Specialized Inhibitory Synaptic Actions Between Nearby Neocortical Pyramidal Neurons

See allHide authors and affiliations

Science  04 May 2007:
Vol. 316, Issue 5825, pp. 758-761
DOI: 10.1126/science.1135468


We found that, in the mouse visual cortex, action potentials generated in a single layer-2/3 pyramidal (excitatory) neuron can reliably evoke large, constant-latency inhibitory postsynaptic currents in other nearby pyramidal cells. This effect is mediated by axo-axonic ionotropic glutamate receptor–mediated excitation of the nerve terminals of inhibitory interneurons, which connect to the target pyramidal cells. Therefore, individual cortical excitatory neurons can generate inhibition independently from the somatic firing of inhibitory interneurons.

In the mammalian brain, neurons integrate synaptic inputs onto their somatodendritic domains, which control the generation of action potentials propagating through the axonal arbor to axon terminals, at which signals are transmitted to postsynaptic neurons. Action potential–dependent transmitter release from axon terminals is modulated by ionotropic glutamate and γ-aminobutyric acid (GABA) receptors that are present, either synaptically or extrasynaptically, on the axon terminals (1, 2).

We used dual whole-cell recording under microscopic observation to study synaptic connections from pyramidal and nonpyramidal neurons to nearby (<75 μm) pyramidal neurons in layer 2/3 of the mouse visual cortex (3). Single action potentials in a pyramidal neuron could produce inhibitory postsynaptic current (IPSC)–like outward currents in another pyramidal neuron held at the reversal potential (0 mV) of excitatory postsynaptic currents (EPSCs) (Fig. 1A). These currents were evoked by individual action potentials with relatively constant latencies that were comparable to those seen in monosynaptic connections (Fig. 1A). The responses were abolished by bath application of the GABA type A (GABAA) receptor antagonist bicuculline methiodide (BMI) (20 μM) and after reversal could then be abolished again by application of the non–N-methyl-d-aspartate (NMDA) glutamate receptor antagonist 2,3-dioxo-6-nitro-1,2,3,4-tetrahydrobenzo[f]quinoxaline-7-sulphonamide (NBQX) (10 μM, n = 6 neuron pairs), indicating that they were polysynaptic IPSCs (Fig. 1B).

Fig. 1.

Comparison of ip IPSCs with unitary EPSCs and IPSCs. (A) Action potentials elicited in a pyramidal cell produced IPSCs, but not EPSCs, with a constant latency in another nearby pyramidal cell. The traces show superimposed (n = 5) spikes elicited by depolarizing voltage pulses in the source cell (top) and superimposed (middle, n = 5) and average (bottom, n = 50) postsynaptic currents recorded from the target cell at the holding potential (Vh) of 0 mV (left) and –70 mV (right). The right confocal image shows the pair of pyramidal neurons stained with biocytin. Histograms plot the distribution of IPSC latency (left) and amplitude (right) in the pair. (B) ipIPSCs were abolished by BMI (20 μM) and NBQX (10 μM). (C and D) Similar to (A), but for uEPSCs between a pair of pyramidal cells (C) and uIPSCs from a nonpyramidal cell to a pyramidal cell (D). (E to G) Comparison of amplitude, latency, and coefficient of variation (CV) of latency in ip IPSC (ip IP), uEPSC (uEP), and uIPSC (uIP). Symbols (triangles and circles) and horizontal bars indicate values in individual pairs and mean values, respectively. Asterisks indicate that values are significantly different from those for ip IPSCs (P < 0.05).

To further characterize these interpyramidal IPSCs (ip IPSCs), we compared them with two kinds of monosynaptic currents: unitary EPSCs (uEPSCs) recorded from pyramidal neuron pairs at the reversal potential (–70 mV) of IPSCs (Fig. 1C) and unitary IPSCs (uIPSCs) from nonpyramidal neurons to pyramidal neurons (Fig. 1D). In pyramidal neuron pairs, the probability of detecting an ip IPSC [28%; 31 out of 110 (31/110)] was slightly higher than that for the detection of an EPSC (22%; 24/110). NBQX blocked ip IPSCs in all of the tested pairs (n =27). Reciprocal interpyramidal inhibitory connections were never observed. Six pairs had both inhibitory and excitatory connections. The direction was the same for three pairs and opposite for three pairs. In these pyramidal neuron pairs, patch pipettes containing a Cs+-based internal solution were used for recording from both pre- and postsynaptic neurons. Similar ip IPSCs were also recorded with a K+-based internal solution (fig. S1). Recordings from pairs involving an inhibitory neuron and a pyramidal neuron had a detection probability for uIPSCs of 32% (19/60), which was slightly higher than that for ip IPSCs.

