| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Laboratory of Perception and Memory, Pasteur Institute, Paris; and Laboratoire d’Ethologie Experimentale et Comparée, Université Paris 13, Villetaneuse, France
ABSTRACT I. INTRODUCTION II. EVOLUTIONARY ASPECTS OF OLFACTION III. FROM ODORANT MOLECULES TO CORTICAL CENTERS A. Olfactory Sensory Neurons and Signal Transduction B. Synaptic Transmission at the First Processing Stage 1. Characteristics of sensory inputs onto bulbar neurons 2. Synaptic processing within olfactory bulb microcircuits C. Decoding Information in Higher Olfactory Centers IV. EARLY COMPUTATIONS AT THE FIRST INFORMATION PROCESSING STAGE A. Odor Maps B. Temporal Coding 1. Network dynamics in the olfactory system 2. Temporal features of olfactory responses 3. Odorant-induced slow oscillations 4. Odorant-induced gamma oscillations 5. Generating odorant-induced gamma oscillations through intrinsic membrane properties 6. Functional implications of synchronization in highly distributed systems C. Roles of Local Inhibition in Active and Dynamical Networks V. NEURONAL REPLACEMENT IN THE ADULT OLFACTORY SYSTEM A. Neurogenesis in the Adult Olfactory System 1. The subventricular zone of rodents and nonhuman primates 2. The human case 3. Guiding newborn neurons in the adult brain A) TANGENTIAL MIGRATION. B) RADIAL MIGRATION. B. Maturation and Functional Integration of Newborn Neurons 1. Maturation of newborn interneurons 2. Newly generated interneuron survival and death 3. Physiological properties of newborn interneurons C. Functional Properties Brought by Adult-Generated Interneurons 1. The olfactory bulb as a locus of memory 2. Newly generated neurons and behavior D. Adult Neurogenesis as an Adaptive Function VI. FROM THE EXTERNAL WORLD OF ODORS TO THE INTERNAL STATE OF AFFECTS VII. CONCLUSIONS ACKNOWLEDGMENTS REFERENCES
 |
ABSTRACT |
|---|
 Top NextReferences |
|---|
|
I. INTRODUCTION |
|---|
|
TopPreviousNextReferences |
|---|
In mammals, the second sensory organ is represented by the olfactory epithelium, which recognizes more than a thousand airborne volatile molecules called odorant compounds (or odorants) (Fig. 1). This neuroepithelium is connected to the next central station for processing olfactory information: the main olfactory bulb (referred to below as the olfactory bulb). While advances in understanding olfactory transduction were taking place, interest in the olfactory bulb was also intensified. This growing interest has been spurred on by discovering the way the sensory organ connects to the olfactory bulb. Finally, whereas the power of olfactory stimuli in memory and the control of animal behavior have long been recognized, the neural mechanisms underlying these processes have only recently received new interest. Here, we summarize the richness and the kinds of interactions that take place in special areas throughout the olfactory system, emphasizing the more recent results from mice, rats, and humans.
|
|
II. EVOLUTIONARY ASPECTS OF OLFACTION |
|---|
|
TopPreviousNextReferences |
|---|
Indeed, chemosensation is a fundamental process shared by most organisms, as it is responsible for recognizing external chemical signals that influence behavior. The origin of molecule detection dates back to prokaryotes and has evolved into four distinct modalities in most vertebrates. The main olfactory system, the accessory olfactory system, the gustatory system, and the so-called common chemical sense mostly carried by trigeminal sensory neurons, all differ with respect to receptor molecules, receptor cells, and wiring of the receptor cells with the central nervous system. The main olfactory system detects only volatile odorants, whereas the accessory system picks up less volatile or even water-soluble odorants (418, 476). It is generally thought that the accessory system specializes in pheromone detection, whereas the main system detects common odorants (56; but see Refs. 109, 204, 375, 436, 443).
Olfaction is applied to chemosensory systems that detect chemicals emanating from a distant source. In contrast, when chemosensory systems require physical contact with the source for detection, they are called gustatory. Because vertebrates have anatomically separate olfactory and gustatory systems, only olfaction will be considered here in analogous systems belonging to different phyla.
In terrestrial environments, chemical signals can be either volatile or nonvolatile. Accordingly, terrestrial vertebrates have two functionally and anatomically distinct olfactory systems: one detecting volatile cues (the main olfactory system) and another thought to process mostly nonvolatile signals (the vomeronasal system) (Fig. 1). Such a dichotomy has been brought into play to support the long-standing hypothesis according to which the vomeronasal system evolved as an adaptation to terrestrial life (51). Today, accumulated evidence rather contests this assumption. For instance, we now know that modern amphibians and amniotes also possess vomeronasal organs. It is thus likely, but yet not demonstrated, that their last common ancestor also possessed a vomeronasal organ. Furthermore, two families of fully aquatic, nonmetamorphosing salamanders, were shown to possess vomeronasal systems, implying that the emergence of the vomeronasal system preceded that of terrestrial life (120). The evolution of a vomeronasal system in aquatic species might rather provide a selective advantage for terrestrial life, and consequently, it could have been retained in many species of terrestrial vertebrates. In spite of this, anatomical studies, and most recently molecular studies, indicate that the selective pressure to retain vomeronasal chemosensory input has been lost in higher primates. As a result, Old World primates, apes, and humans might not have retained a functional vomeronasal system (115, 308). Alternatively, species without a distinct vomeronasal system may still have an accessory olfactory system intermingled within the main system. Thus it is yet possible that the accessory system did not "arise" at some point of the vertebrate evolution, but rather it just became anatomically separated from the main system.
As our knowledge about the neurobiology of olfaction is growing, it is becoming incredibly evident that the main olfactory systems of animals in disparate phyla have many striking features in common (121, 194, 253). For instance, vertebrate and insect olfactory systems display common organizational and functional characteristics. Further recent works that were undertaken to broaden this scope to include nematodes, mollusks, and crustaceans have only strengthened this assumption. Existence of these similarities either results from sharing inheritance or might come from independent evolution through convergence (1, 194, 431). Whatever the mechanism is, the initial common event, shared by all odorant detection systems, requires the specific interaction of odorant molecules with specific receptors expressed on the cilia of sensory olfactory neurons before conveying information to central structures.
