NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Menini A, editor. The Neurobiology of Olfaction. Boca Raton (FL): CRC Press/Taylor & Francis; 2010.

Cover of The Neurobiology of Olfaction

The Neurobiology of Olfaction.

Show details

Chapter 13Temporal Coding in Olfaction

, , and .


Knowledge about the molecular organization principles of the sense of smell in different species has greatly improved in the last decade (Bargmann 2006; Mombaerts 2004a, 2004b; Rodriguez 2007). It is now well established that in many species, odorant molecules are detected by large families of G-protein-coupled receptors (Buck and Axel 1991), whose molecular sequence and structure may vary across species and phyla, but that essentially implement the same function (Bargmann 2006). Interestingly, the insect olfactory receptors display a unique and unconventional membrane topology in comparison to the mammalian receptors, questioning the existence of a coupling with G-proteins (Benton et al. 2006; Vosshall and Stocker 2007). Nevertheless, understanding how odorant information generated by these large arrays of receptors is interpreted by the brain to produce a great variety of behaviors will be the challenge of the next decade. A few questions, which may appear basic with regard to the complexity of the entire olfactory system, are still not answered. Among these, how olfactory information is encoded in brain networks downstream to receptors, remains poorly understood. In recent years, there have been strong debates on this question and it seems that the answer is not as simple as recording from the neurons of these networks. The ambition of this chapter is not to provide a definitive answer, but to present the most relevant results on this question and put them in perspective, helping the reader to appreciate where the field stands in terms of olfactory coding. Since a certain similarity in the olfactory system organization has been observed across species (Kay and Stopfer 2006), we will endeavor to compare between different animal models. Our focus will primarily be on temporal coding, as temporal dynamics, in our opinion, are currently the main aspect of neuronal activity in the olfactory system that is difficult to integrate in a convincing and unanimously recognized theory of olfactory coding.

In most, if not all species, the olfactory system has a first stage downstream to receptor neurons, where sensory axons converge in a receptor-specific fashion onto projection neurons (PNs) dendritic tuft, forming segregated anatomical structures called glomeruli. These downstream structures are found in the main olfactory bulb (OB) in vertebrates or the antennal lobe (AL) in insects, which are considered as functionally analogous in many studies (Kay and Stopfer 2006). So, the first step into the system goes with a clean separation of the different information channels. The information then goes forward via the output neurons of the OB (mitral and tufted cells) or the AL (PNs) to several downstream areas. At this level, the system starts to diverge into the brain. However, in vertebrates as in insects, one of these areas receives more massive projections and is considered to have a more central role than others. This would be the piriform cortex in mammals or the mushroom body (MB) in insects. Most of the present debates about olfactory coding focus on the first stage (bulb or AL), wondering how the circuitry of these networks might transform the spatially segregated sensory input. An increasing number of studies also started to address the question of coding in the main target areas of the OB and AL. Both levels will be reviewed in this chapter.

The first basis of olfactory coding corresponds to the fact that the large receptor repertoire is expressed in a very controlled manner, as one sensory neuron usually expresses only a single olfactory receptor (Bargmann 2006). In most cases, receptors are sensitive to many chemical compounds and have overlapping receptive fields (Firestein 2001; Hallem and Carlson 2006). This suggests that odor identity is represented by complex combinations of receptor activations rather than by the activity of a specific receptor. However, there are also some cases of highly selective receptors. For example, the two receptors, Gr21a and Gr63a, expressed in the Drosophila olfactory epithelium, uniquely respond to CO2 (Jones 2007; Kwon et al. 2007) and drive innate avoidance behavior (Suh et al. 2004, 2007). In mice, the existence of a “specialist glomerulus” narrowly tuned to a compound present in urine was reported (Lin et al. 2005), though the receptor may still be activated by other nontested chemicals. The concentration seems to be a factor modulating the sensitivity range. With higher concentration of an odorant, more receptors become activated (Hallem and Carlson 2006). Hence, the initial olfactory code is sparser at low than at high concentrations.

This discrepancy between specialists and generalists olfactory receptors is the subject of another debate. Some researchers defend a view of the olfactory system where activity of an olfactory receptor is transmitted further in a dedicated “labeled line” pathway, which receives only very limited or no interactions with other pathways. This idea is supported by the observation that, in some animals such as mammals or Drosophila, second order neurons (mitral cells, PNs) receive sensory inputs from a single receptor type. Others oppose that at each level of the olfactory system, olfactory information is largely distributed in the network, in part due to lateral interactions between neurons. This idea is supported by the existence of lateral, multisynaptic connections between second order neurons both in the OB (Shepherd 1972) and the AL (Olsen et al. 2007; Olsen and Wilson 2008). The answer to this debate is probably that both schemes coexist. It was recently shown in Drosophila that among two different narrowly tuned olfactory receptors, one was connected to a very specific downstream neuron of the AL, suggesting a “labeled line” pathway, and the other was connected to a broadly tuned neuron, suggesting contribution to a distributed code (Schlief and Wilson 2007). This outcome may be only partially surprising if one considers that the olfactory system might be involved both in simple, stereotyped, and eventually innate behaviors, as well as in more complex behaviors learned throughout life. “Labeled line” circuits might be present to fulfill simple but important functions, such as carbon dioxide detection in insects (Jones 2007). An innate avoidance circuit, only relying on the dorsal glomeruli of the OB, has recently been described in mice (Kobayakawa et al. 2007). If these glomeruli are genetically inactivated, mice can still learn avoidance to the innately repulsive odors (i.e., can still somehow recognize them), but do not show the innate avoidance behavior anymore.

All this must be kept in mind when addressing the question of odor coding. The OB and the AL are probably less functionally homogenous than one might think, and have several different targets. Each target network does not read out the same code as other targets, meaning that several “codes” may coexist, either in different neurons or in the same neurons. Hence, the coding schemes that we will describe in the following paragraphs have to be interpreted as potentially specific to a particular function and one does not necessarily exclude the other.

This chapter has three parts. After reviewing the temporal constraints on olfactory perception (Section 13.2), we will describe the different types of temporal dynamics observed in the olfactory system (Section 13.3). Finally, we will expose and discuss the current hypotheses on how these dynamics might contribute to the odor code (Section 13.4).


All perceptions occur in time, and time is essential for proper perception and discrimination. Therefore, it is obvious that temporal constraints will most likely influence the selection of a code for sensory information transfer. The purpose of this section is to introduce the concept of temporal coding, to characterize the temporal constraints, and figure how they might determine the choice of a neural code used for odor perception.

13.2.1. Temporal Coding: Definition and Controversy

It is a rather difficult task to define what is meant by temporal coding, as different authors may have implicitly used different definitions. Here, we suggest using the definition proposed by Dayan and Abbott (2001). For these authors, a temporal code is a code based on temporal relationships in the neural response. Temporal relationships can either be timing of the response and, more generally, temporal sequence of the response relative to some clock signal (e.g., stimulus onset, oscillation), or timing of neurons with respect to each other in a population (e.g., synchrony, neuronal activation sequences). It is important to note that a firing rate computed in a specific time window does not make use of any temporal relationship in the responses, and is, therefore, not a temporal code. However, a code based on a series of rate measures in time is a temporal code, as it uses the temporal sequence of neural activations. In consequence, Dayan and Abbott argued that temporal coding and rate coding should not be opposed, as they might be nested in the same scheme. Only the special case (although used in many studies on sensory coding) of a firing rate measure in a single time window can plainly be opposed to temporal coding.

Unfortunately, for some authors, the existence of temporally patterned neural responses does not by itself indicate a temporal code, as neural signals can be read in many ways that do not necessarily include temporal relationships. However, Dayan and Abbott (2001) proposed a method to assess if temporal coding is plausible or not. Temporal coding is only possible if meaningful (i.e., information rich) temporal relationships are present on a timescale smaller than the scale of relevant temporal fluctuation of the stimulus. For example, it is difficult to imagine a mechanism used for encoding visual scenes that would be slower than the actual temporal accuracy of visual perception (VanRullen et al. 2005). If the same argument holds for any type of code, it must be used with care since, as already mentioned in the introduction, sensory systems and perception have multiple facets. Several circuits might implement in parallel several features of stimulus perception (e.g., novelty, noxiousness, quality, usefulness). Consequently, assessing the optimal perception and discrimination times of an animal to different stimuli might be used to define and put constraints on a minimal code that may be implemented by some part of a sensory system to compute simple behavioral responses. Addressing the question of the neural code used then leads to assessing which kind of perception or task this code might be used for.

Having this definition of temporal coding in mind, we will review, further in this chapter, different coding schemes in olfactory circuits that deal with temporal fluctuations of the neural activity. Some of them are temporal codes and others are not. To help the reader further in making up his/her mind about which coding schemes seem most plausible, we will first describe first what is known about the temporal constraints on olfactory perception.

13.2.2. The Intrinsic Temporal Fluctuation of Smells

In a natural environment, odorant molecules are carried by air or water. These mediums undergo constant fluctuations, often incoherent and chaotic. Odors enter into contact with the olfactory receptor neurons (ORNs) depending on these fluctuations. Due to this constraint, the question is whether network activity is regulated by internal mechanisms, such as oscillations, or by these fluctuations. In this case, the olfactory system might have to encode this temporal feature in addition to odor identity and intensity information. It is noteworthy to emphasize an important difference. According to the temporal coding hypothesis, neurons might encode odor identity and intensity by generating a code that is temporally related. In the case of encoding fluctuations of the odor plume, the timing component is now carried by the stimulus itself and is therefore external to the system.

