ARTICLE doi:10.1038/naturell712 Non-synaptic inhibition between grouped neurons in an olfactory circuit Chih-Ying Su1, Karen Menuz1, Johannes Reisert2 & John R. Carlson1 Diverse sensory organs, including mammalian taste buds and insect chemosensory sensilla, show a marked compart -mentalization of receptor cells; however, the functional impact of this organization remains unclear. Here we show that compartmentalized Drosophila olfactory receptor neurons (ORNs) communicate with each other directly. The sustained response of one ORN is inhibited by the transient activation of a neighbouring ORN. Mechanistically, such lateral inhibition does not depend on synapses and is probably mediated by ephaptic coupling. Moreover, lateral inhibition in the periphery can modulate olfactory behaviour. Together, the results show that integration of olfactory information can occur via lateral interactions between ORNs. Inhibition of a sustained response by a transient response may provide a means of encoding salience. Finally, a C02-sensitive ORN in the malaria mosquito Anopheles can also be inhibited by excitation of an adjacent ORN, suggesting a broad occurrence of lateral inhibition in insects and possible applications in insect control. An intriguing feature of a number of sensory systems is the compart -mentalization of their primary sensory cells. These cells are housed together in specialized structures such as the taste buds of vertebrates and the chemosensory sensilla of invertebrates. The compartmentalized primary sensory cells often respond to diverse stimuli. The functional consequence of such organization is unknown. Olfactory receptor neurons (ORNs) are the primary units of odour perception1. ORNs are widely believed to function as autonomous units, each responding to odorants independent of other ORNs. In some organisms, such as insects, ORNs are compartmentalized into sensilla (Fig. la). An individual sensillum encapsulates the dendrites of neurons2 4. The neighbouring ORNs exhibit differing spike amplitudes and odorant sensitivities5. In Drosophila melanogaster, each ORN is assigned a designation indicating the type of sensillum in which it is housed and its relative spike amplitude among the ORNs of the sensillum. Thus, the ab3A neuron is located in antennal basi-conic sensilla of type 3, and the 'A' indicates that its spike amplitude is greater than that of the neighbouring 'B' neuron. In fruitflies, moths and mosquitoes, ORNs are grouped in stereotyped combinations59. The functional significance of this widespread pattern of ORN organization is unknown. In Drosophila, neighbouring ORNs do not have obvious functional relationships10, and they do not project to adjacent regions in the brain11. In certain sensilla of flies, moths and beetles, the activation of neighbouring ORNs elicits opposing beha-viours6-8-9-12-16. There are theoretical predictions based on electrical circuit modelling that the transient activation of one ORN may interfere with the signalling of a neighbouring ORN17, and there is precedent for olfactory stimuli that activate one neuron and inhibit its neighbour1516, but in the absence of molecular genetic analysis it is difficult to determine whether such stimuli act uniquely on one ORN or directly on both. Similar examples can also be found in insect taste sensilla1822, but in Drosophila some bitter compounds have been shown to act directly both on a sugar neuron and on a bitter neuron, inhibiting one and exciting the other23. Here we use the molecular genetics of Drosophila to examine the coding of pairs of odours by the ORNs of olfactory sensilla. We find that the prolonged activation of one ORN is inhibited by the transient excitation of its neighbour. This lateral inhibition is observed within diverse types of Drosophila sensilla, and the activation of a mosquito ORN laterally inhibits the response of a neighbouring ORN to C02, a key cue used by mosquitoes to find their human hosts. The communication between neurons does not require a synapse, and probably proceeds via ephaptic coupling. Finally, we find that this lateral inhibition at the periphery of the olfactory circuit can modulate olfactory behaviour. Together, our results indicate that ORNs do not signal cell-autonomously in all circumstances, but rather their responses can be regulated by the activity of their ORN neighbours in a