1, but with no significant difference (P = 0.31, Wilcoxon rank sum test, Fig. 3E).

We next calculated the percentage of PV cells that form autapses. In recorded PV cells, we applied high-frequency stimuli (20–60 APs, 150–200 Hz), with or without Sr²⁺ in ACSF, and monitored the occurrence of post-train autaptic events. With this method of AR detection, however, we found that the percentage of PV cells with autaptic AR in Sr²⁺ solution (50.0%, n = 21/42) was similar to that in control conditions (without Sr²⁺, 46.6%, n = 61/131, Fig. 3F), and also similar to that found in cortical PV neurons [20]. These probabilities are underestimated because slice preparation reduced the complexity of neuronal dendrites and axons. Together, these results indicate that autapses are abundant in striatal PV cells and autaptic AR occurs in almost every autaptic cell.

Autaptic AR Strength Depends on Stimulation Intensity

Next, we investigated the dependence of autaptic AR strength on the intensity of neuronal activity. We changed the number and the frequency of voltage pulses and monitored the strength of autaptic AR by measuring the PT-AR duration and counting AR events. Consistent with previous findings, the strength of autaptic AR was positively correlated with the number or frequency of stimuli [19] (Fig. 4A, B). When the number of stimulus pulses (200 Hz) increased from 20 to 40 and 60, the average duration of PT-AR increased from 50.6 ± 24.7 to 151 ± 58 and 251 ± 81 ms (ANOVA and post hoc Tukey’s test, n = 12, P = 0.034), and the average number of AR events increased from 2.31 ± 1.28 to 5.33 ± 2.30 and 9.67 ± 3.44 (ANOVA and post hoc Tukey’s test, n = 12, P = 0.044). A similar dependence of AR strength was found when we increased the frequency of stimulus pulses. We stimulated the PV cells with a range of frequencies (from 20 to 200 Hz) and found a progressive increase in the PT-AR duration (ANOVA and post hoc Tukey’s test, n = 15, P = 8.66×10⁻⁷) and event number (n = 15, P = 5.68×10⁻⁶, Fig. 4C, D). In response to 60 APs at 50 Hz, the average duration and event number of PT-ARs was 19.8 ± 9.2 ms and 0.81 ± 0.38, respectively, significantly less than those at 200 Hz (204 ± 41 ms, P = 2.14×10⁻⁴; 6.50 ± 1.48, P = 5.56×10⁻⁴, n = 15).

Fig. 4.

The strength of autaptic AR is dependent on the spiking intensity of PV cells. A Representative trace showing autaptic currents with different numbers of stimulus pulses (at 200 Hz). Arrowheads indicate the end time of PT-AR. B Group data showing the PT-AR duration and event numbers in response to different numbers of stimulus pulses at 200 Hz. C, D As in (A) and (B), but with 60 pulses at different stimulus frequencies. *P < 0.05; **P < 0.01; ***P < 0.001. Error bars represent the SEM.

The Autapse Regulates Spiking Activity of a Single PV Cell

Next, we explored the physiological roles of autapses in striatal PV cells. We first examined the firing patterns and input-output f-I curves under relatively physiological conditions using patch pipettes containing low Cl^– internal solution, then the same cells were re-patched using a high Cl^– internal solution to determine whether they had autapses. In this way, we had two groups of PV cells, one with and one without autapses. In another set of experiments, we obtained recordings from PV cells with low Cl^– internal solution with autapses blocked by 50 μmol/L PTX. Then, we compared the neuronal firing patterns with and without autapses in these three groups of cells (Fig. 5).

Fig. 5.

Autapses alter the firing pattern and input-output properties of PV cells. A Representative traces showing voltage responses of PV cells (Aut vs No Aut, with and without autapses) to moderate (350–650 pA) and strong (1050–1350 pA) current injections. B f-I curves of PV cells under three conditions including those recorded in the presence of 50 μmol/L PTX. C Group data comparing the spike frequencies (boxed regions in B) in response to moderate and strong current injections. D Comparison of the burst duration in the three groups of PV cells. E Group data comparing the burst duration in boxed regions shown in D. *P < 0.05; ns, no significant difference. Error bars represent the SEM.

We found that neurons with autapses had lower firing rates (P = 0.014, ANOVA. Fig. 5A–C) and shorter bursts (P = 0.051, Fig. 5A, D, E) when stimulated with moderate current injection (350–650 pA). However, in response to strong current injection (1050 pA–1350 pA), PV cells with autapses were resistant to depolarization block (a physiological phenomenon due to Na⁺ channel inactivation) and able to maintain a higher firing frequency (P = 0.011) and longer bursts (P = 0.049). These findings indicate that autaptic self-inhibition makes PV neurons less active when they receive moderate inputs, but more active in response to strong inputs.

