|Home | About | Journals | Submit | Contact Us | Français|
Sensation is an active process involving the selective sampling and central processing of external stimuli in space and time. Olfaction in particular depends strongly on active sensing due to the fact that - at least in mammals - inhalation of air into the nasal cavity is required for odor detection. This seemingly simple first step in odor sensation profoundly shapes nearly all aspects of olfactory system function, from the distribution of odorant receptors to the functional organization of central processing to the perception of odors. The dependence of olfaction on inhalation also allows for profound modulation of olfactory processing by changes in odor sampling strategies in coordination with attentional state and sensory demands. This review discusses the role of active sensing in shaping olfactory system function at multiple levels and draws parallels with other sensory modalities to highlight the importance of an active sensing perspective in understanding how sensory systems work in the behaving animal.
Sensory systems gather and process information about the external world. For most modalities, sensation is an active operation in which the detection, representation and processing of sensory information is heavily modulated during behavior. Active sensing allows an animal to selectively sample regions in space and epochs in time, to regulate stimulus intensity and dynamics in order to optimize sensory processing, to extract features of interest from a complex stimulus and to protect sensory neurons from excessively strong or harmful stimuli. Classic examples of active sensation include finger movements during object manipulation, whisking in rodents and saccadic eye movements in vision. Each of these examples involves movement of the sense organs in order to optimally sample an area or object of interest. Active stimulus sampling can profoundly affect patterns of sensory neuron activation and, consequently, the postsynaptic processing of sensory inputs. In addition, active sensing involves the coordination of ‘bottom-up’ effects on sensory inputs with ‘top-down’ modulation of processing at multiple synaptic levels. Thus active sensation is a multi-level, system-wide process affecting sensory system function.
Olfaction, while not as extensively studied as other modalities, is in many respects an ideal model system for active sensing. First, for terrestrial vertebrates, olfactory sensation depends on stimulus acquisition by the animal; the inhalation of air into the nose is a necessary first step in olfaction. Second, mammals in particular have impressively complex behavioral repertoires for odorant sampling; this behavior - typically termed ‘sniffing’ - is precisely and strongly modulated as a function of task demands, behavioral state and stimulus context (Welker, 1964; Wesson et al., 2009; Youngentob et al., 1987). Finally, the olfactory system has in recent years matured into a highly tractable system in which its molecular, cellular and circuit-level organization can be examined, manipulated and integrated with behavioral experiments.
A central thesis of this review is that the active components of olfactory sensation are closely woven with fundamental processes of olfactory system function at levels ranging from receptor expression patterns, sensory neuron response properties, circuit dynamics in the olfactory bulb and cortex, and centrifugal systems. As a result, the reliance of olfaction on transient, active sampling of odors is manifest even in reduced experimental preparations that are far removed from an actively sampling animal. Thus considering olfaction as an active sense is not only essential to understanding how this system works in the behaving animal, it is a useful framework for understanding olfaction in many experimental contexts. A second point made here - and substantiated by examples from other sensory modalities - is that even descriptions of olfactory system function in the awake animal would benefit from considering sampling behavior as a primary factor in shaping how the brain represents and processes olfactory input. In general, considering sensory systems in the context of active sensing provides an important avenue for understanding key principles of sensory system function in the behaving animal.
In terrestrial vertebrates the olfactory epithelium is housed deep within the nasal cavity, such that inhalation of air is required for odorants to access olfactory receptor neurons (ORNs). Typically, this can only occur during the course of resting respiration or by the voluntary inhalation of air in the context of odor-guided behavior - i.e., sniffing. A sniff - like a whisk or a saccade - represents the basic unit of active odor sensing. Analogs of sniffing occur across the animal kingdom, with groups as diverse as crustaceans (Snow, 1973), fish (Nevitt, 1991), semi-aquatic mammals (Catania, 2006) and insects (Suzuki, 1975) showing active, intermittent odorant sampling. The persistence of sniffing behavior in different species and ecological settings together with its strong modulation during odor-guided behaviors suggests that intermittent sampling of odorant is fundamentally important to olfaction (Dethier, 1987).
Sniffing - while highly dynamic from cycle to cycle is precisely controlled during behavior (Figure 1). For example, when sampling odorant from a port in an odor discrimination task, rats show a brief bout of 6 – 10 Hz sniffing precisely timed to just precede odorant delivery and a slightly higher-frequency sniff bout (9 – 12 Hz) just prior to receiving a reward; each of these bouts is repeated with a temporal jitter of only a few hundred msec across hundreds of trials (Kepecs et al., 2007; Wesson et al., 2009). Humans also show stereotyped and task-dependent sniffing patterns and also can rapidly modulate sniffing in response to sensory input (Johnson et al., 2003; Laing, 1983).
