INTRINSIC AND SYNAPTIC PROPERTIES OF OLFACTORY BULB NEURONS AND THEIR RELATION TO OLFACTORY SENSORY PROCESSING RAMANI BALU

Transcription

1 INTRINSIC AND SYNAPTIC PROPERTIES OF OLFACTORY BULB NEURONS AND THEIR RELATION TO OLFACTORY SENSORY PROCESSING by RAMANI BALU Submitted in partial fulfillment of the requirements For the degree of Doctor of Philosophy Thesis Advisor: Dr. Ben W. Strowbridge Department of Neurosciences CASE WESTERN RESERVE UNIVERSITY May, 2007

2 CASE WESTERN RESERVE UNIVERSITY SCHOOL OF GRADUATE STUDIES We hereby approve the thesis/dissertation of Ramani Balu Candidate for the Ph.D. degree *. (signed) Iain Robinson (chair of committee) Ben Strowbridge Ruth Siegel Stefan Herlitze R. John Leigh (date) October 20, 2006 *We also certify that written approval has been obtained for any proprietary material contained therein. ii

3 Table of Contents Table of Contents...iii List of Figures... v Acknowledgements... vi List of Abbreviations... vii Abstract... ix Chapter 1 : Introduction... 1 Anatomical Organization of the Mammalian Olfactory System... 2 Synaptic Mechanisms of Mitral Cell Spike Patterning... 4 Reciprocal dendrodendritic inhibition between mitral and granule cells... 5 Gating of dendrodendritic inhibition under normal physiological conditions... 8 Regulation of action potential propagation in mitral and granule cell dendrites Plasticity at dendrodendritic synapses Other synaptic mechanisms of mitral cell spike patterning Intrinsic Mechanisms of Mitral Cell Spike Patterning Overview of Dissertation Chapter 2 : Phasic Stimuli Evoke Precisely Timed Spikes in Intermittently Discharging Mitral Cells Summary Introduction Materials and Methods Results Discussion Mechanisms of Spike Clustering and Phase Locking Functional Implications for Odor Coding Chapter 3 : Opposing Inward and Outward Intrinsic Currents Control Rebound Discharges in Mitral Cells Summary Introduction Materials and Methods Results Discussion Chapter 4 : Multiple Modes of Synaptic Excitation Onto Granule Cells of the Olfactory Bulb Summary Introduction Experimental Procedures Slice preparation and recording Two-Photon Imaging Two-Photon Guided Microstimulation Imaging Synaptically Evoked Calcium Transients DiI Injections Results iii

4 Two distinct classes of excitatory inputs onto granule cells Distal and proximal excitatory synapses are functionally distinct Do granule cells have silent synapses? What is the source of the proximal excitatory input to granule cells? Discussion Multiple synaptic mechanisms for activating GABAergic granule cells Source of the proximal axo-dendritic input onto granule cells Implications for the long-term plasticity at dendrodendritic synapses Chapter 5 : Discussion Intrinsic regulation of synaptic integration of excitatory inputs Rebound spike generation and the regulation of local inhibitory inputs on mitral cells The sniff as a fundamental unit of olfactory bulb processing Intrinsic regulation of action potential backpropagation Synaptic mechanisms of mitral cell patterning Plasticity in the olfactory bulb Modulation of intrinsic currents and plasticity Plasticity at dendrodendritic synapses Bibliography iv

5 List of Figures Figure 1-1. Glomerular organization of sensory inputs in the olfactory bulb Figure 1-2. Local circuits and synaptic processing in the olfactory bulb Figure 1-3. Ultrastructure of olfactory dendrodendritic synapses Figure 2-1. Intermittent firing and subthreshold oscillations in olfactory bulb mitral cells Figure 2-2. Variability of clustered spike discharge timing in mitral cells Figure 2-3. Variability in mitral cell firing patterns does not reflect initial conditions Figure 2-4. Transient repolarizations promote precise phase locking in mitral cells Figure 2-5. Precise spike timing in mitral cells activated by phasic stimuli Figure 2-6. Different effects of first and second sepsps Figure 2-7. Disruption of precise spiking by 4-AP Figure AP sensitive K + currents regulate intermittent discharges in mitral cells. 63 Figure 3-1. Transient hyperpolarizing stimuli evoke rebound discharges in mitral cells.84 Figure 3-2. Voltage dependence of rebound spiking in mitral cells Figure 3-3. Subthreshold Na currents boost mitral cell responses to depolarizing stimuli and mediate rebound spiking Figure 3-4. Rebound spiking is regulated by the duration of hyperpolarizing inputs Figure 3-5. Prolonged hyperpolarizing steps recruit a slowly-inactivating K current in mitral cells Figure 3-6. Brief hyperpolarizations control mitral cell discharges in an all-or-none manner Figure 3-7. Summary of intrinsic mechanisms regulating rebound discharges in mitral cells Figure 4-1. Two classes of spontaneous excitatory postsynaptic currents in granule cells Figure 4-2. Two types of granule cell EPSCs evoked by 2-photon guided minimal stimulation Figure 4-3. Kinetic differences between distal and proximal minimal EPSCs Figure 4-4. Distal and proximal excitatory synapses have different forms of short-term plasticity Figure 4-5. Frequency-dependent modulation at distal and proximal excitatory synapses Figure 4-6. Both proximal and distal excitatory synaptic inputs activate NMDA and non- NMDA glutamate receptor subtypes Figure 4-7. Tests for AMPAR and NMDAR silent excitatory synapses on granule cells Figure 4-8. Mitral cells contact nearby granule cells primarily through dendrodendritic synapses in the external plexiform layer Figure 4-9 Cortical feedback projections generate facilitating, fast-rising EPSCs in granule cells v

6 Acknowledgements I would first and foremost like to thank my advisor Ben Strowbridge for his patient guidance and scientific insight. Ben is truly committed to the scientific development of his students, and I have benefited greatly from my interactions with him during my PhD training. I would also like to thank the members of my thesis committee for all of their help and advice. My time as a graduate student was greatly enriched by my daily interactions with members of the Strowbridge lab and the other students, faculty and staff of the Neuroscience department. I would especially like to thank Todd Pressler, for great discussions about Neuroscience, politics, Dungeons and Dragons, Star Wars and everything in between. It is my sincere hope that we will stay in touch, both as fellow scientists and as friends, wherever our lives take us. While I can never thank them enough, I must thank my parents, whose love, support and intellectual curiosity made me who I am today. Any compliments I ever get should be sent directly to them. Thanks mom and dad. Finally, I want to thank Janani Rangaswami for coming into my life. Her love and faith in me gives me a strength that I never knew I had (and helped me go on during those late nights writing this thesis). vi

7 List of Abbreviations ACSF: artificial cerebrospinal fluid AMPA: alpha-amino-3-hydroxy-5-methylisoxazole-4proionate AON: anterior olfactory nucleus AP: action potential APC: anterior piriform cortex BATPA: O,O -Bis(2-aminophenyl)ethyleneglycol-N,N,N,N -tetraacetic acid D-APV: D-2-Amino-5-phosphonovalerate DDI: dendrodendritic inhibition EGTA: O,O -Bis(2-aminoethyl)ethyleneglycol-N,N,N,N -tetraacetic acid EPL: External Plexiform Layer EPSC: excitatory postsynaptic current EPSP: excitatory postsynaptic potential GABA: gamma-aminobutyric acid GC: granule cell GCL: granule cell layer IPSC: inhibitory postsynaptic current IPSP: inhibitory postsynaptic potential JGC: juxtaglomerular cell LOT: lateral olfactory tract MC: mitral cell NBQX: 1,2,3,4-tetrahydro-6-nitro-2,3-dioxo-benzo[f]quinoxaline-7-sulfonamide vii

8 NMDA: N-methyl-D-aspartate OB: olfactory bulb PGC: periglomerular cell PTX: picrotoxin R in : input resistance TTX: tetrodotoxin VSCC: voltage sensitive calcium channels viii

9 Intrinsic and Synaptic Properties of Olfactory Bulb Neurons and Their Relation to Olfactory Sensory Processing Abstract by RAMANI BALU The elucidation of how sensory information is represented in the brain by distributed patterns of electrical activity is a fundamental challenge in neuroscience. Numerous theories exist to explain the encoding of sensory perceptions by brain activity; however, the cellular mechanisms of sensory perception remain a mystery. Because of its stereotyped and relatively simple anatomy, the olfactory bulb represents an ideal system to study sensory coding. This project used cellular electrophysiological and optical imaging methods to investigate how local circuits within the olfactory bulb process information from olfactory sensory afferents to produce a coded representation of smell that is relayed to higher centers. First, I studied the intrinsic membrane currents in mitral cells the principal output neurons of the olfactory bulb that shape their response properties. Mitral cells have unique intrinsic electrophysiological properties that actively sculpt their responses to depolarizing and hyperpolarizing stimuli. One class of slowly inactivating voltage gated potassium currents (D-type) controls the generation of action potential clusters in response to depolarizing stimuli and ensures precise spiking in response to phasic depolarizations that mimic trains of olfactory sensory nerve mediated excitatory postsynaptic potentials (EPSPs). A different class of inactivating voltage gated potassium currents (A-type) regulates the ability of mitral cells to fire rebound action potentials in response to inhibitory postsynaptic potentials (IPSPs) and ix

10 hyperpolarizing stimuli. These results suggest that the intrinsic electrophysiological properties of mitral cells actively regulate the temporal pattern of mitral cell action potentials during odor processing I next investigated the properties of excitatory glutamatergic inputs onto granule cells. Granule cells are the most common interneuron in the olfactory bulb and make reciprocal dendrodendritic synapses with mitral cells. These interneurons possess two functionally distinct classes of synapses that differ in their short term plasticity properties: dendrodendritic inputs from mitral cells that show prominent depression in response to trains of stimuli, and feedback inputs from the piriform cortex that strongly facilitate. These results suggest that oscillatory activity in the piriform cortex a prominent feature of odor processing may gate the activity of dendrodendritic inputs, and has important consequences for how feedback inhibition onto mitral cells is regulated. x

11 Chapter 1 : Introduction Understanding how sensory information is encoded by distributed patterns of electrical activity in brain circuits is a fundamental challenge in neuroscience. The mammalian olfactory bulb represents an ideal system to study how sensory input is transformed through intrinsic neuronal properties and local circuit interactions into a coded representation of the outside world that is relayed to higher order brain centers. The olfactory bulb has a relatively simple structure (Shepherd, 1972; Cajal, 1995; Shepherd and Greer, 1998) and well-defined innervation pattern from sensory receptors in the nose (Vassar et al., 1994; Mombaerts et al., 1996; Shepherd and Greer, 1998; Mori et al., 1999b). No synaptic interactions have been reported between olfactory sensory receptors (Shepherd, 1972; Shepherd and Greer, 1998), so olfactory information appears to arrive to the bulb from the sensory periphery unaltered. In addition, unlike other sensory modalities, olfactory information flows directly from the olfactory bulb to higher order cortical centers without further processing in the thalamus (Shepherd and Greer, 1998). Thus, the transformation of olfactory information into distinct spatiotemporal patterns of activity that are unique for different odors can be effectively studied by investigating the inputs, outputs, and synaptic properties of neurons in a single brain area. Finally, because it is situated only one synapse away from the sensory periphery, the olfactory bulb presents a unique opportunity to study how cellular changes in synaptic strength and neuronal excitability might relate to learning and behavioral changes mediated by sensory experience. Previous work has shown a rich repertoire of plasticity in olfactory mediated 1

12 behaviors that involve changes in olfactory bulb function (Kaba et al., 1994; Kaba and Nakanishi, 1995; Brennan and Keverne, 1997; Wilson et al., 2004a); however, the cellular mechanisms underlying these changes remain unclear. Anatomical Organization of the Mammalian Olfactory System Olfactory sensory transduction begins in the nose, where volatile odorant molecules bind to G-protein coupled receptors (GPCR) on olfactory sensory neurons located in the nasal epithelium. Odorant binding to olfactory GPCRs depolarizes sensory neurons and causes them to fire action potentials. Olfactory sensory neurons synaptically activate groups of mitral and tufted cells the principal cells of the olfactory bulb which then send projections to higher order centers such as the piriform cortex, anterior olfactory nucleus, and olfactory tubercle (Shepherd and Greer, 1998; Mori et al., 1999b; Zou et al., 2001). Different classes of olfactory sensory neurons project to specific groups of mitral cells at olfactory bulb glomeruli. Each glomerulus contains presynaptic terminals from a subpopulation of receptor cells that express a single type of olfactory receptor protein (Vassar et al., 1994; Mombaerts et al., 1996). There are approximately 2000 glomeruli in the rodent olfactory bulb and 1000 types of olfactory receptors (Buck and Axel, 1991; Buck, 1996a, b; Shepherd and Greer, 1998; Mori et al., 1999a); most receptor subtypes are represented by two glomeruli in each olfactory bulb. Because olfactory sensory neurons express only a single type of GPCR, and odorants bind to a subset of the total complement of olfactory receptor proteins in the nasal epithelium, different odorants can cause distinct spatial patterns of glomerular (and therefore mitral cell) activation (Fig. 1-1). Interestingly, this well defined anatomical pattern does not lead to the specific activation of one or few glomeruli by single molecular species (Belluscio 2

13 and Katz, 2001; Luo and Katz, 2001; Meister and Bonhoeffer, 2001; Spors and Grinvald, 2002; Bozza et al., 2004) (but see (Lin da et al., 2005; Lin da et al., 2006) for contrary viewpoint). Rather, because odorant receptors show broad tuning to multiple odorant classes (Duchamp-Viret et al., 1999; Malnic et al., 1999; Araneda et al., 2000), even single odorants can activate multiple glomerular modules. Thus, the initial spatial pattern of mitral cell activation shows broad overlap between many different odorant classes. The initial activation of glomerular modules is refined and transformed by local synaptic interactions between mitral cells and bulbar interneurons. These interactions occur between mitral cells both at their glomerular tufts and at mitral cell secondary dendrites (Shepherd and Greer, 1998; Mori et al., 1999a; Schoppa and Urban, 2003). Evidence suggests that these bulbar network interactions both sharpen the initial broad spatial activation of glomerular modules through lateral and self-inhibition (Yokoi et al., 1995; Isaacson and Strowbridge, 1998) and impose temporal patterns on mitral cell outputs that are unique for different odorants (Wellis et al., 1989; Mori et al., 1999a; Laurent, 2002). Temporal patterning of spiking occurs both at the level of the entire olfactory bulb and at the level of individual mitral cell discharges. In the insect antennal lobe an olfactory structure with glomerular organization analogous to the mammalian olfactory bulb (Laurent, 2002) individual projection neurons (analogous to mitral cells) fire unique temporal sequences of spikes in response to different odorants. Spike patterns are repeatable across trials of repeated odorant presentation and are precisely phase locked to the odorant evoked local field potential (Laurent, 1996; Laurent et al., 1996; Wehr and Laurent, 1996). Interactions with local inhibitory interneurons are critical for temporal patterning in the antennal lobe (Wilson et al., 2004b; Wilson and Laurent, 3

14 2005), since blocking inhibitory neurotransmission affects projection neuron spike patterns and disrupts both the transient synchronization of projection neuron assemblies and the ability to discriminate between closely related odors (Stopfer et al., 1997). In response to sensory stimulation, mammalian mitral cells also generate temporally modulated action potential sequences unique for different odorants (Wellis et al., 1989). These spike clusters are phase-locked to the inspiratory rhythm (Macrides and Chorover, 1972; Sobel and Tank, 1993; Belluscio et al., 2002; Cang and Isaacson, 2003; Margrie and Schaefer, 2003) and produce prominent γ-frequency local field potential oscillations that reflect transient synchronization of mitral cell assemblies (Adrian, 1950; Eeckman and Freeman, 1990; Friedman and Strowbridge, 2003; Lagier et al., 2004). Spike timing is especially precise in repeated odorant applications that generate spike clusters with the same number of action potentials (Margrie and Schaefer, 2003). Thus, temporal coding of mitral cell spike times may be a critical aspect of the neuronal encoding of odorants. Synaptic Mechanisms of Mitral Cell Spike Patterning Relatively little is known about the mechanisms that enable mitral cells to respond to synaptic stimulation with precisely timed patterns of action potentials that are unique for different odorants. The different classes of network interactions between mitral cells and local inhibitory interneurons within the olfactory bulb are likely important in sculpting the incoming spatial pattern of mitral cell activation to produce an evolving spatio-temporal olfactory code (Fig. 1-2). Of these, the unique reciprocal dendrodendritic synapses between mitral cell secondary dendrites and granule cells (Price and Powell, 1970d; Shepherd and Greer, 1998) represent the most common synapse in the olfactory 4

15 bulb and are thus likely critical for mitral cell spike patterning and information processing in the olfactory bulb. Reciprocal dendrodendritic inhibition between mitral and granule cells Granule cells are axonless GABAergic interneurons organized in lamina immediately underneath the mitral cells (Price and Powell, 1970b; Shepherd and Greer, 1998). These interneurons have a single apical dendrite which bifurcates extensively after it crosses the mitral cell layer (Price and Powell, 1970b; Shepherd and Greer, 1998). Bifurcated granule cell dendrites in the external plexiform layer contain numerous large spines (termed gemmules) that contact mitral cell secondary dendrites (Price and Powell, 1970b, d; Shepherd and Greer, 1998). Gemmules possess both NMDA and non-nmda type glutamate recepetors (Sassoe-Pognetto and Ottersen, 2000) and receive glutamatergic input from mitral cell dendrites (Rall et al., 1966; Landis et al., 1974; Shepherd and Greer, 1998). In addition, they also have GABA-containing vesicles that mediate feedback GABA release onto mitral cells (Rall et al., 1966; Landis et al., 1974; Shepherd and Greer, 1998) (Fig. 1-3). Depolarization of mitral cell secondary dendrites causes calcium influx through voltage-sensitive calcium channels and subsequent glutamate release onto granule cell spines from presynaptic release sites immediately apposed to gemmules. Released glutamate opens glutamate receptors on gemmules which then produces depolarization and local calcium influx that drives feedback GABA release. Dendrodendritic inhibition has been studied extensively using whole-cell voltage clamp recordings from single mitral cells (Isaacson and Strowbridge, 1998; Schoppa et al., 1998; Halabisky et al., 2000; Isaacson, 2001). In the presence of tetrodotoxin (TTX) 5

16 to block fast Na-channel dependent action potentials and axo-dendritic transmitter release, short depolarizing steps evoke dendritic calcium influx that mediates glutamate release onto granule cell spines. The subsequent feedback GABA release from granule cells can be recorded as an asynchronous barrage of inhibitory postsynaptic currents (IPSCs) onto the recorded mitral cell (Isaacson and Strowbridge, 1998; Schoppa et al., 1998; Halabisky et al., 2000). These studies have revealed several fundamental properties of dendrodendritic synapses in the olfactory bulb. First, depolarization of a single mitral cell can produce both self inhibition (seen as feedback IPSCs) and lateral inhibition of neighboring mitral cells (seen as evoked IPSCs in adjacent neurons during paired whole-cell recordings) (Yokoi et al., 1995; Isaacson and Strowbridge, 1998). Second, unlike traditional synaptic transmission, dendrodendritic inhibition has a unique requirement for NMDA receptor activation to produce feedback GABA release (Isaacson and Strowbridge, 1998; Schoppa et al., 1998; Chen et al., 2000a; Halabisky et al., 2000; Isaacson, 2001). Dendrodendritic inhibition is strongly facilitated by removal of extracellular Mg 2+ (Isaacson and Strowbridge, 1998; Halabisky et al., 2000; Isaacson, 2001; Schoppa and Urban, 2003) or by pairing presynaptic glutamate release with postsynaptic granule cell depolarization (Chen et al., 2000a; Halabisky and Strowbridge, 2003) to remove the voltage dependent blockade of NMDA receptors by Mg 2+ (Mayer and Westbrook, 1987). In contrast, the selective NMDA receptor blocker D-APV largely abolishes dendrodendritic inhibition (Isaacson and Strowbridge, 1998; Schoppa et al., 1998; Chen et al., 2000a; Halabisky et al., 2000; Isaacson, 2001). These studies raise the intriguing possibility that NMDA receptors and presynaptic GABA containing vesicles may exist in close proximity such that calcium influx through open NMDA receptors 6

17 directly facilitates vesicle fusion. Early ultrastructural studies suggested that NMDA receptors and GABA release sites may be as far as 1 µm apart in granule cell spines, which implies that calcium microdomains around open NMDA receptors may be too far away from release sites to directly trigger neurotransmitter release (Price and Powell, 1970d). More recent studies using serial section electron microscopy, however, showed that NMDA receptors and GABA release sites can be as close as a few nanometers apart, lending credence to the view that NMDA receptor mediated calcium influx triggers GABA release (Woolf et al., 1991b; Sassoe-Pognetto and Otterson, 2000). Indeed, recent physiological studies studies have shown that calcium influx through NMDA receptors, in the absence of other sources of calcium influx, is sufficient for dendrodendritic inhibition (Chen et al., 2000b; Halabisky et al., 2000). Alternatively, voltage sensitive calcium channels involved in granule cell GABA release may be preferentially opened by an NMDA receptor mediated depolarization. The slow kinetics of NMDA receptor opening would produce a prolonged depolarization in granule cell spines, which might be required to elicit enough calcium influx to evoke GABA release. For example, since action potentials are not required for GABA release from granule cells, local release events not dependent on granule cell spiking would likely only reach a maximum depolarization of near 0 mv during granule cell activation (Koch and Poggio, 1983; Schoppa et al., 1998). These events, which depolarize gemmules to a lesser extent than during an action potential, would probably only evoke neurotransmitter release with sustained NMDA receptor mediated EPSPs. A prediction of this model is that prolonged AMPA receptor activation should support dendrodendritic inhibition when NMDA receptors are blocked. Indeed, addition of cyclothiazide (CTZ) 7

18 to inhibit AMPA receptor desensitization can restore dendrodendritic inhibition after blockade of NMDA receptors with APV (Isaacson, 2001). Action potential initiation in granule cells also depends critically on NMDA receptor activation (Schoppa et al., 1998; Schoppa and Westbrook, 1999). Prolonged NMDA receptor mediated EPSPs (unlike fast AMPA receptor mediated EPSPs) overcome the inhibitory effect of inactivating A-type potassium channel activation in granule cells (Schoppa and Westbrook, 1999). Spiking, while not required for GABA release in vitro, may be particularly important for gating lateral and self-inhibition in vivo by providing a global signal for GABA release from multiple granule cell spines (Isaacson and Strowbridge, 1998; Egger et al., 2003). Taken together, a picture emerges in which NMDA receptors have a unique and critical role in mediating dendrodendritic inhibition, while AMPA receptors may serve to provide a local depolarization that helps to relieve voltage dependent Mg-block of NMDA receptors and recruit additional calcium influx through voltage-sensitive calcium channels. Calcium influx through NMDA receptors is sufficient to trigger feedback GABA release in vitro; however, the relative importance of NMDA receptor mediated calcium influx versus calcium influx through voltage gated calcium channels in triggering dendrodendritic feedback inhibition in vivo remains unclear. Calcium influx through NMDA receptors may also facilitate transmitter release evoked by voltage-gated calcium channel activation during repetitive inputs. Gating of dendrodendritic inhibition under normal physiological conditions Because of its requirement for NMDA receptors, most studies investigating dendrodendritic inhibition have used Mg 2+ -free extracellular solutions to maximize 8

19 permeation through the NMDA receptor. These studies, however, do not address how NMDA receptors might be activated under normal physiological conditions, where Mg 2+ blocks ion flux through these channels. One possibility is that GABA release from granule cells might require the coincident activation of numerous dendrodendritic synapses to provide enough depolarization to unblock NMDA receptors. Alternatively, if dendrodendritic mitral-granule synapses facilitate, GABA release from single gemmules may occur after a train of mitral cell action potentials. Finally, in addition to dendrodendritic synapses, granule cells have numerous excitatory axo-dendritic synaptic contacts onto proximal spines located on granule cell primary dendrites before they bifurcate in the EPL. Glutamate release onto these spines may trigger backpropagating action potentials that transiently unblock NMDA receptors in gemmules, allowing reciprocal dendrodendritic inhibition in response to glutamate release from mitral cell secondary dendrites. The identity and functional properties of these proximal axodendritic inputs is unclear; they may be due to mitral cell axon collaterals, feedback projections from cortical efferents, or centrifugal inputs from other brain nuclei. Recent studies suggest that coincident activation of proximal axodendritic and distal dendrodendritic inputs gates reciprocal feedback inhibition onto mitral cells. First, pairing intracellular calcium uncaging in a single mitral cell with gamma-frequency (50 Hz) extracellular stimulation of proximal inputs produced a prolonged barrage of feedback IPSPs that was not seen with uncaging or granule cell layer stimulation alone (Chen et al., 2000a). Second, pairing a mitral cell action potential with 50 Hz granule cell layer stimulation results in a prolonged shunting inhibition and decrease in mitral cell input resistance that is blocked by the selective GABA A receptor antagonist picrotoxin 9

