Two GABAergic Intraglomerular Circuits Differentially Regulate Tonic and Phasic Presynaptic Inhibition of Olfactory Nerve Terminals

GAD65-expressing neurons are PG cells

The classic view that glomeruli are surrounded by a uniform population of inhibitory interneurons has been supplanted by the recognition that JG cells are morphologically and neurochemically heterogeneous with complex synaptic interconnections that are only partially understood (Brinon et al. 1999; Halasz et al. 1985; Kosaka et al. 1997, 1998; Parrish-Aungst et al. 2007; Rogers 1992; Toida et al. 1998, 2000). The largest population of JG (JG) neurons, ≥55% of all mouse JG cells, are GABAergic interneurons (Parrish-Aungst et al. 2007), which are further subdivided with 42% expressing GAD65 only, 29% GAD67 only, and 30% expressing both isoenzymes. Other essentially nonoverlapping JG cell populations express calbindin, calretinin, neurocalcin, and other markers (Parrish-Aungst et al. 2007). To investigate one of these JG cell classes, we used a transgenic mouse in which GFP is expressed under the control of the GAD65 promoter (GAD65-GFP mice). Thus while these cells are homogeneous for GAD65, there will be some heterogeneity for other neurochemicals (e.g., 40% of GAD65+ cells should co-express GAD67).

Whole cell patch-clamp recordings were obtained from 137 JG neurons in OB slices; 114 cells expressed GFP (Fig. 1, A and B), and 23 were GFP negative. GAD65-positive cells were filled with biocytin and 20 reconstructed to reveal their 3D morphology (Fig. 1, C and D). GAD65+ cells have a small soma size, 7.8 ± 0.28 μm diam; range 6–11 μm; equivalent to 50.53 ± 3.82 μm² soma surface area (Parrish-Aungst et al. 2007). Reconstructed GAD65+ cells had 2.5 + 0.3 primary dendrites and a total dendrite length of 803 ± 91 μm. These dendrites project a relatively short distance from the soma ramifying in a restricted subregion of a single glomerulus with occasional branches entering a neighboring glomerulus (∼50% of cells contained these side branches that comprise <20% of their total dendritic length and never formed tufts). The distributions of branches as a function of distance from the soma (Scholl analysis) revealed that for 50% of GAD+ cells their longest dendrite extends only 55–70 μm; only 10% extend a dendrite >90 μm from the soma (Fig. 1G). GAD65+ cells had numerous branch points (36.4 ± 5.0), terminal dendrite endings (40.9 ± 5.1) and spines (mean 22.9 ± 3.4, density 29 ± 5.5 spine/mm). The dendrites were relatively convoluted as measured by average tortuosity, the ratio of the actual length of each dendrite segment to the straight-line length between each end of the segment (1.38 ± 0.02). Taken together, these morphological properties are consistent with the classic definition of PG (PG) cells (Cajal 1911; Hayar et al. 2004a; Pinching and Powell 1971a), thus we conclude that GAD65+ JG neurons are PG cells.

GAD65+ PG cells had a resting membrane potential of −57 ± 1.0 mV (range: −45 to −73 mV; n = 45 cells), input resistance of 673 ± 53 MΩ (range: 235–1,786; n = 85 cells), and a small hyperpolarizing activated I_h current −19.9 ± 2.8 pA (range: 0 to −80; n = 44; measured by a step from −60 to −110 mV; not shown). I_h current amplitude was calculated as the difference between the instantaneous and the steady–state current (Yao et al. 2003). All GAD65+ PG cells were capable of generating action potentials; 60% (45/75 cells) generated spontaneous action potentials at rest. The frequency of spontaneous action potentials was 1.9 ± 0.3 Hz, (range: 0.1–8 Hz; n = 45 cells). Depolarizing current evoked action potentials in the remaining 40% “silent” PG cells and further depolarization caused regular firing (Fig. 1E). Rarely, GAD65+ PG cells responded with a sustained plateau potential to brief intracellular injection of depolarizing current (5/75 cells; 6.6%; not shown). Spontaneous burst firing, a defining characteristic of ET cells (Hayar et al. 2004a,b, 2005), was never observed in GAD65+ PG cells.

We compared the GAD65+ PG cells with GAD65-negative cells in the same mice. Of the recorded GAD65-negative cells, 23 were selected representing examples of the major cell types in the glomerular layer (ET, SA, and PG); 10 exhibited spontaneous bursts of action potentials and had a morphology consistent with ET cells (Cajal 1911; Hayar et al. 2004a,b, 2005; Pinching and Powell 1971a), 4 had the morphology of SA cells (Aungst et al. 2003; Price and Powell 1970), and 9 were similar in morphology to GAD65+ PG cells. Thus while all GAD65+ neurons are PG cells, not all PG cells express GAD65.

GFP+ and GFP-negative PG cells did not differ with respect to resting membrane potential (GAD65+ = −57 ± 1.0 mV; GAD65 negative = −57 ± 2.3 mV), input resistance (GAD65+ = 673 ± 53 MΩ; GAD65-negative = 735 ± 136 MΩ), or I_h current (GAD65+ = −19.9 ± 2.8 pA; GAD65-negative = −19.7 ± 7.5 pA). All GAD65-negative PG cells generated action potentials on intracellular injection of depolarizing current; 56% (5/9) had spontaneous action potentials at rest (Fig. 1F). The spontaneous frequency of the five GAD65-negative cells was 3.2 ± 1.0, which was not different from the GAD65+ PG cell firing rate. Thus GAD65-positive and -negative PG cells have many similar physiological properties.

