You are here: Home > Unit on Sensory Coding and Neural Ensembles

Olfactory Coding and Decoding by Ensembles of Neurons

Mark Stopfer, PhD, Head, Unit on Sensory Coding and Neural Ensembles
Stacey Brown Daffron, MS, Technician
Iori Ito, PhD, Postdoctoral Fellow
Joby Joseph, PhD, Postdoctoral Fellow
Rose Chik Ying Ong, MS, Graduate Student
Baranidharan Raman, PhD, Postdoctoral Fellow
Kui Sun, MD, Technician
Nobuaki Tanaka, PhD, Postdoctoral Fellow

All animals need to know what is going on in the world around them; thus, brain mechanisms have evolved to gather and organize sensory information in order to build transient and sometimes enduring internal representations of the environment. Using relatively simple animals and focusing primarily on olfaction, we combine electrophysiological, anatomical, behavioral, and other techniques to examine the ways in which intact neural circuits, driven by sensory stimuli, process information. In the past year, we investigated mechanisms, including transient oscillatory synchronization and slow temporal firing patterns of ensembles of neurons, that underlie information coding and decoding, how spontaneous activity arises in a sensory system, how it is regulated, and how innate sensory preferences are determined. Our work reveals basic mechanisms by which sensory information is transformed, stabilized, and compared as it makes its way through the nervous system.

Sparse odor representation and olfactory learning

In moths, as in other animals, learning experiences readily adjust the meanings of odors. Needless to say, within the brain, the odorants themselves are not matched with conditioning reinforcements; rather, neural representations of odors, presumably spiking activity in olfactory neurons, must undergo the matching process. The mushroom bodies have long been linked with associative learning and memory. In many insects, the bodies are sites of multimodal convergence that include olfactory and gustatory inputs. Many studies indicate that Kenyon cells, the intrinsic neurons of the mushroom bodies, play a critical role in olfactory learning. To understand how neural representations of odors become associated with reinforcement stimuli, we first characterized physiological responses of neurons along the olfactory pathway to odor pulses within the context of an associative learning procedure. The moth Manduca sexta has proved accessible for intracellular recording and is capable of performing the proboscis extension reflex (PER) conditioning, an appetitive olfactory learning task. Therefore, we used the moth to examine neural representations of odor and performed PER training under identical conditions.

We used lengthy odor pulses (4–18s) in our investigations; such pulses, often used in studies of olfactory conditioning, reflect the odor exposures encountered by moths while feeding on flowers. With intracellular and multi-unit recordings, we found that Kenyon cells were almost silent at rest and that odor responses typically consisted of single spikes in a small population of the cells. Interestingly, spiking in Kenyon cells occurred almost entirely at an odor pulse’s onset and sometimes offset, with few spikes in between.

After characterizing the responses of Kenyon cells to the odor stimuli, we used a set of behavioral studies to test a key requirement of a form of Hebbian learning called spike-timing–dependent plasticity (STDP), in which pre- and postsynaptic neurons must both fire spikes nearly simultaneously. To test the relationship between odor-evoked spikes and olfactory learning in Kenyon cells, we used several behavioral procedures with different intervals between odor and reward. Our results indicate that reinforcement delivered many seconds after the conclusion of spiking responses was able to support the formation and recall of associative memory. Thus, the acquisition of short-term memory does not require the concurrence of spikes in Kenyon cells with activation of a reward pathway in the moth. Further, we found that reinforcement delivered after the off-response did not support associative learning.

Our physiological and behavioral studies indicate that spikes in Kenyon cells cannot in and of themselves constitute the odor representation that coincides with appetitive reinforcement. Further, our results reveal that appetitive associative conditioning cannot occur by a Hebbian spike timing–dependent mechanism alone within Kenyon cells. We suggest instead that the odor representation in Kenyon cells that is paired with reward may be a sustained biochemical, perhaps second-messenger, response triggered by highly transient spiking or perhaps even subthreshold activity.

