Dendrodendritic Inhibition in the Olfactory Bulb Is Driven by NMDA Receptors

The slow time course of the dendrodendritic IPSC is determined by the asynchronous release of GABA from granule cells

To activate dendrodendritic synapses in mitral cells, a bipolar electrode was placed on the surface of the slice in the glomerular layer, where olfactory nerve axons contact the primary dendrites of mitral cells (Fig. 1A). Mitral cells were voltage clamped at −70 mV with KCl-containing patch pipettes. To allow full expression of glutamate receptors that might be involved in the activation of dendrodendritic synapses, we first examined synaptic responses in magnesium-free saline. Maximal glomerular stimulation (100 μsec; 100 V) elicited a large, slowly decaying inward whole-cell current. The current was completely blocked by bicuculline (Fig. 1A; 50 μm;n = 7) or picrotoxin (50 μm;n = 2) but not by the GABA_B receptor antagonist 2-hydroxysaclofen (100 μm; n = 3), indicating that the current was mediated by mitral cell GABA_A receptors. The slow IPSCs had a peak amplitude of 1.4 ± 0.3 nA and a decay time constant of 0.41 ± 0.05 sec (n = 18). The IPSC amplitude was proportional to stimulus intensity (n = 3; data not shown). Given the 200 μm tip separation of the stimulating electrode and the diameter of a single glomerulus (130 μm) (Scott et al., 1993), it seems likely that maximal stimulation activated the ∼25 mitral cells that project to a single glomerulus (see Scott et al., 1993). TTX (1 μm) blocked the IPSC, suggesting that normal afferent pathways were activated by simulation.

The prolonged IPSCs we observed are consistent with activation of reciprocal dendrodendritic synapses between mitral and granule cells (Rall et al., 1966; Nowycky et al., 1981; Jahr and Nicoll, 1982). However, activation of granule cells by mitral cell axon collaterals (Shepherd and Greer, 1990) could contaminate the dendrodendritic IPSC. To test this possibility, we recorded IPSCs in response to direct somatic stimulation of a single mitral cell in the presence and absence of TTX (Jahr and Nicoll, 1982). A 2 msec depolarization to 0 mV applied to the mitral cell evoked an action potential that was followed by a flurry of brief inward synaptic currents (Fig. 1B). The synaptic currents were completely blocked by bicuculline (n = 8) or picrotoxin (n = 3). The ensemble average of a series of single mitral cell depolarizations produced a composite IPSC (Fig. 1B) that was smaller than the IPSC evoked in the same cell by glomerular stimulation (73 ± 4% reduction; n = 17) but that had the same decay time course (τ, 0.46 ± 0.05 vs 0.41 sec;n = 21). TTX (1 μm) blocked the IPSC elicited by a 2 msec depolarization (n = 7), but longer somatic depolarizations (10–500 msec) evoked IPSCs (1.2 ± 0.4 nA) that did not differ from the IPSC before the addition of TTX (1.1 ± 0.2 nA). The TTX-insensitive IPSCs are likely to be dendritic in origin because mitral cell dendrites should be much more effectively depolarized than are axons by somatic voltage injections. The TTX-insensitive IPSCs, taken together with morphological data showing that only a subset of mitral cells have axon collaterals in the bulb (Orona et al., 1984), suggest that axon collaterals made little contribution to the IPSCs under our conditions.

The decay of the dendrodendritic IPSCs was much slower than was the kinetics of commonly observed stimulus-evoked GABA_Areceptor-mediated IPSCs (e.g., Edwards et al., 1990). The long duration of the IPSC did not seem to represent prolonged deactivation kinetics of the GABA_A receptors because, as noted above, single mitral cell stimulation revealed discrete synaptic events with rapid decay time courses. In five cells, the decay time constant of the unitary IPSCs was 18 ± 1 msec (Fig. 1C), which is similar to that of IPSCs observed in other pathways. Likewise, direct stimulation of granule cells with a patch pipette elicited a monophasic IPSC in mitral cells that occurred with a short latency (time-to-peak from stimulus, 5.8 and 7.7 msec; n = 2 cells) and with rapid decay (τ, 15 and 22 msec; n = 2 cells). These results indicate that the slow kinetics of the dendrodendritic IPSC elicited by mitral cell stimulation reflects the prolonged and asynchronous release of GABA from granule cells.

