The majority of γ-aminobutyric acid (GABA)ergic interneurons have smooth dendrites with no or only few dendritic spines, but certain types of spiny GABAergic interneurons do actually contain substantial numbers of spines. The explanation for such spines has so far been purely structural: They increase the dendritic surface area and thus provide the opportunity to accommodate larger numbers of synapses. We reasoned that there may be specific functional properties for these spines and therefore, undertook to characterize interneuron spines functionally. We find a remarkable similarity to pyramidal cell spines: They receive excitatory synapses with calcium impermeable α-amino-3-hydroxy-5-methyl-4 isoxazolepropionic acid (AMPA) and N-methyl-d-aspartate (NMDA) receptors, compartmentalize biochemical signals, and display activity-dependent morphological plasticity. Nevertheless, notable differences in spine density, neck length, and spine–dendrite coupling exist. Thus, dendritic spines on inhibitory interneurons have a number of important functional properties that go substantially beyond simply expanding the dendritic surface area. It therefore seems likely that spiny and aspiny interneurons may have very different roles in neural circuit function and plasticity.
Excitatory and inhibitory neurons possess a number of characteristic morphological features. One of the most prominent ones is that by and large excitatory neurons carry spines, whereas inhibitory neurons do not. This rule of thumb, however, has its exceptions. A number of inhibitory neurons also carry spines (e.g., Gulyas et al. 1992; Pitkanen and Amaral 1993; McBain et al. 1994; Kawaguchi et al. 2006; Takacs et al. 2008; Keck et al. 2011). Spines on pyramidal cells are considered a central and important component for both synaptic transmission and synaptic plasticity (Shepherd 1996; Yuste and Bonhoeffer 2001; Chklovskii et al. 2004; Yuste 2011). They compartmentalize biochemical signals (e.g., Muller and Connor 1991; Yuste and Denk 1995; Svoboda et al. 1996) contain molecular signaling complexes regulating synaptic strength (Kennedy 2000; Hering and Sheng 2001), and they increase the “reach” of the dendrite to potential presynaptic partners (Stepanyants et al. 2002). The retraction and formation of spines observed in vivo (e.g., Trachtenberg et al. 2002; Hofer et al. 2009; Xu et al. 2009; Yang et al. 2009) is considered to be a key component in the rewiring of neural circuits and thus, essential for endowing the brain with its ability to constantly adapt its function to altered environmental requirements (Chklovskii et al. 2004).
The important role that dendritic spines are thought to have for pyramidal cell function (Shepherd 1996; Chklovskii et al. 2004; Yuste 2011) would beg the question whether dendritic spines on inhibitory neurons may subserve similar or different roles and thereby, define spiny interneurons as a special interneuron class.
So far dendritic spines of spiny γ-aminobutyric acid (GABA)ergic interneurons have been characterized only morphologically (e.g., Gulyas et al. 1992; Pitkanen and Amaral 1993; McBain et al. 1994; Kawaguchi et al. 2006). Ultrastructural data have suggested that they simply serve to increase the dendritic surface area and thereby, the capacity for synapse formation without specific functionality (Gulyas et al. 1992). Recently, in vivo time lapse imaging has shown that—like in pyramidal cells—sensory deprivation enhances spine turnover also on spiny GABAergic cells (Keck et al. 2011). This suggests that spines on GABAergic interneurons, too, may play a role in circuit wiring and rewiring.
To provide a functional characterization of spines on inhibitory neurons, we used transgenic mice expressing green fluorescent protein (GFP) under control of the GAD65 promotor (Lopez-Bendito et al. 2004), the same mice in which we have previously analyzed spine turnover in spiny GABAergic interneurons in visual cortex in vivo (Keck et al. 2011). Here, we investigated spiny hippocampal interneurons, in order to compare the functional properties of their spines with the extensively characterized spines of CA1 pyramidal cells (e.g., Svoboda et al. 1996; Matsuzaki et al. 2004; Bloodgood and Sabatini 2005; Scheuss et al. 2006). Our results show that spines of these interneurons contain synapses with α-amino-3-hydroxy-5-methyl-4 isoxazolepropionic acid (AMPA) and N-methyl-d-aspartate (NMDA) receptors, can compartmentalize biochemical signals, and display morphological plasticity. Dendritic spines therefore seem to provide these spiny interneurons with similar functionality in neuronal signaling and plasticity as those in pyramidal cells.
Materials and Methods
All experimental procedures were performed in compliance with the institutional guidelines of the Max Planck Society, the local authorities (Regierung von Oberbayern), and the European Communities Council Directive of 24 November 1986 (86/609/EEC).
Brain Slice Preparation and Electrophysiology
Acute transverse hippocampal slices (300 μm thick) were prepared from 35- to 90-d-old mice in chilled dissection solution (in mM: 110 choline chloride, 25 NaHCO3, 25 d-glucose, 11.6 Na-ascorbate, 7 MgCl2, 3.1 Na-pyruvate, 2.5 KCl, 1.25 NaH2PO4, and 0.5 CaCl2) as described previously (Scheuss et al. 2006). Wild-type C57BL/6 mice or transgenic C57BL/6 mice-expressing enhanced green fluorescent protein (eGFP) under control of the GAD65 promotor (GAD65 mice, Lopez-Bendito et al. 2004) or the Thy-1 promotor (line M) (GFP-M mice, Feng et al. 2000) were used as indicated. Slices were incubated in artificial cerebrospinal fluid (ACSF) (in mM: 127 NaCl, 25 NaHCO3, 25 d-glucose, 2.5 KCl, 1 MgCl2, 2 CaCl2, and 1.25 NaH2PO4) bubbled with carbogen (95% O2 and 5% CO2) at 35°C until use. In the recording chamber, the extracellular solution was ACSF bubbled with carbogen at room temperature, unless stated otherwise. Somatic whole-cell recordings (pipette resistance, 3–4 MΩ) were performed on eGFP-expressing interneurons in CA1 or visually identified CA1 pyramidal neurons. Data were acquired via a Multiclamp 700B patch amplifier (Molecular Devices, Sunnyvale, CA, USA) controlled with custom software written in Labview (National Instruments, Austin, TX, USA). In voltage-clamp recordings, the internal solution contained (in mM): 125 CsMeSO4, 10 N-2-hydroxyethylpiperazin-N-2-ethansulfonsäure (HEPES), 10 Na-phosphocreatine, 