Abstract
The olfactory bulb (OB) serves as a relay region for sensory information transduced by receptor neurons in the nose and ultimately routed to a variety of cortical areas. Despite the highly structured organization of the sensory inputs to the OB, even simple monomolecular odors activate large regions of the OB comprising many glomerular modules defined by afferents from different receptor neuron subtypes. OB principal cells receive their primary excitatory input from only one glomerular channel defined by inputs from one class of olfactory receptor neurons. By contrast, interneurons, such as GABAergic granule cells (GCs), integrate across multiple channels through dendodendritic inputs on their distal apical dendrites. Through their inhibitory synaptic actions, GCs appear to modulate principal cell firing to enhance olfactory discrimination, although how GCs contribute to olfactory function is not well understood. In this study, we identify a second synaptic pathway by which principal cells in the rat (both sexes) OB excite GCs by evoking potent nondepressing EPSPs (termed large-amplitude, nondendrodendritic [LANDD] EPSPs). LANDD EPSPs show little depression in response to tetanic stimulation and, therefore, can be distinguished other EPSPs that target GCs. LANDD EPSPs can be evoked by both focal stimulation near GC proximal dendrites and by activating sensory inputs in the glomerular layer in truncated GCs lacking dendrodendritic inputs. Using computational simulations, we show that LANDD EPSPs more reliably encode the duration of principal cell discharges than DD EPSPs, enabling GCs to compare contrasting versions of odor-driven activity patterns.
SIGNIFICANCE STATEMENT The olfactory bulb plays a critical role in transforming broad sensory input patterns into odor-selective population responses. How this occurs is not well understood, but the local bulbar interneurons appear to be centrally involved in the process. Granule cells, the most common interneuron in the olfactory bulb, are known to broadly integrate sensory input through specialized synapses on their distal dendrites. Here we describe a second class of local excitatory inputs to granule cells that are more powerful than distal inputs and fail to depress with repeated stimulation. This second, proximal pathway allows bulbar interneurons to assay divergent versions of the same sensory input pattern.
Introduction
The classic view of the olfactory bulb (OB) that has evolved from anatomic and functional studies is of an array of semi-independent glomerular modules, each driven by a distinct subgroup of olfactory sensory neurons. By interpreting the patterns of activated glomerular modules, animals can determine which odor is present and initiate the complex processing steps that underlie olfactory learning. The principal neurons for conveying olfactory sensory neuron input to downstream cortical regions are mitral cells (MCs) and tufted cells (TCs), the only known glutamatergic cell types in the OB aside for a small population of interneurons in the glomerular input layer (Fremeau et al., 2001; Herzog et al., 2001; Lein et al., 2007; Brill et al., 2009; Ohmomo et al., 2009; Tatti et al., 2014). The OB also contains a large abundance of inhibitory interneurons, including small axonless granule cells (GCs). These GABAergic interneurons vastly outnumber principal neurons and are excited along their distal apical dendrites by dendrodendritic (DD) synapses formed in the external plexiform layer (EPL) (Rall et al., 1966; Isaacson and Strowbridge, 1998; Shepherd et al., 2007; Pressler and Strowbridge, 2017). The distal apical DD pathway provides a mechanism for individual GC to sample a large number of glomerular input modules since these connections are formed along the expansive lateral dendrites of both MCs and TCs. Multiple lines of evidence suggest that inhibition mediated by GCs through these reciprocal DD synaptic local circuits plays a critical role when animals are making fine distinctions between similar odors or odor mixtures (Abraham et al., 2010; Nunes and Kuner, 2015). Exactly how GC inhibition onto principal cells facilitates odor discrimination is not known, but the mechanism likely involves modulation of principal cell firing patterns to enhance subtle differences evident at the glomerular input stage in responses to different stimuli (Fukunaga et al., 2014).
Critical to understanding how GC activity facilitates distinct neural representations of sensory inputs is the question of how these interneurons are excited. Researchers have often studied how neural circuits differentially respond to sensory stimuli through plasticity in one class of orthodromic synaptic inputs (e.g., Fiete et al., 2010). Neurons in many brain regions, however, often compare activity in two or more excitatory pathways (e.g., parallel and climbing fiber inputs to Purkinje cerebellar neurons), providing an alternate model for understanding the organization of circuits within the olfactory system. Simultaneously comparing several distinct representations of a sensory input arriving via different local circuit pathways would enable GCs to reliably develop specific associations between odorants and the diffuse patterns of activity across glomerular modules. The presence of a second excitatory synaptic input that complements the widely sampling DD input could explain the paradoxical ability of broadly sampling GCs to generate sparse and selective responses to individual odors (Kato et al., 2012; Cazakoff et al., 2014).
Previous studies (Price and Powell, 1970a; Kishi et al., 1984; Mori et al., 1999; Nagayama et al., 2004) have speculated that local axon collaterals of MCs and/or TCs might synapse on GCs, providing a potential second local excitatory connection to these interneurons. By using a novel approach of eliminating the abundant DD inputs from individual GCs that could obscure other types of connections (Pressler and Strowbridge, 2019), we have revealed this new local excitatory pathway to GCs that selectively targets proximal dendrites and likely arises from axons of OB principal cells. The large-amplitude, nondepressing EPSPs evoked through this pathway contrast with both depressing DD and facilitating inputs to the same interneurons (Balu et al., 2007; Gao and Strowbridge, 2009; Pressler and Strowbridge, 2017). Both DD and axo-dendritic circuits appear to be at least partially maintained within in vitro brain slices, enabling sensory inputs to excite GCs through both pathways. Because DD and axo-dendritic inputs to GCs have contrasting forms of short-term plasticity, proximal axo-dendritic inputs more reliably reflect the duration of discharges in presynaptic principal cells. Such differential processing may facilitate detecting aspects of the sensory input that modulate discharge duration, such as odor concentration (Cang and Isaacson, 2003; Wienisch and Murthy, 2016).
Materials and Methods
Animals
Horizontal OB slices (300 μm thick) were prepared from postnatal day (P) 14-25 Sprague Dawley rats (both sexes) anesthetized with ketamine, as described previously (Pressler and Strowbridge, 2006, 2019) using a Leica Microsystems VT1200 vibrotome. Parasagittal anterior piriform cortex (APC) slices, 350 μm thick, were prepared from rats in the same age range anesthetized with isoflurane, as described by Suzuki and Bekkers (2011). All experiments were conducted in accordance with the guidelines approved by the Case Western Reserve University Animal Care and Use Committee.
Electrophysiology
Both OB and APC slices were incubated initially at 30°C for 30 min and then maintained at room temperature until use. During recording sessions, slices were placed in a submersion chamber and perfused with oxygenated ACSF at a rate of 1.5 ml/min. Recordings were made between 29°C and 31°C. ACSF consisted of the following (in mm): 124 NaCl, 3 KCl, 1.23 NaH2PO4, 1.2 MgSO4, 26 NaHCO3, 10 dextrose, 2.5 CaCl2, equilibrated with 95% O2/5% CO2. Except where noted, drugs were added to the submerged recording chamber by changing the external solution source. In some experiments, intracellular blockers were included in the recording pipette loading solution. Whole-cell current-clamp patch-clamp recordings were made using AxoPatch 1D amplifiers (Molecular Devices) and borosilicate glass pipettes (3-7 mΩ) under IR-DIC video microscopy. Recording electrodes contained the following (in mm): 140 K-methylsulfate, 4 NaCl, 10 HEPES, 0.2 EGTA, 4 MgATP, 0.3 Na3GTP3, 10 phosphocreatine di-Tris. Alexa-594 or Alexa-488 (100-150 μm; Invitrogen) was also included routinely to permit visualization of the recorded neuron using live 2-photon imaging, as described previously (Pressler and Strowbridge, 2006, 2017). Cell-attached recordings used the same internal solution to enable subsequent breakthrough to whole-cell recording model. All internal solutions were adjusted to pH 7.3 and 290 mOsm.
