Compartmentalization of GABAergic Inhibition by Dendritic Spines

See allHide authors and affiliations

Science  10 May 2013:
Vol. 340, Issue 6133, pp. 759-762
DOI: 10.1126/science.1234274

Dendritic Precision Strikes

The effects of excitatory synaptic inputs are considered to be highly compartmentalized because of the biophysical properties of dendritic spines. Individual inhibitory synapses, however, are thought to affect dendritic integration in a more extended spatial region. Combining optogenetic stimulation of dendrite-targeting γ-aminobutyric acid—mediated interneurons with two-photon calcium imaging in postsynaptic pyramidal cell dendrites, Chiu et al. (p. 759) challenge this latter view. The findings suggest that the effect of an inhibitory synapse can be as compartmentalized as that of an excitatory synapse, provided that the synapses are localized on spine heads.


γ-aminobutyric acid–mediated (GABAergic) inhibition plays a critical role in shaping neuronal activity in the neocortex. Numerous experimental investigations have examined perisomatic inhibitory synapses, which control action potential output from pyramidal neurons. However, most inhibitory synapses in the neocortex are formed onto pyramidal cell dendrites, where theoretical studies suggest they may focally regulate cellular activity. The precision of GABAergic control over dendritic electrical and biochemical signaling is unknown. By using cell type-specific optical stimulation in combination with two-photon calcium (Ca2+) imaging, we show that somatostatin-expressing interneurons exert compartmentalized control over postsynaptic Ca2+ signals within individual dendritic spines. This highly focal inhibitory action is mediated by a subset of GABAergic synapses that directly target spine heads. GABAergic inhibition thus participates in localized control of dendritic electrical and biochemical signaling.

A challenge to elucidating the function of synaptic inhibition is the diversity of γ-aminobutyric acid–releasing (GABAergic) interneurons found in cortical circuits (13). Several interneuron classes, including those that express somatostatin (SOM-INs), target the dendrites of excitatory, glutamatergic pyramidal cells (35). SOM-INs regulate the initiation of action potential bursts generated via active currents in postsynaptic dendrites (68). We hypothesized that these inputs might also exert focal influence over dendritic signaling. Here, we used electrophysiological, optical, and computational approaches to investigate the localized actions of GABAergic inhibition in pyramidal cell dendrites.

To activate dendritic GABAergic synapses, we used a somatostatin-Cre mouse line (9) (Fig. 1A and fig. S1, A and B) to conditionally express channelrhodopsin-2 (ChR2) (10) in SOM-INs of the prefrontal cortex (fig. S1, C and D). In acute brain slices prepared 2 to 3 weeks after viral injection, pulses of light (5 ms, 473 nm) delivered through the microscope objective evoked action potentials (APs) in fluorescently identified SOM-INs (fig. S2, A to C). Whole-cell recordings in layer 2/3 pyramidal neurons revealed corresponding inhibitory postsynaptic potentials (IPSPs) (Fig. 1, B and C, and fig. S2, D to F). For subsequent experiments, type A GABA receptor (GABAAR)–mediated IPSPs were isolated by including the selective type B GABA receptor (GABABR) antagonist CGP-55845 in the perfusate (Fig. 1C). IPSPs exhibited a reversal potential of –69.9 ± 1.5 mV (mean ± SEM, n = 5) that did not differ significantly from the value recorded via gramicidin-based perforated patch (–72.4 ± 1.7, n = 6, P = 0.3, fig. S2, G and H).

Fig. 1 SOM-INs mediate inhibition of dendritic Ca2+ signals.