Sniffing patterns thus reflect a particular strategy for olfactory sampling, chosen for a particular task and context. Sniffing strategies can also be individual-specific: both rodents and humans show individual differences in sniffing behavior when sampling odorants (Laing, 1983; Wesson et al., 2009). A compelling example of context-specific sampling strategies occurs in bird-hunting dogs: when tracking the scent of prey on the ground, dogs sniff at up to 4 – 6 Hz but when tracking the same scent in the air, the animal will raise its head and run forward, forcing a continuous stream of air into the nose for up to 40 sec (Steen et al., 1996). The presumed advantage of the latter strategy is to enable continuous odorant sampling while moving at high speed and to decouple sampling from respiration during a time of heavy load on the respiratory system.
Sniffing patterns - like saccadic eye movements in visual scene analysis and repeated whisking during somatosensory object identification - likely reflect strategies for optimally extracting and processing sensory information (Laing, 1983). How, then, does sniffing affect the detection, representation and processing of odor information by the nervous system? This question remains largely unanswered but crucial to understanding the role of active sensing in olfactory system function.
Addressing this question involves some important caveats, however. First, unlike in other sensory systems, active sampling in olfaction is confounded with an arguably more important function: respiration. In rodents, which are obligatory nose-breathers, odorants are unavoidably sampled with each inhalation (Verhagen et al., 2007) and sniffing strategies will be constrained within the limits required for proper respiratory function. For example, sniff frequency may increase in animals that are actively engaged with their environment due simply to increased respiratory demand. Autonomic or reflex-mediated effects on respiration might also be confused with active sniffing. Second, in the freely-moving animal, sniffing is expressed as part of a larger behavioral repertoire which may include head movements, whisking (in rodents), licking, and locomotion (Bramble and Carrier, 1983; Welker, 1964). The strong coupling between sniffing and other active sampling behaviors can confound interpretation of the role that sniffing plays in olfaction. Rodents increase respiration frequency prior to receiving a reward and when otherwise engaged in motivated behavior, independent of an olfactory context (Clarke, 1971; Kepecs et al., 2007; Wesson et al., 2008b) (Figure 1D). Rodents also increase respiration frequency (and initiate whisking) in response to unexpected stimuli of any modality (Macrides, 1975; Welker, 1964) and when inserting their nose into a port - even when performing non-olfactory tasks (Wesson et al., 2008b; Wesson et al., 2009) (Figure 1E). Finally, rodents and humans can make odor-guided decisions after only a single sample of odorant, which can occur via an inhalation that is indistinguishable from that of resting respiration (Verhagen et al., 2007). Thus, while in this review we use ‘sniffing’ to imply a voluntary inhalation (or repeated inhalations) in the context of odor-guided behavior, we include passive respiration as an effective means of olfactory sampling.
The most important function of sniffing is to control access of olfactory stimuli to the receptor neurons (ORNs) themselves. At least in awake rodents, ORNs are not activated when odorant is simply blown at the nose; the animal must inhale for odorant to reach the olfactory epithelium (Wesson et al., 2008a) (Figure 2A). Inhalation-driven ORN responses are transient, with each inhalation evoking a burst of ORN activity lasting only 100 – 200 ms (Carey et al., 2009; Chaput and Chalansonnet, 1997; Verhagen et al., 2007) (Figure 2B). Up to several thousand ORNs - each expressing the same odorant receptor - converge onto a single glomerulus in the OB (Mombaerts et al., 1996). An important aspect of inhalation-driven sensory activity is that the activation of the ORN population that converges onto one glomerulus is not instantaneous but instead develops over 40 – 150 msec (Carey et al., 2009). As a result, patterns of sensory input to OB glomeruli dynamically develop over the 50 – 200 milliseconds following an inhalation (Figure 2A).