Intracellular recordings were low-pass filtered at 5 kHz (FLA-01, Cygus Technology) and acquired at 10 kHz using an ITC-18 interface (Instrutech/HEKA) and custom Visual Basic.NET and Python software on a Windows 7-based personal computer. Intracellular recordings were not corrected for the liquid junction potential. All cell type classifications were confirmed by intracellular dye filling followed by fluorescence visualization. Unless noted, all drugs were obtained from Tocris Bioscience, except for TTX, which was purchased from Calbiochem.
In this study, we stimulated afferent inputs to GCs using three different methods: focal stimulation near a visualized dendritic section, extracellular stimulation within a specific layer (either the olfactory sensory nerve [OSN] input layer or the GC layer [GCL]), or stimulation within one glomeruli. Visualized focal stimulation (e.g., see Figs. 2D, 3B–E) (Balu et al., 2007; Gao and Strowbridge, 2009) was performed using live 2-photon imaging to position a second patch pipette filled with 124 mm NaCl, 3 mm KCl, and 10 mm HEPES, pH 7.4, and 100 μm Alexa-594 within 5-10 μm of an identified dendritic segment. We used sharpened monopolar tungsten electrodes (FHC) for layer- and glomerulus-specific stimulation. In MC recording experiments involving glomerular-specific stimulation (e.g., see Fig. 5A,B), the glomerulus containing the MC apical dendritic tuft is indicated as “glom0” (confirmed in all experiments by filling MCs with Alexa-594 dye via breaking through to whole-cell recordings).
In Figure 6H–K, we used bath application of the GABAA receptor agonist muscimol (1-2 μm) to test whether evoked synaptic responses were mediated by monosynaptic or polysynaptic pathways, paralleling previous work by others (Najac et al., 2011; Fukunaga et al., 2012; Budzillo et al., 2017). We found similar effects of 1 and 2 μm muscimol on OSN-evoked responses (conversion of prolonged EPSP barrages to single EPSPs in most responses). A third truncated GCs was only tested using 1 μm muscimol and also a similar conversion of an EPSP barrage to single evoked EPSPs in most episodes.
Data analysis and statistics
We selected for GCs by recording from neurons with small, elongated (“egg-shaped”) cell bodies in the GCL imaged using IR-DIC video microscopy and a 60× water immersion objective and confirmed GC-like morphology in all recordings using live 2-photon imaging. Some experiments involved recording from GCs with apical dendrites severed in the GCL (“truncated GCs”), as described in a recent publication from our group (Pressler and Strowbridge, 2019). Truncated GCs appeared to be healthy neurons with high input resistances (1450 mΩ vs 960 mΩ for intact GCs, as reported previously in Pressler and Strowbridge, 2019) and long membrane time constants (49 vs 43 ms in intact GCs). Truncated GCs were not specifically targeted for recording; GC subtype (intact or truncated) was determined only after intracellular filling and visualization.
Summary results are presented as mean ± SEM, except for the computation simulation (mean ± SD). Statistical comparisons used Student's t test unless otherwise indicated and were performed in Python (version 3.6) using the SciPy.Stats library. Cumulative distributions were compared using the Kolmogorov–Smirnov test. The assay of two-dimensional clustering presented in Figure 3F used the Hopkins statistic (Banerjee and Dave (2004) with values >0.75 indicating significant clustering. Jitter estimates represent the SD of the onset latency to the first EPSP (or action potential [AP]).
Computational model
Computation simulations of EPSP summation in GCs used the same methodology described in a recent publication from our group (Pressler and Strowbridge, 2017). In brief, we initially created a set of presynaptic firing patterns by either randomly selecting patterns from a library of previously recorded MC (or TC) discharges evoked by glomerular stimulation (similar to Fig. 5A–D) or by constructing synthetic discharges that matched the statistical properties of recorded MC/TC discharges (the same average interspike interval and variability [coefficient of variation] of the interspike interval). Each spike within the MC/TC discharge was then processed to simulate probabilistic EPSP transmission (for details, see Pressler and Strowbridge, 2017) to generate a new postsynaptic depolarization waveform. Estimates of GC depolarization assume linear EPSP summation from each successive MC/TC spike train. The resting membrane potential in simulations was −70 mV and firing threshold was set at −42 mV based on the most depolarized stable membrane potential we recorded in GCs (−42.0 ± 2.5 mV from Pressler and Strowbridge, 2017). Simulations included a spontaneous EPSP rate of 13 Hz based on the mean rate of spontaneous EPSPs in our intact GCs recordings. Estimates of the number of active, odor-triggered presynaptic MC/TC discharges required to trigger GC APs (see Fig. 8A) reflect conditions in which 95% of simulations using that parameter set depolarized to at least −42 mV and, therefore, would have triggered at least one AP. Simulations performed to assay GC AP latency (see Fig. 8B,C) also continuously increased the number of presynaptic MC discharges until the GC reached firing threshold in 95% of trials.
Results
Variation in spontaneous EPSP amplitudes and kinetics
Abundant previous work suggests that olfactory GCs receive excitatory synaptic input through DD synapses they form with MC and TC dendrites (Rall et al., 1966; Isaacson and Strowbridge, 1998; Schoppa et al., 1998). Our group recently reported properties of these synapses assayed using paired intracellular recordings (Pressler and Strowbridge, 2017) (example DD EPSPs shown in Fig. 1A). In that dataset, we found multiple examples of spontaneous EPSPs that were larger than any evoked DD EPSP recorded in the 6 paired recordings. (This initial 2017 report on paired MC/GC recordings focused exclusively on evoked DD EPSPs in GCs, not on the properties of spontaneous EPSPs.) Figure 1B illustrates examples of large-amplitude spontaneous EPSPs along with spontaneous putative DD EPSPs and evoked DD EPSPs, all recorded in the same GC. These large-amplitude spontaneous EPSPs also had very rapid rising phase kinetics and reached their maximal amplitude ∼4 ms before evoked DD EPSPs.
Olfactory GCs receive large-amplitude spontaneous EPSPs. A, Diagram of experimental configuration for DD paired recording experiment with one intracellular recording in an MC and another in a postsynaptic GC. Thin red traces indicate repeated GC responses to evoked MC APs (bottom black trace). One trial failed to trigger a DD EPSP (labeled failure). Top thick red trace indicates average GC response. B, Example spontaneous EPSPs from the same postsynaptic GC shown in A with more rapid kinetics (all have >1000 mV/s rising phase slopes, 5%-50%) and larger amplitude than the average unitary-evoked DD EPSP shown at bottom. Other example spontaneous EPSPs recorded in the same GC (middle set of traces) resembled the evoked DD EPSP (bottom red trace). C, Plot of the cumulative distribution of spontaneous (black curve) and MC spike-evoked (red curve) EPSP rising phase slopes (5%-50%) in the 6 GCs with monosynaptic DD connections with MCs. Arrows indicate the largest evoked DD EPSP and the largest spontaneous EPSP recorded in the same set of neurons. D, Cumulative plots of EPSP amplitudes from the same two datasets shown in C.