(A) td-Tomato expression in the prefrontal cortex of SOM-Cre;Ai9 mice. Scale bar indicates 200 μm. (B) Recording configuration. (C) Light-evoked IPSPs (ACSF) are abolished by picrotoxin (PTX). Scale bars, 1 mV, 50 ms. (Inset) Light-evoked APs in a SOM-IN. Scale bars, 20 mV, 50 ms. (D) (Left) 2PLSM image of a layer 2/3 pyramidal neuron. Scale bar, 25 μm. (Right) AP-evoked ΔCa2+ in the spine (Spn) and dendritic shaft (Dnd) indicated by the dashed line. Scale bars, 1 μm, 50 ms. (E) (Left) Vm during AP (black), IPSP (blue), and IPSP-AP (red). Scale bars, 2 mV, 100 ms. (Middle and right) ΔCa2+ (Spn, Dnd) in response to AP (black, blue) or IPSP-AP (red, orange). Scale bars, 1% ΔG/Gsat (where G indicates the amount of green fluorescence), 100 ms. (F) Ca2+ inhibition for dendritic shafts versus spines. Gray region indicates significant spine inhibition. (G) Average Ca2+ signals (±SEM) evoked by AP or IPSP-AP for locations showing significant inhibition. Scale bars, 1% ΔG/Gsat, 50 ms. (H) (Left) Ca2+ transients from spine in (E), recorded in PTX. Scale bars, 1% ΔG/Gsat, 100 ms. (Right) Average Ca2+ inhibition before (ACSF) and after GABAA block (PTX). *P < 0.05 (paired Student’s t test).

To determine how inhibition influences dendritic activity in pyramidal neurons, we used two-photon laser scanning microscopy (2PLSM) to image calcium (Ca2+) in apical dendritic spines and shafts. Ca2+ transients (ΔCa2+) were evoked by somatic APs (Fig. 1, D and E, and fig. S3, A and B) and were mediated by voltage-gated Ca2+ channels (VGCCs) (fig. S3C). We compared AP-evoked Ca2+ signals under control conditions (ΔCa2+ctl) and when preceded by an IPSP (ΔCa2+inh) (15-ms interval) evoked by a light pulse targeting the imaged region (Fig. 1E). In 57% (73/127) of randomly imaged spines, optical activation of SOM-INs produced a significant reduction (>15%, see methods and fig. S3, D to F) in the AP-evoked ΔCa2+. At these locations, the average Ca2+ inhibition (ΔCa2+inh/ΔCa2+ctl) was significantly greater for spines than for neighboring dendritic shafts (0.60 ± 0.02 versus 0.78 ± 0.03, P < 0.001; Fig. 1, F and G). The inhibition of ΔCa2+ was abolished by application of the GABAA antagonist picrotoxin (n = 8, P < 0.05, Fig. 1H). Similar Ca2+ inhibition was seen in basal dendrites (23/49 spines; 0.73 ± 0.02 versus 0.87 ± 0.02 for spines and shafts, respectively, P < 0.01; fig. S4).

We frequently observed inhibited and uninhibited spines in close proximity, suggesting compartmentalized GABAergic control of Ca2+ signaling. We therefore imaged Ca2+ inhibition within a small dendritic region. Spines adjacent to an inhibited reference spine typically showed little modulation despite the presence of a somatic IPSP (Fig. 2, A and B, and fig. S5, A and B). We generated “maps” demonstrating heterogeneous inhibition over short distances (Fig. 2C). There was significantly greater inhibition for each reference spine than for its adjacent neighbor (0.58 ± 0.03 versus 0.82 ± 0.03, P < 0.001, n = 22 maps), and inhibition between neighbors was not correlated (Pearson r2 = 0.12, P = 0.09, Fig. 2H). Inhibition in individual spines was not correlated to the magnitude of ΔCa2+ctl (fig. S5C) and was unchanged for experiments conducted at near-physiological temperature or with GABABR function intact (fig. S5D).

Fig. 2 GABAergic dendritic inhibition is highly compartmentalized.

(A) Inhibition mapping using ChR2 stimulation (asterisk) of SOM-INs. Scale bar, 1 μm. (B) ΔCa2+ evoked by AP and IPSP-AP for spines (black and red, respectively) and dendritic shafts (blue and orange, respectively) indicated in (A). Scale bars, 2% ΔG/Gsat, 50 ms. (Inset) Somatic IPSP. Scale bars, 1 mV, 100 ms. (C) Population data for Ca2+ inhibition versus distance from the reference spine. Average binned (5 μm) data shown in red. Gray region indicates significant inhibition. (D) Inhibition mapping using GABA uncaging. Scale bar, 1 μm. (E) ΔCa2+ evoked by AP and IPSP-AP. Scale bars, 2% ΔG/Gsat, 50 ms. (Inset) Somatic IPSP. Scale bar, 1 mV, 100 ms. (F) As in (C) for GABA uncaging. (G) Average Ca2+ signal (±SEM) for GABA uncaging experiments. Scale bars, 2% ΔG/Gsat, 50 ms. (H) Lack of correlation between ChR2-evoked inhibition for reference (n) and adjacent (n + 1) spines. (I) As in (H) for GABA uncaging.