Synaptic learning rules and sparse coding in a model sensory system

Understanding how the brain encodes, processes, transforms, and stores sensory information is a fundamental issue in systems neuroscience. An important question is how several mechanisms such as neural oscillations, synchrony, population coding, and sparseness interact in the process of transforming and transferring information. Another question is how synaptic plasticity interacts efficiently with the various coding strategies to support learning and information storage. We approached these questions, which are rarely accessible to direct experimental investigation, by combining electrophysiological recordings with computational modeling of the olfactory system of the locust.

The transformation of odor responses from the antennal lobe to the mushroom body requires a number of coordinated processes, including an encoding function (the synchronization of groups of projection neurons through oscillatory dynamics) and a decoding function (the ability of Kenyon cells to respond as coincidence detectors, activated only by correlated input from the antennal lobe). For the transformation of odor responses to operate efficiently, encoding and decoding processes must be properly matched across a broad range of conditions. We are investigating the mechanisms that underlie the match and permit the appropriate selection of coherent signals.

We developed a biologically plausible model to test whether synaptic plasticity could control and tune synaptic weights of inputs to the mushroom body. Further, the model tested the relative advantages of various learning rules based on either spike rate– or spike timing–dependent induction of synaptic plasticity. We found that plasticity at the input afferents to the mushroom body can efficiently mediate the tuning necessary for selectively filtering intense sensory input arriving at the mushroom body. More specifically, we found that spike timing–dependent plasticity was more efficient than rate-dependent plasticity. Our results suggest a general mechanism for how plasticity could efficiently promote sparse representations in olfactory and other sensory systems, such as the visual system. More broadly, our findings illustrate a potential central role for plasticity in the efficient transfer of information between brain areas employing different coding strategies within neural systems.

Odor-evoked neural oscillations in Drosophila are mediated by widely branching interneurons

Stimulus-evoked neural oscillatory synchronization is commonly observed in a wide and diverse range of species. However, important questions about how oscillations are generated, and what functions they serve, remain unanswered. Drosophila offers a great and growing variety of powerful, experimental genetic tools that permit the labeling and functional manipulation of specific classes of neurons while providing many advantages for analyzing the structure and functions of olfactory circuitry. In addition, given that Drosophila has become an important model system worldwide for studying olfaction and mechanisms underlying neural function, we investigated odor-elicited oscillations in Drosophila.

We found that odors evoked periodic oscillations similar to those we previously observed in locusts, honeybees, cockroaches, wasps, and moths. A wide assortment of odorants could elicit oscillatory responses at around 10Hz local field potential (LFP), including dilute monomolecular chemicals such as hexanol, natural fly attractants such as ripe banana, and the yeast paste that serves as fly food—all delivered at normal concentrations. Puffs of clean air elicited no oscillations.

To determine the origins of the LFP oscillations that we had recorded in the calyx of the mushroom body, we made paired recordings from the brains of intact, healthy flies and simultaneously monitored the odor-elicited LFP from the mushroom body calyx and the intracellular responses from the PNs that supply most of the olfactory input to the mushroom body. Our recordings routinely revealed subthreshold membrane potential oscillations correlated with the LFP recorded simultaneously in the mushroom body. We also noted that odor-elicited spikes in both projection neurons and inhibitory local neurons were tightly phase-locked to LFP, supporting the idea that the oscillations were generated in the antennal lobe and then transferred to the mushroom body calyx by projection neurons. Consistent with this observation, the oscillations could be reversibly abolished by applying the GABA_A blocker picrotoxin, which interferes with the output of local neurons.

To examine the roles of the local neurons with unprecedented precision, we used genetic tools to target specifically and manipulate different types of local neurons. We found two classes of local neuron. One class has widely branching processes and is identified with the GAL4 strain NP2426; the other has narrow branch patterns and is identified with the GAL4 strain NP1227. We next analyzed the extent to which the two classes of local neuron contribute to the generation of odor-elicited oscillations. To gain control over the output of the neurons, we generated lines of flies expressing the shibire^ts1 (shi) gene in each type of local neuron. The shi gene encodes a temperature-sensitive dynamin mutant protein that can conditionally and reversibly block chemical synaptic transmission at a restrictive temperature.