20 (Halabisky and Strowbridge, 2003). Interestingly, lower frequency stimulus trains are ineffective at gating dendrodendritic inhibition, suggesting that proximal granule cell inputs may facilitate with stimuli occurring at the frequency of local field potentials in the olfactory bulb recorded during odor perception. Finally, pairing focal AMPA application in the granule cell layer with a depolarizing voltage step in mitral cells potentiates feedback dendrodendritic inhibition in normal Mg 2+ ACSF (Halabisky and Strowbridge, 2003). Regulation of action potential propagation in mitral and granule cell dendrites Numerous questions remain about reciprocal dendrodendritic synapses and how they might be regulated in vivo. First, many fundamental properties of dendritic glutamate release from mitral cells remain unexplored, and it is unclear whether dendritic release is similar or distinct from conventional synaptic transmission in other brain areas. The probability of vesicle fusion in response to a single back-propagating action potential into mitral cell secondary dendrites is unknown. In addition, the reliability and regulation of action potential backpropagation into mitral cell secondary dendrites is only beginning to be understood. Both primary and secondary dendrites in mitral cells possess active Na + and Ca 2+ conductances which support the propagation of somatically generated action potentials. Action potentials backpropagate relatively unattenuated into primary dendrites, and very strong (non-physiological) inhibitory inputs onto mitral cells can shift the site of spike initiation from the axon hillock to distal dendritic sites near glomerular tufts (Chen et al., 1997). At mitral cell secondary dendrites, however, the extent of backpropagation and its regulation by synaptic inputs is still unclear. A recent study (Margrie et al., 2001) suggests that action potentials decrement as they spread through 10

21 secondary dendrites, limiting the spatial extent of inhibition to one or a few glomerular modules. This model, however, is difficult to reconcile with anatomical data that show dendrodendritic synapses at distal sites far away from the mitral cell soma (Price and Powell, 1970a). Another study suggests that action potentials can propagate fully down secondary dendrites, but are attenuated relative to somatic action potential amplitude (Lowe, 2002). A third study reports full unattenuated backpropagation of somatically generated spikes down the entire secondary dendritic tree (Xiong and Chen, 2002). Inhibitory inputs from granule cells can limit the extent of spike propagation, suggesting that spatial domains of lateral inhibition can be dynamically shifted during odor processing (Xiong and Chen, 2002). Further work is required to clarify the regulation of spike propagation in presynaptic mitral cell dendrites, and how spiking is coupled to glutamate release at dendritic presynaptic active zones. The factors regulating action potential initiation and propagation at granule cells, and their relation to transmitter release, are virtually unknown. Granule cells fire action potentials in response to depolarizing current injection and synaptic activation both in vitro (Schoppa et al., 1998; Schoppa and Westbrook, 1999; Halabisky et al., 2000) and in vivo (Wellis and Scott, 1990; Cang and Isaacson, 2003), however, the site of action potential initiation, the efficacy of backpropagation into granule cell dendrites, and the importance of action potential firing for dendrodendritic neurotransmitter release remain a mystery. One possibility is that granule cell spines may function as independent units during odor processing that do not require spikes to trigger neurotransmitter release. A recent study showed localized calcium accumulations in gemmules following mitral cell glutamate release, consistent with localized dendritic activation of granule cells during 11

22 odor processing (Egger et al., 2005). Other theoretical studies suggest that granule cell spines and dendrites are biochemically and electrically compartmentalized and therefore may be largely unaffected by somatic changes in membrane potential (Woolf et al., 1991a; Woolf and Greer, 1994). The persistence of reciprocal dendrodendritic inhibition after blocking sodium channels with TTX supports this notion (Isaacson and Strowbridge, 1998; Halabisky et al., 2000); however, local transmitter release after granule cell spine activation should largely provide only self-inhibition of mitral cells. Strong activation of granule cell dendrites may elicit action potentials that propagate to the soma and throughout the dendritic tree, providing sufficient depolarization for global GABA release and lateral inhibition. Alternatively, activating multiple granule cell spines may provide a large depolarization that passively travels to the cell body, eliciting somatically generated spikes that then backpropagate into dendrites and facilitate transmitter release. Finally, global GABA release from multiple gemmules may be gated by synaptic activity at proximal axodendritic inputs onto granule cells. These inputs, located close to the cell body, may selectively elicit somatic action potentials that then travel down dendrites to evoke transmitter release simultaneously from many granule cell spines. Granule cell dendrites have active conductances that support calcium spike propagation following a somatic action potential (Egger et al., 2003); however, further work is required to unravel the regulation of transmitter release from granule cells under normal physiological conditions during odor processing. Plasticity at dendrodendritic synapses Many forms of olfactory learning require changes in olfactory bulb function (Kaba et al., 1994; Kaba and Nakanishi, 1995; Brennan and Keverne, 1997; Wilson et al., 12

23 2004a), however, the cellular bases for synaptic plasticity in the olfactory bulb are largely unexplored. Dendrodendritic synapses may represent an important locus for olfactory bulb plasticity. Repeated tetanic stimulation of olfactory nerve inputs increases coherent γ-frequency oscillations in the bulb through increased activation of granule cells (Friedman and Strowbridge, 2003). Additionally, in insects, repeated odor presentations produces increased synchronous activity through changes in the strength of inhibitory local neurons (Stopfer and Laurent, 1999). The high concentration of NMDA receptors at granule cell spines suggests that long-term potentiation of dendrodendritic transmission may be an important mechanism for synaptic plasticity in the olfactory bulb. In many brain regions, including the hippocampus (Bliss and Collingridge, 1993), cerebral cortex (Katz and Shatz, 1996; Feldman et al., 1999; Feldman, 2000), amygdala (Rodrigues et al., 2004; Rumpel et al., 2005), and thalamus (Mooney et al., 1993), NMDA receptors function as coincidence detectors whose activation induces long-term enhancement of synaptic strength. Pairing presynaptic glutamate release with postsynaptic depolarization (usually by coincident activation of pre- and postsynaptic neurons) (Gustafsson et al., 1987; Bi and Poo, 1998) removes Mg 2+ -dependent voltage block of NMDA receptors and allows calcium influx into postsynaptic spines. This calcium influx initiates a signal transduction cascade that leads to synaptic strengthening, first by inserting more AMPA receptors into the postsynaptic membrane (Bredt and Nicoll, 2003; Collingridge et al., 2004) and then by producing long lasting changes in spine morphology and number (Bolshakov et al., 1997; Luscher et al., 1999) 13

24 Long-term potentiation at central synapses often occurs through the insertion of AMPA receptors into postsynaptic spines that originally contain only NMDA receptors (Isaac, 2003). These silent synapses are normally non-functional since NMDA receptors are largely blocked at resting membrane potential; AMPA receptor insertion converts silent synapses to functional ones and increases the efficacy of presynaptic transmitter release on postsynaptic firing. Silent synapses play a role in both adult synaptic plasticity (Isaac et al., 1995; Liao et al., 1995; Isaac et al., 1996a) and the normal activity dependent maturation of central synapses during development (Durand et al., 1996; Isaac et al., 1997). In both Schafer collateral inputs onto hippocampal CA1 pyramidal cells (Durand et al., 1996) and thalamocortical inputs (Isaac et al., 1997), postsynaptic spines initially contain only NMDA receptors. AMPA receptors are subsequently incorporated by insertion following NMDA receptor mediated calcium influx. In the olfactory bulb, the cellular bases for long-term synaptic plasticity remain a mystery. The sequential incorporation of NMDA and AMPA receptors during normal granule cell development is unclear, leaving open the possibility of AMPA receptor silent synapses. Granule cells, however, continue to be produced through adulthood in the subventricular zone and migrate into the olfactory bulb to be incorporated into existing olfactory bulb circuits (Lois and Alvarez-Buylla, 1994; Alvarez-Buylla and Garcia- Verdugo, 2002; Petreanu and Alvarez-Buylla, 2002; Carleton et al., 2003; Lledo et al., 2006). These adult-born granule cells initially contain AMPA receptors and later incorporate NMDA receptors (Carleton et al., 2003; Lledo et al., 2006), implying that a significant fraction of granule cells are actually NMDA-receptor silent in adulthood. This 14

25 suggests that long-term synaptic enhancement at dendrodendritic synapses, if it exists, may occur through a mechanism that is qualitatively different from canonical long-term potentiation. Despite these data, conclusive tests for silent synapses in the olfactory bulb have not been published by any laboratory. The existence of silent synapses between granule and mitral cells is one of the main questions that I will address in my thesis. The short term plasticity of dendrodendritic synapses is also unclear. Many central synapses exhibit profound short-term (seconds to minutes) changes (such as depression and facilitation) in synaptic efficacy with repeated stimulation (Zucker and Regehr, 2002; Blitz et al., 2004). These short-term changes in synaptic function have important effects on the integration of synaptic inputs and subsequent spike generation. A recent study investigated short term plasticity at both sides of the dendrodendritic synapse and found that granule to mitral cell connections were largely depressing, while mitral to granule cell connections showed either depression or facilitation (Dietz and Murthy, 2005). These authors concluded that two types of granule cells, one with depressing inputs and another with facilitating inputs, exist in the olfactory bulb; however it is possible that these two types of synapses reflect functionally distinct populations on individual granule cells. Since granule cells contain anatomically distinct dendrodendritic and axodendritic excitatory inputs, it is tempting to speculate that these anatomically defined classes may have different functional properties. Other synaptic mechanisms of mitral cell spike patterning In addition to reciprocal dendrodendritic inhibition through granule cells, a variety of other network interactions shape mitral cell outputs. In the external plexiform layer, extrasynaptic NMDA autoreceptors on mitral cell secondary dendrites are activated 15

26 during glutamate release events in mitral cells and may provide feedback excitation during spiking (Aroniadou-Anderjaska et al., 1999; Isaacson, 1999; Friedman and Strowbridge, 2000). In the granule cell layers, a unique class of interneurons originally described by Blanes (Blanes, 1890) receives excitatory input from mitral cells and inhibits granule cells (Pressler and Strowbridge, 2006). Blanes cells are morphologically distinct from granule cells and possess extensive axonal processes that mediate longrange feedforward inhibition of distantly located granule cells. Interestingly, strong activation of these neurons produces regular, persistent spiking lasting many minutes. This persistent spiking may tonically suppress granule cell mediated inhibition of mitral cells during odor processing, or may temporally pattern granule cell outputs which in turn will dynamically regulate mitral cell firing patterns (Pressler and Strowbridge, 2006). In the glomerular layer, neurons surrounding mitral cell glomerular tufts make synaptic contacts with both mitral/tufted cells and olfactory nerve terminals (Shepherd, 1972; Shepherd and Greer, 1998). These juxtaglomerular neurons sculpt both sensory inputs onto mitral cells and mitral cell response patterns to sensory stimulation. Periglomerular neurons are a prominent class of inhibitory juxtaglomerular cells that receive excitatory glutamatergic input from olfactory nerve terminals and mitral cell dendrites (Shepherd, 1972; Shepherd and Greer, 1998). Like granule cells, these cells make reciprocal dendrodendritic synapses with mitral and tufted cell dendrites (Shepherd, 1972; Shepherd and Greer, 1998). Activation of mitral cell dendritic tufts by olfactory nerve inputs presumably evokes glutamate release onto periglomerular cells, which in turn provides feedback inhibition onto activated mitral cells. Because periglomerular neurons are less numerous than granule cells, and since direct mitral cell depolarization 16

27 results in larger amounts of glutamate release from secondary dendrites than glomerular tufts (Shepherd and Greer, 1998), reciprocal dendrodendritic inhibition of mitral cells through periglomerular cells has not been demonstrated. However, depolarization of single voltage-clamped external tufted neurons which, unlike mitral cells, do not possess secondary dendrites results in an asynchronous barrage of feedback IPSCs that is virtually indistinguishable, albeit smaller in amplitude, from granule cell mediated DDI (Murphy et al., 2005). A subclass of periglomerular neurons also receives direct excitatory input from olfactory sensory neurons, which will likely produce feedforward inhibition of mitral/tufted cell dendrites (Murphy et al., 2005). In addition to inhibiting mitral and tufted cells, periglomerular cells can also inhibit themselves (Smith and Jahr, 2002). Depolarization of a single periglomerular neuron evokes GABA release that activates dendritic GABA A receptors, resulting in a prolonged inhibitory current. Periglomerular neuron self-inhibition is thought to occur through spillover of released GABA onto neighboring GABA A -receptors, rather than through autaptic synaptic connections (Smith and Jahr, 2002). Interestingly, because of the high-internal chloride concentration in periglomerular cells, self-inhibitory currents are depolarizing. Despite this, they inhibit action potential generation through shunting. Finally, GABA spillover from periglomerular cells activates GABA B receptors on olfactory nerve terminals (Murphy et al., 2005). Activation of these metabotropic receptors reduces glutamate release from olfactory sensory neuron axons, thereby damping the excitatory drive from the olfactory nerve onto mitral and tufted cells. In addition to short-range synaptic interactions within glomeruli mediated by periglomerular cells, other juxtaglomerular neurons mediate long-range interactions 17

28 between distant glomeruli. These juxtaglomerular short-axon cells are activated by olfactory nerve input and make glutamatergic synapses with inhibitory periglomerular cells in neighboring glomeruli (Pinching and Powell, 1971a, c, 1972). Despite their name, recent studies suggest that short-axon cells in fact possess an extensive axon plexus that forms synapses with inhibitory interneurons located in distant glomeruli up to hundreds of micrometers to a millimeter away (Aungst et al., 2003). Activation of these neurons suppresses activity in neighboring glomerular modules, and is thought to mediate lateral inhibition and an initial contrast enhancement of spatial patterns of glomerular activation (Aungst et al., 2003). Finally, specialized chemical and electrical synaptic interactions exist between mitral cells projecting to the same glomerulus (Schoppa and Westbrook, 2001, 2002). These interactions synchronize spiking patterns from mitral cells from a single glomerular module, and thus ensure the coordinated response of groups of mitral cells receiving common olfactory sensory neuron input. Olfactory sensory nerve evoked action potentials in mitral cells elicit glutamate release from glomerular tufts which activates AMPA autoreceptors on neighboring mitral cell tufts. This autoreceptor potential is propagated electrically to coupled mitral cells projecting to the same glomerulus through gap junctions, ensuring coordinated responses of glomerular modules following sensory inputs (Schoppa and Westbrook, 2002). Intrinsic Mechanisms of Mitral Cell Spike Patterning In addition to synaptic mechanisms, mitral cells also possess intrinsic membrane properties that sculpt and pattern their responses evoked by olfactory sensory neuron activation. Intrinsic electrophysiological properties are crucial for synaptic integration 18

29 and response properties in a variety of neurons (Johnston et al., 1999; Bekkers, 2000). Previous work has shown that mitral cells fire clusters of spikes interspersed with periods of fast gamma-frequency subthreshold oscillations in response to DC current injection (Chen and Shepherd, 1997; Desmaisons et al., 1999). Spike clustering is most likely due to intrinsic membrane properties of mitral cells, since it persisted in the presence of ionotropic glutamate and GABA receptor blockers. This intrinsic behavior is similar to spike clustering in response to DC current injection seen in neurons of other brain areas, including stellate cells of the medial entorhinal cortex (Alonso and Llinas, 1989; Alonso and Klink, 1993; Klink and Alonso, 1993) and inhibitory interneurons of the basal forebrain (Alonso et al., 1996; Wang, 2002). Intermittent firing is thought to be critical for allowing these cells to serve as pacemakers of coherent network oscillations, especially in the theta-band (3-10 Hz) and gamma-band frequency ranges (Wang, 2002). While these intrinsic spike patterning processes have been documented in a variety of cell types, the mechanisms that underlie them are still unclear. Previous theoretical and experimental work suggests that spike clustering reflects the dynamic interplay of intrinsic inward and slowly-inactivating outward currents (Wang, 1993, 2002). Depolarization initially activates outward potassium conductances, which inhibit action potential generation. An action potential cluster occurs after a significant fraction of potassium channels inactivate. Spike clusters then terminate because of cumulative recovery from inactivation of outward currents during action potential afterhyperpolarizations. Subthreshold inward currents may bring neurons into a critical membrane potential window where alternating epochs of potassium current activation and inactivation can occur (Wang, 1993). A specific prediction of this model 19

30 is that the duration and frequency of spike clusters reflects the unique kinetics of activation, inactivation and recovery of voltage dependent potassium channels in mitral cells. The specific conductances mediating spike clustering behavior in mitral cells are not well characterized; however, mitral cells express a variety of inactivating potassium currents that may regulate clustered action potential discharges (Wang et al., 1996a; Wang et al., 1996b; Bischofberger and Jonas, 1997). The slowly inactivating potassium channel subunit Kv1.3 is expressed strongly in mitral cell bodies, the external plexiform layer (likely in mitral cell secondary dendrites) and glomerular tufts (Kues and Wunder, 1992; Fadool et al., 2000) and constitutes the dominant outward conductance in cultured olfactory bulb neurons (Fadool and Levitan, 1998). Kv1.3 knockout mice have dramatic alterations in olfactory-mediated behaviors and glomerular anatomy; in addition, cultured neurons from olfactory bulbs of knockout animals show profound alterations in their voltage responses to current steps (Fadool et al., 2004). Thus, these slowly inactivating currents represent an attractive candidate for controlling the intrinsic spiking behavior of mitral cells. Mitral cell intrinsic conductances may also regulate action potential generation following inhibitory inputs. Previous studies showed that intrinsic subthreshold oscillations are reset by spontaneous IPSPs or hyperpolarizing current steps, and that oscillation resets are frequently accompanied by rebound action potentials (Desmaisons et al., 1999). This suggests that granule cell inputs, rather than only inhibiting action potential generation, may also promote spiking depending on the depolarization state of mitral cells. The ability of IPSPs to reset spiking and subthreshold oscillations in mitral 20

31 cells is not dependent on other synaptic inputs. However, the specific intrinsic conductances mediating this behavior are unknown and are the subject of a chapter of this dissertation. Overview of Dissertation For my Ph.D. thesis, I have investigated both intrinsic and synaptic properties of olfactory bulb neurons and their effects on regulating olfactory bulb output. I first analyzed the intrinsic membrane currents that regulate action potential generation in mitral cells in response to depolarizing and hyperpolarizing inputs. I found that the unique spike clustering behavior of mitral cells is controlled by a single class of slowlyinactivating (I D -like) voltage gated potassium currents. Spike clustering and subthreshold oscillations in mitral cells depend on the interplay between slowly-inactivating potassium channels and subthreshold persistent sodium currents. I D -like currents produce variably timed spike discharges in mitral cells in response to depolarizing current steps. In response to phasic depolarizations that mimic mitral cell excitatory inputs during sniffing, however, I D -like currents ensure precise spike timing over repeated stimuli and are thus likely important for temporal coding strategies employed by olfactory bulb slices. I next investigated the intrinsic mechanisms that regulate spike generation following hyperpolarizing IPSPs onto mitral cells. I found that mitral cells readily fire rebound action potentials in response to IPSPs or small hyperpolarizing stimuli. Rebound spiking is mediated by recovery from inactivation of sodium currents and occurs in a narrow window of depolarized membrane potentials. Large IPSPs and hyperpolarizing steps recruit a distinct, fast-inactivating A-type potassium channel that 21

32 inhibits spike generation. The interplay of these two opposing intrinsic currents thus dynamically regulates the effect of granule cell mediated inhibitory inputs. Finally, I investigated the functional properties of excitatory glutamatergic inputs onto granule cells. I found that granule cells possess two functionally distinct classes of excitatory inputs: distal dendrodendritic inputs from mitral cells which display prominent paired pulse depression, and proximal axodendritic inputs from cortical feedback projections that facilitate. Neither distal nor proximal inputs showed any evidence for silent synapses, suggesting that long term synaptic plasticity at granule cell excitatory inputs occurs through novel cellular mechanisms. Cortical feedback projections may gate the activation of dendrodendritic feedback inhibition in vivo. These results have important implications for how the olfactory bulb processes incoming sensory information. 22

33 Figure 1-1. Glomerular organization of sensory inputs in the olfactory bulb Olfactory sensory neurons have stereotyped anatomical connections with output neurons in the olfactory bulb. In the olfactory epithelium, sensory neurons expressing different G-protein coupled receptors are semi-randomly distributed in four distinct zones. All olfactory sensory neurons expressing the same G-protein coupled receptor synapse onto the same group of mitral cells at discrete glomeruli. (From (Mori et al., 1999b)) 23

34 Figure

35 Figure 1-2. Local circuits and synaptic processing in the olfactory bulb Diagram shows multiple classes of synaptic interactions within the olfactory bulb. Olfactory sensory neurons synapse onto mitral and tufted cells at glomeruli. Mitral cell activation is sculpted and transformed by synaptic interactions with periglomerular cells and other juxtaglomerular neurons (not shown) in the glomerular layer and by network interactions between mitral cells and granule cells at dendrodendritic synapses. (From (Mori et al., 1999b)) 25

36 Figure

37 Figure 1-3. Ultrastructure of olfactory dendrodendritic synapses Transmission electron micrograph shows reciprocal arrangement of dendrodendritic synapses between mitral and granule cells. The mitral cell secondary dendrite is seen at the bottom, while the granule cell spine is at the top. Both mitral and granule cells contain presynaptic active zones with small, clear synaptic vesicles containing glutamate (at mitral cells) and GABA (at granule cells). They also contain postsynaptic densities containing glutamate receptors (at granule cell spines) and GABA receptors (on mitral cell dendrites) respectively. Arrows show direction of neurotransmitter release at both sides of the reciprocal synapse (From Peters et al., 1991). 27

38 Granule Cell Spine Mitral Cell Dendrite Figure

39 Chapter 2 : Phasic Stimuli Evoke Precisely Timed Spikes in Intermittently Discharging Mitral Cells Summary Mitral cells, the principal cells of the olfactory bulb, respond to sensory stimulation with precisely timed patterns of action potentials. By contrast, the same neurons generate intermittent spike clusters with variable timing in response to simple step depolarizations. We made whole cell recordings from mitral cells in rat olfactory bulb slices to examine the mechanisms by which normal sensory stimuli could generate precisely-timed spike clusters. We found that individual mitral cells fired clusters of action potentials at Hz interspersed with periods of subthreshold membrane potential oscillations in response to depolarizing current steps. Tetrodotoxin (1 µm) blocked a sustained depolarizing current and fast subthreshold oscillations in mitral cells. Phasic stimuli that mimic trains of slow EPSPs that occur during sniffing evoked precisely timed spike clusters in repeated trials. The amplitude of the first simulated EPSP in a train gated the generation of spikes on subsequent EPSPs. 4-aminopyridine sensitive K + channels are critical to the generation of spike clusters and reproducible spike timing in response to phasic stimuli. Based on these results, we propose that spike clustering is a process that depends on the interaction between a 4-AP sensitive K + current and a subthreshold TTX-sensitive Na + current; interactions between these currents may allow mitral cells to respond selectively to stimuli in the theta frequency range. These intrinsic properties of mitral cells may be important for precisely timing spikes evoked by phasic stimuli that occur in response to odor presentation in vivo. 29

40 Introduction Olfactory information is encoded through spatio-temporal patterns of activity in mitral cells located in the olfactory bulb (Fig. 2-1A). Intracellular recordings from salamander olfactory bulb in vivo have shown that individual mitral cells fire clusters of action potentials interspersed with periods of inhibition in response to sensory stimulation (Hamilton and Kauer, 1985, 1989). Such temporal patterning of spikes has also been shown for projection neurons (analogous to vertebrate mitral cells) of the insect antennal lobe (Laurent, 1996; Laurent et al., 1996; Wehr and Laurent, 1996). In this system, individual projection neurons were found to fire unique temporal sequences of spikes in response to different odorants. Spike patterns were repeatable across trials of repeated odor presentation and were precisely phase locked to the odorant evoked local field potential. In response to sensory stimulation, mammalian mitral cells generate spike clusters that are phase-locked to the inspiratory rhythm (Macrides and Chorover, 1972; Sobel and Tank, 1993; Cang and Isaacson, 2003; Margrie and Schaefer, 2003). Spike timing is especially precise in repeated odorant applications that generate clusters with the same number of spikes (Margrie and Schaefer, 2003). Thus, temporal coding of mitral cell spike times may be a critical aspect of the neuronal encoding of odorants. Relatively little is known about the mechanisms that enable mitral cells to respond to synaptic stimulation with precisely timed patterns of action potentials that are unique for different odorants. One possibility is that interactions between mitral cells and local inhibitory interneurons within the olfactory bulb might sculpt the incoming spatial pattern of mitral cell activation and produce an evolving spatio-temporal olfactory code. Reciprocal dendrodendritic synapses between mitral cells and granule cells can produce 30