Spontaneous synaptic input patterns define two populations of GAD65+ PG cells

Rat PG cells recorded without regard to neurochemical phenotype comprised two groups differing in patterns of spontaneous synaptic input: bursts of spontaneous EPSCs (sEPSCs) versus single, isolated sEPSCs (Hayar et al. 2004b). This same difference was present in mouse GAD65+ PG cells and was analyzed here in greater detail. Spontaneous EPSCs were recorded from 56 randomly selected GAD65+ PG cells and spontaneous activity analyzed over 2–5 min recording epochs. sEPSCs were detected on the basis of amplitude and area having ≥3:1 signal-to-noise ratio.

A burst of sEPSCs was defined as previously (Hayar et al. 2004b) to be a minimum of four EPSCs separated by <15-ms intervals. As the frequency of sEPSCs increases, the probability that four sEPSCs occur at this frequency by chance increases (methods). To be classified as a cell receiving bursts of sEPSCs, observed bursts had to exceed chance by 2.5 SD (i.e., 99% confidence). Based on this criterion, 67% of GAD65+ PG cells (37/56) exhibited recurring bursts of sEPSCs as well as isolated, single sEPSCs (Fig. 2, A and B) and the remaining 33% (19 of 56) do not exhibit bursts of sEPSCs (Fig. 2, A and B). Overall burst sEPSC PG cells received 14.0 ± 1.8 sEPSC/s and single sEPSC PG cells received 3.4 ± 0.4 sEPSC/s. We examined sEPSC kinetics for single-sEPSC versus burst-sEPSC cells (n = 12 single-sEPSC and n = 12 burst-sEPSC cells). There was a modest difference in the decay time constant between single-sEPSC (4.2 + 0.3 ms) and ETd (3.2 + 0.3 ms) cells but no significant changes in 10–90% rise time (2.1 + 0.5 ms for burst-sEPSC vs. 2.0 + 0.2 ms for single-sEPSC). For burst-EPSC cells, we also examined isolated sEPSCs versus those that are part of a burst. Isolated sEPSCs and sEPSCs that were part of a burst were statistically indistinguishable in 10–90% rise time and decay constant.

We considered the possibility that these differences in synaptic properties (sEPSC input rate and/or single/burst patterns) are due to differing degrees of dendritic or axonal truncation in the slice. The morphology of these cell types presented in greater detail in the following text, but the results indicate that differences in dendrite truncation do not account for these differences in synaptic properties. To have approximately equal numbers of burst and single sEPSC PG cells for further analyses, we recorded additional cells to obtain a total of 56 GAD65+ PG cells that received bursts of sEPSCs (“burst-sEPSC cells”) and 58 that had only single sEPSCs (“single-sEPSC cells”; Fig. 2C).

Patterns of spontaneous synaptic input predict ON-evoked responses

The finding that GAD65+ PG cells differ sharply with respect to their profiles of spontaneous EPSCs suggest that burst-EPSC cells (67% of the population) and single EPSC PG cells (33% of the population) receive different sources of synaptic input and participate in different circuits.

Another class of JG neuron, the ET cell, generates spontaneous bursts of action potentials and provides monosynaptic excitatory input to a subpopulation of PG cells (Hayar et al. 2004a,b, 2005; Liu and Shipley 2008). Thus it is likely that the spontaneous bursts of EPSCs that define the burst-sEPSC PG cells derive from ET cells. ET cells also generate bursts of action potentials in response to monosynaptic input from the ON. Thus if burst-sEPSC PG cells receive sensory inputs via a disynaptic ON→ET→PG circuit they should respond to ON stimulation with a burst of EPSCs.

To investigate this set of hypotheses, we analyzed the responses of 56 burst-sEPSC and 58 single-sEPSC PG cells to ON stimulation (Fig. 3). We used the minimum stimulus intensity necessary to elicit EPSCs in ≥95% of trials, i.e., minimum effective stimulus (MES). The MES for single-sEPSC PG cells (201 ± 30 μA) and for burst-sEPSC PG cells (184 ± 22 μA; P = 0.38) did not differ statistically. ON stimulation evoked bursts of EPSCs in all burst-sEPSC cells (mean: 6.8 ± 0.6 EPSC/stimulation) and never evoked bursts of EPSCs in single sEPSC cells (1.6 ± 0.1 EPSC/stimulation; Fig. 3, A–C). Thus a GAD65+ PG cell's pattern of spontaneous synaptic input predicts its pattern of ON-evoked responses.

We also analyzed the latency and “jitter” (the SD of the latency) of responses to ON stimulation. Across the entire GAD65+ PG cell population latencies ranged from 1.8 to 8.2 ms (mean: 3.61 ± 0.22 ms; Fig. 3D) and jitter ranged from 30 to 2,616 μs (mean: 470 ± 70 μs; Fig. 3E). However, cells with spontaneous and ON-evoked single EPSCs had significantly lower latency (2.63 ± 0.06 ms, range: 1.79–4.08 ms) and jitter (173 ± 9 μs, range: 31–298 μs) than cells with spontaneous and ON-evoked EPSC bursts (latency: 4.62 ± 0.3 ms, P < 0.000001, range: 2.16–8.20 ms; jitter 774 ± 119 μs, P < 0.000001; range: 309–2,616 μs).