sensillum. Activation of an ORN inhibits its neighbour To analyse the relationship between two ORNs in a sensillum, we used a paradigm that allows us to deliver two odours, one for each neuron (Fig. lb, c). One odorant, the 'background odorant', is provided continuously via an airstream and elicits the sustained firing of one ORN, the A neuron in most experiments. Superimposed on this background stimulus, a short pulse of a second odorant is delivered to activate the other ORN, usually the B neuron. This paradigm of odour presentation is distinct from the single-odorant paradigm used commonly in many studies510-24, but it simulates a coding problem that the system encounters in its natural environment, for example when a fly receiving sustained olfactory input from a local source receives a superimposed, transient stimulus from a distant source delivered by a gust of wind. When the ab3 sensillum is stimulated with a prolonged dose of methyl hexanoate, the ab3A neuron responds with a sustained train of action potentials (large action potentials in Fig. Id). When a pulse of 2-heptanone is superimposed on this background, not only does ab3B fire (small action potentials) but there is a marked reduction in the firing of ab3A (Fig. Id). This inhibitory effect could, in principle, be due to direct inhibition of OR22A, the receptor of ab3A, by 2-heptanone. However, ablation of ab3B by expression of the cell death gene reaper (rpr) completely abolished the inhibition of ab3A (Fig. Id, bottom). This result indicates that the inhibition of the A neuron depends on the excitation of the B neuron. department of Molecular, Cellular and Developmental Biology, Yale University, New Haven, Connecticut 06520, USA. 2Monell Chemical Senses Center, Philadelphia, Pennsylvania 19104, USA. 66 I NATURE I VOL 4 9 2 I 6 DECEMBER 2012 ©2012 Macmillan Publishers Limited. All rights reserved |^^^ RESEARCH Sensillum ;\ Main air stream Odour pulse B-neuron Odorant Figure 1 | Lateral inhibition of ORNs. a, An olfactory sensillum that houses two ORNs, A and B. Inset: a single-unit recording. 'A' has a larger spike amplitude than 'B'. b, The two-odour paradigm, c, The ab3 sensillum, the ORNs of which express the OR22 A and OR85B receptors, which are sensitive to the indicated odorants. d, Top: a sustained stimulus of methyl hexanoate (m-hex, 10 7 dilution, long blue bar) elicits a response from ab3A (large spikes, —37 spikes s l). A 500-ms pulse of 2-heptanone (2-hep, 10 4 dilution, orange bar) activates ab3B (small spikes). The response of ab3A is inhibited by the 2-heptanone stimulus. Right, averaged responses. Grey traces indicate responses when a pulse of diluent is delivered instead of 2-heptanone. Shaded areas represent s.e.m. Inset: blue dots indicate ab3A spikes. Bottom: genetic ablation of ab3B prevented inhibition, e, In flies expressing ChR2* in ab3B (top), a 500-ms pulse of blue light (473 nm, — 10mWmm~2) excited ab3B, which inhibited the response of ab3A to methyl hexanoate (—32 spikes s \ 10 6). The more phasic inhibition is probably due to the kinetics of ChR2-dependent activation. Bottom: flies without ChR2*. f, Top: activation of ab3A by a pulse of methyl hexanoate (10 6) inhibited the response of ab3B to 2-heptanone (—38 spikess ',5X10 7). Bottom: genetic ablation of ab3A prevented inhibition. Inset: orange dots indicate ab3B spikes. Very large spikes represent the coincidence of A and B spikes, g, ChR2* expressed in ab3A. A pulse of blue light (—25 mWmm 2) excited ab3A, inhibiting the response of ab3B to 2-heptanone (—35spikess_1, 5 X 10~7). n = 12 in d-g. To test further the possibility that activation of the ab3B neuron can inhibit the ab3A neuron, we expressed Channelrhodopsin2 (H134R-ChR2)25 in ab3B. As expected, blue light elicited an excitatory response in ab3B of these engineered flies (Fig. le). Activation of ab3B by light also inhibited the tonic firing of ab3A elicited by methyl hexanoate. Blue light had no effect on ab3A firing in control flies lacking H134R-ChR2 (Fig. le, bottom), indicating that it does not inhibit ab3A directly. The simplest interpretation of these results is that activation of ab3B inhibits the firing of ab3A. We next asked whether activation of ab3A can inhibit ab3B. We first elevated ab3B activity by delivering 2-heptanone as the background odorant and then presented a pulse of methyl hexanoate to