A Single Population of Olfactory Sensory Neurons Mediates an Innate Avoidance Behavior in Drosophila PDF

letters to nature.............................................................. by associative learning require the mushroom body4, whereas...

letters to nature.............................................................. by associative learning require the mushroom body4, whereas unconditioned avoidance responses to chemical repellents do A single population of olfactory not5,6. We asked whether the mushroom body is required for the sensory neurons mediates an innate avoidance response to dSO, by ablating this structure through hydroxyurea (HU) treatment at a critical stage of development7. avoidance behaviour in Drosophila Alternatively, we have used a mushroom-body-specific Gal4 enhancer trap line to drive the expression of UAS-Shi ts, a domi- Greg S. B. Suh1,2, Allan M. Wong1,3, Anne C. Hergarden1,2, Jing W. Wang1,3, nant-negative mutant of dynamin that inhibits neurotransmitter Anne F. Simon2*, Seymour Benzer2, Richard Axel1,3 & David J. Anderson1,2 release at a non-permissive temperature 1. Neither treatment impaired the avoidance response to dSO (Fig. 1c), although a deficit 1 Howard Hughes Medical Institute, 2Division of Biology 216-76 and 156-29, in olfactory learning confirmed that the lesion was successful California Institute of Technology, Pasadena, California 91125, USA (Fig. 1c, lower panel). These data suggest that the avoidance 3 Columbia University College of Physicians and Surgeons 701 West 168th Street, response to dSO does not require brain structures necessary for New York 10032, USA learned olfactory avoidance. * Present address: Brain Research Institute University of California, Los Angeles, California, USA Analysis of the chemical components of dSO by gas chromato-............................................................................................................................................................................. graphy and mass spectrometry (GC/MS) revealed a small peak of All animals exhibit innate behaviours in response to specific 44 daltons that corresponds to CO2, which was present in samples of sensory stimuli that are likely to result from the activation of air from conditioned tubes (Fig. 2a, left panel, arrow). Further developmentally programmed neural circuits. Here we observe analysis using a CO2 respirometer indicated that flies emit about that Drosophila exhibit robust avoidance to odours released by three- to fourfold more CO2 during shaking than do undisturbed stressed flies. Gas chromatography and mass spectrometry flies (Fig. 2b). By comparison to signals obtained by injecting identifies one component of this ‘Drosophila stress odorant known amounts of pure CO2 into the respirometer, the concen- (dSO)’ as CO2. CO2 elicits avoidance behaviour, at levels as low tration of CO2 in conditioned tubes was estimated at ,0.2% (data as 0.1%. We used two-photon imaging with the Ca21-sensitive not shown). We next asked whether CO2 alone evoked avoidance fluorescent protein G-CaMP to map the primary sensory neurons behaviour in a T-maze assay. Flies avoid CO2 in a dose-dependent governing avoidance to CO2. CO2 activates only a single glome- manner, at concentrations far below anaesthetic levels (30%) rulus in the antennal lobe, the V glomerulus; moreover, this (Fig. 2c; see also Supplementary Fig. S1). A concentration as low glomerulus is not activated by any of 26 other odorants tested. as 0.1% above the ambient CO2 level (0.0376%) evoked a statisti- Inhibition of synaptic transmission in sensory neurons that cally significant avoidance response (performance index, innervate the V glomerulus, using a temperature-sensitive PI ¼ 29.6 ^ 10.9, P , 0.001 by ANOVA, n ¼ 3, see Supplementary Shibire gene (Shi ts)1, blocks the avoidance response to CO2. Fig. S1). A concentration comparable to that estimated in dSO Inhibition of synaptic release in the vast majority of other (,0.2%), although it did elicit significant avoidance, evoked a olfactory receptor neurons has no effect on this behaviour. weaker response than did dSO itself (Fig. 2c, ‘ þ CO2 0.02 ml’ These data demonstrate that the activation of a single population versus ‘CS shaken’, where ‘CS’ indicates the wild-type Canton S of sensory neurons innervating one glomerulus is responsible for Drosophila strain), suggesting that stressed flies release additional an innate avoidance behaviour in Drosophila. repellent compound(s) together with CO2. We observed that Drosophila tend to avoid chambers in which We analysed the pattern of glomerular activity in the antennal other flies have previously been subjected to stress by mechanical lobe elicited by CO2, using a calcium-sensitive fluorescent indicator shaking or electric shock. To investigate the basis of