In consequence, as most experimental studies are performed using constant and long-lasting odor applications, particularly in nonbreathing animal models, we may wonder if an experimental bias has not been introduced. Nevertheless, some studies have specifically addressed the question of odor plume fluctuations. Vickers and colleagues (2001) have used an electroantennogram (EAG)—reflecting the activity of ORNs population—to monitor the activity of a moth antenna in response to pheromone plume fluctuations in a laboratory wind tunnel. The recorded preparation was moved around the center of the tunnel, changing its relative position in comparison to the pheromone plume wind flow. At low wind speed, the largest burst and the most variable EAG activity occurred in the central zone of the plume. As speed changed, the odor plume became more dispersed. The largest amplitude and the highest frequency fluctuations shifted to the periphery. The authors concluded that a very minor shift in the position relative to the odor plume can dramatically change the ORNs activity. They also performed intracellular recordings of PNs and observed that neurons are strongly time-locked to stimulus dynamics. The PNs firing was strongly correlated with EAG onset and their frequency of spiking increased with EAG bursts amplitude. Applying a varying odor plume in amplitude and duration, they observed a large range of the PNs frequency of discharge (0–150 Hz), indicating that activity is strongly dependent on the stimulus dynamics (Vickers et al. 2001). Other studies have suggested that the network dynamics are built to preferentially follow and encode these fluctuations. Christensen and colleagues presented a set of odor pulses at 1 Hz to the moth and recorded projection neuron (PN) single-unit activity (Christensen et al. 2000). They suggested that the PN firing pattern not only depended on the chemistry, but also on the physical context. By presenting a blend of odor, they observed that two particular PNs responded in synchrony to a weak odor concentration, but displayed the complete contrary, a desynchronization when concentration was increased. These so-called emergent properties are thought to be strongly adapted to constantly changing odor plume (Christensen et al. 2000). A more recent study has also addressed the question of encoding fluctuating stimulus by the locust PNs (Brown et al. 2005). Intracellular recordings of the AL neurons and single-unit recordings of the AL and MB neurons have been performed. Varying numbers of odor pulses were applied at different frequencies. The authors analyzed the data at a neural ensemble level, monitoring the population activity by implementing a population vector analysis (see Figure 13.2A). Briefly, it consists of taking the number of spikes in a certain temporal window (= time bin), each vector row representing an individual neuron. The time-course of population activity is represented by a time series of the population vector. Authors have shown that independently of the number and frequency of pulses, the correlation between PN population vector activity of different protocols was high during some part of the odor presentation. These results suggest that the responses of multiple and rapid pulses are sufficiently correlated to one another to allow odor discrimination. Moreover, when pulses were brief, each new pulse-evoked activity truncated the previous one. Therefore, the network is adapted to be rapidly reset, allowing the next odor pulse to be encoded. Finally, authors have recorded the response of MB neurons (Kenyon cells, KCs) to such stimuli. They noticed that when the interpulse interval was brief, the largest response arose at the pulse onset, the frequency of discharge decreasing in the following pulses, and increasing again at the offset. This part of the olfactory system, at least in locusts, could therefore be more adapted to detect the odor appearance, the continuous odor flow fluctuations and its disappearance (Brown et al. 2005).

FIGURE 13.2. Odor-evoked population dynamics.


Odor-evoked population dynamics. (A) Construction of a population vector. For any time bin (t0 to tN), the nth dimension of the vectors corresponded to the average firing rate (FR) of the nth-recorded cell. (B) Average trajectory of the population vector (more...)

Brain oscillations have been proposed to be an important feature for odor discrimination (see Section 13.3). But oscillations sometimes develop with long-lasting odor application, which do not relate to the fast fluctuation timescales observed for odor plumes. Some studies have emphasized that these constant fluctuations may avoid the appearance of such oscillations (Christensen et al. 2000). In the moth, a study has described this phenomenon (Christensen et al. 1998). Intracellular recordings of the PN have been done, and 2 Hz pulsed as well as 5 s continuous odor stimulations were presented to the insect. The authors observed two different firing behaviors. When applying the pulses, no oscillations were observed and the authors suggested that information could be carried by a simple rate code. However, when the stimulus duration was increased, more temporally complex firing properties of PNs and oscillatory mechanisms emerged (Christensen et al. 1998). The same authors have also described that local field potential (LFP) oscillations and PNs spiking were not temporally correlated when applying brief odor pulses, questioning the role of oscillations in odor coding (Christensen et al. 2003).

In summary, these studies highlighted the importance of fluctuating odor plume in the olfactory coding. Neuronal population activity is able to track these dynamical changes with high precision. The coding strategy of the olfactory system may be strongly imposed by these dynamics. Theoretical work has suggested that temporally fluctuating stimuli are more reliable than constant ones, leading to more reproducible spiking behavior (de Ruyter van Steveninck et al. 1997). Experimental data tend to confirm this theoretical finding. Flying moths cease to make upwind progress and start to cast or counterturn across the wind as soon as they are exposed to a constant pheromone stimulus. This behavior may enhance the retrieval of the odor trace (Baker and Haynes 1989).

The results of higher efficiency of encoding fluctuating stimuli are subject to discrepancy. Other theoretical studies have shown that spiking variability of neurons in the fly visual system do not change between both conditions of stimulation, meaning that both are encoded with the same reliability (Warzecha and Egelhaaf 1999). However, these differences may be explained by variations among sensory systems. Different species that have a different ecological environment and different behavior may also need different coding strategies. Thus, a flying insect may need a system that can follow rapid fluctuation of the odor plume in order to quickly adapt and modify the flying trajectories, while a walking insect may need less rapid adaptation to follow an odorant trace.

13.2.3. Time and Olfactory Behavior

Processing sensory inputs is performed with a certain delay due to receptor activation onset and transfer of information along sensory axons to the first brain relay. The subject of temporal coding brings up the question of the timescale needed for the olfactory system to segregate two odors. To address this question, researchers have used behavioral paradigms in mice and rats. Uchida and Mainen (2003) trained rats using an operant conditioning to perform two-alternative forced choices. They monitored the speed of discrimination between rewarded and unrewarded odors by video tracking. The rats had to discriminate between mixtures containing different proportions of two odors. First, the authors noted that the speed, but not accuracy, of discrimination is independent of mixture difficulty. In addition, accuracy increases with sampling time, but only up to 200 ms. A longer sampling time above 200 ms did not lead to further increases in accuracy, but, on the contrary, had the tendency to disimprove the performances. This study suggests that at 200 ms, the rat is able to discriminate between two odorants and discrimination time is independent of discrimination difficulty (Uchida and Mainen 2003). However, this two-choice discrimination task has some limitations. It measures the moment of head retraction, namely the motor command that makes the rat move from the odor port to the reward port. It has been claimed that the decision for such a movement and, more particularly, the olfactory discrimination process is taking place much earlier. Hence, some authors trained mice to a go/no-go task, which is thought to be more accurate for measuring the olfactory processing time. For an unrewarded odor, the mouse retracted its head and continued exploring the cage. However, for a rewarded odor, the mouse started licking the water reward. The head movement was tracked with a beam in the port. If odor was rewarded, the mouse stayed in the port and the beam was continuously interrupted. But, for the unrewarded odor, the beam was resealed when the mouse retracted its head. The speed of the discrimination task was determined by comparing both beam traces. The time point of head retraction (breaking point) significantly different from the time point of unbroken beam trace was defined as the decision time point. Authors have found a quite similar discriminating time for easy task meaning simple odors: ~200 ms. However, these authors have found a contrary result compared to Uchida. Indeed, they have noted that discrimination time significantly increased with binary mixture, even more with very close concentration of the two components (Abraham et al. 2004).

A more recent study seems to conciliate both views with the concept of speed-accuracy tradeoff (SAT), which has been described in both vision and audition. Authors claimed that both precedent studies might be biased by the fact that the mouse can choose the spending time in the port to better discriminate between two odors. Thus, they conceived a paradigm in which the odor sampling was chosen by the authors. They have observed that the accuracy depended not only on the difficulty of the task but also on the sampling time (Figure 13.1A). Indeed, the mice were more accurate when they were forced to discriminate with a longer odor sample. Hence, mice were able to discriminate in ~300 ms for the easy task, but the discrimination time reached ~600 ms for the hardest tasks (Rinberg et al. 2006). In terms of neural coding, these results would mean that more computational time is needed in the brain to differentiate odor-evoked neural responses and, therefore, make a decision. As such waiting more time allows collecting more information that could be used for increasing discrimination performances, as it can simply be observed by changing the size of the temporal window in which neuronal ensembles firing information is read (Figure 13.1B).

FIGURE 13.1. Speed-accuracy tradeoff in odor discrimination.


Speed-accuracy tradeoff in odor discrimination. (A) Relationship between odor discrimination accuracy and mouse sampling time during discrimination behavioral task. The curve is dependent of the task difficulty (increased difficulty indicated by progressive (more...)