activate ab3A. Indeed, the pulse of methyl hexanoate inhibited the activity of ab3B (Fig. If, top). Genetic ablation of ab3A demonstrated that this inhibition depended on ab3A (Fig. If, bottom). Similarly, when H134R-ChR2 was expressed in ab3A, a blue-light stimulus activated ab3A and inhibited the tonic firing of ab3B (Fig. lg). Lateral inhibition in other sensilla There are four morphological types of antennal sensilla: large basico-nic sensilla, small basiconic sensilla, coeloconic sensilla and trichoid sensilla1-26-27. ab3 is a large basiconic sensillum containing two ORNs. We analysed four other sensilla, chosen for their morphological diversity and their functional specificities. Their ORNs express receptors that have been functionally characterized, and odorants have been identified that at certain concentrations selectively activate the receptor of only one ORN in each sensillum10-24. Lateral inhibition between ORNs was observed in all sensillar types examined: a large basiconic sensillum containing four ORNs (abl); a large basiconic sensillum with two ORNs (ab2); a small basiconic sensillum (ab5); and a coeloconic sensillum (ac3). In each case, a short odorant pulse that activated one target ORN inhibited the tonic firing of a neighbouring ORN (Fig. 2a-d). When the targeted ORN was ablated or non-functional, the short odorant pulse showed no inhibition of the neighbouring ORN (Supplementary Fig. 1). We note also that the pulsed odorant alone did not directly inhibit the spontaneous firing of the A neuron (Supplementary Fig. 2). These results indicate that lateral inhibition is observed broadly in the Drosophila antenna. Lateral inhibition in a mosquito sensillum ORNs are compartmentalized in sensilla in a wide variety of insects. We examined a sensillum of the malaria vector Anopheles gambiae that responds to C02 (ref. 7), a human volatile that attracts many mosquito species28. This sensillum contains an ORN, cpA, that responds to C02, and a neighbouring ORN, cpB, that is excited by l-octen-3-ol (ref. 7). We used a prolonged C02 stimulus to elicit a sustained response from cpA. When a short pulse of l-octen-3-ol was superimposed, the cpB neuron was excited and cpA was robustly inhibited (Fig. 2e). We note that when l-octen-3-ol was delivered in the absence of C02, it did not inhibit the spontaneous firing of the C02-responsive cpA neuron directly (Supplementary Fig. 2d), consistent with previous results7. Taken together, our results show that lateral inhibition occurs in olfactory sensilla of multiple insect species, in sensilla of radically different morphology, and in sensilla containing two, three or four ORNs. Inhibition is dose-dependent When ab3A was tonically excited with a constant concentration of methyl hexanoate, increasing doses of 2-heptanone produced increasing excitation of ab3B and increasing inhibition of ab3A (Fig. 3a, b). When the scales of the firing ranges are adjusted (Fig. 3b), the dose-response functions seem to be symmetrical. 6 DECEMBER 2012 I VOL 4 9 2 I NATURE I 67 ©2012 Macmillan Publishers Limited. All rights reserved RESEARCH Mosquito sensillum (cp] mm Figure 2 | Lateral inhibition in diverse sensilla. Odorants at the tested concentrations activate only one ORN in each sensillum. a-d, Drosophila sensilla. Activation of the target ORN (orange) inhibited the response of the neighbouring ORN (blue) to the background odorant. In a, ablA and ablB spikes could not be sorted reliably and were grouped, e, In the capitate-peg sensillum of Anopheles, activation of the cpB neuron by l-octen-3-ol (10 4) inhibited the response of cpA to C02. cpB and cpC spikes were combined. n = 11-13. Odour dilutions and A neuron basal activities are in Supplementary Table 2. m-hex dilution Figure 3 | Lateral inhibition is dose-dependent, a, Responses of ab3A and ab3B to a 500-ms pulse of 2-heptanone (orange) superimposed on a background odorant, methyl hexanoate (10 7 dilution;—37 spikes s l). At these low concentrations, methyl hexanoate and 2-heptanone selectively activate ab3A and ab3B, respectively. 