4 MgCl2, 4 Na-ATP, 0.4 Na-GTP, 3 Na-l-ascorbate, and 30 μM Alexa-594. In the experiments about the glutamate receptor identity, the internal solution contained in addition (in mM): 5 ethylene glycol-bis(2-aminoethylether)-N,N,N′,N′-tetraacetic acid, 5 QX-314, and 0.1 spermine. The extracellular solution contained in addition (in mM): 1 Trolox, 10 tetraethylammonium chloride (TEA), 2 4-aminopyridine (4-AP), 0.2 (S)-α-methyl-4-carboxyphenylglycine (MCPG), 0.05 4-ethylphenylamino-1,2-dimethyl-6-methylaminopyrimidinium chloride (ZD 7288), 0.05 picrotoxin, 0.01 d-serine, and 0.001 tetrodotoxin (TTX) in glutamate receptor pharmacology and AMPA/NMDA colocalization experiments, and (in mM) 1 Trolox, 10 TEA, 2 4-AP, 0.2 MCPG, 0.05 ZD 7288, 0.05 picrotoxin, 0.01 3-((R)-2-carboxypiperazin-4-yl)-propyl-1-phosphonic acid (CPP), and 0.001 TTX in the AMPA rectification experiments. Firing patterns were recorded at 32°C in modified ACSF (containing in mM: 127 NaCl2, 15 NaHCO3, 1.25 NaH2PO4, 25 d-glucose, 2.5 KCl, 15 sucrose, 2 CaCl2, and 1 MgCl2). The internal solution for current-clamp recordings contained (in mM): 135 K-methylsulfonate, 10 HEPES, 10 Na-phosphocreatine, 4 MgCl2, 4 Na-ATP, 0.4 Na-GTP, 3 Na-l-ascorbate, and 30 μM Alexa-594. First, −5 mV test pulses were applied in the voltage clamp for recording the input resistance, and then 1 s long current pulses in steps of 100 pA from −300 to 600 pA for recording spiking patterns in current clamp. Chemicals and drugs were from Sigma-Aldrich (München, Germany) except for 4-AP, TTX, MCPG, ZD 7288, CPP (Biotrend, Köln, Germany); NaHCO3, NaH2PO4, NaCl (Merck, Darmstadt, Germany); HEPES, d-Glucose (Roth, Karlsruhe, Germany); 4-methoxy-7-nitroindolinyl-caged-L-glutamate (MNI-caged glutamate) (Tocris, Bristol, UK); Alexa-594 (Invitrogen, Karlsruhe, Germany); QX-314 (Alomone Labs, Jerusalem, Israel); and GYKI 53655 (Biozol, Eching, Germany).
Two-photon laser-scanning microscopy was performed with a custom microscope (objective: ×60, 0.9 numerical aperture; Olympus, Tokyo, Japan). The light beams from 2 Ti:Sapphire lasers, one for imaging (Mai Tai; Newport-Spectra Physics, Santa Clara, CA, USA) and the other for glutamate uncaging (Millennia/Tsunami; Newport-Spectra Physics), were combined with a polarizing beam splitting cube and scanned by the same scanner (Yanus IV; Till Photonics, Gräfelfing, Germany). The intensity of each beam was independently controlled with electro-optical modulators (350-80 LA; Conoptics, Danbury, CT, USA). Photomultipliers (Hamamatsu, Tokyo, Japan) recorded both epi- and transfluorescence. Image acquisition and uncaging was controlled by custom software written in Labview. Slices were screened for eGFP-positive spiny interneurons (imaging at λ = 930 nm).
Spiny interneurons and pyramidal cells were labeled with Alexa-594 by single-cell electroporation (Nevian and Helmchen 2007). A patch pipette was filled with an aqueous solution of 0.1 mM Alexa-594. The pipette was placed close to the soma. Electroporation was performed by applying one square voltage pulse (20 V, 10 ms duration, negative polarity) from an isolated stimulator (HG106, HI-Med Instruments, UK). After waiting for 5 min, 2-photon imaging stacks were acquired from the whole cell and at higher magnification from individual dendritic branches. In spiny interneurons, we included first to fourth order branches. In pyramidal cells, we restricted the analysis to apical branches (second and third order branches).
The apparent head width was measured from the fluorescence intensity profile along a line through the approximate center of the spine in the maximum intensity projections as the width of the profile with intensity >30% of the peak (Bloodgood and Sabatini 2005). Spine neck length was estimated in maximum intensity projections by measuring the linear distance from the junction with the dendritic shaft to that with the spine head (Bloodgood and Sabatini 2005). Two-dimensional (2D) spine length was measured in the maximum intensity projections as the radial distance from the dendritic shaft to the outer spine head edge. The average 3D spine length was estimated from the 2D spine length by assuming that there are no preferred directions of spine outgrowth, and considering that is given by the integral over all projections of according to , which gives .
Spiny interneurons were labeled with Alexa-555 (0.1 or 0.5 mM in water; Invitrogen, 2-photon excitation at 730 nm) by single-cell electroporation as described above in slices from 14 animals. Image stacks of 22 × 22 µm were recorded to document the spine density on a selection of the dendritic tree. Subsequently, the tissue was fixed overnight in 4% paraformaldehyde (PFA) in 0.1 mM phosphate-buffered saline (PBS) at 4°C. Slices were rinsed 3 times for 10 min each in PBS, transferred to 30% sucrose in PBS, and cut at 50 µm on a sliding microtome. The 50-µm slices containing the labeled neurons were incubated overnight in a blocking solution of 1% Triton and 10% normal goat serum (NGS) in PBS at 4°C. Primary antibodies were applied in PBS with 1% Triton and 5% NGS at 4°C overnight. After rinsing slices 3 times for 10 min each in PBS, secondary antibodies were applied at a concentration of 1 : 200 in PBS with 1% Triton and 5% NGS at 4°C overnight or for 3 h at room temperature. The following primary antibodies were used: Rabbit anti-VIP (Immunostar #20077; 1 : 500), rat antisomatostatin (Chemicon MAB354; 1:500), rabbit anticalretinin (Swant #7699/3H; 1:1000), rabbit anti-neuropeptide Y (NPY; Immunostar #22940; 1:1000), mouse anti-CCK (Dr Ohning, UCLA Cure #9303; 1:1000), rabbit anticalbindin (Swant CB-38a; 1:5000), mouse anti-GluA2/GluR2 (clone L21/32, UC Davis/NIH NeuroMab Facility; 1:100 or 1:200). Secondary antibodies were conjugated with Alexa-405 (Molecular Probes, Karlsruhe, Germany; Invitrogen), DyLight-405 (Jackson ImmunoResearch Europe, Newmarket, Suffolk, UK), or Cy5 (Dianova, Hamburg, Germany). Immunostained sections were analyzed on a confocal microscope (SP2, Leica Microsystems, Wetzlar, Germany). The spiny interneurons marked with Alexa-555 were located and image stacks (375 × 375 mm, 512 × 512 pixels; z-steps = 2 µm) on 4 channels (Alexa-405/DyLight-405, GFP, Alexa-555, and Cy5) were successively acquired. The staining of the cells of interest was only evaluated if labeling of neighboring cells indicated successful immunostaining.