We recorded 2695 spontaneous EPSPs in the same 6 postsynaptic GCs with monosynaptic DD EPSPs evoked via paired intracellular recordings. The largest amplitude evoked DD EPSP in this dataset was 5.1 mV, >40% smaller than the largest spontaneous EPSP recorded in the cells (8.9 mV; cumulative histograms presented in Fig. 1C). We found a similar increase in the range of the EPSP initial slopes with the fastest evoked DD EPSP representing ∼50% of the slope of the fastest spontaneous EPSP recorded (Fig. 1D). When analyzed individually within each paired recording, the fastest spontaneous EPSPs had initial slopes were 2.04 ± 0.25 times greater than the fastest MC-evoked DD EPSP (spontaneous:evoked ratio ranged from 1.4 to 2.9, mean ratio significantly different from 1, p < 0.005, T = 4.103).
This initial result suggests that the evoked DD EPSPs we assayed in our six monosynaptic paired recordings represented only a subset of the range of excitatory synaptic inputs GCs receive. While varying degrees of electrotonic attenuation and filtering may contribute to the deviations between evoked and spontaneous distributions, electrotonic effects are unlikely to be the explanation for the excess large-amplitude spontaneous EPSPs recorded in GCs. The experiments described below will test the hypothesis that many of these large-amplitude EPSPs represent a new type of synaptic response reflecting glutamatergic contacts on the proximal apical and basal dendrites of GCs. Our results are divided into three sections, beginning with testing whether large-amplitude EPSPs are distinct from classic DD EPSPs by attempting to record them in GCs lacking distal apical dendrites (“truncated GCs”). We then ask whether large-amplitude EPSPs are distinct from previously demonstrated (Balu et al., 2007; Gao and Strowbridge, 2009; Sun et al., 2020) facilitating inputs that also target proximal GC dendrites. And finally, we ask whether local circuits evoke large-amplitude EPSPs.
Large-amplitude EPSPs arise from proximal GC synapses
If large-amplitude EPSPs arise from excitatory synapses on proximal dendrites, then they should still be present in “truncated” GCs that have no dendritic arbors in the EPL and, therefore, no DD synaptic inputs (Fig. 2A,B) (Price and Powell, 1970b; Balu et al., 2007; Pressler and Strowbridge, 2017). We recorded from 27 truncated GCs that lacked EPL dendrites and found spontaneous large-amplitude EPSPs in 82% of those recordings. As shown in the example recordings in Figure 2C, spontaneously occurring large-amplitude EPSPs in truncated GCs had similar large amplitudes (>2 mV) and fast rising phase kinetics as spontaneous large-amplitude EPSPs in intact GCs (Fig. 1B). Focal stimuli near truncated GCs elicited large-amplitude EPSPs with short, presumably monosynaptic latencies (1.8 ms in the example in Fig. 2D) at threshold stimulus intensities that triggered both successes and failures. We also could evoke large-amplitude EPSPs using extracellular GCL stimulation while recording from truncated GCs (N = 2, not shown).
Large-amplitude EPSPs evoked by focal microstimulation of proximal GC dendrites. A, Left, 2-photon maximal intensity z stacks demonstrating extensive dendritic arborization of a recorded intact GC within the EPL, the lamina where DD synapses are located. Image was obscured by debris on the surface of the slice located on the MCL/EPL boundary. Right, z stacks of a truncated GC with apical dendrite severed in the GCL. B, Responses to depolarizing and hyperpolarizing current steps in the truncated GC shown in A (right). C, Spontaneous non-DD EPSPs recorded in the truncated GC shown in A. These example spontaneous EPSPs all have >1000 mV/s initial slopes (5%-50%). D, Top, Focal microstimulation near the basal dendrite of the truncated GC shown in A-C evoked large-amplitude non-DD EPSPs with fast kinetics. Dashed line indicates the average initial rising phase slope of EPSPs (5%-50%; all > 1000 mV/s). Asterisks indicate stimulus timing (and in subsequent figures). Bottom, Histogram of EPSP initial slopes from 9 truncated GC recordings with focal dendritic stimulation. Typical recording configuration shown in inset. E, DD EPSPs evoked in monosynaptic MC/GC recordings (top). Dashed line indicates the average EPSP slope. Bottom, Histogram of MC AP-evoked DD EPSP initial slopes paired MC/GC recordings. Diagram of MC/GC paired recording configuration shown in inset.
Over 9 focal stimulation experiments, we evoked both small-amplitude EPSPs that facilitated with repeated stimulation and large-amplitude EPSPs in truncated GCs. The mean amplitude of evoked EPSPs in truncated GCs was 1.9 mV, significantly larger than the mean amplitude of evoked DD EPSPs assayed in MC/GC paired recordings (p < 0.01, T = 2.38, unpaired t test). We also found a large distribution of EPSP initial slope kinetics (Fig. 2D, bottom) with >54% of evoked EPSPs in truncated GCs having more rapid initial slopes than found in all DD EPSPs assayed in our paired MC/GC recordings (Fig. 2E).
We next asked whether large-amplitude EPSPs also had a unique form of short-term plasticity that would help define a functional signature for these inputs. Over 7 focal microstimulation experiments that evoked large-amplitude EPSPs, most recordings revealed near neutral paired-pulse ratios (PPRs) (neither depressing nor facilitating), a result that, while uncommon in CNS excitatory synapses, has been reported in a subset of glutamatergic synapses in piriform cortex (Suzuki and Bekkers, 2011). In the example recordings in Figure 3A–E, we evoked different monosynaptic EPSPs at two different focal stimulation positions in the same intact GC: small-amplitude EPSPs that facilitated with paired stimulation when stimulating the proximal apical dendrite (Fig. 3B,D) and large-amplitude EPSPs with neutral PPRs when stimulating near a basal dendrite (Fig. 3C,E). Over 17 microstimulation experiments, plots of EPSP unitary amplitude versus PPR showed two clusters: Hopkins (H) statistic = 0.94, greater than the 0.75 threshold for significant clustering (Banerjee and Dave, 2004) (Fig. 3F), corresponding to small-amplitude facilitating (SAF) EPSPs (green symbols) and large-amplitude EPSPs with near-neutral PPRs (orange symbols). When analyzed solely by short-term plasticity, small-amplitude EPSPs had significantly larger PPRs than large-amplitude EPSPs (2.24 ± 0.21 vs 1.21 ± 0.1 for large-amplitude EPSPs evoked at 20 Hz; p < 0.0002, T = 4.69). We found similar, near neutral PPRs when assaying large-amplitude EPSPs evoked at 40 Hz (0.91 ± 0.05; N = 7).
