Sniffing controls an adaptive filter of sensory inputto the olfactory bulbJustus V Verhagen1,3, Daniel W Wesson1, Theoden I Netoff3, John A White2& Matt Wachowiak1Most sensory stimuli are actively sampled, yet the role of sampling behavior in shaping sensory codes is poorly understood.Mammals sample odors by sniffing, a complex behavior that controls odorant access to receptor neurons. Whether sniffing shapesthe neural code for odors remains unclear. We addressed this question by imaging receptor input to the olfactory bulb of awakerats performing odor discriminations that elicited different sniffing behaviors. High-frequency sniffing of an odorant attenuatedinputs encoding that odorant, whereas lower sniff frequencies caused little attenuation. Odorants encountered later in a sniff boutwere encoded as the combination of that odorant and the background odorant during low-frequency sniffing, but were encoded asthe difference between the two odorants during high-frequency sniffing. Thus, sniffing controls an adaptive filter for detectingchanges in the odor landscape. These data suggest an unexpected functional role for sniffing and show that sensory codescan be transformed by sampling behavior alone.Neural representations of external stimuli first occur as spatiotemporalpatterns of activity across receptor neurons. Transformation of thisprimary receptor code is typically thought to arise from synapticprocessing and anatomical reorganization in the CNS, with peripheralevents having a minor role in shaping receptor coding1,2. For sensorymodalities in which stimulus acquisition requires active sampling bythe animal, sampling behavior alone has the potential to alter receptorneuron responses and the neural representation of stimulus identity.However, the extent to which sampling shapes peripheral receptorcodes is not well understood.In olfaction, odorant access to olfactory receptor neurons (ORNs) iscontrolled by the rhythmic inspiration and expiration of air throughthe nose, a process that is dependent on behavioral state3. For example,rodents respire at frequencies of 1–3 Hz when at rest in a familiarenvironment4–6but increase respiration to 4–12 Hz under a variety ofcircumstances4,6–9. The transition from resting respiration to activeodor sampling (typically termed ‘sniffing’) can occur in one respirationcycle and can cease just as rapidly5,10. Other parameters of therespiratory cycle, such as amplitude and duration, are also heavilyand rapidly modulated7. In odor-guided tasks performed in thelaboratory, rats consistently show high-frequency sniffing at the timeof odor sampling7,8,11, and animals in a less constrained setting increasesniff frequency when tracking an odor source, scanning a substrate orinvestigating any novel stimulus4,6,9. Thus, high-frequency sniffing ispresumably important in olfactory function.Proposed roles for sniffing include increasing ORN responsivenessby increasing odorant flux into the nose7,12, increasing the rate of odor‘snapshots’ conveyed to the brain to facilitate rapid decision-making8and modulating ORN activity patterns to optimize the detection ofparticular odorants7,13. Sniffing also plays a critical role in olfactoryinformation processing by imposing a rhythmic temporal structure onthe input to the olfactory bulb14–16. Rhythmic inputs driven by sniffingare crucial for many models of olfactory bulb network function17–19,and the timing of postsynaptic activity relative to the sniff cycle ishypothesized to encode odor information14,20,21.Despite the fundamental role of sniffing in olfaction, experimentaldata on the relationship between sniffing, odor representations andolfactory processing remain largely absent. Even a basic characteriza-tion of the response properties of ORNs during natural sniffing has notbeen reported in mammals. Here, we ask how sniffing shapes themammalian receptor code for odors by imaging receptor input to theolfactory bulb in awake, head-fixed rats performing a simple odordiscrimination. We imaged olfactory bulb input using calcium-sensitive dyes loaded into ORNs, a technique that yields sufficientspatial resolution to map inputs to individual glomeruli and yieldsa temporal resolution of tens of milliseconds14,22.Tospecificallyaddress the role of sniffing in odor coding, we used a behavioralprotocol that allowed us to compare neural representations of thesame odorant sampled during low-frequency passive respiration andhigh-frequency sniffing.RESULTSOur goal was to image odorant representations during differentsampling behaviors. We used a head-fixed behavioral protocol thatallowed us to image receptor input to the olfactory bulb whilesimultaneously measuring respiration through a chronic intranasalcannula (Supplementary Figs. 1 and 2 and Supplementary Resultsonline). To ensure that the rats attended to the stimulus, we trainedReceived 31 January; accepted 14 March; published online 22 April 2007; doi:10.1038/nn18921Departments of Biology and2Biomedical Engineering, Boston University, 24 Cummington Street, Boston, Massachusetts 02215, USA.3Current addresses: The JohnB. Pierce Laboratory, 290 Congress Avenue, New Haven, Connecticut 06519, USA (J.V.V.); Department of Biomedical Engineering, University of Minnesota, 312 ChurchSt. SE NHH 7-105, Minneapolis, Minnesota 55455, USA (T.I.N.). Correspondence should be addressed to M.W. ([email protected]).NAT URE NEUROSCIENCE VOLUME 10[NUMBER 5[MAY 2007 631ARTICLES© 2007 Nature Publishing Group http://www.nature.com/natureneurosciencethem to carry out a lick/no-lick two-odor discrimination task (Fig. 1a)for up to 90 min per daily session. Head-fixed rats showed respirationrates that were typical of unrestrained rats in a familiar environment5,with a mean frequency of 2.37 ± 1.79 Hz (n ¼ 51,619 cycles, four rats,four sessions per rat) and a variable instantaneous frequency of 1–10Hz (Fig. 1b). Because of the continuous distribution of respirationrates observed throughout a session (Fig. 1b), we use ‘sniffing’ to referto all respiratory activity regardless of frequency10. When discriminat-ing familiar odorants, rats showed either no change or slight (B1Hz)increases in respiration rate around the time of odor discrimination(Fig. 1c, Supplementary Fig. 3 and Supplementary Results online).Despite a lack of high-frequency (44 Hz) sniffing during the dis-crimination of familiar odorants, task performance in head-fixed ratswas highly accurate and qualitatively