In vision and touch, information central to perception is ordered in space in the external world and this order is maintained from the peripheral sense organs to the cortex. Olfactory information, however, does not exhibit a discernible spatial order in the physical world and this poses the question of how odors are represented in the brain. In mammals, olfactory perception is initiated by the recognition of odorant molecules by a large repertoire of receptors in the olfactory sensory epithelium1
. Individual olfactory sensory neurons express one of approximately 1,000 receptors2–4
, and each receptor interacts with multiple odorants. Neurons expressing a given receptor, although randomly distributed within zones of the olfactory epithelium, project with precision to two spatially invariant glomeruli in the olfactory bulb5–7
. Thus, the randomly distributed population of neurons activated by an odorant in the olfactory epithelium is consolidated into a discrete stereotyped map of glomerular activity in the olfactory bulb8,9
How is this highly ordered map of spatially invariant glomeruli represented in the cortex? The projection neurons of the olfactory bulb, mitral and tufted cells, extend an apical dendrite into a single glomerulus and send axons to several telencephalic areas, including a significant input to the piriform cortex and cortical amygdala10–15
. Electrophysiological studies and optical imaging reveal that individual odorants activate subpopulations of neurons distributed across the piriform without spatial preference16,17
. The piriform therefore discards the spatial segregation of the bulb and returns to a highly dispersed organization in which different odorants activate unique ensembles of cortical neurons. However, the patterns of neural activity do not allow us to discern whether mitral and tufted cell projections from a given glomerulus to cortical neurons are segregated or distributed, and whether they are random or determined. Distinguishing between these possibilities is important for understanding odor perception because a random representation of odor identity could accommodate learned olfactory behaviors, but is unlikely to specify innate behaviors. Rather, innate olfactory behaviors are likely to result from the activation of genetically determined, stereotyped neural circuits. We have therefore developed a strategy to trace the projections from identified glomeruli in the olfactory bulb to higher olfactory cortical centers.
Mitral and tufted cells that innervate a single glomerulus were labeled by electroporation of tetramethylrhodamine (TMR) dextran under the guidance of a two photon microscope. This technique labels mitral and tufted cells that innervate a single glomerulus and is sufficiently robust to allow the identification of axon termini within multiple higher order olfactory centers (; ; Suppl. Figs.1–4
). Labeling of glomeruli in the olfactory bulbs of mice that express GFP under the control of specific odorant receptor promoters permits us to examine potential stereotypy of projections from identical glomeruli (MOR 28-IRES-GFP (n = 8), MOR 1–3-IRES-GFP (n= 13), and MOR 174-9-IRES-GFP (n = 10))18
. Labeling of random unidentified glomeruli was performed in mice expressing synapto-pHluorin (OMP-IRES-spH), a fluorescent marker that allows us to visualize individual glomeruli9
. This permits us to sample the projection patterns from multiple different glomeruli ().
Targeted electroporation of tetramethylrhodamine (TMR) dextran labels cells that innervate a single glomerulus in the olfactory bulb
Mitral/tufted cells connected to a single glomerulus exhibit distinct patterns of projections to several areas of the olfactory cortex
Electroporation of a single glomerulus results in the labeling of 6–17 neurons in the mitral cell layer (mean = 9.2 +/− 0.8; ; Suppl. Fig. 1c, 2
; see Methods). We have demonstrated that all mitral and tufted cells labeled in this manner innervate a single glomerulus by electroporating TMR dextran into one glomerulus and fluorescein (FITC) dextran into a neighboring glomerulus (). Examination of the mitral and tufted cells following this two-color electroporation reveals either red or green mitral and tufted cells with only a rare cell labeled simultaneously with the two dyes (1/60 cells, n= 4) (). This labeling strategy therefore restricts incorporation of tracer to mitral and tufted cells innervating only one glomerulus.
We observe that projections from individual glomeruli extend to all major olfactory cortical regions including the accessory olfactory nucleus, piriform cortex, olfactory tubercle, cortical amygdala and lateral entorhinal cortex (, Suppl. Fig. 3
). Visualization of the extent of axonal projections was facilitated by the development of a flattened hemi-brain preparation that enables high-resolution imaging of all olfactory centers except the anterior olfactory nucleus, which is obscured by the overlying lateral olfactory tract. The glomeruli we have examined project to all the major olfactory cortical regions independent of the spatial location of the glomerulus within the olfactory bulb (n=21 different glomeruli in the flattened preparation; see below). Each of the different higher olfactory centers receives a qualitatively unique pattern of input from the olfactory bulb (, Suppl. Fig. 3
). In the piriform cortex a distributive representation is observed, whereas in the amygdala mitral cell projections are broad but spatially segregated.