ET cells provide monosynaptic burst input to ∼70% of PG cells (Hayar et al. 2004b). Thus the burst-sEPSC PG cell responses to ON input may derive from the disynaptic, ON→ET→PG circuit. However, mitral/tufted (M/T) cells could also contribute to this burst response: Back propagation of action potentials initiated near the M/T cell soma into the apical dendrites might trigger transmitter release onto PG cells. To explore this possibility, we made a cut through the external plexiform layer separating M/T cell apical dendrites from their cell bodies and then analyzed spontaneous and ON-evoked responses in nine PG cells. There were no significant differences in PG cell sEPSCs from cut or intact slices. In cut slices, six of the nine cells exhibited bursts of sEPSCs (analyzed from 2- to 3-min recording epochs as in the preceding text) and three exhibited single sEPSCs. Overall, the six burst-sEPSC cells in cut slices received 19.6 ± 4.4 sEPSC/s compared with 14.0 ± 1.8 sEPSC/s in intact slices; the three single-sEPSC cells in cut slices received 2.1 ± 0.4 sEPSC/s compared with 3.4 ± 0.4 sEPSC/s in intact slices. In all six burst-sEPSC cells from cut slices, ON stimulation evoked bursts of EPSCs (mean: 7.3 ± 0.9 EPSC/stimulation; Fig. 3G) that were statistically indistinguishable from those elicited from intact slices (mean: 6.8 ± 0.6 EPSC/stimulation). EPSCs evoked in all three single sEPSC cells were similar to responses observed in the intact slice (1.0 ± 0.0 EPSC/stimulation in cut slice; Fig. 3F; vs. 1.6 ± 0.1 EPSC/stimulation in intact slice). The latency and jitter of EPSCs evoked by ON stimulation were also indistinguishable between cells in intact versus cut slices (burst-sEPSC PG cell latency 4.29 ± 0.28 ms and jitter 727 ± 193 μs in cut slices vs. 4.62 ± 0.3 ms and 774 ± 119 μs in intact slices; single-sEPSC cell latency 2.08 ± 0.08 ms and jitter 119 ± 23 μs in cut slices vs. 2.63 ± 0.06 ms and 173 ± 9 μs in intact slices). These results indicate that back-propagated action potentials into M/T cell apical dendrites make little or no contribution to ON-evoked bursts in PG cells. We cannot exclude regenerative events arising within M/T dendritic tufts independently of the cell body. In the accessory OB, M/T cells have a unique multi-tuft morphology with the multiple tufts eclectically remote from the soma. These tufts can receive both back propagation signals as well as local regenerative events activated by synaptic input (Urban and Castro 2005), raising the possibility that local processing could occur within the uni-tufted M/T cells of the main OB, although this has yet to be reported. Thus we conclude the most likely source for spontaneous and ON-evoked burst-sEPSC PG cell responses arises from the disynaptic, ON→ET→PG circuit.

These results are consistent with the idea that at MES intensity, single-sEPSC cells receive monosynaptic ON input and burst-sEPSC PG cells receive ON input predominantly via the disynaptic, ON→ET→PG circuit. It is possible, however, that burst-EPSC PG cells receive monosynaptic ON inputs that were not activated at MES, and single-sEPSC cells might receive ET cell burst inputs that that are not activated by MES stimulation. Thus we examined the responses of an additional 26 burst-sEPSC and 17 single-sEPSC cells to ON stimulation intensities ranging from 1-10X MES. For single-EPSC cells, increasing stimulus intensity never produced EPSC bursts (Fig. 3H). This indicates that single-EPSC PG cells are unlikely to receive input via the ON→ET→PG circuit. In 20 of 26 (77%) burst-sEPSC cells, increased stimulus intensity did evoke shorter latency (mean at 1× MES = 4.79 ms, 3–4× MES = 2.87 ms, ≥5× MES = 2.56 ms; regression coefficient = 0.5, P = 0.0001; Fig. 3, I and J), and lower jitter (mean at 1× MES = 842 μs, 3–4× MES = 271 μs, ≥5× MES = 161 μs; regression coefficient = 0.4, P = 0.0005; Fig. 3, I and K) responses consistent with monosynaptic ON input, although we cannot rule out reduced latency/jitter is the result of increasing the fidelity of a disynaptic circuit at higher stimulus intensity. Stimulus intensity had to be increased to 2.65 ± 0.21× MES to evoke shorter latency, lower jitter responses in burst-sEPSC cells. Next we considered the possibility that the higher intensity stimulation required for monosynaptic ON activation of burst-sEPSC cells was due to suboptimal placement of the simulating electrode. However, this explanation would require that stimulation locus was suboptimal to activate ON synapses in all 26 burst-sEPSC cells but optimal in all 17 single-sEPSC cells. Notwithstanding its a priori unlikelihood, we evaluated this possibility by testing two stimulation electrode placements in 18 burst-sEPSC and 4 single-sEPSC cells. Stimulating at a second location (80–200 μm anterior to the 1st site) elicited short-latency, low jitter responses in all four single-sEPSC cells tested and never elicited a burst with increasing stimulus intensity. By contrast, ON stimulation at a second location still elicited bursts with long latency and high jitter in all 18 burst-sEPSC cells tested. Stimulus intensity at this second site had to be increased to 2.99 ± 0.45× MES to evoke shorter-latency, lower jitter responses in burst-sEPSC cells. This suggests that the longer, variable latency burst responses in burst-sEPSC PG cells is not due to suboptimal stimulation electrode placement.