41 feedback inhibition onto activated mitral cells (Jahr and Nicoll, 1980, 1982; Isaacson and Strowbridge, 1998) and laterally inhibit neighboring mitral cells (Yokoi et al., 1995; Isaacson and Strowbridge, 1998). Recurrent dendrodendritic inhibition also can modulate the pattern of mitral cell activity evoked by phasic stimuli (Halabisky and Strowbridge, 2003). In addition, extrasynaptic NMDA autoreceptors on mitral cell dendrites (Aroniadou-Anderjaska et al., 1999; Isaacson, 1999; Friedman and Strowbridge, 2000) may provide feedback excitation during spiking. The interaction of these inhibitory and excitatory mechanisms could modulate the pattern of spikes in single mitral cells to produce temporal codes for odors. In addition to synaptic mechanisms, mitral cells may possess intrinsic membrane properties that sculpt and pattern their responses evoked by olfactory sensory neuron activation. Previous work has shown that mitral cells fire clusters of spikes interspersed with periods of fast gamma-frequency subthreshold oscillations in response to DC current injection (Chen and Shepherd, 1997; Desmaisons et al., 1999). Spike clustering is most likely due to intrinsic membrane properties of mitral cells, since it persisted in the presence of ionotropic glutamate and GABA receptor blockers. This intrinsic behavior is strikingly similar to spike clustering in response to DC current injection seen in neurons of other brain areas, including stellate cells of the medial entorhinal cortex (Alonso and Llinas, 1989; Alonso and Klink, 1993; Klink and Alonso, 1993) and inhibitory interneurons of the basal forebrain (Alonso et al., 1996; Wang, 2002). This behavior has been proposed to be critical for allowing these cells to serve as pacemakers for the theta rhythm. 31

42 Using whole cell recordings, we found that mitral cells generated intermittent, irregularly-timed spike clusters at slow theta frequencies (1-5 Hz). By constrast, phasic current stimuli mimicking the trains of slow EPSPs that occur during sniffing evoked precisely timed spike clusters. Both spike clustering during step depolarizations and precise timing evoked by phasic stimuli are likely due to the interplay of 4-AP-sensitive potassium current and a subthreshold inward current. The ability of mitral cells to fire precisely timed spikes in response to phasic stimuli suggests that their intrinsic membrane properties may allow them to act as filters converting incoming olfactory sensory neuron activity into precise temporal patterns of spikes that are relayed to higher brain centers. Materials and Methods Horizontal slices (300 µm) through the olfactory bulb were prepared from anesthetized (ketamine, 140 mg/kg ip) P14-25 Sprague-Dawley rats using a modified Leica (Nussloch, Germany) VT1000S vibrotome, as described previously (Isaacson and Strowbridge, 1998; Halabisky et al., 2000). Olfactory bulb slices were incubated at 30 O C for 25 min then maintained submerged at room temperature until needed. Whole-cell patch-clamp recordings were made in mitral cells visualized under infrared-differential interference contrast optics (Zeiss Axioskop 1 FS) using an Axopatch 1D amplifier (Axon Instruments). During recordings, olfactory bulb slices were superfused with artificial cerebrospinal fluid (ACSF) that contained (in mm): NaCl 124, KCl 3, NaH 2 PO , NaHCO 3 26, dextrose 10, CaCl 2 2.5, and MgSO 4 1.2, equilibrated with 95% O 2 /5% CO 2 and warmed to 30 C (flow rate, 1-2 ml/min). A modified ACSF solution was employed when making slices and in the holding chamber that contained reduced CaCl 2 32

43 (1 mm) and elevated MgSO 4 (3 mm). Patch electrodes used for current clamp recordings (3-5 MΩ resistance) contained (in mm): K-methylsulfate 140, NaCl 8, HEPES 10, EGTA 0.2, MgATP 4, Na 3 GTP 0.3, and phosphocreatine 10. All recording were obtained in the presence of NBQX (5 µm) and D-APV (25 µm) in the bath solution to block ionotropic glutamate receptors. Voltage records were low-pass filtered at 2 khz and then digitized at 5 khz using a 16-bit A/D converter (ITC-18, Instrutech). In some experiments, a current injection waveform consisting of a train of 2-8 temporally-overlapping EPSP-like waveforms was injected into mitral cells. Each simulated EPSP in the train was generated using a single alpha function with a decay time constant of ms. This stimulus train was modeled after respiration-evoked calcium and voltage oscillations recorded from mitral cell glomerular tufts in vivo (Charpak et al., 2001). In these in vivo experiments, oscillations at the beginning of odor application tended to be larger than those occurring later. For this reason, in many of our experiments, we used simulated EPSP trains where the last three EPSPs were gradually reduced in amplitude (see Fig. 2-5A). Electrophysiological data were recorded and analyzed using custom software written in Visual Basic 6 (Microsoft) and Origin 7 (OriginLab). Spike latencies were determined using a threshold crossing (0 mv) algorithm implemented in Origin and confirmed by visual inspection in most cells. Variability in spike timing across trials was measured by calculating the standard deviation (S.D.) of the first spike latency from repeated current stimulus presentations (10-30 trials). Spike timing variability is generally assayed either by measuring the regularity or the reproducibility of spiking at particular times across repeated trials. Regularity can be measured either by the 33

44 coefficient of variation (C.V.) of the interspike interval distribution or the ratio of the variance of spike count to the mean spike count in a fixed time interval (the Fano Factor) (Dayan and Abbott, 2001). Because mitral cell firing is intrinsically intermittent, these measures were not well suited for our analyses of spike timing. Reproducibility of spike timing has been studied previously both in vitro (Mainen and Sejnowski, 1995; Nowak et al., 1997; Harsch and Robinson, 2000) and in vivo (Reinagel and Reid, 2002) by measuring the precision (S.D. of spike times) for spikes that occur within repeatable spike events. Repeatable spike events often are defined by analyzing the peristimulus time histogram (PSTH) and identifying peaks in the PSTH that exceed a defined threshold value (for example, three times the mean spike rate). These methods also are less suited to analyze spike timing in mitral cells which generate clusters of near-regular firing intermittently. In order to separate inter-cluster and within-cluster sources of variability, we choose to focus on the variability (S.D.) of the latency to the first spike cluster. Membrane potentials indicated are not corrected for the liquid junction potential. All chemicals were obtained from Sigma (St. Louis, MO) except for tetrodotoxin (TTX; Calbiochem). Data are shown as the mean ± SEM. Statistical significance was determined using paired t-tests except where noted. Results Mitral cells generate clusters of action potentials intermittently in response to stepwise depolarizing current injections. Intermittent firing was maintained across a large (2-fold) variation in current step amplitude (Fig. 2-1B). These pauses did not reflect recurrent synaptic interactions since ionotropic glutamate receptor blockers (5 µm NQBX and 25 34

45 µm D-APV) were included in these and subsequent experiments. While the number of spikes per cluster and intra-cluster firing frequency increased with current step amplitude, altering the depolarizing stimulus intensity had little effect on the mean pause duration (Fig. 2-1C). This characteristic clustered spiking pattern was observed at a range of resting membrane potentials (from -75 to -55 mv, data not shown); in these voltage ranges there was almost no discernible effect of resting potential on spike patterning. We were able to induce tonic firing with current steps only by holding mitral cells very near spike threshold, where subthreshold oscillations were prominent (see below). Intermittent firing has been reported previously in mitral cells (Chen and Shepherd, 1997; Desmaisons et al., 1999; Friedman and Strowbridge, 2000) and other cell types (Alonso and Llinas, 1989; Llinas et al., 1991; Alonso and Klink, 1993; Gutfreund et al., 1995; Pedroarena and Llinas, 1997; Bracci et al., 2003); however, the mechanism underlying this behavior is unclear. Mitral cells also generate prominent subthreshold membrane potential oscillations near firing threshold (Fig. 2-1D-E). Large subthreshold oscillations are often correlated with intermittent firing in a variety of neurons (mitral cells: (Desmaisons et al., 1999); entorhinal neurons: (Klink and Alonso, 1993); thalamocortical projection neurons: (Pedroarena and Llinas, 1997)). Computational models of subthreshold oscillations suggest they are mediated by opposing low-threshold inward and outward currents (Wang, 1993, 2002). We first sought to test whether mitral cells generate a sustained Na + current and if this current is involved in subthreshold oscillations. We found that bath application of tetrodotoxin (TTX; 1 µm) reversibly reduced the steady-state depolarization produced by step current injection by 5.5 ± 0.7 mv (Fig. 2-1D; n = 5 35

46 cells). This effect could be due to either blocking persistent Na + currents or to subthreshold inactivating Na + currents. Tetrodotoxin also blocked subthreshold oscillations (Fig. 2-1E; membrane potential variance decreased from 0.57 ± 0.1 mv 2 at mv to ± mv 2 at mv; p < 0.05), suggesting that these oscillations may result from the interaction between K + currents and subthreshold Na + currents. Since TTX also blocks action potentials, it is difficult to determine directly if these Na + currents are also necessary for intermittent firing. The timing of spike clusters was highly variable even when mitral cells were activated by constant current steps (first spike S.D. = 169 ± 32 ms; n = 11 mitral cells; Fig. 2-2A). The distribution of interspike intervals shows two distinct peaks: one at < 100 ms that reflects intervals within spike clusters and one centered at 470 ms that represents inter-cluster pauses (Fig. 2-2B; n = 5 cells). Mitral cells typically generated a small afterhyperpolarization (AHP) following each spike cluster. As shown in Fig. 2-2C, these cluster AHPs decay exponentially with a mean time constant of 202 ± 40 ms (n = 4 cells) and were associated with a transient decrease in input resistance estimated by responses to small hyperpolarizing test pulses (73.9 ± 7.7 % of pre-cluster input resistance; n = 3 cells). This decrease in input resistance at the end of a spike discharge suggests that the buildup of an outward current may be responsible for cluster termination and may contribute to low frequency of short inter-cluster pauses (less than 250 ms.) Spike timing remained highly variable in repeated responses to both small and large amplitude step current injections (Fig. 2-3A; mean first spike latency S.D. for weak stimuli = 365 ± 48 ms; mean S.D. for strong stimuli = 273 ± 47 ms; n = 5 cells; not statistically significant). Imprecise firing in mitral cells was not limited to the first spike 36

47 cluster; the duration of the first pause between spike clusters also was variable (mean S.D. = 195 ± 27 ms; n = 6 cells) and was not correlated with the latency to the first spike (R = ± 0.09; n = 4 cells). Figure 2-3B shows the superposition of two responses to the same current step in one mitral cell. The initial membrane potential trajectory of both responses was similar despite the 420 ms difference in first spike latencies. We found no correlation between first spike latency and resting membrane potential (R = ± 0.092; n = 9 cells) or input resistance (R = ± 0.07; n = 9 cells). We tested whether the large variability in first spike latencies reflected differences availability of voltagedependent channels by evoking depolarizing step responses following hyperpolarizing prepulses, which should reset the resting activation state of Na + and K + channels. We found no difference in the variability of spike timing with either 500 ms or 1 sec duration hyperpolarizing prepulses (Fig. 2-3C, D), suggesting that the variability likely reflects properties of voltage-gated channels activated by the depolarizing step. We next attempted to regularize mitral cell firing by resetting ionic currents activated by the step depolarization by introducing brief (25-75 ms) repolarizations back to rest. As shown in Fig. 2-4A, this protocol eliminated most of the variability in first spike latency (mean S.D. = 0.50 ± 0.07 ms; n = 5 cells) while responses to step depolarizations remained highly variable. Action potential threshold was reduced by the brief repolarizing steps (see Fig. 2-4A inset), suggesting that the brief repolarizations recovered more inactivated Na + current than K + current. In addition to controlling the onset of spiking, a brief repolarization often could terminate clusters. This phenomenon, however, was not as robust (successful in 204 of 300 attempts in 5 cells) as the first repolarization initiating firing, which worked with 100% reliability in 9 mitral cells. Fig. 37

48 2-4B shows that brief repolarizations can initiate and terminate firing at different times during the step depolarization. The ability of transient repolarizing steps to evoke precise firing in mitral cells suggests that mitral cells may respond selectively to slow time-modulated or oscillatory stimuli. In the intact animal, mitral cells receive phasic EPSPs in the theta frequency band (2 7 Hz) from upstream olfactory receptor neurons that are coupled to the respiratory rhythm (Macrides and Chorover, 1972; Charpak et al., 2001; Cang and Isaacson, 2003; Margrie and Schaefer, 2003). We therefore tested whether phasic EPSP-like stimuli evoke reproducible spiking in control mitral cells. In these experiments we injected a train of 6 simulated EPSPs at 1-5 Hz; these stimuli were designed to mimic the natural response of mitral cells during respiration (Charpak et al., 2001) and has been used in other in vitro studies (Halabisky and Strowbridge, 2003). Trials with phasic waveforms were alternated with simple step depolarizations. As shown in Fig. 2-5A, phasic stimuli generated precise spike timing (first spike SD = 1.6 ±.33 ms; 2.5 Hz; n = 7 cells), compared with the large variability of spike latencies that resulted from the alternating step trials (first spike SD = 230 ± 34 ms; n = 18 cells). The first simulated EPSP failed to generate action potentials but decreased the apparent action potential threshold for subsequent sepsps. The facilitating effect of the first EPSP on later responses was time dependent; increasing the delay between four identical sepsps (Fig. 2-5B) rapidly diminished the total number of spikes evoked by the sepsp train. This frequency filtering effect (Fig. 2-5C) was observed in 5/5 mitral cells and was not simply due to temporal summation since firing was facilitated at relatively low frequencies (1.3 Hz, compared with 1 Hz) in which the membrane potential recovered completely between sepsp 38

49 cycles. At higher frequencies, firing occurred at threshold membrane potentials more hyperpolarized than reached during the response to the first sepsp (at 2 and 2.5 Hz). Higher frequency sepsp trains also were effective in phase-locking mitral cell spikes (first spike SD = 1.89 ± Hz; n = 3 cells; data not shown). Our results show that mitral cells, which fire intermittently in response to step stimuli, can generate spikes with reproducible timing in response to phasic stimuli repeated at relatively low frequencies. These properties enable mitral cells to act as highpass filters, responding selectively to stimuli repeated at > 1 Hz and ignoring single simulated EPSP (sepsp) events (unless they are extremely large amplitude). As shown in Fig. 2-6, the first sepsp in a train controls spiking in subsequent sepsps. In this experiment we varied the amplitude of either the first (Fig. 2-6A) or second (Fig. 2-6B) sepsp in a 2 sepsp train stimulus; results from these experiments are summarized in Fig. 2-6C. Interestingly, the first sepsp controlled spiking in the subsequent sepsp in an all-or-none manner. In this neuron, no spikes were evoked by either sepsp if the first sepsp amplitude was less than 17 mv. Increasing the amplitude of the first sepsp enabled spiking on the second sepsp; increasing the amplitude of the first sepsp further did not change the frequency or number of spikes evoked by the second sepsp appreciably (Fig. 2-6B-C; n = 3 cells). By contrast, varying the amplitude of the second sepsp modulated both firing frequency and spike number. Suprathreshold responses also could be gated by short trains of small-amplitude simulated EPSPs (Fig. 2-6D; n = 4 cells), similar to those recorded in resting mitral cells in vivo (Spors and Grinvald, 2002; Cang and Isaacson, 2003; Margrie and Schaefer, 2003). 39

50 We next tested whether activation of transient K + currents facilitated phase locking in response to repeated phasic stimuli. We found that low concentrations of 4- aminopyridine (4-AP; 5 µm) enhanced the response to first sepsp in a train (Fig. 2-7A), which was usually subthreshold in mitral cells. In the presence of 5 µm 4-AP, the first sepsp now triggered multiple spikes with variable latencies (mean first spike SD in sepsp 1 = 11.9 ± 3.0 ms; n = 5 cells). Spikes evoked by second sepsp in 5 µm 4-AP remained precisely timed (first spike SD = 1.8 ± 0.4 ms; not different from control.) Increasing the 4-AP concentration to 100 µm slowed the average firing rate and decreased the temporal precision of spikes evoked by the second sepsp (SD = 6.6 ± 2.1 ms; different from control, p < 0.05; unpaired t-test; n = 5 cells). These results are summarized in Fig. 2-7B and suggest that 4-AP sensitive K + currents facilitate precise timing in mitral cells driven by phasic stimuli. Besides facilitating phase locking during phasic stimuli, 4-AP-sensitive K + currents also are required for intermittent firing following step depolarizations. As shown in Fig. 2-8A, 5 µm 4-AP gradually eliminated pauses between spike clusters, eventually producing tonic firing (n = 8 cells). Intermittent firing could be restored upon washout of 4-AP (Fig. 2-8A right). As shown in B, 5 µm 4-AP did not eliminate the initial delay before spiking; this delay was reduced by increasing the concentration of 4-AP to 100 µm (from 208 ± 28 ms in 5 µm 4-AP to 31.9 ± 13 ms in 100 µm 4-AP; n = 5 cells). Low concentrations of 4-AP also dramatically decreased the variability in first spike latencies in responses to step depolarizations (first spike SD = 17.2 ± 3.5 ms in 5 µm 4- AP versus 169 ± 32 ms in control conditions; n = 4 mitral cells; P <.05; Fig.2-8C). While 40

51 4-AP reduced the first spike latency, this effect did not explain the reduction in S.D. observed with 4-AP (first spike latency CV control = 25.5 ± 2.8 %; CV 5 μm 4-AP = 8.54 ± 0.74 %; P < 0.05). This concentration of 4-AP did not alter the input resistance (Fig. 2-8D), suggesting that 4-AP did not block K + channels active at rest. Higher concentrations of 4- AP (100 μm) caused an even greater reduction (3.64 ± 1.3 ms; n = 5 cells) in the standard deviation of first spike latency. The effects of different concentrations of 4-AP on spike timing are summarized in Fig. 2-8E. Discussion In this study we employed in vitro brain slices to investigate the nature of intermittent firing in mitral cells. We made three principal conclusions from this study. First, the intermittent firing pattern normally recorded in mitral cells activated with depolarizing current steps is highly sensitive to K + channel blockers specific for slowly inactivating I D - like currents and could be converted into tonic firing by less than 10 µm 4-AP. Second, the timing of individual spike clusters was highly variable with repeated steps; this variability also was reduced by low concentrations of 4-AP. Finally, when activated by phasic stimuli, mitral cells function as high-pass filters and generate precisely-timed spike clusters in response to inputs that mimic the natural 2-5 Hz sensory synaptic drive to these neurons in vivo during sniffing. We found that mitral cells fire clusters of action potentials at Hz interspersed with periods of subthreshold membrane potential oscillations at Hz. This spike clustering was dependent on intrinsic membrane properties, since it persisted in the presence of blockers of fast synaptic transmission. This finding is consistent with earlier reports on the intrinsic behavior of mitral cells (Chen and Shepherd, 1997; 41

52 Desmaisons et al., 1999; Friedman and Strowbridge, 2000). Spike clustering in response to step depolarizing stimuli or tonic depolarization has been observed in neurons from numerous brain areas, including stellate cells of the medial entorhinal cortex (Alonso and Llinas, 1989; Alonso and Klink, 1993; Klink and Alonso, 1993), non-cholinergic inhibitory interneurons of the basal forebrain (Alonso et al., 1996), striatal fast spiking interneurons (Bracci et al., 2003), and layer IV frontal cortex neurons (Llinas et al., 1991; Gutfreund et al., 1995). The usefulness of temporal coding as a strategy to represent information in the central nervous system requires that the timing of individual spikes in single neurons be highly reproducible across repeated identical stimuli. The reproducibility of spiking in response to repeated stimuli has generally been quantified by measuring spike precision (Mainen and Sejnowski, 1995; Nowak et al., 1997). Precision refers to the temporal jitter of spiking across multiple trials, and is measured as the standard deviation of spike latency. Previous in vitro studies have suggested that regular spiking cortical neurons have very low intrinsic noise and can respond with high precision to the onset of a step current stimulus (Mainen and Sejnowski, 1995; Nowak et al., 1997); noisy stimuli increase the precision of later spikes. By contrast, the intrinsic properties of mitral cells give rise to highly variable, unreliable spiking in response to simple step depolarizations but enable mitral cells to respond reproducibly to phasic stimuli in the theta frequency range. In many cell types, K + currents that are sensitive to very low concentrations of 4- AP have relatively slow deactivation and inactivation kinetics and are frequently termed I D -like (Storm, 1988; Wu and Barish, 1992; Fadool and Levitan, 1998; Coetzee et al., 42

53 1999; Mitterdorfer and Bean, 2002; Saviane et al., 2003). While the subunit composition of I D has not been established, this current may reflect heteromultimers containing Kv1- family subunits (Coetzee et al., 1999). Kv1.3 subunits have been shown to be expressed strongly in the olfactory bulb (Kues and Wunder, 1992). A recent study has shown that Kv1.3 protein is initially expressed in all layers of the rat olfactory bulb in early postnatal development (P1-P10), including mitral cell somata, but becomes progressively greater in the external plexiform layer, where the primary and secondary dendrites of mitral cells reside. While this staining pattern could be due in part to channel subunits localized to granule cell dendrites, dendrites terminating in glomeruli (presumably mitral/tufted cell primary dendrites) are especially heavily stained (Fadool et al., 2000). Other studies have shown that Kv1.3 mediated currents constitute the dominant outward conductance in cultured olfactory bulb neurons, and that these currents decay with a time constant of several hundreds of milliseconds (Fadool and Levitan, 1998)). Olfactory bulb cultures contain two morphologically distinct types of neurons: small bipolar neurons that are thought to be granule and periglomerular cells and larger pyramidal shaped neurons with prominent apical and secondary dendrites that are putative mitral/tufted cells (Trombley and Westbrook, 1990; Egan et al., 1992; Fadool and Levitan, 1998). Both neuronal types express large amounts of Kv1.3 currents; however there are subtle differences in the rate of inactivation of voltage-dependent currents in these subtypes (Fadool and Levitan, 1998). This may reflect different Kv1.3 containing heteromultimeric channels that are present in output versus local interneurons in the olfactory bulb. Fadool and colleagues (Fadool et al., 2004) recently investigated Kv1.3 knockout mice and found dramatic alterations in olfactory mediated behaviors and glomerular anatomy. In addition, 43

54 cultured neurons from olfactory bulbs of knockout animals show profound alterations in their voltage responses to current steps. These results underscore the potential importance of I D -like currents which involve Kv1.3 subunits in the function of the olfactory bulb. The long initial spike latency (Storm, 1988) and the sensitivity of intermittent discharges to specific blockers of slowly inactivating Kv1 family members that we have found in mitral cells suggest that I D -like currents play a critical role in patterning mitral cell responses. Preliminary mitral cell voltage clamp recordings indicate that mitral cells express at least two 4-AP sensitive transient potassium currents with decay kinetics that range from 40 to >500 ms (in response to steps from -80 mv to 0 mv; data not shown). A parallel study is underway in our laboratory with the goal of identifying the molecular basis of the transient K + currents in mitral cells that enable intermittent firing in response to step stimuli and phase locking in response to phasic stimuli. Mechanisms of Spike Clustering and Phase Locking Based on our results, we propose that spike clustering in mitral cells depends on the interplay between slowly-inactivating I D -like K + channels and a subthreshold TTXsensitive Na + current. Intermittent firing has been investigated previously using computational (Wang, 1993, 2002) and experimental (Klink and Alonso, 1993) studies to explain the genesis of fast subthreshold membrane potential oscillations and spike clustering in cells of the medial entorhinal cortex and basal forebrain (Alonso et al., 1996). These studies have proposed that cluster initiation depends on the level of inactivation of outward currents, while cluster termination depends on the buildup of potassium currents during a burst. Our studies support the view that spike cluster 44