this behaviour, protein (GCaMP) and two-photon microscopy8. Electrophysiologi- about 70 ‘emitter’ flies were subjected to mechanical stress by cal recordings have previously identified CO2-responsive ORNs in vortexing them in a test tube (see Methods). We removed the the ab1 basiconic sensilla of the antenna9–11, but the receptor stressed flies, and allowed naive ‘responder’ flies to choose between expressed in these neurons, and their projections, have not been this ‘conditioned tube’ and a fresh tube, in a T-maze apparatus2. The established. Experiments in Lepidoptera12,13 and Diptera14,15 have majority (80–95%) of responder flies avoided the conditioned tube traced the glomerular targets innervated by axons from sensilla in a one-minute trial (Fig. 1a). A similar avoidance response was containing CO2-responsive ORNs, but these sensilla contain other observed when emitter flies were stressed using electric shock ORNs as well. Moreover, the studies in Dipteran species have (Fig. 1a). To determine whether the mere presence of flies in a differed with respect to the number and identity of the glomeruli tube causes emission of the avoidance-promoting substance, flies innervated14,15. Ca2þ imaging permits an analysis of the glomerular were gently introduced into a tube using positive phototaxis, and activation pattern of CO2-responsive sensory neurons in the anten- removed after several hours of occupancy. Despite the evident nal lobe. We first examined flies in which the GCaMP indicator presence in the tube of fly waste, responder flies showed no (UAS-GCaMP) is driven in all neurons by the Elav-GAL4 activator. avoidance response to the tube (Fig. 1a, ‘no stress’). This suggests At concentrations of CO2 up to 10%, only the most ventral pair of that avoidance is elicited by a substance emitted in response to glomeruli, the V glomeruli2, were activated (Fig. 3a). Activation was mechanical or electrical stress. The emission of the substance is not detected by as little as 0.05% CO2 (Supplementary Fig. S1). These observed when anaesthetized flies are vortexed, indicating that such glomeruli were not activated by any of 26 other odorants tested emission requires neural activity. (data not shown). Surgical removal of the third antennal segment, which houses the We have previously shown16 that axonal projections to Voriginate olfactory receptor neurons (ORNs), eliminated the avoidance from antennal sensory neurons expressing the candidate gustatory response (Fig. 1b). By contrast, removal of the aristae or maxillary receptor GR21A (Fig. 3d, left). GR21Aþ neurons are located in the palps had no effect. These data suggest that the olfactory system dorso-medial portion of the antenna (Fig. 3d, right), the region mediates the avoidance response. We therefore operationally refer where CO2-responsive ab1 neurons are positioned in basiconic to the substance evoking the avoidance response as Drosophila stress sensilla10,11. Calcium imaging was thus performed with flies in odorant (dSO). which the UAS-GCaMP reporter was driven by a GR21A promo- Olfactory sensory neurons in the antennae project axons to ter-Gal4 activator (Fig. 3b). CO2 (Fig. 3b), as well as air from a tube glomeruli in the antennal lobe. Projection neurons then connect conditioned by traumatized flies (Fig. 3c, left), activated GR21A the antennal lobe to the mushroom body and lateral proto- sensory termini in the V glomeruli. Air from a tube that had cerebrum3. Conditioned olfactory avoidance responses produced contained undisturbed flies produced significantly lower levels of 854 ©2004 Nature Publishing Group NATURE | VOL 431 | 14 OCTOBER 2004 | www.nature.com/nature  letters to nature activation (Fig. 3c, right). These results indicate that both CO2 and dSO activate neurons that express the GR21A receptor and project to the V glomerulus. We next asked whether the GR21Aþ sensory neurons are necess- ary for the avoidance responses to CO2 and dSO. For this, we employed Shibire ts to reversibly inactivate these neurons at increased temperature1. Flies bearing either of two independent GR21A-Gal4 insertions and UAS-Shi ts no longer exhibited the avoidance response to ,1% CO2 at a non-permissive temperature (that is, a temperature at which neurotransmitter release cannot occur), but revealed a normal response at a permissive temperature (Fig. 3e, red versus blue bars labelled ‘2’, respectively). Control flies expressing either of the GR21A-Gal4 drivers, but not UAS-Shi ts, showed normal CO2 avoidance at the non-permissive temperature (Fig. 3e). Furthermore, flies expressing UAS-Shi ts and either of two Gal4 drivers broadly expressed in other ORNs, but not in GR21Aþ neurons (OR83b-Gal4, expressed by ,80% of ORNs, or Or47b- Gal4; L. Vosshall, personal communication), exhibited normal CO2 avoidance at the non-permissive