Basically, four features are shared by all olfactory systems. They include 1) the presence of odorant binding proteins in the fluid overlying the receptor cell dendrite; 2) the requirement of G protein-coupled receptors as odorant receptors (even though some sensory neurons may use transmembrane guanylate cyclase receptors such as in Caenorhabditis elegans and mammals); 3) the use of a two-step signaling cascade in odorant transduction; and 4) the presence of functional structures at the first central target in the olfactory pathway (194). All these characteristics may represent adaptations that have evolved independently, and therefore might provide us with valuable information about the way the nervous system processes odorant stimuli. Alternatively, these shared properties may instead reflect underlying homology or could have arisen independently due to similar constraints (121).
The detection capacity of a wide variety of olfactory systems arises from invariant series of information-processing steps that occur in anatomically distinct structures. In mammals, the olfactory epithelium contains several thousands of bipolar olfactory sensory neurons, each projecting to one of several modules in the olfactory bulb. These discrete and spherical structures, called olfactory glomeruli, are considered to be both morphological and functional units made of distinctive bundles of neuropil (416) (Fig. 1). This term reflects both the homogeneity of the sensory inputs received by the cells in the glomerular unit and the degree to which the neurons in the same glomerular unit are interconnected. Their number varies in different species: rodent olfactory bulbs contain several thousand glomeruli, and fish and insects have 10-fold fewer. In different species, each glomerular structure results from the convergence in the olfactory bulb of 5,000–40,000 axon terminals that establish synapses with dendrites of bulbar output neurons and of various classes of local bulbar interneurons. Because each group of glomerulus-specific output neurons is odorant receptor specific, they form a morphological defined network somewhat analogous to ocular-dominance columns in visual cortex or to barrels in the somatosensory cortex. It is also worth noting that a number of mechanisms have evolved to ensure that only a single odorant receptor is expressed per sensory cell. In rodents, tight transcriptional control results in the choice of one among a possible thousand odorant receptor genes (292). This extremely large repertoire of odorant receptors is undergoing rapid evolution, with at least 20% of the genes lost to frame-shift mutations, deletions, and point mutations that are the hallmarks of pseudogenes (81, 383; reviewed in Refs. 313, 454). Facing a changing environment, this characteristic may reflect the pressure made on a gene family to diversify and generate large numbers of new receptors that might confer new selective advantages. Interestingly, 
50% of human odorant receptor genes carry one or more coding region disruptions and are therefore considered pseudogenes (157, 291, 384). In fact, this massive pseudogenization of the odorant receptors repertoire in humans and Old World primates is preceded by a moderately high level of pseudogenes (30%) (153). In contrast, of the thousand odorant receptor sequences in the mouse genome, 20% are pseudogenes (158, 490). Thus there has been a decrease in the size of the intact odorant receptor repertoire in apes relative to other mammals, with a further deterioration of this repertoire in humans (153, 383). Because such decline occurred concomitant with the evolution of full trichromatic vision in two separate primate lineages, it has recently been proposed that the weakening of olfaction might result from the evolution of full color vision in our primate ancestors (154).
|
III. FROM ODORANT MOLECULES TO CORTICAL CENTERS |
|---|
|
TopPreviousNextReferences |
|---|
A. Olfactory Sensory Neurons and Signal Transduction
The olfactory sensory organ is made of a specialized olfactory neuroepithelium that directly interacts with inhaled odorants (116, 127, 312). This organ is exquisitely tuned to recognize an immense variety of molecules, different in shape, size, or chemical function that will be further encoded by neuronal circuits. It is made up of three major cell types: sensory neurons, supporting sustentacular cells, and several types of basal cells including the olfactory stem cells (Fig. 1). The former are unusual in that they are short-lived cells that exist for only 30–60 days (174, 326). Once mature, the sensory bipolar neurons extend a single dendrite to the neuroepithelial surface from the apical pole. Numerous cilia protrude from this dendrite and extensively invade the mucus lining of the nasal cavity. Odor molecules that dissolve in the nasal mucus bind to specific receptors on the cilia of olfactory sensory neurons. Therefore, the first step takes place at the interaction between odorant molecules and their respective receptors in sensory neuron dendrites (56, 57, 312, 313, 490). Odorant receptors can bind a number of volatile compounds with rather moderate affinity despite the overall high sensitivity of the system (230). Because each individual receptor is substantially cross-reactive for different ligands, the receptor repertoire might evolve according to the concentration and the mixture of odorants (113). Odorant receptors belong to the family of the G protein-coupled seven-transmembrane proteins (382), and several findings suggest that these receptors signal through the tissue-specific downstream components, the heterotrimeric G protein subunit Golf (368), type III adenylyl cyclase (348, 381, 423), and a cyclic nucleotide-gated ion channel (78, 130, 265, 330, 497) to mediate odor detection. In addition to this major pathway, several other second messenger cascades [e.g., Ca2+, inositol trisphosphate (IP3), or cGMP] that are activated upon odorant detection are thought to regulate secondary events such as odorant adaptation for instance (159, 401).
According to recent genome-wide analysis, there are between 1,000 and 1,300 odorant receptor genes, with an intact open reading frame number, in the mouse. This constitutes the largest gene family so far in the mammalian genome, perhaps in any genome (482, 490, 491). These genes have a compact gene structure and are scattered throughout the genome in clusters of various sizes. Olfactory receptor genes have been isolated from several vertebrate species including rat, mouse, human, catfish, zebrafish, dog, frog, chicken, pig, opossum, mudpuppy, and lamprey. Homologies are not recognizable between insect and vertebrate olfactory receptors (455) and are rather minor between fish and mammalian receptors (335, 462).
Olfactory receptor genes form also the largest category of genes with monoallelic expression. This principle was originally demonstrated by single-cell reverse transcription-polymerase chain reaction (RT-PCR) on pools of olfactory sensory neurons using limiting dilution and polymorphic alleles (81) and led to the one receptor-one neuron hypothesis. However, an alternative model proposing that a single neuron expresses during differentiation zero, one, or a few odorant receptor genes has recently challenged this view. According to this hypothesis termed "oligogenic expression," the developmental phase of oligogenic expression is followed by positive and negative selections resulting in cells with one expressed receptor (314).