Temporal coupling between the dynamics of neural activity in the olfactory pathway and rhythmic odor sampling is the most distinctive feature of odorant-evoked activity in the CNS (Adrian, 1942; Buonviso et al., 2006; Macrides and Chorover, 1972). While long controversial, most recent evidence suggests that respiration-related rhythms in the CNS are dependent on the activation of ORNs during inhalation, as opposed to reflecting a central signal related to respiratory pattern-generation (Buonviso et al., 2006). In rodents, eliminating ORN activation or decoupling nasal airflow from respiration disrupts respiratory rhythms in the olfactory pathway in favor of nasal airflow rhythms (Grosmaitre et al., 2007; Sobel and Tank, 1993; Spors and Grinvald, 2002). Thus, olfactory network dynamics are primarily driven by the dynamics of inhalation-driven ORN input (Figures 2C, D). For example, the rise-time of odorant-evoked EPSPs in mitral/tufted (MT) cells - the principal OB output neuron - of anesthetized rats is approximately 100 msec, similar to that of the ORN response transients (Cang and Isaacson, 2003; Margrie and Schaefer, 2003). MT cells also show variation in temporal response patterns (e.g. - latency, rise-time and duration of an excitatory burst) that is unit- and odorant-specific (Bathellier et al., 2008; Macrides and Chorover, 1972) and varies over a range similar to that of ORNs (Carey and Wachowiak, 2011). Finally, pyramidal neurons in piriform cortex (PC) - a major target of OB output - also show strong inhalation-coupled dynamics in their spike output and in subthreshold synaptic inputs (Poo and Isaacson, 2009; Rennaker et al., 2007) (Figure 2D).
Given the temporal constraints on ORN responses imposed by respiration it seems likely that postsynaptic networks will be optimized for such input dynamics. Indeed, while the canonical view of the OB network has been that it shapes MT response properties in the spatial domain - e.g. relative to activity in other glomeruli and their associated MT cells (Johnson and Leon, 2007; Yokoi et al., 1995), recent data suggest that postsynaptic processing may primarily function to shape responses in the temporal domain relative to inhalation-driven bursts of input (Figure 3). Work supporting this view comes largely from experimental paradigms far removed from ‘active’ sensing. For example, in OB slice preparations, delivering patterned olfactory nerve stimulation at frequencies that mimic resting respiration amplifies MT responses to ORN input and leads to increased synchrony of MT firing and the emergence of gamma-frequency oscillations in MT cell membrane potential (Hayar et al., 2004b; Schoppa, 2006b).
Neurons in PC - the major cortical target of OB output neurons - also appear optimized to process information in a temporal domain organized around inhalation-driven bursts of input from MT cells. For example, MT cell axons from the OB provide direct but selective excitation to pyramidal neurons in PC while also driving more widespread feedforward inhibition via GABA-ergic local interneurons (Poo and Isaacson, 2009). For sparse and temporally unstructured MT cell inputs to PC, this strong feedforward inhibition creates an extremely short (5 – 10 ms) time window during which pyramidal neurons may integrate M/T inputs from the OB. However, if MT cell inputs to PC pyramidal neurons occur in sustained bursts - such as those as evoked by inhalation - this feedforward inhibition depresses while MT cell input to pyramidal neurons facilitates, allowing MT cell inputs to more effectively drive pyramidal neuron spiking (Stokes and Isaacson, 2010)(Figures 3B, C). This dynamic synaptic organization may act as a filter for OB inputs to PC that show strong inhalation-coupled temporal patterning.; this prediction could be tested in vivo by comparing PC neuron responses to inhalation-driven, odorant-evoked inputs with responses to brief electrical stimulation of MT axons.
There is strong behavioral evidence that a single, inhalation-driven ‘packet’ of activity can encode odor information sufficiently to support odor discrimination. Rodents (and humans) can discriminate two familiar odors after a single sniff and in as little as 150 – 250 msec (Abraham et al., 2004; Laing, 1986; Rinberg et al., 2006b; Uchida and Mainen, 2003). In fact, behavioral measurements of odor perception times in awake rats performed simultaneous with imaging of ORN inputs to the OB, indicate that a novel odor can be distinguished from a familiar one before the initial ORN response burst - as inferred from presynaptic calcium imaging - has even finished (Wesson et al., 2008a). In addition, rats performing an operant, two-choice odor discrimination task tend to make their choice after only a single sniff when that sniff evokes strong neural responses within an optimal time-window after inhalation but not when it evokes activity at later times (Cury and Uchida, 2010). Thus, the initial onset phase of the inhalation-evoked burst of ORN activity appears particularly important for olfactory processing and odor perception. Rodents require more time - an additional 100 – 200 msec - to discriminate highly similar odors (Abraham et al., 2004; Rinberg et al., 2006b)(Figure 3D). This additional time roughly matches the time-window over which patterns of ORN input and MT cell activity evolve after an inhalation (Figure 3A) (Cury and Uchida, 2010; Shusterman et al., 2011; Wesson et al., 2008a). Thus, the dynamics of inhalation-evoked ORN inputs to the OB may set an upper limit on the time-window for integration of odor information in the behaving animal (Schaefer and Margrie, 2007).