Two types of excitatory responses onto proximal GC dendrites. A, 2-photon z-stack montage showing intact GC recorded in experiments in B–E. Same GC following placement of a dye-filled focal stimulating electrode near either the proximal apical dendrite (B) or a basal dendrite (C). D, Evoked monosynaptic EPSPs following trains of stimuli applied to the proximal apical dendritic stimulating electrode (trains of 3 shocks at 20 Hz). Individual responses superimposed at bottom; average response shown above (green trace). E, Responses in the GC to focal stimulation of a basal dendrite in the same GC. Individual responses superimposed at bottom; average response shown above (orange trace). Trains of two shocks at 20 Hz. These example spontaneous EPSPs all have > 1000 mV/s rising phase slopes (5%-50%). F, Summary plot of responses to GCL stimulation in visualized GCs. Filled green symbols represent SAF EPSPs. Filled orange symbols represent LANDD EPSPs. Diamond symbols represent 2-photon-guided microstimulation near visualized dendritic segments. Circles represent stimuli within the GCL but without visualization of the stimulation electrode in both groups. (Nonvisualized stimulation was only tested in truncated GCs to exclude activation of DD inputs.) Clusters were statistically different based on the Hopkins statistic > 0.75. Mean ± SEM for PPR estimates indicated inside the y axis and unitary amplitude inside the x axis. G, Plot of EPSP initial slope (5%-50% of rising phase) for three different classes of GC EPSPs. For LANDD versus SAF EPSPs, ***p < 1.2 × 10−4, T = 5.03. For LANDD versus DD EPSPs, ***p < 1.3 × 10−4, T = 4.75; both unpaired t tests. H, Plot of onset latency for the same three EPSP classes. **p < 0.01, T = 2.65 (unpaired t test). n.s.; p > 0.05.
Because they appear to reflect a distinct input class based on both kinetics and short-term plasticity, we refer to the large-amplitude, fast-rising EPSPs (initial slopes > 800 mV/s) synaptic potentials as “large-amplitude nondendrodendritic” (LANDD) EPSPs. We used a primary classification metric based on EPSP rising-phase slope, rather than amplitude, because of the difficulty in assessing the baseline membrane potential during trains of summating EPSPs. Prior work suggests that small-amplitude facilitating EPSPs, referred to as “SAF EPSPs” in this study, are likely to be heterogeneous, representing both cortical feedback (de Olmos et al., 1978; Uva et al., 2006; Balu et al., 2007; Gao and Strowbridge, 2009) and a subset of intrinsic OB inputs (Sun et al., 2020). Over our series of focal microstimulation experiments, there was a trend for LANDD EPSPs to be evoked by stimulation near basal dendrites (in 5 of 7 of experiments) and a similar trend for SAF EPSPs to be evoked by stimulation near proximal apical dendrites in the GCL (in 3 of 4 experiments).
The results presented thus far suggest that GCs receive three types of excitatory synaptic inputs: (1) weak and depressing DD EPSPs on distal apical processes, (2) weak and facilitating EPSPs on proximal dendrites, and (3) large-amplitude LANDD EPSPs, also likely arising from excitatory synapses on proximal GC dendrites. Initial rising phase slope in LANDD EPSPs differed significantly from both SAF and DD EPSPs (Fig. 3G). All three GC EPSP responses appeared to be monosynaptic based on short (<3 ms) onset latencies (Fig. 3H).
Although SAF EPSPs facilitate with repeated stimulation, we find greater temporal summation in LANDD EPSPs, as shown in Figure 4A–C. In the example recording in Figure 4A that compares two stimulation sites on the same intact GC, the maximal depolarization reached following a four-shock stimulus train was larger following LANDD stimulation despite the absence of paired-pulse facilitation in this synapse. As shown in Figure 4B, the average amplitude of DD EPSPs continuously depresses during similar 40 Hz stimulus trains and SAF EPSPs begin to depress following the third stimulus. Figure 4C summarizes the results of a series of similar 4-shock experiments across all three types of excitatory synapses onto GCs and demonstrates the degree of large temporal summation inherent in LANDD inputs. The minimal depression following prolonged stimulus trains likely explains the ability of summating LANDD EPSPs to generate large postsynaptic depolarizations. Consistent with the large degree of temporal summation we find with LANDD EPSPs, trains of these EPSPs could reliably trigger GC spiking as shown in the example recording in Figure 4D.
Contrasting forms of short-term plasticity in SAF and LANDD EPSPs. A, Comparison of short-term plasticity in responses to trains of 40 Hz stimuli applied near the proximal apical (top green trace) or basal (bottom orange trace) of the same intact GC. B, Summary plot of short-term plasticity assayed using trains of 40 Hz stimuli that triggered SAF EPSPs (green symbols; N = 6 experiments), DD (blue; N = 6), or LANDD (orange; N = 8) symbols. C, Plot of EPSP summation following the final stimulus in the 4 × 40 Hz stimulation trains illustrated in A. ***p < 1.03 × 10−4, T = 5.67; **p < 3.3 × 10−4, T = 4.86, both unpaired t tests. D, Example recordings in which trains of LANDD EPSPs triggered APs with low jitter (0.32 ms; latency SD). E, Blockade of AMPARs with 10 μm DNQX abolishes LANDD EPSPs recorded at −70 mV evoked by focal stimuli applied near the basal dendrite.
LANDD EPSPs appeared to represent conventional ionotropic glutamate receptor-mediated synaptic responses. Evoked LANDD EPSPs were blocked by DNQX, an AMPAR antagonist (Fig. 4E; N = 3; 10 μm). LANDD EPSPs could still be evoked when fast voltage-gated Na+ channels were blocked by perfusion with QX-314 (N = 3; 5 mm). We also observed spontaneous LANDD EPSPs in following both intracellular QX-314 perfusion and when Na+ channels were blocked extracellularly by bath application of TTX (see Fig. 7). In addition, we could evoke LANDD EPSPs in GCs filled with 10 mm BAPTA, which disrupts Ca2+-mediated intracellular signaling (N = 3). Together, these results indicated that, despite their large amplitude and the propensity of GCs to generate intrinsic spikes (Egger et al., 2005; Bywalez et al., 2015), LANDD EPSPs resemble AMPAR-based classic cortical and hippocampal synaptic potentials that do not appear to be dependent on voltage-gated Na+ channels or Ca2+-activated conductances.
Polysynaptic networks mediate LANDD EPSPs
We next conducted a series of experiments to test whether LANDD EPSPs arise from centrifugal axons or whether they are mediated by local (bulbar) excitatory neurons, presumably MCs and/or TCs. If LANDD EPSPs arise from presynaptic MCs or TCs, then glomerular layer stimulation should evoke polysynaptic LANDD EPSPs (by activating distal MC/TC dendrites directly and/or stimulating OSN excitatory inputs to MCs and TCs). To test this possibility, we first verified that focal stimulation in the glomerular layer could bring MCs to threshold. As shown in Figure 5A, graded single-shock stimulation applied to one glomerulus recruited repetitive firing assayed with a cell-attached MC recording. Once the threshold stimulus level was reached, doubling the stimulus intensity did not recruit faster firing or prolong MC discharges. MC responses were highly selective to which glomeruli was stimulated. As shown in Figure 5B, of three neighboring glomeruli tested with suprathreshold stimuli (3× threshold intensity), only one stimulus position elicited MC firing (the same glomerulus we found contained the apical dendritic tuft on subsequent dye filling). In both MC and TC cell-attached recordings, we observed a variety of responses to focal stimuli in the glomerular layer stimulation with some cells responding to the stimuli with a single short-latency spike (Fig. 5C). Other principal cells responded to similar glomerular stimuli with a delayed discharge (Fig. 5D).