Based on mathematical models of individual PV cells and striatal networks containing not only the PV interneurons but also the principal cells, D1 and D2 SPNs [25], together with models of SR and AR [26, 27], we examined the specific functions of synchronous GABA release or its combination with AR (i.e. SR alone, or SR+AR) in regulating the spiking activity of PV cells and SPNs.

We added autapses (with or without AR) to the dendritic compartment of the PV neuron model and compared the differences in firing rate and profile of the AP burst (Fig. 6A–C). Similar to previous studies [35], our PV cell model also showed two distinct electrophysiological characteristics, stuttering and γ resonance (Fig. 6B, C). The strength of autaptic transmission was set to the experimental findings (Fig. 3A) [20, 26]. Considering that the AR strength was underestimated because some of the autaptic contacts were lost during slicing procedures, we set the standard AR parameters (AR = 1, corresponding to the model parameter τ_AR = 150 ms) similar to the strongest AR found in our experiments, reflecting a condition in which the dendrite branches and axon collaterals were relatively better preserved (Fig. 6D). In agreement with the experimental findings, the strength of simulated AR (both PT-AR event number and duration) in the PV cell model showed dependence on the stimulus number and frequency (Fig. 6E, F).

Fig. 6.

Autapses regulate spiking activity in a single PV cell (simulations). A Schematics showing three simulation conditions: a PV cell without an autapse (No Aut, i), with an autapse (Aut, i.e. SR alone, ii), and autaptic AR (SR+AR, iii), in a model of a single PV cell. The PV cell receives tonic excitation (14 μA/cm²) and Poisson noise with a rate of 100 inputs per second. B Spiking activity of the PV cell model under corresponding conditions shown in (A). C Spectrograms of the voltage traces in (B). D An example trace of autaptic currents with AR strength = 1 (i.e. τ_AR = 150 ms) when the PV neuron is allowed to discharge 60 APs at 200 Hz. E Plots of PT-AR duration and event number as a function of the number of stimuli at 200 Hz (n = 100 trials). Note the increase in duration and event number as the AR strength increases from 0.5 to 1 and 2. F As in (E), but with 60 APs at different stimulation frequencies. G Group data showing the effects of autapses (SR only) on the spiking frequency, burst duration, and interval in a single PV cell. H Group data comparing the firing rate and burst profile with different AR strengths. **P < 0.01; ***P < 0.001; ns, not significantly different.

In order to mimic the two physiological conditions, baseline and high dopamine (DA) states, we injected two background currents (7 and 14 μA/cm²) together with the same noise current (Poisson noise) into the modeled PV cell. We then compared the firing patterns with (two conditions: SR only, SR+AR) and without autapses (Fig. 6B). At the baseline DA level (Fig. S2), autaptic SR alone decreased the spike frequency and the duration of spike bursts, but increased the interval between bursts. However, adding AR had no further effect on these spiking properties. At the high DA level (Fig. 6), we found that SR alone reduced the average firing rate (n = 10 trials, P = 5.75×10⁻⁵, two-sample Student’s t-test) and the average duration of bursts (P = 2.00 × 10⁻³, two-sample Student’s t-test), but had no significant effect on the burst interval (P = 0.47, Wilcoxon rank sum test, Fig. 6G). These results are consistent with the experimental results showing that autapses decreased the burst duration with moderate stimulation. When we added AR to the autapse (SR+AR), we also found a slight decrease in the firing rate (n = 40 trials, P = 1.21×10⁻⁹, ANOVA) and the burst duration (P = 1.31×10⁻³) as the AR strength increased from 0 to 2. Increasing AR strength also significantly prolonged the interval between bursts (P = 7.87 × 10⁻³, Fig. 6H).

PV Cell Autapses Regulate SPN Firing and Striatal Oscillations

Next, we investigated the functional role of autapses in PV cells in the regulation of striatal neuronal and network activity. In a simulated network composed of 50 PV neurons, 100 D1 SPNs, and 100 D2 SPNs, we also set the input currents of PV cells to 7 and 14 μA/cm² to mimic the baseline and high DA states, respectively. At the baseline DA level (Fig. S3), PV cells discharged at low frequencies (~9 Hz); adding autaptic SR alone (SR alone, $U_{\mathit{sr}}$ = 1, ${\tau }_{\mathit{sr}}$ = 20 ms) had a marginal effect on PV cell spiking activity and the power in the beta and low gamma bands. As expected, adding AR (SR+AR, τ_AR = 150 and 300 ms for AR strength 1 and 2, respectively) had no further effect on both neuronal and network activity (Fig. S3), due to the absence of AR at low firing rates (Figs. 4, 5).

