There are two levels of processing in the olfactory bulb. The first, described in chapter 9, consists of the glomerular layer, which forms an image representing the smell molecules and performs signal-to-noise operations and lateral interactions to begin the processing of the image. The image is then sent to the second level within the olfactory bulb. The connection between levels is by means of a large cell, the mitral cell, and its smaller companion, called a tufted cell (see figure 7.1). These cells collect the input in their dendritic branches in the glomerulus, transfer the processed signal to the next level through a long primary dendrite, and send it out through their long axon to the olfactory cortex. But before it goes out, the image is subjected to an additional stage of processing. This is necessary because the glomerular interneurons have had their actions on the odor image limited by their axon connections to specific surrounding glomeruli.
At least two operations are required before the image can be sent further. First, there must be coordination with all other glomerular modules. This takes place through long so-called secondary dendrites of the mitral and tufted cells, which reach out sideways for long distances, branching and terminating across many neighboring glomerular modules. These dendrites do not interact with each other, but rather with the special type of interneuron called a granule cell (see figure 7.1). Through it, the second operation—lateral inhibition between coordinated glomerular modules—is carried out. There are about a hundred granule cells for every mitral cell, so this is strong inhibition. The granule cell is thus a key to this coordinated inhibitory processing, whose function is to format the odor image for output to the next stage, the olfactory cortex. How does it do it?
Solving the Granule Cell Problem
I first encountered the granule cell as a graduate student with Charles Phillips in Oxford when I was studying the physiological responses of cells in the olfactory bulb. My main finding, simultaneous with the same finding in two other laboratories, was that mitral cells are subjected to very strong and long-lasting lateral inhibition.
But there were no known inhibitory cells in the olfactory bulb at the level of the mitral cell secondary dendrites, only this curious little granule cell with prickly looking spine-covered dendrites, and no axon. Without an axon one could not even be certain that it was a nerve cell, much less one that could be an inhibitory interneuron for the mitral cells. Nonetheless, it seemed to be in the right place, so we suggested that it could be activated by side branches (collaterals) of the mitral cell axon and inhibit the mitral cell through its long dendrite mingled among the secondary dendrites.
The next step in my training consisted of postdoctoral studies with Wilfrid Rall at the National Institutes of Health (NIH) in Bethesda, Maryland. Rall was establishing himself as the pioneer of computational neuroscience, constructing the first computer models of nerve cells to elucidate the functions of their enigmatic dendritic branches. He had to endure much opposition from those who believed that the dendrites were not involved in information processing but were there primarily for nutritional support. We decided that the mitral and granule cells of the olfactory bulb were excellent subjects for his approach to support his theory of the importance of dendrites and my theory that they were important for processing smells.
Our aim was to build computational models of the mitral cell and the granule cell to test whether we could account for my experimental recordings. Almost two years of work produced a model for impulse spread in the mitral cell dendrites, excitation of the granule cells, and inhibition of the mitral cells. Unfortunately the model didn’t give any new insights into how the two kinds of cells might actually interact. I had another couple of months before leaving for my next position. The inhibitory action of the granule cell on the mitral cell seemed clear enough, but how was it activated to begin with? The more we struggled with the problem, the more the constraints of the model indicated that the excitation of the granule cell dendrites must occur at the same narrow level as the subsequent inhibition of mitral cell dendrites. But how?
More “Aha!” Moments
As we discussed this problem one afternoon, the idea hit us that the excitation of the granule cell dendrites must come from the same mitral cell dendrites that the granule cell dendrites then inhibit. It sounded crazy, but it seemed the only solution. I knew from the classical anatomical literature that there was no precedent for this kind of interconnection between dendrites, and we both knew there was no precedent in the physiological literature for this kind of functional interaction. So Rall duly wrote down our hypothesis in his green protocol book (on August 26, 1964, to be exact): “dendrodendritic interactions” between mitral and granule cells likely occur, and these would function to cause self and lateral inhibition of the mitral cells. We suggested that this would be similar to the lateral inhibition in the retina. I then went on to a further research fellowship in Sweden.
The only sure way to test this hypothesis was with the electron microscope, which would enable one to prove that such synapses existed. As it happened, Tom Reese and Milton Brightman were working at NIH in a nearby building. I had encouraged them to look for synapses between the mitral and granule cell dendrites, and they soon found them. On being shown that the synapses were unusual in being situated side by side and oriented in opposite directions, Rall had another “aha!” moment. He immediately told Reese and Brightman that these were precisely the kinds of contacts to mediate the interactions we had postulated. It was an example of how perplexing data can be interpreted instantly by the prepared mind.
My next “aha!” moment came in Stockholm on receiving the letter with this news. Two other investigators reported the “atypical configurations” in the olfactory bulb, but without the physiology and the model it wasn’t possible to infer a function for the synapses because they seemed to be in direct opposition to each other.
Shaping Spatial and Temporal Processing
The four of us excitedly wrote up our initial results and submitted them to the leading journal Science. The reviews came back: “Rejected; not of general interest.” We could have fought against the rejection, as is the custom these days, but Wil was too courteous for that. He found, instead, another journal that would take our paper.
We subsequently published the full details of the model in a second paper. This made explicit the way the self-inhibitory interactions always proceed in a sequence from excitation of a granule cell to feedback inhibition, so that there is no opposition, whereas lateral inhibition is mediated only by the granule-to-mitral synapse onto a neighboring mitral cell. This provided a new type of circuit, for mediating more localized self and lateral inhibition compared with the classical pathway. We pointed out that the lateral inhibition was likely to be important in the spatial organization of activity in the olfactory bulb as well as contribute to oscillatory activity in the mitral-granule cell populations. Thus, the same interactions are at the core of both the spatial and the temporal properties involved in processing the odor images.