NMDA receptors, and not AMPA receptors, mediate GABA release from granule cells

The slowly decaying IPSCs in mitral cells are driven by the excitatory synaptic response in granule cells. Most central excitatory synapses have both AMPA and NMDA receptors. Granule cells also express multiple AMPA and NMDA receptor subunits (Keinänen et al., 1990;Petralia and Wenthold, 1992; Watanabe et al., 1993; Petralia et al., 1994). However, bath application of the NMDA receptor antagonistd,l-AP-5 (50–100 μm) completely abolished the IPSC elicited by glomerular stimulation (Fig.2A,B), whereas the IPSC was insensitive to the AMPA receptor antagonist NBQX (10 μm). The IPSCs typically displayed multiple kinetic components (Fig. 2A), but NBQX did not affect any of these components. The less-selective AMPA receptor antagonist CNQX (10 μm) produced a 24 ± 5% reduction in the IPSC charge (n = 7), consistent with its role as a glycine antagonist at NMDA receptors (Lester et al., 1989). A similar glutamate receptor antagonist profile was observed for the IPSC elicited by single mitral cell stimulation (Fig. 2C). Because single mitral cell stimulation bypasses the activation of the olfactory nerve–mitral cell synapse, the complete block by d,l-AP-5 indicates that NMDA receptors located at dendrodendritic synapses are an absolute requirement for activation of IPSCs in mitral cells.

Because EPSCs mediated by NMDA receptors typically persist for hundreds of milliseconds (McBain and Mayer, 1994), the duration of the granule cell EPSC may be responsible for the long duration of the dendrodendritic IPSC. To test this possibility, we examined the effect of the noncompetitive NMDA receptor antagonist MK-801, which can reduce the duration of NMDA receptor-mediated EPSCs by ∼50% (Rosenmund et al., 1993). For IPSCs elicited by single mitral cell stimulation, MK-801 (10 μm) reduced the IPSC amplitude while producing a 23 ± 12% (n = 10) acceleration of the decay time constant. The faster IPSC kinetics suggests that the kinetics of the granule cell NMDA receptor-mediated EPSC contributes to the long duration of the dendrodendritic IPSC.

AMPA and NMDA receptors are present at dendrodendritic synapses

Although AMPA and NMDA receptors generally are colocalized, there are several examples of “pure” NMDA receptor synapses (Dale and Roberts, 1985; Liao et al., 1995; Durand et al., 1996; Wu et al., 1996; O’Brien et al., 1997). Although granule cells have been reported to have both AMPA and NMDA components in their excitatory responses (Trombley and Shepherd, 1992; Wellis and Kauer, 1994), the disparate effects of NMDA and AMPA receptor antagonists on mitral cell IPSCs could imply that granule cells have many more NMDA receptors than AMPA receptors. To test this possibility, we recorded granule cell EPSCs in response to focal stimulation of a nearby mitral cell (Fig.3A). The EPSCs had a short latency and displayed a fast component that was completely blocked by CNQX (10 μm; τ = 5.5 ± 1.2 msec;n = 7), as well as a slow component that was blocked byd,l-AP-5 (50 μm). The slow component was well described by two exponentials with time constants of 52 ± 10 and 343 ± 48 msec (n = 9). Glomerular stimulation typically elicited short-latency EPSCs with multiple peaks, presumably reflecting asynchrony in the release of glutamate from mitral cells (data not shown). In two cells, the mean latency to the first peak after stimulation was 6.7 and 7.0 msec, whereas the latency to all peaks averaged 27 and 17 msec. The fast components of these EPSCs were clearly blocked by NBQX (n = 4 cells). The peak amplitude of the NMDA component of the focally evoked EPSC was smaller than that of the AMPA component (Fig. 3B;I_NMDA/I_AMPA = 0.26 ± 0.05; n = 24), whereas the relative current amplitudes were similar to those of other excitatory synapses (Forsythe and Westbrook, 1988; Hestrin et al., 1990; Silver et al., 1992; Jonas et al., 1993). Thus, the effectiveness of the NMDA receptor in driving dendrodendritic IPSCs is not simply because of a larger number of functional NMDA receptors.