55 termination is controlled by potassium current buildup since blocking I D -like currents increases cluster duration (and eventually abolishes intermittent firing). In addition, spike threshold increases slightly during the course of a cluster (see Fig. 2-2C), which suggests that outward currents increase during spike clusters. The first spike in a cluster is always smaller than later spikes; however, there is very little (< 3 mv) modulation of spike amplitude during a cluster. This suggests that processes which control Na + channel availability, such as cumulative inactivation during a train of spikes, may not play a prominent role in cluster termination. While elevated K + currents are likely to be responsible for cluster termination, the precise biophysical mechanisms involved have not been determined experimentally. Potassium currents may increase during clusters as result of very rapid recovery from inactivation between individual spikes within a cluster (Wang, 1993). Alternatively, macroscopic potassium currents may increase throughout each cluster, reflecting the slow deactivation kinetics of individual I D channels (Mitterdorfer and Bean, 2002). The origin of the variability in spike timing across repeated trials of step current is unlikely to reflect subtle changes in membrane properties of mitral cells from trial to trial. We found no correlation between the first spike latency and membrane potential or input resistance immediately preceding the depolarizing stimulus. One possible explanation for spike time variability is if spike initiation in mitral cells is controlled by a small number of ion channels, such that spike variability is a reflection of noise from channel gating events. Several studies have addressed this issue using computational (Schneidman et al., 1998; White et al., 1998; Jones, 2003) and experimental (Johansson and Arhem, 1994) 45

56 approaches. However, the functional significance of stochastic channel gating in controlling spike timing in mitral cells has not been established and may not apply to neurons as large as mitral cells. Alternatively, variability in discharge times may reflect the complex oscillatory dynamics of opposing inward and outward macroscopic currents active near threshold. Our preliminary voltage clamp studies indicate that mitral cells express high levels of transient 4-AP sensitive K + currents, suggesting that oscillating inward and outward currents may be a more important mechanism for controlling spike timing than stochastic channel gating events. Previous studies on the role of transient 4- AP sensitive K + currents in controlling spike timing in neurons have focused on the characteristic delay in the timing of the first spike (Storm, 1988; Saviane et al., 2003). Blockade of 4-AP sensitive K + currents reduces this delay (Storm, 1988; McCormick, 1998; Saviane et al., 2003); however, it is not known whether expression of I D -like currents in these neurons leads to variable spike timing. Precise spike timing evoked by phasic stimuli likely arises because of differences in the kinetics of recovery from inactivation of transient outward currents and voltagedependent Na + currents. The initial depolarization from the first simulated EPSP causes a rapid activation of outward currents that inhibit spiking. During the falling phase of the first EPSP, fast transient Na + currents should recover rapidly from inactivation (Kuo and Bean, 1994). By contrast, slowly inactivating, I D -like currents, will likely de-inactivate at much slower rates (Fadool and Levitan, 1998), enabling a subsequent depolarization to trigger a cluster of spikes. Spike clusters are terminated either by repolarization during the falling phase of the EPSP or buildup of outward currents. This scheme is supported 46

57 by our experiments that show that the amplitude of the first EPSP gates the generation of spikes on subsequent EPSPs. Small amplitude simulated EPSPs may not inactivate sufficient K + currents to allow firing on subsequent simulated EPSPs (as shown in lower traces in Fig. 2-6A). Larger initial simulated EPSPs presumably inactivate more I D -like K + current, thereby facilitating firing following a repolarization/depolarization cycle. Preferential recovery of Na + versus K + currents during the repolarization phase is likely to account for the decreased firing threshold following brief repolarizing steps (Fig. 2-4) and during responses to the trains of simulated EPSPs (Fig. 2-5). Functional Implications for Odor Coding Several theoretical studies have suggested that action potential timing may be important for representing sensory stimuli (Hopfield, 1995; Rieke et al., 1997). Temporal coding appears to be especially important in olfactory processing (Laurent et al., 1996; Wehr and Laurent, 1996) where single olfactory receptor neurons have broad specificity for many odorants (Duchamp-Viret et al., 1999; Araneda et al., 2000) to increase the number of odorants that can be uniquely identified. Recent studies in insects have shown that downstream neurons that receive information from projection neurons act as coincidence detectors (Perez-Orive et al., 2002). Such a coding scheme requires that incoming spike trains be highly reproducible across repeated trials. Our study suggests that intrinsic ionic mechanisms in mitral cells promote precise spiking in response to phasic stimuli in the theta-frequency range. Prominent 2-7 Hz activity coupled to the respiratory rhythm has been observed in mitral cells in vivo using extracellular unit (Macrides and Chorover, 1972; Belluscio et al., 2002) and whole-cell intracellular (Margrie and Schaefer, 2003) recordings, voltage dye imaging (Spors and 47

58 Grinvald, 2002) and calcium imaging in mitral cell apical dendritic tufts (Charpak et al., 2001). The genesis of this respiratory-coupled activity is likely to reflect changes in odorant concentration (and thus, activation of olfactory receptor neurons) in the olfactory epithelium during inspiration (Sobel and Tank, 1993). Our results suggest that mitral cells are tuned to receive synaptic input in the theta-band frequency range in which rodents normally sniff. The intrinsic properties of mitral cells allow them to filter olfactory information by controlling the generation of spikes that are evoked by inspiration-induced theta activity. By this mechanism, the activation of a broad subset of olfactory receptor neurons would result in precisely timed trains of spikes in a small subset of mitral cells. Weak or transient stimuli may not evoke spiking at all, whereas sustained stimuli that are not modulated in time might produce spikes that are highly variable from trial to trial, presumably impairing downstream coincidence detection mechanisms. These intrinsic filtering mechanisms might act in concert with synaptic mechanisms that synchronize theta oscillations in adjacent mitral cells (Schoppa and Westbrook, 2001, 2002; Urban and Sakmann, 2002) to ensure that mitral cells which project to the same glomerulus act as distinct functional units. 48

59 Figure 2-1. Intermittent firing and subthreshold oscillations in olfactory bulb mitral cells (A) Schematic cartoon of olfactory bulb circuitry showing relative position of mitral cells and patch pipette. (B) Responses to graded step depolarizations ( pa, 4 sec duration). Mitral cells fire intermittent clusters of action potentials in response to each step with intervening periods of subthreshold membrane potential oscillations. Responses recorded in NBQX (5 µm) and D-APV (25 µm); RMP = -69 mv. (C) Plots of the relationship between current step amplitude and within cluster firing frequency (left), number of spikes per cluster (center) and mean pause duration (right). Data points represent mean ± SD of at least 3 trials. (D) Plot of the effect of TTX (1 µm) on the steady-state depolarization reached during a 200 pa current step (3 sec duration). Example traces shown above in control conditions (left) and after exposure to TTX for 2.3 and 3.5 minutes (right). Arrow marks point during current step where steady state voltage was measured. Action potentials clipped in example traces. Note that TTX reduces both the depolarizing extent of the step response and the amplitude of the subthreshold oscillations normally present near firing threshold. (E) Enlargements of steady-state responses in control, after 2.3 and 3.5 minutes exposure to TTX, and during washout of TTX. 49

60 50

61 Figure 2-2. Variability of clustered spike discharge timing in mitral cells (A) Variable responses in a single mitral cell to four 250 pa step depolarizations, repeated every 20 sec; RMP = -65 mv. (B) Histogram of interspike intervals in 5 mitral cells (7789 interspike intervals). The distribution of pauses between spike clusters (interspike intervals > 100 ms, determined by visual inspection; n = 511) was well fit by a Gaussian function with a mean of 470 ms (R 2 = 0.95). Note that the Y axis was clipped to reveal the distribution of long interspike intervals. (C) Example response to a step depolarization in a different mitral cell (RMP = 68 mv). Expansion on right shows the afterhyperpolarization that normally follows each cluster of action potentials. The cluster AHPs in this cell could be fit using a single exponential function that decayed with a time constant of 179 ± 49 ms. 51

62 52

63 Figure 2-3. Variability in mitral cell firing patterns does not reflect initial conditions (A) Responses to repeated weak (150 pa) and strong (300 pa) depolarizing steps in a mitral cell. Vertical lines on raster plot represent times of individual action potentials during repeated trials. Responses were variable to both weak and strong depolarizing steps. RMP = -69 mv. (B) Superposition of two responses to the same depolarizing current step in one mitral cell in which the latency to the first action potential varied by 420 ms. (C) Hyperpolarizing prepulses (either 500 ms or 1 sec duration) did not reduce the spike timing variability to subsequent depolarizing steps. RMP = -63 mv. (D) Summary of the effect of hyperpolarizing prepulses on first spike latencies (mean and S.D.). Data from 5 mitral cells. 53

64 54

65 Figure 2-4. Transient repolarizations promote precise phase locking in mitral cells A, Responses to repeated trials of step depolarizations alone (left) or with four transient repolarizations (50 ms duration) (right). Spikes were triggered reliably following the offset of the repolarizing pulse (first spike S.D. = 0.50 ± 0.07 ms) compared with the large variability in first spike latency in response to simple step stimuli (S.D. = 230 ± 34 ms). Both set of records from the same mitral cell; RMP = -68 mv. Periods of tonic firing could be halted by a second transient repolarizing pulse. Insert on right shows expansion of recording during the first repolarization; dashed line represents the steady-state membrane potential achieved before the repolarization. Note that the repolarizing step altered the action potential threshold (arrow) now occurred 6.1 mv below the previous steady-state level. Calibration bar in insert: 10 mv, 25 ms. B, Tonic firing initiated and halted at arbitrary times. Four responses to the same depolarizing step stimuli (280 pa) with repolarizing pulses occurring at different latencies (offset by 50 ms in each trace). RMP = -68 mv. 55

66 56

67 Figure 2-5. Precise spike timing in mitral cells activated by phasic stimuli (A) Alternating responses to either a step depolarization (left) or a train of 6 simulated EPSPs (sepsps) at 2.5 Hz (center; see text for details). Traces on right show expansion of responses to the second sepsp. Variability in first spike latency was reduced with phasic stimuli to 1.18 ms (S.D.), compared with 258 ms in responses to step depolarizations in this cell. Note the reduction in action potential threshold by the first sepsp. RMP = - 66 mv. (B) Responses to trains of 4 uniform sepsps at different frequencies. Repeated sepsp stimuli at > 1 Hz evoked firing in mitral cells, which increased with increasing sepsp frequency. Note the absence of spikes triggered by the first sepsp in all examples responses. RMP = -65 mv; action potentials clipped. (C) Summary plot of the relationship between sepsp frequency and the total number of spikes evoked by the 4-sEPSP train (data from 5 mitral cells). 57

68 58

69 Figure 2-6. Different effects of first and second sepsps (A) Effect of varying the first sepsp in a 2-sEPSP train (sepsp 2 = 450 pa). The first sepsp regulated spiking evoked by sepsp 2 in an all-or-none manner. (B) Varying the second sepsp in a 2-sEPSP train (sepsp1 = 450 pa) altered both number and frequency of action potentials evoked by sepsp 2. Responses in A and B from the same mitral cell; RMP = -62 mv. (C) Plots of the relationship between the amplitude of sepsp 1 (left) and sepsp 2 (right) and spike frequency (top) and the number of action potentials evoked by sepsp 2 (bottom). Each point represents data from one trial. (D) Gating by smallamplitude sepsp trains. No spikes were evoked by the first sepsp (arrow) in a 4 Hz sepsp train (left). Preceding this train with a second 4 Hz train composed of smallamplitude sepsps (5-6 mv) enabled spiking on first sepsp (middle). No spikes were evoked by the first sepsp (arrow) if the small-amplitude train occurred 500 ms before the first sepsp (right). 59

70 60

71 Figure 2-7. Disruption of precise spiking by 4-AP (A) Responses to repeated trials using a 6 sepsp train at 2.5 Hz showed phase locking. Low concentrations of 4-AP (5 µm) allowed the first sepsp to trigger spikes which were poorly phase locked; action potentials evoked by sepsp 2 remain time-locked on repeated trials. Higher concentrations of 4-AP (100 µm), disrupted phase locking to both sepsp 1 and sepsp 2. Control and 100 µm 4-AP example responses are from the same mitral cells (RMP = -70 mv) while the example responses in 5 µm 4-AP are from a different cell (RMP = -65 mv). (B) Summary of variability in first spike latencies to sepsp 1 (open bars) and sepsp 2 (closed bars) in control conditions and in 5 µm and 100 µm 4-AP. Data from 5 7 mitral cells in each condition. * p < 0.05; ** p <

72 62

73 Figure AP sensitive K + currents regulate intermittent discharges in mitral cells (A) Responses of a mitral cell to a step depolarization in control conditions and in 5 µm 4-AP (RMP = -65 mv; 260 pa current step; example traces 4 and 5 mins after exposure to 4-AP). This concentration of 4-AP reliably converted the intermittent discharge response pattern into tonic firing. (B) Higher concentrations of 4-AP (100 µm) decreased the initial delay before tonic firing. (C) Raster display of suprathreshold responses to step depolarizations repeated every 20 sec before and after bath application of 5 µm 4-AP. Step amplitude = 200 pa, RMP = -67 mv. (D) Plot of reduction in first spike latency by 5 µm 4-AP. This concentration of 4-AP did not affect resting input resistance as measured by responses to small amplitude hyperpolarizing test pulses applied immediately before the step depolarization (bottom graph). (E) Summary plot of the variability in first spike latency in control and in different concentrations of 4-AP. Data from 14 mitral cells; each point represents the mean from at least 4 cells. 63

74 64

75 Chapter 3 : Opposing Inward and Outward Intrinsic Currents Control Rebound Discharges in Mitral Cells Summary The unique reciprocal dendrodendritic synapses between mitral cells and granule cells represent the most common synapse in the olfactory bulb, and are critical for sculpting mitral cell output to higher brain centers. Numerous studies using whole-cell patch clamp techniques have revealed fundamental properties of these reciprocal synapses; however, the functional significance of granule cell mediated inhibition on mitral cell firing patterns is still largely unknown. We used whole cell patch clamp recordings from mitral cells in olfactory bulb slices to investigate the mechanisms by which granule activity might regulate mitral cell spike discharges. Mitral cells have unique intrinsic membrane properties which support rebound spike generation in response to small (3-5 mv) hyperpolarizing current injections or granule cell IPSPs. Rebound spiking occurred in depolarized (-45 to -40 mv) cells and was dependent on recovery of subthreshold persistent sodium currents, and could be blocked by tetrodotoxin (TTX, 1 µm) or the more selective persistent sodium channel blocker riluzole (10 µm). Surprisingly, larger amplitude hyperpolarizing stimuli impeded spike generation by recruiting a Ba 2+ - sensitive outward current. The interplay of voltage gated sodium channels and Ba 2+ - sensitive outward current produces a narrow range of IPSP amplitudes that are effective at generating rebound spikes. We also found that persistent, noninactivated sodium channels boost subthreshold excitatory stimuli to voltage ranges where granule cell mediated IPSPs can produce rebound spikes. Finally, we used dual whole cell recordings 65

76 from pairs of mitral cells to show that granule cell mediated IPSPs can transiently synchronize mitral cell activity. These results show how the unique intrinsic membrane properties of mitral cells may interact with synaptic stimulation to produce temporally correlated patterns of spiking, and shed light on how odor information is transformed in the olfactory bulb. 66

77 Introduction Sensory activation of neurons in the nasal epithelium by odorant molecules causes synaptic activation of groups of mitral and tufted cells, the principal cells of the olfactory bulb, which send projections to higher order centers such as the piriform cortex (Shepherd and Greer, 1998). Computations performed in the olfactory bulb involve synaptic interactions with local inhibitory interneurons in both the glomerular layer (with juxta- and periglomerular cells) (Pinching and Powell, 1971a, b, 1972; Smith and Jahr, 2002; Aungst et al., 2003; Schoppa and Urban, 2003; Murphy et al., 2005) and with GABAergic interneurons in the granule cell layer (Price and Powell, 1970b, d; Isaacson and Strowbridge, 1998; Schoppa and Urban, 2003; Pressler and Strowbridge, 2006). In vivo intracellular recordings (Hamilton and Kauer, 1985, 1989; Wellis and Scott, 1990) directly demonstrated large amplitude IPSPs mediated by these local circuit pathways in mitral cells following sensory stimulation. These studies illustrated a classic function of GABAergic inhibition in the CNS sculpting the firing pattern of output neurons (Eccles et al., 1967). However GABAergic inhibition also has other functions in different brain regions, including triggering rebound spiking (Jahnsen and Llinas, 1984a; McCormick, 1998) and synchronized oscillations (Steriade et al., 1993; Buzsaki, 2002; Buzsaki and Draguhn, 2004). While there is some evidence that GABAergic inhibition may also subserve these functions in the olfactory bulb (Desmaisons et al., 1999; Friedman and Strowbridge, 2003; Lagier et al., 2004), little is known about the cellular mechanisms underlying these responses. In the present study we examined the cellular mechanism responsible for rebound activity in olfactory mitral cells. 67

78 Most GABAergic inhibition onto principal neurons in the olfactory bulb arises from dendrodendritic microcircuits between mitral cells and local interneurons, predominately granule cells (Shepherd and Greer, 1998). The unique reciprocal dendrodendritic synapses between mitral cells and granule cells represent the most common synapse in the olfactory bulb (Shepherd and Greer, 1998). Depolarization of mitral cell secondary dendrites causes glutamate release onto granule cell spines, which then leads to subsequent GABA release onto both the originally depolarized mitral cell (selfinhibition) and neighboring mitral cells (lateral inhibition). One hypothesis for the function of lateral inhibition in the olfactory bulb is to sharpen the spatial pattern of mitral cell activity (Yokoi et al., 1995; Isaacson and Strowbridge, 1998). Alternatively, local inhibitory processing may promote rebound discharges which can transiently synchronize mitral cell assemblies. In other brain areas (Jahnsen and Llinas, 1984a, b; McCormick and Bal, 1997; McCormick, 1998) such as the thalamus, neocortex and inferior olive, IPSPs often evoked post-inhibitory rebound action potential generation in principal neurons that exert a profound influence on spike timing. Several recent studies suggest that granule cell mediated IPSPs may act in a similar fashion to promote correlated spiking of groups of mitral cells during odor processing. Activation of olfactory sensory nerve afferents causes long-lasting gamma-frequency local field potential (LFP) oscillations in the olfactory bulb both in vivo and in vitro that reflect synchronous mitral cell activity (Friedman and Strowbridge, 2003; Lagier et al., 2004). These LFP oscillations are dependent on granule cell activity, and can be abolished by blocking GABA A receptors (Friedman and Strowbridge, 2003). Other studies in insects showed that local inhibitory processing was necessary for LFP oscillations and transient 68

79 synchronization of projection neuron assemblies (Stopfer et al., 1997). Blocking local IPSPs abolished both synchronization and the ability to discriminate between closely related odorants. Finally, a recent study has shown that both transient hyperpolarizing stimuli and spontaneous IPSPs can elicit rebound spikes in depolarized mitral cells, providing a clue as to how granule cell mediated IPSPs may promote action potential synchronization in mitral cells (Desmaisons et al., 1999). Despite these results, the cellular mechanisms of rebound spike generation in mitral cells are still unknown. We used whole cell patch clamp recordings from mitral cells in olfactory bulb slices to define the conditions under which mitral cells generate rebound discharges and investigate the mechanisms by which local inhibitory circuits might promote correlated mitral cell activity. We found that mitral cells depolarized to near spike threshold can produce rebound discharges that are dependent on voltage gated Na-channel recovery in response to small (3-5 mv) hyperpolarizing current injections or unitary granule cellmediated IPSPs. Surprisingly, larger amplitude hyperpolarizing stimuli impeded spike generation by recruiting a transient K current that is blocked by high concentrations of 4- AP and Ba. We also found that persistent subthreshold sodium channels boost subthreshold excitatory stimuli to voltage ranges where granule cell mediated IPSPs can produce rebound spikes. The interplay of opposing inward and outward intrinsic currents produces a narrow window of IPSP amplitudes that are effective at generating rebound spikes and allows IPSPs to bidirectionally control spike output depending on which intrinsic currents are preferentially recruited. 69

80 Materials and Methods Slice preparation and recording Horizontal slices (300 µm) through the olfactory bulb were prepared from anesthetized (ketamine, 140 mg/kg ip) P14-21 Sprague-Dawley rats using a modified Leica (Nussloch, Germany) VT1000S vibratome, as described previously (Isaacson and Strowbridge 1998; Halabisky et al 2000). Olfactory bulb slices were incubated at 30 O C for 25 min then maintained submerged at room temperature in a holding chamber until needed. Wholecell patch-clamp recordings were made in mitral cells visualized under infrareddifferential interference contrast optics (Zeiss Axioskop 1 FS) using an Axopatch 1D amplifier (Axon Instruments). During recordings, olfactory bulb slices were superfused with artificial cerebrospinal fluid (ACSF) that contained (in mm): NaCl 124, KCl 3, NaH 2 PO , NaHCO 3 26, dextrose 10, CaCl 2 2.5, and MgSO 4 1.2, equilibrated with 95% O 2 /5% CO 2 and warmed to 30 C (flow rate, 1-2 ml/min). During experiments examining the effect of evoked IPSPs on mitral cell spiking, we used ACSF containing 5 mm KCl to increase the probability of finding functional inhibitory synapses. A modified ACSF solution was employed when making slices and in the holding chamber that contained reduced CaCl 2 (1 mm) and elevated MgSO 4 (3 mm). Patch electrodes used for current clamp recordings (3-5 MΩ resistance) contained (in mm): K-methylsulfate 140, NaCl 4, HEPES 10, EGTA 0.2, MgATP 4, Na 3 GTP 0.3, and phosphocreatine 10. Recordings using somatic current injections to examine mitral cell intrinsic membrane properties were obtained in the presence of NBQX (5 µm) and D-APV (25 µm) in the bath solution to block ionotropic glutamate receptors and recurrent synaptic activity. 70

81 Extracellular stimulation Granule cell mediated GABAergic inhibitory postsynaptic potentials (IPSPs) were evoked by monopolar extracellular stimulation using a fine tungsten microelectrode (9-12 MΩ impedance, Frederick Haer & Co.) placed either in the granule cell layer or proximal external plexiform layer approximately µm lateral to the recorded mitral cell. 200 µsec duration constant current stimuli were given using a stimulus isolator (World Precision Instruments). In experiments using paired mitral cell recordings, we evoked IPSPs using a bipolar stimulating electrode which consisted of a pair of tungsten microelectrodes (tip separation 305 µm, Frederick Haer & Co.) placed in the granule cell layer directly beneath or just lateral to the recorded mitral cell. Data acquisition and analysis Voltage records were low-pass filtered at 2 khz and then digitized at 5 khz using a 16-bit A/D converter (ITC-18, Instrutech). In some experiments, a current injection waveform consisting of a train of 4 temporally-overlapping EPSP-like waveforms was injected into mitral cells (Halabisky and Strowbridge, 2003; Balu et al., 2004). Each simulated EPSP in the train was generated using a single alpha function with a decay time constant of 80 ms. This stimulus train was modeled after respiration-evoked calcium and voltage oscillations recorded from mitral cell glomerular tufts in vivo (Charpak et al., 2001). 71

82 Electrophysiological data were recorded and analyzed using custom software written in Visual Basic 6 (Microsoft) and Origin 7 (OriginLab). We quantified the degree of rectification in mitral cell voltage responses by first calculating the voltage difference from immediately before the onset of the hyperpolarizing pulse to 50 ms into the hyperpolarizing pulse, and then subtracting this quantity from the voltage difference from immediately before the offset of the hyperpolarizing pulse to 50 ms after then end of the pulse. Using this formula, a standard RC circuit gives a rectification of 0 mv whereas any outward current activated during hyperpolarization will result in a delay in repolarization after the end of the hyperpolarizing pulse and a rectification of > 0 mv. Membrane potentials indicated are not corrected for the liquid junction potential. All chemicals were obtained from Sigma (St. Louis, MO) except for tetrodotoxin (TTX; Calbiochem). Data are shown as the mean ± SEM. Statistical significance was determined using paired Student s t-tests except where noted. Results Focal extracellular stimulation in the granule cell layer evoked hyperpolarizing IPSPs in mitral cells held near firing threshold. Inhibitory postsynaptic responses persisted in D- APV (25 µm) and NBQX (5 µm) but were blocked by picrotoxin (PTX; 50 µm) and reversed at approximately -70 mv (Fig 3-1B), consistent with activation of GABA A receptors. As previously reported, we found that GABAergic synaptic inputs can mediate two roles in mitral cells conventional inhibition, expressed by a reduction in spiking 72