Sensory input evokes LANDD EPSPs. A, Left, Diagram of recording configuration with cell-attached MC recording and focal microstimulation of three nearby glomeruli. Subsequent whole-cell recording and 2-photon imaging was used to determine the “on-beam” glomerulus for that MC. Right, Example MC responses to on-beam (“glom 0”) stimulation recorded at three intensities. B, Only the stimulation of the on-beam glomerulus triggered MC discharges. Same recording as in A; stimulus intensity 3 times threshold level determined in A in glom0 and two adjacent glomeruli (glom-1 and glom1). C, Example single short-latency (7.7-9.0 ms) AP response to glomerular stimulation in a different MC. D, Example long-latency (53-71 ms) discharge following glomerular layer stimulation in a different MC. E, Long-latency LANDD EPSPs evoked by glomerular stimulation in a truncated GC. Diagram of recording configuration shown on left. Experiment in an OB slice with the AON removed. F, Examples of short-latency single LANDD EPSPs evoked by glomerular stimulation in a different truncated GC. G, Long-latency clusters of LANDD EPSPs evoked by glomerular stimulation in a different truncated GC. H, Plot of onset latency versus onset latency jitter in cell-attached recordings from MCs (filled square symbols) and TCs (filled circles) following on-beam glomerular stimulation. I, Analogous plot of latency versus jitter in EPSP responses to glomerular layer (blue symbols) or GCL (orange symbols) stimulation in truncated GCs. All GCL stimuli experiments generated presumptive monosynaptic EPSPs with onset latencies <3 ms. J, Plot of initial slopes of LANDD EPSPs evoked by GCL (orange) or glomerular layer (blue) stimulation. Mean initial slope values are not statistically different (n.s.; p > 0.05).
We evoked polysynaptic LANDD EPSPs in 8 truncated GCs in a separate set of recordings using the same type of focal glomerular stimulation used to assay MC and TC responses. We observed LANDD EPSP patterns that matched all three MC/TC discharge patterns we observed, including short-latency discharges (Fig. 5E), short-latency single EPSP responses (Fig. 5F), and long-latency discharges (Fig. 5G). In all cases, the latency to elicit LANDD EPSPs was >3 ms, suggesting that the focal glomerular layer stimulation activated a local bulbar excitatory neuron that was presynaptic to the GC, generating a polysynaptic EPSP in the truncated GC.
Since all LANDD EPSP recordings in this set of experiments were from truncated GCs that lacked distal apical dendrites in the EPL, these excitatory responses could not arise from DD synapses. Projections from nearby olfactory cortical regions also were unlikely to generate polysynaptic LANDD EPSPs since most OB slices had no cortical targets that projected back to the OB (e.g., anterior olfactory nucleus [AON], APC, or posterior piriform cortex). While no slices in this study included piriform cortex, which requires a special slicing configuration to obtain OB-APC connections (Balu et al., 2007; Apicella et al., 2010), a minority of OB slices included a section of the AON. In two experiments, we dissected the AON region away from these slices and were still able to record polysynaptic LANDD EPSPs in truncated GCs (e.g., Fig. 5E).
The onset latencies and barrage patterns of polysynaptic LANDD EPSPs evoked by glomerular stimulation closely resembled the discharge patterns we demonstrated in cell-attached MC and TC recordings, suggesting that LANDD EPSPs likely arise from axon collaterals of a subset of OB principal cells. [Both MCs and TCs are polarized neurons with separate laminar zones for their dendritic and axonal arborization (Price and Powell, 1970a) with only the axons of MCs and TCs entering the GCL.] We tested this hypothesis further by quantifying the latency/jitter relationship for MC/TC discharges (Fig. 5H) and polysynaptic LANDD EPSPs (Fig. 5I, blue symbols), both evoked by similar focal glomerular stimulation protocols. Consistent with previous studies using similar in vitro recording conditions (De Saint Jan et al., 2009; Gire and Schoppa, 2009) and in vivo discharges evoked by odors (Kashiwadani et al., 1999; Shusterman et al., 2011; Fukunaga et al., 2012), we find a wide range of onset latencies for both MC and TC discharges. We also found a positive correlation between mean onset latency and variability of train latency, discharges that began relatively long after the stimulus tended to have highly variable start times. As shown in Figure 5I, there was a similar range of onset latencies and latency variation in LANDD EPSPs recorded in truncated GCs providing additional evidence that LANDD EPSPs likely arise from axon collaterals of MCs and/or TCs. All the EPSP barrages evoked glomerular stimulation in truncated GCs appeared to be polysynaptic based on their onset latencies (all >3 ms). By contrast, monosynaptic LANDD EPSPs evoked by focal stimulation near-proximal dendritic segments had relatively low latency jitter (Fig. 5I, orange symbols, all with latencies <2.6 ms). LANDD EPSPs evoked monosynaptically by focal stimulation and polysynaptically via glomerular layer stimulation had similar properties, including nearly identically average initial slopes (Fig. 5J).
While we demonstrated that glomerular layer stimulation triggered long-latency LANDD EPSPs, the glomerular layer contains a variety of external afferents fibers (Shipley and Adamek, 1984; McLean and Shipley, 1987, 1991; Boyd et al., 2012), which might also contribute to evoked EPSPs we recorded in truncated GCs. To address this issue, we asked whether barrages of LANDD EPSPs could also be evoked by stimulation OSN axons. In 4 of 4 truncated GC recordings, OSN stimulation triggered barrages of EPSPs. Since these GCs lacked distal dendrites in the EPL, these EPSPs could not arise from the more common DD pathway. The AON was dissected off the OB slice in 3 of 4 experiments recording OSN responses in truncated GCs, including the example response shown in Figure 6A, eliminating the possibility that OSN stimulation activated long-range OB/cortical pathways mediated the responses. The individual EPSPs within the synaptic barraged evoked by OSN stimulation in truncated GCs resembled LANDD EPSPs with large amplitudes and rapid initial slopes.
Sensory input stimulation excites GCs through a polysynaptic circuit. A, A single OSN shock evokes an EPSP barrage recorded in a truncated GC from an OB slice without the AON. Latency to initial EPSP was 20.8 ms. Schematic of experimental configuration shown on left. B, Responses to 40 Hz tetanic stimulation of the LOT in APC SLs (EPSP latency = 2.4 ms). C, Analogous 40 Hz OSN stimulation evokes long-lasting EPSP barrages in OB GCs. D, LOT stimulation triggered monosynaptic EPSP onto an APC SL (top, 2.3 ms latency), while OSN stimulation triggered a polysynaptic EPSP onto an intact OB GC (bottom, 5.8 ms latency). Dashed line indicates onset of SL EPSP. Blue arrow indicates onset of GC EPSP. E, Summary plot of initial EPSP latency in SL, intact, and truncated GCs. **p < 0.02, T = 2.41; ***p < 0.001, T = 4.638 (y axis broken to illustrate range of data points). F, Summary plot of EPSP latency jitter in the same three cell types. *p < 0.05, T = 2.0; ***p < 0.01, T = 2.74 (also with y axis broken). G, Plot of the cumulative distribution of EPSP latencies found in the first 100 ms of time following stimulation, triggered by LOT stimulation in SLs (black curve) and OSN stimulation in intact GCs (blue curve) and OSN stimulation in truncated GCs (orange curve). The plots of intact GC and truncated GC EPSPs are significantly different from the plot of EPSPs onto SLs (N = 9) versus intact GC (N = 6): p < 1.9 × 10−17; SL (N = 9) versus truncated GC (N = 4): p < 3.7 × 10−13; both Kolmogorov–Smirnov tests. H, Activation of GABAA receptors with muscimol does not affect the latency to the initial EPSP in an SL cell. I, Muscimol abolishes late spikes in OSN-evoked TC response while preserving the initial, short-latency (1.8 ms) spike. J, Muscimol abolishes late EPSPs evoked by OSN stimulation in an intact GC while preserving the initial EPSP (enlarged in inset). K, Muscimol also abolishes late EPSPs evoked by OSN stimulation in a truncated GC. Inset, Initial muscimol-resistant EPSP.