temperature (Fig. 3e). Similarly, flies expressing UAS-Shi ts and another Gal4 driver, GH146-Gal4, which is expressed in about two-thirds of antennal lobe projection neurons5,17 (but not in those innervating V), also showed robust CO2 avoidance at the non-permissive temperature (Fig. 3e). These data indicate that GR21Aþ sensory neurons that project to the V glomerulus are probably the sole population of ORNs responsive to Figure 1 Drosophila exhibits innate avoidance of odorants released by stressed flies. a, Avoidance of tubes containing air from stressed flies in a T-maze, quantified as the PI (see Methods) of avoidance. The ‘No stress’ tube was conditioned by flies gently introduced and removed by phototaxis. The ‘Control’ is an empty tube. Bars indicate the mean ^ s.e.m. of seven independent experiments. ***, P , 0.001 in this and all subsequent figures (ANOVA). b, Effect of ablating different sensory organs on dSO avoidance. c, dSO-avoidance is MB-independent. Three MB-specific Gal4 lines, 103Y Figure 2 CO2 is a component of dSO. a, Mass spectrometry of air sample derived from (1; http://www.fly-trap.org), c747 (ref. 29) (2), and OK107 (ref. 29) (3), each crossed with ,250 shaken flies. Arrow (44 daltons) indicates CO2. b, Respirometer analysis of CO2 UAS-Shi ts and tested at 28 or 32 8C. Flies were also treated with HU (5, 6) or without HU emission from undisturbed and shaken flies. Double arrowheads indicate the CO2 level of (4) to chemically ablate the MB. The HU ablation prevents classical olfactory avoidance an unoccupied tube. c, Flies avoid CO2 in a dosage-dependent manner. The indicated conditioning (‘Learning assay’). Fluorescent micrographs indicate successful volumes of pure CO2 were infused into a 15-ml tube immediately before the choice tests. HU-mediated MB ablation (arrowheads), visualized in 253Y;UAS-mCD8GFP brains. Scale ‘CS shaken’ indicates avoidance PI obtained with tube conditioned by ,70 Canton S flies. bar, 100 mm. Error bars in b and c indicate the s.e.m. ‘Control’ indicates empty tube. Data represent mean ^ s.e.m. (n ¼ 6). NATURE | VOL 431 | 14 OCTOBER 2004 | www.nature.com/nature ©2004 Nature Publishing Group 855 letters to nature CO2, and are required for the avoidance response. Flies expressing defects in dSO-avoidance, using a collection of Gal4 enhancer trap UAS-Shi ts in GR21Aþ neurons still avoided dSO (Fig. 3e, red bars lines (K. Kaiser). A pilot screen of ,250 lines yielded 12 exhibiting labelled ‘1’). Although a reduced response was observed using one reduced dSO avoidance at non-permissive temperatures (G.S.B.S., of the two driver lines (Fig. 3e, GR21aG4(2);UAS-Shi ts), this unpublished work). Several of these dSO-unresponsive lines also reduction did not reach statistical significance. These data support exhibited a strong and specific reduction in CO2 avoidance in the notion that dSO contains other repellent(s), in addition to CO2. subsequent tests, including one designated c761 (Fig. 4a, b). To characterize further the neural substrates mediating dSO- Analysis of the c761 expression pattern revealed that it includes a responsiveness, we conducted a screen for UAS-Shi ts-dependent subset of ORNs in the third antennal segment (Fig. 4c, right), but Figure 3 CO2 avoidance is mediated by ORNs that project to the V glomerulus. a, Calcium 16). Right, GR21Aþ sensory neuron cell bodies in the dorso-medial region of the antenna. response to CO2 imaged using a pan-neuronally expressed GCaMP reporter. Left, Scale bar, 15 mm. e, Inhibition of synaptic transmission in GR21Aþ neurons using Shi ts prestimulation image; right, ,1% CO2. Colour scale indicates activation level (red is the blocks CO2 avoidance. Red and blue bars indicate non-permissive and permissive highest). Arrowheads in a–c indicate V glomerulus. Scale bar, 10 mm. b, Activation by temperatures, respectively. Bars labelled ‘1’ and ‘2’ are responses to dSO and CO2, CO2 of presynaptic terminals of GR21Aþ neurons innervating V glomerulus. c, Activation respectively, of flies of the indicated genotypes. ***, P , 0.001, by ANOVA. Error bars of GR21Aþ neurons by dSO (left) versus control (right). d, GR21Aþ ORNs project to V indicate the s.e.m. (arrow). Left, double-labelling of antennal lobe with anti-GFP and mAb nc82 (see also ref. 856 ©2004 Nature Publishing Group NATURE | VOL 431 | 14 OCTOBER 2004 | www.nature.com/nature  letters to nature not projection neurons (not shown). That this line is deficient in conspecifics in the wild, and the conditions under which they might avoidance of CO2, as well as of dSO, suggested that these ORNs do so, are not yet clear. We have used experimental stimuli such as might include those projecting to V, and others projecting to mechanical agitation and electrical current to elicit release of dSO additional glomeruli. The projections