Taken together, these data indicate that spontaneous input, latency, jitter, and pattern of responses to ON stimulation are correlated. Thus GAD65+ PG cells separate into distinct clusters in multidimensional space along these parameters (Fig. 4, A–C). We determined how many separate clusters could be identified using a K-means test correlating all four parameters. The test for two clusters assigned all 58 single-sEPSC PG cells to one cluster and the 56 burst-sEPSC cells to a second cluster (Fig. 4D). When a K-means test was performed forcing separation into three clusters, only one cell was assigned to the third cluster. This demonstrates that GAD65+ PG cells comprise two highly distinct populations.

Two GABAergic intraglomerular circuits

Taken together, these results suggest that GAD65+ PG cells form two distinct circuits (Fig. 4E): 1) single-sEPSC cells receive single, isolated EPSCs and are driven primarily, if not exclusively, by monosynaptic ON input. We refer to these as ON-driven (ONd)-PG cells. ONd-PG cells represent ∼33% of the GAD65+ PG cell population. 2) Burst-sEPSC cells receive spontaneous bursts of EPSCs. Although most of these PG cells also receive monosynaptic ON input, the stimulus intensity necessary to drive a monosynaptic (short latency, low jitter) response is ∼2.7-fold greater than that required to elicit ON-evoked EPSC bursts. Therefore we suggest that the predominant input to this population of PG cells is via the disynaptic ON→ET→PG circuit. These cells, which we refer to as ET-driven (ETd)-PG cells, comprise ∼67% of the GAD65+ PG cell population.

Because ETd and ONd GAD65+ PG cells differed in their patterns of spontaneous and ON-evoked synaptic inputs, we wondered if they might have different intrinsic properties as well. However, there were no significant differences in resting membrane potential (−56 ± 1.2 mV, n = 21, for ETd vs. −55 ± 0.7 mV, n = 24, for ONd), input resistance (652 ± 50 MΩ, n = 50, for ETd vs. 704 ± 62 MΩ, n = 55, for ONd), spontaneous action potential frequency at rest (2.1 ± 0.3 Hz, n = 13, ETd vs. 3.7 ± 0.4 Hz, n = 15, for ONd), or I_h current (18.5 ± 2.9 pA, n = 21, for ETd vs. 21.2 ± 3.8 pA, n = 23, for ONd). Thus ETd and ONd GAD65+ PG cells are indistinguishable along these dimensions.

Could the different synaptic characteristics of ETd and ONd GAD65+ PG cells be the result of different degrees of dendritic truncation in the slice preparation? If ONd cells had many dendrites pruned during slicing, they might have lost ET cell inputs; conversely if ETd cells lost dendrites, they might have lost ON inputs. To investigate this, we harvested physiologically characterized, biocytin-filled GAD65+ PG cells, made 3D reconstructions and compared the dendritic features of 10 ETd cells and 10 ONd cells (Fig. 1D shows an ETd cell on the right and an ONd cell on the left). There were no significant differences in number of primary dendrites (2.4 ± 0.4 ETd vs. 2.7 ± 0.5 ONd), total dendritic length (721 ± 96 μm ETd vs. 884 ± 162 μm ONd), number of dendritic branches (34 ± 6 ETd vs. 38 ± 8 ONd), number of dendrite endings (38 ± 6 ETd vs. 43 ± 8 ONd), tortuosity (1.37 ± 0.03 ETd vs. 1.38 ± 0.02 ONd), or spine number (25.2 ± 4.9 ETd vs. 20.5 ± 5.2 ONd) although there was a tendency for ETd cells to have a slightly higher spine density (spines/dendrite length; 36.2 ± 5.8 ETd vs. 22.1 ± 4.1 ONd; P < 0.05). The distributions of branches as a function of distance from the soma (Scholl analysis) were statistically indistinguishable between ONd- and ETd-PG cells (not shown). Could different degrees of olfactory nerve truncation underlie sparse ON inputs to some ETd cells? Both ONd- and ETd-PG cells have multiple intact dendrites, and our recordings show these cells can be separated on the basis of spontaneous input alone (single/burst sEPSCs). Thus if the sparse ON input to ETd cells is due to severing on the olfactory nerve, we would have to postulate that we systematically severed more nerve inputs to the 56 cells that exhibited spontaneous EPSC bursts and systematically severed fewer inputs to the 58 cells that lack spontaneous EPSC bursts. Statistically this is highly unlikely. Thus we conclude that the differences in ETd- and ONd-PG cell synaptic physiology are due to the fact that the two cell types operate in two distinct intraglomerular GABAergic circuits and not to selective pruning of olfactory nerve inputs.