Once expressed in the membrane of the sensory neuron, the activation of olfactory receptors induces a cascade of intracellular events resulting in an influx of both Na+ and Ca2+ (129, 331) that culminate in the generation of a graded receptor potential in the soma of the sensory neuron (151, 347). Electrophysiological studies indicate that odorant sensitivity and the odorant-induced current are uniformly distributed along the cilia, suggesting that all the components of the immediate responses to odorants are localized to the cilia. From its basal pole, the sensory neuron sends a single axon through the basal lamina and cribiform plate (of the ethmoid bone) to terminate in the olfactory bulb, the first central relay (Fig. 1). The unmyelinated sensory neuron axons merge into densely packed fascicles to form the olfactory nerve, which transmits the electrical signals to the bulb. Throughout the olfactory pathway, a unique population of glia, the ensheathing cells, forms the bundles of axons that make up the olfactory nerve. The gathering of sensory axons may lead to ephaptic transmission that allows synchronizing action potential in neighboring fibers, with submillisecond precision, by extracellular electrical field (38). Upon reaching the olfactory bulb, axon terminals from olfactory sensory neurons that express the same odorant receptor converge on a specific glomerulus (310) and arborize to form 15 synapses with target dendrites (184, 242). Axonal projections from the sensory neurons to the olfactory bulb form reproducible patterns of glomeruli in two widely separated regions of each bulb, creating two mirror-symmetric maps of odorant receptor projections (311). Surprisingly, it has been shown that odorant receptor identity in epithelial neurons determines not only glomerular convergence and function, but also organizes the neural bulbar circuitry (29).
It is important to note that there is no strict spatial relationship between the arrangement of excitatory projections of the olfactory sensory neurons in the olfactory bulb and the regions of mucosa from which they originate. This feature contrasts with the spatial organization of other sensory systems where afferent inputs are organized in a rather precise topographical mode. Similarly, as described below, much evidence indicates that bulbar outputs do not have point-to-point topographical projections to their target structures, which are characteristic of all other sensory systems. In mammals, the convergence ratio of sensory neurons-to-olfactory bulb output neuron is very large: 1,000:1. A bulbar output neuron thus forms its responses to odors from very large numbers of converging sensory inputs, ensuring that postsynaptic averaging increases signal-to-noise ratios. Interestingly, it has recently been shown that the mechanism underlying the extensive organization and targeting, between the olfactory sensory neurons and the olfactory bulb, implicates an activity-dependent mechanism. With the use of random inactivation of the X-chromosome in a genetic model to establish competition between normal and channel-deficient olfactory neurons, an activity-dependent competition between sensory neuron terminals was indeed revealed (492). Experience-dependent selection for trophic factors, as has been described in other neuronal systems, could account for this selection process (369). Two recent reports highlight further mechanisms based on spontaneous and odorant-evoked neuronal activity in the establishment and maintenance of the sensory projections. First, it was proposed that spontaneous activity of olfactory sensory neurons plays a permissive rather than instructive role in this process (483). Then, it was demonstrated that neuronal activity could also help to weed out weak synaptic connections, thereby contributing to spatial refinement and plasticity of the sensory projections (495). Clearly, more work will be needed to clarify the mechanism by which neuronal activity might be able to refine and stabilize sensory projection to the olfactory bulb.
B. Synaptic Transmission at the First Processing Stage
Unlike other sensory neurons, axonal termini of olfactory sensory neurons synapse directly onto second-order neurons within the forebrain (90, 173, 325). There, they form synapses impinging onto both output (second-order) neurons and local interneurons of the olfactory bulb (124, 250). At least in mammals, this makes the olfactory bulb the major site of integration for the olfactory information. The topography of these connections has been the subject of extensive studies revealing that sensory neurons expressing a given receptor project to a given subset of glomeruli (Fig. 1). It was concluded that the processing of olfactory information adheres to a certain spatial distribution, at least between sensory neurons and the olfactory bulb itself (see below), and remarkably such topographical organization is conserved in different species across phyla.
1. Characteristics of sensory inputs onto bulbar neurons
In the glomeruli, olfactory nerve terminals form excitatory glutamatergic synapses with the apical dendrites of the bulbar output cells (16, 32, 124), and with periglomerular interneurons (22, 235, 249, 357). These two olfactory nerve-evoked excitatory response types comprise fast amino-3-hydroxy-5-methyl-4-isoaxozole-propionic acid (AMPA) and slow N-methyl-D-aspartate (NMDA) components (16, 22, 32, 80, 85, 124). The latter is particularly long lasting and thus plays an important role in the bulbar output by maintaining a pattern of sustained discharge of output neurons (69, 193).
It should be kept in mind that a given glomerulus can respond to multiple odorants, and a given odorant activates multiple glomeruli. As a result, odor identity is represented rather combinatorially by patterns of glomerular activation that might rely, at least in part, on the properties of synaptic transmission between sensory neurons and their postsynaptic targets in the bulb.