of c761þ neurons to the from Drosophila. Although such conditions are artificial, they afford antennal lobe were consistent with this expectation; labelling was us the ability to maintain tight control over the stimulus and the observed both in the V glomerulus, and in several other glomeruli organism’s response, as well as to apply molecular genetic tools to (Fig. 4c, left; Fig. 4d, left). Calcium imaging of c761;UAS-GCaMP monitor and perturb neural activity. Although these stimuli flies revealed activation of V by CO2 (Fig. 4d) as well as by dSO (not increase physical and metabolic activity, it is possible that dSO shown), confirming expression of this enhancer trap in GR21Aþ could also be emitted in response to threats. neurons. We have identified CO2 as one active component of dSO. CO2 is Together, these results indicate that c761 is expressed in, among known to be an important chemical messenger for many insect others, CO2-responsive sensory neurons that project to V (Fig. 4c). species19. In mosquitoes, for example, CO2 is an attractant that directs Because c761 was isolated in a screen for dSO-unresponsive lines, the insect towards warm-blooded animals20. At all the concentrations these data provide additional evidence that CO2 is a behaviourally of CO2 that we tested, we detected only avoidance responses. The relevant component of dSO. Moreover, the observation that c761 different behavioural responses to CO2 exhibited by mosquitoes and expresses in additional populations of ORNs besides GR21Aþ Drosophila are likely to reflect hard-wired, species-specific differences neurons suggests that these ORNs may respond to other active in neural circuitry; nevertheless, we cannot exclude that these components of dSO. behavioural differences may be context-dependent. The behavioural We have shown that Drosophila, when stressed, emits an odorant and physiologic responses to CO2 that we have measured in the mixture that elicits avoidance in other flies, and have identified CO2 laboratory are seen at several-fold increases above ambient as one active component of this mixture. Calcium imaging data (,0.0376%) that are well within the range measured for other suggest that a single population of primary olfactory receptor insects19. The role of CO2 in the ethology and ecology of Drosophila neurons, which projects to the V glomerulus16, is activated by remains to be explored, but the highly specific olfactory circuitry CO2. Specific inhibition of neurotransmission in these GR21Aþ revealed by our experiments suggests that it may be important. sensory neurons, but not in other sensory neurons, abrogates CO2 Current data suggest that in Drosophila, most odorant com- avoidance behaviour. Together, these data identify a single popu- pounds excite multiple populations of olfactory sensory neurons, lation of olfactory sensory neurons that mediates robust avoidance each expressing a single olfactory receptor gene. A given odorant to a naturally occurring odorant, and provide initial insight into the will therefore activate multiple glomeruli in the antennal lobe8,21. By neural circuitry that underlies this innate behaviour. contrast, our data suggest that a single population of ORNs, and Many insect species, when stressed or threatened, emit semio- therefore a single glomerulus (V), are involved in sensing and chemicals that evoke aggressive or avoidance behaviour in con- avoiding CO2. Previous electrophysiological studies identified specifics18. Whether Drosophila actually uses dSO to signal stress to ORNs uniquely responsive to CO2, located exclusively in ab1 Figure 4 A dSO-unresponsive enhancer trap line is also defective in its CO2 response. repellents (yellow and blue bars) and an attractant (orange). c, c761-Gal4 is expressed in a, Response of c761;UAS-Shi ts flies to dSO (labelled ‘1’) or CO2 (labelled ‘2’) at the antennal sensory neurons (right panel, nuclear-GFP reporter) that project to V (arrows), non-permissive (red bars) and permissive (blue bars) temperatures. ***, P , 0.001). and other glomeruli. d, Calcium imaging of CO2 responses in c761;UAS-GCaMP flies b, Wild-type (1) and c761;UAS-Shi ts (2) flies exhibit equivalent responses to other reveals activation in V (arrowhead). NATURE | VOL 431 | 14 OCTOBER 2004 | www.nature.com/nature ©2004 Nature Publishing Group 857 letters to nature basiconic sensilla10,11. These and the present data suggest that CO2 Gas chromatography and mass spectrometry activates a single population of ORNs, and that these ORNs respond Air samples from tubes containing dSO and from fresh tubes were analysed using a gas chromatograph (Agilent 6890) interfaced with a quadruple mass spectrometer (Agilent only (or primarily) to CO2. We cannot exclude, however, that other 5970B). Five microlitres of air from a tube in which 250 flies had been