Spiny interneurons and pyramidal cells were patched and loaded with 100 µM Alexa-594 and 2 mg/mL neurobiotin tracer (Linaris, Wertheim, Germany). The patch pipette was retracted after 15 min. An overview of the neuron and higher magnification images of spiny dendritic segments were acquired. Slices were transferred earliest 15 min after pipette retraction into fixative (4% PFA, 0.2% glutaraldehyde, 0.1 M phosphate buffer, pH 7.4). Slices were processed for electron microscopy following standard procedures (e.g., Knott et al. 2009). Slices were fixed at 4°C for 24–48 h. After cryoprotection in 30% sucrose (2 × 30 min), slices were freeze-thawed in liquid nitrogen for permeabilization. Endogenous peroxidase activity was quenched by incubation in 0.3% H2O2 in PBS buffer, pH 7.4 (20 min) and blocked with BSA-c (Aurion, Wageningen, Netherlands) in 0.1 M tris(hydroxymethyl)-aminomethan (Tris)-buffered saline (TBS buffer), pH 7.4 (2 × 15 min). Slices were incubated in ABC solution 1:100 in 0.1 M TBS buffer, pH 7.4 (90 min; Vectastain Elite ABC Kit, Standard, Vector Laboratories, Burlingame, USA) and reacted using diaminobenzidine (0.4 mg/mL, 0.015% H2O2 in 0.05 M TBS buffer, pH 7.4; 8–10 min, Sigma). To enhance staining contrast, slices were postfixed in 1% osmium tetroxide in cacodylate buffer, pH 7.4 (1 h), then dehydrated, and embedded in Epon (Serva Electrophoresis, Heidelberg, Germany). Semi-thin sections (2 µm) were cut (Ultramicrotome Leica Ultracut E, Leica Microsystems) to locate the stained neuron with a light microscope and to identify regions of interest (ROIs) for further analysis. Selected sections were reembedded in Epon. Ultrathin serial sections (60 nm) were cut from ROIs with an ultramicrotome (EM UC6i, Leica Microsystems). Serial sections were mounted on Formvar-coated copper grids, counterstained with lead and uranyl acetate (EM-stain, Leica Microsystems). Images were acquired with a CCD camera (Orius SC1000, Gatan, Pleasanton, CA, USA) on a Jeol JEM-1230 electron microscope (80 kV; Jeol, Tokyo, Japan). Electron micrographs were analyzed and dendritic segments reconstructed with the RECONSTRUCT software (version 220.127.116.11., http://synapses.clm.utexas.edu/tools/index.stm, last accessed 21/06/2013) (Fiala 2005).
Glutamate Receptor Identity and AMPAR Rectification
Direct stimulation of individual spines with 2-photon glutamate uncaging was performed in frame scan mode. For glutamate uncaging, the intensity of the laser with λ = 720 nm was set high (40–80 mW at the back aperture of the objective) for 0.4 ms when the beam passed the tip of the spine head (∼0.5 μm from the center of the spine head in the direction away from the dendritic shaft) and the uncaging-evoked excitatory postsynaptic current (uEPSC) was recorded from the soma.
To determine if uEPSCs are mediated by AMPA receptors (AMPAR) when interneurons were clamped at −65 mV, we recorded 10–15 uEPSC before and after application of either 10 µM NBQX or 30 µM GYKI 53655. To determine if uEPSCs are mediated by NMDA receptors (NMDAR) when interneurons were clamped at +40 mV, we recorded 10–15 uEPSC in the presence of 10 µM NBQX and after additional application of 10 µM CPP.
To explore if AMPA and NMDA receptors occur in combination, for each analyzed spine sets of 10–15 responses were recorded while the parent cell was clamped at −70, 0, and +40 mV. After changing the holding voltage, we waited for ≥1 min before resuming acquisition. In some experiments, we observed a significant inward current at nominal 0 mV holding potential indicating an offset in the pipette offset potential. We then determined the actual reversal potential, which corresponds to about 0 mV for AMPAR, and if the offset was <20 mV, continued recording at +40 mV relative to the reversal potential and scaled the −70 mV data accordingly. After baseline subtraction, 5–10 traces were averaged. For the AMPAR component, the peak amplitude of the average current at −70 mV was determined. The NMDAR component was computed as the mean of the average current at +40 mV between 20 and 40 ms after the stimulus.
In the AMPAR rectification, experiments for each analyzed spine sets of 10–15 responses were recorded while the parent cell was clamped at −65, 0, +60 or +40, and −30 mV. This sequence was chosen to control for unspecific deviations from a linear current–voltage (I–V) relationship at positive potentials. After a number of successful experiments in interneurons, cells tended to die when clamped to +60 mV, therefore we used +40 mV in subsequent experiments. After changing the holding voltage, we waited for ≥1 min before resuming acquisition. After baseline subtraction, 5–10 traces were averaged. For the I–V relationship, the amplitude and time of peak of the average currents were determined, except for the current at 0 mV. The latter was computed as the mean current over a 1-ms interval starting at the average time of peak from the currents at the other voltages. A linear I–V relationship was fit to the 3 current values at the holding voltages ≤0 mV. The rectification index was then computed from the ratio of the recorded current at positive membrane potential over the expected value from extrapolation of the I–V curve.
Fluorescence Recovery After Photobleaching and Spine Volume
Fluorescence recovery after photobleaching (FRAP) recordings of GFP were performed in the line scan mode (2.57 ms line duration) at a wavelength of 910 nm zoomed in on the spine of interest (Fig. 4A,B). The photomultiplier dark current was recorded before shutter opening. After a baseline interval of 120 ms, the excitation intensity was set high (90–110 mW in the objective back aperture) during one line to bleach the GFP fluorescence in the spine head by 20–50% and subsequently, the fluorescence recovery was recorded. An ROI around the spine head was set with boundaries at 50% of the peak of the fluorescence intensity profile along the line scans. The fluorescence for each time point (see Fig. 4A,B, lower panel) was computed by integrating the fluorescence signals of all pixels within this ROI for each scanned line. The fluorescence recovery was fit with a mono-exponential to determine the recovery time constant. Only spines were included in the analysis, which showed >90% steady-state recovery.
For measuring spine volume stacks were recorded, which contained the spines, on which the FRAP recordings were performed, and part of a thick dendrite or the soma of the same cell. Spines are below the resolution of the 2-photon microscope such that the brightness of a spine is proportional to its volume, assuming a homogenous distribution of the dye. Therefore, spine volumes were quantified by their brightest pixel normalized to the peak brightness in a large volume structure of the same cell in the maximum intensity projections after median filtering in a 5-by-5 neighborhood (Nimchinsky et al. 2004).