Since the olfactory system can detect extremely faint sensory signals, some mechanisms might occur to reliably transmit information contained in the odorant-evoked firing of sensory neurons to the brain. In other sensory systems, some devices such as the synaptic ribbons in the retina or the cochlea indeed enable sustained and reliable synaptic transmission (147, 350). Because such a presynaptic specialization is not present in the terminals of olfactory sensory neurons, other features supporting a reliable synaptic transmission might be involved. Detailed analyses of olfactory nerve-evoked excitatory responses in olfactory bulb slices have shown a marked paired-pulse depression, supporting that glutamate release from sensory neuron terminals is high under normal conditions (17, 200, 235, 328, 409). The presence of presynaptic cyclic nucleotides (both cAMP and cGMP) provides a mechanism by which afferent inputs to the bulb are highly reliable (328). To characterize this feature, the properties of transmitter release from olfactory nerve terminals have been further examined. With the use of stationary fluctuation analysis of AMPA receptor excitatory postsynaptic currents (EPSCs) and the progressive blockade of NMDA receptor EPSCs by (5R,10S)-(+)-5-methyl-10,11-dihydro-5H-dibenzo[a,d]cyclohepten-5–10-imine hydrogen maleate (MK-801), it was demonstrated that the probability of glutamate release from nerve terminals is unusually high (
0.8) (327). This suggests that olfactory nerve terminals have specialized features that may contribute to the reliable transmission of sensory information from the olfactory sensory organ to the first central relay. This might be critical to amplify extremely faint signals. In the presence of threshold odorant concentrations, such features would serve to ensure transmission of information contained in the sparse firing of sensory neurons. In contrast, the strong paired-pulse depression likely constrains the saliency of bursts evoked by high concentrations of odorant to the first spike. Supporting the notion that the probability of release from olfactory nerve terminals is near maximal when odorants evoke action potentials is the fact that activation of any presynaptic metabotropic receptors, known to be present on olfactory nerve terminals, has been shown, so far, to reduce but never strengthen the synaptic transmission between olfactory sensory neuronsand the olfactory bulb (e.g., Refs. 17, 123). Hence, periglomerular local neurons release dopamine and GABA within olfactory bulb glomeruli (301, 425). These local neuromodulators, which reduce both the transmitter release from olfactory nerve terminals and paired-pulse depression, via activation of presynaptic D2 dopamine and GABAB receptors (17, 123, 200, 328), could further maintain relative response magnitudes across a wide range of input intensities, reduce sensory noise, and improve contrast between neighbor-activated glomeruli.
In addition to this unique presynaptic feature of olfactory receptor neurons, an alternative mechanism for signal amplification coming from the sensory organ itself relies on synchronizing glutamate release from sensory neuron terminals together. This might occur when presynaptic action potentials are synchronized (38). The mammalian olfactory nerve is arranged in configurations that may favor such synchronization. These axons lack myelin, and they are arranged in densely packed fascicles (94, 110, 175). Each fascicle contains between 10 and 200 axons (297), with each axon having a diameter of 0.2 µm (175). Axons within a fascicle are oriented parallel to each other and do not branch before they reach their termination site in the glomeruli of the olfactory bulb (94, 317). The high packing density and the geometry of these axons suggest that neighboring axons can be synchronized with each other through several forms of interactions. Although there are no known chemical synapses between sensory neurons, there are other potential means of communication between these nonmyelinated axons, including gaseous messengers (44), gap junctions (488), ephatic interactions (direct transmission of electrical signals) (38), and extruded potassium (36). Such synchronization might help to produce an oscillatory drive onto bulbar output neurons (188, 296) that is subsequently filtered and amplified by intrinsic subthreshold voltage oscillations occurring in the depolarized state of bistable output neurons (193).
2. Synaptic processing within olfactory bulb microcircuits
Because of its relatively simple anatomical organization and easy accessibility, the olfactory bulb has been a favored model system for investigating neural processing of sensory information. Odors elicit a well-organized pattern of activation in glomeruli across the surface of the olfactory bulb, but the mechanisms by which this map is transformed into an odor code are still unclear. With the advance in recent years of in vitro brain slice preparation, as well as in vivo techniques that can be applied on live animals, recently complex processing of the olfactory information has started to be revealed.
Since Cajal's pioneering studies, it is known that the main output neurons of the bulb, the so-called mitral cells, are located in a single lamina, the mitral cell layer. Their primary (or apical) dendrite, extending vertically from its soma, contacts one glomerulus (at least in mammals), where massive interactions with bulbar interneurons and olfactory nerve terminals occur. Most of the local interneurons have dendrites restricted to one glomerulus and impinge onto olfactory nerve terminals or mitral cell primary dendrites. In contrast to the primary dendrite, mitral cell secondary (or basal) dendrites radiate horizontally, up to 1,000 µm, to span almost entirely the olfactory bulb (318, 346). In the external plexiform layer, they interact with inhibitory axonless interneurons, named granule cells, the most numerous cellular populations within the bulb. Both sensory excitatory inputs and the intrabulbar circuit, which mainly includes two distinct connections, between primary dendrites and periglomerular cells, and between secondary dendrites and granule cells (415), tightly controlled the firing activity of output neurons. The main difference between periglomerular and granule cells is that the former mediate mostly interactions between cells affiliated with the same glomerulus while granule cells mostly mediate interactions between output neurons projecting to many different glomeruli. However, periglomerular neurons might exert also a distant action by interacting with another class of bulbar interneurons. The functional consequences of this synaptic organization will be described below.
The synaptic mechanisms that play a key role in the circuits of the olfactory bulb have two unusual features. First, many bulbar neurons communicate via reciprocal dendrodendritic synapses (357, 361, 365). The reciprocal circuit provides inhibition that forms the basis for a reliable, spatially localized, recurrent inhibition (365). A mitral cell's synaptic depolarization, driven by the long-lasting excitatory input from the sensory neurons, triggers glutamate release by dendrites and thus depolarizes interneuron dendrites and spines (118, 208, 402, 444). This, in turn, elicits the release of GABA directly back onto output neurons (208, 210, 339, 402). Second, several bulbar neuronal types are known to modulate their own activity through the transmitters that they themselves release (207, 403, 425), and transmitter release might occur, in some cell types, through an action potential-independent manner (208, 210, 402).
Because secondary dendrites have large projection fields and extensive reciprocal connections with interneurons (415), each bulbar local neuron may contact the dendrites of numerous output neurons. This suggests that not only do dendrodendritic interactions provide a fast and graded feedback inhibition, but they also offer a unique mechanism for lateral inhibition between output neurons that innervate different glomeruli (208, 295, 402, 404 407, 481). Consequently, it has been demonstrated that bulbar projecting neurons connected to different glomerular units, and which respond to a wide range of related odor molecules, also receive inhibitory inputs from neighboring glomerular units via lateral inhibition at dendrodendritic connections (321). Thus the propagation of action potential into the lateral dendrites, and the possible spread of excitation through granule cell dendrites, contribute to a "spatial" contrast mechanism that sharpens the tuning of output neuron odorant receptive fields (447, 481, but see Refs. 258, 260).