Although the above results indicate that granule cells have both AMPA and NMDA receptors, it is possible that the receptors are segregated such that dendrodendritic inhibition is driven by pure NMDA receptor synapses, whereas the AMPA component of the evoked EPSC could reflect contributions from mitral cell axon collaterals or centrifugal fibers. Such a segregation of synapses should be apparent in mEPSCs. However, the small amplitude of pure NMDA receptor mEPSCs makes them difficult to detect (Bekkers and Stevens, 1989). To circumvent this problem, we compared evoked EPSCs with mEPSCs selected using the rapid kinetics characteristic of AMPA receptor-mediated mEPSCs. As shown in Figure 3C (left), mEPSCs selected in this way showed a prominent NMDA receptor-mediated slow component. If the evoked EPSC reflects a composite of different populations of synapses, including pure NMDA receptor synapses, the ensemble average of these mEPSCs should have a smaller NMDA component than does the evoked EPSC because our mEPSC selection criteria excluded pure NMDA receptor-mediated mEPSCs. However, the averaged mEPSCs and the evoked current had the same decay time course (Fig. 3C;n = 2), suggesting that most synapses on one granule cell have a similar complement of AMPA and NMDA receptors.

NMDA receptor activation causes stronger granule cell excitation than AMPA receptor activation

At conventional central synapses, the depolarization caused by AMPA receptor-mediated EPSPs drives the short-latency response of the postsynaptic cell. The failure of AMPA receptors to activate dendrodendritic inhibition, despite the presence of typical dual-component EPSCs, suggests that granule cells process incoming excitatory signals in a distinctive manner. One possibility is that the short duration of the AMPA receptor-mediated EPSC limits the voltage response of granule cells. We examined this issue directly in current-clamp recordings in granule cells using potassium gluconate-containing patch pipettes. The amplitude and duration of the evoked EPSP as well as the fraction of stimuli that elicited action potentials were used as measures of the effectiveness of incoming mitral cell activity.

Glomerular stimulation (100 V; 100 μsec), identical to that used to record IPSCs in Figure 1, evoked dual-component EPSPs (Fig.4A,B) with large peak amplitudes and slow decay kinetics (18 ± 2 mV;t_½ = 204 ± 24 msec;n = 21). d,l-AP-5 (50 μm) had a variable effect on the peak EPSP amplitude, with an average reduction of 32 ± 18% (n = 7), but blocked stimulus-induced action potentials in all cells tested (Fig.4A,C). In some cells, the peak of the remaining AMPA EPSP component was as large as that of the dual-component EPSP, but the AMPA component nevertheless failed to elicit spiking [Fig. 4A,B(top)]. d,l-AP-5 reduced the duration of the EPSPs by 62 ± 8% (n = 7), leaving an AMPA component that had a decay time constant similar to the membrane time constant (τ = 62 ± 15 and τ_m = 59 ± 14 msec, respectively; n = 12), as expected for the voltage response to a fast-decaying current input. In contrast, NBQX (10 μm) did not affect the peak amplitude or duration of the EPSP. Furthermore, the remaining NMDA receptor-mediated EPSP was just as effective at eliciting action potentials as was the dual-component response (Fig. 4C). These results indicate that AMPA receptors are less effective at eliciting granule cell excitation than are NMDA receptors, despite their similar amplitudes.