ON- and ET-driven PG cells differentially regulate phasic and tonic presynaptic inhibition of ON inputs

Olfactory nerve terminals have a high probability of transmitter release with the result that the second of two paired shocks leads to a reduced postsynaptic response because of reduced vesicle availability, i.e., PPD (Murphy et al. 2004). In addition, ON terminals express GABA_B receptors (Bonino et al. 1999; Kratskin et al. 2006; Priest and Puche 2004), which when activated, reduce presynaptic Ca²⁺ influx, transmitter release, and postsynaptic responses (Aroniadou-Anderjaska et al. 2000; McGann et al. 2005; Pirez and Wachowiak 2008; Wachowiak et al. 2005). ON activation of GABAergic PG cells releases GABA, which activates these GABA_BRs resulting in reduced glutamate release to any subsequent ON activation (i.e., PPD). ON terminals also express DA D₂Rs, DA reduces glutamate release from ON terminals, and ∼10% of PG cells contain DA (Berkowicz and Trombley 2000; Ennis et al. 2001; Koster et al. 1999; Maher and Westbrook 2008; Parrish-Aungst et al. 2007; Wachowiak and Cohen 1999). Thus ON-evoked PPD can be due to at least two distinct presynaptic mechanisms: reduced vesicle availability and transmitter release due to the high release probability of ON terminals and presynaptic inhibition via activation of GABA_BRs and/or DA D₂Rs on ON terminals. The relative contribution of these two mechanisms can be distinguished pharmacologically: PPD due to GABA_BR/D₂R-mediated presynaptic inhibition can be attenuated by GABA_BR or D₂R antagonists (Maher and Westbrook 2008) and PPD due to reduced vesicle availability is unaffected by GABA_BR/D₂R antagonists (Murphy et al. 2004).

Activation of GABA_BRs should reduce ON inputs to both ONd and ET cells, both of which receive monosynaptic ON input, as well as to ETd-PG cells, which are primarily driven by the ON→ET→PG circuit. Indeed the GABA_BR agonist, baclofen, strongly attenuated EPSCs elicited by ON stimulation in all three cell types (87.0 ± 6.6% reduction in EPSC amplitude in the presence of baclofen in ONd-PG cells, n = 4; 86.3 ± 8.0% reduction in ETd-PG cells, n = 4; and 80.5 ± 3.0% reduction in ET cells, n = 4). This indicates that ON terminals have functional GABA_BRs. Next we investigated whether both ETd-and ONd-GAD65+ PG cells contribute to presynaptic inhibition of ON terminals.

We first examined ONd-PG cells (Fig. 5A). Paired ON shocks caused PPD in all ONd-PG cells tested (Fig. 5B; n = 8 cells). The amplitude and area (integrated current) of the EPSC in response to the test stimulus was significantly reduced compared with the conditioning response at test intervals from 25 to 1,000 ms (maximum reduction at 75 ms in both amplitude, 49.4 ± 2.9%; Fig. 5, B and C, and area, 55.9 ± 6.1%; n = 8 cells). In the presence of 10 μM CGP55845, PPD was significantly attenuated at test intervals 50–800 ms (Fig. 5, B and C). Thus ON-evoked GABA release from ONd-PG cells presynaptically inhibits ON terminals. However, there was significant residual CGP-insensitive PPD at intervals from 25 to 1,000 ms (maximal at 25 ms in both amplitude, −41.1 ± 4.5%; Fig. 5C, and area, −37.3 ± 6.9%). This residual PPD may be due to activation of D₂Rs (1) or vesicle depletion (Murphy et al. 2004), although at intervals ≤50–75 ms, it is most like due to the latter.

A previous study reported that intracellular stimulation of some PG cells reduced the magnitude of subsequent ON-evoked excitatory postsynaptic potentials (EPSPs) and that this effect could be attenuated by blocking GABA_B receptors (Murphy et al. 2005). Could these be ONd-PG cells? To address this question, we employed the same approach taken by Murphy et al. (2005) using a brief depolarizing step (from −70 to 0 mV membrane potential for 5 ms) to evoke GABA release from GAD65+ ONd-PG cells 50–200 ms prior to ON stimulation. Depolarization reduced subsequent ON-evoked EPSC amplitude by 37 ± 7% (n = 8 cells, P < 0.001; Fig. 5, D–F) and area by 38.8 ± 5%; this effect was prevented by CGP55845. Taken together, these findings show that ON synaptic input to GAD65+ ONd-PG cells evokes GABA release, which presynaptically inhibits ON terminals synapsing with that cell. The ON→PG cell synaptic site appears to very close to the PG cell GABA release sites because intracellularly evoked GABA release reaches sufficient concentration to reduce subsequent transmitter release from ON terminals.

In addition to reducing PPD, CGP55845 also increased the magnitude of the conditioning response in the PPD paradigm suggesting that the ON→PG synapse is under tonic GABA_BR inhibition. To investigate this further, we tested additional cells comparing single ON shock responses to those recorded following addition CGP55845. Blocking GABA_BRs increased ON-evoked EPSCs by 27.6 ± 9% (n = 5 cells; P < 0.005; Fig. 5, G and H). This is consistent with a previous report of tonic GABA_B inhibition of ON terminals measured by local field potentials (Aroniadou-Anderjaska et al. 2000). Thus during phasic activation of the ON→PG circuit, transmitter release from ON terminals is subject to both GABA_BR-dependent and -independent PPD. In addition, the ON-PG synapse is under significant GABA_BR-dependent tonic presynaptic inhibition.