We found that flies expressing shi either in widely branching neurons alone or in both wide- and narrow-branch neurons showed a significantly greater temperature-sensitive decrease in oscillatory power than that in the wild type; we observed no significant differences between the narrow- and combined narrow- and wide-branching strains. The results indicate that only the widely branching local neurons contribute to the generation of odor-elicited oscillations.

Together, our results establish that Drosophila uses an oscillatory synchronization mechanism as part of its responses to odors. Further, the results reveal, with unprecedented precision, the mechanism underlying the oscillations. It will now be of great interest to use these new insights and tools to examine the functions of neural oscillations in odor coding and decision making.

Temporally diverse firing patterns in olfactory receptor neurons underlie neural codes for odors

Odorants are represented as spatio-temporal patterns of spiking in the antennal lobes of insects and the olfactory bulbs of vertebrates. These odor-evoked ensemble responses, which contain information about odor identity, intensity, and timing, are reliable over repeated trials and are sometimes remarkably elaborate, consisting of sequences of excitation and inhibition that together can outlast the eliciting odor stimulus. These patterns change most rapidly during the odor’s onset and offset. During the middle portions of lengthy odor presentations, the firings of populations of principal neurons tend to settle into stable patterns that have been described, in the language of dynamical systems analysis, as a “fixed point.” These activity patterns provide all the information that the brain receives about odors in the environment. The question is how the neural codes for odors are generated.

We hypothesized that the codes depend on a diversity of output from olfactory receptor neurons (ORN). To examine the characteristics of the input from the antenna to the antennal lobe, we first produced electroantennograms (EAG) from isolated locust antennae; the EAG provides an assay of total ORN output. EAG responses evoked by different odorants all showed relatively similar time courses. Earlier models of antennal lobe function were based on input similar to these EAG patterns. However, given that the EAG sums population activity, it obscures any odor-specific temporal structure contributed by individual receptor neurons. To characterize the responses of individual ORNs that underlie the EAG response (and that provide olfactory input to the antennal lobe), we made the first systematic recordings from ORNs in the locust. We found that the responses of individual ORNs showed a surprising diversity of temporal structure, including periods of inhibition and sequences of excitation and inhibition. Viewed as a population, the response characteristics of the ORNs were in many ways remarkably similar to those observed downstream in the responses of the well-described projection neurons. Thus, many of the characteristics of odor codes thought to arise in the antennal lobe actually arise earlier in the periphery and include odor-elicited spatio-temporal patterning of the principal neuron activity, the decoupling of odor identity from intensity, and the formation of fixed points for long odor pulses. The question arises as to the roles these complex peripheral responses might play in establishing neural codes for odors in the antennal lobe.

To examine systematically the significance of the diverse odor responses in the ORNs, we constructed a two-part computational model. The first part simulated a population of ORN odor responses while the second part realistically modeled the responses of the antennal lobe circuitry to the diversity of input. We found that our combined model generated responses that accurately matched the complexity, duration, and fixed-point characteristics of spatiotemporal odor codes recorded from the antennal lobe.

Further, given the surprising diversity and complexity of input provided by the receptor neurons, we considered what additional information processing roles the antennal lobe might play. With our combined computational model, we found that the network dynamics of the antennal lobe impose oscillatory synchronization, transform the spatio-temporal input from ORNs into a higher-dimensional representation, and evenly redistribute odor codes to make better use of coding capacity. In several instances, our results demonstrated that apparently complex forms of information processing described in other systems may be explained by simple interactions between receptor neurons and their followers.

In view of the strong structural and functional parallels between insect and vertebrate olfactory systems, it seems likely that similar mechanisms operate across species. We therefore look forward to determining how ORNs generate diverse temporal responses to odors.