Experiments using freely moving animals might lead to some bias. First, behavioral paradigms generally measure the precise moment when the animal is moving as a discrimination time. However, such behavior implies other processes, like decision making and motor planning. These processes take time and olfactory discrimination might be processed upstream and earlier. Second, the onset of odor sampling is generally the instant when the valve opens to release the odor. Considering that the odor runs into the tubing, the exact odor-sampling onset can be biased by this delay. Finally, complex behaviors, such as decision making, are known to be influenced via feedback processes. Hence, some researchers have tried to sidestep this bias and addressed the question of odor discrimination by reading the process time of rat OB using the calcium-imaging technique. The rat was head fixed and trained to an operant conditioning task, in which they associated an odor to a reward. The rat actively sniffed when they received the odor and started to lick if the odor was a rewarded one. Wesson and colleagues measured the delay between the onset of the first sniff and the odor-evoked calcium response (Wesson et al. 2008a, 2008b). Input arrived 100–150 ms after inhalation begin. Previous work by the same group has shown that the rat can discriminate an odor in a single sniff. The intersniff interval was ~75 ms, which is considered by the author as the central processing time of the OB to discriminate two odors. The total time of processing would then be 175–225 ms (Wesson et al. 2008a).

In conclusion, the consensus brought by these studies is that olfactory perception and discrimination occur very rapidly. The rodents are most likely using a single sniff to collect odor information in order to make a decision. Therefore, these behavioral data imposes temporal constraints that have to be taken into account in order to assess proposed odor codes.


13.3.1. Gamma and Beta Range Oscillations in the Olfactory System Phenomenology

Fast oscillations were probably the first example of temporal dynamics observed in the brain and, more specifically, in the olfactory system, described in the pioneering LFP studies of Adrian (1942, 1950). While recording in the OB of the anesthetized hedgehog, he noticed that prominent oscillatory variations of the extracellular potential occurred when the animal smelled an odor. These oscillations typically had a frequency in the range of 40–60 Hz, which is in the so-called gamma-frequency band. This observation has since been repeated several times in many mammals, both in anesthetized (Buonviso et al. 2003; Neville and Haberly 2003) and awake preparations (Kay and Laurent 1999).

It has also been shown that relatively fast oscillations are also triggered by odor stimulation in the OB of fish (Friedrich et al. 2004), in invertebrates such as the limax (Gelperin and Tank 1990), and in the insect AL (Laurent and Davidowitz 1994). Interestingly, the oscillation frequency is lower in fish and insects, reducing to around 20 Hz. This would correspond to the beta-frequency band if one follows the established nomenclature. However, 20 Hz oscillations in insects and fish are often referred to as gamma oscillations in analogy to the mammalian oscillations. Gamma-band oscillations were not only observed in the OB or AL, but also in downstream areas, such as the olfactory cortex in vertebrates (Freeman 1978) and the MB in insects (Laurent and Naraghi 1994). In rodents, careful analysis both in the OB and the cortex also revealed some odor-induced activity in the beta frequency band (15–40 Hz) (Boeijinga and Lopes da Silva 1989; Buonviso et al. 2003) in combination with gamma-band activity. Neither beta- nor gamma-band oscillations were ever described in ORNs.

Single or multiunit recordings performed in combination with LFP revealed that in insects (Perez-Orive et al. 2002), fish (Friedrich et al. 2004), and mammals (Buonviso et al. 2003; Eeckman and Freeman 1990; Litaudon et al. 2008), the action potentials of most principal neurons occur around a given phase of the extracellular oscillation. However, a synchronized neuron does not necessarily fire at gamma frequency and can skip several oscillation cycles (Bathellier et al. 2006; Friedrich et al. 2004; Lagier et al. 2004; Perez-Orive et al. 2002). In all cases, the preferred phase seems to be homogeneous among neurons, indicating that many neurons in the network fire in synchrony during the fast oscillations episodes (Eeckman and Freeman 1990; Friedrich et al. 2004; Laurent and Davidowitz 1994). Hence, the fact that gamma (or beta) oscillations are a collective behavior of a large number of neurons also explains why they can be observed in field potential recordings. Mechanism

Since the works of Rall and Shepherd (1968) and later Freeman (1975), it has been mainly hypothesized that fast oscillations are self-generated by the OB and the cortex due to strong inhibitory feedback loops (e.g., inhibitory interneurons pre- and postsynaptically connected to excitatory neurons). The self-generation hypothesis is supported by the absence of gamma oscillation in nasal epithelium activity and the fact that an isolated OB slice is able to generate gamma oscillation following even single and short olfactory nerve stimulation (Halabisky and Strowbridge 2003; Lagier et al. 2004, 2007).

Theoretical studies on neural networks endowed with inhibitory feedback loops have shown that, provided with a strong enough feedback and a large enough “delay” between firing and feedback, such networks can generate fast oscillations of their population firing rate. In this case, the network alternates between a period of decreased firing transiently imposed by inhibition and a period of increased firing where feedback inhibition builds up to start the next cycle (Bathellier et al. 2008b; Brunel and Wang 2003; Freeman 1975). Although other mechanisms, such as synchronization of oscillating neurons via gap junctions or excitatory synapses, can also reliably generate collective oscillations in neural networks, the inhibitory feedback loops provide a more robust mechanism. Importantly, it is not required that the neurons fire themselves at gamma frequency (Bathellier et al. 2006), which is, indeed, not the rule for the olfactory system neurons (Bathellier et al. 2008a; Friedrich et al. 2004; Perez-Orive et al. 2002). It has also been shown in artificial networks with inhibitory feedback loops that oscillation frequency remains stable when inhibition strength is globally changed in the network. This has been experimentally observed in slice preparations of the OB (Bathellier et al. 2006), but is not predicted by other oscillation generation mechanisms. Models of the insect AL also point toward a feedback loop mechanism that involves local interneurons (Bazhenov et al. 2001). In the OB of mammals, candidate interneurons for driving the gamma oscillation are the granule cells (Halabisky and Strowbridge 2003; Lagier et al. 2004, 2007), while it is hypothesized that the slower beta oscillations in the bulb are generated by a feedback loop originating from the olfactory cortex (Buonviso et al. 2003; Neville and Haberly 2003). Fast Oscillation and Behavior

The physiological relevance and importance of oscillations for sensory coding remain unclear. Oscillations themselves may be involved in odor coding mechanisms; on the other hand, they may just represent a side-product of the action of some inhibitory loops, which are themselves important for odor coding. It is evident that addressing this question is very important, but an experimental approach based on a pharmacological suppression of inhibitory loops to remove the oscillation would not succeed to simply clarify this point.

The fact that gamma-band (or beta-band) oscillations occur simultaneously with odor perception may suggest that they are in some way implicated in odor processing. This idea was further supported by the observation that the amplitude of gamma-band activity over the OB is heterogeneous and that its spatial distribution depends on presented odors and behavioral contexts (Freeman and Schneider 1982). It was recently established that the power of gamma-band oscillations in the bulb increases in behaving animals when they have to perform difficult olfactory discriminations (Beshel et al. 2007). In the insect, suppression of fast inhibitory feedback by picrotoxin has been shown to suppress gamma oscillations and to deteriorate olfactory performance (Stopfer et al. 1997). It has also been observed in the rat that the relative power of gamma-band and beta-band oscillations can change dramatically with the animal’s experience. While novel perception of an odor is mainly associated with gamma-band activity in the bulb and cortex, beta-band oscillations become more prominent when the animal has learnt this odor (Ravel et al. 2003).

However, these results cannot yet help in deciding whether an increase or decrease in oscillation amplitude is the consequence of the differential involvement of some inhibitory feedback mechanisms, or correspond to a direct role of the oscillation in olfactory coding. Further work is needed to answer this important question.

13.3.2. Oscillatory Dynamics due to Sniffing in Mammals Phenomenology and Mechanism

Other prominent neuronal oscillations observed in the olfactory system of terrestrial vertebrates are the slow oscillations synchronized to the animal’s respiratory cycle. In rats and mice, these oscillations range between 2 and 10 Hz, depending on the animal’s sniffing behavior, and can be clearly separated from gamma and beta oscillations. Unlike gamma and beta oscillations, these slow oscillations are present in all three main stages of the olfactory system: in the sensory neurons (Spors and Grinvald 2002; Spors et al. 2006; Verhagen et al. 2007; Wachowiak and Cohen 2001), in the OB (Bathellier et al. 2008a; Buonviso et al. 2003; Cang and Isaacson 2003; Macrides and Chorover 1972; Margrie and Schaefer 2003; Onoda and Mori 1980), and in the olfactory cortex (Rennaker et al. 2007). They can be observed, although with a weaker amplitude, even in the absence of any odor input, but are absent when animals are tracheoto-mized and do not breath via the nose (Onoda and Mori 1980). Conversely, imposing an artificial sniffing cycle via a tube plugged in the trachea and directed to the nose is sufficient to induce slow oscillations in the olfactory system at desired frequency. Hence, the slow oscillations originate from the constant modulation of airflow at the nasal epithelium, leading to sensory neurons mechanical activation (Grosmaitre et al. 2007), but are not due to a synaptic drive internal to the brain.

The slow oscillations are clearly visible in both the LFP recordings (Buonviso et al. 2003) and in the firing patterns of receptor neurons (Duchamp-Viret et al. 2005), mitral cells (Bathellier et al. 2008a; Buonviso et al. 2003; Macrides and Chorover 1972; Onoda and Mori 1980), and pyramidal cells of the olfactory cortex (Rennaker et al. 2007). It is a global modulation of neural activity. However, different olfactory receptors can fire at different phases of the respiratory cycle (Spors et al. 2006). Likewise, different mitral cells can fire at different phases of the breathing cycle in the OB (Bathellier et al. 2008a), but the same mitral cell can also exhibit different phasing, depending on the odor and concentration presented to the animal (Bathellier et al. 2008a; Macrides and Chorover 1972). In general, several spikes are fired in a breathing cycle, but the number of spikes and their timing are extremely variable across cells and presented odors, giving rise to a large diversity of temporal firing patterns (Bathellier et al. 2008a).