One longstanding - and still unresolved - question is whether the precise timing of odorant-evoked activity relative to the timing of inhalation plays a role in odor perception. Modeling and experimental data support the idea that spike timing relative to inhalation can robustly represent odor information (Chaput, 1986; Hopfield, 1995; Schaefer and Margrie, 2007; Shusterman et al., 2011). However, whether animals actually use a sniff-based temporal code remains unclear. Important evidence in support of such a coding strategy comes from a recent study using optogenetics in awake, head-fixed mice to activate the same ORN inputs at different times relative to inhalation or exhalation onset (Smear et al., 2011). Mice were able to perceive ORN inputs activated at different times after inhalation, with some mice able to discriminate input latency differences of as little as 10 msec; responses of individual OB units also showed sensitivity to the timing of ORN inputs relative to inhalation. Thus, the timing of sensory inputs relative to odorant sampling is, by itself, sufficient to mediate odor discrimination in the awake animal.
The fact that odor encoding and perception can occur after a single inhalation begs the question of why behaving animals modulate their sniffing behavior so profoundly when sampling odors. Here we discuss several hypotheses on how active control of sniff parameters shape the initial odor representations formed by ORNs; the following section discusses the consequences of changing sniffing patterns for the central processing of olfactory inputs.
One longstanding hypothesis is that animals actively shape ORN response patterns by modulating the rate of air flow over the olfactory epithelium and subsequently altering how odorant distributes across it (Adrian, 1950; Mozell, 1964). This idea - which we will call the sorption hypothesis - arises from the fact that the nasal cavity of most vertebrates – mammals in particular - is anatomically complex and forms a narrow space lined with epithelium and mucus onto which odorant molecules absorb as they flow through the cavity (Yang et al., 2007; Zhao et al., 2006). This arrangement causes a ‘chromatographic effect’ in which odorants are preferentially absorbed in different locations depending on their solubilities and their flow rate (Mozell and Jagodowicz, 1973; Yang et al., 2007). The topography of odorant receptor expression across the olfactory epithelium correlates with the areas of maximal sorption for the receptors’ respective ligands, suggesting that receptors are optimally localized to take advantage of the chromatographic effect (Schoenfeld and Cleland, 2006; Scott et al., 2000). Because the strength, duration and frequency of respiration can change dramatically during odor-guided behavior and because these parameters affect the rate and total volume of airflow into and out of the nasal cavity, sampling behavior has the potential to alter odorant sorption and, as a consequence, patterns of ORN activation (Mozell et al., 1987; Youngentob et al., 1987).
The sorption hypothesis makes specific predictions about how flow rate should shape activity in the intact animal, and applies to both rodent models and humans (Hahn et al., 1994; Mozell et al., 1987). The most directly testable is the following: at low flow rates, strongly-sorbed odorants - for example, polar compounds such as alcohols - will be largely removed from the air stream as they pass through the nasal cavity, resulting in fewer odorant molecules available to activate ORNs, particularly those positioned later in the path of airflow. Because less sorption occurs at higher flow rates, sniffs that elicit higher flows will bring more odorant molecules to ORNs and so evoke larger responses. In contrast, weakly-sorbed odorants - for example the terpene d-limonene, a principal component of orange odor - absorb slowly onto the epithelium and so tend to remain in the air stream as inhaled odorant passes through the nasal cavity. For these compounds, increasing flow rate will have little effect on odorant deposition. Thus, responses to a strongly-sorbed odorant should increase as flow rate increases, while responses to a weakly-sorbed odorant should remain constant or decrease (Hahn et al., 1994). Such effects have been measured at the level of the olfactory epithelium in reduced rodent preparations (Kent et al., 1996; Scott-Johnson et al., 2000) and, recently, in the olfactory bulb using artificial inhalation (Oka et al., 2009). The sorption hypothesis remains untested during natural odor sampling, however, with earlier studies relying primarily on steady-state flow rates, not the transient changes in airflow that occur during natural respiration and active sniffing. Thus, whether animals modulate sniff flow rate in order to actively modulate odorant response patterns remains unclear.
