OSNs expressing the same odorant receptor converge onto one or a few glomeruli within the olfactory bulb/antennal lobe (Ressler et al. 1994; Vassar et al. 1994; Mombaerts et al. 1996) and appear to be functionally equivalent in their odor response properties (Wachowiak et al. 2004), thus establishing a spatial map of receptor expression. Multiple axon guidance mechanisms cooperate in this very precise targeting of OSN axons, including the odorant receptor itself (Mombaerts 2001). Recent results indicate that at least the final precision of glomerular targeting is controlled by homotypic interactions between odorant receptors expressed on OSN axon terminals (Mombaerts and Feinstein, this volume).
The number of glomeruli is correlated with the number of functional odorant receptor genes in different species. Drosophila melanogaster has at least 61
odorant receptor genes and 43 glomeruli, mice have ^1000 odorant receptor genes and ^2000 glomeruli, whereas rats have ~1500 functional odorant receptor genes and ~ 3000 glomeruli. In mammals, each odorant receptor is associated with, on average, glomeruli in each olfactory bulb. The roughly 1:1 correspondence between the number of odorant receptor genes and the number of functionally different glomeruli lead to the model that each glomerulus integrates the input to the olfactory bulb/antennal lobe conveyed by one receptor type. Glomeruli, therefore, represent separate input channels or dimensions that are, however, not orthogonal to each other.
In mammals, but not in lower vertebrates and invertebrates, the map of receptor expression is mirror symmetric about a roughly vertical plane; that is, most idiotypic OSNs project to glomeruli at similar coordinates in the medial and lateral hemisphere of each olfactory bulb. Moreover, external tufted cells receiving input from a given glomerulus project to locations in the granule cell layer in the vicinity of the homotypic glomerulus in the other olfactory bulb hemisphere. Currently, the functional importance, if any, ofthe mirror-symmetric organization of the mammalian bulb is unresolved.
The spatial coordinates of idiotypic glomeruli are preserved, but not with exquisite precision. Rather, the position of a given glomerulus can vary within 1-2% ofthe surface ofthe olfactory bulb across individuals and between hemispheres ofthe same olfactory bulb. As a consequence, immediate neighborhood relationships between glomeruli are variable (Strotmann et al. 2000).
Odor-evoked activity across the array of glomeruli has been visualized by a variety of techniques, including 2-deoxyglucose uptake, c-fos expression, fMRI, intrinsic signal imaging, and calcium imaging (Stewart et al. 1979; Guthrie et al. 1993; Friedrich and Korsching 1997; Johnson et al. 1999; Rubin and Katz 1999; Sachse et al. 1999; Meister and Bonhoeffer 2001; Wachowiak and Cohen 2001; Xu et al. 2003). Even single chemical compounds activate multiple glomeruli and single glomeruli respond to multiple odorants, presumably because each odorant receptor can be activated by multiple compounds (Araneda et al. 2000). Odor information is, therefore, contained in a combinatorial pattern of activity across the array of glomeruli. Patterns evoked by chemically related odorants are often similar. Thus, microcircuits in the olfactory bulb/antennal lobe must analyze spatially distributed activity patterns to extract stimulus information.
An important question is whether the spatial relationships between glomeruli in the map reflect similarities between the respective odorant receptors' response profiles. Such an organization could create a chemotopic map, in which features of the chemical stimulus space are spatially mapped onto the array of glomeruli. Glomeruli responding similarly to a subset of odors sharing obvious chemical features are sometimes clustered spatially. For example, in experiments using 2-deoxyglucose uptake, intrinsic signal imaging, and fMRI, aliphatic aldehydes were found to activate glomeruli in an anteromedial region of the dorsal olfactory bulb. However, even within this region only a subset of glomeruli responded to aliphatic aldehydes, and these odors also stimulated glomeruli in other locations. Studies using imaging of calcium indicators or a transgenic fluorescent activity probe in OSN axon terminals (Wachowiak and Cohen 2001; Bozza et al. 2004) yielded only weak evidence for a chemotopy of aldehyde responses in the dorsal olfactory bulb. Moreover, individual glomeruli within a region loosely defined by its response to a class of odorants can also respond to other, dissimilar sets of stimuli. As a result, the spatial proximity of glomeruli appears to be only weakly correlated with the similarity of their overall response profile (Friedrich and Stopfer 2001). The structure of chemotopic maps is, therefore, not well understood and deserves further experimental attention. Nonetheless it is clear that a chemotopic organization, if it exists, is much more fractured than topographic maps in other sensory systems, possibly relating to the complexity and high dimensionality of the stimulus space (Friedrich and Stopfer 2001). It is also interesting to note that from the perspective of neuronal wiring, spatial distance between all glomeruli is equal in insects, because all interglomerular connections pass through the central area of the spherical antennal lobe (Sachse and Galizia, this volume).