We next examined presynaptic inhibition in the ON→ET→PG circuit (Fig. 6A). The ET-driven PG cells in this circuit receive strong disynaptic ON input via ET cells and weaker monosynaptic ON input. ON-evoked GABA release from ETd-PG cells might, perhaps via spillover (De Saint and Westbrook 2007; Isaacson 1999; Rossi and Hamann 1998), achieve concentrations sufficient to reduce subsequent ON transmitter release onto ET cells thus reducing ET cell drive onto ETd-PG cells. Because ON stimulation generates a burst of action potentials in ET cells resulting in a compound burst of EPSCs in ETd-PG cells (Fig. 6, B and C; n = 6 cells) we measured the integrated current rather than the amplitude of EPSCs in ETd-PG cells. Moreover, because the duration of the burst of EPSCs elicited by ON stimulation typically lasts longer than 25–50 ms, we could reliably measure PPD only for conditioning-test intervals ≥100 ms. Significant PPD was present at all intervals from 100 to 1,000 ms (33.9 ± 12% maximum reduction at 200 ms; Fig. 6C). In contrast to ONd-PG cells, CGP55845 had no effect on ON-evoked PPD in ETd-PG cells (Fig. 6, B and C). Thus PPD across the ON→ET→PG circuit is independent of GABA_B receptor activation.

ONd-PG cells receive direct ON input and intracellular stimulation of ONd-PG cells reduced the magnitude of subsequent ON-evoked EPSPs. By contrast, ETd cells receive primarily ET cell input and only weak ON inputs, hence we reasoned that intracellular stimulation of a single ETd PG cell might not release sufficient GABA to affect ON→ET synapses. To address this question, a brief depolarizing step (from −70 to 0 mV membrane potential for 5 ms) was applied to evoke GABA release from GAD65+ ETd-PG cells 50–200 ms prior to ON stimulation. Depolarizing steps had no effect on subsequent ON-evoked EPSCs (n = 7 cells Fig. 6, D and E), consistent with the idea that GABA release from one ETd PG cell does not reduce subsequent transmitter release from ON terminals onto ET cells.

Is the GABA_BR-independent PPD across the ON→ET→PG circuit due to PPD at the ON→ET synapse? To address this, we examined paired ON-evoked EPSCs in ET cells. There was significant PPD at all stimulation intervals from 50 to 1,000 ms (40.6 ± 6.0% maximum reduction at 200 ms; n = 5 cells Fig. 7, B and C). However, CGP55845 had no effect on this PPD at any test interval (Fig. 7, B and C). Thus PPD at the ON→ET synapse is not due to GABA_BR-dependent presynaptic inhibition and is consistent with the idea that short-term plasticity at the ON→ET synapse is due primarily to reduced vesicle availability in ON terminals (Murphy et al. 2005) although there may be some contribution by D₂Rs.

Tonic GABA_BR-dependent presynaptic inhibition of ON terminals

The lack of GABA_BR-dependent PPD across the ON→ET→PG circuit was somewhat puzzling. ON-evoked monosynaptic EPSCs in ET cells are followed by GABA_AR-mediated feedback inhibitory PSCs (IPSCs) from PG cells (Hayar et al. 2004a) Thus phasic activation of the ON→ET→PG circuit evokes GABA release from ETd-PG cells. Why, then, doesn't this activate GABA_BRs on ON terminals and/or ET cell dendritic release sites and inhibit responses to a subsequent ON shock? We reasoned as follows: ETd-PG cells receive high-frequency spontaneous EPSCs from ET cells; this spontaneous bombardment may generate sufficient GABA spillover to tonically activate GABA_BRs on all ON terminals producing tonic presynaptic inhibition of sensory inputs. Thus the incremental GABA spillover elicited by a single phasic stimulation of the ON→ET→PG circuit might not achieve sufficient concentration to further activate GABA_BRs against this background of high spontaneous GABA release.

If there is tonic inhibition in the ON→ET→PG circuit, CGP55845 should relieve ON terminals from GABA_BR inhibition and increase the amplitude of ON-evoked EPSCs in ETd-PG cells. This prediction was confirmed: the magnitude of ON-evoked EPSCs in ETd-PG cells was increased 25.0 + 4.4% in the presence of CGP55845 (n = 5 cells; P < 0.01; Fig. 6, F and G).

Tonic GABA_BR-dependent presynaptic inhibition across the ON→ET→PG circuit could be due to tonic inhibition at the ON→ET synapse, the ET→PG synapse, or both. If there is tonic GABA inhibition at the ON→ET synapse, then CGP55845 should increase the magnitude of ON-evoked monosynaptic EPSCs in ET cells. Figure 7, D and E, shows that ON-evoked EPSCs in ET cells were increased 26.1 + 2.8% in the presence of CGP55845 (n = 5 cells; P < 0.001). Therefore the ON→ET synapse is under tonic GABA_BR-dependent inhibition, and the magnitude of this inhibition is very similar to the amount of tonic inhibition across the ON→ET→PG circuit. These results do not exclude the possibility of additional tonic GABA_BR inhibition at the ET→PG synapse. However, while a recent study showed that ET cells have GABA_BRs that can be activated by exogenous baclofen, ET cells showed no evidence of tonic inhibition by endogenous GABA (Karpuk and Hayar 2008). This suggests that the tonic inhibition across the ON→ET→PG circuit is primarily due to tonic GABA_BR inhibition at the ON→ET synapse.