The amplitudes of ip IPSCs were significantly larger (P < 0.01) than those of uIPSCs (Fig. 1E), and their time course was similar to that of uIPSCs (Fig. 1, A and D, and fig. S2). Although the average latency of ip IPSCs was significantly (P < 0.02) longer than that of either uIPSCs or uEPSCs, it was distributed in a wide range that included latency values for the two monosynaptic connections (Fig. 1F). If ip IPSCs resulted from conventional polysynaptic activation involving action-potential generation at the somata of inhibitory neurons, the expected variation in latency for each pair should be far larger than that in monosynaptic connections. However, the coefficient of variation of their latency was indistinguishable (P > 0.2) from those for either uEPSCs or uIPSCs (Fig. 1G), suggesting that they were unlikely to be mediated by the generation of somatic action potentials in inhibitory interneurons. Consistent with this supposition, the failure rate of ip IPSCs was not significantly different (P > 0.1) from that of uIPSCs or uEPSCs (fig. S3). This interpretation is also supported by the observation that unitary excitatory inputs alone induce only small postsynaptic responses that are subthreshold for action-potential generation in inhibitory interneurons (47). Thus, we hypothesized that ip IPSCs are generated by direct excitation of the presynaptic terminals of inhibitory neurons, which in turn connect to the target pyramidal neuron (Fig. 2A). This mechanism implies that the axo-axonic synaptic transmission must be strong enough to release GABA immediately from the inhibitory terminals. If this synaptic transmission is very strong, extraordinarily quick depolarization would occur at the terminals because of their small volume and lack of strong filtering effects on input signals seen in dendrites. This may, at least in part, explain the short latency of ip IPSCs, together with the absence of conduction time in interneurons. We tested this hypothesis, as described below.

Fig. 2.

Hypothesized interpyramidal inhibitory connection and presence of AMPA and kainate receptors on presynaptic inhibitory terminals. (A) Schematic illustration of interpyramidal inhibitory connection from pyramid A to pyramid B. (B) Glutamate (10 μM) decreased the interval but not the amplitude of mIPSCs. The traces show mIPSCs recorded from a pyramidal cell before (top) and during (bottom) glutamate application in the presence of TTX (1 μM). The histograms plot cumulative probability distributions of amplitude and inter-event intervals of mIPSCs for the cell before (interrupted line) and during (solid line) glutamate application. (C) Similar to (B), but NBQX (10 μM) increased the interval but not the amplitude of mIPSCs. (D and E) Summary of the effects of glutamate (n = 7 cells), AMPA (1 μM, n = 7), domoic acid (200 nM, n = 6), ATPA (1 μM, n =7), and NBQX (n = 7) on mIPSC amplitude (D) and interval (E). Open and striped bars indicate values before and during the application of agents, respectively. Asterisks indicate that these values are significantly different from control values (P < 0.05). Error bars indicate SEM.

If such excitatory axo-axonic synapses are present, the frequency of miniature IPSCs (mIPSCs) recorded from pyramidal cells in the presence of 1 μM tetrodotoxin (TTX), a sodium channel blocker, may be affected by glutamatergic agents. Bath application of glutamate (10 μM) significantly (P < 0.02) increased the frequency of mIPSCs without any significant (P > 0.6) changes in their amplitude (Fig. 2, B, D, and E). Similar facilitative effects were produced by the selective activation of AMPA receptors with AMPA (1 μM) and kainate receptors with (RS)-2-amino-3-(3-hydroxy-5-terf-butylisoxazol-4-yl) propanoic acid (ATPA) (1 μM) (8) or a low dose (200 nM) of domoic acid (9), suggesting that both AMPA and kainate receptors contribute to the facilitation of mIPSC frequency (Fig. 2, D and E). We confirmed this supposition with a pharmacological blockade of these receptors (fig. S4). The effect of these receptors may be mediated by the depolarization of nerve terminals, because the facilitation of mIPSC frequency was not found in the presence of Co2+, which blocks voltage-gated Ca2+ channels, and because the metabotropic action of kainate receptors was not involved in this process (fig. S5).