Spontaneous odor receptor neuron activity determines follower cell response properties

Noisy or spontaneous activity poses a challenge to neural systems. Activity in the absence of obvious stimuli occurs throughout the central and peripheral nervous systems. Such spontaneous activity has been shown to play several useful roles; in some situations, however, it may not obviously benefit the organism and instead imposes limits on perception and behavior. We used the locust olfactory system to investigate fundamental properties of spontaneous activity at points along the sensory pathway. In insects, ORNs are distributed along the antenna. Our recordings from these neurons and their immediate and more distant followers revealed high levels of spontaneous activity in the ORNs themselves, local neurons, and projection neurons, but little spontaneous activity in Kenyon cells. Indeed, in a variety of species, spontaneous activity has been observed in ORNs and their immediate follower neurons (e.g., mitral cells in vertebrates); however, neurons one synapse farther (e.g., cortical neurons in vertebrates), such as Kenyon cells in insects, are typically almost silent at rest.

Where and how does this spontaneous activity originate? What rules govern its propagation from one group of neurons to the next, and how and why is such activity sharply limited two steps removed from the ORNs? What effects does spontaneous activity exert on olfactory coding? The locust olfactory system, where sensilla-containing ORNs are located both externally and accessibly, provided us with the opportunity to answer these questions directly.

We found that we could reversibly silence the ORNs by cooling them and that such silencing nearly abolished spontaneous and odor-elicited spiking in the projection neurons as well. Our investigation resolved a long-standing issue: spontaneous activity does not arise within the circuitry of the antennal lobe but rather is inherited entirely from the output of the ORNs. With a series of whole-cell patch-clamp recordings, we found that silencing the ORNs also significantly decreased the resting membrane potentials of projection neurons, local neurons, and Kenyon cells, indicating that spontaneous activity originating in the odor receptors exerts a constant influence on the response thresholds of follower cells.

To investigate the causes of spontaneous activity in ORNs, we controlled the delivery of odorant and the purity of air surrounding the antenna. In insects, as in vertebrates, absorbent fluid lymph surrounds ORNs. We found that isolating the antenna from its environment by coating it with an oily substance such as Vaseline® had little impact on spontaneous ORN output. However, manipulations designed to accelerate the removal of odorants from the sensillar lymph (surrounding the antenna with ultra-purified bottled air or directing high-speed streams of room air toward the antenna) significantly reduced the amount of output from the ORNs. Taken together, our results show that odorants or other ligands lingering in the sensillar lymph trigger spontaneous activity in the olfactory system. Environmental odorants continually presented to the antenna appear to play only a small role. Events within the ORNs’ transduction machinery may make additional contributions to spontaneous activity.

To understand why spontaneous activity originating in ORNs is passed to the secondary projection neurons but is then sharply attenuated before reaching tertiary Kenyon cells, we developed a simple receiving operator characteristic (ROC) model to simulate the success of Kenyon cells in discriminating signal from noise, when presented with varying degrees of input convergence and ORN signal strength and various threshold set points. We found that, given the ongoing barrage of activity from ORNs, odor detection is optimal when projection neurons have a low response threshold; Kenyon cells, however, have a high response threshold. Our exploration of noise sources in the locust olfactory system provides a specific example of how a sensory system, bombarded with noise at the first stage of processing, balances the competing challenges of maintaining sensitivity to a wide range of stimuli and setting thresholds to eliminate noise and sparsen neural codes. We expect that such strategies apply to other sensory systems that employ several stages of processing and circuitry convergence to achieve optimal detection.

Sensory neuron responses determine innate olfactory behaviors in the locust

How do the sensory capacities of animals develop? Through their innate sensory preferences, animals often demonstrate the existence of inborn information. We are investigating how such information is encoded and how it differs from information acquired through direct experience. To conduct our studies, we use hatchling locusts, which offer several experimental advantages.