A second way in which sniffing behavior can alter ORN response patterns is through changes in sniff frequency. High-frequency (6 – 10 Hz) sniff bouts lasting up to several seconds are one of the most distinctive odor sampling strategies in mammals, particularly during exploratory behavior (Macrides, 1975; Welker, 1964). High-frequency sniffing shapes ORN responses in unexpected ways. An intuitive prediction is that increases in sniff frequency lead to increased ORN responses - and perhaps recruitment of activation of new ORN populations - due to an increased odorant influx. This prediction has been tested using presynaptic calcium imaging from ORN axon terminals in the OB of awake rats, which sampled the same odorant during low-frequency (1 – 2 Hz) respiration or during high-frequency (4 – 8 Hz) exploratory sniffing (Verhagen et al., 2007). Surprisingly, sampling an odorant at high-frequency only weakly enhanced the initial response to the odorant and did not recruit activation of new ORN populations. More importantly, sustained high-frequency sniffing of odorant led to a strong attenuation of ORN response magnitude (Figure 4A). Sniff frequency-dependent attenuation is rapidly reversible, with ORN response magnitudes recovering within one second after sniffing returns to below 4 Hz. A likely cellular mechanism mediating the frequency-dependent attenuation of ORN inputs to the OB is simple adaptation. At low respiration rates, ORNs can recover from adaptation in the interval between successive inhalations, but higher sniff frequencies allow less time for recovery between cycles (Reisert and Matthews, 2001).
What is the functional significance of peripheral frequency-dependent attenuation? The attenuation of ORN responses is specific to those glomeruli receiving odorant-evoked input, leaving other glomeruli free to respond to other odorants encountered during a sniff bout - for example, as an animal explores its environment. Thus, this attenuation constitutes an ‘adaptive filter’ of sensory input to the OB in which ORNs activated by odorants present at the beginning of exploratory sniffing (i.e. - ‘background’ odorants) are selectively suppressed in the representation of subsequently-sampled odorants (Verhagen et al., 2007). In contrast, during low-frequency sampling, odorants encountered against a background are represented as the sum of the background and ‘foreground’ response maps (Figure 4B). This filtering can enhance the contrast between odorants having overlapping molecular features (or mixtures with shared components). An equally important function of frequency-dependent attenuation may be to increase the salience of temporally dynamic or spatially localized odorants relative to broadly distributed background odorants. This effect is similar to that seen in active vision, in which repeated scanning of a complex visual scene induces adaptation to the scene statistics and increases the salience of novel stimuli appearing against this background (McDermott et al., 2010). Thus sniffing provides a bottom-up mechanism for the active modulation of odor salience.
Finally, odor representations may depend on whether odorants are sampled via inhalation of odorant through the nose - ‘orthonasal’ sampling - or via the oral cavity and through the nasopharynx - ‘retronasal’ sampling (Hummel, 2008). Retronasal odor sampling can occur during odorant exhalation or, as is more typically considered, after the release of odorant vapor from ingested liquids or solids; retronasally-sampled odorants are large contributors to flavor perception in humans (Murphy et al., 1977). Evidence from humans suggests that odors sampled orthonasally are perceived differently from those sampled retronasally, with retronasal odors perceived as less intense and originating from the oral cavity rather than externally (Murphy et al., 1977; Small et al., 2005). Ortho-versus retronasally sampled odors differentially activate brain areas involved in odor and flavor perception, suggesting that the route of odorant sampling can also impact central processing of odor information (Small et al., 2005). The specific role that retronasal olfaction plays in odor and flavor perception, including whether it is under active control during behavior, remains unclear, however. Retronasal odor sampling may also represent an important difference between human and rodent olfaction: in humans, both inhaled and exhaled air pass over the olfactory epithelium, while in rodents and other macrosmatic animals exhaled air largely bypasses the olfactory recess, severely limiting retronasal access of odorants to ORNs (Zhao et al., 2004; Craven et al., 2010).
Because the central circuits that process olfactory inputs are themselves dynamic, changes in the strength and temporal structure of ORN input during active sniffing should also change how these networks process olfactory information. For example, in the rodent somatosensory system, high-frequency inputs such as those that occur during active whisking lead to reduced responsiveness in cortical pyramidal neurons due to dynamic network properties such as short-term depression at the thalamocortical synapse and changes in the driving force of excitatory versus inhibitory inputs (Chung et al., 2002; Crochet et al., 2011). Frequency-dependent effects on olfactory network dynamics have primarily been studied in the OB (Figure 5), although olfactory processing in the PC likely also depends on sniff frequency.