We next considered whether the longer duration of the NMDA receptor-mediated EPSP was critical for granule cell excitation. One clue to this possibility was their long latency to action potential firing. Even with suprathreshold dual-component depolarizations, there was a pronounced delay (72 ± 14 msec; n = 7 cells) before the onset of action potentials, as seen in the responses to five stimuli in Figure 4A. This suggests that the AMPA receptor-mediated EPSP was not of sufficient duration to elicit an action potential, although the peak amplitude of the AMPA component in this cell was “suprathreshold.” Short somatic current injections (2–5 msec; 50–100 pA; n = 3) were similarly ineffective. The depolarizations induced by these current pulses were larger than those elicited by long current injections (400–500 msec; 10–20 pA) that were effective in generating spikes. Thus, the granule cell membrane appears to have an intrinsic shunt that prevents action potential firing in response to a short-duration current input.

Although the complete block of EPSP-evoked action potentials in granule cells by d,l-AP-5 matched the block of the dendrodendritic IPSCs in mitral cells, this correlation does not imply that NMDA receptor-driven action potentials are required for dendrodendritic inhibition. For example, as discussed above, IPSCs could be elicited by single mitral cell stimulation in the presence of TTX. However, natural odorant stimulation does elicit action potentials in granule cells (Wellis and Scott, 1990), as did the same glomerular stimulation that elicited the IPSCs in our experiments. In 11 of 18 granule cells, glomerular stimulation drove action potentials in at least 20% of the trials. Interestingly, dendrodendritic inhibition evoked by both natural stimuli and glomerular stimulation involves both reciprocal and lateral components. Action potentials should markedly augment the dendritic spread of a voltage signal in the granule cell, which may be particularly important for lateral inhibition (Woolf et al., 1991;Scott et al., 1993).

Dendrodendritic inhibition in the presence of extracellular magnesium

Our experiments thus far have shown that NMDA receptor activation is required for dendrodendritic inhibition. However, no magnesium was added to the extracellular bath in these experiments. Magnesium causes a potent voltage-dependent block of the NMDA receptor channel (Mayer et al., 1984; Nowak et al., 1984), but the addition of magnesium (30–1000 μm) did not eliminate the mitral cell IPSC evoked by glomerular stimulation (Fig.5A,B). Both the amplitude and duration of these IPSCs were reduced by 1 mm Mg²⁺; the remaining charge was 23 ± 4% of control (n = 5). Furthermore, dual-component granule cell EPSPs in 1 mmMg²⁺ were large (15 ± 2 mV; n= 26) and frequently elicited action potential firing in granule cells (Fig. 4C). Thus, granule cell excitation and subsequent mitral cell inhibition in response to glomerular stimulation remain intact in the presence of physiological concentrations of magnesium.

Extracellular magnesium could nevertheless change the relative effectiveness of AMPA and NMDA receptors in mediating dendrodendritic inhibition. However, d,l-AP-5 (50 μm) caused a nearly complete (89 ± 4%) block of mitral cell IPSC in 1 mm Mg²⁺ (Fig. 5C;n = 11), with similarly large effects on granule cell excitation (Fig. 4C). The small remaining current had a rapid decay time course (τ = 6.8 ± 2.6 msec; n= 3), suggesting that it was attributable to an AMPA receptor-mediated EPSC at the olfactory nerve–mitral cell synapse. In contrast to Mg²⁺-free conditions, NBQX (10 μm) caused a modest reduction in the IPSC (34 ± 9%;n = 18), as well as reductions in granule cell excitability (Fig. 4C). The effect of NBQX on the IPSC was, however, highly variable, as shown for two cells in Figure5C. There was a modest negative correlation (r = −0.44; n = 18) between the magnitude of the NBQX-induced block of the IPSC and the amplitude of the IPSC (Fig. 5D). These results imply that dendrodendritic inhibition in the presence of Mg²⁺ relies completely on the activation of granule cell NMDA receptors, whereas AMPA receptors play a facilitatory role by causing relief of the magnesium block of NMDA receptors.