We found that newly hatched locusts, literally just crawling out of their egg hatching cups, immediately move toward fresh grass. In a series of behavioral studies using thousands of locusts, tested individually or in groups, we established that the hatchlings choose real grass over visually similar but odorless plastic grass; paper rubbed with fresh grass over clean paper of the same color; and paper dabbed with colorless monomolecular odorants that are components of grass odor over paper with other colorless odorants (even when the odorants were diluted to have identical vapor pressures). The results indicate that naive locusts, which had never eaten, touched, or otherwise encountered their natural food source, have a built-in preference for its odor.

Such preference could be attributed to a peripheral mechanism; for example, hatchling locusts could have a surplus of odor receptors for grass odors. We tested such a hypothesis by making electroantennograms from hatchlings and found that the antennae generate especially strong signals when stimulated by hexanol and octanol, which are volatile chemicals released by many green plants. Other odorants generated weaker responses. The same result obtained when the various odors were all diluted to provide equal vapor pressure, indicating either that the hatchling and adult antennae contain a surfeit of hexanol and octanol receptors or that each hexanol and octanol receptor provides a particularly strong output. In addition, we found that, in hatchling antennae, sensory adaptation occurs for grass odors with the same timing and extent as for non-plant odors. Our results strongly suggest that the locust’s innate sensory preference for grass odors is encoded peripherally, in the antenna. We also found that the response profiles of adult locust antennae are virtually identical to those of the new hatchlings, indicating that no experience or “tuning” is required to prepare locusts to detect grass odors preferentially.

Frequency shifts reveal basic mechanisms for odor-evoked neural oscillations

Accumulating evidence suggests that in many animals, from insects to mammals, olfactory information is represented by the temporally structured, synchronized firing of a spatially distributed population of neurons in the olfactory systems. We found that a wide range of odorants and stimulus pulse durations all evoked clear oscillatory activity in the moth Manduca sexta. Interestingly, relatively brief, plume-like pulses (shorter than 750 ms) produced only fast oscillations (30–40 Hz), whereas long pulses (longer than 1 s), such as those that effectively induce associative learning, produced an oscillatory burst that was initially fast and then slow (10–20 Hz). These results provide the first clear evidence that moths, like locusts, produce oscillatory responses to odors. These findings also provide the first indication that (1) brief and lengthy presentations of the same odorant elicit somewhat different types of neural responses and (2) the oscillatory mechanism can operate stably in fast and slow modes.

What determines the oscillation frequency of a circuit, and how can a single neural circuit operate stably under two frequency regimes? We found that, in the moth, a lengthy odor pulse elicits an EAG deflection that, over the response, decreases in amplitude because of adaptation in ORNs. We also found that oscillation frequency roughly tracked EAG amplitude, suggesting that frequency may depend on the intensity of input to the circuitry of the antennal lobe.

However, we also discovered that, over a wide range of odor concentrations, the initial odor-elicited oscillation frequency was invariant, whereas EAG amplitudes varied greatly with concentration. Thus, some evidence suggested that oscillation frequency was dependent on input intensity, but other evidence suggested the opposite. To explore this apparent contradiction, we developed a computational model of the moth antennal lobe that mimics the sharp transition between discrete fast and slow oscillatory states when input intensity gradually decreases. The recruitment of additional but less well-tuned ORNs to simulate responses to higher concentrations did not affect oscillation frequency. Our recordings from ORNs showed that long odor pulses caused most individual ORNs to adapt their firing rates rapidly; the firing-rate change closely matched the time course of the shift in oscillation frequency. Our results suggest that oscillation frequency can shift between two stable states, depending on the varying output intensity of adapting receptors rather than on odor concentration. Our model indicates that such a shift in oscillation frequency is possible if the ORNs that are highly tuned for a given odor fire at near-saturating rates, even when presented with low odor concentrations; higher concentrations recruit additional but less tuned ORNs. The model also revealed that oscillation frequency is regulated flexibly by the intensity of input rather than rigidly by the duration of the inhibitory post-synaptic potential from inhibitory neurons, as is often assumed.