It is now firmly believed that the identity of active units in the combinatorial pattern contains essential stimulus information. Currently unresolved, however, is the question whether the position per se of glomeruli in the map is important for the decoding of glomerular activity patterns. It is, for example, possible that the given arrangement of glomeruli simply minimizes the total wiring length of circuits in the olfactory bulb/antennal lobe or is a byproduct of the axon guidance mechanisms underlying glomerular targeting, that is, a developmental process. In a thought experiment, the shuffling of glomerular positions in the map does not affect the information conveyed by activity patterns. Moreover, the information could potentially be extracted in the same way after shuffling if all connections in the network were kept intact. However, the system may still require positional information for its function. For example, interactions through gap junctions or electrotonic mechanisms may require a particular spatial arrangement of functional units within the circuit. Furthermore, it is formally possible that the establishment of correct synaptic connections between target neurons and their inputs relies on axon guidance mechanisms that read positional cues and would be fooled when glomerular positions are scrambled. Currently, there is no conclusive evidence arguing either for or against a role for spatial position in olfactory system function. This is clearly one important open question. Ideally, the problem should be approached by (genetic) manipulation of glomer-ular positions without otherwise disturbing the system and subsequent tests of the olfactory system's performance. This is, however, beyond the current realms of experimental possibility.
In the deeper layers of the olfactory bulb, focal excitation of a few glomeruli produces a cone of activity that fans out with increasing depth (Guthrie et al.
1993). Hence, glomerular activation spreads laterally within the olfactory bulb and probably also in the antennal lobe (Wilson et al. 2004). It is still unresolved whether the olfactory bulb shares a columnar functional organization with other brain structures. A "reductionist" analysis of basic properties of olfactory microcircuits using focal stimuli maybe fruitful to derive insights into the mechanisms by which microcircuits process more complex glomerular activity patterns evoked by realistic stimuli.
TEMPORAL PATTERNING OF OUTPUT FROM THE OLFACTORY BULB/ANTENNAL LOBE
Olfactory microcircuits are renowned for their temporal dynamics. Temporal patterning of the activity of output neurons on at least two timescales has been observed in all species studied, vertebrates and invertebrates alike: (a) slow, aperiodic modulations of firing rate on a timescale of one or a few hundred of milliseconds, and (b) fast oscillatory synchronization with a precision of a few milliseconds. Further temporal patterns are observed in some species.
Output neurons respond to odor stimuli with modulations of their firing rate during the first hundreds of milliseconds after stimulus onset. These firing rate modulations can include successive excitatory and inhibitory epochs resulting from circuit interactions in the olfactory bulb/antennal lobe. In mammals, odor-specific modulations of firing probabilities occur during each breathing cycle. The mechanisms underlying slow temporal patterning of output neurons are still elusive. Candidate pathways mediating these effects include all of the interglomerular microcircuits mentioned above. In locusts, slow temporal patterns were not substantially affected by GABAa or GABAb antagonists (MacLeod and Laurent 1996).
The slow temporal modulation of firing probability evoked by one odor is different between output neurons, and different odors evoke distinct slow temporal firing patterns in the same output neuron. As a result, the pattern of activity (firing rate) across the population of output neurons evolves in an odor-specific manner after stimulus onset. After a rapid change during the initial phase of the odor response, activity patterns asymptotically approach a relatively stable state after a few hundred milliseconds.
In zebrafish, locusts, and possibly moths, it has been shown that the dynamic change of spiking activity patterns evoked by similar odors results in a decor-relation of activity patterns evoked by related odorants (Friedrich and Laurent 2001; Stopfer et al. 2003; Daly et al. 2004; Friedrich and Laurent 2004): immediately after response onset, activity patterns evoked by related odors are similar, possibly because output neurons are driven to a large extent by their sensory inputs, which respond similarly to related stimuli. Subsequently, however, patterns of output activity change, following trajectories that are specific for each stimulus and diverge over time. As a result, activity patterns evoked by related stimuli become more distinct, and the discrimination of patterns becomes significantly more reliable during the first few hundred milliseconds of the response. Hence, olfactory microcircuits perform a computation (pattern decorrelation) that appears important for the discrimination of "odor images" across glomeruli.
Odor-evoked population activity in the olfactory bulb/antennal lobe has an oscillatory component with frequencies ranging between 15 and 40 Hz in insects and lower vertebrates, and in the beta (15-30 Hz) and gamma (30-100 Hz) range in mammals. This oscillatory synchronization is mediated by reciprocal interactions between principal neurons and inhibitory interneurons in inter-glomerular microcircuits (see above). The spatial pattern of this oscillatory activity is widespread and only weakly reflects the discrete pattern of glomerular input. Within each oscillation cycle, only an odor-specific subset of output neurons synchronizes, while others fire without any apparent temporal relation to the oscillation. Hence, odor-specific subsets of spikes transmitted to higher brain regions are synchronized.