The negligible effect that NBQX had on IPSCs in many cells, even in magnesium, implies that NMDA receptor activation is capable of eliciting IPSCs in the absence of AMPA receptor activation. This result was somewhat surprising given that at most central synapses, 1 mm Mg²⁺ blocks nearly all of the NMDA receptor-mediated EPSC at voltages near the granule cell resting potential (−66 ± 2 mV; n = 32). However, some recombinant NMDA receptors have a reduced sensitivity to magnesium (seeMcBain and Mayer, 1994). To test the magnesium sensitivity of granule cell NMDA receptors, we measured granule cell EPSCs across a wide voltage range in response to focal mitral cell stimulation in the absence and presence of magnesium. The degree of block of the NMDA component attributable to 1 mm Mg²⁺, as estimated from the current at 30–40 msec after stimulus, was voltage-dependent, averaging 44 ± 5% at −28 mV (n = 5), 78 ± 5% at −48 mV (n = 3), and 96 ± 2% at −78 mV (n = 3). This block is very similar to that of the highly magnesium-sensitive receptors containing NMDAR-2A and 2B subunits (Monyer et al., 1994), as well as to the magnesium sensitivity of EPSCs at other synapses (Hestrin et al., 1990). The convergence of many mitral cells onto single granule cells, however, may provide a mechanism for relieving the magnesium block of the NMDA receptor during glomerular stimulation. Given the high input resistances of granule cells (0.98 ± 0.17 GΩ; n = 12), the activation of several mitral cell inputs should provide sufficient current through NMDA receptors to depolarize the granule cell. Consistent with such a mechanism, the block of mitral cell IPSCs elicited by glomerular stimulation was much less than that of IPSCs induced by single mitral cell stimulation (Fig.5B). These results imply that NMDA receptors can drive dendrodendritic inhibition not because of a low intrinsic magnesium sensitivity but because of the position of the granule cell within the circuitry of the olfactory bulb.

MK-801 blocks dendrodendritic inhibition in vivo

Although the above experiments provide convincing evidence that NMDA receptor activation is required for dendrodendritic inhibition in brain slices, it is difficult to mimic precisely natural stimulation conditions in vitro. For example, the stimulus strength and other network properties of the bulb could conceivably influence the impact of voltage-dependent NMDA receptors on dendrodendritic inhibition. Thus, we examined the effects of NMDA receptor blockade on the activation of olfactory bulb circuits in 21-d-old male rats using the immediate early gene c-fos as a measure of cellular activation. Because c-fos is induced by a rise in intracellular calcium (Sheng et al., 1990), its induction has been widely used as an assay for studying the excitability of populations of neurons within many different brain regions (Morgan and Curran, 1991). Although c-fos mRNA expression is not a linear function of synaptic activity or action potential generation (Fields et al., 1997), it can be used to compare the relative activation of different cell groups. In particular, if NMDA receptors are required for dendrodendritic inhibition, then block of these receptors should remove inhibition and result in increased c-fos mRNA expression in mitral cells.

As seen in the low-power autoradiographs in Figure6, the basal c-fos levels of rats exposed to clean air in a laminar flow apparatus (“ambient”) were low with occasional glomeruli labeled, as well as patchy labeling of the granule cell layer. We then exposed rats to the general odorant isoamyl acetate (IAA) that induces c-fos in periglomerular and granule cells throughout the bulb (Guthrie et al., 1993). After a 5 min exposure to a 1:10 dilution of IAA, there was intensec-fos mRNA expression in scattered glomeruli and broad expression throughout the granule cell layer. In rats pretreated with MK-801 (1 mg/kg, i.p.), a ring of increased c-fos expression corresponding to the mitral cell layer was apparent under ambient air conditions, similar to that observed by Wilson et al. (1996). In addition, there was a rather uniform increase in c-fos mRNA expression in the glomerular and granule cell layers. This pattern of increased activity was even more striking in rats injected with MK-801 and exposed to IAA (Fig. 6, lower right). The expression in the granule cell layer obscured separation of the mitral cell layer at this magnification. A similar pattern was observed after injection with the competitive NMDA antagonist CPP (n = 2; data not shown).