The magnitude of tonic inhibition at the ON→ET synapse (26.1%) is similar to the tonic inhibition at the ON→ONd-PG cell synapse (27.6%), suggesting a uniform level tonic of GABA_BR-dependent presynaptic inhibition of all olfactory nerve terminals. If this is true, then spontaneous EPSCs in ET and ONd PG cells should be tonically reduced as both receive monosynaptic input from the ON. Indeed CGP55845 increased the frequency and amplitude of spontaneous EPSCs in both ET and ONd PG cells. In ET cells, there was a 101 + 40% increase in frequency (Fig. 7G; n = 5; P < 0.01) and a 22 + 1.8% increase in the amplitudes of sEPSCs (Fig. 7H; n = 5; P < 0.001); for ONd PG cells, there was a 15% increase in sEPSC frequency and a 41% +11% in amplitude (n = 4 cells; P < 0.05; not shown). These data indicate that there is tonic GABA_B receptor-mediated inhibition of all olfactory nerve terminals.

Two functional types of GAD65+ GABAergic PG cells

Glomerular GAD65+ cells have the morphological (Cajal 1911; Golgi 1875; Pinching and Powell 1971a–c, 1972) and intrinsic physiological (Hayar et al. 2004a; Murphy et al. 2005) properties of PG cells. We identified two subpopulations of GAD65+ PG cells with markedly different synaptic properties: ONd-PG and ETd-PG cells. These two GABAergic cell types are sharply distinguished by their patterns of spontaneous and ON-evoked activity. ONd-PG cells, comprising 33% of the GAD65+ population, receive single, isolated sEPSCs and respond to ON stimulation with single EPSCs of short, relatively invariant latency. ETd-PG cells, comprising 67% of GAD65+ PG cells, receive bursts of sEPSCs along with single isolated events and respond to ON stimulation with longer, more variable latency EPSC bursts. Increased ON stimulus intensity never produced bursts in ONd-PG cells. By contrast, higher stimulus intensity evoked shorter, less variable latency single EPSCs in many ETd-PG cells. Thus ONd-PG cells appear to receive monosynaptic ON input but no input from bursting ET cells. ETd-PG cells receive their most effective sensory input via ET cells with less effective monosynaptic ON input. These differences between ON- and ET-driven PG cells do not appear to be due to differential loss of dendrites or ON axons in the slice preparation but rather to differences in circuitry.

Previous electron microscopy (EM) studies distinguished two types of PG cells, termed type I, which receive ON and dendrodendritic synapses, and type II, which receive only dendrodendritic and no ON synapses (Kosaka et al. 1997; Toida et al. 1998, 2000); GABAergic and DAergic PG cells are type I and calbindin- and calretinin-positive cells are type II. ONd-PG cells likely correspond to type I cells. ETd-PG cells receive prominent dendrodendritic ET input and weaker ON input; the presence of some ON input suggests they might have been classified as type I cells. A smaller population, comprising ∼20% of the ETd-cells and thus ∼15% of all GAD65+ PG cells, did not respond to strong ON stimulation. These may be type II cells that do not receive ON terminals. A small percentage of GAD65+ PG cells express calbindin or calretinin (6 and 9%, respectively) (Parrish-Aungst et al. 2007) and thus may be calbindin or calretinin type II cells.

The contrasting synaptic properties of ONd- and ETd-PG cells strongly argue that they operate in two functionally distinct circuits: an ON→PG monosynaptic circuit (ONd-PG cells) and an ON→ET→PG disynaptic circuit (ETd-PG cells). These two circuits differentially regulate tonic and phasic presynaptic GABAergic inhibition of ON terminals.

Tonic and phasic presynaptic inhibition of sensory input

ON terminals express GABA_B receptors and glutamate release from ON terminals is dramatically reduced by the GABA_BR agonist, baclofen (Aroniadou-Anderjaska et al. 2000; Murphy et al. 2005; Wachowiak et al. 2005); present study). This has led to the idea that phasic ON activation causes GABA release from PG cells resulting in feedback GABA_BR-mediated presynaptic inhibition of ON-terminals (Aroniadou-Anderjaska et al. 2000; McGann et al. 2005; Murphy et al. 2005; Wachowiak et al. 2005). The present results show that ON terminals are also under strong tonic GABA_BR-mediated inhibition. ON-evoked monosynaptic EPSCs in ET cells, ETd-PG cells, and ONd-PG cells were increased by ∼27% when GABA_B receptors were blocked. This is consistent with an earlier report of tonic GABA_BR presynaptic inhibition measured by local field potentials (Aroniadou-Anderjaska et al. 2000). The most parsimonious explanation is that this tonic presynaptic inhibition arises from the ON→ET→PG circuit. ET cells burst spontaneously and generate monosynaptic bursts of EPSCs in ETd-PG cells (Hayar et al. 2004a). Indeed, the present experiments showed that ETd-PG cells receive 14.0 ± 1.8 sEPSC/s compared with 3.4 ± 0.4 sEPSC/s for ONd-PG cells. We propose that this constant bombardment of EPSCs on ETd-PG cells causes sustained GABA release that achieves sufficient steady state spillover concentrations to tonically inhibit all ON terminals. Pirez et al. (2008) recently used optical imaging to show that odor-evoked calcium influx into ON terminals is suppressed by as much a 40% by tonic activation of GABA_B receptors in vivo. The slightly higher level of tonic inhibition (40 vs. 27%) could be due to differences in the measure used, presynaptic Ca²⁺ signals in vivo versus postsynaptic EPSCs in slices, or to higher tonic GABA release in vivo. Notwithstanding this small difference, both approaches agree that spontaneous GABA release activates GABA_BRs to tonically inhibit ON terminals in vivo and in slices. Intrabulbar circuits or centrifugal inputs that modulate the ON→ET→PG circuit can thus potentially regulate the “gain” of sensory input at the first synapse in the olfactory system. Serotonin, for example, increases GABA release from the ON→ET→PG circuit (Aungst and Shipley 2005), suggesting that ON terminals are modulated by behavioral state.