More recently, a distinct and novel intrabulbar network was found to fulfill a similar function. The so-called "short axon" cells, located near glomeruli, send interglomerular axons over long distances to excite inhibitory periglomerular neurons (20). This interglomerular center-surround inhibitory network, along with the mitral-granule-mitral inhibitory circuit above described, forms a serial, two-stage inhibitory circuit that could enhance spatiotemporal responses to odors. In this case, information is transmitted not only vertically across the glomerular relay between sensory neurons and output neurons, but also horizontally through local interneuron connections that are activated in odor-specific patterns. Such a model based on lateral connection, originally introduced into neuroscience to explain visual contrast enhancement in the retina (187), has been mathematically extensively characterized (14, 148, 451, 453). Both anatomical and functional analyses support the existence of lateral inhibitory mechanisms, in the olfactory bulb, through which activity in few stimulated output neurons may lead to suppression of other neighboring neurons innervating distinct glomeruli. This inhibition was proposed to refine the process of odor information. For instance, examination of the responses of individual bulbar neuron to inhalation of aliphatic aldehydes reveals that many individual cells are excited by one subset of these odorants, inhibited by another subset, and unaffected by yet a third subset (321). Glomerular unit has thus the potential to respond to a wide range of related odorants but also receives inhibitory inputs from neighboring glomerular units through lateral inhibition. The quality of the odor stimulus is first encoded by a pattern of sensory input activity across glomeruli that are, at least in first approximation, not shaped by inhibition. The odor stimulus is further encoded by mitral cell activity patterns within the olfactory bulb by a specific combination of activated mitral cells that critically depend on GABAergic inhibition.
In addition to the lateral inhibition model, the long-range projections of secondary dendrites could support a novel perspective that has emerged recently and views bulbar microcircuits as a nonlinear dynamical system (260). According to this view, the olfactory bulb transforms stationary input patterns into time-varying output patterns, moving along input-specific trajectories in coding space. In this framework, the main function of bulbar microcircuits would be to enable odor-specific dynamics that can decorrelate input patterns. Such a decorrelation function would distribute clustered input patterns more evenly in coding space, thus optimizing the use of the coding space for discrimination and other olfactory tasks (Fig. 2). It is here important to note that not only decorrelation can be achieved by lateral inhibition but also that inhibition is crucially involved in the decorrelation. Supporting this notion is the fact that the bulbar inhibition is not restricted to near neighbors, but can extend at least over medium spatial ranges. In other words, because secondary dendrites project to long distances, the olfactory bulb networks aim to reformat combinatorial representations so as to facilitate their readout by downstream centers. This model is consistent with experiments showing that GABAergic reciprocal inhibition contributes to synchronizing the output neuron activity (60, 100, 228) mainly through granule cell spine activity (254).
|
In addition to inhibitory inputs arising from granule cells, it has been reported that output neuron secondary dendrites receive large excitatory inputs when either inhibition was antagonized or magnesium was removed from the external medium (18, 142, 207). In fact, the first evidence for self-excitation in mitral cells came from the turtle olfactory bulb, when injection of a depolarizing current into a mitral cell elicited a prominent glutamate receptor-dependent afterdepolarization (337). On the basis of its resistance to blockade of action potentials with tetrodotoxin, it was suggested that self-excitation originated from dendrodendritic synapses rather than from recurrent axon collaterals. Patch-clamp studies have shown that both AMPA and NMDA glutamate receptors can mediate self-excitation (393, 405) on both primary and secondary dendrites (393). More recent studies performed in slices have demonstrated that mitral cells receiving synaptic inputs within a common glomerulus laterally excite one another through chemical and/or electrical transmission (69, 207, 406, 448). Thus secondary dendrites also serve the more classical function of somatic input devices, integrating both excitation and inhibition from nearby local interneurons. In mammals, excitatory synapses onto mitral cells have been localized exclusively to the apical dendritic tufts that receive primary sensory afferents (249, 357, 411). However, glutamate autoreceptors have been localized anatomically on output neurons near dendrodendritic synapses (316), but their exact location with respect to the glutamate release site is still not known (395). The origin of the excitatory inputs to the secondary dendrites is therefore highly debated. It has been proposed that glutamate released from mitral cells could activate glutamate autoreceptors that can in turn enhances their firing (142, 207, 393). Self-excitation can drive an afterdischarge in output neurons that last for hundreds of milliseconds, reflecting the fact that a significant component of the response is mediated by slow NMDA autoreceptors (403). In addition, using a combination of in vitro whole cell recordings and immunogold detection of glutamate, it was demonstrated that ionotropic glutamate receptors on secondary dendrites of bulbar output neurons could also be activated by interneurons located in the granule cell layer (101).
It is noteworthy that remote synaptic contacts on the secondary dendrite are probably electronically decoupled from the soma and have minimal impact on somatic firing, but they might nevertheless affect local outgoing spikes. Supporting this view is the finding that inhibitory synaptic events, elicited by stimulating subjacent granule cells, completely abolish spike propagation (477). Alternatively, by participating in spike traffic, the excitatory synaptic events received by secondary dendrites might be considered as safety factors for propagation of action potentials. Together, dual excitation-inhibition responses impinging onto secondary dendrites might offer a unique mechanism for stability of bulbar network activity to recruitment of additional glomeruli with increasing odorant concentration (144, 303, 389, 456). The relative amplitudes of both excitatory and inhibitory responses received by output neurons during odor responses in vivo depend on their activation states and the network of granule cells (295).
Taking into account all these findings, it is obvious that the balance between excitation and inhibition in the olfactory bulb, and thus the interplay between local interneurons and output neurons, provides a combinatorial device to the representation of olfactory information. Contacts mediated by local interneurons and bulbar output neurons may contribute together to a global reformatting of odor representations, in the form of a stimulus-dependent, temporal redistribution of activity across the olfactory bulb (Fig. 2). Depending on which sensory neuron afferents are stimulated and which local interneuron connections become active as a result, different topologies of inhibitory circuits are expected to assemble. From this framework, two combinatorial encoders that take part in information processing within the bulbar network could be distinguished (260). The first one consists of the olfactory receptor repertoire expressed by the sensory neuron ensemble that transduces receptor activation patterns into glomerular odor maps throughout a highly reliable synaptic transmission (stimulus space; Fig. 2). The secondary encoder lies in the intricate interneuron network that extracts, or at least prepares representations, for higher order features, from the odor images to convert them as timing relationships across the firing output neuron ensemble (e.g., whether neurons are simultaneoulsy active or not) (representation space; Fig. 2).