To examine c-fos mRNA expression within cell groups, we analyzed the sections shown in Figure 6 at higher magnification. Figure7 shows bright-field and dark-field views of thionin-stained, emulsion-dipped sections, along with profile plots of mean intensity values along the horizontal axis of the dark-field image. Consistent with the low-magnification views, the sections from rats exposed to ambient air and not treated with MK-801 had low levels of c-fos mRNA expression in granule cells and periglomerular cells. In IAA-stimulated rats, glomeruli were surrounded byc-fos mRNA-expressing periglomerular cells, as were granule cells in the superficial half of the granule cell layer. MK-801 produced a striking activation of cells in the mitral layer both in ambient air and after IAA stimulation. Additionally, cells in the external plexiform layer, most likely tufted cells, show higher expression of c-fos mRNA in MK-801. MK-801 increased the absolute level of granule cell activation as well as mitral cell activation, but this was to be expected because blocking granule cell NMDA receptors should relieve granule-cell-to-granule-cell inhibition (Wellis and Kauer, 1994). The ratio of the mean intensity values of the mitral cell layer to a layer of granule cells deep to the internal plexiform layer confirmed that MK-801 produced a significant increase in mitral cell c-fos expression. MK-801 increased the mitral/granule intensity ratio from 0.86 ± 0.12 to 1.36 ± 0.16 in ambient air and from 0.47 ± 0.01 to 1.03 ± 0.06 in odor-stimulated animals (n = 4, all groups). These data indicate that NMDA receptors play a key role in dendrodendritic inhibition in vivo.

Is the NMDA receptor dependence a general property of dendrodendritic synapses?

NMDA receptors were required for triggering GABA release from granule cells, whereas AMPA receptor activation by itself was functionally ineffective. We considered several factors that could have influenced the dependence of dendrodendritic IPSCs on NMDA receptor activation in our experiments including the stimulus strength, the effects of extracellular magnesium, and the age of the animals. However, these factors did not affect the basic result. For example, in experiments done in the absence of extracellular magnesium, stimulation of a single mitral cell or many mitral cells by glomerular stimulation produced IPSCs that were completely blocked by d,l-AP-5. As expected, extracellular magnesium reduced the NMDA response in granule cells, but the remaining glomerular stimulation-evoked IPSC was still completely blocked by d,l-AP-5. On the other hand, NBQX reduced the IPSC only in magnesium-containing solutions. The magnesium dependence of the NBQX sensitivity indicates that an AMPA receptor-mediated depolarization does not trigger GABA release but instead facilitates the NMDA response by reducing magnesium block. For technical reasons, we used slices from young rats (P9–P16). Granule cell maturation and synaptogenesis are still ongoing during this period (Rosselli-Austin and Altman, 1979), and NMDA receptors may precede AMPA receptors at some synapses (Durand et al., 1996; Wu et al., 1996). However, granule cell EPSCs had both AMPA and NMDA components (see also Trombley and Shepherd, 1992; Wellis and Kauer, 1994). Likewise, the disinhibition of mitral cell c-fos expression by MK-801 was actually more prominent at 21 d than at 12 d (data not shown), indicating that the critical role of NMDA receptors extends at least to the young adult.