Mitral and tufted cell axons extend to the piriform cortex via the lateral olfactory tract (LOT). We observe that axonal branches exit the LOT at right angles and extend upward to densely and diffusely project to the piriform cortex along the entire anteroposterior axis (), with no apparent spatial preference in any dimension. High resolution multiphoton imaging reveals varicosities likely to be axonal boutons (Suppl. Fig. 4
). The spatial distribution of these varicosities is similar in every field imaged and is independent of glomerular origin, suggesting that mitral and tufted cell synapses with piriform neurons are distributed throughout the piriform cortex (Suppl. Fig. 5
). The density of these varicosities within the piriform cortex is also similar regardless of the identity of the electroporated glomerulus, further suggesting that each glomerulus makes a similar number of synapses (MOR1–3: 10.2 +/− 0.57 µM of axon per varicosity; M72 (1): 9.9 +/− 0.65 and M72 (2): 10.1 +/− 0.36 µM of axon per varicosity; Suppl. Figs. 4–5
). The highly dispersed pattern of projection to the piriform cortex is observed from every glomerulus examined independent of its identity or location within the olfactory bulb (n = 21; Suppl. Fig. 6
). On visual inspection, the patterns of projection from two identical glomeruli are no more similar than the patterns of projection observed from two different glomeruli.
Projections from single glomeruli to piriform cortex are disperse, homogeneous and indistinguishable
We performed hierarchical and k-means clustering to determine whether the observed patterns of projections from different glomeruli are quantitatively distinguishable (see Methods). We were unable to identify any parameters, including axon fiber positions, density of TMR labeling, center of mass X and Y coordinates and centroid X and Y coordinates, that reliably distinguish the projection patterns from different glomeruli (Suppl. Table 1
). All measured parameters were similar upon comparison of the projection patterns from identical or different glomeruli (Suppl. Fig. 7
Cross-correlation analysis was performed to compare the patterns of piriform projections from different glomeruli (see Methods, Suppl. Fig. 8
for detailed explanation of method and interpretation). The correlograms comparing identical and distinct glomeruli display an extended region of moderate correlation (). These data suggest that the dispersed pattern of projections is largely homogeneous in density over several spatial scales (Suppl. Fig. 9
) and indicate that the patterns are similar for each of the 24 glomeruli we have examined. The similarity of correlograms from identical and different glomeruli provide further evidence that the pattern of piriform projections does not differ for each of the distinct glomerulus types. Thus, the mitral cells innervating an individual glomerulus discard the insular and invariant spatial segregation of the bulb and project dense, dispersed axons to the piriform cortex with no discernible spatial bias.
We next examined the patterns of projections of single glomeruli to the cortical amygdala. The cortical amygdala consists of three nuclei: the anterior cortical, posterolateral cortical, and posteromedial cortical nuclei. The posteromedial cortical nucleus, a major site of innervation from the accessory olfactory bulb19
, receives no discernible input from any of the glomeruli of the main olfactory bulb we have examined (, n = 33 glomeruli in the en face preparation, see Methods), and we observe relatively sparse projections anterior to the posterolateral nucleus. We therefore restrict our analysis to the posterolateral cortical nucleus. The patterns of projection from individual glomeruli in the posterolateral nucleus reveal dense, patchy axonal projections that exhibit a focal nexus surrounded by a less dense halo of fibers (). Despite the diffuse nature of projections, different glomeruli appear to send fibers to anatomically distinct and spatially invariant regions of the posterolateral cortical amygdala (). For example, projections from the MOR 1–3 glomerulus consistently occupy the most medial aspect of the posterolateral nucleus, whereas projections from the MOR28 glomerulus terminate more laterally. These conclusions are evident on visual inspection and are supported by more quantitative analysis.
Projections from single glomeruli to the cortical amygdala are broad, patchy and stereotyped
K-means clustering using relevant parameters extracted from the amygdala projection patterns after image alignment (center of mass X coordinate, X position of medial-most fiber, absolute medial fiber density, ratio of lateral/medial fiber density) was performed on the projections from three identified glomeruli, MOR 1–3 (n=5), MOR 174-9 (n =5), and MOR 28 (n=4). This cluster analysis correctly assigns glomerular identity for 79% of the samples examined (one-way MANOVA, p = 0.0006, α = 0.05).