shaken, or from a ORNs, (or antennal lobe projection neurons innervating glomeruli fresh tube, were injected with a Hamilton syringe into the GC column at 50 8C. The other than V), are also activated by CO2 at levels below the detection column was then heated to 270 8C at a rate of 10 8C min21 with normal inlet temperature limit of our imaging technology. Nevertheless, previous studies of 250 8C (splitless mode). The GC column was equipped with a column from J&W, DB5- MS, 30 cm £ 0.25 mm (i.d.) £ 0.25 mm film thickness. Each molecule eluted from the GC using this method have shown that multiple glomeruli are activated column was detected and its molecular mass and abundance measured by the MS. The by most odorants tested8, so it is striking that just a single operating conditions for the MS were:10 to 500 m/z; 1.64 scans s21; ionization energy glomerulus is activated by CO2. Furthermore, the behavioural 70 eV. response to CO2 is extinguished by genetic silencing of GR21Aþ sensory neurons. The fact that the avoidance response to CO2 is Respirometer measurements unaffected by genetic silencing of Or83b or GH146 neurons further Emission of carbon dioxide was measured using a Sable System TR-2 carbon dioxide gas respirometry system (Model LI-6251). Groups of 20 flies were placed in a 2.2-ml glass suggests that large populations of ORNs and projection neurons not chamber, which was flushed with a constant flow of CO2-free air through a CO2 detector. directly innervating V are unnecessary for CO2 avoidance. Taken The amount of CO2 produced by each group of flies was calculated by using DATACAN together, these data suggest that a dedicated circuit, which involves software (Sable Systems International). a single population of ORNs, mediates detection of CO2 in Visualization of murine CD8GFP Drosophila. The simplicity of this early-stage olfactory processing Adult fly brains were dissected, fixed in 2% paraformaldehyde, and mounted in offers a great advantage in further tracing the circuits that translate Vectashield (Vecta Labs). Native green fluorescence protein (GFP) fluorescence of whole- CO2 detection into an avoidance response. mount brains was visualized by confocal microscopy (Leica). Olfactory axonal projections In general, recognition of many odorants in insects probably of flies bearing GR21A-Gal4 and UAS-mCD8GFP were visualized by fluorescent requires the decoding of combinatorial patterns of glomerular immunohistochemistry, as described in ref. 16. activation8,22–24 , perhaps combined with complex temporal Hydroxyurea treatment dynamics in the antennal lobe25. Nevertheless, our data suggest The HU protocol7 was used to block development of the mushroom body (MB) structure. that there is also a set of olfactory stimuli, including CO2, that To check the ablation of the MB, we used the 253Y enhancer trap line carrying UAS- release innate behaviours by activating a single class of primary mCD8GFP, which expresses in the MB, as well as in other regions. The survival of the other sensory neurons and their associated glomerulus. In Drosophila, structures after treatment serves as an internal control for the specificity of HU ablation. these stimuli may also include mating pheromones which, in other Learning assay insect species, are known to activate specialized glomeruli26,27. In The ‘Long Program’ training protocol described in ref. 30 was used to train flies. Flies were Caenorhabditis elegans, the activation of a single chemosensory exposed to 60 s of odour A associated with a 90-V 1.5-s shock delivered every 5 s for 60 s neuron can elicit a repulsive behaviour28. Uni-glomerular circuits (CS þ ) followed by 60 s of odor B with no shock (CS 2 ). The trained flies were then dedicated to the detection of certain odorants may have evolved to given a choice in the T-maze between odours A and B. Both training and testing were done at room temperature (23–25 8C) and humidity (20–50%). provide innate behavioural responses to these stimuli, which are essential to survival or reproduction of the species. A Received 13 July; accepted 1 September 2004; doi:10.1038/nature02980. Published online 15 September 2004. 1. Kitamoto, T. Conditional modification of behavior in Drosophila by targeted expression of a Methods temperature-sensitive shibire allele in defined neurons. J. Neurobiol. 47, 81–92 (2001). Flies 2. Dudai, Y., Jan, Y. N., Byers, D., Quinn, W. G. & Benzer, S. dunce, a mutant of Drosophila deficient in Unless otherwise indicated, Canton-S flies (Caltech stock) were used for all experiments. learning. Proc. Natl Acad. Sci. USA 73, 1684–1688 (1976). Flies carrying the UAS-Shi ts transgenes on the X and third chromosomes were reared at 3. Stocker, R. F. The organization of the chemosensory system in Drosophila melanogaster: a review. Cell 20 8C. For reversible neuronal silencing, flies were incubated in a 28 8C room for ,90 min Tissue Res. 275, 3–26 (1994). and then in a 328 or 34 8C water bath for 5 min, before performing the behavioural 4. Heisenberg, M. Mushroom body memoir: from maps to models. Nature Rev. Neurosci. 