We were initially surprised that the IPSCs were completely dependent on NMDA receptor activation, although previous results did suggest that NMDA receptors are involved in dendrodendritic inhibition. For example, α-aminoadipate, a somewhat selective NMDA antagonist, completely blocked dendrodendritic IPSCs in the turtle olfactory bulb (Nicoll and Jahr, 1982). Also, Wilson et al. (1996) found that MK-801 increased basal c-fos expression in mitral cells in rats exposed to clean air. However, incomplete block of mitral cell IPSCs by AP-5 was observed by Wellis and Kauer (1993) in the salamander olfactory bulb. Whether species differences are significant in this regard remains unclear. Recently, Isaacson and Strowbridge (1998) reported incomplete block of mitral cell IPSCs by AP-5 in the rat bulb. However, these authors measured TTX-insensitive IPSCs in response to prolonged mitral cell depolarizations using CsCl-filled patch pipettes. The properties of glutamate release under these conditions may differ from that in our experiments.

Dendrodendritic synapses compared with conventional synapses

The critical role of NMDA receptor activation in dendrodendritic inhibition differs fundamentally from the properties of most excitatory synapses. At conventional axosomatic or axodendritic synapses, postsynaptic AMPA receptor activation is generally sufficient to drive short-latency firing of action potentials, whereas NMDA receptors require coincident depolarization or a train of depolarizations to relieve magnesium block and become fully activated. The requirement for ongoing electrical activity fits well with a modulatory role of NMDA receptors (Bliss and Collingridge, 1993). In contrast, at dendrodendritic synapses, NMDA receptors are necessary for information transfer, whereas AMPA receptors serve a facilitatory role.

Conventional axodendritic synapses do exist in which NMDA receptor activation is critical for information transfer, for example, in the thalamus, spinal cord, and cerebellum (Salt, 1986; Dickenson and Sullivan, 1990; D’Angelo et al., 1995). Dendrodendritic synapses are unusual, however, in that they are critical for action potentials as well as inhibition evoked by a single stimulus, whereas a train of stimuli are required for NMDA receptor-mediated action potentials at other synapses. The basis for this difference may be the anatomical convergence of a large number of mitral cells onto single granule cells, allowing a buildup of an NMDA response in granule cells during a single stimulus. Dendrodendritic NMDA receptors may also integrate responses to a stimulus train, as seen at other synapses.

Despite the specialized anatomical features of dendrodendritic synapses, many functional aspects of synaptic transmission appear similar to other central synapses. For example, focal mitral cell stimulation elicited fast-deactivating AMPA receptor-mediated EPSCs in granule cells (Fig. 3A), implying that glutamate release is brief, as observed at other fast-acting synapses (Clements et al., 1992). Focal granule cell stimulation also elicited fast-decaying IPSCs in mitral cells, suggesting that mitral cell GABA_Areceptors have conventional kinetics and, moreover, that the mechanics of GABA release from granule cell dendrites are conventional, given the appropriate stimulus.

What is special at dendrodendritic synapses—the NMDA receptor or the granule cell response?

Based on the properties of the granule cell EPSCs, the number of NMDA receptors or their magnesium sensitivity did not explain their preferential activation of dendrodendritic inhibition. NMDA receptors also have a high calcium permeability (Mayer and Westbrook, 1987;Ascher and Nowak, 1988), whereas most AMPA receptors including those found in bulb interneurons (Jardemark et al., 1997) have a low calcium permeability (Hollman and Heinemann, 1994). The proximity between granule cell glutamate receptors and GABA release sites on dendrodendritic synapses raises the possibility that calcium influx through NMDA receptors directly triggers GABA release. This mechanism would, however, be inconsistent with the calcium sensitivity of exocytosis at other synapses, where vesicle fusion occurs in response to >100 μm calcium within 100 nm of voltage-gated calcium channels (Llinas et al., 1992; Zucker, 1993). In contrast, GABA release sites on granule cell dendritic spines are up to 1 μm from the postsynaptic density (Price and Powell, 1970).