Induction of Spine Plasticity
Morphological spine plasticity was analyzed at 32°C in ACSF (in mM: 127 NaCl2, 15 NaHCO3, 1.25 NaH2PO4, 25 d-glucose, 2.5 KCl, and 15 sucrose) with 1 mM Trolox, 1 µM TTX, 4 mM CaCl2, nominally no MgCl2, and 2.5 mM MNI-caged glutamate. An infusion pump was used to balance evaporation of ACSF to maintain constant osmolarity. Spines located between 10 and 50 µm deep in slices from GAD65 mice for interneurons and from GFP-M mice for pyramidal cells were imaged 4 times at approximately 30 s intervals, then stimulated with 60 uncaging pulses at 2 Hz with 0.5 ms pulse duration and 75 mW laser power in the back aperture of the objective, before proceeding with imaging at 1 min intervals for 5 min and subsequently, at 5 min intervals with 2 acquisitions about 1 min apart. Control data were acquired with the same imaging protocol but without stimulation.
Data were analyzed with custom routines written in Matlab (The Mathworks, Natick, MA, USA). Data are presented as mean ± standard error of the mean (SEM) and error bars are SEM except for Fig. 5D,H. Statistical comparisons were performed with the 2-tailed Student's t-test.
Spiny Interneurons and Pyramidal Cells
We used acute hippocampal slices from transgenic GAD65-GFP mice (Lopez-Bendito et al. 2004). These express GFP under control of the GAD65 promotor in approximately 20% of GABAergic hippocampal interneurons (Lopez-Bendito et al. 2004; Wierenga et al. 2010). At postnatal day (P) 70–90, a subset of GFP-positive GABAergic interneurons in CA1 (∼20%) carry dendritic spines. We found these interneurons to have mostly multipolar morphologies (Fig. 1A) and to be located in the outer half of stratum radiatum and stratum lacunosum-moleculare. We immunostained tissue in which a GFP-positive cell had been electroporated with Alexa-555 to unequivocally reveal the spiny dendrites. These immunostainings showed that a large fraction of the spiny interneurons was reelin positive (11 of 15 cells, Fig. 1E) like the majority of GFP-expressing neurons in CA1 stratum radiatum and lacunosum-moleculare in this mouse line (Wierenga et al. 2010). A small fraction stained for vasoactive intestinal peptide (VIP, 3 of 15 cells) and for somatostatin (SOM, 2 of 15 cells). Furthermore, the majority of examined spiny interneurons were positive for NPY (12 of 15 cells, Fig. 1E) and a minority for calretinin (CR, 1 of 15 cells), calbindin (CB, 3 of 15 cells), and cholecytokinin (CCK, 3 of 15) similar to spiny interneurons in the visual cortex of this mouse line (Keck et al. 2011). Thus, NPY is expressed in a 2-fold higher proportion and the remaining markers in about similar proportions of spiny interneurons when compared with the total population of GFP-positive hippocampal interneurons in this mouse line (reelin, ∼70%; VIP, ∼15%; SOM, ∼3%; NPY, ∼40%; CCK, ∼25%; CB, ∼20%; and CR, ∼20%; Wierenga et al. 2010). We did not explore the expression of parvalbumin, because it was previously shown that parvalbumin expression is effectively absent from GFP-positive interneurons in this mouse line (Lopez-Bendito et al. 2004; Wierenga et al. 2010; Keck et al. 2011). The majority of examined spiny interneurons were strongly adapting and stopped firing prematurely before the end of current pulse injections (5 of 6 cells; resting membrane potential, −61.6 ± 4.3 mV; input resistance, 543 ± 63 MΩ; n = 5, Fig. 1F; Wierenga et al. 2010).
In young animals (P14–20), where most experiments on pyramidal cell spines were performed (e.g., Scheuss et al. 2006; Harvey and Svoboda 2007), we rarely found spiny interneurons. We therefore performed our morphological analysis in P70–90 animals, but—for comparison with the previous studies on pyramidal cell spines—conducted the electrophysiological experiments at an intermediate age of P35–45, where we still found a reasonable number of GABAergic interneurons carrying spines.
For morphological characterization, we electroporated neurons with red Alexa-594 to stain single interneurons in the densely GFP-labeled slices and to stain unlabeled pyramidal cells for comparison. Spiny GABAergic interneurons had both spiny and aspiny dendritic segments. The spine density of spiny branches (0.40 ± 0.02 spines/µm; n = 25 segments, 4 cells, total length of dendrite 801 µm; Fig. 1A) was lower than in pyramidal cells (1.27 ± 0.06 spines/µm; n = 30 segments, 5 cells, total length of dendrite 596 µm; Fig. 1B; P < 0.01; Kawaguchi et al. 2006). Interneuron spines displayed the typical morphologies of thin, mushroom, and stubby spines as described for pyramidal cells (Peters and Kaiserman-Abramof 1970). This is not to imply that there were clear-cut and distinct spine classes. In fact, the morphological characteristics—like in pyramidal cells—formed more a continuum than distinct clusters (e.g., Arellano et al. 2007; Fig. 1C). In maximum projections of 2-photon micrographs, the spine head widths were similar in interneurons (0.71 ± 0.009 µm, n = 203 spines, 4 cells; Fig. 1C, left panel, red line; Table 1) and pyramidal cells (0.66 ± 0.005 µm, n = 483 spines, 5 cells; Fig. 1C, left panel, black line; Table 1; P = 1). The fraction of spines without discernible neck in the 2-photon micrographs (stubby spines) was larger in interneurons (31%, n = 299, 4 cells; Fig. 1C, right panel, red line) than in pyramidal cells (22%, n = 721 spines, 5 cells; Fig. 1C, right panel, black line) as was the fraction of spines with long necks (Fig. 1C, right panel). Thus, the mean spine neck length excluding stubby spines was longer in interneurons (0.59 ± 0.03 µm, n = 207 spines, 4 cells) than in pyramidal cells (0.42 ± 0.01, n = 562 spines, 5 cells; P < 0.01; Table 1). The distributions of these morphological parameters were best fit by log-normal distributions, compared with normal or gamma distributions (for fit parameter see Table 1). The values for pyramidal cell spines are in the range as previously reported data obtained in the same way (Bloodgood and Sabatini 2005). For the functional characterization below, we concentrated on mushroom-like spines.