Later, in an article for Scientific American, I termed these and other interactions like them microcircuits, in analogy with microcircuits in computers. The term has caught on and become useful in describing the organization of the nervous system in terms of distinct and repeatable patterns of connections, for which the olfactory bulb is still one of the best models. It has become an example of how a theoretical model can predict a circuit with a specific function. After more than 40 years of testing, it continues to be helpful in guiding experiments on the microcircuits for odor processing in the olfactory bulb.
How Multiple Glomerular Units Are Coordinated
How are the widely distributed glomerular modules coordinated in processing the image? We have seen that the process begins in the localized connections of periglomerular cells between glomeruli. We now need to see how it is completed by the lateral inhibitory connections through the granule cells.
The problem is this: the odor image extends broadly within the olfactory glomerular layer, even when aroused by a single odor molecule (chapter 8). It requires that the responses of mitral and tufted cells that may be far apart need to be coordinated so that lateral inhibition can occur to enhance the image. How do we get effective lateral inhibition over long as well as short distances?
An answer to that question has come from a new method based on the virus that causes rabies. As is well known, the rabies virus enters nerve cells and is transported throughout their branches, killing them, and also passing into cells to which they are connected in order to kill them as well. Molecular biologists have taken advantage of this property and turned it into a research tool for understanding how nerve cells are connected—by modifying the virus into a “pseudorabies” virus so that it has a much reduced lethality. By attaching a fluorescent label to this virus, its progress through the cell and into other cells can be traced.
David Willhite, a postdoctoral student in our laboratory, used this tool to trace the connections between mitral and granule cells. This showed that an infected mitral cell is connected to a local cluster of granule cells, mitral cells, tufted cells, and periglomerular cells that are all related to a single glomerulus. The cells form a column, somewhat similar to columns of cells that have been seen in areas of the cerebral cortex, for processing the glomerular signal. We have termed this a glomerular unit. Willhite finds that a given mitral cell has connections to other glomerular units that are widely distributed through the olfactory bulb. It appears that this could be the basis for the hypothesized coordinated lateral inhibition.
How can one mitral cell impose strong lateral inhibition on mitral cells of different glomerular units over variable distances between them? A key was provided by two other colleagues, Wen Hui Xiong and Wei Chen, who showed that the impulse generated in the mitral cell body not only travels out into the axon and on to the olfactory cortex, but also travels backward into the lateral dendrites all the way to their tips. This means that the dendrite can act somewhat like an axon in spreading impulses long distances to activate granule cells and their lateral inhibition throughout the olfactory bulb.
Another colleague, Michele Migliore, constructed a computational model in which the impulse activates granule cells belonging to glomerular units at variable distances, thus providing the strong inhibition at arbitrary distances required for processing the distributed odor images. It was the first major change to our original model in 40 years. And just as with the first model, it was rejected as being “not of general significance” the first time around, before eventually being accepted by another journal.
These inhibitory operations not only provided for enhancement of the spatial images, but also for synchronization of the activated mitral cells through inhibitory gating of the mitral cell responses, as predicted in the original model (the odor pattern is therefore one that exists in both space and time). There is much to test in this new proposal, but it provides a working hypothesis that extends the original model and can guide experiments in the future.
The reader will note that this new view has come about from the coordinated research of several groups. This illustrates several important features of doing basic research: the need for stable funding over a sufficient period of time; the need for a critical mass of investigators working on different aspects of the same problem; and the advantage of coordinating computational with experimental approaches. These were behind our success in the original discovery of the dendrodendritic synapses and their mechanisms, and they were behind the recent advance in understanding their coordinated action among the distributed glomerular units.
Modulating Our Appetites
In addition to providing for this important step in processing the sensory input, the granule cells are also a key site of modulation by the behavioral state of the animal. By behavioral state, we mean whether the animal is awake or sleeping, or whether it is hungry or satiated. The olfactory bulb is obviously an important step with respect to the brain mechanisms for feeding; if we are hungry, the smell of food (either the aroma or the flavor) really stimulates our appetite; when we are full, the smell is much less attractive, perhaps even aversive. This modulation is obviously critical for the brain flavor system. It turns out that processing in the olfactory pathway is heavily dependent on whether we are hungry or full, and that this starts in the olfactory bulb itself.
This was first shown in 1978 by a French scientist, Jeanne Pager, who recorded from mitral cells in rats under these two conditions and found that they showed brisk responses if the animals were hungry and weak responses if they had eaten. The most likely interpretation is that this reflects the activity of central sites in the brain that are influenced by feeding signals, such as blood sugar and intestinal sensory fibers. The granule cells receive heavy inputs from these central sites (particularly the site known as the locus ceruleus in the brain stem and the diagonal band in the forebrain). These outwardly directed “centrifugal” fibers are indicated in figure 7.1, coming from the brain stem and from the nucleus of the horizontal limb of the diagonal band (NHLDB) in the forebrain. It has been hypothesized that activity in these fibers functions as a set point for gating the flow of sensory information about smell through the olfactory pathway in relation to the feeding status of the animal. They are increasingly busy modulating retronasal smell patterns through the granule cells and the glomeruli as you proceed through your dinner and go from feeling hungry to feeling full.
Similar modulatory actions take place at every stage deeper in the brain. They make the olfactory pathway probably the most heavily modulated pathway in the brain. The reason for this appears to be that our perception of food smells is heavily dependent on our behavioral state: whether we are hungry or full, angry or sad, craving for something or repulsed by it, suspicious of a new food (called neophobia) or eager for novelty (neophilia). The smell image has to be modified by the behavioral state. The next chapter will discuss further how appetite modulates a central brain region involved in flavor.