At a population level, as has been recently shown for mitral cells, the result of this diversity is the emergence over time of complex neuronal ensemble activation patterns (Bathellier et al. 2008a). Different snapshots of the mitral cell population activity at different time points of the breathing cycle can be as dissimilar from each other as two snapshots taken during presentation of different odors (see also Section 13.3.3). Similar conclusions can actually be drawn for ensembles of ORNs derived from calcium imaging of glomeruli responses (Spors et al. 2006). Hence, unlike gamma oscillations, temporal complexity at breathing frequency originates, at least in part, from odor detection and transduction mechanisms themselves. Airflow, odor adsorption, and receptor dynamics might therefore play a significant role (Scott et al. 2006). In the bulb and cortex, internal network dynamics might also shape temporal patterns of principal neuron activity. However, because no thorough comparison of cyclic dynamics at different levels of the olfactory system has been conducted yet, it is hard to evaluate the respective contribution of intrinsic and extrinsic mechanisms in these dynamics. Breathing Cycle Dynamics and Behavior

If many studies of the breathing cycle dynamics have been carried out in anesthetized animals, for which sniffing frequency is rather constant, one must not forget that behaving animals constantly adjust their sniffing (Kepecs et al. 2007). Odorant inflow in the nasal cavity via sniffing is therefore an extraordinary generator of temporal variability (although under the control of the animal) rather than a precise clocking system. Some results, developed in Chapter 12, show that sniffing frequency can strongly change the patterns of inputs to the OB (Verhagen et al. 2007). This must be taken into account when defining any theory evaluating the role of breathing cycle dynamics in olfactory coding. In addition, a wide range of sniffing frequencies should be explored at each level of the olfactory system in order to be conclusive.

Even if interaction between respiration and sense of smell is only relevant for part of the animal reign, the olfactory system of the other species has also to deal with temporal fluctuations of the odor input. In some cases, fluctuations can also be actively controlled, as for some arthropods able to oscillate their antennae during odor perception (Koehl et al. 2001). More generally, the fluctuations are imposed by the structure of odorant stimuli, which contact odor sensors in a temporally discontinuous fashion. Interestingly, the response of AL neural ensembles to oscillating inputs are very similar to the rapidly evolving cycle observed in the rodent OB (Brown et al. 2005). This suggests a broad relevance for the study of slow oscillating or fluctuating dynamics (2–10 Hz frequency range) in most olfactory systems.

13.3.3. Slow, Nonoscillatory Patterning in Fish, Insects, and Mammals Phenomenology and Mechanism

There is a third type of temporal dynamics in the olfactory system. When a step odor stimulus is given to an animal, recorded neural responses are usually not following a stepwise time-course, even in the absence of breathing or if one averages neuronal activity over each breathing cycle. Instead, many neurons exhibit strong changes in their firing rate over time after odor onset and offset. For example, some neurons can be first inhibited and start firing with some delay, but the opposite is also possible. These changes are robust across trials. This was first evidenced in the OB of tracheotomized rats (Meredith 1986). Later on, complex temporal fluctuations of individual PN firing activity were shown in the insects AL (Laurent 1996; Wilson et al. 2004) and in the fish OB (Friedrich and Laurent 2001). In order to overcome the diversity of single neuron responses and try to better understand odor coding, some studies have analyzed odor responses in neural cell assemblies. To do so, the firing responses of a large number of recorded neurons can be simultaneously considered by putting them together in a population vector (Figure 13.2A). As mentioned earlier, it consists of taking the number of spikes in a certain temporal window (= time bin), each vector row representing an individual neuron. Every vector, therefore, represents a snapshot of the neuronal population activity in a defined time bin. It is important to note the difference with an average population firing, as population vectors preserve the specificity of individual neural responses. The time-course of population activity is represented by the time series of the population vector, which can be plotted as trajectories in a multidimensional space (space of all recorded neurons) (Figure 13.2B).

This analysis was first conducted in the fish OB, demonstrating that the activity of mitral cell ensembles could significantly change over a period of at least 1 s after steady odor application onset (Friedrich and Laurent 2001). In the locust AL, population activity is in a resting state that is left after odor onset, and typically evolves during roughly 1 s and then settles in a steady state called a fixed point, which lasts until the end of the stimulus application. Thereafter, it takes approximately 1 s for the population activity to settle back in its resting state (Mazor and Laurent 2005). A similar phenomenon was also recently observed in the mouse OB, superposed with the breathing cycle oscillation (Bathellier et al. 2008a) (Figure 13.2B). During the first second of odor presentation, population activity changed from cycle to cycle (Figure 13.2C), and after this initial period, population activity repetitively described the same cycle as long as the stimulus was sustained. These dynamic transitions could be better visualized when computing the rate at which the vectors changed over time (i.e., vector velocity, representing the distance between the population vectors of consecutive time bins; Figure 13.2C). After a steep change at odor onset, population activity kept evolving significantly above “noise” (i.e., baseline velocity) for ~1 s, and then reached a steady state. Another transient evolution was observed at odor offset and then the population settled in a poststimulus state that slowly drifted back to the resting state (within ~20 s).

A slow convergence of the firing rate has also been observed in fish (Friedrich and Laurent 2001) and in insect (Hallem and Carlson 2006) receptor neurons. However, at this level, all neurons display the same phasic time-course contrary to the OB or AL principal neurons, which exhibit various time-courses. The complexity of the response can be quantified by computing the correlation of the series of population vectors with the first vector after odor onset. The correlation stays very close to 1 for a population of receptor neurons, and can dramatically drop close to 0 for OB or AL neurons. It must be noted here that the decrease of the correlation is less pronounced in the mouse OB (Bathellier et al. 2008a), than in the fish bulb (Friedrich and Laurent 2001) and locust AL (Brown et al. 2005; Stopfer et al. 2003). This indicates a reduced complexity of slow, nonoscillatory dynamics at population levels in mammals.

The mechanisms of the slow nonoscillatory dynamics have not been clearly established yet. In ORNs, the phasic response profile seems to correspond to a rate adaptation mechanism. These input dynamics should play a role in the dynamics observed at the next level, although the increased complexity there suggests the involvement of other mechanisms, such as synaptic interactions. It was shown in the AL of insects that the GABAA receptor antagonist has no effect on slow temporal dynamics (Stopfer et al. 1997; Wilson and Laurent 2005). On the contrary, GABAB antagonists seem very potent in reducing the complexity of single neuron temporal firing patterns (Wilson and Laurent 2005). Hence, slow inhibitory synapses in the AL probably contribute substantially to the slow dynamics, as also suggested by theoretical work (Bazhenov et al. 2001). It is not known to what extent these results can be applied to the OB. Slow Dynamics and Behavior

The impact of slow convergence of neural activity on olfactory perception has never been directly studied. Nevertheless, comparison of perceptual delays with the 1 s time constant typically observed for the slow dynamics of neuronal activity can help set some constraints on the potential role of the latter phenomenon. Mice and rats were observed to discriminate between two odor pairs within 100–500 ms (Abraham et al. 2004; Uchida and Mainen 2003). This indicates that full convergence of neural activity to its steady state may not be required for perception. However, a study in mice suggests that discrimination of close odor pairs requires longer delays that for simple odor pairs. Therefore, one could imagine that slow convergence has a role in improving the discriminability of two odor percepts. This question will be discussed further in Section 13.4.3.

13.3.4. Short-Term Plasticity in Insects

Up to now, we have reviewed temporal schemes that shape neuronal activity in the olfactory system at different, clearly separable timescales: fast (gamma and beta from 20 to 80 Hz), intermediate (breathing cycle, 2–10 Hz), and slow (slow convergence, ~1 Hz). A fourth phenomenon, only observed in the locust, should be mentioned to end this list. When several pulses of an odorant stimulus are successively presented to a locust, neuronal responses change drastically from one pulse to another (Stopfer and Laurent 1999). The amplitude of gamma oscillation in the AL increases, while the number of PN spikes decreases, but their temporal coherence and accuracy increases. These modifications are not due to receptor neurons, and should be intrinsic to the AL neural circuits. The characteristics of these modifications (decrease of activity, increase in synchrony) suggest that they could originate from an increase in the strength of some inhibitory feedback loops, which is known to produce exactly the same changes in networks similar to the AL (Bathellier et al. 2006). Interestingly, discontinuous odor inputs are more potent than continuous inputs for triggering changes in the neural response. The phenomenon is quite insensitive to the interval between odor pulses (in the range of 2.5–20 s) or to the duration of these pulses (0.2–2 s). Unfortunately, it is not known how the system behaves when odor pulses become temporally very close, as occurs, for example, in an actively sniffing mammal. The phenomenon is also odor-specific, suggesting that it is a form of odor memory. But the “memory” of the received odor lasts no longer than 10 min. However, to the best of our knowledge, this peculiar form of temporal dynamics has never been described in other species.