Finally, the organization of pattern activity in the bulb that relies on a proper synaptic interaction between local interneurons and output neurons depends on precise ontogenetic mechanisms that tightly control the bulbar wiring. Recent studies have brought some new insights into how intrabulbar connections are fine-tuned by neuronal activity (29, 274). According to these works, the extraordinary specificity of sensory axon convergence in the olfactory bulb was recapitulated in additional aspects of the bulb connectivity. Using a combination of genetically tagged receptors and focal dye injection into glomeruli, it was demonstrated that not only bulbar output neurons project axons across the medial-lateral axis of the bulb to a similar region on the opposing side, a finding previously reported (271), but also that they were extremely flexible according to the degree of sensory inputs. These authors reported that the intrabulbar circuit was rewired with the arrival of novel olfactory sensory neurons (271). It was therefore proposed that the postsynaptic targets of sensory neurons are not prespecified but instead are instructed and organized by their presynaptic partners. Altogether, the ability of the odor environment to affect the projection and survival of sensory neurons as well as the intrabulbar connectivity suggests that evoked activity contributes to the connectivity as well as the maintenance of the olfactory circuit (369, 483, 495).
C. Decoding Information in Higher Olfactory Centers
Odor information received by the olfactory bulb is first processed and refined before being transmitted to downstream centers. As described above, two potential sites of odor processing can be distinguished according to the topographical organization of the bulbar circuit. The first one resides in the glomeruli where local interneurons shape excitatory inputs coming from sensory neurons. The second one lies in the external plexiform layer of the olfactory bulb where reciprocal dendrodendritic synapses, between dendritic spines of local interneurons and the dendrite of output neurons, are heavily distributed. The final processing occurs in higher-order brain structures comprising the primary and accessory olfactory cortex. The axons of bulbar output neurons project in the olfactory tract to higher-order brain structures without contacting the thalamus. These higher centers include the anterior olfactory nucleus, which connects the two olfactory bulbs through a portion of the anterior commissure, the olfactory tubercle, the piriform cortex (considered to be the primary olfactory cortex), the cortical nucleus of the amygdala, and the entorhinal area (416). Nevertheless, olfactory information is ultimately relayed through the thalamus to the neocortex, similar to other sensory systems. The olfactory tubercle projects to the medial dorsal nucleus of the thalamus, which in turn projects to the orbitofrontal cortex, the region of cortex thought to be involved in the conscious perception of smell (377).
Olfactory information must be relayed from convergent synapses in the olfactory bulb to higher brain centers, where it is decoded to yield a coherent odor image. Experiments in mice that traced the olfactory circuitry of olfactory sensory neurons expressing a given odorant receptor suggest a distributed olfactory code in the olfactory cortex (496). This study employing genetically encoded transneuronal tracers in the mouse olfactory system has shown that output neurons which receive input from a single glomerulus project to defined regions of the piriform cortex that are more extensive than the glomerular segregation. These tracing studies are consistent with functional studies that demonstrate a characteristic response property of individual mitral cells that reflects the molecular receptive range of the glomerulus from which it receives olfactory input (281). Overlap between the projection patterns of different glomeruli affords the opportunity for integration of olfactory information at higher olfactory centers. In both the olfactory bulb and higher centers, odor information seems to be encoded by activity across the entire neuronal network. This divergent connectivity is reminiscent of the persistence in discriminating odorants even after ablation of a large fraction of the olfactory bulb (279).
Despite recent progress, considerable functional analysis has yet to be performed before we will completely understand how odorant molecules are represented in higher olfactory centers. The nature of the olfactory stimulus itself seems crucial for olfactory processing in higher centers. For instance, the lateralization of odor information processing in the human brain is thought to depend on odor identity. Most odorants simultaneously activate two systems: the olfactory system, which generally projects ipsilaterally, and the trigeminal system, most of which crosses to the contralateral side. Thus hemispheric asymmetry for olfactory function strongly depends on the quality of the nasal stimulus, more specifically its potential olfactory or trigeminal stimulating properties (42).
|
IV. EARLY COMPUTATIONS AT THE FIRST INFORMATION PROCESSING STAGE |
|---|
|
TopPreviousNextReferences |
|---|
Among the different relays along the olfactory pathway, the olfactory bulb first plays a central role in relaying information from the sensory organ to several central targets (55, 114, 127, 312, 321). There has been some controversy over the relative importance of spatial versus temporal patterns of odor-induced activity (see, for instance, Refs. 56, 108, 248, 258, 260). The following issue reviews some recent findings regarding both spatial and temporal dimensions in the mammalian olfactory bulb. Because a critical or comprehensive assessment of all studies is not provided here, the reader may want to consult one of several recent reviews for a more complete perspective on this area (82, 143, 146, 182, 230, 247, 253, 259, 261, 469).
For all species, sensory stimuli are detected by specialized receptors located in sensory organs, and the signals hence generated flow through multiple and interconnected centers of the brain where they are analyzed and processed into sensory perception. Such a process requires the coding of sensory inputs into specific patterns of neuronal activity. A common mechanism for sensory information coding is provided by the topographic organization of sensory neurons and their axonal projections, such that sensory centers represent an internal map of the external sensory world: the body surface, the frequency of sounds, and the visual world. In that manner, the nature of sensory stimuli is encoded by the spatial coordinates of neuronal activity in high sensory centers of the brain, and the discrimination between sensory signals results from the stimulation of topographically distinct subsets of neurons.