83 (Isaacson and Strowbridge, 1998; Friedman and Strowbridge, 2000), and activation of mitral cells through rebound discharges (Desmaisons et al., 1999). While the inhibitory function of IPSPs has been studied previously (Hamilton and Kauer, 1985, 1989), little is known about the cellular mechanisms and functional properties of rebound excitation in mitral cells. At depolarized membrane potentials, mitral cell IPSPs often triggered rebound depolarizations (Fig. 3-1C) that could trigger multi-spike discharges (Fig. 3-1D). Rebound responses trigged by IPSPs were voltage dependent and were abolished by a moderate (~ 5 mv) hyperpolarizing shift in the membrane potential (n = 4 cells; Fig. 3-1C). The ability of hyperpolarizing IPSPs to evoke rebound firing enabled these inputs to trigger correlated discharges in populations of mitral cells, as illustrated with the dual recording in Fig. 3-1E. Simple hyperpolarizing steps mimicked the ability of GABAergic IPSPs to trigger short-latency rebound discharges (87 of 93 mitral cells tested), suggesting that these responses reflect properties of the voltage-gated ionic currents present in mitral cells. Rebound spiking was tightly synchronized to the offset of the hyperpolarizing steps (S.D. of first spike latency = 1.60 ± 1.08 ms, n = 9 cells), suggesting that divergent inhibitory synaptic input may function to synchronize subpopulations of mitral cells. While both synaptic IPSPs and direct hyperpolarizations effectively triggered rebound discharges, evoked IPSPs often triggered additional, longlatency spiking activity (Fig. 3-1F 1 ) that were not observed after hyperpolarizing current pulses (Fig. 3-1F 2 ). Subthreshold rebound depolarizations depicted in Fig. 3-1C resembled lowthreshold Ca spikes typically found in thalamic relay neurons (Jahnsen and Llinas, 1984a; 73

84 McCormick and Bal, 1997) and many other CNS cell types. However rebound discharges persisted in mitral cells following blockade of non-selective voltage-gated Ca channels with Cd (200 µm; n = 4 cells; data not shown) and in low Ca / high Mg ACSF (0.25 and 6 mm, respectively; n = 4 cells), suggesting that rebound activity was not due to de-inactivation of low-threshold Ca channels. Bath application of the low-threshold Ca channel blocker Ni (100 µm; n = 3 cells) also failed to attenuate rebound activity. Also unlike thalamic neurons (Jahnsen and Llinas, 1984a; McCormick and Bal, 1997), the duration of rebound discharges was not modulated by the amplitude of the hyperpolarizing pulse except for very large amplitude pulses (Fig. 3-1G). Instead, we found that a large range of step amplitudes (50 to 400 pa) triggered stereotyped rebound discharges composed of the same number of action potentials. Rebound firing slowed following large amplitude steps (see bottom trace in Fig. 3-1G; n = 6 cells), a finding that also is inconsistent with rebound spikes mediated by low-threshold Ca spikes. Mitral cells showed a distinctive biphasic response to a graded series of relatively weak 100 ms hyperpolarizing pulses, as shown in Fig. 3-2A. When held near firing threshold, small amplitude (6 25 pa) steps, which caused hyperpolarizations between 0.5 and 2 mv, rarely produced rebound spikes but often triggered subthreshold rebound depolarizations. Moderate amplitude steps (30 65 pa; generating 3-7 mv hyperpolarizations) evoked rebound spikes with high probability. Surprisingly, increasing the hyperpolarization step amplitude further (generating hyperpolarizations > 7 mv; n = 7 cells) reduced the probability of triggering rebound spikes. Rebound activity (spiking and subthreshold depolarizations) also was abolished when the membrane potential was hyperpolarized by 6 mv (open circles in Fig. 3-2A 3 ). We observed similar 74

85 results in which rebound discharges were triggered by weak (< 5 mv) but not largeamplitude (> 15 mv) hyperpolarizing steps (Fig. 3-2B; n = 4 cells). Membrane repolarization following large hyperpolarizing steps was slowed (see arrow in Fig. 3-2B), suggesting that strong hyperpolarization recovered an outward current that was activated at the step offset. The previous results suggest that rebound spiking may be controlled by two opposing processes that are recruited during a hyperpolarizing step: one that promotes spike generation and another that inhibits spiking following a hyperpolarizing stimulus. We first focused on identifying the ionic mechanisms which promote rebound spiking in response to smaller hyperpolarizing steps. We then investigated possible factors that contribute to spike inhibition following large amplitude hyperpolarizations. There are at least three common mechanisms that generate rebound discharges in CNS neurons: (1) de-inactivation of low-threshold Ca current, (2) de-inactivation of subthreshold Na current and (3) activation of I H (refs). As discussed above, blockade of Ca currents in mitral cells did not abolish rebound depolarizations such as those shown in Fig. 3-1C and 2A. Similarly, reducing Ca influx by switching to a low Ca / High Mg ACSF increased rather decreased the number of rebound spikes triggered by hyperpolarizing pulses (from 4.40 ± 1.6 to 16.1 ± 2.7 spikes; P < 0.05; n = 4 cells), suggesting that voltage-gated Ca channels are not required to trigger rebound activity in mitral cells. Mitral cells have a small membrane potential sag during prolonged hyperpolarizing steps (Fig. 3-2C 1 ), indicative of a weak I H current. The I H blocker Cs (4 mm) eliminated membrane potential sag in 7 mitral cells tested (see insert in Fig. 3-2C). However Cs did not reduce rebound discharges following hyperpolarizing steps near threshold (Fig. 3-2C 2 ; 4 mm; n 75

86 = 4 cells). These results suggest that neither voltage-gated Ca currents nor I H mediate rebound activity in mitral cells. We next tested whether rebound spikes were triggered by de-inactivation of subthreshold Na-channels. Mitral cells show a characteristic prolonged subthreshold period following depolarizing steps; often the initial response is dominated by small amplitude membrane potential oscillations before the first cluster of action potentials is generated (Chen and Shepherd, 1997; Desmaisons et al., 1999; Balu et al., 2004). In addition to blocking fast sodium channel dependent action potentials, TTX attenuated the sustained subthreshold depolarization and blocked subthreshold membrane potential oscillations (Fig. 3-3A; n = 6 cells). Subthreshold sodium currents also boosted depolarizing responses to phasic stimuli that mimic trains of inspiratory EPSPs (Fig 3-3B; (Halabisky and Strowbridge, 2003; Balu et al., 2004)). The depolarizing membrane potential boost due to TTX-sensitive Na channels was smaller during the first simulated EPSP than on subsequent sepsps (mean EPSP 1 boost = 4.27 ± 0.45 mv; mean EPSP 4 boost = 7.63 ± 0.67 mv; p < 0.01; n = 6 cells) suggesting that progressive activation of subthreshold Na currents contributes to EPSP summation during sniffing-like excitatory input. TTX also reversibly blocked rebound spikes and subthreshold depolarizations triggered by graded hyperpolarizing pulses (Fig. 3-3C; n = 7 cells) held at the same membrane potential. Riluzole, a moderately selective blocker of subthreshold Na currents (10 µm; (Del Negro et al., 2005; Wu et al., 2005; Enomoto et al., 2006)) also reduced the membrane potential boost to prolonged depolarizing steps (Fig. 3-3D 1 ) and blocked rebound discharges triggered by hyperpolarizing pulses (Fig. 3-3D 2 ; n = 4 cells) 76

87 without blocking action potentials (AP amplitude before riluzole = 78.8 ± 3.0 mv versus 77.3 ± 1.8 mv after riluzole; not significantly different; Fig. 3-3D 1 insert). While the results presented thus far suggest that rebound spiking is dependent on recovery of subthreshold Na currents, this mechanism does not explain why rebound spiking was inhibited following prolonged or large amplitude hyperpolarization steps. As shown in Fig. 3-4A, rebound bursting can be eliminated by slightly increasing the duration of the hyperpolarizing step from 50 to 100 ms (n = 4 cells). This result paralleled the gating of rebound bursting by hyperpolarization step amplitude illustrated in Fig. 3-2B 1 and suggests that multiple active conductances are recovered by hyperpolarizing steps from near threshold, including an outward current that opposes rebound spiking. Further increases in step duration resulted in a graded slowing of the repolarization following step offset (Fig. 3-4B-C), presumably reflecting increasing activation, followed by inactivation, of K currents that oppose rebound spiking. The maximum repolarization delay was approximately 150 ms (generated by 200 ms duration hyperpolarization steps from -42 to -71 mv), approximately 3 fold longer than the membrane time constant of mitral cells (tau = 50.7 ± 12.8 ms; n = 6). Delayed repolarization was not observed following similar hyperpolarizing steps from the resting membrane potential (-64.3 ± 4.9 mv; n = 6; data not shown) suggesting that activation of voltage-dependent K channels are responsible for the repolarization delay. We next asked if mitral cells express transient K currents that inactivated near rest with time constants that matched the repolarization delay we recorded under current clamp conditions. We previously reported (Balu et al., 2004) that mitral cells express an I D -like transient K current that was sensitive to low concentrations of 4-aminopyridine (4-77

88 AP; 1-10 µm). However these low concentrations of 4-AP did not affect membrane repolarization, suggesting that I D was not responsible for this phenomenon (data not shown). As illustrated in Fig. 3-5A, delayed repolarization was still evident following blockade of Na channels with TTX (1 µm) and voltage-gated Ca channels with Cd (200 µm) and Ni (100 µm), suggesting that this delay was not due to Ca-activated K currents. In the presence of TTX, Cd and Ni, high concentrations of 4-AP (6 mm) abolished repolarization delay (Fig. 3-4A; n = 4) suggesting that recovery of inactivated I A current (Segal and Barker, 1984; Markram and Segal, 1990) may slow repolarization and gate rebound discharges in mitral cells. Supporting that hypothesis we found that 2 mm Ba also blocked repolarization delay (n = 5). At this concentration Ba has several actions, including blockade of erg-family (Saganich et al., 1999; Saganich et al., 2001) and I A -like (Hille, 2001) K currents. However we found that more selective erg-family channel blockers (5 µm E4031 and 10 µm dofetilide) did not affect repolarization delay in mitral cells. We found that the I M blocker XE-991 (10 µm), the I H blocker Cs (6 mm) and the delayed rectifier blocker TEA (25 mm) also did not affect repolarization delay in mitral cells. These results are summarized in Fig. 3-5B and suggest that repolarization delay is due to recovery of inactivated I A current. Mitral cells express both transient and non-inactivating K currents. As shown in the family of voltage clamp responses in Fig. 3-5C, even relatively weak depolarizing steps (from -80 to -40 mv) activated transient K currents that were eliminated by the I A blocker 4-AP (6 mm; n = 5 cells). The kinetics of the 4-AP sensitive current in mitral cells (tau = 139 ± 9.6 ms; mean peak amplitude = 1060 ± 108 pa; steps from -80 to -40 mv;) was similar to the repolarization delay observed following hyperpolarizing steps from near 78

89 firing threshold (152.8 ± 13.3 ms maximum repolarization delay; Fig. 3-4C). As shown in Fig. 3-5D, transient K currents in mitral cells inactivated completely within 1 sec at - 40 mv and were greatly diminished at -50 mv, suggesting that hyperpolarizing responses evoked near firing threshold have the potential to recover inactivated I A current. Together these data suggest that mitral cells express an I A -like transient K current that is blocked by mm concentrations of 4-AP and mediates the delayed repolarization following hyperpolarizing steps from near firing threshold. Finally, we investigated the effect of hyperpolarizing responses on phasicallyactivated mitral cells. The normal sensory drive to mitral cells occurs through inspirationally-linked glutamatergic synaptic inputs in the glomerular layer (Shepherd and Greer, 1998). With relatively weak phasic drive, mitral cells responded intermittently to the phasic stimulation with an all-or-none pattern (Fig. 3-6A). Surprisingly, the timing of the action potential clusters evoked by each alpha function was maintained despite the intermittent nature of responses on preceding cycles (first spike SD = 1.2, 2.7 and 3.1 ms for sepsp 2-4 ). This result, observed consistently in 8/8 mitral cells systematically, suggests that the intrinsic conductances that mediate precisely-timed all-or-none discharges to phasic stimuli are reset in the periods between stimuli. This resetting process presumably involves recovery of inactivated Na and K currents that interact to generate the all-or-none discharge at the peak of the phasic depolarization. We found that brief hyperpolarizing responses injected during the interstimulus period modulated discharges on the subsequent phasic depolarization in an allor-none manner. As shown in Fig. 3-6B, a simulated IPSP (alpha function with a 10 ms time constant) applied 100 ms before a phasic depolarization consistently abolished the 79

90 discharge normally evoked by that depolarization. The amplitude of the phasic depolarizing stimuli was increased to reliably trigger action potential discharges in these experiments. We observed similar results in 5 experiments using brief hyperpolarizing current injections with the timing shown in Fig. 3-6B. As shown in Fig. 3-6C, simulated hyperpolarizing IPSPs that occurred within 125 ms of the onset of the phasic depolarization could abolish firing in an all-or-none manner. Simulated IPSP blocked phasic discharges most effectively when they occurred between 40 and 75 ms before phasic depolarizations (Fig. 3-6D; average of 4 experiments). Discussion In this study, we showed that hyperpolarizing stimuli and granule-cell mediated IPSPs can bidirectionally control spike generation in mitral cells by recruiting opposing inward and outward currents. Smaller hyperpolarizing stimuli and unitary IPSPs, which do not often hypepolarize mitral cells by more than 5 mv, promote spiking in depolarized mitral cells by recovering subthreshold sodium currents that then produce rebound depolarizations. In contrast, larger hyperpolarizations and summating IPSPs also recruit a barium sensitive outward current which counteracts the effects of voltage gated sodium channel recovery and significantly delay the repolarization of the mitral cell membrane. These large hyperpolarizing stimuli exert a powerful inhibitory influence on the generation of spike clusters evoked by phasic stimuli that mimic trains of inspirationevoked slow EPSPs. 80

91 Rebound spiking is regulated by the differential recovery of subthreshold Na and I A -like K currents Previous work in many neurons showed that rebound spike generation after hyperpolarizing stimuli often depends on recovery from inactivation of low-threshold voltage dependent calcium currents (Jahnsen and Llinas, 1984a; McCormick and Bal, 1997). In contrast, our data show that, in mitral cells, rebound spiking uses a mechanism reminiscent of classical anode-break excitation requiring recovery of voltage dependent sodium channels (Johnston and Wu, 1995), especially subthreshold sodium currents, during a hyperpolarizing stimulus (Fig. 3-7). In mitral cells, large (> 10 mv) hyperpolarizations caused a pronounced membrane potential rectificiation that slowed the rate of repolarization after the offset of the hyperpolarizing stimulus (Fig. 3-7). This rectification became more pronounced both with the degree of hyperpolarization and the duration of the hyperpolarizing stimulus, which suggested that it was due either to a hyperpolarization-activated cation current (such as I h ) or by recovery from inactivation of voltage dependent potassium currents. Blocking I h (with Cs + ), delayed rectifier K channels (with TEA), or KCNQ type currents (with XE-991) had little effect on hyperpolarization induced membrane potential rectification. Instead, only high concentrations of 4-AP (2-6 mm) and Ba produced a significant reduction in membrane potential rectification. Implications for olfactory processing 81

92 Our results suggest that, by promoting rebound spike generation, granule cell IPSPs can promote synchronization across populations of mitral cells. After activation of a mitral cell by an olfactory stimulus, dendrodendritic inhibition could recruit other activated mitral cells to synchronously fire together. This synchronization could occur both within a glomerular module, to ensure proper temporal processing of signals at higher centers, or across glomerular modules, to widen the spatio-temporal pattern of activity in the bulb and allow for unambiguous coding of a wider variety of odors (Stopfer et al., 1997; Laurent, 2002; Perez-Orive et al., 2002). In contrast, larger IPSPs, produced by synchronous activation of groups of granule cells during odor processing, would be expected to inhibit groups of mitral cells and limit the spatial extent of mitral cell activation. Thus, IPSPs can serve as a powerful mechanism to bidirectionally control spiking and synchronization of mitral cells, and therefore dynamically control evolving spatiotemporal patterns of activity in the olfactory bulb. Several questions about the functional impact of dendrodendritic inhibition, however, still remain. First, the strength and duration of IPSPs activated by single action potentials in mitral cells is not known. Previous work has shown that synchronized gamma-frequency oscillatory activity in granule cells can gate the strength and selfinhibitory potential of single mitral cell action potentials (Halabisky and Strowbridge, 2003). However, it is unclear what the properties of single spike evoked recurrent IPSPs are during ongoing olfactory processing in vivo. For instance, while we found that single granule cell layer shocks produced small amplitude unitary IPSPs, single mitral cell action potentials in vivo may activate recurrent networks that produce long lasting trains 82

93 of IPSPs and synchronous granule cell activity (Isaacson and Strowbridge, 1998; Schoppa et al., 1998; Lagier et al., 2004). In addition, it is still unclear how processes that control the strength and extent of dendrodendritic synaptic transmission, such as the extent of action potential back propagation in mitral cell secondary dendrites (Margrie et al., 2001; Xiong and Chen, 2002) and the amount of active propagation of excitatory stimuli in granule cells (Egger et al., 2003, 2005), might control the balance between spike initiation and spike inhibition by IPSPs. Further work on these issues will be of critical importance to understand information coding by olfactory neural networks. 83

94 Figure 3-1. Transient hyperpolarizing stimuli evoke rebound discharges in mitral cells. (A) Schematic diagram of olfactory bulb circuitry showing relative positions of recording pipette and extracellular stimulating electrodes. (B) Granule cell layer stimulation evoked IPSPs in mitral cells that were blocked by picrotoxin (50 µm) and reversed polarity at -73 mv. (C) Evoked IPSPs triggered rebound depolarizations (arrow) at membrane potentials near threshold (at -40 mv) but not at more hyperpolarized voltages (-45 mv). Picrotoxin-sensitive IPSPs triggered rebound bursts (D) and evoked correlated discharges in two simultaneously recorded mitral cells (E). (F 1 ) Raster plot of rebound spiking activity triggered by IPSPs in five trials. (F 2 ) Small hyperpolarizing pulses (100 ms duration) also triggered rebound discharges in the same mitral cell. (G) Stereotyped rebound discharges triggered by graded hyperpolarizing steps; rebound firing was slowed following large-amplitude hyperpolarizing steps. 84

95 Figure

96 Figure 3-2. Voltage dependence of rebound spiking in mitral cells. (A 1 ) Responses of a mitral cell held near (top trace, -43 mv) and slightly below (bottom trace, -49 mv) spike threshold to a graded series of hyperpolarizing current steps (100 ms duration). Rebound spikes were evoked only by moderate amplitude steps when the cell was held near spike threshold. (A 2 ) Enlargements of the three rebound responses indicated in A 1. (A 3 ) Graph of relationship between hyperpolarizing step amplitude and spike probability near (closed circles) and slightly below (-5 mv; open circles) threshold in 7 mitral cells. (B 1 ) Rebound discharges triggered by weak but not by large-amplitude hyperpolarizing steps in the same mitral cell. Response to large-amplitude step shows a delayed repolarization following the step offset. (B 2 ) Summary of the number of rebound spikes triggered by weak (< 5 mv) and strong (> 15 mv) hyperpolarizing steps in 4 mitral cells. ** P <

97 Figure

98 Figure 3-3. Subthreshold Na currents boost mitral cell responses to depolarizing stimuli and mediate rebound spiking. (A) Tetrodotoxin (TTX, 1 µm) blocked both sodium-dependent action potentials and a sustained subthreshold depolarization in mitral cells. (B 1 ) TTX-sensitive subthreshold sodium currents boosted mitral cell responses to slow, phasic depolarizations. (B 2 ) Summary plot of TTX-sensitive amplification of each response to a train of phasic depolarizations (sepsp 1-4 ). * P < (C) Rebound spiking is mediated by TTX sensitive sodium currents. Response of a mitral cell to a graded series of hyperpolarizing current steps before and after TTX and following washout of TTX. Insets show enlargements of the three responses indicated by arrows. (D 1 ) The sodium channel blocker riluzole (10 µm) also attenuated the sustained subthreshold Na current and blocked rebound discharges without blocking Na channel mediated action potentials (inset). (D 2 ) Hyperpolarizing steps failed to trigger rebound discharges even when membrane depolarization was increased to compensate for the attenuation of the subthreshold Na current by riluzole. 88

99 Figure

100 Figure 3-4. Rebound spiking is regulated by the duration of hyperpolarizing inputs. (A) Responses to varying duration hyperpolarizing steps. Rebound discharges were triggered only by the shortest step while longer duration steps slowed membrane potential repolarization. (B) Superposition of responses shown in A, aligned by the hyperpolarizing step offset. (C) Plot of the repolarization latency (to 90 % recovery following step offset) versus hyperpolarization step duration. This relationship was fit by a single exponential function with a tau of 88.0 ± 11 ms (solid line; n = 4 cells). The mean maximum repolarization delay was ± 13.3 ms (n = 4). 90

101 Figure

102 Figure 3-5. Prolonged hyperpolarizing steps recruit a slowly-inactivating K current in mitral cells. (A) Delayed repolarization (arrows) persisted following blockade of voltage-gated Na and Ca channels with 1 µm TTX, 200 µm Cd and 100 µm Ni. 4-Aminopyridine (4-AP; 6 mm) blocked the delayed repolarization following hyperpolarizing steps recorded under current clamp. (B) Summary of the effects of K and Na channel blockers on the rectification (V 1 -V 2 ; see inset) caused by delayed repolarization in mitral cells. Only Ba (2 mm) and 6 mm 4-AP significantly reduced rectification; * P < TTX (1 µm), TEA (25 mm), Cs (2 mm), XE-991 (10 µm), E4031 (5 µm), dofetilide (Dof; 10 µm) and 100 µm 4-AP had no significant effects on rectification. (C) Voltage clamp responses of mitral cells recorded from -80 mv in the presence of TTX (1 µm), Cd (200 µm), Cs (2 mm), Ni (100 µm) and nifedipine (100 µm). Most of the transient K current was blocked by 6 mm 4-AP. (D) Transient K currents evoked by steps to -20 mv in mitral cells were completely inactivated by 1 sec duration pre-pulses to -40 mv. 92

103 Figure

104 Figure 3-6. Brief hyperpolarizations control mitral cell discharges in an all-or-none manner. (A) Mitral cell responses to slow phasic depolarizations (4 alpha functions; tau = 80 ms). At low amplitudes, this stimulus waveform evoked all-or-none clusters of action potentials at the peak of the last three sepsps. Action potential timing in 11 successive trials indicated in raster plot above voltage trace. (B 1 ) The same stimulus waveform evoked reproducible discharge patterns when presented at increased amplitude. (B 2 ) Injecting a brief (negative alpha function; tau = 10 ms) hyperpolarization immediate before sepsp 3 abolished the discharge that was normally triggered by that depolarization. (C 1 ) Control response to the phasic depolarization stimulus in another mitral cell. (C 2 ) Varying the timing of the brief hyperpolarization (sipsp; arrows) gated the discharge on sepsp 3 in an all-or-none manner. (D) Plot of the number of action potentials evoked by sepsp 3 as a function of the hyperpolarization latency (sipsp; timing indicated from onset of sepsp 3 ). 94

105 Figure

106 Figure 3-7. Summary of intrinsic mechanisms regulating rebound discharges in mitral cells (A) Weak hyperpolarizations appear to trigger rebound spikes by de-inactivating a subthreshold Na current. (B) Larger hyperpolarizations fail to trigger rebound spikes because they also de-inactivate K currents that delay membrane potential repolarization. 96

107 Figure

108 Chapter 4 : Multiple Modes of Synaptic Excitation Onto Granule Cells of the Olfactory Bulb Summary Granule cells, the most common GABAergic cell type in the olfactory bulb, play a critical role in shaping the output of this brain region. Relatively little is known about the synaptic mechanisms responsible for activating these interneurons. While mitral cells are known to contact granule cell dendrites through specialized dendrodendritic synapses on distal dendrites, the source of the principal excitatory input to proximal dendrites has not been established. Using 2-photon guided minimal stimulation in acute rat brain slices, we found that distal and proximal excitatory synapses onto granule cells are functionally distinct. Proximal synapses arise from piriform cortical neurons and facilitate with paired-pulse stimulation while distal dendrodendritic synapses generate EPSCs with slower kinetics that depress with paired stimulation. Most excitatory synapses we examined activated both NMDARs and AMPARs while a subpopulation appeared to be NMDAR silent. The convergence of two types of excitatory inputs onto GABAergic granule cells provides a mechanism for populations of cortical neurons to regulate lateral inhibition in the olfactory bulb, and thereby the degree of inter-glomerular processing of sensory input. Introduction 98