|Interneurons (IN)||Pyramidal cells (PC)|
|Spine density (spines/µm)a||0.40 ± 0.02||1.27 ± 0.06|
|Spine head width (µm)b||0.71 ± 0.009 (μ = −0.35, σ = 0.18, χ2 = 2 × 10−4)c||0.66 ± 0.005 (µ = −0.42, σ = 0.17, χ2 = 1 × 10−4)c|
|Spine length (µm)d||1.01 ± 0.02 (µ = −0.086, σ = 0.46, χ2 = 1 × 10−4)c||0.83 ± 0.01 (µ = −0.25, σ = 0.40, χ2 = 2 × 10−4)c|
|Spine neck length (µm)e||0.59 ± 0.03 (µ = −0.83, σ = 0.83, χ2 = 3 × 10−4)c||0.42 ± 0.01 (µ = −1.08, σ = 0.70, χ2 = 5 × 10−4)c|
|FRAP time constant (ms)f||162 ± 23||207 ± 16|
|Interneurons (IN)||Pyramidal cells (PC)|
|Spine density (spines/µm)a||0.40 ± 0.02||1.27 ± 0.06|
|Spine head width (µm)b||0.71 ± 0.009 (μ = −0.35, σ = 0.18, χ2 = 2 × 10−4)c||0.66 ± 0.005 (µ = −0.42, σ = 0.17, χ2 = 1 × 10−4)c|
|Spine length (µm)d||1.01 ± 0.02 (µ = −0.086, σ = 0.46, χ2 = 1 × 10−4)c||0.83 ± 0.01 (µ = −0.25, σ = 0.40, χ2 = 2 × 10−4)c|
|Spine neck length (µm)e||0.59 ± 0.03 (µ = −0.83, σ = 0.83, χ2 = 3 × 10−4)c||0.42 ± 0.01 (µ = −1.08, σ = 0.70, χ2 = 5 × 10−4)c|
|FRAP time constant (ms)f||162 ± 23||207 ± 16|
aIN: n = 25 segments, 4 cells, total length of dendrite 801 µm; PC: n = 30 segments, 5 cells, total length of dendrite 596 µm.
bIN: n = 203 spines, 4 cells; PC: n = 483 spines, 5 cells.
cParameters of fit with log-normal distribution.
dIN: n = 299 spines, 4 cells; PC: n = 721 spines, 5 cells.
eExcluding stubby spines; IN: n = 207 spines, 4 cells; PC: n = 562 spines, 5 cells.
fSame range of head volumes and neck lengths; IN: n = 23 spines; PC: n = 25 spines.
By growing out from the dendrite, spines bridge the distance to nearby axons for synapse formation and their length determines the choice of accessible axons (Stepanyants et al. 2002). We measured the radial distances from the parent dendrite to the spine tip in maximum projections of 2-photon micrograph stacks to determine the “2D spine length.” This length was about 20% longer in GABAergic interneurons (1.01 ± 0.02 µm, n = 299 spines, 4 cells; Fig. 1D, red line) than in pyramidal cells (0.83 ± 0.01, n = 721 spines, 5 cells; Fig. 1D, black line; P < 0.01). With a conversion factor from 2D to 3D spine length of 1.57 (see Materials and Methods), these spine length values are in the range of previous reports from interneurons (Kawaguchi et al. 2006) and pyramidal cells (Harris and Stevens 1989). The estimated areas of the sampled neuropil are 20–50% greater for spiny interneurons (average, 12.4 µm2 and maximal, 35.4 µm2) than for pyramidal cells (average, 8.3 µm2 and maximal, 28.6 µm2).
Synapses on Interneuron Spines
To confirm dendritic spines of spiny GABAergic interneurons contain synapses like in pyramidal cells, we filled spiny interneurons with patch pipettes containing Alexa-594 and neurobiotin, for later electron microscopical analysis. Figure 2A,B shows the reconstruction of a distal dendritic segment of a spiny interneuron and 2 examples of single ultrathin sections of labeled interneuron spines. Both have synapses with presynaptic boutons, and in both examples, one can clearly see a clustering of presynaptic vesicles. We found that 66% of identified interneuron spines carry synapses (n = 24 spines, 2 dendritic segments, and 1 cell). This compares well with previous reports on interneuron spines (Kawaguchi et al. 2006). All synaptic interneuron spines which we investigated formed a single synapse with a single bouton except for 1 spine, which had synaptic contacts to 2 boutons. The ratio of spine versus shaft synapses in the 2 reconstructed dendritic segments was 13:2 (distal, small caliper) and 5:7 (proximal, large caliper), suggesting that a significant fraction of the inputs to spiny interneurons is received on their spines.
The inputs to spiny interneuron spines appear to selectively target spines for synapse formation. We observed 10 locations where boutons forming a synapse on an interneurons spine made physical contact also with the dendritic shaft of the interneuron either directly (6 of 17) or via their parent axon (4 of 14). However, at none of these sites of shaft contact, we found a synapse (e.g. in Fig. 2C). Furthermore, the bouton contacting a spine shown in Figure 2C had a sister bouton on the same axon, which also formed a synapse on another spine of the interneuron. On the other hand, boutons forming synapses on the interneuron dendritic shaft displayed similar selectivity but for synapse formation on the shaft. We observed 4 locations where a shaft bouton itself, via a sister bouton (3 of 9) or via its parent axon (1 of 4) touched spines of the interneuron. At none of these sites was a synapse present.
Glutamate Receptor Types in Interneuron Spines
Depending on their precise class, interneurons express different types and combinations of glutamate receptors, which, among other things, determine the learning rules that govern the plasticity of their synapses (Kullmann and Lamsa 2007).
To determine the type of glutamate receptors present in synapses on spines of spiny interneurons, we used 2-photon glutamate uncaging at individual spines and recorded the uncaging uEPSCs at the soma, while clamping the neurons at −65 mV or at +40 mV (Fig. 3A,B; Busetto et al. 2008). The AMPAR component was confirmed by its complete inhibition by both the non-NMDAR antagonist NBQX (10 µM, n = 3 spines/cells; Fig. 3A, lower panel, Fig. 3B, left) and the selective AMPAR antagonist GYKI 53655 (30 µM, n = 2 spines/cells; Fig. 3B, left), which does not block kainate receptors. The NMDAR component was identified by its sensitivity to CPP (10 µM, n = 5 spines/cells), its voltage dependence, and slow time course (Fig. 3A, upper panel and Fig. 3B, right panel). At all interneuron spines tested, both AMPAR and NMDAR were present (Fig. 3C,D; n = 13 spines, 7 cells) just like in pyramidal cells (Fig. 3C,D; n = 15 spines, 6 cells) at this age (Busetto et al. 2008).