The application of NBQX significantly (P < 0.02) reduced the frequency of mIPSCs without any significant (P > 0.9) changes in their amplitude in control solution (Fig. 2, C, D, and E), indicating that the basal level of extracellular glutamate facilitates mIPSC frequency. The activation of AMPA and kainate receptors on inhibitory presynaptic terminals increases or decreases evoked-IPSCs in various connections (1018). It is likely that inhibitory synaptic transmission is not affected by the basal level of extracellular glutamate in our experimental condition because NBQX application produced no change in uIPSCs (fig. S6).

Excitatory synaptic transmission seems to be mediated mostly by AMPA receptors in neocortical inhibitory interneurons (6). We confirmed that AMPA (but not kainate) receptors mediated excitatory synaptic transmission from pyramidal to nonpyramidal neurons (fig. S7). Thus, an investigation of glutamate receptors involved in ip IPSCs may provide important information on the validity of our hypothesis. We conducted additional dual whole-cell recordings from pyramidal neurons. In all of the tested pairs with latencies shorter than 3 ms (n = 6 pairs), ip IPSCs were not affected by the AMPA receptor antagonist SYM 2206 (30 μM) (19, 20), but they were reduced substantially by the subsequent application of SYM 2081 (1 μM, n = 6), a kainate receptor ligand that causes potent receptor desensitization (19, 21) (Fig. 3A and fig. S8A). In two cases of these pair recordings, the responses were completely abolished by further application of UBP 301 (50 μM), a kainate receptor antagonist (22).

Fig. 3.

Non-NMDA receptors on inhibitory terminals mediate ip IPSCs. (A) Short-latency (2.2 ms) ip IPSCs were not affected by SYM 2206 (30 μM), but they were greatly reduced by the subsequent application of SYM2081 (1 μM). The traces show average (n = 50) spikes in a source cell (top) and average (n = 50) postsynaptic currents in a target cell held at 0 mV (bottom). (B) An example of long-latency (5.2 ms) ip IPSCs, which were completely abolished by SYM 2081. (C) An example of long-latency (5.6 ms) ip IPSCs, which were blocked by SYM 2206. (D) Experimental arrangements of stimulation (s) and recording (r) electrodes and an NBQX-containing (10 μM) patch pipette used in the experiments shown in (E) to (H). (E) The upper and lower traces show superimposed average evoked IPSCs (n = 5) before (black) and during (red) bath application of 10 μM NBQX and those scaled to the same amplitude, respectively. (F and G) Similar to (E), but for local NBQX application from a patch pipette placed near [<5 μm in (F)] and far from [50 to 70 μm in (G)] the soma of the recorded cell. (H) NBQX-induced reduction of IPSC amplitude for bath application (n = 6 cells) and local application from near (N, n = 5) and far (F, n = 6) pipettes. Asterisk indicates that values are significantly (P < 0.05) different from those for bath application. Error bars indicate SEM. (I) Experimental arrangements for iontophoretic application of glutamate receptor agonists in the presence of TTX (1 μM). (J) The traces show raw current responses recorded from pyramidal neurons at 0 (left) and –70 mV (right) in response to glutamate (Glut), AMPA, kainate, and ATPA application. Arrows indicate the onset of iotophoretic application. Time and current calibrations are common to all traces.

In the pairs with latencies longer than 5 ms (n = 10 pairs), most ip IPSCs were completely abolished by SYM 2081 (n = 5) (Fig. 3B) or UBP 301 (n = 3) (fig. S8B). In the remaining two pairs, ip IPSCs were partially blocked by SYM 2081 (about half of the control amplitude), and the remaining responses were completely abolished by the subsequent application of SYM 2206. In addition, long-latency ip IPSCs were completely abolished by the application of SYM 2206 alone (n = 6) (Fig. 3C and fig. S8C). These results strongly suggest that ip IPSCs with short latencies are mediated mostly by kainate receptors, whereas those with long latencies are mediated by both AMPA and kainate receptors to a comparable degree. The consistent contribution of kainate receptors to ip IPSCs strongly supports the intervention of axo-axonic synapses in these responses. Although SYM 2081 incompletely blocked short-latency ip IPSCs, it completely blocked most long-latency ip IPSCs (Fig. 3, A and B, and fig. S8), suggesting that the latency depends, at least in part, on the number of kainate receptors implicated.