2-Heptanone dilutions are at the right of the panel, b, Activities of ab3A and ab3B during 2-heptanone pulses. Fit is with the Hill equation; n = 12. c, Responses to a pulse of 2-heptanone (10 4) in the presence of varying levels of methyl hexanoate, indicated at the right of the panel, d, Responses of ab3A during 500-ms exposures to paraffin oil (control) or 2-heptanone with varying concentrations of background methyl hexanoate; n = 12. In the absence of sustained stimulation of the A neuron ('no bkg'), strong activation of the B neuron elicited a small increase in the firing of A, which may represent passive depolarization of A resulting from close apposition of the neuronal membranes43'44. This effect seems to be overwhelmed by the passive hyperpolarization produced by ephaptic interactions (discussed below) when B is activated during sustained stimulation of A. Differences are significant in all conditions (P< 0.002, paired f-test); n = 12. When the background odorant, methyl hexanoate, was delivered at increasing concentrations, the rate of ab3A tonic firing increased across a range of —15 spikes s 1 to —50 spikes s 1 (Fig. 3c, d, and Supplementary Table 2). Inhibition by a strong ab3B stimulus was potent across all these concentrations; in all of these cases the rate of firing was reduced to approximately the same level. A genetic ablation experiment confirmed that these reductions depended on ab3B (Supplementary Fig. 3). We note that 2-heptanone alone did not directly inhibit ab3A spontaneous activity (Fig. 3c, d, 'no bkg'). Transmission without a synapse Next we asked whether the intrasensillar communication is mediated by synapses. First we used tetanus toxin (TNT)29 to block synaptic transmission. We expressed TNT in ORNs using the Oreo promoter and the GAL4/UAS system, which is expected to drive expression in all basiconic ORNs30 except the C02-sensitive ablC neuron31-32. Activation of ab3B inhibited the tonic excitation of ab3A in these TNT-expressing flies (Fig. 4a, top). Moreover, the degree of inhibition was comparable to that in control flies (Fig. 4a, bottom). T-maze behavioural tests confirmed that synaptic transmission was blocked in the targeted ORNs (Fig. 4b). Second, we performed single-unit recordings from isolated antennae, severed from the heads of flies. Activation of ab3B again inhibited the tonic excitation of ab3A (Fig. 4c), supporting the conclusion that lateral inhibition between neighbouring ORNs occurs in the periphery without involvement of central synapses. Third, we tested the possibility of axo-axonic synapses between ORNs with a cross-correlation analysis33. Analysis of ab3A and ab3B spontaneous spikes did not reveal coordinated spiking patterns and thus provided no evidence for axo-axonic synaptic interactions (Fig. 4d), similar to what has been found between homotypic ORNs in Drosophila34. Finally, we used Cd2+ to block synaptic neurotransmission35. We included a high concentration of Cd2+ in the recording pipette so as to allow Cd2+ to diffuse into the sensillum lymph and block any peripheral dendro-dendritic synapses in sensilla of Orco-GAL4; UAS-TNT flies. We observed little if any effect on the inhibition of ab3A firing after ab3B excitation (Fig. 4e; compare with Fig. 4a). To verify the efficacy of our drug delivery method, we applied the Oreo agonist VUAA1 (ref. 36) via the recording pipette and observed elevated ORN spike activities, as expected (Fig. 4f). Together, these results indicate that lateral inhibition does not depend on chemical synapses. Intrasensillar communication could, in principle, be mediated via gap junctions; however, the activation of one ORN would then probably lead to the activation, rather than the inhibition, of its neighbour. Moreover, we found that nitric oxide signalling inhibitors had 68 I NATURE I VOL 4 9 2 I 6 DECEMBER 2012 ©2012 Macmillan Publishers Limited. All rights reserved |^^^ RESEARCH b\ Orco::TNT m-hex 2-hep Cd2* DD 2-Heptanone CG 2-Heptanone Q Ctrl TNT Ctrl TNT ab3A versus ab3B (spontaneous) 0.0 nterspike time (s) Figure 4 | Lateral inhibition does not require synapses, a, ab3 sensilla in flies expressing TNT in ORNs via the Oreo promoter. Neurons were exposed to a 500-ms pulse of the ab3B odorant, 2-heptanone (orange, 10 4), superimposed on the background ab3A odorant, methyl hexanoate (blue, 10 7). IN, representative interneuron. Right: ab3A activity during a 500-ms exposure to paraffin oil (PO) or 2-heptanone in the presence of methyl hexanoate. Error bars indicate s.e.m.; *P < 0.05, one-way repeated measures ANOVA, multiple comparison versus control group (PO) with Dunnett's method (n = 12). b, T-maze choice between water and 25% ACV or between air and 0.67% CO,. 