Inputs and local circuitry of several sensory systems are organized in modular fashion (e.g., the rodent barrel cortex), with similar inputs being received by groups of cells that are highly interconnected. In the olfactory system, while other parameters are likely to be as important, spatial activity patterns have been hypothesized to play a key role in sensory information encoding. Several studies have provided experimental evidence illustrating the relationship between spatial patterns of olfactory bulb activity and features of odor stimuli (reviewed in Refs. 232, 312, 442, 478). The analyses of the glomerular convergence of olfactory axons using genetically modified mice suggest that individual glomeruli represent only one or at most a few type(s) of odorant receptors. The axonal projections of sensory neurons expressing the same receptor converge onto a few spatially fixed glomeruli within a large glomerular array on the surface of the bulb, thus creating a stimulus-specific two-dimensional spatial representation of olfactory receptor activation (56, 311, 315). According to this spatial map, signals from different types of odorant receptors are sorted into different glomeruli and individual output neurons associated with a single glomerulus are tuned to specific features of odor molecules (205, 320, 322).
Together, these findings support the idea that olfactory image of a given odorant object is coded by a specific combination of activated glomeruli. In mammals, odor-specific spatial activation patterns have been reported using several distinct methods including analysis of immediate early gene expression such as c-fos, c-jun, zif268 and Arc (21, 177–179, 206, 215, 400), 2-deoxyglucose mapping (19, 88, 212–214, 219, 387, 417, 422, 429), and functional magnetic resonance imaging (fMRI) (427, 479). More recently, optical imaging based on intrinsic signals (28, 281, 303, 389, 446), calcium signaling (41, 456), or voltage-sensitive dye (235, 428) have been successfully applied to map individual glomeruli. These physiological approaches revealed that the initial mapping of olfactory stimuli across the spatial dimension of the olfactory bulb arises from the precise convergence of receptor neurons expressing the same odorant receptor onto only two glomeruli located in a stereotyped manner. They also demonstrated that physiological responses of the bulb were odor specific, bilaterally symmetric, reproducible over multiple trials, conserved among conspecifics (suggesting precise ontogenetic control), and independent of the context of a given odor in an odor sequence. Third, they demonstrated that the specific detection of odorant is achieved using devices of rather low specificity. Nevertheless, the combination of responses of several receptors, each with low but different specificity, generates a unique fingerprint for each odorant or odor mixture (303, 389). With these characteristics taken into account, it has been possible to link map differences to differences in perception within the vertebrate olfactory system (268, 390, 400). Only perceptually distinguishable olfactory stereoisomers elicit recognizably different odor images (268), supporting the relevance of spatial activation patterns in encoding olfactory information. In addition, imaging data indicate that higher odor concentrations increase glomerular response intensity and recruit additional glomeruli (303, 389, 429), consistent with the observation that bulbar output neuron responses change both quantitatively and qualitatively with odor concentration (231, 461). Remarkably, single vertebrate olfactory receptor neurons are reported to have a dynamic range of only one to two log units (128, 370, 445), yet optical recordings often showed continuous increases in receptor neuron input spanning nearly three log units (457). Individual sensory neurons expressing the same odorant receptor may have identical odorant response profiles but differing thresholds, thus increasing the dynamic range of population terminals converging into a single glomerular (86, 128).
In fact, odor encoding is a spatially distributed process. It was initially thought that olfactory stimulation evokes distributed, odor-specific spatial patterns of activity in the olfactory bulb (177, 212–214, 216, 320, 321, 429). However, because single odors can activate broad overlapping regions, it is clear that individual neurons can be activated by many odorants, including ones that belong to different chemical families, underlying different odor qualities. Calcium imaging on dissociated olfactory sensory neurons (41, 292) and older studies in situ (150, 222) are consistent with such results. Furthermore, recent imaging studies indicated that the spatial pattern of bulbar activity is not only distributed but also extremely dynamic, thus providing a picture of how odor identity and concentration could be represented by a combination of spatial and temporal coding (285, 303, 389, 446, 480). In this way, action potential timing or rate could code nontemporal stimulus features such as quality or intensity of odorants. More recently, detailed insights into these patterns have come from measuring changes in membrane potential within populations of neurons using a voltage-sensitive dye that nonselectively stains all olfactory bulb neurons (428). In this study, the dye signal, which is thought to primarily reflect postsynaptic neuronal activity, revealed that odor-evoked responses were widespread and not localized to individual glomeruli. The extensive nature of the voltage-sensitive dye signals may reflect the widespread lateral branching of bulbar output secondary dendrites, and further supports the hypothesis that processing of olfactory input involves distributed activity across much, if not all, of the bulb. In addition, they showed that odorant-evoked spatial patterns were not simply repeated during the breathing cycle (428). Some glomeruli responded less strongly during the second breathing cycle, suggesting that adaptation occurs, whereas others responded more strongly, indicating that other processes also contribute to the dynamics observed. Together, these results demonstrate that activity patterns in the bulb are not static, but rather evolve within a respiration cycle and from one cycle to the next.
In summary, odors generally activate an array of receptor types and, hence, elicit a well-organized pattern of activation in glomeruli distributed across the surface of the olfactory bulb. Different odors activate different combinations of glomeruli; the odor identity code across population of sensory neurons thus possesses a critical combinatorial component. These activity patterns depend on the identity of the active glomeruli and have to be considered as a function of time. Such odor map dynamic may be shaped by synaptic interactions within glomeruli, between neurons in adjacent glomeruli, or between neurons in subglomerular layers, as the olfactory input is transferred to higher-order neurons, during learning for instance. As a result, both spatial distribution and the temporal structure of neuronal activity should therefore not be studied in isolation but rather considered as a single entity of the same coding process.
Several experimental and theoretical points clearly argue against the single-neuron coding hypothesis for the sense of smell (23). One of the most obvious reasons relies on the number of neurons in the olfactory system, which is simply insufficient to represent the tremendous amount of information that an animal processes in its lifetime. In contrast, the biased allocation of channel bandwidth in favor of timing is especially useful for sensory systems that rely on large populations of neurons to convey the signal: one can readily increase the precision of the stimulus estimate by simply pooling more neurons (for reviews of population coding, see Refs. 61, 345). It is noteworthy that coding sensory information by cell assemblies (392, 465) offers four important advantages: 1) overlapping set coding of information items (according to this property, the same neuron is a part of many different assemblies made of overlapping sets of neurons that encode different information); 2) sparse coding of information items; 3) dynamic coding implying correlation and decorrelation since neurons are interconnected by flexible functional synapses that evolve with time; and 4) dynamic reverberant property since activation of a cell assembly persists for a time much after the offset of sensory inputs. This long-lasting effect results from complex feedback synaptic interactions. As we shall see below, the four characteristics apply to the olfactory system circuit.