Predicted effects of sniff frequency on OB processing arise from experiments in anesthetized animals or slice preparations in which sniff frequency is mimicked with pulsed electrical stimulation or direct current injection (Balu et al., 2004; Hayar et al., 2004b; Margrie and Schaefer, 2003). These studies have led to predictions that increasing sniff frequency will have distinct, cell-type-specific effects on the strength of odorant-evoked activity and the coherence of activity across a population of neurons within the OB. For example, granule cells - GABA-ergic interneurons thought to mediate feedback- and lateral inhibition of MT cells - show increased synchrony and stronger inhibition onto MT cells at synaptic input frequencies corresponding to active sniffing (Young and Wilson, 1999; Schoppa, 2006a). In addition, MT cells themselves show increased spike output and temporal precision as input frequency increases into the range of active sniffing (Balu et al., 2004) (Figure 5C).
Another important element mediating sniff frequency-dependent changes in OB processing is the external tufted (ET) cell - an excitatory interneuron in the glomerular-layer. ET cells can drive direct feedforward excitation as well as indirect (disynaptic) feedforward inhibition of MT cells and are thus potent regulators of MT excitability (Hayar et al., 2004a; Najac et al., 2011). ET cells show spontaneous spike bursts but their bursts become increasingly entrained to rhythmic ORN inputs as input frequency increases (Hayar et al., 2004b), leading to an increase both in their excitation of MT cells and their activation of inhibitory periglomerular interneurons (PG cells) (Hayar et al., 2004a). In vivo, this effect is predicted to generate an increasingly sharp time-window over which MT cells integrate ORN inputs and may also increase the strength of lateral inhibition between glomeruli (Wachowiak and Shipley, 2006).
Overall, the consensus prediction from these circuit-level studies is that frequency-dependent effects within the OB network serve to enhance the inhalation-driven temporal patterning of ORN inputs and increase the reliability and temporal precision of MT cell firing relative to inhalation onset (Balu et al., 2004; Schaefer et al., 2006; Wachowiak and Shipley, 2006). MT cell recordings from anesthetized rats during artificial sniffing at different frequencies have found frequency-dependent changes in MT response dynamics consistent with this prediction (Bathellier et al., 2008; Carey and Wachowiak, 2011) (Figure 5D). Thus, the circuit organization and dynamics of central olfactory networks appear optimized to process sensory inputs organized by inhalation.
While data from slice experiments, anesthetized animals and computational studies all point to the fundamental importance of sniff-driven dynamics in shaping odor information processing, integrating these results with data from awake animals in which sampling behavior is truly ‘active’ (and highly variable) remains a major challenge. For example, no studies in awake animals have systematically explored the relationship between a particular parameter of sniffing behavior and circuit interactions in the OB or PC. In addition, slice experiments that mimic sniffing with electrical or optogenetic stimulation typically use synchronous activation of many neurons to mimic a sniff (Hayar et al., 2004b; Young and Wilson, 1999) rather than the slowly-rising, inhalation-driven packets of ORN input that develop over ~100 msec in vivo (Carey et al., 2009). Significant changes in synaptic transmission can develop during this time window - for example, synaptic depression and presynaptic inhibition of transmitter release from ORNs (Murphy et al., 2004; Wachowiak et al., 2005); these effects are not apparent following single shocks to the olfactory nerve. Extrapolating response properties from slice experiments or anesthetized animals to behaving animals is also complicated by differences in the spontaneous activity of MT cells, other interneurons, and centrifugal inputs to the OB and PC in awake versus anesthetized or slice preparations (Davison and Katz, 2007; Rinberg and Gelperin, 2006).
Nonetheless, many of the basic response properties of ORNs as well as MT and PC neurons are similar in anesthetized and awake animals. Inhalation-driven ORN responses show identical latencies and burst durations in awake and anesthetized rodents and similar degrees of frequency-dependent attenuation of response strength (Carey et al., 2009; Verhagen et al., 2007). Likewise, MT cells recorded from anesthetized and awake rodents show nearly identical response dynamics relative to inhalation in terms of their range of response latencies, duration and precision of spike timing (Carey and Wachowiak, 2011; Shusterman et al., 2011). Strategies of odor identity coding also appear similar in awake and anesthetized preparations, with MT cells showing roughly similar response specificities (Davison and Katz, 2007). Importantly, many of these similarities only become apparent when considered relative to inhalation or sniffing (Cury and Uchida, 2010; Shusterman et al., 2011); earlier studies that did not precisely monitor sniff timing noted significant differences in response features between anesthetized and awake animals (Rinberg et al., 2006a). Thus, while understanding olfactory system function in any realistic context will require work in behaving animals, anesthetized and slice preparations remain important tools by enabling systematic exploration of network dynamics and more direct probing of olfactory circuits. Considering these results from the perspective of active sensing and, specifically, sniff timing, appears key to integrating data across paradigms.