To further test for the possibility that ip IPSCs are mediated by the excitation of inhibitory terminals in close proximity to the postsynaptic pyramid, we used a paradigm that allowed local glutamatergic blockade, far from the likely location of the dendrites of inhibitory neurons that might mediate the IPSCs (Fig. 3D). Monosynaptic-like IPSCs were evoked in layer-2/3 pyramidal neurons by stimulation with bipolar metal electrodes placed in layer 4. They were decreased by bath application of NBQX (10 μM) without substantial changes in their time courses, as shown in the superimposed traces of control and reduced IPSCs, scaled to the same amplitude (Fig. 3E). Thus, these IPSCs probably comprise a mixture of ip IPSCs (arising from antidromic activation of layer-2/3 pyramidal neurons) and uIPSCs (due to direct input from inhibitory interneurons). Brief pressure application of NBQX (10 μM) from another patch pipette placed near (<5 μm) the soma of the recorded neuron produced similar reductions in IPSCs, whereas the same NBQX application far from the soma (50 to 70 μm) failed to produce such reductions (Fig. 3, E to H), suggesting that most non-NMDA receptors mediating ip IPSCs are located in close proximity to the cell body of the target pyramidal neuron. This observation further implicates inhibitory basket cells as likely mediators of at least some ip IPSCs (23, 24).

The hypothesis that excitatory axo-axonic synapses mediate ip IPSCs would be strengthened if iontophoretic application of glutamate to inhibitory nerve terminals onto pyramidal cells induces IPSCs in the presence of TTX, which localizes glutamate-induced depolarization to the applied area (Fig. 3I). The application of glutamate, at high intensities of ejection currents, to the soma of pyramidal cells produced slow inward currents at –70 mV, presumably mediated by AMPA and kainate receptors on the soma of recorded pyramidal cells, but it produced large outward currents at 0 mV (Fig. 3J and fig. S9). The outward currents had a much faster time course than did the inward currents (P < 0.03) (Fig. 3J and fig. S9). In addition, glutamate application at low intensities of ejection currents evoked asynchronous outward currents exhibiting a fast time course. These currents were comparable to mIPSCs, but their amplitudes were much larger than those of mIPSCs (fig. S10). Thus, glutamate-evoked synchronous outward currents are considered to be compound IPSCs, resulting from the simultaneous release of GABA from multiple inhibitory nerve terminals quickly depolarized by glutamate. This suggests that glutamatergic transmission at the GABAergic terminal is very effective. Similar outward currents were also induced by AMPA (n = 5 cells), kainate (an agonist for both AMPA and kainate receptors, n =5), and ATPA application (n = 5), suggesting that they are mediated by both AMPA and kainate receptors (Fig. 3J and fig. S9). AMPA (n =5), kainate (n = 5), or ATPA-induced outward currents (n =5) were abolished by bath application of BMI without any reduction in inward currents, and inward and outward currents were both abolished by NBQX (fig. S11). These observations strongly support our hypothesis about ip IPSCs.

Finally, we attempted to demonstrate morphologically, using mechanically dissociated neurons from slices, that GABAergic terminals on the soma of layer-2/3 pyramidal neurons have non-NMDA receptors and that they are apposed by glutamatergic synaptic terminals. Dissociated cells had a pyramidal-like shape, and mIPSCs could be measured (n = 4) (Fig. 4A), indicating that some inhibitory synaptic terminals remained attached to the soma and dendrites. Iontophoretic application of kainate to the soma of these neurons induced outward and inward currents, similar to those observed in slices (n = 3) (Fig. 4B). Thus, we used this preparation for immunocytochemical staining.

Fig. 4.