Secondly, we explored if spiny interneuron AMPAR are calcium impermeable (CI-AMPAR) as they are in pyramidal cells, or calcium permeable (CP-AMPAR) as they are in most interneurons (Freund and Buzsaki 1996; Kullmann and Lamsa 2007). To distinguish the rectifying CP- from the nonrectifying CI-AMPAR, we determined the AMPA current–voltage relationship while blocking NMDA-R with CPP (10 μM) and including spermine (100 μM) in the patch pipette to balance the dilution of endogenous polyamines during the whole-cell recording. We recorded uEPSCs evoked at single spines at voltages of −65, −30, 0, and +40 or +60 mV and found in both spiny interneurons and pyramidal cells a linear current–voltage relationship (Fig. 3E–H). The rectification index (see Materials and Methods) was in both spiny interneurons (1.06 ± 0.09, n = 12 spines, 8 cells) and pyramidal cells (1.02 ± 0.08, n = 10 spines, 4 cells), not significantly different from 1 (Fig. 3I; interneurons, P > 0.4; pyramidal cells, P > 0.7). This suggests that CI-AMPAR are expressed in spiny GABAergic interneuron spines like in pyramidal cell spines.
Interestingly, the GluA2 subunit, which in its edited form renders AMPAR calcium impermeable (e.g., reviewed in Kullmann and Lamsa 2007), induces spines when expressed in smooth interneurons (Passafaro et al. 2003). We found that the AMPAR GluA2 subunit is indeed expressed in spiny interneurons (9 of 10 cells; Fig. 3J). Unexpectedly, in the population of GFP-positive interneurons in outer stratum radiatum and stratum lacunosum-moleculare, about 60% (59 of 100 cells, Fig. 3J) were GluA2-positive although only approximately 20% appear spiny. This suggests that the GluA2 subunit might play a role in the existence of spines on these interneurons, but it appears not to be the only required factor: It may be necessary but not sufficient.
Compartmentalization Within Interneuron Spines
One hallmark of dendritic spine function is their ability to compartmentalize biochemical and electrical signals in the spine head, which is diffusionally isolated from the dendritic shaft by the thin spine neck (Muller and Connor 1991; Yuste and Denk 1995; Svoboda et al. 1996; Bloodgood and Sabatini 2005; Araya, Jiang, et al. 2006; Bloodgood et al. 2009). To compare compartmentalization in spines of spiny interneurons and pyramidal cells, we performed GFP FRAP recordings on spines (Svoboda et al. 1996) of GFP-positive spiny interneurons in slices from GAD65 mice (Fig. 4A) and spines of GFP-positive pyramidal cells in slices from GFP-M mice (Fig. 4B). For better time resolution, spine fluorescence was recorded in the line scan mode. During one line, the laser beam intensity was increased (90–110 mW at the objective back aperture) to achieve photobleaching. The mean time constant τFRAP of exponential fits to the fluorescence recovery was significantly lower by about 30% in spines of spiny interneurons (151 ± 14 ms, n = 40 spines, 9 cells) than in spines of pyramidal cells (209 ± 14 ms, n = 41 spines, 4 cells; Fig. 4C; P < 0.01), which compares well with time constants reported, for example, from measurements with photoactivatable GFP in pyramidal cells (Bloodgood and Sabatini 2005).
The fluorescence recovery time constant τFRAP depends on spine head volume V, spine neck length L, neck cross-sectional area A, and the diffusion coefficient of the fluorescent marker D according to the equation τFRAP = VL/DA (Svoboda et al. 1996). In a subset of the data of Figure 4C, we could estimate absolute spine volumes (see Materials and Methods) and spine neck length from the distance between spine head and parent dendrite in maximum projections of 2-photon micrographs. When limiting the data sets to the same range of spine head volumes and neck lengths (largest minimum to smallest maximum; volume, 0.053–0.32 a.u.; neck length, 0.070–0.87 µm), such that on average there is no difference in head volume (interneurons, 0.15 ± 0.01, n = 23; pyramidal cells, 0.17 ± 0.01, n = 25; P > 0.2) and neck length (interneurons, 0.38 ± 0.04 µm, n = 23; pyramidal cells, 0.47 ± 0.04 µm, n = 25; P > 0.1), there is no significant difference in τFRAP (interneurons, 162 ± 23 ms, n = 23; pyramidal cells, 207 ± 16 ms, n = 25; P > 0.1). In pyramidal cell spines, there was a significant correlation between τFRAP and the product of spine head volume and neck length (Fig. 4D, correlation coefficient R = 0.83, P < 0.01) as well as with each morphological parameter by itself (spine head volume, R = 0.49, P < 0.01; neck length, R = 0.40, P < 0.05; data not shown). In contrast, in interneuron spines, we did not find any significant correlation between τFRAP and spine head volume (P > 0.4) or neck length (P > 0.8) or their product (P > 0.4; Fig. 4D). Assuming that the diffusion coefficient of GFP in the cytosol is constant, this suggests that the last parameter in the equation above, the spine neck cross section, is more variable in interneurons than in pyramidal cells.
Morphological Plasticity of Interneuron Spines
In pyramidal cells, dendritic spine size is a structural signature of synaptic strength (e.g., Harris and Stevens 1989; Yuste and Bonhoeffer 2001). When single spines of pyramidal cells are stimulated with repeated 2-photon glutamate uncaging or by synaptic stimulation, their size and responsiveness to glutamate are enhanced (Matsuzaki et al. 2004; Harvey and Svoboda 2007). We explored if interneuron spines exhibit morphological plasticity similar to pyramidal cell spines.
We imaged spines on interneurons and pyramidal cells before and after stimulation with glutamate uncaging (60 pulses at 2 Hz; Fig. 5A,E) at 32°C. For the given examples, the spine size was increased by about 50% 30 min after stimulation. However, we observed such increase in spine size only in a subset of stimulated spines in both interneurons and pyramidal cells (filled circles, Fig. 5B,F). To define this subset more precisely, we performed the same time lapse imaging on control spines without the uncaging stimulus (open squares, Fig. 5B,F). While the distributions of control and stimulated spine sizes overlap significantly, there is a tail toward increased size in stimulated spines. The cumulative distributions of the size change average over 20–30 min after stimulation diverge for stimulated versus unstimulated spines at larger size changes in both interneurons and pyramidal cells (Fig. 5C,G). We defined an induced spine size change to be significant when it was greater than the mean plus 2 SD of control spine data. With this definition, 4 of the 15 stimulated spines in both interneurons (Fig. 5D) and pyramidal cells (Fig. 5H) displayed a significant increase in size. For these spines, the increase magnitude was similar in interneurons (32 ± 7%, n = 4) and pyramidal cells (37 ± 19%, n = 4; P > 0.6). This success rate and the amplitude of change are smaller than in young hippocampal pyramidal cells (Matsuzaki et al. 2004; Harvey and Svoboda 2007). Therefore, we tested also other induction protocols either reported for experiments on pyramidal cell spines in young animals (30 stimuli at 0.5 Hz, Harvey and Svoboda 2007; 60 stimuli at 1 Hz, Matsuzaki et al. 2004) or adapted from long-term potentiation (LTP) induction protocols for excitatory synapses on interneurons considering a release probability of 0.5 (30 stimuli at 1 Hz, Lamsa et al. 2005; twice 50 stimuli at 50 Hz separated by 20 s, Lamsa et al. 2007), which are also similar to LTP induction protocols reported for adult pyramidal cells (Ngezahayo et al. 2000). None of these was more efficient in plasticity induction (data not shown), suggesting that spines become more stable and resistant against plasticity induction with age as predicted by in vivo recording of spine turnover on pyramidal cells (reviewed in Holtmaat and Svoboda 2009). In conclusion, interneuron spines display morphological plasticity, which is similar to that observed in pyramidal cells.