All OSN-evoked responses recorded in truncated GCs appeared to be polysynaptic based on onset latencies >4 ms for the initial EPSP evoked in the barrage. The similar pattern of delayed EPSPs evoked by OSN (Fig. 6A) and direct glomerular stimulation (Fig. 5E–G) could be explained by axo-dendritic MC/TC synapses onto GCs since both stimulation paradigms robustly activate OB principal cells.
We next performed additional experiments to determine whether OSN-evoked EPSP barrages could reflect inputs from centrifugal glutamatergic axons. Centrifugal inputs in our OB slice preparation primarily represent severed axons as we dissected the only nearby cortical region (the AON) away in most experiments and more distant sources of excitatory centrifugal axons, such as piriform cortex, were not present in our OB slices (Balu et al., 2007).
We initially compared OSN-evoked responses in truncated GCs with a “ground truth” standard for monosynaptic excitation via severed glutamatergic axons, the LOT input to semilunar cells (SLs) in APC (Suzuki and Bekkers, 2006, 2011), and for polysynaptic excitation, the DD input to intact OB GCs (Isaacson and Strowbridge, 1998; Pressler and Strowbridge, 2017). As shown in the examples presented in Figure 6B–D, Each LOT stimuli typically triggered one monosynaptic EPSP in SLs with short latency (<3 ms) and with low jitter. By contrast, OSN stimuli triggered polysynaptic EPSPs in intact GCs with latencies >4 ms and larger jitter than monosynaptic EPSPs.
OSN-evoked responses in truncated GCs more closely resemble polysynaptic EPSPs with long latencies and high jitter than monosynaptic EPSPs (Fig. 6E,F; and for similar latency and jitter results for EPSPs evoked by glomerular stimulation, see Fig. 5G,H). The average latency and jitter of OSN-evoked EPSPs in truncated GCs were significantly larger than monosynaptic LOT-evoked EPSPs in SLs (Fig. 6E,F, orange symbols). Both single and repetitive OSN stimulation evoked prolonged barrages of EPSPs in intact and truncated GCs (Fig. 6A,C; barrage EPSP latencies plotted in Fig. 6G). In SLs, similar repetitive stimulation of severed LOT axons failed to trigger barrages of EPSPs (Fig. 6G, black curve). The mean EPSP frequency in the 1 s window following tetanic LOT stimulation of SLs (2.0 ± 0.3 Hz, excluding the initial monosynaptic evoked EPSP) was not elevated above spontaneous EPSP rates (2.0 ± 0.4 Hz; p > 0.05; N = 9; paired t test) in the same neurons. By contrast, EPSP frequency during the initial 1 s following OSN stimulation was significantly elevated above the spontaneous EPSP frequency (p < 0.005 for intact GCs; p < 0.02 for truncated GCs). These findings provide further evidence suggesting that OSN stimulation evoked EPSPs barrages in truncated GCs by activating OB excitatory local circuits rather than inducing spontaneous spiking in severed centrifugal axons.
We conducted one additional test for polysynaptic versus monosynaptic excitation in truncated GCs using the GABAA receptor agonist muscimol to shunt responses in potential intermediate excitatory neurons. Muscimol shunting should preferentially impact polysynaptic responses by attenuating intracellular depolarization associated with EPSPs and therefore lowering the transmission probability through multisynapse excitatory networks (Najac et al., 2011; Fukunaga et al., 2012; Budzillo et al., 2017).
In control experiments recording monosynaptic SL EPSPs, bath application of muscimol failed to block EPSPs evoked by LOT stimulation in 9 of 9 experiments despite reducing SL input resistance by 53%. As expected from the increasing shunting, muscimol depressed the amplitude of LOT-evoked EPSPs (by ∼30%; Fig. 6H). In MCs and TCs, muscimol reduced input resistance to a similar degree as SLs (∼70%) and abolished most OSN-driven spiking. In most MCs/TCs tested (4 of 5 experiments, including the example shown in Fig. 6I), muscimol eliminated all APs, except the initial AP evoked by OSN stimulation (at ∼2 ms; Fig. 6I, bottom). In one MC experiment, muscimol abolished all OSN-evoked APs. As expected based on these MC/TC responses, muscimol abolished most of the EPSP barrage in 9 of 9 intact GCs tested while preserving the initial EPSP (Fig. 6J). Truncated GCs behaved similarly to intact GCs with muscimol abolishing nearly all of the OSN-evoked EPSP barrage while leaving only the initial EPSP (Fig. 6K; 3 of 3 experiments). In both intact and truncated GCs, the remaining initial EPSP in muscimol occurred at latencies associated with polysynaptic, not monosynaptic, excitation (>6 ms in all GC experiments; see Fig. 6J,K, insets). The ability of prolonged shunting with muscimol to abolish most of the OSN response in truncated GCs provides further evidence that LANDD EPSPs are mediated by local polysynaptic excitatory circuits rather than by severed centrifugal afferents.
Our ability to drive LANDD EPSPs at long, presumably polysynaptic latencies with glomerular layer and OSN layer stimulation suggests that they originate from local MCs and/or TCs. Since the OB principal cells often are spontaneously active even in in vitro brain slices (Heyward et al., 2001; Schmidt and Strowbridge, 2014), we asked whether we could record AP-dependent spontaneous LANDD EPSPs in truncated GCs. We tested the effect of blocking voltage-gated Na+ spikes with TTX in 13 truncated GCs. On average, TTX significantly reduced the frequency of all EPSPs by 60% (Fig. 7A–C). Nearly all truncated GCs (12 of 13) still received spontaneous LANDD EPSPs in TTX, presumably reflecting the random, quantal release events at LANDD synapses. The frequency of spontaneous LANDD EPSPs was significantly reduced by TTX (from 0.25 ± 0.095 to 0.061 ± 0.014 Hz; p < 0.05; T = 1.93, paired t test), consistent with the hypothesis that at least some LANDD EPSPs originated from spontaneously active local presynaptic excitatory neurons, neurons contained within OB slices.
Spontaneous LANDD EPSPs require Na+ spiking. A, Image of filled truncated GC. B, Spontaneous EPSPs before (top, black traces) and after (bottom, orange traces) bath application of 1 μm TTX. Recordings from the truncated GCs shown in A. C, Summary plot of the effect of TTX on spontaneous EPSP rates in 13 truncated GCs. ***p < 0.01, T = 2.86 (paired t test). Inset, Examples of spontaneous LANDD EPSPs before (black) and after (orange) TTX recorded in the same truncated GC.