Thus far we have considered active sensing as a ‘bottom-up’ process in which the physical aspects of stimulus sampling shape sensory neuron activation and, subsequently, central processing. However, active sensing in any modality also involves ‘top-down’ mechanisms, which modulate sensory processing in coordination with stimulus sampling and other behavioral states. While ‘bottom-up’ processes are, as we have seen, amenable to a range of experimental approaches, investigating ‘top-down’ processes ultimately requires work in the awake animal, in which the systems modulating these processes are functioning normally. While the modulation of olfactory processing has been extensively studied - in particular in the rodent OB - much of this work has been performed in anesthetized animals and relatively little has been performed or interpreted in the context of active sensing, in which sensory processing is modulated in precise coordination with sampling behavior. Here we discuss potential pathways underlying the active modulation of olfactory processing, using parallels from other modalities - vision and somatosensation in particular - as instructive examples.
The modulation of sensory processing as a function of focal sampling in space or time has been termed ‘directed’ or ‘selective’ attention (Noudoost et al., 2010). For example, visual saccades involve directed attentional modulation of the responsiveness of visual neurons: responses of neurons with receptive fields in the region of spatial attention (e.g. - the target region of the saccade) show transient increases in sensitivity, while neurons with receptive fields in other regions show decreases in sensitivity (Winkowski and Knudsen, 2007; Noudoost et al., 2010). Similarly, cortical somatosensory neurons change their responsiveness to mechanosensory stimuli in the transition from passive to active touch mediated by reaching (in primates) or whisking (in rodents) (Hentschke et al., 2006; Nelson et al., 1991).
Like saccades and active touch, sniffing can provide an unambiguous and temporally precise behavioral readout of directed attention (Kepecs et al., 2007; Wesson et al., 2008a). In humans, anticipation of sniffing and attention to an olfactory task modulates activity in primary olfactory cortical areas (Zelano et al., 2005). Beyond these initial observations, however, attentional modulation of olfactory processing related to active sniffing remains largely unexplored. One prediction is that individual ‘active’ sniffs or high-frequency sniff bouts modulate odorant-evoked responses. A critical feature of directed attention in other modalities is that attentional modulation is transient and precisely timed to coincide with (and briefly precede) active sampling (Han et al., 2009). Tests for sniff-related modulation are less straightforward than for vision or touch because - at least in the awake mammal - inhalation is required to elicit odorant-evoked responses, precluding odorant presentation at different times relative to a sniff. Optogenetic approaches in which light is used to reliably activate sensory inputs independent of sniff timing (Smear et al., 2011) provide a promising solution to this problem.
What are the neural pathways underlying attentional modulation during active sensing?. In the heavily-studied visual system, multiple cortical as well as thalamic areas have been implicated in directed attention (Noudoost et al., 2010). One major source of attentional control is the frontal eye field - the premotor area controlling eye movements. In nonhuman primates, microstimulation of frontal eye field neurons enhances the responsiveness of visual cortex neurons with spatially overlapping receptive fields (Moore et al., 2003; Noudoost et al., 2010). In the somatosensory system, there are reciprocal connections between somatosensory neurons and the motor areas controlling active touch (Veinante and Deschênes, 2003). In addition, recent evidence has not only demonstrated monosynaptic connections between primary somatosensory and motor cortices corresponding to the same whisker (Ferezou et al., 2007) but direct control of whisker protraction by somatosensory cortex (Matyas et al., 2010). Thus, a tight coordination between the motor systems controlling stimulus sampling and the processing of incoming sensory signals mediated by this sampling is likely a fundamental component of top-down control in active sensing.
There is considerable evidence for coordination between olfactory sensory pathways and the motor systems controlling sniffing. First, in both humans and in rodents, olfactory stimuli can modulate sniffing behavior extremely quickly: humans show differences in the flow rate of inhalation that vary with odorant intensity within 200 msec after beginning an inhalation (Figure 6A) (Johnson et al., 2003); rats show an increase in sniff frequency in response to novel odorants in a similar time after inhalation and in as little as 50 – 100 msec after sensory input arrives at the OB (Figure 6B)(Wesson et al., 2008a). Motor signals related to sniffing also affect odor perception. For example, in human subjects in which odorant is injected into the bloodstream, sniffing can ‘gate’ odor perception (Mainland and Sobel, 2006). In addition, the degree of motor effort expended during a sniff affects perceived odor intensity (Hornung et al., 1997; Teghtsoonian and Teghtsoonian, 1984). Thus, motor information about sniffing is rapidly integrated with incoming sensory information and the motor component of sniffing appears to play an essential role in constructing an odor percept (Mainland and Sobel, 2006).