Several mechanisms extract features from natural odor stimuli

As information moves through the brain, it undergoes dramatic transformation in myriad ways. Our previous work suggested that one of the general mechanisms responsible for such transformation is neural plasticity. To investigate mechanisms underlying the transformations, we delivered rapid, repeated pulses of odors with timing designed to mimic features of natural plumes and monitored, in intact animals, neural activity in several locations: olfactory receptor neurons, ensembles of projection and local first-order interneurons of the antennal lobe (analogous to the olfactory bulb), and the second-order Kenyon cells of the mushroom body (analogous to the pyriform cortex). At each location, we sought to understand responses in terms of the interactions of plasticity occurring at earlier sites. We also sought to understand the transformations’ potential value to the animal.

We found that interneuronal responses to natural odor stimuli are shaped by at least two plastic mechanisms: rapid adaptation in the receptors and relatively enduring facilitation of inhibition within the receptors’ downstream targets. Peripheral adaptation renders the olfactory system relatively insensitive to stimuli that repeat very rapidly. Central facilitation of inhibition increases the reliability and sparseness of stimuli that are encountered repeatedly but are separated by long intervals. Further, these mechanisms constrain the projection neuron ensemble in order to provide relatively stable output to its downstream followers, thereby allowing the encoding of information about odor identity and concentration with firing patterns that are not confounded by the timing patterns of the stimulus.

How do the follower neurons decode this time-varying ensemble activity? Intracellular and extracellular recordings from Kenyon cells showed that the cells’ firing rates change dramatically throughout trains of odor pulses in a timing-dependent manner; for brief inter-pulse intervals, the great majority of action potentials fire at the beginning of the train and again following the train’s conclusion. We found that the Kenyon cells’ firing threshold can be met when projection neurons fire at rates that are relatively low but with spikes that are highly synchronized across the population by the oscillatory mechanism of the antennal lobe (as occurs during the onset of the pulse train). On the other hand, the threshold can be met when the instantaneous firing rate of the projection neuron ensemble is high in the absence of pronounced synchronization (as occurs following the offset of the train).

Together, our work suggests that the non-associative plasticity elicited by odor plumes leads to responses in the projection neuron ensemble that combine an instantaneous report of sensory input with a record of recent input, allowing the extraction of high-level features.

Additional Funding

JSPS fellowships to I.I. and N.T.
NRC fellowship to B.R.

Publications

Ito I, Ong RC, Raman B, Stopfer M. Sparse odor representation and olfactory learning. Nat Neurosci 2008 11:1177-1184.
Finelli LA, Haney S, Bazhenov M, Stopfer M, Sejnowski TJ. Synaptic learning rules and sparse coding in a model sensory system. PLoS Comput Biol 2008 4:e1000062.
Tanaka NK, Ito K, Stopfer M. Odor-evoked neural oscillations in Drosophila are mediated by widely branching interneurons. J Neurosci 2009 29:8595-8603.
Ito I, Bazhenov M, Ong R, Raman B, Stopfer M. Frequency transitions in odor-evoked neural oscillations. Neuron 2009, in press.

Collaborators

Maxim Bazhenov, PhD, Howard Hughes Medical Institute, The Salk Institute for Biological Studies, La Jolla, CA
Luca A. Finelli, PhD, Howard Hughes Medical Institute, The Salk Institute for Biological Studies, La Jolla, CA
Kei Ito, PhD, Institute of Molecular and Cellular Biosciences, University of Tokyo, Tokyo, Japan
Terrence J. Sejnowski, PhD, Howard Hughes Medical Institute, The Salk Institute for Biological Studies, La Jolla, CA

Contact

For more information, email stopferm@mail.nih.gov or visit http://neuroscience.nih.gov/Lab.asp?Org_ID=491.

Top of Page