13.4.1. Temporal Coding with Theta or Gamma Frequency

Observation of temporal dynamics in neural responses is a fact, but their significance and contribution to odor coding is still unclear. Some hypotheses have been described in coding schemes that only use temporal features to build the odor code and discard classical measure of neural activity, such as firing rate or spike counts. Even if, in light of recent experimental evidences, these hypotheses appear as such rather unlikely, they contain key concepts that could be involved in more complex theories and are worth presenting in this chapter. Coding with Spike Timing in the Breathing Cycle Oscillation

As demonstrated by Hopfield (1995), oscillations are ideal to build up temporal codes because a neuron driven by an oscillatory current fires with a timing that varies with the amplitude of its input. This makes it possible to transfer any analog information (e.g., odor concentration) via another analog variable (spike timing), thereby avoiding the loss of precision induced by spike count coding schemes (discrete variable).

Studying mitral cell responses to odors from in vivo patch-clamp recordings, Margrie and Schaeffer (2003) found that the latency of the first spike fired in a breathing cycle is dependent on odor identity and intensity. The authors also found that the number of spikes per cycle depended on odor identity and intensity, but not interspikes intervals, which led them to conclude that the instantaneous firing rate of a cell cannot efficiently code for the stimulus. They proposed instead that spike latency could encode the stimulus much better. Their claim was further supported by a simple computational model in which latency-based coding was shown to have a larger capacity (i.e., number of odors that can be encoded with a given number of neurons) than spike count or instantaneous firing-rate-based coding. However, more recently, a direct measure of the efficiency of latency-based coding in the mouse OB showed that contrary to the prediction of Margrie and Schaeffer (2003), it is much less efficient than a firing-rate-based coding (Bathellier et al. 2008a). Two explanations for this result can be given. First, latency does not systematically vary with odors and concentrations. There is a significant negative correlation between concentration and firing latency, but it appeared to be rather loose (Bathellier et al. 2008a). Second, the precision of firing latencies from trial-to-trial is rather poor, which directly limits the precision of the coding scheme. Noise constraints were not taken into account in the model proposed by Margrie and Schaeffer (2003), explaining why they could find better performance for spike latency-based coding. In physiological noise levels, the first spike latency is very variable. Therefore, a coding based on this sole parameter would not be very efficient and seems rather unlikely in the OB. Coding with Synchrony in a Neural Population

Another key phenomenon that can be used in a temporal coding scheme is synchrony between different neurons. As mentioned earlier in this chapter, gamma oscillations observed in the LFP are due to synchronous activity of many neurons. It has long been recognized that synchrony can be a way of transmitting information across neural networks, and many experimental studies support the idea that the brain might make great use of synchrony (Varela et al. 2001).

Recently, Brody and Hopfield (2003) have proposed an odor-coding scheme for the OB based solely on synchrony (Figure 13.3A). This scheme uses the fact that in a population of integrate-and-fire neurons receiving the same oscillatory drive and firing roughly at the same frequency, there exists a range of input current amplitudes in which neurons are well synchronized. When an odor enters the nose, a specific group of mitral cells would receive currents that are in the appropriate range and would be synchronized. Other cells would still fire asynchronously. Synchronized neurons could then be easily detected by downstream neurons or neuronal circuits endowed with appropriate bandpass filtering properties (e.g., circuits with delayed feed-forward inhibition are bandpass filters).

FIGURE 13.3. Oscillations and neuronal synchrony in odor coding.


Oscillations and neuronal synchrony in odor coding. (A) Comparison between the temporal coding scheme proposed by Brody and Hopfield (2003) and typically observed firing in the olfactory bulb or the antennal lobe. In the Brody and Hopfield model, all (more...)

The interesting aspect of this theory is that mitral cell assemblies could encode odor information without changing their firing rate. However, it has now been clearly demonstrated that in insects (Stopfer et al. 2003), fish (Friedrich and Laurent 2001), and mammals (Bathellier et al. 2008a), mitral cells actually do change their firing rate during odor stimulation, and that firing rate changes carry odor information. Therefore, the reality of the olfactory system is more complex than the simple model of Brody and Hopfield, and it is, in fact, unlikely that synchrony alone carries odor information in the olfactory system. In order to match experimental observations more closely, several theories have tried to conciliate firing rate changes and neural synchrony in a single coding scheme. We review some of them in the following sections.

13.4.2. Gamma Synchrony: Clocking of Neuronal Integration

The role of neural synchrony in the gamma frequency band has been particularly emphasized in the insect olfactory system, mostly in the locust and in the honeybee. One hypothesis that is now supported by several pieces of evidence is that each cycle of the gamma oscillation serves as a time window for integration of upstream activity by downstream networks.

As described in Section 13.3, gamma oscillations (~20 Hz) appear in the AL, where they are most probably intrinsically generated due to inhibitory feedback loops (Figure 13.3C). Odordependent neural activity in the AL is also very dynamic on a slower timescale with a transient lasting up to 1 s, during which ensemble activity is strongly reshaped in time. Due to these temporal dynamics, it is crucial to know on which timescale activity is integrated by downstream networks. Perez-Orive and colleagues (2002) have suggested that integration of AL activity by KCs in the MB occurs over single cycles of the gamma oscillation. The proposed mechanism is the following. Both AL PNs and KCs tend to synchronize on the gamma cycle. On average, PNs lead KCs by almost half a cycle, a delay that probably corresponds to the propagation of spikes from the AL to the MB. A population of inhibitory neurons in a separate region of the insect brain (lateral horn) also receives feed-forward excitation from PNs and fires roughly at the same phase of the gamma oscillation as the KCs. Lateral horn inhibitory neurons project onto the KCs, sending a strong barrage of inhibition that arrives in the last part of the cycle. This inhibitory input prevents further firing of the KCs and tends to reset integration of excitatory inputs before the next cycle begins, suggesting that the integration time window of the Kenyon cell (KC) only spans a gamma cycle. The consequence is that all temporal features of AL activity slower than 20 Hz can theoretically be retained by the MB.

The role of feed-forward inhibition in the MB is not restricted to the establishment of a precise integration window. It also drastically reduces the firing of KCs (Perez-Orive et al. 2002), contributing to the transformation of a dense representation odor in the AL (i.e., each PN responds to many odors) into a sparse representation in the MB (i.e., each KC responds to few odors; Perez-Orive et al. 2002). Indeed, a pharmacological block of inhibition in the MB considerably reduces sparseness. Sparse representations are thought to offer many advantages for memory storage in neural networks. Hence, the feed-forward inhibition from the lateral horn is probably a crucial mechanism for efficient olfactory function. It is, however, not clear whether the fact that it occurs in a rhythmic fashion is crucial to MB function or whether it is just the consequence of the presence of inhibitory loops in the locust olfactory system that downregulates activity and generates an oscillatory behavior.

Recently, some experimental evidence started to provide an answer for this question. It was recently demonstrated that gamma synchrony is actively transferred across the synapse between KCs and their downstream targets (the beta-lobe neurons) by a spike timing-dependent plasticity mechanism (Cassenaer 2007). This mechanism acts so that beta-lobe cells, which fire too late with respect to the gamma cycle, have their synaptic input increased so that they fire earlier, and conversely. The existence of such a mechanism supports the idea that the clocking of neural activity by gamma oscillation is important for the function of the olfactory system. It is, however, not a definitive proof. Spike timing-dependent plasticity is a general mechanism that is thought to underlie some of the learning phenomena in the nervous system (Kepecs et al. 2002). If some theoretical studies have shown that it mechanically increases synchrony of neural activity (Suri and Sejnowski 2002; Zhigulin et al. 2003), many other studies point out other crucial functions, such as optimization of information transfer (Toyoizumi et al. 2005). Currently, it is not possible to rule out that spike timing-dependent plasticity is present in this synapse to fulfill one of these other functions and that, as a byproduct, it also sharpens gamma synchrony.

13.4.3. Slow Dynamics and Decorrelation

As we have detailed in previous sections, aside from gamma oscillations, olfactory system activity also exhibits slow temporal patterns. The idea that integration across different stages of the system might occur at the timescale of a gamma cycle would imply that slower patterns can be entirely read out by downstream structures. Nonetheless, the function of slow temporal patterning is not yet clear. We have mentioned earlier that coding solely based on spike timing with respect to the breathing cycle oscillation was unlikely. The breathing cycle is also not present in all species. But what about the slow, nonoscillatory dynamics of neural activity, which is conserved across phyla?

Slow dynamics are, in some form, present in all stages of the olfactory system, but gain in complexity in the OB or equivalently in the AL. In both networks, these dynamics are characterized by a reshaping of ensemble activity over time, which ends roughly 1 s after odor onset (Bathellier et al. 2008a; Mazor and Laurent 2005). One possibility that we will explore in the last section is that this slow patterning builds a temporal code, and that the entire trajectory of neural population activity is needed to determine which odor the animal received. But another interpretation would be that slow dynamics reshape neural activity to perform some kind of processing of the odor information.

An hypothesis along this line was proposed by Friedrich and Laurent (2001). It was observed that in the fish OB, ensemble activity evolves in time toward less correlated and thereby more easily separable representations of different odors. This result was obtained by computing the correlation between all population vectors representing the ensemble response to a set of amino acid odors. It shows that the mean correlation decreases progressively in time after odor onset, taking approximately 1 s to reach a minimum. The odor classification success based on population vectors also increased in time. Interestingly, this outcome was not observed in a population of receptor neurons (Friedrich and Laurent 2001), suggesting that the improvement of odor representations is generated by the bulb circuitry. Hence, it was proposed that slow dynamics in the OB aims at improving the discriminability of odor percepts.