Potential pathways underlying sensorimotor integration during sniffing are outlined in Figure 6D. However, much of the neural circuitry mediating motor-related control of olfactory processing remains unclear, largely because the premotor control of sniffing is poorly understood. One important structure may be the cerebellum, which is activated during sniffing, may receive olfactory input from PC, and is involved in optimizing motor output for sensory acquisition in other modalities (Mainland and Sobel, 2006; Robinson, 1976; Sobel et al., 1998b). Sniffing is also likely under cortical control: several cortical areas, including the insular and infralimbic cortices, send projections to brainstem nuclei (such as the nucleus of the solitary tract) involved in respiratory pattern generation (Bianchi et al., 1995), and electrical stimulation of insular and infralimbic cortices alters respiration in anesthetized rats and can elicit increases in respiration frequency that mimic exploratory sniff bouts (Aleksandrov et al., 2007) (Figure 6C). Whether activation of these or other areas modulate processing in olfactory areas such as OB and PC remains untested.
Classical neuromodulatory pathways likely also play a role in directed attention. For example, cholinergic inputs to visual cortex alter visual responses via muscarinic acetylcholine receptors (Goard and Dan, 2009; Herrero et al., 2008), and dopaminergic signaling modulates the top-down control of visual responses by frontal eye field neurons (Noudoost and Moore, 2011). In the olfactory system, cholinergic, noradrenergic and serotonergic projections all target OB and PC (McLean and Shipley, 1992) and are capable of modulating olfactory responses (Chaudhury et al., 2009; Petzold et al., 2009; Shea et al., 2008) (Figure 6D). In addition, several olfactory cortical areas send strong centrifugal projections to the OB which are hypothesized to modulate odorant processing by affecting the strength of inhibition within the OB network (Strowbridge, 2009). Importantly, several of these areas are activated during sniffing in the absence of odorant, presumably by the airflow-driven, somatosensory component of a sniff (Adrian, 1942; Sobel et al., 1998a; Grosmaitre et al., 2007). Sniff-induced feedback from olfactory cortical areas to OB or PC might thus provide an alternate, ‘bottom-up’ mechanism by which a sniff can modulate olfactory processing.
The active control of stimulus sampling and sensory information processing is fundamental to all sensory systems, and investigating sensory function from a perspective of active sensing us important not only for understanding sensation in the behaving animal, but also for integrating data obtained with different experimental approaches and at different levels of the nervous system. In the mammalian olfactory system, the act for odor inhalation through respiration or sniffing is a fundamental feature that is reflected in its anatomical, cellular and systems-level organization and thus can serve as a framework for unifying work at many levels. Considering olfaction as an active sense also serves to highlight areas ripe for future investigation.
For example, in the periphery, it seems important to gain a greater understanding of airflow patterns in the nasal cavity during the range of sniffing strategies expressed during behavior, analogous to the detailed descriptions of whisker movements described for the rodent somatosensory system (Ritt et al., 2008). Centrally, it is important to better understand how sniff-driven inputs shape the transformation of odor representations in the OB and its cortical targets - such questions have been addressed in other systems through replay of naturalistic stimuli (Goard and Dan, 2009) or recording from central neurons while carefully monitoring sampling behavior in the awake animal (Han et al., 2009; Nelson et al., 1991); to date only a handful of studies have used such approaches in the olfactory system (Cury and Uchida, 2010; Shusterman et al., 2011; Verhagen et al., 2007; Wesson et al., 2008a). Finally, a key to understanding how top-down pathways actively shape odor processing is understanding how and when these pathways are activated during odor sensing; this question has also been difficult to address in other modalities. While challenging, these and related questions outline a path towards achieving a more complete - and realistic - understanding of sensory system function during behavior.
I would like to thank the past and present members of the Wachowiak lab - in particular D. Wesson, J. Verhagen and R. Carey - for contributing to the views expressed here and for performing critical experiments described from our laboratory. I would also like to thank T. Bozza, D. Rinberg, M. Shipley, D. Katz, A. Yamaguchi, and K. Zhao for valuable discussions. The laboratory has been supported by the National Institutes of Health (NIDCD), Boston University and the University of Utah USTAR initiative.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.