4, 266–275 experiments at 28 8C. The permissive temperature used in experiments with Shi ts was (2003). 21 8C. To generate antennaless, palpless or aristaless flies, extirpations were performed 5. Heimbeck, G., Bugnon, V., Gendre, N., Keller, A. & Stocker, R. F. A central neural circuit for using fine tweezers under CO2 anaesthesia, and the animals were allowed to recover for experience-independent olfactory and courtship behavior in Drosophila melanogaster. Proc. Natl ,24 h before testing. Acad. Sci. USA 98, 15336–15341 (2001). 6. Wang, Y. et al. Blockade of neurotransmission in Drosophila mushroom bodies impairs odor attraction, but not repulsion. Curr. Biol. 13, 1900–1904 (2003). Behavioural tests 7. de Belle, J. S. & Heisenberg, M. Associative odor learning in Drosophila abolished by chemical ablation For behavioural testing, ,40 ‘responder’ and 70 ‘emitter’ adult flies (3- to 6-days post- of mushroom bodies. Science 263, 692–695 (1994). eclosion, mixed gender) were anaesthetized using CO2 and sorted into separate vials 8. Wang, J. W., Wong, A. M., Flores, J., Vosshall, L. B. & Axel, R. Two-photon calcium imaging reveals an 24–48 h before use, or were sorted using an aspirator. A 16-inch 15-W fluorescent bulb was odor-evoked map of activity in the fly brain. Cell 112, 271–282 (2003). horizontally centred on the bench-top, behind the testing apparatus, in an otherwise dark 9. Stange, G. & Stowe, S. Carbon-dioxide sensing structures in terrestrial arthropods. Microsc. Res. Tech. room. Responder flies were transferred into the T-maze by first placing them into a 15-ml 47, 416–427 (1999). plastic tube (Fisher no. 149598), and tapping them into the elevator of the T-maze. While 10. Stensmyr, M. C., Giordano, E., Balloi, A., Angioy, A. M. & Hansson, B. S. Novel natural ligands for the responders were in the elevator, emitter flies were vortexed in a clean 15-ml tube in 3-s Drosophila olfactory receptor neurones. J. Exp. Biol. 206, 715–724 (2003). bouts, at 5-s intervals, over a 1-min period. As an alternative method of stressing flies, 11. de Bruyne, M., Foster, K. & Carlson, J. R. Odor coding in the Drosophila antenna. Neuron 30, 537–552 electric shocks were applied across a copper grid at 60 V (direct current) for 1 min. The (2001). emitter flies were discarded, the conditioned tube immediately inserted into one side of 12. Kent, K. S., Harrow, I. D., Quartararo, P. & Hildebrand, J. G. An accessory olfactory pathway in the T-maze, a fresh tube being inserted on the other side. The elevator containing Lepidoptera: the labial pit organ and its central projections in Manduca sexta and certain other sphinx responder flies was lowered, and the flies given one minute to choose between the two moths and silk moths. Cell Tissue Res. 245, 237–245 (1986). tubes, after which the elevator was partially lifted to block any further choices, and the 13. Bogner, F., Boppre, M., Ernst, K. D. & Boeckh, J. CO2 sensitive receptors on labial palps of number of flies in each tube counted. For testing responses to other odorants, 10 ml of Rhodogastria moths (Lepidoptera: Arctiidae): physiology, fine structure and central projection. 0.01% octanol, 10 ml of 0.01% methylcyclohexanol (MCH) or 10 ml of 0.3 mg ml21 freshly J. Comp. Physiol. A 158, 741–749 (1986). prepared freeze-dried instant coffee (Taster’s Choice) were dripped onto pieces of 14. Distler, P. & Boeckh, J. Central projections of the maxillary and antennal nerves in the mosquito Aedes Whatman filer paper (0.5 £ 0.25 in), placed at the end of the 15-ml plastic tube inserted aegypti. J. Exp. Biol. 200, 1873–1879 (1997). into the T-maze apparatus. The control tube contained filter paper and vehicle alone. The 15. Anton, S. Central olfactory pathways in mosquitoes and other insects. Ciba Found. Symp. 200, 184–192; discussions 192–196, 226–232 (1996). avoidance response was analysed by calculating the PI, that is, the percentage of flies 16. Scott, K. et al. A chemosensory gene family encoding candidate gustatory and olfactory receptors in avoiding minus the percentage of flies entering. PI ¼ 0 indicates an equal distribution of Drosophila. Cell 104, 661–673 (2001). flies between the two tubes. PI ¼ 100% indicates that all flies avoided the conditioned 17. Stocker, R. F., Heimbeck, G., Gendre, N. & de Belle, J. S. Neuroblast ablation in Drosophila P[GAL4] tube. Statistical significance was calculated using analysis of variance, ANOVA. lines reveals origins of olfactory interneurons. J. Neurobiol. 32, 443–456 (1997). 18. Shorey, H. H. Behavioral responses to insect pheromones. Annu. Rev. Entomol. 