Surprisingly, somewhat dissimilar results were obtained in the locust AL. Indeed, in the locust, the PNs response to long odor pulse decreases in precision over the same time period (~1 s), while the population converges to a fixed point of the dynamics (Mazor and Laurent 2005). The similarity between population vectors representing responses to different odor also increases. This suggests that slow dynamics in the locust produces opposite effects to those observed in fish. Careful analysis of locust ensemble activity, however, shows that representations decorrelate from 0 to ~100 ms after odor onset, as observed in fish on a much longer timescale. The discrepancy in decorrelation speed makes it questionable whether the two phenomena are comparable. It is, however, possible that slow dynamics, which develop both in fish and insects over the timescale of 1 s and yield comparable temporal patterns of neural activity, would represent different species-specific mechanisms. In mice, current data show that discriminability between representations of different odors in the OB does not improve on the slow timescale, but does not decrease either (Bathellier et al. 2008a). To this end, care should be taken with the interpretation of these results. Discrepancies exist in the experimental conditions and in the set of stimuli tested, which could also explain the apparent contradictions. It should be mentioned, for example, that mice experiments were performed in freely breathing animals, for which breathing cycle dynamics interact with the slow convergence of ensemble activity. In insects receiving fluctuating inputs so that ensemble activity of PNs does not reach a fixed point, no decrease of discriminability has been observed (Brown et al. 2005). Careful parallel analyses remain to be done to decide whether the impact of slow dynamics is really dissimilar in different species. Overall, slow dynamics appears as a complex phenomenon whose purpose and mechanism are far from being resolved yet.

13.4.4. Multiplexing: Combining Slow and Fast Dynamics

Even more complexity can be added if one combines fast and slow dynamics in the odor code. It was recently shown that the temporal behavior of odor representations in the fish OB depends on phase locking of spikes to the ensemble oscillation (Friedrich et al. 2004). In fish, spikes that are not phased-locked to gamma oscillations turned out to be the majority. Considered separately, they form neural representations that decorrelate over time, as described in the previous section for an unsorted population. On the contrary, the minority of phased-locked spikes yields ensemble representations that become more correlated over time.

This observation led the authors to propose that phased-locked spikes carry information about odor categories that is lost in non-phased-locked spikes during the decorrelation process. In other word, two codes could be used in parallel. A firing-rate-based code for odor identity, and a synchrony-based code for odor categorization. This interesting hypothesis on how different codes might serve different purposes in the same neural system deserves to be explored further, in particular to test whether downstream targets of mitral cells in fish actually make use of the two codes.

13.4.5. Information Contained in Temporal Sequences

If the response of a population of neurons to odors has the form of a complex temporal sequence, there are three alternatives for coding:

  • Neurons firing rates in a single and fixed time window of the sequence could be used for read out, while other time bins of the sequence would just represent elements of the brain dynamics that lead to the read out time bin. This is the simplest scheme, which is used in most artificial sensor systems.
  • Different time windows could equivalently be used for reading out firing rates. The brain could “choose” the time window to adapt to the requirement of the tasks it is engaged in (e.g., earlier time window for speed, best time window for accuracy).
  • The combination of all spike times in the sequence could be read out during a certain time window (whose duration could be flexible). This temporal scheme is equivalent to counting the number of spikes falling in successive time bins, with the duration of the time bin representing the accuracy of the spike time measure.
As long as the “decoding” mechanisms in brain networks downstream to the considered population are not known, it is hard to figure out which scheme this population actually implements. But, some hints can be obtained from the analysis of the structure of the neural signal. The first possibility would be evident if information about odors was clearly concentrated in a specific time window. The second and third possibilities would, on the contrary, require the neural response to be informative on several time windows. The third possibility would be more likely if spikes time-based coding was more informative than firing-rate-based coding on the same time interval.

Recordings from neuronal populations in the insect AL (Brown et al. 2005; Mazor and Laurent 2005; Stopfer et al. 2003) and the fish OB (Friedrich and Laurent 2001) indicate that enough information about the stimulus is present in the firing rates of 50–200 cells to discriminate between eight and 16 odors with less than 20% error in any time bin of 50–400 ms from stimulus onset to 1 s after stimulus offset (Figure 13.4A). Here, the effect of time bin duration was not studied. It is clear that all time bins contain non-negligible information about the stimulus. Of course, as mentioned earlier, some time bins contain a bit more information than others. They are found 1 s after odor onset in fish, and in 100–200 ms to 1 s after odor onset or offset in locusts (Mazor and Laurent 2005). Data, therefore, indicate that different time bins can be used for decoding. The same qualitative outcome was obtained in the mouse OB (Bathellier et al. 2008a) (Figure 13.4B and C): all breathing cycles (a cycle represents ~300 ms in this study) following odor onset were equally informative (~10% error for 101 cells). However, within a breathing cycle, the prediction error for time bins of ~40 ms could vary from 20 to 60% for 15 odorant stimuli (chance level = 93.7% prediction error). This is a quantitative difference to locusts or fish, which do not show such large and rapid variations of prediction error. It is interesting that in mice, time bins that are preferential for decoding appear only at a fast timescale. In the mouse bulb, progressively increasing the time bin duration from 40 ms to one breathing cycle (300 ms) could monotonically decrease the prediction error from ~20% down to ~10% (Figure 13.1B, Bathellier et al. 2008a). This shows that averaging out rapid temporal fluctuations actually produces no loss of information for a population code. Instead, it increases read out information because time averaging reduces noise levels.

FIGURE 13.4. Neuronal ensembles coding and input dynamics in locusts and mice.


Neuronal ensembles coding and input dynamics in locusts and mice. (A) Ensemble responses can be used to classify the set of odor trials presented to the animal, regardless of the change in the odor application patterns (indicated by the gray bars and (more...)

To date, the information contained in true temporal coding schemes has only been evaluated in mice (Bathellier et al. 2008a). In mice, the strongest changes in population activity are observed within breathing cycles. When breathing cycles were divided into 16 time windows to describe the temporal sequence of spikes, the combined information of all 16 time windows yielded a prediction error of 8% for 15 odor stimuli and a population of 101 mitral cells. In comparison, the mean firing rate computed over one breathing cycle yielded 13% prediction error. Hence, for the 15 odorant stimuli of this study, the information contained in the full temporal sequence of population activity added little to the information that could already be extracted from mean firing rates. In addition, most of the temporal information was contained in slow temporal features. Odor prediction based on the first Fourier coefficient of the activity sequence (i.e., mean phase and modulation amplitude of neuronal activity in a breathing cycle) yielded only 9% error, but the error dramatically increased when subsequent Fourier coefficients (i.e., finer temporal features) were used alone for odor prediction. Hence, some temporal information is clearly present in OB activity in mice. Current data are, however, not sufficient to decide whether this information is really useful to the system, because for the small number of odorant stimuli tested, roughly the same performance was obtained whether temporal information was used for read out (sequence) or not (average firing rate). It is clear from existing data that temporal information is not necessary for simple odor discrimination. But temporal information could improve olfactory coding in two different ways. Either it could bring more sensitivity, helping in difficult odor discriminations, or it could increase the information capacity of the olfactory system, allowing encoding of more odors with less neurons. Further investigation should be carried out to address these questions.


Recordings of neurons in the olfactory system have now highlighted the diversity and temporal complexity of odor-evoked firing patterns. In this chapter, we endeavored to give an overview of the temporal dynamics that has been observed so far. It is interesting to note that the published studies support analogies in the dynamics of odor representations in evolutionary very different species, such as mammals and insects. It is, therefore, tempting to think that these dynamics may be underlying for some part of the odor code, at least in the first relay of the olfactory system. However, this remains a hypothesis that has to be carefully tested. Indeed, as long as the “decoding” mechanisms in downstream brain networks are not better known, it will be hard to figure out which encoding schemes are used. The challenge for the next years will be to record neural activities in different networks of the olfactory system and try to link them to the animal behavior. In addition, unraveling the molecular, cellular, and network mechanisms underlying the different observed dynamics should help us understand what are the actual coding principles used in the brain in olfaction and assess whether they are optimal.


We thank the University of Geneva and the Swiss National Science Foundation for their financial support.