However, NMDA receptors could be located at presynaptic terminals of granule cells, as reported at some axon terminals (Berretta and Jones, 1996). Moreover, some central synapses display a slow component of transmitter release, perhaps mediated by highly calcium-sensitive protein(s) responding to residual calcium (Goda and Stevens, 1994). In principle, residual calcium derived from the NMDA receptor could mediate such a response. Although we cannot exclude the NMDA “calcium hypothesis,” we favor the possibility that the calcium that triggers GABA release is derived from voltage-gated calcium channels that are preferentially opened by an NMDA receptor-mediated depolarization. The failure of TTX to block mitral cell IPSCs evoked by single mitral cell stimulation implies that action potentials are not required for this release. In the absence of action potentials, the voltage in an activated spine would be expected to be near 0 mV (Koch and Poggio, 1983). Because such a depolarization is less than an action potential, a longer EPSP might be required to account for slower calcium channel-opening kinetics.

Because of the lag to action potential generation in granule cells, a prolonged NMDA receptor-mediated EPSP was also required for spiking, which might be particularly critical for lateral inhibition. The firing lag raises the possibility that an inactivating potassium conductance in the granule cell, like that reported in some other neurons (Storm, 1988; Hoffman et al., 1997; Rusznák et al., 1997), may shape the granule cell response. Thus, the intrinsic properties of the whole granule cell membrane might be better matched to a slow NMDA depolarization than to the brief AMPA receptor-mediated EPSP.

Implications for olfactory processing

The long duration of dendrodendritic IPSCs and their dependence on NMDA receptors have important implications for sensory processing in the olfactory bulb. Although the c-fosexperiments are qualitative, they do provide a means to evaluate dendrodendritic inhibition in response to natural stimuli. The MK-801-induced enhancement of c-fos expression in mitral cells is consistent with a reduction of dendrodendritic inhibition. Although we did not test the effects of AMPA receptor antagonistsin vivo, the disinhibition observed in MK-801 indicates that granule cell AMPA receptors drive less dendrodendritic inhibition than does NMDA receptor activation. MK-801 did not prevent the odorant-induced increase in c-fos expression in granule cells, indicating that AMPA receptor-mediated depolarizations resulted in sufficient calcium entry to activate c-fos mRNA (<1 μm peak [Ca²⁺]; Fields et al., 1997); however, this calcium may not be sufficient to trigger GABA release. Only a subset of glomeruli were activated by isoamyl acetate in control animals, but there was a generalized activation of glomeruli after MK-801. The loss of spatial specificity suggests that NMDA receptor activation is critical for lateral inhibition in the glomerular layer (see also Guthrie et al., 1993). The uniform activation by odorant of the granule cell layer after MK-801 also suggests that elements of both reciprocal and lateral inhibition depend on NMDA receptors.

The properties of the NMDA response seem well adapted to influence sensory processing in the olfactory bulb. For example, their voltage dependence may insure that lateral inhibition occurs only from mitral cells that respond strongly to an odor. Spontaneously active mitral cells or those that respond weakly to an odor may not provide sufficiently robust excitation of granule cells to relieve the magnesium block of the NMDA receptor. The slow kinetics of the NMDA receptor-mediated EPSC is a major determinant of the duration of dendrodendritic IPSCs. Prolonged IPSCs are also expected to augment the spatial signal-to-noise ratio. Mitral cells can fire spontaneously at 10–20 Hz (e.g., Harrison and Scott, 1986); thus, their effective suppression requires inhibition lasting ≥100 msec. Prolonged IPSCs might also more effectively suppress odor-induced activity in other mitral cells, whose odor-induced spiking can be delayed or asynchronous (Hamilton and Kauer, 1989). The time course of the reciprocal inhibitory output might also be critical for the synchronized oscillatory pattern observed in odor responses (Tank et al., 1994) that have been shown in honeybees to be essential for odor discrimination (Stopfer et al., 1997).