Functional differences between DD and LANDD synaptic inputs
These experimental results suggest that GCs receive two functionally distinct classes of excitatory drive from OB principal cells: depressing DD inputs onto their distal apical dendrites and nondepressing inputs via axo-dendritic synapses on their proximal dendrites. We used a computational approach similar to that used in Pressler and Strowbridge (2017) to address the significance of these two converging excitatory pathways. We first repeated the computational assay our group used to estimate the number of presynaptic odor-activated MCs and/or TCs required to bring a GC to spiking threshold. In these simulations, we used a large library of cell-attached recordings of MCs and TCs responses to glomerular layer stimulation (similar to responses shown in Fig. 5A–D; N = 120 TC responses and 84 MC responses). We simulated EPSP generation and temporal summation triggered by each AP in the MC/TC discharge using synaptic release probability, unitary amplitude, and short-term plasticity parameters estimated in this report and earlier, related studies (Pressler and Strowbridge, 2017). As expected based on the large unitary amplitude and absence of short-term depression, summating LANDD inputs from relatively few presynaptic MC or TCs (∼11) are required to bring a GC to threshold and trigger at least one AP (Fig. 8A). When GCs were excited exclusively through the DD pathway, GC spiking required sensory-evoked discharges in approximately 3 times as many MCs or TCs (∼29).
Differential sensitivity to MC discharge duration in DD and LANDD pathways. A, Summary plot of the number of presynaptic MCs (or TCs assayed in separate simulations) required to bring a GC to firing threshold in 95% of model runs. Error bars indicate mean ± SD. B, Plot of the length of the presynaptic MC discharge versus latency to the initial GC spike in simulations using DD-like (blue symbols) or LANDD-like (orange) short-term plasticity. Simulations used the same (DD-like) EPSP properties (e.g., amplitude, probability, response template). Results using simulated MC discharges based on mean interspike interval in recorded glomerular-evoked MC discharges (21.3 ms) and the measured coefficient of variation of the interspike interval (1.83). The number of active (discharging) presynaptic MCs required to trigger at least one GC AP in the DD-like STP simulations varied from 30 (3 spike discharges) to 22 (12 spike discharges). With LANDD-like STP, the number of active presynaptic MCs required to bring the GC to the AP threshold varied from 25 (3 spike discharges) to 12 (12 spike discharges). C, Similar results were obtained when increasing the onset latency jitter from 2 to 20 ms.
In addition to requiring different size active presynaptic ensembles, inputs through the DD and LANDD pathways appear specialized to convey different types of odor-related information. Increasing odor concentration prolongs the duration of both MC and TC discharges (Fukunaga et al., 2012), an effect that appears to be minimized in downstream GCs by the short-term depression inherent in the DD pathway. Figure 8B,C illustrates the effect of varying the duration of synthetic MCs discharges (see Materials and Methods). In each simulation, the size of the presynaptic MC ensemble was increased until the GC reached firing threshold in 95% of trials (see Materials and Methods). The LANDD and DD curves represent the functional effects of different short-term plasticity experimentally measured in both types of EPSPs (synaptic amplitude, probability, and response time course were kept identical [DD-like] in both sets of simulations). While GC AP latency was sensitive to the duration of MC discharges for relatively brief trains (up to 6 APs) with activated synapses with DD-like short-term plasticity, the GC spike latency was relatively insensitive to later MC spikes. In vivo, MC discharge duration triggered by the initial sniff is highly variable and often extends beyond 150 ms, including ≥10 spikes (e.g., Fukunaga et al., 2012, their Fig. 5; see also Bolding and Franks, 2017, 2018; Wilson et al., 2017). Excitatory drive with LANDD-like short-term plasticity, by contrast, preserves more information related to the duration of presynaptic MC discharges, and presumably related sensory information, such as odor concentration. The divergence in the effect of variable MC discharge duration was evident both when relatively little jitter was added to the onset of each MC discharge (Fig. 8B) and when there was substantial random variation in when MCs responded to the same stimulus (Fig. 8C). These results suggest that GCs may exploit differences in synaptic properties in multiple excitatory pathways from OB principal cells to create parallel representations of the same sensory input (e.g., with different degrees of sensitivity to odor concentration).
Discussion
We make three conclusions based on the results presented in this study. First, we identified a new excitatory synaptic input to olfactory GCs with rapid onset kinetics (termed LANDD EPSPs). The novel LANDD EPSPs we identified are distinguished from previously described DD (Pressler and Strowbridge, 2017) and SAF EPSPs (Balu et al., 2007; Gao and Strowbridge, 2009; Sun et al., 2020) by unitary EPSP properties and by markedly different forms of short-term plasticity. Second, we find that LANDD EPSPs likely arise from synaptic inputs onto the proximal dendrites of GCs, especially the basal dendrites. Because of their proximal location, we were able to record spontaneous and evoked LANDD EPSPs in truncated GCs that lack distal apical dendrites, the primary site of DD synaptic inputs. The large amplitude of LANDD EPSPs likely reflects minimal electrotonic attenuation of AMPAR-mediated currents that are proximal to the somatic recording electrode rather than amplification by intrinsic currents. And finally, LANDD EPSPs likely originate from local MCs and/or TCs since we find the frequency of spontaneous LANDD EPSPs was reduced when spontaneous Na+ spiking was suppressed with TTX. Also, orthodromic stimulation in the glomerular layer and OSN triggered barrages of polysynaptic LANDD EPSPs, even in truncated GCs that lack DD synaptic inputs. Together, these results suggest that at least some MCs and/or TCs form axo-dendritic synapses on the proximal dendrites of GCs, providing a second othrodromic pathways to activate these interneurons (Fig. 9). Because of their different forms of short-term plasticity, DD and LANDD EPSPs likely reflect different aspects of MC/TC firing patterns with depressing DD inputs primarily signaling the onset of an MC/TC spike discharge while temporally summating LANDD EPSPs enable GCs to monitor the duration and frequency of the discharge.
Diagram of hypothesized routes of GC excitation leading to DD and LANDD EPSPs. Diagram of proposed local circuits responsible for MC/TC-to-GC excitation via the DD pathway (targeting the distal apical dendrite) and through the LANDD pathway (targeting the basal dendrite). GCs are also excited by SAF EPSPs, shown terminating on the proximal apical dendrite in the diagram. SAF EPSPs originate from both intrinsic bulbar circuits and long-range feedback projections from piriform cortex. Blue text represents synaptic pathways.
LANDD EPSPs likely reflect GC excitation from synapses formed by MC/TC axon collaterals
Our results provide six lines of evidence that MCs/TCs activate GCs via axo-dendritic synapses onto proximal dendrites. First, we find that focal stimulation near proximal dendrites, especially basal dendrites, monosynaptically activates LANDD EPSPs. LANDD synaptic responses appear to be unitary EPSPs with distinct failures and successes with irreducible latencies >1.2 ms, suggesting that they are not mediated by intrinsic currents activated by passive depolarization from electrical stimulation. Second, we evoked LANDD EPSPs by focal stimulation in both intact GCs and in truncated GCs that lack distal apical dendritic arbors in the EPL and, therefore, DD EPSPs. This result suggests that LANDD EPSPs are not a large-amplitude variant of DD synaptic responses. Third, monosynaptic LANDD EPSPs are abolished by the non-NMDAR antagonist DNQX, demonstrated they are driven by a glutamatergic presynaptic neuron. Fourth, LANDD EPSPs have a distinctive functional signature, a neutral PPR evident with repeated stimulation, that distinguishes them from both DD and SAF EPSPs. Excitatory synapses that neither facilitate nor depress with repeated stimulation are relatively rare, although the association inputs to pyramidal cells in APC (Suzuki and Bekkers, 2006) show similar neutral PPRs.