  • Abraham N.M., Spors H., Carleton A., Margrie T.W., Kuner T., Schaefer A.T. Maintaining accuracy at the expense of speed; stimulus similarity defines odor discrimination time in mice. Neuron. 2004;44:865–76. [PubMed: 15572116]
  • Adrian E.D. Olfactory reactions in the brain of the hedgehog. J Physiol. 1942;100:459–73. [PMC free article: PMC1393326] [PubMed: 16991539]
  • Adrian E.D. The electrical activity of the mammalian olfactory bulb. EEG Clin Neurophysiol. 1950;2:377–88. [PubMed: 14793507]
  • Baker T.C., Haynes K.F. Field and laboratory electroantenno-graphic measurements of pheromone plume structure correlated with oriental fruit moth behaviour. Physiol Entomol. 1989;14:1–12.
  • Bargmann C. Comparative chemosensation from receptor to ecology. Nature. 2006;444:295–301. [PubMed: 17108953]
  • Bathellier B., Buhl D.L., Accolla R., Carleton A. Dynamic ensemble odor coding in the mammalian olfactory bulb: Sensory information at different timescales. Neuron. 2008a;57:586–98. [PubMed: 18304487]
  • Bathellier B., Carleton A.
  • Gerstner W. Gamma oscillations in a nonlinear regime: A minimal model approach using heterogeneous integrate-and-fire networks. Neural Comput. 2008b;20:2973–3002. [PubMed: 18533817]
  • Bathellier B., Lagier S., Faure P., Lledo P.M. Circuit properties generating gamma oscillations in a network model of the olfactory bulb. J Neurophysiol. 2006;95:2678–91. [PubMed: 16381804]
  • Bazhenov M., Stopfer M., Rabinovich M., Abarbanel H.D., Sejnowski T.J., Laurent G. Model of cellular and network mechanisms for odor-evoked temporal patterning in the locust antennal lobe. Neuron. 2001;30:569–81. [PMC free article: PMC2907737] [PubMed: 11395015]
  • Benton R., Sachse S., Michnick S.W., Vosshall L.B. Atypical membrane topology and heteromeric function of Drosophila odorant receptors in vivo. PLoS Biol. 2006;4:e20. [PMC free article: PMC1334387] [PubMed: 16402857]
  • Beshel J., Kopell N., Kay L.M. Olfactory bulb gamma oscillations are enhanced with task demands. J Neurosci. 2007;27:8358–65. [PubMed: 17670982]
  • Boeijinga P.H., Lopes da Silva F.H. Modulations of EEG activity in the entorhinal cortex and forebrain olfactory areas during odour sampling. Brain Res. 1989;478:257–68. [PubMed: 2924130]
  • Brody C.D., Hopfield J.J. Simple networks for spike-timing-based computation, with application to olfactory processing. Neuron. 2003;37:843–52. [PubMed: 12628174]
  • Brown S.L., Joseph J., Stopfer M. Encoding a temporally structured stimulus with a temporally structured neural representation. Nat Neurosci. 2005;8:1568–76. [PubMed: 16222230]
  • Brunel N., Wang X.J. What determines the frequency of fast network oscillations with irregular neural discharges? I. Synaptic dynamics and excitation-inhibition balance. J Neurophysiol. 2003;90:415–30. [PubMed: 12611969]
  • Buck L., Axel R. A novel multigene family may encode odorant receptors: A molecular basis for odor recognition. Cell. 1991;65:175–87. [PubMed: 1840504]
  • Buonviso N., Amat C., Litaudon P., Roux S., Royet J.P., Farget V., Sicard G. Rhythm sequence through the olfactory bulb layers during the time window of a respiratory cycle. Eur J Neurosci. 2003;17:1811–19. [PubMed: 12752780]
  • Cang J., Isaacson J.S. In vivo whole-cell recording of odor-evoked synaptic transmission in the rat olfactory bulb. J Neurosci. 2003;23:4108–16. [PubMed: 12764098]
  • Cassenaer S., Laurent G. Hebbian STDP in mushroom bodies facilitates the synchronous flow of olfactory information in locusts. Nature. 2007;488:709–13. [PubMed: 17581587]
  • Christensen T.A., Lei H., Hildebrand J.G. Coordination of central odor representations through transient, non-oscillatory synchronization of glomerular output neurons. Proc Natl Acad Sci USA. 2003;100:11076–81. [PMC free article: PMC196929] [PubMed: 12960372]
  • Christensen T.A., Pawlowski V.M., Lei H., Hildebrand J.G. Multi-unit recordings reveal context-dependent modulation of synchrony in odor-specific neural ensembles. Nat Neurosci. 2000;3:927–31. [PubMed: 10966624]
  • Christensen T.A., Waldrop B.R., Hildebrand J.G. GABAergic mechanisms that shape the temporal response to odors in moth olfactory projection neurons. Ann N Y Acad Sci. 1998;855:475–81. [PubMed: 9929641]
  • Dayan P., Abbott L.F. Theoretical Neuroscience Computational and Mathematical Modeling of Neural Systems. MIT Press; Cambridge, MA: 2001.
  • de Ruyter van Steveninck R.R., Lewen G.D., Strong S.P., Koberle R., Bialek W. Reproducibility and variability in neural spike trains. Science. 1997;275:1805–8. [PubMed: 9065407]
  • Duchamp-Viret P., Kostal L., Chaput M., Lansky P., Rospars J.P. Patterns of spontaneous activity in single rat olfactory receptor neurons are different in normally breathing and tracheotomized animals. J Neurobiol. 2005;65:97–114. [PubMed: 16114031]
  • Eeckman F.H., Freeman W.J. Correlations between unit firing and EEG in the rat olfactory system. Brain Res. 1990;528:238–44. [PubMed: 2271924]
  • Firestein S. How the olfactory system makes sense of scents. Nature. 2001;413:211–18. [PubMed: 11557990]
  • Freeman W. Mass Action in the Nervous System:Examination of the Neurophysiological Basis of Adaptive Behavior Through EEG. Academic Press; New York: 1975.
  • Freeman W.J. Spatial properties of an EEG event in the olfactory bulb and cortex. Electroencephalogr Clin Neurophysiol. 1978;44:586–605. [PubMed: 77765]
  • Freeman W.J., Schneider W. Changes in spatial patterns of rabbit olfactory EEG with conditioning to odors. Psychophysiology. 1982;19:44–56. [PubMed: 7058239]
  • Friedrich R.W., Habermann C.J., Laurent G. Multiplexing using synchrony in the zebrafish olfactory bulb. Nat Neurosci. 2004;7:862–71. [PubMed: 15273692]
  • Friedrich R.W., Laurent G. Dynamic optimization of odor representations by slow temporal patterning of mitral cell activity. Science. 2001;291:889–94. [PubMed: 11157170]
  • Gelperin A., Tank D.W. Odour-modulated collective network oscillations of olfactory interneurons in a terrestrial mollusc. Nature. 1990;345:437–40. [PubMed: 2342574]
  • Grosmaitre X., Santarelli L.C., Tan J., Luo M., Ma M. Dual functions of mammalian olfactory sensory neurons as odor detectors and mechanical sensors. Nat Neurosci. 2007;10:348–54. [PMC free article: PMC2227320] [PubMed: 17310245]
  • Halabisky B., Strowbridge B.W. Gamma-frequency excitatory input to granule cells facilitates dendrodendritic inhibition in the rat olfactory Bulb. J Neurophysiol. 2003;90:644–54. [PubMed: 12711716]
  • Hallem E.A., Carlson J.R. Coding of odors by a receptor repertoire. Cell. 2006;125:143–60. [PubMed: 16615896]
  • Hopfield J.J. Pattern recognition computation using action potential timing for stimulus representation. Nature. 1995;376:33–36. [PubMed: 7596429]
  • Jones W.D., Cayirlioglu P., Kadow I.G., Vosshall L.B. Two chemosensory receptors together mediate carbon dioxide detection in drosophila. Nature. 2007;445:86–90. [PubMed: 17167414]
  • Kay L.M., Laurent G. Odor- and context-dependent modulation of mitral cell activity in behaving rats. Nat Neurosci. 1999;2:1003–9. [PubMed: 10526340]
  • Kay L.M., Stopfer M. Information processing in the olfactory systems of insects and vertebrates. Semin Cell Dev Biol. 2006;17:433–42. [PubMed: 16766212]
  • Kepecs A., Uchida N., Mainen Z.F. Rapid and precise control of sniffing during olfactory discrimination in rats. J Neurophysiol. 2007;98:205–13. [PubMed: 17460109]
  • Kepecs A., van Rossum M.C., Song S., Tegner J. Spike-timing-dependent plasticity: Common themes and divergent vistas. Biol Cybern. 2002;87:446–58. [PubMed: 12461634]
  • Kobayakawa K., Kobayakawa R., Matsumoto H., Oka Y., Imai T., Ikawa M., Okabe M., Ikeda T., Itohara S., Kikusui T., Mori K., Sakano H. Innate versus learned odour processing in the mouse olfactory bulb. Nature. 2007;450:503–8. [PubMed: 17989651]
  • Koehl M.A., Koseff J.R., Crimaldi J.P., McCay M.G., Cooper T., Wiley M.B., Moore P.A. Lobster sniffing: Antennule design and hydrodynamic filtering of information in an odor plume. Science. 2001;294:1948–51. [PubMed: 11729325]
  • Kwon J.Y., Dahanukar A., Weiss L.A., Carlson J.R. The molecular basis of CO2 reception in Drosophila. Proc Natl Acad Sci USA. 2007;104:3574–78. [PMC free article: PMC1805529] [PubMed: 17360684]
  • Lagier S., Carleton A., Lledo P.M. Interplay between local GABAergic interneurons and relay neurons generates gamma oscillations in the rat olfactory bulb. J Neurosci. 2004;24:4382–92. [PubMed: 15128852]
  • Lagier S., Panzanelli P., Russo R.E., Nissant A., Bathellier B., Sassoe-Pognetto M., Fritschy J.M., Lledo P.M. GABAergic inhibition at dendrodendritic synapses tunes gamma oscillations in the olfactory bulb. Proc Natl Acad Sci USA. 2007;104:7259–64. [PMC free article: PMC1855399] [PubMed: 17428916]