2-Heptanone C02 neurons do not express Or genes, (n = 9). c, Recordings were performed as in a except that the antenna was severed from the head (n = 7). d, Cross-correlation analysis of spontaneous spikes from an ab3 sensillum, showing intervals between ab3A spikes and ab3B spikes, binned in 10-ms increments. Each ab3B spike is used as a reference. Another ab3 sensillum gave similar results, e, Recordings made 15min after introduction of Cd2+ (n = 12). f, VUAA1 (1 mM) or vehicle (1% dimethylsulphoxide (DMSO)) was delivered via the recording electrode36. ab3A responses were recorded for 10 s. ***P<0.001, f-test(n= 12). no effect on lateral inhibition (not shown). In summary, conventional forms of neuronal communication are unlikely to mediate lateral inhibition in a sensillum. Lateral inhibition modulates behaviour To determine whether intra-sensillar neuronal inhibition can modulate olfactory behaviour, we examined a pair of neighbouring ORNs, the activation ofwhich leads to opposing behavioural outputs (Fig. 5). ablA mediates attraction to apple cider vinegar (ACV)13, whereas its neighbour ablC mediates aversion to low concentrations of C02 (refs 12, 32, 37, 38). We confirmed that in a T-maze assay, when given a choice between C02 and air alone, flies avoid C02, whereas when faced with a choice between ACV and water, they are attracted to ACV (Fig. 5a, black bars). We then tested whether the two behavioural pathways interact. When both arms of the T-maze contained C02, the flies showed no preference (Fig. 5a). When ACV was added to one of the C02-containing arms, the flies preferred that arm. The preference for the arm containing both C02 and ACV could have two sources: the attraction to its ACV that is mediated by ablA, and a reduction in the avoidance of its C02 that is mediated by ablC. To evaluate the contributions of these sources, we used Orco-GAL4;UAS-TNT, which blocks synaptic transmission from ablA but not ablC. Consistent with the expected specificity of this block, these flies did not respond to ACV but avoided C02 (Fig. 5b, black bars). We note that in these flies, physiological recordings confirmed that abl A neurons respond to ACV (not shown). When given a choice between two arms, one with C02 and one with C02 and ACV, these flies preferred the arm with ACV (Fig. 5b). Because synaptic transmission from abl A neurons is blocked and the flies have no attraction to ACV, the simplest interpretation of these results is that activation of ablA attenuated the response of ablC to C02 via lateral inhibition: the reduced C02 response decreased the avoidance of the arm containing C02 and ACV relative to the arm containing C02 alone, and this decreased avoidance is seen as an attraction to the arm containing C02 and ACV. If this interpretation is correct, and the preference for the arm containing C02 and ACV depends on the activation of abl A, then the preference should be abolished in Oreo mutants, which lack a co-receptor required for the response of ablA but not ablC. Consistent with this prediction, Oreo mutants showed no preference between the arm containing C02 and the arm containing C02 and ACV (Fig. 5c). We note that ACV does not inhibit ablC directly (Supplementary Fig. 4). Taken together, these results provide evidence that lateral inhibition within a sensillum can modulate behaviour. Discussion Integration of olfactory information has long been known to occur in the CNS, and has more recently been shown to occur in individual ORNs39. We have demonstrated that integration also occurs at a third 6 DECEMBER 2012 I VOL 4 9 2 I NATURE I 69 ©2012 Macmillan Publishers Limited. All rights reserved RESEARCH 0.4 0.2 0.0 -0.2 -0.4 -0.6 CO, ACV CO, CO,+ACV CO, ACV CO, CO,+ACV CO, ACV CO, CO,+ACV 0.4 0.2 0.0 -0.2 -0.4 -0.6 0.4 0.2 0.0 -0.2 -0.4 -0.6- Figure 5 | Lateral inhibition modulates behaviour, a-c, Activation of ablA mediates attraction to ACV; activation of ablC mediates aversion to C02. Two of the four ORNs in abl are depicted. Preference indices of control (a), Orco-GAL4; UAS-TNT (b) and Oreo (c) are shown (mean ± s.e.m.). The ablA neurons of the TNT-expressing flies respond to ACV but are expected not to transmit information to postsynaptic neurons, whereas ab 1A neurons in Oreo are expected not to respond to ACV. In each T-maze assay, —50 flies were allowed 2 min