We further characterized inhibitory compartmentalization by using photoactivation of the caged compound RuBi-GABA (11). Brief light pulses (2 ms, 473 nm) evoked IPSPs in pyramidal neurons bathed in RuBi-GABA (10.8 μM) with much smaller amplitudes but similar kinetics as those produced by stimulation of SOM-INs (fig. S6, A to C). By using GABA uncaging, we found that 51% (44/87) of randomly imaged apical spines showed significant inhibition of ΔCa2+. Inhibition was stronger in spines than in dendritic shafts (0.65 ± 0.02 versus 0.79 ± 0.03, P < 0.0001, n = 59; Fig. 2, G and fig. S6D) and was blocked by picrotoxin (fig. S6, E and F). Inhibitory compartmentalization was similar to that seen by using optical stimulation of SOM-INs (Fig. 2, D to F). Ca2+ inhibition for the reference spine was significantly greater than for the adjacent neighbor (0.64 ± 0.05 versus 0.90 ± 0.06, P < 0.01, n = 15 maps), and these values were uncorrelated (Pearson r2 = 0.11, P = 0.22, Fig. 2I).

Increasing intracellular chloride caused IPSPs to be depolarizing from a membrane potential (Vm) of –60 mV and largely eliminated inhibition of ΔCa2+ (n = 24, P < 0.0001 versus control, fig. S7, A and B), suggesting that local membrane hyperpolarization contributes to reduced Ca2+ influx. VGCCs with more depolarized activation thresholds should therefore exhibit greater sensitivity to GABAergic inhibition. Indeed, blockade of high-threshold (L and N/P/Q types) but not lower-threshold (T and R types) channels significantly reduced the amount of Ca2+ inhibition evoked by GABA uncaging (fig. S7C).

Many spines receive direct GABAergic input (5, 12, 13), and we wondered whether SOM-INs might contribute to this pool of synapses. We reconstructed apical dendrites of recorded neurons from ChR2 experiments and found 18.5% of spines (n = 1185 spines, 4 cells) expressed the inhibitory synaptic protein gephyrin, whereas 43.5% (n = 3058 spines, 9 cells) appeared to be contacted by a presynaptic bouton originating from a SOM-IN (fig. S8, A and B; see methods). For a subset of cells (n = 3), we recovered spines with corresponding Ca2+ imaging. In all cases, Ca2+ inhibition was only observed for spines with an apposed SOM-IN terminal (Fig. 3, A to C, and fig. S8, C to H).

Fig. 3 GABAergic synapses on spines mediate local Ca2+ inhibition.

(A) Confocal projection of a dendrite (red) and ChR2–enhanced yellow fluorescent protein–positive boutons (green). Scale bar, 1 μm. (B) Single-section images of spines from (A). Scale bar, 1 μm. (C) ΔCa2+ measured in spines from (B) for AP (black) and IPSP-AP (red). Scale bars, 1% ΔG/Gsat, 50 ms. (D) Inhibition mapping using ChR2 (asterisk) or 2PLUGABA (arrowheads). Scale bar, 1 μm. (E) ΔCa2+ in reference spine (n) and neighbor (n + 1) for AP (black) and IPSP-AP (red). Scale bars, 2% ΔG/Gsat, 50 ms. (F) Correlation between ChR2- and 2PLUGABA-evoked inhibition. (G) Average Ca2+ inhibition (±SEM) evoked by 2PLUGABA. (H) Computational model of dendritic inhibition. (I) Simulated GABAergic input selectively inhibits ΔCa2+ in spine 2. GABAergic input onto dendritic shaft has minimal effect on ΔCa2+. (J) GABAergic input does not inhibit Ca2+ in the dendritic shaft. Scale bars in (I) and (J), 100 nM, 100 ms. *P < 0.05 (paired Student’s t test).