Since the integration of synaptic inputs in neurons can be exquisitely sensitive to temporal proximity, synchronized spiking may transiently establish neuronal ensembles that carry particular information accessible by coincidence detection-based readout mechanisms. Indeed, Kenyon cells in the mushroom body receiving input from projection neurons in insects are efficient coincidence detectors (Laurent andNaraghi 1994; Perez-Orive et al. 2002). The short temporal integration window is established by two mechanisms. First, intrinsic mechanisms, probably involving voltage-gated Ca2+ and possibly Na+ channels, boost synaptic transients. Second, projection neurons also target a small pool of GABAergic neurons elsewhere in the brain, which in turn provides strong and nonspecific feedforward inhibition onto Kenyon cells. This inhibition arrives at the Kenyon cell dendrite with a delay relative to the excitatory projection neuron input during the same cycle, thereby defining a sharp integration time window. Each Kenyon cell receives input from a small fraction
2.5 %) ofthe projection neuron population. Hence, Kenyon cells in the mushroom body analyze selectively synchronized spiking across an evolving subpopulation of neurons during each oscillation cycle (Laurent, this volume).
In vertebrates, little is known about the temporal integration properties of neurons downstream ofthe olfactory bulb. Moreover, while output from the antennal lobe is conveyed to only two target areas in insects, output from the olfactory bulb is transmitted to at least five different areas in vertebrates. Therefore, it is of prime importance to study the properties of neurons innervated by mitral/tufted cells to understand which of the properties of the temporally structured pattern of activity across olfactory bulb outputs may be relevant for further processing.
The mechanisms of readout by Kenyon cells in the mushroom body suggest that these neurons selectively access information from synchronized spikes, while other spikes are discarded. This does not, however, imply that nonsyn-chronized spikes are irrelevant. They could, for example, play important roles in interglomerular microcircuits within the olfactory bulb/antennal lobe. Furthermore, it is possible that nonsynchronized spikes also convey information that may be retrieved by target neurons with longer integration time constants. Indeed, recent results support the hypothesis that patterns of nonsynchronized spikes convey important information accessible by using a longer integration window (Friedrich et al. 2004). These results suggest that due to the synchronization of subsets of output neurons, different messages maybe multiplexed and conveyed simultaneously to higher brain regions by the across-neuron pattern of spiking in mitral/tufted cells.
The above considerations given to operations performed by olfactory microcircuits on afferent glomerular activity have important implications concerning the representation of stimulus information in the olfactory bulb/antennal lobe:
• In the inputs and the outputs of the olfactory bulb/antennal lobe, odor information conveyed by the response of single elements is limited because of their moderate odor selectivity Rather, stimulus information has to be retrieved from the pattern of activity across many elements (glomeruli or output neurons). Microcircuits, therefore, appear to transform one combinatorial representation into another.
• Currently, it is uncertain whether the position of active units is necessary for the function of olfactory circuits. Theoretically, it is possible that the system relies on positional cues (e.g., during the development of connectivity), even though information is contained purely in the identity of active neurons. An experimental approach to this problem has thus far proven difficult.
• One computation appears to be a decorrelation of activity patterns by the dynamic distribution of activity across output neurons. The underlying synaptic mechanisms are, however, not known precisely.
• The transient synchronization of ensembles of output neurons is an important factor in determining the readout of antennal lobe activity by Kenyon cells in locusts. Synchronization may, therefore, play an important role in the transmission of information from the antennal lobe. Further results, however, are required to understand the role of oscillatory synchronization in other insect species and in vertebrates.
• Available evidence indicates that temporal activity patterns observed in output neurons reflect the dynamic reorganization of instantaneous activity across the population. Theoretically, downstream neurons may also detect the temporal evolution of firing in single neurons or ensembles. However, there is currently no evidence for mechanisms supporting this hypothesis.
These considerations indicate that odor information resides in (a) the identity and instantaneous activity of elements in activity patterns and (b) their synchronization. These features would, therefore, be considered part of the "code." Other properties of odor-evoked activity in the olfactory bulb/antennal lobe, such as the position of glomeruli/neurons or the slow temporal patterning of activity, may not be analyzed directly by downstream targets; further results are needed to resolve this question. If so, they would constitute a "format" of odor-encoding activity patterns. Moreover, they are likely to play essential roles in important computations within the olfactory bulb/antennal lobe that affect the "code," such as the decorrelation of sensory inputs.