109 The olfactory bulb plays a critical role in transforming monotonic sensory inputs into complex spatio-temporal patterns of action potentials that are transmitted to cortical regions. The output of this second-order brain region is determined by firing patterns of its principal neurons, mitral and tufted cells (Shepherd and Greer, 1998; Fig. 4-1A). Their firing patterns, in turn, are governed by a large array of unusual and poorly understood intrinsic and synaptic conductances that affect how mitral and tufted cells respond to sensory input. Inhibitory synaptic interactions play a central role in shaping mitral and tufted cell responses to sensory stimuli. Large inhibitory postsynaptic potentials (IPSPs) are a common feature of intracellular recordings from principal cells in the olfactory bulb (Hamilton and Kauer, 1985, 1989) and most likely arise from local bulbar synaptic circuits since the afferent input from sensory neurons is purely excitatory (Aroniadou-Anderjaska et al., 1997). Locally-generated IPSPs play a critical role in patterning mitral and tufted cell output (Hamilton and Kauer, 1985, 1989) and also may contribute to the genesis of the large amplitude gamma-frequency oscillations frequently recorded in the olfactory bulb (Adrian, 1950; Friedman and Strowbridge, 2003; Lagier et al., 2004). Blockade of inhibitory postsynaptic responses potentiates mitral cell responses to odorants (Yokoi et al., 1995). Related experiments using picotoxin to block GABA A receptor mediated synaptic transmission in the antennal lobe in honeybees, the brain structure analogous to the mammalian olfactory bulb, impairs olfactory discrimination (Stopfer et al., 1997), suggesting that higher brain regions make use of the temporal information conveyed by principal cell spike patterns when interpreting complex sensory stimuli. 99

110 In contrast to the well documented effects of inhibition on principal neuron firing patterns, little is known about how the activities of the local interneurons that generate principal cell inhibition are regulated. Granule cells, the most abundant GABAergic interneuron in the olfactory bulb (Shepherd and Greer, 1998), receive two anatomically distinct classes of excitatory input on their proximal and distal dendrites (Fig. 4-1B). Reciprocal dendrodendritic synapses with mitral cells are the primary distal source of excitatory input (Rall et al., 1966; Price and Powell, 1970b, c). Granule cell activation through distal dendrodendritic synapses, however, depends on NMDA receptors which are tonically blocked by extracellular Mg ions (Isaacson and Strowbridge, 1998; Schoppa et al., 1998; Chen et al., 2000a). Tetanic stimulation of axons in the granule cell layer not only activates granule cells but also relieves the Mg blockade of NMDA receptors at distal dendrodendritic synapses (Halabisky and Strowbridge, 2003). These results suggest that the proximal excitatory inputs to granule cells may play an important role in gating recurrent and lateral dendrodendritic inhibition in the olfactory bulb. The primary source of these proximal excitatory inputs is unclear; previous studies have suggested that both mitral cell axon collaterals (Price and Powell, 1970c; Orona et al., 1984) and centrifugal cortical axons (de Olmos et al., 1978; Haberly and Price, 1978; Shipley and Adamek, 1984) innervate granule cells at proximal synapses. The explanation for the selective ability of gamma frequency stimuli, and not single stimuli or lower frequency trains, to gate dendrodendritic inhibition (Halabisky and Strowbridge, 2003) also is unclear. Presumably this relates to specific forms of shortterm plasticity (facilitation or depression) at proximal excitatory synapses on granule cells. The one study that addressed short-term plasticity in granule cells (Dietz and 100

111 Murthy, 2005) found evidence for both facilitating and depressing excitatory synapses onto granule cells. However, it was not clear if individual granule cells receive different types of excitatory input or if this heterogeneity reflects multiple functionally-defined subpopulations of granule cells, as suggested in the Dietz and Murphy study. In this study, we defined the functional properties and short-term plasticity in the two known types of excitatory synapses onto granule cells in the rat olfactory bulb. We circumvented the primary problem encountered using focal stimulation in complex brain regions the lack of specificity in the type of presynaptic process activated by employing 2-photon imaging to position fine stimulating electrodes very close to proximal or distal dendritic segments of granule cells recorded under whole-cell voltageclamp conditions. Using this method, we found that proximal axo-dendritic and distal dendrodendritic excitatory synapses form two homogeneous classes of inputs that have different kinetics and different forms of short-term plasticity. Using 2-photon guided minimal stimulation we also examined the origin of the short-term plasticity and tested for AMPA and NMDA receptor silent synapses on granule cells. In the hippocampus and several other brain regions, AMPA receptor silent synapses are intimately associated with expression mechanisms for long-term plasticity (Malinow and Malenka, 2002; Isaac, 2003). The low abundance of AMPA receptor silent synapses we found on granule cells may help explain why classical protocols that readily induce long-term potentiation in the hippocampus fail to trigger plasticity in the olfactory bulb. Finally, we used slices with spontaneously bursting mitral cells, as well as a novel combined olfactory bulb/anterior piriform cortex slice preparation, to test whether proximal excitatory inputs to granule cells arise from the piriform cortex or from mitral cell axon collaterals. Together these 101

112 results suggest that much of the inhibition in the olfactory bulb is governed by the relative timing of two independent excitatory inputs to granule cells: distal dendrodendritic synapses with the mitral cell secondary dendrites and proximal inputs from cortical pyramidal cells that project back to the olfactory bulb. Experimental Procedures Slice preparation and recording Horizontal slices (300 µm) through the olfactory bulb or ventral hippocampus were prepared from anesthetized (ketamine, 140 mg/kg ip) P10-21 Sprague-Dawley rats using a modified Leica (Nussloch, Germany) VT1000S vibratome, as described previously (Isaacson and Strowbridge, 1998; Halabisky et al., 2000; Halabisky and Strowbridge, 2003). An artificial cerebrospinal fluid (ACSF) dissection solution with reduced Ca was used when preparing and storing slices. This solution contained 124 mm NaCl, 2.6 mm KCl, 1.23 mm NaH 2 PO 4, 3 mm MgSO 4, 26 mm NaHCO 3, 10 mm dextrose, and 1 mm CaCl 2, equilibrated with 95% O 2 /5% CO 2 and was chilled to 4 C during slicing. For experiments investigating cortical feedback projections, we made larger horizontal slices including both the olfactory bulb and a section of anterior piriform cortex. Combined olfactory bulb/piriform cortex slices were prepared by attaching the ventral surface of a block of brain tissue containing both olfactory bulbs and frontal lobes to the vibratome stage. Horizontal slices prepared from this block at least 1500 µm from the ventral surface contained both the olfactory bulb and a portion of the anterior piriform cortex. Some slices also included some of the anterior olfactory nucleus (AON) located medial to 102

113 the piriform cortex. In slices we first identified the AON/piriform cortex boundary to ensure that all stimulation and dye injection sites were confined to piriform cortex and did not include the AON. Anterior piriform cortex was readily identified by its lateral location, laminated structure, and cellular morphology using infrared-differential intereference contrast (IR-DIC) microscopy. Brain slices were incubated in a 30 C water bath for 30 min and then maintained at room temperature. During experiments, slices were superfused with ACSF at room temperature that contained 124 mm NaCl, 3 mm KCl, 1.23 mm NaH 2 PO 4, 1.2 mm MgSO 4, 26 mm NaHCO 3, 10 mm dextrose, and 2.5 mm CaCl 2, equilibrated with 95% O 2 /5% CO 2. Whole-cell patch-clamp recordings were made from mitral and granule cells in the olfactory bulb and CA1 hippocampal pyramidal neurons visualized under IR-DIC optics using an Olympus BX51WI fixed-stage upright microscope and an Axopatch 1D amplifier (Axon Instruments). Patch electrodes used for granule and hippocampal cell voltage clamp recordings (5-7 MΩ resistance) contained (in mm): Cs-methanesulfonate 115, NaCl 4, TEA-methanesulfonate 25, QX-314 5, HEPES 10, EGTA 1, MgATP 4, Na 3 GTP 0.3, and phosphocreatine 10. For experiments investigating hippocampal silent synapses and LTP, we evoked EPSCs onto single CA1 pyramidal cells through Schafer collateral inputs using a glass monopolar stimulating electrode filled with HEPES buffered saline (124 mm NaCl, 3 mm KCl, 10 mm HEPES, ph adjusted to 7.4) placed in the stratum radiatum connected to a constant current stimulus isolation unit (WPI). In most experiments, 100 µm Alexa594 was added to the pipette solution to visualize neuronal morphology. In other experiments, 150 µm Oregon Green BAPTA-1 was added to the patch solution in place of EGTA to visualize synaptically-evoked calcium 103

114 transients. For whole cell current-clamp recordings from mitral cells, patch pipettes (3-5 MΩ resistance) containing (in mm) K-methylsulfate 140, NaCl 4, HEPES 10, EGTA 0.2, MgATP 4, Na 3 GTP 0.3, and phosphocreatine 10 were used. All recordings were obtained in the presence of gabazine (10 µm) to block GABA A -receptor mediated synaptic events. All chemicals were obtained from Sigma except for Alexa594 hydrazide and Oregon Green BAPTA-488 hexapotassium salt (Invitrogen). Two-Photon Imaging Live imaging experiments utilized a custom two-photon microscope based on the Verdi V10 pump laser, Mira 900 Ti-sapphire laser (both from Coherent, Santa Clara, CA) and a high-speed XY galvanometer mirror system (6210; Cambridge Technology). Intracellularly loaded fluorescent dyes were excited at 830 nm through a 60 waterimmersion objective (Olympus). Emitted light was detected through an epifluorescent light path that included a 700DCLPXR dichroic mirror, a BG39 emission filter (both from Chroma Technology), and a cooled PMT detector module (H7422P-40; Hamamastu). Photomultiplier output was converted into an analog voltage by a highbandwidth current preamplifier (SR-570; Stanford Research Systems). Custom Visual Basic software written by BWS controlled the scanning system and image analysis functions. Laser beam intensity was controlled electronically through a Pockels cell attenuator (ConOptics) and a shutter (Uniblitz). In most experiments, the output of the Mira laser was attenuated by 90% 95%. 104

115 Two-Photon Guided Microstimulation We used two-photon microscopy to guide the focal stimulation of different populations of excitatory inputs on granule cells. After waiting for minutes to allow the dye to diffuse from the patch pipette into distal processes, we used a fast scanning mode (3200 lines/sec) to visualize granule cell morphology and spines. We then placed a glass stimulating pipette (tip opening ~ 1 µm) containing HEPES buffered saline (124 mm NaCl, 3 mm KCl, 10 mm HEPES, µm Alexa594, ph adjusted to 7.4) approximately µm away from distal spines located in the external plexiform layer or proximal spines located in the granule cell layer under two-photon guidance. This stimulating electrode was connected to a constant-current stimulus isolation unit (WPI) and used to evoke neurotransmitter release from presynaptic terminals located near granule cell dendrite segments of interest. We used 2-photon Ca imaging in some experiments to confirm that this focal stimulation method reliably activated spines near the stimulating electrode. Subsequent experiments using glutamate receptor antagonists (e.g., Fig. 4-4A) confirmed that these spines were activated synaptically and not by passive depolarization from the stimulating electrode. We used 2-photon imaging to position Alexa594-filled pipettes near visualized dendritic segments for both minimal and supraminimal stimulation experiments. We evoked supraminimal responses by increasing the stimulus intensity until there were no failures and the response amplitude was approximately twice the amplitude of successes evoked by minimal stimulation (typically 2-3X the intensity used for minimal stimulation in the same experiment). 105

116 Imaging Synaptically Evoked Calcium Transients To image synaptically evoked calcium transients, we filled granule cells voltage-clamped at -70 mv with 150 µm Oregon Green BAPTA-488. After waiting for minutes to allow the dye to diffuse into distal processes, we placed an Alexa594-filled stimulating pipette near a group of granule cell spines (see above) to evoke neurotransmitter release onto spines of interest. Calcium transients were imaged by taking sequential images in a fast-scanning mode (25 ms frame rate, 3200 lines/sec) during extracellular stimulation and then calculating the percent change in fluorescence over baseline. Regions of interest were placed over dendrites and spines to calculate changes in dendritic and spine calcium levels, respectively. For these experiments, we used a Mg-free ACSF to maximize calcium influx through NMDA receptors (Isaacson and Strowbridge, 1998). DiI Injections To visualize the trajectory of cortical axons as they entered the olfactory bulb, we injected a bolus of DiIC 18 (3) (3 mm in ethanol) into the anterior piriform cortex using a patch pipette connected to a picospritzer (2 psi, 500 ms) in combined olfactory bulb/piriform cortex slices. After recording, slices were fixed in phosphate buffered saline (PBS) containing 4 % paraformaldehyde overnight at 4 C. Fixed slices were then placed in PBS and kept at room temperature for 2 weeks to allow DiI to diffuse through axonal processes. To visualize DiI labeled processes, slices were whole-mounted onto microscope slides, coverslipped and imaged using epifluorescence microscopy (Zeiss Axioskop 2). Fluorescent images were digitized using an Olympus DP70 CCD camera. 106

117 Data Acquisition and Analysis Electrophysiological data were recorded and analyzed using custom software written in Visual Basic 6 (Microsoft) and Origin 7.5 (OriginLab). Current and voltage records were low-pass filtered at 2 khz and then digitized at 5 khz using a 16-bit A/D converter (ITC- 18, Instrutech). Series resistance was typically <20 MΩ and was routinely compensated by >80% in voltage-clamp experiments. Evoked and spontaneous EPSCs were detected by the first derivative (slope threshold = 3 pa/ms) using custom software written in Visual Basic 6 and verified by visual inspection. For evoked events, EPSC amplitudes were measured by calculating the average value during a 1 ms window surrounding the peak relative to the average baseline value in a 5 ms window immediately before the stimulus. For spontaneous events, the baseline value was measured by calculating the average value during a 5 ms window immediately before the EPSC onset. For both evoked and spontaneous events, the peak time was measured by calculating the time point when the 1 st derivative crossed from negative to positive. Evoked responses were categorized as failures if the threshold for slope change was not reached in a 20 ms window following the stimulus. For these events, the EPSC amplitude was measured as the average value during a 1 ms window 10 ms after the stimulus relative to the average baseline value immediately before the stimulus % rise times were calculated by subtracting the time after EPSC onset to reach 90% of the peak value from the time to reach 10% of the peak value. Failure rates were calculated by dividing the number of trials with no slope change by the total number 107

118 of responses. In a subset of neurons, we verified this method by also calculating failure rate by first measuring the EPSC amplitude for all trials in a 1 ms time window 10 ms after the stimulus onset, doubling the number of responses with amplitude > 0 pa and dividing by the total number of trials (Liao et al., 1995). These two methods did not show any significant differences. In order to provide a visual estimate of the proportion of failures, we included dashed vertical lines at 2X the noise S.D. in the amplitude histograms in Fig. 4-7B. Some olfactory bulb slices exposed continuously to GABA A receptor antagonists developed spontaneous mitral cell discharges. This spontaneous activity was obvious in intracellular recordings from mitral cells and also could be detected as barrages of slow spontaneous EPSCs in granule cells with dendrites in the EPL. While we did not explore the cellular basis of this spontaneous activity in this study, this activity did not appear to represent all-or-none synchronous discharges of large populations of mitral cells. In most slices in which spontaneous discharges occurred, these discharges were relatively infrequent and did not contaminate the evoked response analysis. We also confirmed the major findings from this study (different kinetics and short-term plasticity for proximal and distal granule cell synapses, lack of AMPA receptor silent synapses on granule cells) in slices that did not exhibit spontaneous discharges. Except for the experiments presented in Fig. 4-8, all data were obtained for slices in which spontaneous discharges were either not evident or occurred with intervals of at least 45 sec. In a small subset of slices with more frequent discharges, we took advantage of this periodic synaptic drive to granule cells to test whether mitral cell axon collaterals contact nearby granule cells (Fig. 4-8). 108

119 In the minimal stimulation experiments we assumed activation of a single presynaptic axon or dendrite if we observed: (1) all-or-none EPSCs, that (2) gradually increasing stimulus intensity resulted in an abrupt transition from all failures to all-ornone responses, and that (3) small changes in the stimulus intensity beyond that threshold had no change in the amplitude of successes. For these experiments, we calculated paired-pulse ratios by measuring both the ratio of average EPSC amplitudes (both failures and successes) and by calculating the ratio of failure rates. We typically analyzed trials for each cell. For supraminimal stimulation experiments, the paired-pulse ratio was measured by averaging trials and calculating the ratio of average EPSC amplitudes. Data are presented as mean ± S.E.M. Unless otherwise noted, statistical significance was determined using Student s t-test. Results Two distinct classes of excitatory inputs onto granule cells Granule cells receive frequent spontaneous excitatory synaptic responses. We first sought to define the properties and presynaptic source of these synaptic inputs by voltage-clamping granule cells at -70 mv, near the reversal potential of spontaneous inhibitory currents, and by adding the GABA A receptor antagonist gabazine (10 µm) to the extracellular solution. In most granule cells, we recorded a large range of both spontaneous excitatory postsynaptic current (sepsc) amplitudes and rise times, as illustrated by the sweeps shown in Fig. 4-1C. All spontaneous synaptic events we 109

120 recorded under these conditions were blocked by the non-nmda receptor antagonist NBQX (10 µm; n = 5 granule cells), suggesting that they were glutamatergic. While this broad range of sepsc may reflect differing degrees of electrotonic attenuation (Jack et al., 1983), the bimodal distribution of sepsc rise times (Fig. 4-1D) suggests that this may not be the primary explanation for the diversity of spontaneous EPCSs. The rise time distribution in this granule cell was well fit by the sum of two Gaussian distributions (peaks at 1.4 and 3.8 ms) and contained a clear gap between the two peaks at ~ 2.5 ms. We found that the amplitude and rise time of individual sepscs were not correlated (R = 0.43), which also argued against a simple electrotonic attenuation explanation. While the sepsc amplitude distribution was not biomodal, the mean amplitudes of fast- and slowrising EPSCs were significantly different (-22.0 ± 1.2 pa for fast-rising sepscs with rise times < 2.5 ms versus ± 0.4 pa for sepscs with rise times > 2.5 ms; P < 0.01). The bimodal rise time distribution we found for the granule cell shown in Fig. 4-1B, and for three other visualized granule cells that had dendritic processes in the external plexiform layer (EPL), suggested that granule cells receive two types of excitatory inputs that generate kinetically distinct postsynaptic responses. This hypothesis is consistent with anatomical studies that showed that granule cells form two morphological types of dendritic synapses (Price and Powell, 1970b, c). As discussed previously, granule cells with dendrites that were truncated before the EPL had a different, unimodal distribution of sepsc rise times, presumably because they did not receive the class of excitatory inputs that preferentially innervate distal dendrites. We used 2-photon guided minimal stimulation (2PGMS) to determine if fastrising EPSCs (< 2.5 ms 10-90% rise time) corresponded to proximal inputs and the slow- 110

121 rising EPSCs (> 2.5 ms rise time) to distal dendrodendritic inputs. In most of these experiments, both the granule cell recorded under voltage-clamp and the focal stimulating electrode were filled with Alexa594 and were visualized using 2-photon microscopy. We positioned the stimulating pipette near a visualized dendritic segment to selectively activate distal (in the EPL) or proximal (in the GCL) excitatory synaptic inputs (Fig. 4-2A). To verify that this method effectively activated spines near the stimulating pipette we first conducted a series of experiments with granule cells filled with 150 µm Oregon Green BAPTA 1 (OGB-1) instead of Alexa594. As shown in Fig. 4-2B and C, supraminimal stimulation (200 µs, 43 µa; single shock) using this technique selectively triggered Ca influxes in two of three imaged proximal dendritic spines that were near (~ 20 µm) the stimulation pipette. At this stimulus intensity no Ca accumulation was detected in the third spine or in the neighboring dendrite shaft, suggesting that the nearby spines were activated by synaptic inputs and not by passive depolarization from the stimulating electrode. We also show below that glutamate receptor antagonists (NBQX and D-APV) completely block the electrical response of granule cells to this form of focal stimulation. This localized pattern of Ca accumulation was repeatable across multiple trials (Fig. 4-2B), suggesting that this microstimulation protocol reliably evoked neurotransmitter release from a localized group of presynaptic terminals near the stimulating electrode. We observed similar results showing Ca transients in subsets of spines near the stimulating pipette, and the absence of Ca accumulations in dendritic shafts, with both proximal and distal 2-photon guided stimulation (n = 4 cells). Using 2PGMS with both the recording and stimulating electrodes filled with Alexa594, weak focal stimuli evoked unitary EPSCs at both distal (Fig. 4-2D 1 ) and 111

122 proximal (Fig. 4-2D 2 ) stimulus sites in an all-or-none manner. Both stimulus sites showed abrupt response thresholds above which unitary responses (successes) could be clearly distinguished from failures. While the unitary response amplitude was relatively constant following small increases in stimulus intensity in each cell, response amplitudes were variable across the population of cells tested (mean unitary amplitude was ± 1.7 pa (n = 10 cells) for proximal and ± 0.7 pa (n = 10 cells) for distal 2PGMS). As expected, unitary responses evoked by 2PGMS near proximal dendritic locations showed little latency jitter (mean latency S.D. = 1.09 ±.05 ms; n = 10 cells). By contrast, most responses evoked at distal sites showed pronounced jitter (see example traces in Fig. 4-2D 1 ; mean latency S.D. = 2.29 ± 1.0; n = 10 cells; significantly different from the S.D. of proximal 2PGMS response latencies; P < 0.01). The homogenous nature of the distal 2PGMS responses, the relatively low intensity stimulus intensity used in these experiments, and the absence of any anatomical evidence for recurrent excitatory pathways in the external plexiform layer (Schoppa and Urban, 2003) suggest that the distal responses were monosynaptic. The origin of this distal EPSC latency jitter may reflect biophysical differences (e.g., possibly lower Na channel densities and longer membrane time constants) in presynaptic dendritic compartments, compared with presynaptic axon segments. Theoretically, brief ( µs) extracellular stimuli should more efficiently excite thin neuronal structures with small chronaxes, such as axons, than larger diameter dendrites (Ranck, 1975). Unitary EPSCs evoked by proximal and distal 2PGMS had different kinetics and closely resembled fast and slow-rising spontaneous EPSCs, respectively (compare example traces in Fig. 4-2D with spontaneous examples in Fig. 4-1C). The rise time 112

123 distribution for proximally-evoked EPSCs (Fig. 4-3A 2 ) was smaller than the analogous plot for EPSCs evoked by distal 2PGMS (Fig. 4-3A 1 ). Fig. 4-3B shows the overall rise time distribution for 10 granule cells with distal and 10 different granule cells with proximal 2PGMS. The combination of the two Gaussian distributions in this plot closely resembles the bimodal distribution of spontaneous EPSC rise times (Fig. 4-1D), suggesting that the diversity of excitatory synapses onto granule cells activated by 2PGMS is similar to that found in spontaneous synaptic inputs. Across our population of granule cell recordings, the mean distal 2PGMS EPSC rise time (4.20 ± 1.3 ms; n = 10) was significantly greater than the proximal EPSC rise time (1.36 ± 0.42 ms; n = 10; P < 0.01). Distal and proximal excitatory synapses are functionally distinct Synaptic responses with different kinetics may arise from multiple mechanisms. Differences in EPSC kinetics may reflect different positions of activated synapses along the dendritic tree that generate different degrees of electrotonic attenuation or they may reflect biophysical differences between different types of synapses (e.g., receptor subunit composition), or a combination of these two mechanisms. We used paired-pulse stimulation to determine if the differences between proximal and distal 2PGMS responses reflected multiple types of excitatory synapses with different functional properties. As shown in Fig. 4-4A, responses to distal stimulation near dendritic segments in the EPL (using both 2PGMS and 2-photon guided supraminimal stimulation methods) showed paired-pulse depression (mean distal 2PGMS paired-pulse ratio (PPR) = 0.74 ± 0.09; n = 113