We next confirmed the spatial precision of GABAergic inhibition by using diffraction-limited two-photon laser uncaging (2PLU) of 4-carboxymethoxy-5,7-dinitro-indolinyl (CDNI)–GABA (14) (methods). The Ca2+ inhibition evoked by 2PLUGABA was similar in magnitude to that produced by ChR2 activation, highly sensitive to the precise location of the uncaging spot around the spine perimeter, and isolated from neighboring spines (Fig. 3, D to G).

To see whether GABAergic synapses onto spines are necessary for compartmentalized inhibition, we simulated AP-evoked Ca2+ influx into dendritic spines and shafts (Fig. 3H, fig. S9, and methods). GABAergic input to a single spine head inhibited ΔCa2+ only in the targeted spine, whereas inhibition targeting the dendritic shaft had minimal effect on nearby spines (Fig. 3I). Moreover, ΔCa2+ in the dendritic shaft was unaffected by GABAergic input to either the spine head or shaft (Fig. 3J). Ca2+ inhibition was mediated by a compartmentalized reduction in input impedance, reducing AP amplitude in the targeted spine (fig. S9, A to D). The magnitude of inhibition was influenced by spine neck resistance, chloride reversal potential, and VGCC activation threshold but was independent of VGCC density (fig. S9, E to G). Inhibition (>15%) of dendritic ΔCa2+ could only be obtained by increasing the dendritic GABAergic conductance 10-fold (fig. S9H). A current-based inhibitory synapse that generated a similar IPSP in the absence of a conductance change produced minimal Ca2+ inhibition (fig. S9I).

Lastly, we asked whether localized inhibition occurred for synaptic Ca2+ transients and excitatory postsynaptic potentials (EPSPs). We combined one-photon RuBi-GABA uncaging with 2PLU of CDNI-glutamate [2PLUGlu (15) (methods)] and imaged ΔCa2+ in spines (Fig. 4, A to D). Inhibition of synaptic ΔCa2+ was strongly compartmentalized with no correlation between neighboring spines (0.60 ± 0.05 versus 0.98 ± 0.05 for reference spine and adjacent neighbor, respectively, P < 0.01; n = 12; Pearson r2 = 0.18; P = 0.17; Fig. 4, A to D). The 2PLUGlu-evoked EPSPs were similarly inhibited (Fig. 4, B and D), exhibiting reductions in both amplitude and duration (Fig. 4, D and E) that suggested inhibition might influence synaptic integration. GABAergic input significantly reduced the summation of responses evoked by glutamate uncaging on neighboring spines (P < 0.05, n = 11, Fig. 4, F to H). The effect on summation was eliminated after blocking N-methyl-d-aspartate–type glutamate receptors (NMDARs, n = 7, Fig. 4, G and H). Furthermore, summation was not reduced for cases where local GABAAR activation evoked an IPSP but did not inhibit spine ΔCa2+ (n = 8, fig. S10).

Fig. 4 Dendritic inhibition regulates synaptic integration.

(A) Inhibition of responses evoked by 2PLUGlu (arrowheads) using one-photon GABA uncaging (asterisk). Scale bar, 1 μm. (B) ΔCa2+ and voltage transients for EPSP (black) and IPSP-EPSP (red) for spines indicated in (A). IPSP alone has been subtracted from paired response. Scale bars, 4% ΔG/Gsat, 50 ms (top) or 0.1 mV, 5 ms (bottom). (C) Ca2+ inhibition is not correlated between neighboring spines. (D) Average ΔCa2+ and somatic voltage transients (±SEM) for EPSP alone (black) and IPSP-EPSP (red) pairing. Scale bars, 2% ΔG/Gsat, 50 ms (left) or 0.1 mV, 10 ms (right). (E) Average amplitude and duration (±SEM) of voltage transients for EPSP (black) and IPSP-EPSP pairing (red). (F) Integration of responses evoked by 2PLUGlu (arrowheads). Scale bar, 1 μm. (G) Average voltage transients (±SEM) for EPSP (black) and IPSP-EPSP (red) evoked by 2PLUGlu on three neighboring spines, recorded in control ACSF (left) or with NMDARs blocked by carboxypiperazin-4-yl-propyl-1-phosphonic acid (CPP) (right). Scale bars, 0.25 mV, 20 ms. (H) Relative summation of EPSPs (black) or IPSP-EPSPs (red), recorded in control ACSF or with CPP. *P < 0.05 (Wilcoxon matched pairs test); n.s., not significant.