18, 349–380 (1973). Ca21 imaging and odour delivery 19. Stange, G. in Advances in Bioclimatology (ed. Stanhill, G.) 223–253 (Springer, Berlin, 1996). Sample preparation and calcium imaging were as described in ref. 8. CO2 was diluted to 20. Enserink, M. What mosquitoes want: secrets of host attraction. Science 298, 90–92 (2002). 1% and delivered at a flow rate of 81 ml min21. dSO and air from non-traumatized flies 21. Galizia, C. G., Sachse, S., Rappert, A. & Menzel, R. The glomerular code for odor representation is were prepared as described above, and delivered undiluted. species specific in the honeybee Apis mellifera. Nature Neurosci. 2, 473–478 (1999). 858 ©2004 Nature Publishing Group NATURE | VOL 431 | 14 OCTOBER 2004 | www.nature.com/nature  letters to nature 22. Wang, Y. et al. Genetic manipulation of the odor-evoked distributed neural activity in the Drosophila light, you see pedestrians waiting to cross the street. Effortlessly, you mushroom body. Neuron 29, 267–276 (2001). 23. Ng, M. et al. Transmission of olfactory information between three populations of neurons in the decide whether one of them is your spouse, your boss or a stranger, antennal lobe of the fly. Neuron 36, 463–474 (2002). and connect the percept with the appropriate action, so that you will 24. Wilson, R. I., Turner, G. C. & Laurent, G. Transformation of olfactory representations in the either be waving frantically, greeting respectfully or taking another Drosophila antennal lobe. Science 303, 366–370 (2004). sip of coffee. During a rainstorm, however, the sensory input is 25. Laurent, G. Olfactory network dynamics and the coding of multidimensional signals. Nature Rev. Neurosci. 3, 884–895 (2002). noisier, and thus you have to look longer to gather more sensory 26. Christensen, T. A., Harrow, I. D., Cuzzocrea, C., Randolph, P. W. & Hildebrand, J. G. Distinct data to make a decision about the person at the light and the projections of two populations of olfactory receptor axons in the antennal lobe of the sphinx moth appropriate behavioural response. Manduca sexta. Chem. Senses 20, 313–323 (1995). This type of decision-making has been studied in single-unit 27. Hansson, B. S., Carlsson, M. A. & Kalinova, B. Olfactory activation patterns in the antennal lobe of the sphinx moth, Manduca sexta. J. Comp. Physiol. A 189, 301–308 (2003). recording studies in monkeys performing sensory discriminations5–8. 28. Troemel, E. R., Kimmel, B. E. & Bargmann, C. I. Reprogramming chemotaxis responses: sensory Shadlen et al. proposed that perceptual decisions are made by neurons define olfactory preferences in C. elegans. Cell 91, 161–169 (1997). integrating the difference in spike rates from pools of neurons 29. Connolly, J. B. et al. Associative learning disrupted by impaired Gs signaling in Drosophila mushroom selectively tuned to different perceptual choices9. For example, in a bodies. Science 274, 2104–2107 (1996). 30. Beck, C. D., Schroeder, B. & Davis, R. L. Learning performance of normal and mutant Drosophila after direction-of-motion task, in which the monkey must decide repeated conditioning trials with discrete stimuli. J Neurosci 20, 2944–53 (2000). whether a noisy field of dots is moving upward or downward, a decision can be formed by computing the difference in responses Supplementary Information accompanies the paper on www.nature.com/nature. between lower-level neurons that are sensitive to upward motion and those sensitive to downward motion1–4. Similarly, in a somatosensory Acknowledgements We thank J.-S. Chang for technical assistance, L. Vosshall for providing task, in which the monkey must decide which of two vibratory stimuli Or83b-Gal4 and Or47b-Gal4 flies and for other unpublished information, D. Armstrong for Gal4 enhancer trap lines 103Y, 253Y, c747 and c761, T. Kitamoto for UAS-Shi ts flies, U. Heberlein for has a higher frequency, a decision can be formed by subtracting the the HU protocol and R. I. Wilson for discussion of unpublished data and comments on the activities of two populations of sensory neurons that prefer low and manuscript. G.S.B.S. is a recipient of a National Research Service Award. A.C.H. is supported by a high frequencies, respectively8,10. These findings suggest that a com- Howard Hughes Predoctoral fellowship. This work was supported by the HHMI (R.A. and D.J.A.) parison of the outputs of different pools of selectively tuned lower- and by the NSF (S.B.). R.A. and D.J.A. are Investigators of the Howard Hughes Medical Institute. level sensory neurons could be a general mechanism by which higher- Author contributions S.B., R.A. and D.J.A. made equally minimal contributions to this work. level cortical regions compute perceptual decisions1,2,11. However, it is still unknown whether such a mechanism is at work for more Competing