124 7; mean supraminimal PPR = 0.67 ± 0.07; n = 7). By contrast, proximal stimulation near dendritic segments in the GCL showed paired-pulse facilitation (mean proximal 2PGMS PPR = 1.74 ± 0.22; n = 5; mean supraminimal PPR = 1.50 ± 0.13; n = 10). The proximal/distal difference in paired-pulse ratio was statistically significant for both 2PGMS and supraminimal stimulation (P < 0.01; Fig. 4-4B). We also noted similar mean paired-pulse ratios using both 2PGMS and supraminimal stimulation in the same stimulus location, suggesting that the cellular mechanisms responsible for these forms of shortterm plasticity are unlikely to reflect neurotransmitter spillover (Isaacson et al., 1993). We also verified that the differences we found in proximal and distal stimulation experiments did not reflect functionally distinct subpopulations of granule cells by demonstrating that paired stimulation in the EPL induced depressing responses while stimulation in the GCL induced facilitating responses in the same granule cell (see red circles in Fig. 4-4B, right). We found a very high correlation between the form of shortterm plasticity and the stimulus position. In a survey of 43 granule cells tested with 2- photon guided supraminimal paired stimulation, 95 % (21 of 22) of the experiments with the stimulus electrode positioned near proximal apical dendrites showed facilitating responses and 86 % (18 of 21) with distal stimulation in the EPL showed paired-pulse depression. Responses to both proximal and distal stimulation at -70 mv were blocked completely by the non-nmda receptor antagonist NBQX (10 µm; grey traces in Fig. 4-4A), indicating that both stimulation sites activated purely glutamatergic postsynaptic responses. The different forms of short-term plasticity evident in these focal stimulation experiments suggest that proximal and distal stimuli activate different types of glutamatergic synapses onto granule cells. 114

125 Both forms of short-term plasticity appear to result from changes in presynaptic release properties. The paired-pulse depression observed with distal 2PGMS was associated with a statistically significant increase in the failure rate (from 0.49 ± 0.04 on stim 1 to 0.62 ± 0.04 on stim 2 ; P < 0.05; n = 7; Fig. 4-4C left) without a change in the average amplitude of successes (mean potency ratio (R 2 /R 1 ) = 0.99 ± 0.05; n = 7). Similarly, the facilitation of responses with proximal 2PGMS was associated with a significant decrease in the failure rate (from 0.59 ± 0.10 on stim 1 to 0.42 ± 0.09 on stim 2 ; P < 0.05; n = 5; Fig. 4-4C right), also without a change in potency (potency ratio = 1.03 ± 0.07; n = 5). For both proximal and distal 2PGMS, the PPR calculated by mean response amplitudes was strongly correlated with the PPR calculated by changes in failure rate (R = 0.97; Fig. 4-4D), consistent with presynaptic expression mechanisms for both forms of short-term plasticity. Proximal and distal glutamatergic synapses also differed in their degree of depression during high-frequency stimulus trains. As shown in Fig. 4-5A, responses to distal stimulation in the EPL were silenced by the fourth stimuli in a 50 Hz stimulus train. Responses to proximal stimulation initially showed facilitation, followed by steady-state depression and persisted following each stimuli in the train. Fig. 4-5B shows a summary of 4 proximal and 4 distal experiments using similar stimulus trains. Proximal synapses showed significantly less steady-state depression at the end of these stimulus trains than did distal synapses (40.0 ± 3.8 % versus 19.9 ± 2.3 % percent of initial response; P < 0.01; Fig. 4-4B inset). We also tested whether proximal and distal synapses showed differences in the frequency dependence of their paired-pulse modulation. As shown in Fig. 4-5C, paired-pulse modulation of both proximal and distal synapses was maximal 115

126 with inter-stimulus intervals (ISI) less than 200 ms; depression was maximal with very short ( < 20 ms) ISIs while facilitation was maximal with slightly longer ISIs (20-50 ms). Both forms of paired-pulse modulation were abolished with ISIs of 1 second or greater. The results presented thus far suggest that proximal and distal excitatory synapses onto granule cells are functionally distinct since they have different kinetics, different forms of short-term plasticity and different degrees of steady-state depression. We next examined responses to proximal and distal stimulation at different membrane potentials to determine if there were also differences in the NMDA receptor components of the EPSCs. As expected, blockade of NMDA receptors with D-APV (50 µm) had little effect on the response to either proximal or distal focal stimulation when granule cells were held at hyperpolarized membrane potentials (Fig. 4-6A-B). APV blocked only a small fraction of -70 mv EPSC current integral in both responses to proximal (9.41 ± 2.2 % of control; n = 4) and distal (12.2 ± 3.5 % of control; n = 4) supraminimal stimulation, suggesting that the difference in kinetics between these EPSCs (see insert in Fig. 4-6B) was not due to a differential contribution of NMDAR activation at -70 mv. However, large APV-sensitive components were evident in both responses when the granule cells were held at depolarized membrane potentials (Fig. 4-6A-B). The EPSC response at +50 mv to focal proximal stimulation decayed rapidly and was nearly abolished at 500 ms (7.6 ± 2.0 % of peak current; n = 4). By contrast, the distal EPSC evoked at the same holding potential decayed more slowly, reaching approximately half its peak amplitude after 500 ms (42.6 ± 9.5 % of peak current; n = 4; significantly different from proximal EPSP; P < 0.05). The rectification ratios of the AMPAR-mediated EPSCs were similar in proximal (ratio of -70/+50 mv EPSC amplitude in APV = 3.85 ± 0.64; n = 4) and 116

127 distal (3.25 ± 0.36; n = 4) focal stimulation experiments. The different rates of decay of the EPSCs recorded at +50 mv appeared to reflect differences in NMDAR-mediated currents as both proximal and distal AMPAR-mediated responses recorded at +50 mv had returned to baseline within 100 ms. The relatively slow kinetics of these responses suggests that these differences may reflect other functional differences (e.g., different NMDA receptor subunit composition) rather than just differences in electrotonic filtering. Do granule cells have silent synapses? Not all glutamatergic synapses in the CNS contain both functional AMPA and NMDA receptors (Isaac et al., 1995; Liao et al., 1995; Isaac et al., 1997). This phenomenon, typically referred to as silent synapses, reflects different distributions of glutamate receptor subunits at dendritic sites in the vicinity of postsynaptic active zones (Malinow and Malenka, 2002; Isaac, 2003). Differences in glutamate receptor composition can have dramatic consequences on the nature of the postsynaptic response; AMPAR silent synapses, an extreme example, generate no postsynaptic response at hyperpolarized membrane potentials. Also, several forms of long-term plasticity appear to be mediated by movement of spare receptors into the postsynaptic zone following synaptic activity where they then can contribute to the postsynaptic response (Bredt and Nicoll, 2003). We used 2PGMS to investigate the receptor subunit composition in excitatory synapses on granule cells and to ask whether granule cells have AMPAR or NMDAR silent synapses. One approach to test for silent synapses, shown in Fig. 4-7A, is based on comparing minimal stimulation failure rates at -70 and +50 mv. The plots shown in Fig. 117

128 4-7A 1 and 4-7A 2 illustrate the most common results we observed: dual-component EPSCs interspersed with failures. Given the strong rectification of AMPAR-mediated synaptic currents (see Fig. 4-6A-B), we expected that most successes recorded at +50 mv reflected currents through NMDAR receptors. We confirmed this in two granule cell experiments in which we found that bath application of 50 µm D-APV abolished all successes recorded at +50 mv using 2PGMS. The same data sets shown as time plots in Fig. 4-7A are replotted in Fig. 4-7B as response amplitude histograms. In the two experiments shown in Fig. 4-7B 1 and 4-7B 2, the proportion of responses categorized as failures was approximately the same when the granule cell was held at -70 and at +50 mv, suggesting that these responses included both AMPAR and NMDAR-mediated components (i.e., dual-component). Approximately three quarters of our 2PGMS experiments on granule cells fit this pattern (69 % with proximal 2PGMS and 70 % with distal 2PGMS) and were classified as dual-component EPSCs. None of the granule cell experiments categorized as dual-component had failure rates greater than 90% at either - 70 or +50 mv holding potentials or statistically significant changes in the failure rate at the two potentials (Chi squared test; significance threshold of P < 0.01). A minority of granule cell synapses activated using 2PGMS appeared to be NMDAR silent. An example of this type of synapse is shown in Fig. 4-7A 3. In this experiment, proximal 2PGMS evoked clear successes at -70 mv but almost no successes when the granule cell was held at +50 mv. We confirmed that this cell was not damaged by the transient depolarization to +50 mv by verifying that successes still occurred when the cell was returned to -70 mv. Fig. 4-7B 3 shows the amplitude distributions calculated from this experiment and demonstrate clearly separable successes and failures at -70 mv 118

129 but not at +50 mv, suggesting that this proximal synapse contained AMPA receptors but no functional NMDA receptors. Approximately one quarter of the granule cell 2PGMS experiments resembled this example (30 % proximal 2PGMS and 25 % distal 2PGMS) and were categorized as NMDAR silent. All of the distal (n = 3) and one of the synaptic responses categorized as NMDAR silent were not completely silent at +50 mv but rather had intermediate failures rates at +50 mv which were significantly decreased at - 70 mv. The remainder of the proximal synapses categorized as NMDAR silent (n = 3) had failure rates greater than 90 % at +50 mv. We found a very small incidence of AMPAR silent synapses onto granule cells (0/10 for distal 2PGMS and 1/16 for proximal 2PGMS), a synaptic phenotype that is very common in the hippocampus (Isaac et al., 1995; Liao et al., 1995; Isaac, 2003). We replicated the results from these hippocampal studies by recording responses to minimal stimulation in CA1 pyramidal cells under the same conditions as in our olfactory bulb slice experiments. We found evidence for AMPAR silent synapses in approximately half (58 %; 11/19) of our hippocampal experiments. Fig. 4-7A 4 shows an example of an AMPAR silent hippocampal synapse with a very high failure rate at -70 mv but frequent successes at +50 mv. Only one 2PGMS olfactory bulb experiment (out of 26 experiments) resembled the pre-pairing conditions shown in Fig. 4-7B 4, suggesting that the incidence of AMPAR silent synapses is very low in granule cells. The results from all of the olfactory bulb and hippocampal silent synapse experiments are plotted in Fig. 4-7C-D. While there was a statistically significant change in the mean failure rate in our set of hippocampal experiments (from 68.8 ± 4.4 % at -70 mv to 44.8 ± 4.1 % at +50 mv; n = 19; P < 0.01; paired t test), there was no difference in 119

130 the mean population failure rate in either set of proximal or distal olfactory bulb 2PGMS experiments (P > 0.05). As discussed above, five proximal and three distal olfactory bulb experiments and 11/19 hippocampal experiments showed statistically significant changes in the failure rate when analyzed individually (Chi squared test; P < 0.01; thick lines in Fig. 4-7C-D). The overall proportion of AMPAR silent, NMDAR silent and dualcomponent EPSCs we recorded is shown in Fig. 4-7E. We performed several additional experiments to test for AMPAR silent synapses on granule cells. In 9 experiments, we initially searched for minimal responses in granule cells held at -70 mv and then gradually reduced the stimulus intensity until we recorded no successes (all failures). We tested for successes at +50 mv at this stimulus intensity but found no evidence for AMPAR silent responses. We also verified that pairing synaptic stimulation with intracellular depolarization (shown by horizontal bar in Fig. 4-7A 4 ) could convert an AMPAR silent hippocampal synapse into a dual-component synapse. Pairing successfully revealed AMPAR-mediated EPSCs in 3 of 5 hippocampal experiments. As shown in the hippocampal amplitude histograms in Fig. 4-7B 4, before pairing there were no responses at -70 mv with amplitudes greater than -3 pa while after pairing most of the response amplitudes were between -5 and -30 pa. A similar pairing protocol was not successful in revealing an AMPAR-mediated component in the one AMPAR silent granule cell synapse we found (data not shown). We also tested whether pairing protocols altered the -70/+50 mv failure ratio in 7 granule cell 2PGMS experiments. The same pairing protocol that was effective with hippocampal synapses failed to modulate the -70 / +50 mv failure rate ratio in all of these olfactory bulb 120

131 experiments, suggesting that long-term plasticity may be mediated by different cellular mechanisms in these two brain regions. What is the source of the proximal excitatory input to granule cells? Previous work suggested two potential sources of excitatory, glutamatergic inputs to the proximal dendrites of granule cells: local collaterals of mitral cell axons and centrifugal feedback projections from cortical regions. We used 2-photon imaging and a combined olfactory bulb/piriform cortex slice preparation to determine the relative importance of these two potential proximal inputs. In the first set of experiments we took advantage of the ability of 2-photon imaging to visualize entire dendritic arbors to classify granule cells into one of two categorizes: (1) granule cells with apical dendrites that entered and bifurcated in the EPL and (2) granule cells with apical dendrites that were truncated at the top or bottom surface of the slice before they entered the EPL. As shown in Fig. 4-8A, both types of granule cells received spontaneous EPSCs. However, the two types of granule cells differed dramatically in their range of spontaneous EPSC kinetics; all granule cells with truncated dendrites (n = 10) lacked spontaneous slow-rising EPSCs (10-90% rise times greater than 2.5 ms). An example of the sepsc rise time histogram calculated from a granule cell with a truncated apical dendrite is shown in Fig. 4-8B 3. The rise time distribution in this example, and in two other visualized and reconstructed granule cells with truncated dendrites, was unimodal and contained only fast-rising sepscs. The mean sepsc rise time in the example shown in Fig. 4-8B 3 was 1.61 ± 0.02 ms and closely matched both the first peak in the bimodal rise-time sepsc distribution of 121

132 granule cells with EPL dendrites (1.8 ms; Fig. 4-1D) and the mean rise-time of EPSCs evoked by proximal 2PGMS (1.4 ms; Fig. 4-3B). These results are consistent with the 2PGMS experiments presented above and strongly suggest that slow-rising EPSCs arise from distal dendrodendritic mitral/granule cell synapses located in the EPL. We also found that a subset of granule cells received intermittent barrages of spontaneous EPSCs. An example of granule cell with EPL dendrites that received synaptic barrages is shown in the middle set of traces in Fig. 4-8A. While we observed isolated barrages in a majority of granule cells with EPL dendrites, we selected a small subset of experiments with frequent barrages (> 0.1 Hz; 7 of 61 experiments) to analyze in detail. We found that all mitral cells tested under current clamp conditions (7/7) in these slices were bursting spontaneously (see top traces in Fig. 4-8A). While we did not examine the cellular mechanisms underlying this spontaneous activity in this study, it is likely that prolonged application of gabazine (present throughout these experiments) generates a hyperexcitable state that promotes mitral cell bursting. As shown in Fig. 4-8A, spontaneous bursts in mitral cells were irregularly spaced and were not associated with a large post-discharge hyperpolarization. We used the presence of spontaneous discharges in mitral cells to test whether mitral cell axons innervated nearby granule cells. We first analyzed the kinetics of sepscs within barrages recorded from granule cells with EPL dendrites. The vast majority of these sepscs were slow-rising (rise times > 2.5 ms; Fig. 4-8B 1 ). The mean rise time of sepscs within barrages was 3.93 ± 0.8 ms (n = 8 cells; Fig. 4-8C) and matched both the second peak in the sepsc rise time distribution (3.8 ms; Fig. 4-1D) and the rise times of EPSCs evoked by distal 2PGMS (4.2 ms; Fig. 4-3D). As shown in Fig. 122

133 4-8B 2, the same granule cell that received only slow-rising EPSCs during synaptic barrages received both fast- and slow-rising spontaneous EPSCs in the intervals between barrages. We also recorded from 10 granule cells with apical dendrites truncated before the EPL in slices with frequently bursting mitral cells. Granule cells with truncated dendrites in bursting slices had less synaptic noise than granule cells with EPL dendrites (Fig. 4-8D). None of the granule cells with truncated dendrites received synaptic barrages (0 of 10 cells; bottom traces in Fig. 4-8A) while most of the granule cells with EPL dendrites in bursting slices received synaptic barrages (81.3 %; 13 of 16 cells; Fig. 4-8E). We also found the frequency of synaptic barrages recorded in granule cells with EPL dendrites was significantly higher than the mean frequency of spontaneous bursting in mitral cells (Fig. 4-8F), suggesting that spontaneous discharges in mitral cells were not synchronous and that individual granule cells received dendrodendritic synaptic inputs from multiple spontaneously active mitral cells. The observations that spontaneous mitral cells discharges resulted in synaptic barrages only in granule cells with EPL dendrites and that these barrages were composed solely of slow-rising EPSCs strongly suggest that mitral cells innervate nearby granule cells through dendrodendritic synapses in the EPL and not through proximal axo-dendritic synapses in the GCL. Finally, we tested whether cortical axons activated granule cells through proximal synapses. We used horizontal brain slices that contained both the olfactory bulb and the anterior piriform cortex (Fig. 4-9A) for these experiments. We first asked whether these slices maintained any of the feedback projections from piriform cortex that normally innervate the granule cell layer in the olfactory bulb (de Olmos et al., 1978; Haberly and Price, 1978; Shipley and Adamek, 1984). We tested this by making focal DiI injections 123

134 in anterior piriform cortex in fixed slices (3 mm; 500 ms pressure pulse duration; 2 psi; n = 8 slices). After waiting 14 days for DiI to diffuse throughout the axonal arborizations, we were able to visualize abundant labeled axons, many with en passant terminals, in the granule cell layer in most DiI injected slices (7 of 8; see insert in Fig. 4-9A). After verifying that the cortical feedback pathway was at least partially preserved in these horizontal slices, we tested whether focal stimulation in layers 2-3 of anterior piriform cortex activated EPSCs on granule cells. We recorded EPCSs in granule cells in approximately one third of the combined OB/APC slices tested. Fig. 4-9B illustrates EPSCs evoked by minimal stimulation in the layers 2-3 of anterior piriform cortex. Cortical stimulation evoked purely fast-rising EPSCs in granule cells (mean rise time = 1.10 ± 0.03 ms; n = 6 cells) that resembled the EPSCs evoked by proximal 2PGMS. Cortical EPSCs also facilitated with paired-pulse stimulation (mean PPR = 1.62 ± 0.21; Fig. 4-9C). By contrast, minimal stimulation in the superficial layer of APC, activating axons in the lateral olfactory tract, evoked purely slow-rising EPSCs (mean rise time = 3.73 ± 0.15 ms; n = 5; Fig. 4-9B) that depressed with paired-pulse stimulation (mean PPR = 0.55 ± 0.14) and resembled distal 2PGMS responses. Both the changes in EPSC rise time (Fig. 4-9D) and paired-pulse ratio (Fig. 4-9E) were significantly different between the APC and LOT stimulation sites. The distal-like minimal responses evoked by LOT stimulation in the combined APC/OB slices did not have the large latency jitter observed in responses to 2PGMS in the EPL (see Fig. 4-2D 1 ). This difference may reflect differences in the reliability of spike initiation by weak focal axonal (in the LOT) versus dendritic (in EPL) stimulation. Together, these results suggest that most proximal excitatory inputs to granule cells arise from feedback cortical projections. 124

135 Discussion In this study, we investigated the functional properties of excitatory glutamatergic inputs in the olfactory bulb using a novel technique, 2-photon guided minimal stimulation, to evoke transmitter release onto small groups of dendritic spines at defined positions along granule cell dendritic trees. We made three principal conclusions in this study. First, distal dendrodendritic and proximal axo-dendritic excitatory inputs form functionally distinct synapses. Dendrodendritic inputs from mitral cells have slow kinetics and show paired-pulse depression while proximal axonal inputs have fast kinetics and facilitate. Second, unlike other brain regions such as the hippocampus (Isaac et al., 1995; Liao et al., 1995; Durand et al., 1996) and neocortex (Isaac et al., 1997; Feldman et al., 1999; Rumpel et al., 2005), both distal and proximal granule cell synapses generally are not AMPAR silent but surprisingly can be NMDAR silent, in addition to the more commonly observed dual-component phenotype. This inversion of the general pattern of ionotropic glutamate receptor expression found in other brain regions places unique constraints on possible mechanisms of activity-dependent synaptic modifications that might mediate experience-dependent plasticity in the olfactory bulb. Finally, we find that centrifugal inputs originating in piriform cortex generate facilitating proximal axo-dendritic synapses onto granule cells. Mitral cells appear to contact nearby granule cells predominately through distal dendrodendritic synapses. These results suggest that the piriform cortex may play a crucial role in controlling granule cell activity and gating lateral inhibition in the olfactory bulb. 125

136 Most previous studies using minimal stimulation to define the functional properties of specific types of synapses took advantage of anatomically-defined fiber tracts, such as parallel fiber inputs onto cerebellar Purkinje cells or the Schafer collateral inputs onto hippocampal CA1 pyramidal neurons, to trigger transmitter release from a homogeneous population of synapses. Even in these cases, presynaptic processes are activated at relatively large distances away (often hundreds of microns) from the actual synaptic terminal, making identification of the activated synapses along the dendritic tree difficult. Because the olfactory bulb lacks anatomically-defined, homogenous fiber pathways, it is difficult to selectively activate different types of axon terminals in this brain region using conventional extracellular stimulation methods. By employing 2- photon imaging to position an extracellular simulating electrode near a specific dendritic segment, we were able to activate relatively homogenous populations of presynaptic processes, judging from the functional properties of the resulting postsynaptic responses. Previous studies have used 2-photon imaging to position stimulating electrodes (Skeberdis et al., 2006) and to record quantal-like postsynaptic Ca transients (Oertner et al., 2002) in response to relatively large (supraminimal) stimuli. In our study, we combined 2-photon imaging with minimal stimulation methods to define, for the first time, the differences at the single-synapse level (e.g., failure rates) between proximal and distal inputs to granule cells. In future studies, the usefulness of this method might be enhanced by employing brain slices with genetically encoded or extracellularly injected fluorescent labels that mark specific fiber tracts. This method may enable 2PGMS to define the functional properties of more closely spaced synaptic inputs onto the same postsynaptic cell. 126

137 Multiple synaptic mechanisms for activating GABAergic granule cells We find that the large degree of variability among both spontaneous and evoked EPSCs in granule cells is not solely due to different degrees of electrotonic filtering, but rather reflects multiple types of functionally distinct glutamatergic inputs. We base this conclusion on the large differences in EPSC kinetics, especially the differences apparent in the decay of the NMDAR-mediated component at depolarized membrane potentials, the different forms of short-term plasticity, and the different degrees of steady-state depression we recorded following proximal versus distal focal stimuli. While differences in EPSC kinetics may be partially explained by electrotonic filtering (Jack et al., 1983; Spruston et al., 1994), electrotonic effects cannot explain differences in short-term plasticity and steady-state depression. These results strongly suggest that granule cells receive multiple types of excitatory inputs that have different functional properties. The differences in the kinetics of the NMDAR-mediated EPSCs recorded at +50 mv may also reflect functional differences, rather than simply electrotonic filtering. The slow time course of these responses are unlikely to be strongly influenced by electrontonic filtering, compared with fast AMPAR-mediated EPSCs (Jack et al., 1983). Instead, this difference may reflect different NMDAR subunits in distal and proximal synapses on granule cells (Petralia et al., 1994b; Petralia et al., 1994a). Our results showing EPSCs with different kinetics evoked at proximal and distal synapses are consistent with a previous report of spontaneous fast- and slow-rising EPSCs in granule cells (Carleton et al., 2003). One previous study (Dietz and Murthy, 127

138 2005) investigated short-term plasticity at glutamatergic synapses in the olfactory bulb and found evidence for two classes of granule cells: one that received excitatory inputs that showed paired-pulse depression and another type that received facilitating inputs. While we also found evidence for two classes of excitatory synapses, we found no evidence for separate subpopulations of granule cells that receive facilitating and depressing inputs. One explanation for this discrepancy is the different stimulating methods employed in these studies. The 2PGMS method we used enabled us to selectively activate presynaptic processes relatively close to synaptic terminals on specific postsynaptic neurons. By contrast, extracellular stimulation at sites relatively distant from the postsynaptic dendrite might activate a combination of distal and proximal inputs, depending on the location of the stimulating electrode and the stimulus intensity. Responses to extracellular stimulation in the granule cell layer are especially difficult to interpret since it is relatively easy to antidromically activate mitral cell axons in this layer and thereby trigger release at dendrodendritic synapses in the EPL. Positioning the stimulating pipette close to the relevant dendritic segment should introduce a bias toward activating either proximal or distal synapses, depending on the location of the stimulating electrode. The relatively homogenous responses we observed after positioning fine-diameter stimulating pipettes very near (typically µm) specific dendritic segments argues strongly that different functional properties we observe (short-term plasticity, steady-state depression) are correlated with distal dendrodendritic and proximal axo-dendritic synapses and not with functionally defined subpopulations of granule cells. 128