Our results indicate that dendritic spines compartmentalize GABAergic inhibition, limiting both AP- and synaptically evoked Ca2+ influx and regulating NMDAR-dependent synaptic integration. These findings establish a previously unknown mechanism for the synapse-specific control of Ca2+ signaling and downstream cellular processes such as synaptic plasticity.

Theoretical studies suggested that inhibition might regulate dendritic signaling near synaptic contacts [(1618), but see (19)]. Subsequent experimental data demonstrated that GABA receptors can inhibit regenerative voltage-dependent dendritic spikes, controlling the production of AP bursts at the soma (68, 20). These findings were mediated in part by GABAB-dependent modulation of VGCCs and NMDARs (8, 21) and suggested that inhibition acts with lower spatial resolution than glutamatergic excitation, which exhibits compartmentalization of electrical and biochemical signals within single spines (22, 23). However, our data indicate that the spine head similarly restricts GABAA-mediated inhibition. The model further suggests that, in addition to the chloride reversal potential, spine neck resistance influences the efficacy of GABAergic synapses onto spine heads as occurs for glutamatergic inputs (2426). Notably, our experimental data was closely modeled by using a neck resistance of 520 Mohm, similar to the value reported for hippocampal pyramidal neurons (26). Both neck resistance and chloride reversal are modulated by development and experience (27, 28), suggesting the impact of dendritic inhibition may be similarly regulated.

Dendritic Ca2+ influx plays a key role in the induction of plasticity at glutamatergic synapses (29), and inhibition can serve as a negative regulator of plasticity (3032). Our results suggest that this control occurs at a previously unappreciated spatial scale, enabling dendrite-targeting interneurons to influence individual glutamatergic inputs. This observation is particularly relevant given the growing attention to links between perturbed GABAergic inhibition, alterations in developing neuronal circuits, and neuropsychiatric disorders such as schizophrenia and autism (33, 34).

Why do certain spines receive GABAergic inputs? One possibility is that GABA receptors are recruited by the presence of specific glutamatergic afferents, as proposed for thalamorecipient spines in frontal or visual cortex (35, 36). Additionally, recruitment of GABAA receptors might be activity-dependent (37). This hypothesis is supported by evidence that spine-targeting GABAergic inputs exhibit distinctly high rates of turnover in vivo (12, 36). Future experiments are necessary to determine the existence of feedback loops between dendritic Ca2+ signals and the formation and stabilization of GABAergic synapses.

Supplementary Materials

Materials and Methods

Figs. S1 to S10

References (3842)

References and Notes

  1. Acknowledgments: The authors thank G. M. Shepherd for discussions of spine function and P. DeCamilli, S. Tomita, J. Cardin, and members of the Higley laboratory for comments during the preparation of this manuscript. ChR2 plasmids were a gift from K. Deisseroth, Stanford University; TTA-A2 was a gift from B. Bean, Harvard Medical School; SOM-Cre mice were provided by J. Huang, Cold Spring Harbor Laboratory. The work was funded by grants from the Epilepsy Foundation (C.Q.C. and M.J.H.), the Yale Brown-Coxe Memorial Fund (G.L.), the Klingenstein Foundation (M.J.H.), the Sloan Foundation (M.J.H.), the Human Frontier Science Project (G.C.R.E.-D.), and awards from the NIH: NS011613 (N.T.C.), DC009977 (N.T.C. and T.M.M.), MH099045 (M.J.H.), and GM053395 and NS069720 (G.C.R.E.-D.).
View Abstract

Navigate This Article