  • Laurent G. Dynamical representation of odors by oscillating and evolving neural assemblies. Trends Neurosci. 1996;19:489–96. [PubMed: 8931275]
  • Laurent G., Davidowitz H. Encoding of olfactory information with oscillating neural assemblies. Science. 1994;265:1872–75. [PubMed: 17797226]
  • Laurent G., Naraghi M. Odorant-induced oscillations in the mushroom bodies of the locust. J Neurosci. 1994;14:2993–3004. [PubMed: 8182454]
  • Lin D.Y., Zhang S.Z., Block E., Katz L.C. Encoding social signals in the mouse main olfactory bulb. Nature. 2005;434:470–77. [PubMed: 15724148]
  • Litaudon P., Garcia S., Buonviso N. Strong coupling between pyramidal cell activity and network oscillations in the olfactory cortex. Neuroscience. 2008;156:781–87. [PubMed: 18790020]
  • Macrides F., Chorover S.L. Olfactory bulb units: Activity correlated with inhalation cycles and odor quality. Science. 1972;175:84–87. [PubMed: 5008584]
  • Margrie T.W., Schaefer A.T. Theta oscillation coupled spike latencies yield computational vigour in a mammalian sensory system. J Physiol. 2003;546:363–74. [PMC free article: PMC2342519] [PubMed: 12527724]
  • Mazor O., Laurent G. Transient dynamics versus fixed points in odor representations by locust antennal lobe projection neurons. Neuron. 2005;48:661–73. [PubMed: 16301181]
  • Meredith M. Patterned response to odor in mammalian olfactory bulb: The influence of intensity. J Neurophysiol. 1986;56:572–97. [PubMed: 3537224]
  • Mombaerts P. Genes and ligands for odorant, vomeronasal and taste receptors. Nat Rev Neurosci. 2004a;5:263–78. [PubMed: 15034552]
  • Mombaerts P. Odorant receptor gene choice in olfactory sensory neurons: The one receptor-one neuron hypothesis revisited. Curr Opin Neurobiol. 2004b;14:31–36. [PubMed: 15018935]
  • Neville K.R., Haberly L.B. Beta and gamma oscillations in the olfactory system of the urethane-anesthetized rat. J Neurophysiol. 2003;90:3921–30. [PubMed: 12917385]
  • Olsen S.R., Bhandawat V., Wilson R.I. Excitatory interactions between olfactory processing channels in the Drosophila antennal lobe. Neuron. 2007;54:89–103. [PMC free article: PMC2048819] [PubMed: 17408580]
  • Olsen S.R., Wilson R.I. Lateral presynaptic inhibition mediates gain control in an olfactory circuit. Nature. 2008;452:956–60. [PMC free article: PMC2824883] [PubMed: 18344978]
  • Onoda N., Mori K. Depth distribution of temporal firing patterns in olfactory bulb related to airintake cycles. J Neurophysiol. 1980;44:29–39. [PubMed: 7420137]
  • Perez-Orive J., Mazor O., Turner G.C., Cassenaer S., Wilson R.I., Laurent G. Oscillations and sparsening of odor representations in the mushroom body. Science. 2002;297:359–65. [PubMed: 12130775]
  • Rall W., Shepherd G.M. Theoretical reconstruction of field potentials and dendrodendritic synaptic interactions in olfactory bulb. J Neurophysiol. 1968;31:884–915. [PubMed: 5710539]
  • Ravel N., Chabaud P., Martin C., Gaveau V., Hugues E., Tallon-Baudry C., Bertrand O., Gervais R. Olfactory learning modifies the expression of odour-induced oscillatory responses in the gamma (60–90 Hz) and beta (15–40 Hz) bands in the rat olfactory bulb. Eur J Neurosci. 2003;17:350–58. [PubMed: 12542672]
  • Rennaker R.L., Chen C.F., Ruyle A.M., Sloan A.M., Wilson D.A. Spatial and temporal distribution of odorant-evoked activity in the piriform cortex. J Neurosci. 2007;27:1534–42. [PMC free article: PMC2291208] [PubMed: 17301162]
  • Rinberg D., Koulakov A., Gelperin A. Speed-accuracy tradeoff in olfaction. Neuron. 2006;51:351–58. [PubMed: 16880129]
  • Rodriguez I. Odorant and pheromone receptor gene regulation in vertebrates. Curr Opin Genet Dev. 2007;17:465–70. [PubMed: 17709237]
  • Schlief M.L., Wilson R.I. Olfactory processing and behavior downstream from highly selective receptor neurons. Nat Neurosci. 2007;10:623–30. [PMC free article: PMC2838507] [PubMed: 17417635]
  • Scott J.W., Acevedo H.P., Sherrill L. Effects of concentration and sniff flow rate on the rat electroolfactogram. Chem Senses. 2006;31:581–93. [PMC free article: PMC2225541] [PubMed: 16740644]
  • Shepherd G.M. Synaptic organization of the mammalian olfactory bulb. Physiol Rev. 1972;52:864–917. [PubMed: 4343762]
  • Spors H., Grinvald A. Spatio-temporal dynamics of odor representations in the mammalian olfactory bulb. Neuron. 2002;34:301–15. [PubMed: 11970871]
  • Spors H., Wachowiak M., Cohen L.B., Friedrich R.W. Temporal dynamics and latency patterns of receptor neuron input to the olfactory bulb. J Neurosci. 2006;26:1247–59. [PubMed: 16436612]
  • Stopfer M., Bhagavan S., Smith B.H., Laurent G. Impaired odour discrimination on desynchronization of odour-encoding neural assemblies. Nature. 1997;390:70–74. [PubMed: 9363891]
  • Stopfer M., Jayaraman V., Laurent G. Intensity versus identity coding in an olfactory system. Neuron. 2003;39:991–1004. [PubMed: 12971898]
  • Stopfer M., Laurent G. Short-term memory in olfactory network dynamics. Nature. 1999;402:664–68. [PubMed: 10604472]
  • Suh G.S., Ben-Tabou de Leon S., Tanimoto H., Fiala A., Benzer S., Anderson D.J. Light activation of an innate olfactory avoidance response in Drosophila. Curr Biol. 2007;17:905–8. [PubMed: 17493811]
  • Suh G.S., Wong A.M., Hergarden A.C., Wang J.W., Simon A.F., Benzer S., Axel R., Anderson D.J. A single population of olfactory sensory neurons mediates an innate avoidance behaviour in Drosophila. Nature. 2004;431:854–59. [PubMed: 15372051]
  • Suri R.E., Sejnowski T.J. Spike propagation synchronized by temporally asymmetric Hebbian learning. Biol Cybern. 2002;87:440–45. [PMC free article: PMC2944018] [PubMed: 12461633]
  • Toyoizumi T., Pfister J.P., Aihara K., Gerstner W. Generalized Bienenstock-Cooper-Munro rule for spiking neurons that maximizes information transmission. Proc Natl Acad Sci USA. 2005;102:5239–44. [PMC free article: PMC555686] [PubMed: 15795376]
  • Uchida N., Mainen Z.F. Speed and accuracy of olfactory discrimination in the rat. Nat Neurosci. 2003;6:1224–29. [PubMed: 14566341]
  • VanRullen R., Guyonneau R., Thorpe S.J. Spike times make sense. Trends Neurosci. 2005;28:1–4. [PubMed: 15626490]
  • Varela F., Lachaux J.P., Rodriguez E., Martinerie J. The brainweb: Phase synchronization and large-scale integration. Nat Rev Neurosci. 2001;2:229–39. [PubMed: 11283746]
  • Verhagen J.V., Wesson D.W., Netoff T.I., White J.A., Wachowiak M. Sniffing controls an adaptive filter of sensory input to the olfactory bulb. Nat Neurosci. 2007;10:631–39. [PubMed: 17450136]
  • Vickers N.J., Christensen T.A., Baker T.C., Hildebrand J.G. Odour-plume dynamics influence the brain’s olfactory code. Nature. 2001;410:466–70. [PubMed: 11260713]
  • Vosshall L.B., Stocker R.F. Molecular architecture of smell and taste in Drosophila. Annu Rev Neurosci. 2007;30:505–33. [PubMed: 17506643]
  • Wachowiak M., Cohen L.B. Representation of odorants by receptor neuron input to the mouse olfactory bulb. Neuron. 2001;32:723–35. [PubMed: 11719211]
  • Warzecha A.-K., Egelhaaf M. Variability in spike trains during constant and dynamic stimulation. Science. 1999;283:1927–30. [PubMed: 10082467]
  • Wesson D.W., Carey R.M., Verhagen J.V., Wachowiak M. Rapid encoding and perception of novel odors in the rat. PLoS Biol. 2008a;6:e82. [PMC free article: PMC2288628] [PubMed: 18399719]
  • Wesson D.W., Verhagen J.V., Wachowiak M. Why sniff fast? The relationship between sniff frequency, odor discrimination and receptor neuron activation in the rat. J Neurophysiol. 2008b;101:1089–1102. [PMC free article: PMC2657070] [PubMed: 19052108]
  • Wilson R.I., Laurent G. Role of GABAergic inhibition in shaping odor-evoked spatiotemporal patterns in the Drosophila antennal lobe. J Neurosci. 2005;25:9069–79. [PubMed: 16207866]
  • Wilson R.I., Turner G.C., Laurent G. Transformation of olfactory representations in the Drosophila antennal lobe. Science. 2004;303:366–70. [PubMed: 14684826]
  • Zhigulin V.P., Rabinovich M.I., Huerta R., Abarbanel H.D. Robustness and enhancement of neural synchronization by activity-dependent coupling. Phys Rev E Stat Nonlin Soft Matter Phys. 2003;67:021901. [PubMed: 12636709]
Copyright © 2010 by Taylor and Francis Group, LLC.
Bookshelf ID: NBK55968PMID: 21882423


  • PubReader
  • Print View
  • Cite this Page

Other titles in this collection

Related information

  • PMC
    PubMed Central citations
  • PubMed
    Links to PubMed

Similar articles in PubMed

See reviews...See all...

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...