At the high DA level (Fig. 7), PV cells discharged at ~40 Hz, consistent with that found in the movement state [36]. Similar to single PV cell simulations, adding autapses had no significant effect on burst interval (P = 0.070, two-sample Student’s t-test) but significantly reduced burst duration (P = 7.01×10⁻⁹) and the average spiking frequency of PV cells in the network model (n = 10 trials, P = 1.83 × 10⁻⁴ for PV cells, two-sample Student’s t-test, Fig. 7A–G). By contrast, autaptic SR alone increased the activity in both D1 (n = 10 trials, P = 3.59 × 10⁻⁵, two-sample Student’s t-test) and D2 SPNs (n = 10 trials, P = 2.66×10⁻⁴). At the network level, the power density at certain frequencies in the gamma bands (75–85 Hz) showed a dramatic decrease (n = 10 trials, P = 1.83×10⁻⁴, Wilcoxon rank sum test), but those at other frequency bands were significantly increased (Fig. 7E, G).

Fig. 7.

PV cell autapses regulate striatal neuronal and network activity (simulations). A Schematics showing three simulation conditions in the striatal network model: PV cells without an autapse (No Aut, (i), with an autapse (Aut, SR alone, ii), and autaptic AR (SR+AR, iii). The network model contains 50 PV cells, 100 D1, and 100 D2 SPNs. PV cells receive tonic excitation (14 μA/cm²) and Poisson noise, while D1 and D2 SPNs receive tonic excitation at a strength of 1.29 and 1.09 μA/cm², respectively, corresponding to a high dopamine state. AR strength 1 corresponds to τ_AR = 150 ms. B Raster plots of the three types of striatal neurons under the corresponding conditions shown in (A). C Example local field potential (LFP) traces in the three conditions. D Spectrograms of the LFP traces in (C). E Mean power spectral analysis of the LFP traces under the three conditions. F Group data showing the effect of autapses (SR alone) on the firing rates of distinct cell types and the PV cell burst duration and interval. G Changes of the power density at LFP frequencies. H Group data comparing the firing rate, burst duration, and interval with different AR strengths. I Changes of the power density at the indicated frequencies. *P < 0.05; **P < 0.01; ***P < 0.001; ns, Not significantly different.

In subsequent simulations, we added autaptic AR to PV cells and found that autaptic AR slightly decreased the firing rate of PV neurons (n = 40, P = 3.18×10⁻⁵, ANOVA, Fig. 7H). In sharp contrast to SR alone, AR had no effect on burst duration (P = 0.563, ANOVA) but slightly increased the burst interval (P = 6.61×10⁻⁵, ANOVA, Fig. 7H). A marginal increase in firing frequency occurred in D1 SPNs, but not in D2 SPNs. An increase in AR strength significantly enhanced the power density of theta (6–8 Hz, P = 5.78×10⁻³, ANOVA) and gamma bands (30–75 Hz, P = 0.0133; 75–85 Hz, P = 2.07×10⁻⁸), but not those of other bands (Fig. 7I). Together, these simulations suggest a role of autaptic AR in the regulation of striatal neuronal and network activity.

Discussion

In this study, we showed that autaptic contacts occur in PV interneurons in the striatum. By contrast, no functional autaptic connections were found in the striatal principal cell type, SPNs. Synaptic events mediated by GABA_A receptors were recorded after a high-frequency AP burst in PV cells, reflecting asynchronous GABA release at its autapses. We further found that the AR strength was dependent on the frequency and the number of APs. Our simulation results suggested that autapses regulate the spiking activity of PV cells by providing self-inhibition. At the network level, activation of PV cell autapses also regulates the spiking activity of SPNs and shapes striatal network oscillations.

PV-positive cells are found in different brain regions and contribute significantly to information processing in a variety of brain circuits [37]. In the neocortex and hippocampus, PV neurons have been confirmed to have functional autapses, similar to certain types of conventional synapses, these autapses possess two modes of neurotransmitter release, synchronous and asynchronous modes [18, 20, 38]. Recordings from acute cortical slices have revealed that the percentage of autaptic PV cells in rodents (~85%) is more than that found in humans (64.3%) [18, 20]. These percentages could be underestimated because slicing procedures would cut some neurites and thus reduce the complexity of dendritic and axonal branches. In the striatum, we found that about half of the recorded PV cells formed autaptic connections under normal physiological conditions (i.e. normal bath solution at body temperature). Since adding Sr²⁺ to the bath desynchronizes neurotransmitter release and thus delays the occurrence of autaptic events [17], the percentage of autaptic PV cells would be more accurate. However, the percentage with Sr²⁺ was similar to that found in normal bath solutions, indicating that AR occurs under physiological conditions.