Glutamatergic synapses exist on GABAergic terminals presynaptic to pyramidal cells. (A) mIPSCs (Vh = 0 mV) recorded from a mechanically dissociated layer-2/3 pyramidal neuron (upper picture, differential interference image). (B) Bath application of BMI (20 μM) abolished kainate-evoked outward currents (Vh = 0 mV), and NBQX (10 μM) application abolished both inward (Vh = –70 mV) and outward currents in a dissociated pyramidal neuron. Arrows indicate the onset of iontophoretic application. (C and D) Immunocytochemical demonstration of GluR5-containing kainate receptors on GABAergic terminals surrounding the soma of a dissociated neuron. Confocal microscopic single-section images demonstrate colocalization of GAD65/67 with synaptophysin (Syn) (C) or GluR5 (D). (E) GAD65-positive terminals were often adjoined by VGluT1-positive terminals. In (C) to (E), the left pictures show fluorescent images for GAD65/67 (red) and the other proteins (green), and the right (upper and lower) pictures show differential interference contrast images and merged images of the left two images, respectively. Arrows indicate colocalization of two proteins [(C) and (D)] and GAD65 apposed by VGluT1 (E). (F) Percentages of coexpression of synaptophysin with GAD65/67 (GAD/Syn), coexpression of GAD65/67 with GluR5 (GluR5/GAD), and GAD65-positive terminals adjoined by VGlut1-positive terminals (VGluT1/GAD). The numbers of neurons examined were 11 (C), 17 (D), and 14 (E), respectively. Error bars indicate SEM.

Staining for synaptophysin, a presynaptic terminal marker, showed bouton-like structures (presumably presynaptic terminals) surrounding the soma of neurons (Fig. 4C). Double staining for synaptophysin and the GABA synthetic enzymes glutamic acid decarboxylase 65 and 67 (GAD65/67) showed that >50% of synaptophysin-positive terminals expressed GAD65/67, confirming the presence of GABAergic terminals on the dissociated neuron (Fig. 4, C and F). GAD65/67 was absent inside of the dissociated neurons, consistent with the morphological supposition that they were pyramidal neurons. We used an antibody against glutamate receptor 5 (GluR5) as a marker of non-NMDA receptors because the application of ATPA, activating GluR5-containing kainate receptors selectively (8), produced compound IPSCs (Fig. 3J). Double staining for GAD65/67 and GluR5 demonstrated that ∼40% of GABAergic terminals expressed GluR5 receptors (Fig. 4, D and F). Double staining for GAD65 and VGluT1, a vesicular glutamate transporter located at pyramidal neuron nerve terminals (25, 26), demonstrated that more than half of the GAD-positive terminals adjoined VGluT1-positive terminals (Fig. 4, E and F), consistent with an immunocytochemical study in the rat somatosensory cortex (27). These morphological data suggest that glutamate released from the axon terminals of a pyramidal neuron quickly activates non-NMDA receptors on inhibitory nerve terminals that connect to the soma of other pyramidal neurons. In addition, axons of pyramidal neurons stained with biocytin in slice preparations were often found on the soma of presumed pyramidal neurons. Those axons had multiple synaptic terminal-like boutons, apposed by GAD-positive terminals surrounding neurons lacking GAD inside their soma (fig. S12). This anatomical feature of connections may, at least partly, explain the large amplitude of ip IPSCs. However, it is uncertain whether the axo-axonic transmission is mediated by synapses with a structure commonly demonstrated by electron microscopy (28).

We demonstrated that layer-2/3 pyramidal neurons, sending output signals to other cortical areas, exert strong inhibitory effects on nearby pyramidal cells via the direct activation of nerve terminals of inhibitory interneurons, bypassing their somatodendrite domain (Fig. 2A). Our analysis suggests that the activation of inhibitory terminals is mediated by axo-axonic signaling, although we cannot rule out partial involvement of dendro-axonic signaling (29). This inhibition may play a crucial role in the regulation of the cortical output signal. According to traditional views, pyramidal neurons receive inhibitory inputs via action potentials initiated in inhibitory interneurons after integration of synaptic inputs to their somatodendritic domain. Thus, synaptic transmission from inhibitory nerve terminals to layer-2/3 pyramidal cells is driven by two distinct signaling pathways: (i) via an integration of feedforward and feedback signals in inhibitory interneurons (30) and (ii) more directly via output signals of nearby pyramidal cells. The presence of interpyramidal inhibition suggests that the functional influence of inhibitory neurons can be far greater than might be predicted by their relatively small numbers (∼20% of cortical neurons) (23, 31).

Supporting Online Material

Materials and Methods

Figs. S1 to S12

Table S1


References and Notes

Stay Connected to Science

Navigate This Article