These findings argue that LANDD EPSPs arise from axo-dendritic excitatory synapses on GCs. Instead of reflecting an external input from glutamatergic neurons outside the OB, the final two lines of evidence suggest that at least some LANDD EPSPs arise from MCs and/or TCs, the only established glutamatergic cell types in the OB (Fremeau et al., 2001; Lein et al., 2007; Ohmomo et al., 2009). Approximately half of our truncated GC recordings had relatively frequent (>1 Hz) spontaneous EPSPs. All these recordings included spontaneous LANDD EPSPs and blockade of spontaneous spiking with TTX decreased EPSP frequency. Because severed axons are typically not spontaneously active, this result implies that LANDD EPSPs originate from local (bulbar) excitatory neurons that are spontaneously active. Previous studies have demonstrated that both MCs and TCs can be spontaneously active in rodent brain slices (Heyward et al., 2001; Schmidt and Strowbridge, 2014). And finally, stimulation in the glomerular layer and OSN of OB slices drove polysynaptic LANDD EPSPs in truncated GCs. The long latencies of sensory afferent-driven LANDD EPSPs suggests that the stimulation activated intermediate excitatory neurons contained within the OB slice. While our OB slices did not contain piriform cortex, we occasionally made slices that contained part of the AON. In those slices, we could dissect the AON region off the slice and still evoke polysynaptic LANDD EPSPs. Both MCs and TCs exhibit a variety of spiking response patterns to focal glomerular stimulation (e.g., discharges and single spikes). The pattern of LANDD EPSPs evoked by the same type of glomerular stimulation matched both the variation in onset latency and latency jitter we recorded in glutamatergic principal cells.
While our study does not demonstrate LANDD EPSPs in monosynaptic paired recordings, such a study is likely to be exceeding difficult given the large number of potential postsynaptic GCs there are for each principal cell. With live 2-photon visualization of MC/TC processes (via intracellular recording and dye filling), most filled MC and TC axons either were truncated at the slice surface or failed to bifurcate within the superficial portion of the GCL, technical factors that accentuate the already sparse nature of the LANDD pathway. When this visualization approach was successful as part of MC/GC and TC/GC paired recordings, we found GCL axon collaterals more frequently in TCs than MCs (and see Igarashi et al., 2012). Another potential caveat of our work relates to truncated GCs, which have a higher input resistance than GCs with intact apical dendritic arbors (Pressler and Strowbridge, 2019). While this effect may accentuate the amplitude of LANDD EPSPs recorded in truncated GCs, we often recorded similar large-amplitude spontaneous EPSPs in intact GCs. The higher input resistance (and longer membrane time constant) also would not explain very fast kinetics of LANDD EPSPs. In intact GCs, shunting associated with frequent spontaneous EPSPs and IPSPs may also attenuate LANDD EPSPs. Finally, with our limited dataset of polysynaptic evoked LANDD EPSPs, we cannot yet determine the spatial range of glomerular modules that can excite GCs through proximal inputs and whether individual presynaptic neurons excite GCs through both DD and LANDD pathways.
Two previous ultrastructural reports support the hypothesis that MC/TC axon collaterals synapse on GCs. Classic work by Price and Powell (1970b) suggested that the MC/TC axons contacted GC dendrites within the GCL since they found asymmetric synapses on GC dendrites within the GCL. The presynaptic terminals in these systems had vesicles that resembled vesicles in MC axon initial segments and in known synaptic terminals formed by MCs in piriform cortex. More recently, Schaefer and colleagues demonstrated synaptic contacts from a filled TC axon collateral onto the proximal dendrite of a GC (Schwarz et al., 2018, their Supplemental Fig. 8A–C). Our results suggesting that local excitatory bulbar neurons form axo-dendritic synapses is also consistent with findings from a recent study from Sun et al. (2020) who activated subsets of TCs optogenetically and evoked EPSPs in GCs with near-neutral PPRs (∼1.3 average PPR).
Functional significance
Principal cells in the OB respond to olfactory sensory input by generating an AP barrage at γ-band frequencies that typically lasts for 30-100 ms (e.g., Eeckman and Freeman, 1990). Because of their different forms of short-term plasticity, excitatory input through DD and LANDD synapses likely convey two divergent signals to GC interneurons in response to MC/TC discharges. Strongly depressing DD inputs function to detect the onset of a MC or TC discharge since the largest postsynaptic depolarization occurs following the initial AP in the discharge. Synaptic depression reduces the probability that GCs will continue to spike in response to subsequent spikes within the MC/TC discharge. By contrast, LANDD EPSPs temporally summate in response to γ-band tetanic stimulation (Fig. 4A–C), that mimic odor-evoked MC/TC discharges, allowing this pathway to drive GC spiking later within the presynaptic MC/TC discharge.
Whether GCs spike primarily in response to the onset of principal cell discharges or throughout the discharge will have a large impact on both the neural signals that activate GCs and the resulting recurrent inhibition GCs provide back to principal cells. Through the LANDD excitatory pathway, GC responses are likely influenced by the duration of principal cell discharges. Odor concentration modulates the duration of MC and TC discharges (Fukunaga et al., 2012), suggesting that GCs likely receive sensory signals that are less sensitive to odor concentration through the DD pathway, while LANDD EPSPs convey more information-related odor concentration. Therefore, LANDD inputs could facilitate odor concentration coding since they would convey excitatory drive to GCs throughout the odor-evoked MC/TC barrage. In vivo recordings (Cang and Isaacson, 2003; Wienisch and Murthy, 2016) also demonstrate that the intensity and duration of GC discharges are modulated by odor concentration, suggesting that the nondepressing LANDD pathway we identified onto GCs is likely active in vivo.
The ability of LANDD EPSPs to trigger late discharges in GCs, compared with the transient DD excitatory input near the start of MC/TC firing, is likely to influence olfactory function. Accurate discrimination between similar odorants requires more processing time than easier odor classifications (Uchida and Mainen, 2003; Abraham et al., 2004; Rinberg et al., 2006). Interneuron excitation via LANDD EPSPs provides a cellular mechanism for MC/TC inhibition to develop later in the odor response, presumably reflecting longer periods of synaptic integration in GCs, that could help pattern principal cell discharges. The large amplitude of LANDD EPSPs also provides a potential mechanism to explain sparse GC odor responses (relative to MC/TC responses) (Kato et al., 2012, 2013; Cazakoff et al., 2014; and see Koulakov and Rinberg, 2011). GCs express fewer dendritic spines on their basal dendritic arbor, the likely site of LANDD inputs, than on their distal apical dendrites where DD inputs are found (Price and Powell, 1970c; Carleton et al., 2003; Whitman and Greer, 2007; Kelsch et al., 2008). These anatomic differences suggest that the LANDD input pathway may reflect a narrower range of presynaptic glomerular modules and, therefore, sparser odor coding.
Dual LANDD/DD synaptic inputs may also facilitate learning the constellation of active glomerular modules activated by an odor through spike timing-dependent plasticity mechanisms if a subset of active principal cells drives GC spiking through proximal LANDD EPSPs. The GC could then use conventional long-term plasticity mechanisms to associate the larger population of active principal cells that form DD synapses with the sparser set of “primary” glomeruli that initiated GC firing via LANDD inputs. Further studies will be necessary to determine whether such a “teacher/learner” learning model, reminiscent of classic models proposed by Marr (1969) and others, operates in the OB.
Footnotes
This work was supported by National Institutes of Health Grant R01-DC04285. We thank Dr. Dan Wesson for helpful discussions related to this project; and Drs. Chris Ford and Joel Zylberberg for providing helpful comments on this manuscript.
The authors declare no competing financial interests.
- Correspondence should be addressed to Ben W. Strowbridge at bens{at}case.edu