1. Network dynamics in the olfactory system
Since the pioneering work of Lord Adrian (2), it has been proposed that the coding of odor information requires the activation of a large fraction of bulbar output neurons (232, 260, 321). Subsequently, numerous studies have shown that odor molecules are able to activate numerous glomeruli and a large number of output neurons, each broadly tuned to odor molecules (84, 144, 151, 389, 446). This led many authors to propose that spatial assembly of activated output neurons encodes odors in the olfactory bulb. The existence of a precise topography of glomeruli supports this view (55, 315, 452) and implies that a specific assembly of output neurons whose activity is determined by its connectivity can code each odor. However, stimulation of the olfactory epithelium with a mixture of odorant compounds or, even with a single compound, causes activation of bulbar output neurons that are, in many cases, distributed over several discrete regions of the bulb (322, 412, 429). Thus already at the first central stage that processes sensory inputs, the neuronal circuitry cannot be viewed only as a passive relay of odor-related information but rather as richly interconnected circuits in which excitation and inhibition are physically widespread. As a consequence, its global mode of action might not be captured accurately by either static or isolated samples.
Electrophysiological studies have revealed many forms of temporal patterning of activity of the bulbar output neurons that are relevant of the olfactory pattern-recognition task (60, 289, 306, 324, 461). Through such spatiotemporal patterns of neuronal activation, the circuit dynamic offers a large coding space that spreads odor representations of chemically related odor (e.g., acting as a decorrelator) and simultaneously optimizes this distribution within it throughout oscillatory synchronizations (e.g., formatting odor space) so that odor representations can be sparsened in the next olfactory station.
In general, changes in the format of sensory-related information rely on coordinated neural firing patterns in large-scale neural networks (122, 420, 441). The coordinated electrical activity emerges from collective behavior of neurons across large groups of neurons (165), and it is reliably correlated with a variety of behavioral states and cognitive processes such as attention, working memory, or perception (155, 167, 269, 450). In sensory systems, evoked oscillations are thought to encode stimuli (329, 420; reviewed in Refs. 260, 419) and to participate in fine discrimination (430). They have been reported in neural assemblies of various sensory systems such as vision (e.g., Ref. 333 for the retina and Ref. 166 for the visual cortex), audition (e.g., Ref. 98), and olfaction (reviewed in Ref. 257). The temporal correlation hypothesis from the "binding" theory remains a matter of active debate. Among the contentious issues are how prevalent are the oscillatory activities in evoked neural responses, and to what extent neural oscillations are correlated with sensory stimulation (410, 419). Before answering these questions, it is important to note that temporal patterning comprises different components. Some are imposed onto odor stimuli by the dynamics of the carrier medium and by active sampling, whereas others are generated internally by neuronal dynamics in the olfactory bulb. Odor information in the bulb is represented not only by the combination of active neurons in the network, but also by the slow temporal sequence of activity patterns and by the synchronization of neuronal subpopulations (143). Another property of population activity that can subserve odor processing is the chemotopic spatial format (146). There is no doubt that exploring the extent to which these features of population activity are decoded by the brain, and by what mechanisms, will be the major issues for the near future.
2. Temporal features of olfactory responses
By recording the local field potential (LFP), which is an indirect reflection of neuronal activity since it is generated by current flows through the extracellular space linking inward and outward membrane currents, both spontaneous and odor-induced LFP oscillations were found to be a universal feature of olfactory processing systems from a wide variety of vertebrates (2, 3, 50, 72, 134, 347, 396, 405, 434). For a given odor stimulus, the "induced rhythm" (as first termed by E. D. Adrian) is synchronized transiently in time and only among a neural subpopulation that is selectively responsive to that particular odorant. The inhalation of odor molecules has first been reported to trigger oscillations in the olfactory bulb with different frequency ranges (e.g., Refs. 2, 47, 50, 117, 137–139, 168, 233). A robust pair of fast oscillations defined by their frequencies as gamma (30–80 Hz) and beta (15–40 Hz), as well as a slower one called theta rhythm (3–12 Hz), are classically observed in the olfactory system. Whereas gamma and beta waves are induced by odor inputs, the theta frequency band seems rather to be phase-locked with respiration (3, 47, 50, 117, 131, 138, 233). This respiration-coupled theta rhythm displays spatiotemporal dynamics in response to odor stimuli (428) and is also present in downstream piriform cortex (49, 467, 468), indicative of a potential function in the representation of odor stimuli.
The amplitude and frequency of these oscillations may reflect previous olfactory experience and the behavior of the animal (37, 48, 72, 134, 140, 233, 366). Hence, previous studies focusing on gamma frequencies have shown that the amplitude of oscillatory bursts associated with odor sampling defined maps over the bulbar surface. Yet, those maps were more related to the behavioral content of the odor than to its chemical nature (48, 103, 138). More recently, LFP recordings revealed that odor sampling enhanced oscillatory activity in beta frequency range when odors were experimentally associated with a reward (37, 366) or after repeated presentations of the same odor (72, 168). It is worth noting that gamma and beta frequency bands never coexist in response to odorant stimulation. Hence, recent studies examining spatiotemporal evolution of odor-induced LFP oscillations in the bulb during training, reported an alternation of the two rhythms during odor processing (59, 298, 334). From these studies, it is clear that odor sampling and learning are associated with a decrease in gamma burst power followed by an increase in beta oscillatory activity.
Interestingly, all rhythms observed in the first relay of mammals have also been seen in their functional analog in invertebrates and are thought to encode stimuli (reviewed in Ref. 260) as well as to participate in the fine discrimination of close stimuli (430). Extension of the results o