Phasic regulation of transmitter release from ON terminals is due to at least two mechanisms operating concurrently. ON synaptic terminals have a high release probability (∼0.9) such that postsynaptic responses to the second of two paired ON shocks is attenuated due to reduced vesicle availability (Murphy et al. 2004). In addition, phasic ON input activates ONd-PG cells triggering GABA release that activates GABA_BRs reducing glutamate released by a second ON shock. Thus in addition to tonic presynaptic inhibition, ON input to ONd-PG cells is further regulated by both GABA_BR-dependent presynaptic inhibition and GABA_BR-independent PPD.

ET cells and ETd-PG cells exhibited only GABA_BR-independent PPD. Thus while ON inputs to the ON→ET→PG circuit are under tonic GABA_BR-dependent presynaptic inhibition, the major source of phasic plasticity in this the circuit appears to be vesicle availability at the ON→ET synapse (Murphy et al. 2004). We suggest that the ON→ET→PG circuit lacks phasic GABA_BR-dependent presynaptic inhibition because the ON→ET synaptic release sites are under tonic GABAergic inhibition and are too remote from the ET-PG GABA release sites to be influenced by the incremental GABA released by a single ON shock. By contrast, the ON→PG synaptic release sites appear to be sufficiently proximal to the PG→ON release sites that GABA released by low, perithreshold ON activation reaches concentrations sufficient to further inhibit subsequent ON transmitter release. Is there anatomical evidence for such segregation of synaptic contacts? Glomeruli were classically viewed as relatively homogenous structures consisting of axons, dendrites, synapses, and glial processes. However, more recent studies indicate that glomeruli exhibit a degree of subcompartmentalization that is manifested as a segregation of different types of synaptic contacts (Chao et al. 1997; Halasz and Greer 1993; Kasowski et al. 1999; Toida et al. 2000). ORN axons entering the glomerulus synapse on target dendrites in “axonal” subcompartments or islands within the glomerular neuropil, whereas dendrodendritic synapses occur in bundles of 4–100 dendrites that are segregated from the ORN axonal islands by the processes of glia (Chao et al. 1997; Kasowski et al. 1999). It is reasonable to conjecture that ON→PG synapses are located in these axonal domains whereas ET→PG synapses are in the dendritic domains. Thus while there is ample evidence for functional PG→ON synapses, classical morphological synaptic specializations have not identified.

In summary, vesicle availability in ON synapses and tonic GABA release from ETd-PG cells appear to regulate glutamate release from ON terminals; phasic GABA_BR-mediated presynaptic inhibition influences ON terminals that monosynaptically contact PG cells. Phasic presynaptic inhibition may provide negative gain control of ON terminals, regulating transmitter release adaptively as a function of input frequency. Tonic presynaptic inhibition levels may be adapted over longer time scales ranging from sniff cycles to behavioral states such as arousal/attentiveness, sleep/wakefulness, or motivation.

Dual GABAergic intraglomerular circuits

GAD65+ PG cell dendrites are primarily localized to a single glomerulus, thus in addition to their roles in presynaptic inhibition of ON terminals, both the ON→PG and ON→ET→PG circuits likely contribute to intraglomerular postsynaptic inhibition.

The ON→ET→PG circuit drives ∼67% of GABAergic GAD 65 PG cells. ET cells are entrained by repetitive ON stimulation over the range of sniff frequencies (1–8 Hz) such that more ET cells burst synchronously at higher input frequencies. Synchronous bursting should increase the drive on ETd-PG cells thus adding to the tonic pool of GABA generated by spontaneous ET bursting. The net effect may be a relatively gradual modulation of tonic inhibition in the pre- and postsynaptic targets of the ETd-PG cells. Potential postsynaptic targets of the ON→ET→PG circuit include mitral/tufted or other PG cells.

The ON→PG circuit is driven primarily by the olfactory nerve and ONd-PG cells generate feedback presynaptic inhibition of their ON inputs. ONd-PG cells may also postsynaptically inhibit mitral/tufted or other PG cells. The ON→PG circuit, which drives ∼33% of GABAergic PG cells, may be well-suited to generate pre- and postsynaptic inhibition that is more tightly linked than the ON→ET→PG circuit to the instantaneous frequency of ON inputs.

There is also evidence for synaptic interactions among GABAergic PG cells (Murphy et al. 2005). Thus there may be inhibitory cross-talk between the ON→PG and ON→ET→PG circuits. The finding that ONd-PG cells and ETd-PG cells can be unambiguously identified based on their patterns of spontaneous and ON-evoked EPSCs should facilitate future investigation these two circuits and their roles in intraglomerular signal processing.