Do the principal SPNs form autapses? Early studies in the 1970s reported that some SPN axon collaterals target the soma or proximal dendrites of the same cell in the monkey striatum [6]. Later, Park and colleagues made intracellular recordings in rats and found the occurrence of recurrent inhibition in the recorded SPNs when they stimulated the substantial nigra, providing indirect evidence for the possible existence of autapses [24]. In cultured SPNs, Shi and Rayport recorded autaptic PSPs [23]. In our experiments, we revisited this early question in striatal slices from young adult mice. Surprisingly, we failed to detect any autaptic responses in the recorded SPNs in both normal and Sr²⁺-containing bath solutions. Therefore, our results indicate that SPNs tend not to form functional autapses, distinct from those previous observations. The reasons for the contradictory findings could be as follows. First, the morphologic intersection of axons and dendrites observed in previous studies does not necessarily mean a synaptic structure. Second, in vivo recording of recurrent inhibition in SPNs may result from the polysynaptic transmission. Last, cultured neurons may form redundant synaptic connections including autapses. Therefore, we believe that SPNs do not form functional autapses. However, it remains to be further determined with electron microscopy whether SPNs form silent autaptic contacts [39].

What are the functional roles of PV autapses? It has long been thought that autapses regulate the firing pattern of neurons, thus affecting network oscillations [40]. Synchronous GABA release enhances the temporal precision of APs in neocortical PV cells [41]. Autapses promote synchronized firing in neocortical PV cells, allowing them to follow gamma oscillations [42]. Meanwhile, the autaptic AR decreases PV cell spike reliability and desynchronizes the local network [21]. In our simulations, we demonstrated that autaptic SR of GABA shortened the burst duration in PV neurons, allowing earlier rebound firing in SPNs. In contrast, adding autaptic AR prolonged the burst interval of PV cells and thus allowed a wider time window for SPN to discharge. The activation of SPNs and PV cells leads to a change in network oscillation. Indeed, the high gamma band power was significantly decreased if we added SR. However, in the presence of AR, both theta and gamma power were significantly increased (Fig. 7). Since AR only occurred at high frequencies of PV cells, it should only play a role at high DA levels. As expected, adding AR had no additional effect on neuronal and network activity when PV cells discharged at low frequencies (i.e. baseline DA level). Therefore, we mainly focused on the physiological contributions of autapses in states when PV cells discharged at high frequencies.

Feedback inhibition provided by autaptic SR would reduce PV cell spike frequency and shorten its burst duration. Because AR occurs not only during the burst but also after it, AR-induced prolonged hyperpolarization would prevent the emergence of the next burst in PV cells, thus increasing the burst interval and prolonging SPN firing. Since PV cells show a fast-spiking firing pattern and are able to discharge up to 220 Hz (Fig. 2), they contribute largely to high-frequency LFP oscillations. Indeed, we found a decrease in PV cell spike frequency and a shift of power density from high gamma to other bands after adding autaptic SR (Fig. 7E–G). Since the frequency of AR events would change progressively after a PV cell burst, it would shape LFP oscillations in different bands. We found increases in gamma band power after introducing AR into the network model. The accumulation of AR events as a whole would cause slow hyperpolarization after a PV cell burst, which may contribute to the enhancement of theta band power. Therefore, PV cell autapses shape neuronal activity in both PV cells and SPNs and regulate network activity via both SR and AR.

Different bands of network activity may play distinct roles in brain functions. The theta band oscillation has been linked to cognitive behavioral states, such as working memory and decision-making [43, 44]. Gamma oscillation is associated with the initiation of movement [45]. Striatal theta oscillation coherence between other brain regions such as the hippocampus and amygdala has been shown to facilitate information exchange, and gamma oscillation helps to organize the active neurons in various brain regions [46]. The contribution of PV cell autapses to these bands may play critical roles in proper motor execution and cognition. It has been speculated that dysfunction of interneurons in the striatal network could be an important mechanism for neurological diseases such as Parkinson's disease [47]. It remains to be determined whether the strength of autaptic connections and their role in the regulation of network oscillations are pathologically altered in different brain disorders.

Together, our results show that striatal PV interneurons, but not SPNs, develop functional autapses. Considering the important role of PV cells in the striatal network, we believe that their autapses and the two GABA release modes are fundamental circuit elements and physiological mechanisms contributing to basal ganglia functions such as motor control and emotion processing. In addition, PV cell autapses could be a key target for the development of new therapies for striatum-related diseases. Our findings also suggest that autaptic connections should be considered when interpreting the function of the basal ganglia and building computational models.

Supplementary Information

Below is the link to the electronic supplementary material.

Data Availability

All relevant data and computation codes are available from the corresponding authors upon request.

Conflict of interest

The authors claim that there are no conflicts of interest.