interests statement The authors declare that they have no competing financial complex cognitive operations in the human brain and, if so, where interests. in the brain this computation might be performed. We used functional magnetic resonance imaging (fMRI) while Correspondence and requests for materials should be addressed to D.J.A. ([email protected]). subjects decided whether an image presented on a screen was a face or a house (Fig. 1). Previous neuroimaging studies have identified regions in the human ventral temporal cortex that are activated more by faces than by houses, and vice versa12–16. Increases in the.............................................................. blood-oxygen-level-dependent (BOLD) signal have been shown to be proportional to changes in neuronal activity in a given region17,18. A general mechanism for perceptual Therefore larger BOLD responses to faces than to houses and vice versa in specific voxels in the ventral temporal cortex reflect the decision-making in the human brain change in activity in a population of neurons that are more responsive to faces than to houses, and vice versa. Our task thus H. R. Heekeren1, S. Marrett2, P. A. Bandettini1,2 & L. G. Ungerleider1 enabled us to identify two brain regions, one more sensitive to faces and another to houses, and to test whether there are higher-level 1 Laboratory of Brain and Cognition, NIMH, 2Functional MRI Facility, NIMH, cortical regions whose output is proportional to the difference in NIH, Bethesda, Maryland 20892-1148, USA activation in the face- and house-selective regions, respectively.............................................................................................................................................................................. We based our hypotheses on results from single-unit recording Findings from single-cell recording studies suggest that a com- studies in monkeys, which have shown that neuronal activity in parison of the outputs of different pools of selectively tuned areas involved in decision-making gradually increases and then lower-level sensory neurons may be a general mechanism by remains elevated until a response is given, with the rate of increase which higher-level brain regions compute perceptual decisions. being slower during more difficult trials1,2. These studies have also For example, when monkeys must decide whether a noisy field of shown that higher-level cortical regions, such as the dorsolateral dots is moving upward or downward, a decision can be formed prefrontal cortex (DLPFC), might form a decision by comparing the by computing the difference in responses between lower-level output of pools of selectively tuned lower-level sensory neurons4,9. neurons sensitive to upward motion and those sensitive to Therefore, we hypothesized that higher-level cortical regions downward motion1–4. Here we use functional magnetic resonance computing a decision would have to fulfil two conditions. First, imaging and a categorization task in which subjects decide they should show the greatest activity on trials in which the evidence whether an image presented is a face or a house to test whether for a given perceptual category is greatest, for example, a greater a similar mechanism is also at work for more complex decisions fMRI response during decisions about suprathreshold images of in the human brain and, if so, where in the brain this compu- faces and houses than during decisions about perithreshold images tation might be performed. Activity within the left dorsolateral of these stimuli. Second, their activity should be correlated with the prefrontal cortex is greater during easy decisions than during difference between the output signals of the two brain regions difficult decisions, covaries with the difference signal between containing pools of selectively tuned lower-level sensory neurons face- and house-selective regions in the ventral temporal cortex, involved; that is, those in face- and house-responsive regions. and predicts behavioural performance in the categorization task. To test the model of decision-making, we added noise to the face These findings show that even for complex object categories, the and house stimuli, which made the task arbitrarily more difficult by comparison of the outputs of different pools of selectively tuned reducing the sensory evidence available to the subject (Fig. 1b). In neurons could be a general mechanism by which the human brain the fMRI experiment, subjects viewed images that were either easy computes perceptual decisions. (suprathreshold, Fig. 1b top) or difficult (perithreshold, Fig. 1b Consider driving home from work in clear weather. Stopping at a bottom) to identify as faces or houses. NATURE | VOL 431 | 14 OCTOBER 2004 | www.nature.com/nature ©2004 Nature Publishing Group 859