Anatomical studies have shown that, in various species, a subset of GABAergic interneurons in the hippocampus and neocortex carry dendritic spines (Gulyas et al. 1992; Pitkanen and Amaral 1993; McBain et al. 1994; Kawaguchi et al. 2006; Takacs et al. 2008; Keck et al. 2011). Because dendritic spines are thought to be functionally important structures in pyramidal cells, it seems plausible to assume that they also serve important functions in GABAergic interneurons beyond simply increasing the dendritic surface area available for synapse formation (Gulyas et al. 1992). Indeed, just like pyramidal cells, also GABAergic interneurons show experience-dependent turnover of dendritic spines (Keck et al. 2011). The present study shows that also other key properties of pyramidal cell spines are shared by GABAergic interneuron spines: They carry synapses containing CI-AMPAR and NMDA receptors, compartmentalize biochemical signals, and display morphological plasticity. Yet, important differences also exist in spine density, spine length, and spine–dendrite coupling.
The interneuron spines of our study had the standard morphologies of thin, mushroom, and stubby spines (Peters and Kaiserman-Abramof 1970), which are more of a continuum rather than distinct classes as in most of the pyramidal cells (e.g., Arellano et al. 2007). These “classical” morphologies have previously been described in interneurons together with more elaborate structures, for example, spines with multiple swellings or branched spines with multiple heads (McBain et al. 1994; Kawaguchi et al. 2006). While interneuron and pyramidal cell spines had similar spine head dimensions, interneuron spines had on average longer necks and thus greater total length, thereby providing a potential contact with a larger choice of excitatory axons (Stepanyants et al. 2002). These morphological parameters compared well with reported data for spines of the standard morphologies in pyramidal cells and spiny interneurons (Harris and Stevens 1989; Bloodgood and Sabatini 2005; Kawaguchi et al. 2006). The late emergence of spines on the spiny GFP-labeled interneurons between P35 and P90 seems to be in contrast to the apparently completed maturation of hippocampal interneurons by the third postnatal week (Lang and Frotscher 1990). However, in the strata, where they are located, that is, in stratum lacunosum-moleculare and in the outer radiatum, it has been shown that also pyramidal cell dendrites still develop up to P90 (Pokorny and Yamamoto 1981a), and that the overall number of spines and axo-spinous synapses increases up to P48 (Pokorny and Yamamoto 1981b).
We mostly found only one synapse per interneuron spine similar to the rule of “one spine equals one synapse” for the typical spines on hippocampal CA1 and cortical pyramidal cells (Harris and Stevens 1989; Arellano et al. 2007). The ultrastructural reconstruction of 2 spiny interneuron dendritic segments showed that a significant number of synapses is formed on spines compared with the shaft. The selective targeting of either spines or the dendritic shaft by axons might reflect afferent specific innervations of spiny interneuron spines (Toth and McBain 1998).
The spiny interneurons, like pyramidal cells, expressed CI-AMPAR and NMDAR. This is different from the majority of interneurons, which express CP-AMPAR receptors and often no NMDAR (Freund and Buzsaki 1996; Kullmann and Lamsa 2007). The presence of CI-AMPAR at interneuron spines may explain why these spines exist in the first place. The AMPAR GluA2 subunit, which in its edited form is part of CI-AMPAR (Kullmann and Lamsa 2007), induces spines when expressed in smooth interneurons (Passafaro et al. 2003). We confirmed the expression of the AMPA GluA2 subunit in spiny interneurons by immunohistochemistry. The fraction of GluA2-positive cells in the total population of GFP-labeled interneurons was larger than that of spiny cell. As expected GluA2 alone is not sufficient to explain spine formation on interneurons, but other factors also appear to play a role. They may or may not be some of the regulators of spinogenesis described for pyramidal cells (reviewed in Ethell and Pasquale 2005).
The interneuron spines were also similar to pyramidal cell spines with respect to compartmentalization of biochemical signals (Muller and Connor 1991; Yuste and Denk 1995; Svoboda et al. 1996; Bloodgood and Sabatini 2005). The range of FRAP time constants as a measure of diffusional spine–dendrite coupling was similar in interneuron and pyramidal cell spines and comparable with earlier studies (Bloodgood and Sabatini 2005). This suggests also that interneuron spine necks display an electrical resistance in the range of a few up to hundreds of MΩ as reported for pyramidal cell spines (Svoboda et al. 1996; Bloodgood and Sabatini 2005; Palmer and Stuart 2009). A difference is the absence of a significant correlation between FRAP time constant, spine neck length, and head volume in interneurons, which suggests that neck resistance is more variable in interneurons and not as tightly controlled as in pyramidal cell spines (Bloodgood and Sabatini 2005; Grunditz et al. 2008).
Finally, interneuron and pyramidal cell spines displayed similar structural plasticity when stimulated with repetitive 2-photon glutamate uncaging. In both the fraction of stimulated spines, which increased in size, and the average increase in amplitude was smaller than in previous reports from pyramidal cells in young animals (P14–18, here P35–45; Matsuzaki et al. 2004; Harvey and Svoboda 2007). We tested a number of other protocols (Ngezahayo et al. 2000; Matsuzaki et al. 2004; Lamsa et al. 2005; Harvey and Svoboda 2007; Lamsa et al. 2007), and none of them proved to be more effective. Although we cannot rule out that all these protocols were suboptimal, the more likely explanation is a reduced responsiveness to plasticity induction with age. This is supported by results on in vivo spine turnover on excitatory neurons, in particular pyramidal cells, which also show a reduced fraction of plastic spines and increased fraction of stable, persistent spines with age (reviewed in Holtmaat and Svoboda 2009).