FUNCTION OF THE OLFACTORY BULB/ANTENNAL LOBE
Much contemporary research revolves around the general computations performed by microcircuits in the olfactory bulb/antennal lobe. It is, therefore, worth considering the function ofthe olfactory bulb/antennal lobe in more general terms. It is currently contended that in the periphery, odors are represented by distributed patterns of activity across OSNs or glomeruli. Due to overlapping response profiles of odorant receptors, and possibly due to correlations in the world of natural stimuli, these patterns are not evenly distributed within the neural space (in which each dimension represents the activity level of one OSN type/glomerulus). Patterns are instead clustered, complicating the discrimination of individual stimulus representations. This inherent structure in the world of peripheral odor representations leads to the assumption that one function of olfactory processing in the olfactory bulb/antennal lobe is to promote the separation of overlapping odor representations.
In the architecture of interglomerular microcircuits within the olfactory bulb, interactions resulting in inhibition of output neurons are prominent (see above). This fact gave rise to the hypothesis that lateral inhibition in the olfactory bulb/antennal lobe may contribute to odor discrimination. Drawing upon existing knowledge about other systems, the radial and horizontal processing of visual information in the retina has been suggested as a conceptual model. Horizontal cells modify the output of photoreceptors, whereas amacrine cells interact more directly with bipolar/ganglion cells (Shepherd and Greer 1998). In this analogy, the periglomerular cells are synonymous with horizontal cells and granule cells are equivalentto amacrine cells. This analogy leads to the hypothesis that the olfactory bulb/antennal lobe enhances contrast or detects edges in odor-evoked patterns of sensory input by narrowing the response profiles of output neurons as compared to their inputs. Some experimental evidence supporting this hypothesis exists (Yokoi et al. 1995) but this evidence has recently been challenged (Laurent 1999). Moreover, recent analyses of the dynamics and reorganization of activity patterns are not consistent with a simple refinement of afferent "odor images." Despite some similarity in the general layout of circuits, the spatial retinal processing as a conceptual model for system function is, therefore, under debate. In another analogy to the retina, it has been proposed that odor processing in the olfactory bulb/antennal lobe may be more akin to the processing of colors by opponent channels. This proposal is intriguing and receives support from calcium imaging data (Sachse and Galizia, this volume). Due to the complexity of the olfactory stimulus space and the large number of channels (glomeruli), it is difficult to determine whether interglomerular microcircuits in the olfactory bulb establish "odor-opponency channels." Moreover, it does not account for the dynamic properties of olfactory bulb/antennal lobe output.
Another perspective has emerged recently, which views microcircuits in the olfactory bulb/antennal lobe as a (nonlinear) dynamical system (Laurent et al. 2001). The considered class of dynamical systems transforms stationary input patterns into time-varying output patterns, moving along input-specific trajectories in coding space (see Laurent, this volume). In this framework, a primary function of olfactory microcircuits would be to enable odor-specific dynamics that can decorrelate input patterns. Such a system would distribute clustered input patterns more evenly in coding space and, thus, optimize the use of the coding space for discrimination and other tasks. In this framework, the olfactory bulb/antennal lobe would reformat combinatorial representations so as to facilitate their readout. This view is generally consistent with the reorganization of odor-evoked olfactory bulb/antennal lobe output observed experimentally (Friedrich and Laurent 2001, 2004; Laurent 2002; Stopfer et al. 2003). Such a redistribution of activity in coding space would be considered successful (or "optimized") if single downstream neurons could immediately extract any relevant information from it. In other words, after evenly distributing representations in coding space, it should be possible to extract desired information by a simple classifier, such as a support vector machine (Fernandez Galan et al. 2004). Indeed, the extreme specificity of Kenyon cell odor responses in the locust indicates that very specific and high-level information can be extracted from the antennal lobe output in one synaptic step (Laurent, this volume).
An intriguing parallel is apparent between the dynamical systems view of the olfactory bulb/antennal lobe and liquid state machines, which have been proposed as a theoretical framework for the function of cortical circuits (Maass et al. 2002; Maas and Markram, this volume). While it is problematic to apply the liquid state machine model in its generalized form to the specific computations performed by the olfactory system, a more specialized form may lead to valuable theoretical insights into olfactory system function.
In summary, alternative general views of the function of olfactory microcircuits have emerged and are presently being debated vigorously. Common to these views is the notion that the olfactory bulb/antennal lobe does not extract highly specific information by creating outputs tuned very narrowly to particular stimuli or features. Rather, microcircuits appear to reformat odor representations for further use. Precisely how representations are reformatted, and what the use ofthe operations is for further processing, is controversial. According to one view, the output would be a refinement of afferent inputs without the need for dynamics, whereas under the other view, the output would be a fundamental reorganization of activity patterns requiring dynamics. Further research will certainly address these questions. Moreover, many views are motivated by the (perhaps subconscious) assumption that one primary goal ofthe olfactory system is to achieve fine odor discrimination, although olfactory circuits may, in addition, have evolved to achieve other tasks.
Was this article helpful?