We have performed normalized cross-correlation analysis to further compare the projection patterns from different glomeruli. Cross-correlation analysis reveals a single peak in the correlogram that reflects the more focal nature of projections to this brain region than in the piriform (). The correlograms between the projection patterns of identical glomeruli reveal peaks that exhibit a small displacement from the center (; Suppl. Figs. 10–11a–c, 12g–h, Suppl. Tables 2 and 3
). Cross-correlation using images of projection patterns from different glomeruli exhibit more varied and often very large displacements (; Suppl. Figs. 10–11d–j and 12i, Suppl. Tables 2 and 3
). These data indicate that the cortical amygdala receives spatially stereotyped projections from individual glomeruli. Although individual glomeruli project to fixed positions, extensive overlap is observed for the projections from different glomeruli. The apparently random pattern of projections in the piriform and the determined pattern in the amygdala are likely to provide the anatomic substrates for distinct olfactory-driven behaviors mediated by these two brain regions.
Insight into the logic of olfactory perception will depend upon an understanding of the how the highly ordered glomerular map is represented in higher olfactory centers. Previous experiments have employed the injection of tracer molecules into the bulb or cortex, to relate the spatial position of projection neurons in the bulb with their targets in higher olfactory centers10–15
. These experiments predate the ability to identify specific glomeruli,6
precluding a determination as to whether projections from a single mitral/tufted cell or a single glomerulus are random or stereotyped.
We have defined a neural circuit that conveys olfactory information from specific glomeruli in the olfactory bulb to the piriform cortex and the cortical amygdala. A distributive representation of neurons in the sensory epithelium is converted into a topographic map in the bulb upon the convergence of like axons onto spatially invariant glomeruli5,6
. The piriform discards this spatial order; axons from individual glomeruli project diffusely to the piriform without apparent spatial preference. Neurons from every glomerulus elaborate similar axonal arbors and quantitative analyses fail to identify features that may distinguish the individual projection patterns. This data is in accord with retrograde tracings using rabies virus that reveal the convergence of multiple, spatially distributed glomeruli on a small number of piriform neurons (Miyamichi et al., this issue). A distributive representation of glomerular projections in piriform is also observed upon anterograde tracing of individual mitral cells following Sindbis virus infection (Ghosh et al., submitted).
Optical imaging and electrophysiological studies of neural responses to odors reflect these anatomic transformations. Distributed neural activity in the sensory epithelium of the nose is transformed in the bulb, with each odor eliciting distinct spatial patterns of glomerular activity8,9
. A second transformation is apparent in the piriform cortex where individual odorants activate unique ensembles of neurons that are distributed without discernible spatial order16,17
. These neurons also exhibit discontinuous receptive fields; neurons within an ensemble responsive to a given odor respond to multiple, structurally dissimilar odors16,20
. The dispersed projections to the piriform provide an anatomic substrate for the generation of these patterns of neural activity.
One model consistent with both the anatomy and physiology invokes the random convergence of excitatory inputs from mitral cells onto piriform neurons such that each piriform neuron would sample a random combination of glomerular inputs. If the connections from bulb to cortex are indeed random, then the representation of the quality of an odorant or its valence in the piriform must be imposed by experience. Odorants, however, can elicit innate behavioral responses, suggesting that a second area of the brain must receive determined inputs from the olfactory bulb. The pattern of projections to the posterolateral amygdala implicate this structure in the generation of innate olfactory-driven behaviors. This suggestion is in accord with the finding that disruption of the amygdala abrogates innate, odor-driven behaviors but leaves learned olfactory responses intact21,22
The olfactory circuits we describe in the mouse are reminiscent of the architecture of the olfactory system in Drosophila23
despite the six hundred million years of evolution that separate the two organisms. In Drosophila
, neurons expressing a given odorant receptor are distributed throughout the antenna and converge on spatially invariant glomeruli in the antennal lobe. Information from the antennal lobe bifurcates with one branch exhibiting spatially invariant projections to the lateral horn, a brain region mediating innate olfactory behaviors. A second branch projects to the mushroom body, a structure required for learned olfactory responses. This anatomic and functional bifurcation provides a context in which to consider the generation of the various forms of olfactory-driven behavior in both flies and mice. Our data suggest that innate olfactory behaviors derive from determined neural circuits selected over evolutionary time, whereas learned behaviors may be mediated by the selection and reinforcement of random ensembles of neurons over the life of an organism.