Various types of sparsely spiny and spiny inhibitory interneuron types have been described (e.g., Kawaguchi et al. 2006). The spiny GFP-labeled interneurons in our study differ from other spiny interneurons such as Martinotti and hippocampal-septal (HS) cells in that they contain reelin and NPY, whereas Martinotti cells are somatostatin-positive (Kawaguchi et al. 2006) and HS cells contain somatostatin and calretinin (Gulyas et al. 1992; Takacs et al. 2008). The cells of our study do, however, display the same molecular markers as the spiny GFP-labeled interneurons in the visual cortex of the same mouse line (Keck et al. 2011). Furthermore, the firing properties of spiny GFP-labeled interneurons (fast adapting) are different from Martinotti cells (nonfast spiking, Kawaguchi et al. 2006).
Nevertheless, we compared spine morphologies and densities between the GFP-labeled interneurons and Martinotti as well as HS cells, choosing these cells, because the morphological properties of their spines have been described in detail (Gulyas et al. 1992; Kawaguchi et al. 2006). Their spine densities are similar (spiny GFP-labeled interneurons, ∼4 spines/10 µm, this study; Martinotti cells, 2–3 spines/10 µm, Kawaguchi et al. 2006; HS cells, 4–7 spines/10 µm, Gulyas et al. 1992; Takacs et al. 2008) and the highest densities reported for any inhibitory interneuron. The types of spine morphologies are largely similar in spiny GFP-labeled interneurons and Martinotti cells, except that we did not encounter multihead spines as they are found on Martinotti cells (Kawaguchi et al. 2006). In contrast, HS cells have predominantly long, sometimes branched spines without a distinct head (Gulyas et al. 1992). Furthermore, the majority of spines on the spiny GFP-labeled interneurons carried only a single synapse, while spines of HS cells receive between 1 and 6 inputs. While the spines of spiny GFP-labeled interneurons turned out to be as efficient in biochemical compartmentalization as pyramidal cell spines, this is most likely not the case for Martinotti cell spines, which have a larger spine neck diameter (260 ± 80 nm, Kawaguchi et al. 2006) than hippocampal pyramidal cells (150 ± 60 nm, Harris and Stevens 1989). Finally, in Martinotti cells, spines are present already at high density by at least P20 (Kawaguchi et al. 2006), and in HS cells by P30 (Takacs et al. 2008). In the spiny GFP-labeled interneurons, we found effectively no spines around P20, intermediate densities around P40, and only after P70 densities comparable with those in Martinotti and HS cells. Spines on HS cells have been suggested to serve as passive structural elements, which expand the dendritic surface area to accommodate high numbers of inputs (Gulyas et al. 1992). For spines of spiny GFP-labeled interneurons in GAD65 mice, our results suggest that they may play similar active roles as spines of pyramidal cells considering their structural and functional similarity.
What then may be the special functional role of spiny interneurons? In pyramidal cells, dendritic spines are considered to support the specificity and plasticity of neuronal circuits (Chklovskii et al. 2004; Yuste 2011). Thus, in contrast to interneurons, which sample the population activity in neural circuits (Freund and Buzsaki 1996; Kerlin et al. 2010), spiny interneurons are most likely part of circuits with highly specific connectivity (Yoshimura and Callaway 2005). Furthermore, in contrast to other interneurons, which often only display limited plasticity (e.g., Nissen et al. 2010), they should—like pyramidal cells—have the ability to support major rewiring, for example, during experience-dependent plasticity (Keck et al. 2011). Spines provide potential contact with a larger choice of excitatory axons than if synapses have to be made directly on the dendritic shaft. Only a small fraction of these potential connections is utilized (Stepanyants et al. 2002), which has been suggested to reflect the specificity of the wiring as well as the ability to generate new functional connections at unused sites (Chklovskii et al. 2004). In smooth interneurons, some specificity can be achieved through tortuous dendritic branches (or axons) with appropriate bends and turns to reach their respective counterparts (Kawaguchi et al. 2006). A distinct advantage of spiny dendrites, however, is their ability to support rapid rewiring. Spine formation and retraction occur all over the dendritic tree within hours in both pyramidal cells (reviewed in Holtmaat and Svoboda 2009) and spiny interneurons (Keck et al. 2011). In contrast, a smooth dendrite would have to undergo much more massive structural rearrangement to make contact with a new set of axons. Therefore, in smooth interneurons under normal conditions and in plasticity paradigms, one observes only the extension and retraction of branch tips by tens of microns over timescales of days to weeks (Chen et al. 2011), providing only very limited plasticity.
Calcium compartmentalization in pyramidal cell spines (Shepherd 1996; Nimchinsky et al. 2002) is thought to warrant synapse-specific synaptic plasticity. Therefore, one may expect different synapse specificity in spiny and aspiny interneurons (Cowan et al. 1998). However, pathway-specific plasticity is also observed in smooth interneurons (e.g., Lamsa et al. 2005; Soler-Llavina and Sabatini 2006; Lamsa et al. 2007), but see Cowan et al. (1998). While fast CP-AMPAR-mediated calcium signals are effectively compartmentalized also in shaft synapses (Goldberg et al. 2003; Soler-Llavina and Sabatini 2006), the underlying mechanism in NMDAR-containing synapses remains unknown (Cowan et al. 1998; Lamsa et al. 2005; Lamsa et al. 2007). This does, of course not preclude that spines in inhibitory interneurons play an important role as compartments for other signaling molecules, in particular those regulating the observed structural spine plasticity (Murakoshi and Yasuda 2012).
The electrical properties and functions of pyramidal cell spines remain controversial (e.g., Araya, Eisenthal, et al. 2006; Araya, Jiang, et al. 2006; Palmer and Stuart 2009; Gulledge et al. 2012). In interneurons, it has been argued that dendritic spines are detrimental for rapid electrical signaling, which is considered an essential component of interneuron function (e.g., Jonas et al. 2004; Klausberger and Somogyi 2008). Thus, there might be a tradeoff in spiny versus aspiny interneurons between the specificity and flexibility of their connectivity and their speed of electrical signaling.
In conclusion, dendritic spines appear to be universal neuronal structures serving in general as flexible synaptic connection elements, which rather than being a distinguishing feature of pyramidal cells and interneurons, that is, excitatory and inhibitory neurons, reflect the specific function of their parent neurons in neural signaling and plasticity (Hübener et al. 1990; Ascoli et al. 2008).
This work was supported by the Max Planck Society, the Deutsche Forschungsgemeinschaft (SFB 870), and the Human Frontier Science Program (V.S.).
The transgenic GAD65 mice were kindly provided by Gábor Szabó (Budapest, Hungary). We thank Moritz Helmstaedter, Tobias Rose, and Rafael Yuste for comments on the manuscript, and Volker Staiger, Claudia Huber, Ursula Weber, and Marianne Braun for expert technical assistance. Conflict of Interest: None declared.