to choose. In single-odour experiments (black bars) the test arm contained either C02 or ACV. *P < 0.05; ***p< 0.001, f-test (n = 16). C02 was 0.13%; ACV was 100%, pH 7.5. In physiological recordings from Oreo flies, ACV did not inhibit the spontaneous firing of the C02 neuron. level, the sensillum, via lateral inhibition between ORNs responding to different components of a mixture. The sensillum thus acts as a processing unit in olfactory computation. Lateral inhibition of a prolonged signal by a transient signal may provide a neural representation of the salience of an odour that has recently reached the fly40. Sustained responses were inhibited more strongly by stronger transient pulses. This graded pattern of lateral inhibition may give rise to a potent form of contrast enhancement in which the output of a sensillum is dominated by a pulse of a strong odour. Graded lateral inhibition may provide a peripheral mechanism for evaluating countervailing signals and allowing one to prevail. We note that in Drosophila, an ORN that responds to a pheromone41-42 is the only ORN that does not have a neighbour, as if to ensure that its sustained response is not inhibited by a pulse of any other odorant. Our finding that lateral inhibition does not require synapses is consistent with anatomical data. Electron microscopy in Drosophila has not revealed synaptic structures or gap junctions between ORNs housed in the same sensillum2-3. Rather, as detailed below, the physiological features of olfactory sensilla suggest another mechanism of lateral information flow: ephaptic transmission, which refers to non-synaptic communication between adjacent neurons through an extracellular electrical field43-44. The ability of either neuron in a two-neuron sensillum to inhibit the other, as well as the grossly similar temporal dynamics of activation and lateral inhibition (Supplementary Fig. 5), are consistent with ephaptic transmission. In insect olfactory sensilla, a substantial electrical potential exists between two isolated compartments: the sensillum lymph, which bathes the dendrites, and the haemolymph, which surrounds the somata (Fig. la and Supplementary Fig. 6). This 'transepithelial' potential serves as the primary driving force for odorant-induced transduction currents of the ORNs17-45. Elaboration of an established electrical circuit model17-45 based on these physiological features predicts that strong activation of one ORN will hyperpolarize the soma of a co-compartmentalized ORN (Supplementary Fig. 6), resulting in a reduced firing rate. This prediction is consistent with the results of our molecular genetic analysis and with our interpretation that lateral inhibition is due to ephaptic interactions. The model further predicts that the magnitude of the hyperpolari-zation of the neighbouring neuron, and hence its reduction in firing rate, is reflected by the change in the transepithelial potential (VA) (Supplementary Fig. 6), measured experimentally as a local field potential (LFP) (Supplementary Fig. 7a). Although strong activation of an ORN can influence the LFP in a neighbouring sensillum46, we found that the magnitude of the LFP change in nearby unstimulated sensilla is small (Supplementary Fig. 7). Consistent with this observation, lateral inhibition does not spread among homotypic sensilla that are in close proximity to one another (Supplementary Fig. 8). These results further support the conclusion that the lateral inhibition is due to local electrical interactions between neighbouring ORNs within a sensillum. The two-odour paradigm used in this analysis, in which a transient odour is superimposed upon a sustained odour, differs from the classic one-odour paradigm in which a transient pulse of a single odour is delivered. A priori one might expect to observe ephaptic effects in the one-odour paradigm if one ORN were excited sufficiently strongly, but the effects may be expected to be less pronounced than in the two-odour paradigm. ORN spike frequency is determined not only by the somatic transmembrane potential Vm, but also by its rate of change, dVm/df (ref. 46). According to the model, transient activation of ORN2 reduces the depolarizing current of ORNi (Supplementary Fig. 6). In the two-odour paradigm, activation of ORN2 has a marked effect on the value of d Vml/df, which changes from 0 to a negative value (dVml/df