139 Source of the proximal axo-dendritic input onto granule cells Using combined olfactory bulb-piriform cortex slices, we found that proximal facilitating inputs onto granule cells arise primarily from cortical feedback inputs and not from mitral cell axon collaterals. We showed that anterior piriform cortical neurons send numerous projections that ramify in the granule cell layers, in agreement with previous studies on the distribution of inputs from piriform cortex in the olfactory bulb (de Olmos et al., 1978; Haberly and Price, 1978; Shipley and Adamek, 1984) and classic work demonstrating that stimulation of deep piriform cortical layers excites olfactory bulb granule cells (Nakashima et al., 1978). We found that stimulating the anterior piriform cortex evoked fast facilitating EPSCs in granule cells that were indistinguishable from proximal synaptic inputs activated by 2PGMS. Antidromically activating mitral cells by LOT stimulation evoked slowly-rising EPSCs that depressed and were similar to EPSCs evoked by distal 2PGMS. Previous studies suggest that feedback inputs from the piriform cortex are extensive and may exceed the density of local excitatory input onto granule cells (Haberly, 2001). The same extensive feedback projection from anterior piriform cortex to granule cells may also mediate the large, long-latency negative field potential recorded in the isolated whole-brain preparation (Uva et al., 2006) that was blocked by perfusion with AMPAR antagonists. Transecting the LOT in this preparation abolished the field potentials associated by dendrodendritic inhibition while sparing the late field response. Current source density analysis demonstrated the late APC-evoked 129

140 potential was associated with a large current sink in the granule cell layer (Uva et al., 2006). Our results using spontaneously bursting olfactory bulb slices suggest that mitral cell axon collaterals do not constitute a large fraction of the proximal input to granule cells. While synaptic contacts from local mitral cell axon collaterals onto granule cells are often included in schematic diagrams of the olfactory bulb, there is little direct evidence for these connections. Price and Powell (1970b) found a subpopulation of asymmetric axo-dendritic synaptic contacts onto granule cells that persisted following large lesions of the ipsilateral anterior olfactory nucleus, a lesion that should trigger degeneration of most types of extrinsic input to the olfactory bulb. Subsequent work (Orona et al., 1984) suggested that local axon collaterals in the GCL may occur only in a subpopulation of mitral cells. Neither study directly demonstrated mitral-to-granule cell axo-dendritic connections. Our results found no evidence for axo-dendritic connections onto nearby granule cells (located in the same olfactory bulb slice). Since our experiments employed acute brain slices, it is impossible to tell from this study whether mitral cell axon collaterals contact distant granule cells. Interestingly, dendrodendritic mitral/granule cell synapses appeared to be preserved in our slice preparation based on the high proportion of granule cells with EPL dendrites that followed spontaneous mitral cell discharges. This result suggests that even if a subpopulation of mitral cells innervates granule cells through proximal axo-dendritic synapses, the incidence of granule cells that receive both distal dendrodendritic and proximal axo-dendritic synapses from the same mitral cell is probably very low. The low apparent incidence of mitral cell axon collaterals contacting nearby granule cells also raises the intriguing possibility that 130

141 other types of interneurons in the granule cell layer, such as Blanes cells (Pressler and Strowbridge, 2006), may be the principal target of these connections. Implications for the long-term plasticity at dendrodendritic synapses The high concentration of NMDA receptors at granule cell spines (Sassoe- Pognetto and Ottersen, 2000) suggests that long-term potentiation of dendrodendritic transmission may be an important mechanism for synaptic plasticity in the olfactory bulb. In many brain regions, including the hippocampus (Bliss and Collingridge, 1993), cerebral cortex (Katz and Shatz, 1996; Feldman et al., 1999; Feldman, 2000), amygdala (Rodrigues et al., 2004; Rumpel et al., 2005), and thalamus (Mooney et al., 1993), NMDA receptors function as coincidence detectors whose activation induces long-term enhancement of synaptic strength. Long-term potentiation at central synapses often occurs through the insertion of AMPA receptors following NMDA receptor activation into postsynaptic spines that originally contain only NMDA receptors (Isaac, 2003). These silent synapses are normally non-functional since NMDA receptors are largely blocked at resting membrane potential; AMPA receptor insertion converts silent synapses to functional ones and increases the efficacy of presynaptic transmitter release on postsynaptic firing. Silent synapses play a role in both adult synaptic plasticity (Isaac et al., 1995; Liao et al., 1995; Isaac et al., 1996b) and the normal activity dependent maturation of central synapses during development (Durand et al., 1996; Isaac et al., 1997). 131

142 Despite considerable effort, the cellular basis for long-term synaptic plasticity in the olfactory bulb is unknown. Ultrastructural studies using immunogold electron microscopy indicated that dendrodendritic synapses in the EPL often contain both AMPAR and NMDAR subunits (Sassoe-Pognetto and Ottersen, 2000). Our study found little physiological evidence for AMPAR silent synapses in granule cells. Our results suggest that activity-dependent changes in synaptic efficacy at dendrodendritic synapses may not occur through classical NMDAR-dependent long-term potentiation mediated by the insertion of AMPA receptors into the postsynaptic membranes. Instead, plasticity at dendrodendritic synapses may occur through presynaptic changes in release probability or through intrinsic changes in granule cell or mitral cell excitability. The presence of NMDAR silent synapses onto granule cells raising the possibility that plasticity in the olfactory bulb also may be mediated by activity-dependent insertion of NMDA receptors in dendrodendritic synapses. Granule cells continue to be produced through adulthood in the subventricular zone and migrate into the olfactory bulb to be incorporated into existing olfactory bulb circuits (Lois and Alvarez-Buylla, 1994; Alvarez-Buylla and Garcia-Verdugo, 2002; Petreanu and Alvarez-Buylla, 2002; Carleton et al., 2003; Lledo et al., 2006). These adult-born granule cells initially express predominately AMPA receptors and later incorporate NMDA receptors (Carleton et al., 2003; Lledo et al., 2006), implying that even in adulthood, a subpopulation of granule cells may have NMDAR silent synapses. However these studies did not examine glutamate receptor distribution at the single synapse level or test for AMPAR and NMDAR silent synapses physiologically. Based on our results alone, it is not possible to know whether the NMDAR silent synapses we 132

143 found reflect a subpopulation of immature granule cells or whether NMDAR silent and dual-component synapses co-exist on fully mature granule cells. To answer this question, future studies will need to combine synapse-specific stimulation methods, such as 2PGMS, with retroviral methods that selectively mark immature granule cells. Functional significance of multiple excitatory inputs onto GABAergic granule cells Reciprocal dendrodendritic synapses between mitral and granule cells provide the dominant source of both recurrent and lateral inhibition onto mitral cells (Rall et al., 1966; Shepherd and Greer, 1998). However most physiological studies of these synaptic microcircuits found the inhibitory output from granule cells was tonically attenuated by extracellular Mg ions that prevented permeation through critical NMDA receptors (Chen et al., 2000; Isaacson and Strowbridge, 1998; Schoppa et al., 1998). Our results suggest that this requirement for NMDAR activation is not due to the absence of AMPARs in dendrodendritic synapses but rather may reflect aspects of NMDAR-mediated responses that facilitate GABA release from granule cells (e.g., presynaptic Ca entry through NMDARs (Halabisky et al., 2000) or long-duration NMDAR EPSCs that outlast transient K currents (Schoppa and Westbrook, 1999)). Several lines of evidence suggest that proximal excitatory inputs to granule cells can reverse the Mg blockade of NMDA receptors at distal dendrodendritic synapses and can gate dendrodendritic inhibition. Pairing Ca transients in mitral cells (from photolyzing caged Ca) with gamma-frequency stimulation of proximal inputs triggered a prolonged barrage of IPSPs not observed with either Ca uncaging or GCL stimulation 133

144 alone (Chen et al., 2000). Halabisky and Strowbridge (2003) demonstrated that pairing mitral cell action potentials with 50 Hz GCL stimulation triggered feedback inhibition that was blocked by the selective GABA A receptor antagonist picrotoxin. The present results suggest that pyramidal neurons in anterior piriform cortex may provide this proximal gating input. The strong facilitation we found in these proximal synapses, which was maximal with inter-stimulus intervals between ms, also provides an explanation of why relatively high frequency GCL tetani, and not single shocks or low frequency trains, are required to gate dendrodendritic inhibition (Halabisky and Strowbridge, 2003). Our results suggest that Hz (beta to gamma band) oscillations that normally occurs in populations of anterior piriform cortical cells (Freeman, 1978) may reflect an endogenous gating signal that enables recurrent and lateral dendrodendritic microcircuits in the olfactory bulb. Additional studies will be necessary to determine the relationship between beta/gamma band oscillations in the subpopulation of piriform cortical neurons that project back to the olfactory bulb and GABA release at dendrodendritic synapses, and also the cellular mechanism by which the proximal input from APC regulates NMDAR activation at distal synapses. Finally, our results suggest that the common conceptual model of dendrodendritic inhibition mediated by local circuits in the olfactory bulb may not be appropriate. Instead, our work suggests that the olfactory bulb and anterior piriform cortex function as a tightly integrated system with piriform cortex providing a critical feedback excitatory input to granule cells that governs their behavior and output. The relative timing between dendodendritic excitation (reflecting backpropagating action potentials in mitral cell secondary dendrites) and high frequency discharges of piriform cortical neurons may 134

145 regulate much of the GABA-mediated inhibition in the olfactory bulb. One prediction from this model is that the degree of lateral inhibition in the olfactory bulb following sensory stimulation may be dynamically modulated by activity in anterior piriform cortex. In one extreme, very little local processing may occur in the olfactory bulb when piriform cortex activity is depressed. This model may explain recent results showing very little difference between mitral cell output patterns (assessed indirectly through glomerular surface intrinsic signal imaging) when groups of specific odorants were tested separately and the resulting activity maps merged or when the odorants were applied as a mixture (Lin et al., 2006). The similarity between the mitral cell activity patterns in these two experiments suggested that very little lateral inhibitory processing occurred when the mixture was presented. However, the isofluorane anesthesia used in these experiments often depresses cortical activity levels (Orth et al., 2006) and thus may have diminished the cortical feedback projection to granule cells that enables lateral bulbar inhibition. Another prediction of our model is that interventions that boost overall activity levels in piriform cortex (e.g., focal stimulation or pharmacological disinhibition) should potentiate lateral dendrodendritic inhibition in the olfactory bulb and increase the dissimilarity between activity patterns of odorant combinations presented separately and as mixtures. 135

146 Figure 4-1. Two classes of spontaneous excitatory postsynaptic currents in granule cells (A) Schematic diagram of excitatory synapses onto granule cells. Granule cells receive excitatory input from mitral cells through reciprocal dendrodendritic synapses (EPSP 1 ) on distal dendrites in the EPL. Granule cells also receive presumptive excitatory input onto spines on proximal dendrites, possibly from both mitral cell axon collaterals (EPSP 2 ) and from centrifugal inputs from cortical areas (EPSP 3 ). (B) 2-Photon reconstruction of an Alexa594-filled granule cell. A second patch pipette containing Alexa594 was used for extracellular stimulation and is visible near the apical dendrite. (Stimulating electrode tip indicated by tan asterisk.) Inset shows magnified views of granule cell dendrites with spines located on both distal bifurcated dendrites in the EPL (1) and the proximal primary dendrite (2). Scale bars are 10 µm for full reconstruction and 5 µm for the insets. (C) Spontaneous synaptic responses recorded from a granule cell at -70 mv in gabazine (10 µm) to block GABA A -receptor mediated IPSCs. Granule cells receive both fast (black dots; rise time < 2.5 ms) and slow (red dots; rise time > 2.5 ms) spontaneous EPSCs. Overlaid fast (black; <2.5 ms) and slow (red; > 2.5 ms rise time) synaptic events aligned by their rising edge shown above sweeps. (D) Rise time distribution of 287 spontaneous EPSCs recorded from the granule cell in C. The rise time distribution is bimodal and is well fit by the sum of two Gaussian distributions (blue curve) centered at 1.5 ms (black curve) and 4.2 ms (red curve). 136

147 Figure

148 Figure 4-2. Two types of granule cell EPSCs evoked by 2-photon guided minimal stimulation. (A) Schematic diagram of experiment in B-C. (B) Plot of OGB-1 Ca transients recorded over 6 trials in the 3 spines and a dendrite shaft segment shown in C. Focal supraminimal stimuli (single 200 µs shock, 43 µa; Alexa594-filled stimulating electrode positioned ~20 µm from imaged dendritic region) reliably triggered Ca accumulations in 2 of the 3 imaged spines but not in the dendritic shaft segment. (Statistically significant increases over baseline indicated by **; p < 0.01.) Example ΔF/F traces from one trial shown in inset. (C) 2-Photon images of baseline OGB-1 fluorescence (left) and ΔF image frames before and immediately after a single focal stimulus. Stimulus-evoked Ca accumulations were restricted to a subset of imaged dendritic spines. Labeled regions of interest correspond to image areas analyzed in B. Acquired images were 392 by 64 pixels; 25 ms/frame; 3200 raster lines/sec. Calibration bar is 3 µm. (D) Plots of amplitudes of unitary EPSCs evoked by 2-photon guided minimal stimulation of distal (D 1 ) and proximal (D 2 ) granule cell dendrites versus stimulus intensity. Insets show 2-photon images of the relationship between the distal stimulating electrode in the EPL (D 1 ) and the proximal stimulating in the GCL (D 2 ) and the recorded granule cells. Both recording and stimulating pipettes were filled with Alexa594. (Different granule cells shown in D 1 and D 2.) Stimulating electrode tip indicated by tan asterisk. Calibration bar in D 1 is 10 µm and 7 µm in D 2. Distal minimal stimulation evoked all-or-none EPSCs with slow rising phases and variable onset latencies (red arrowheads) while proximal minimal stimulation evoked fast-rising EPSCs at constant latencies. Both distal and proximal stimulation responses show sharp activation thresholds with distinguishable successes 138

149 (filled circles) and failures (open circles). Example threshold responses are shown above each plot. 139

150 Figure

151 Figure 4-3. Kinetic differences between distal and proximal minimal EPSCs. (A) Amplitude (middle) and rise time (right) EPSC distributions calculated from distal (A 1, gold shading) and proximal (A 2, purple shading) 2-photon guided minimal stimulation. Data from two different granule cells held at -70 mv. Distribution of failures in the amplitude plots closely matches the noise amplitude distribution calculated from the same cells (left). Note the different EPSC rise time distributions between responses evoked from the two stimulus positions. (B) Summary plot of the rise time distribution from all distal (n = 10 cells; gold shading) and proximal (n = 10 cells; purple shading) stimulation experiments. Both distributions were well fit by Gaussian distributions (smooth curves). Inset shows statistically significant difference in mean EPSC rise time for the two stimulation sites (** p < 0.01). 141

152 Figure

153 Figure 4-4. Distal and proximal excitatory synapses have different forms of short-term plasticity. (A) Granule cell responses to 2-photon guided paired-pulse stimulation (50 ms ISI) of either distal (A 1 ) or proximal (A 2 ) dendrites. Minimal responses to distal stimulation (top traces) show paired-pulse depression while analogous responses to proximal stimulation show paired-pulse facilitation. Increasing the stimulus intensity to recruit additional axons (bottom traces) did not change the type of paired-pulse modulation at either synapse. Both distal and proximal EPSCs were blocked by the non-nmda glutamate receptor antagonist NBQX (10 µm; grey traces). Images above traces show relationship between Alexa594-filled stimulating pipette and the recorded neuron. Calibration bars are 10 µm in A 1 and 5 µm in A 2. Stimulating electrode tip indicated by tan asterisk. (B) Summary graphs showing paired-pulse response ratios (PPR) calculated by mean EPSC amplitudes for 12 minimal stimulation experiments (left) and 17 supraminimal stimulation experiments (right). Paired-pulse ratios were significantly different between proximal and distal stimulus sites at both stimulus intensity ranges (** P < 0.01.) Pairedpulse ratios for individual experiments shown by open circles. Red filled circles in the supraminimal graph represent results from a single granule cell that was stimulated at both proximal and distal dendritic sites. (C) Paired-pulse depression at distal synapses was associated with a statistically significant increase in failure rate (left) at minimal stimulus intensities while paired-pulse facilitation at proximal synapses resulted in a decrease in failure rate (right; * P < 0.05.) (D) Plot of PPR calculated by mean EPSC amplitude (X-axis) versus the PPR calculated by failure rate (Y-axis). Both proximal 143

154 (purple dots) and distal minimal stimulation results (gold dots) fall near the dashed line representing equal PPR ratios. 144

155 Figure

156 Figure 4-5. Frequency-dependent modulation at distal and proximal excitatory synapses. Granule cell responses to a 50 Hz supraminimal distal (top) and proximal (bottom) stimulus train. Note the rapid silencing of the distal response by the fourth stimulus. Proximal synapses initially facilitate then show steady-state depression. (B) Summary of 8 experiments using 2-photon guided 50 Hz stimulus trains (4 proximal and 4 distal). * P < Inset shows amplitude of responses 8-10 normalized to the initial response for the proximal (purple) and distal (gold) stimulus sites. ** P < Inset shows mean steady state depression in last three responses of train. ** P < (C) Plot of the paired-pulse ratio, calculated from the mean supraminimal response amplitude, versus interstimulus interval for 4 proximal 2-photon guided stimulation experiments (purple dots) and 4 distal (gold dots) experiments. * P <

157 Figure

158 Figure 4-6. Both proximal and distal excitatory synaptic inputs activate NMDA and non- NMDA glutamate receptor subtypes. (A) Supraminimal responses to 2-photon guided distal stimulation at -70 and +50 mv. Blockade of NMDA receptors with D-APV (50 µm) attenuated the response at +50 mv but did not affect the response at -70 mv. (B) Similar experiment as A using proximal stimulation. Images in A-B show relationship between the Alexa594-filled stimulation pipette and the recorded neuron. Scale bars are 10 µm. Stimulating electrode tip indicated by tan asterisk. Inset shows distal and proximal normalized -70 mv responses in D-APV. (C) Summary plot of the APV-resistant current integral (% of control; integral over 700 ms) for the different stimulus conditions. (D) Plot of the component of the EPSC remaining 500 ms after onset (% of peak +50 mv control response) for proximal and distal supraminimal stimulation. * P <

159 Figure

160 Figure 4-7. Tests for AMPAR and NMDAR silent excitatory synapses on granule cells. (A) Plots of response amplitude to 2-photon guided minimal stimulation at 0.2 Hz of a distal EPL synapse (A 1 ), proximal GCL synapse (A 2-3 ) and st. radiatum stimulation of a hippocampal CA1 pyramidal cell (A 4 ). Membrane potentials are indicated above each plot. Plots A 1 and A 2 illustrate examples of dual (NMDA and non-nmda receptor) component minimal stimulation responses. Plot A 3 illustrates an example of NMDA receptor silent granule cell response. Plot A 4 illustrates an example of an AMPA receptor silent response in a CA1 pyramidal cell that was converted into a dual component response by pairing 50 minimal stimuli with intracellular depolarization to 0 mv. (B) Amplitude histograms generated from the responses plotted in A. Separate histograms are shown for responses at -70 and +50 mv; B 4 shows the change in the -70 mv response amplitude distribution before (open bars) and after (shaded bars) pairing. Vertical dashed line represents 2 x S.D. of the noise distribution. Example traces are shown above each plot. (C-D) Summary plots of the results from experiments using proximal (C, right; n = 16 cells) and distal (C, left; n = 10 cells) 2-photon guided minimal stimulation of granule cells and st. radiatum stimulation of CA1 pyramidal cells (D; n = 19 cells). Each summary plot shows the response failure rate at -70 and +50 mv (inside dots connected by lines) and the overall failure rate for the each group of experiments at -70 and +50 mv (outside dots with error bars). Only the group of experiments using CA1 pyramidal cells showed a statistically significant difference in mean failure rate at -70 and + 50 mv (** p < 0.01; paired t-test). Analysis of the failure rate in each individual experiment showed statistical significant differences between -70 and +50 mv in 5/16 proximal and 3/10 distal granule cell minimal stimulation experiments and 11/19 CA1 150

161 pyramidal minimal stimulation experiments (thick lines; Chi squared test; P < 0.01; three experiments with 100% failures at -70 mv included in significant difference category). (D) Summary plot showing the proportion of experiments classified as NMDAR silent (blue; statistically higher failure rate at +50 than -70 mv; P < 0.01), AMPAR silent (grey; statistically greater failure rate at -70 than +50 mv; P < 0.01) and dual component (hashed; no statistically significant difference in failure rates at -70 and +50 mv; P > 0.01) for the three stimulation sites presented in A-C. 151

162 Figure

163 Figure 4-8. Mitral cells contact nearby granule cells primarily through dendrodendritic synapses in the external plexiform layer. (A) Recordings from three different cells in olfactory bulb slices that exhibited frequent spontaneous bursting in gabazine. Top set of traces are current-clamp recordings from a mitral cell that generated spontaneous action potential discharges. The bottom two sets of voltage-clamp traces are from a granule cell with dendrites in the EPL (middle traces) and a granule cell with dendrites that were truncated in the GCL (bottom traces). Granule cells with dendrites in the EPL often generated spontaneous barrages of EPSCs (red asterisks) in bursting slices. Granule cells with dendrites that extended within the GCL but not the EPL had isolated spontaneous EPSCs and did not receive periodic barrages of synaptic input in bursting slices. Numbered rectangular boxes correspond to the sweep segments enlarged in B. (B) Spontaneous rise time distributions from EPSCs within synaptic barrages in the granule cell with EPL dendrites (B 1 ), isolated, inter-barrage EPSCs from the same granule cell (B 2 ) and EPSCs in the granule cell with a truncated dendrite (B 3 ). Expanded sweeps shown above each distribution from the regions indicated in A. Black dots represent spontaneous EPSCs with rise times < 2.5 ms; red dots > 2.5 ms. Note the absence of EPSCs with fast rise times within synaptic barrages in the granule cell with EPL dendrites and the absence of slow spontaneous EPSCs in the granule with truncated dendrites. (C) Plots of the mean rise time of EPSCs within synaptic barrages in 8 granule cells with EPL dendrites (white bar) and mean rise time of spontaneous EPSCs in 5 granule cells with truncated dendrites (hashed bar). Mean evoked distal (gold shading) and proximal (purple shading) EPSC rise time data from Fig. 4-3B are replotted for comparison. Spontaneous EPSCs within barrages have slow 153

164 rise times that are similar to distal evoked responses and are significantly different from proximal evoked responses. Spontaneous EPSCs in granule cells with truncated dendrites resembled proximal evoked responses and had significantly faster rise times that distal evoked response. ** P < 0.01; one-way ANOVA. (D) Plot of mean current variance in 16 granule cells with dendrites in the EPL and 10 granule cells with dendrites truncated in the GCL. ** P < (E) Plot of the proportion of cells that showed spontaneous bursts/barrages in bursting slices. All mitral cells tested (7/7) generated action potential bursts in bursting slices. Most (13/16) granule cells with EPL dendrites had spontaneous barrages of EPSCs while no (0/10) granule cells with dendrites truncated in the GCL had spontaneous barrages in bursting slices. (F) Frequency of spontaneous barrages/bursts was significantly greater in granule cells with EPL dendrites than mitral cells. ** P < 0.01; * P <

165 Figure

166 Figure 4-9 Cortical feedback projections generate facilitating, fast-rising EPSCs in granule cells. (A) Diagram of the combined olfactory bulb-anterior piriform cortex slice preparation. Mitral cell axon collaterals and cortical feedback projections were independently activated by stimulating the lateral olfactory tract (LOT) or the deep pyramidal cell layer of the anterior piriform cortex (APC), respectively. Focal DiI injections (3 mm; 2 psi for 500 ms) in the APC confirmed that the combined OB-APC brain slice contained cortical axons that innervated the olfactory bulb. Inset shows DiI-labeled axons with en passant boutons in the GCL. (B) Granule cell responses to minimal LOT stimulation in the APC (left) and focal APC stimulation (right). Antidromic mitral cell activation (LOT stimulation) evoked slow-rising EPSCs that resembled the distal responses shown in Fig. 4-2D 1 while APC stimulation evoked fast-rising EPSCs that resembled proximal responses shown in Fig. 4-2D 2. (C) Both minimal and supraminimal LOT stimulation evoked EPSCs that depressed (left) while EPSCs evoked by APC stimulation facilitated (right). (D) Summary plot of mean EPSC rise time for minimal LOT (n = 5) and APC (n = 6) stimulation experiments. Corresponding results from 2-photon guided focal stimulation of distal (gold shading) and proximal (purple shading) also are replotted from Fig. 4-3B. APC-evoked EPSCs had significantly faster rise times than both LOT- and distal OB-evoked minimal EPSCs (** P < 0.01; one way ANOVA.) (E) Summary graph of paired-pulse ratio, calculated from mean EPSC amplitude, for 5 LOT and 6 APC stimulation experiments. Corresponding results from distal and proximal OB stimulation are replotted from Fig. 4-4B for comparison. The PPR of APC-evoked EPSCs was 156

167 significantly greater than either LOT- or distal OB-evoked EPSCs. ** P < 0.01; one way ANOVA. 157

168 Figure