Odor-Induced Multi-Level Inhibitory Maps in Drosophila

Abstract Optical imaging of intracellular Ca2+ influx as a correlate of neuronal excitation represents a standard technique for visualizing spatiotemporal activity of neuronal networks. However, the information-processing properties of single neurons and neuronal circuits likewise involve inhibition of neuronal membrane potential. Here, we report spatially resolved optical imaging of odor-evoked inhibitory patterns in the olfactory circuitry of Drosophila using a genetically encoded fluorescent Cl- sensor. In combination with the excitatory component reflected by intracellular Ca2+ dynamics, we present a comprehensive functional map of both odor-evoked neuronal activation and inhibition at different levels of olfactory processing. We demonstrate that odor-evoked inhibition carried by Cl- influx is present both in sensory neurons and second-order projection neurons (PNs), and is characterized by stereotypic, odor-specific patterns. Cl--mediated inhibition features distinct dynamics in different neuronal populations. Our data support a dual role of inhibitory neurons in the olfactory system: global gain control across the neuronal circuitry and glomerulus-specific inhibition to enhance neuronal information processing.


Introduction
Inhibition of neural excitability is a ubiquitous feature of all neuronal circuits. Neurons that release inhibitory transmitters are present in all parts of the nervous system. In the olfactory systems of both insects and vertebrates, inhibition is crucial for stimulus gain control (Olsen and Wilson, 2008;Root et al., 2008), synchronizing neural networks (Laurent et al., 2001), generating precise timing (Schoppa and Westbrook, 1999;Margrie and Schaefer, 2003), broadening transmission of olfactory signals (Nagel et al., 2015), odor mixture interactions (Mohamed et al., 2019), and enhancing contrast between similar odor representations (Mori et al., 1999;Sachse and Galizia, 2002;Urban, 2002). In the mammalian olfactory bulb, inhibition is largely mediated by dendrodendritic synaptic connections between excitatory mitral cells and inhibitory granule cells (Schoppa and Urban, 2003;Egger and Urban, 2006). Despite these important roles of inhibition for odor processing, most studies analyzing olfactory coding at the level of spatially distributed neuronal populations focused on monitoring neuronal excitation. Therefore, odor representations at the level of the insect antennal lobe (AL) or the vertebrate olfactory bulb typically have been characterized as patterns of excitation. Here, we aimed at monitoring spatially distributed maps of odor-evoked inhibition at different levels of processing in Drosophila melanogaster.
In the fly, odors are detected by olfactory sensory neurons (OSNs) located on the antennae and maxillary palps. Each OSN typically expresses one or very few chemo-receptor genes, and each OSN projects its axon to the AL, the insect analog of the vertebrate olfactory bulb. In the AL, those OSNs expressing the same odorant receptor (OR) stereotypically converge to the same spatially invariant olfactory glomeruli (Couto et al., 2005;Fishilevich and Vosshall, 2005), each of which can be unambiguously identified (Laissue et al., 1999;Grabe et al., 2015). The AL is densely innervated by local interneurons (LNs) that mediate both intraglomerular and transglomerular inhibition (Wilson and Laurent, 2005;Seki et al., 2010). Olfactory projection neurons (PNs) convey the olfactory signals to higher-order brain centers.
The morphologic structure of the AL network specifies the physiologic logic of how odors are encoded: Each odorant evokes a characteristic, spatiotemporal activity pattern leading to a combinatorial, stereotypic activation of glomeruli in the AL (Fiala et al., 2002;Wang et al., 2003). Inhibitory LNs provide both feedforward synaptic inhibition of PNs and feedback inhibition of OSNs (Wilson and Mainen, 2006;Olsen and Wilson, 2008;Root et al., 2008). However, it still remains elusive how spatially distributed, odor-evoked inhibition interferes with and relates to the well-described excitation-based odor maps.
In Drosophila, functional imaging has mainly relied on genetically expressed Ca 2ϩ sensors that detect intracellular Ca 2ϩ dynamics as a correlate of neuronal excitation (Grienberger and Konnerth, 2012). In this study, we monitored odor-induced inhibitory maps in the olfactory circuitry using a DNA-encoded indicator for Cl -, the main ionic mediator of synaptic inhibition in mature neurons (Owens and Kriegstein, 2002). The FRET-based indicator Clomeleon consists of a Cl --sensitive yellow fluorescent protein (YFP) and a Cl --insensitive cyan fluorescent protein (CFP; Kuner and Augustine, 2000). Binding of Clto YFP reduces its absorbance, which results in a change of the YFP/CFP emission ratio proportional to [Cl -] i . The applicability of Clomeleon in vivo has been demonstrated in hippocampal slices (Berglund et al., 2006), retinal bipolar cells (Haverkamp et al., 2005;Duebel et al., 2006), thalamo-cortical neurons of mice (Glykys et al., 2009) and cerebellar granule cells (Berglund et al., 2016).
We genetically expressed Clomeleon in defined olfactory neurons and characterized odor-evoked inhibition at different levels of olfactory processing in comparison with Ca 2ϩ -mediated activity using the likewise FRET-based Ca 2ϩ -sensitive protein Cameleon 2.1 (Miyawaki et al., 1999). First, we observed odor-evoked Cl --influx in dendrites of OSNs. Second, we generated a comprehensive functional map of both odor-evoked activation and inhibition of the fly AL. We demonstrate that odor-evoked inhibition carried by Clinflux is characterized by stereotypic odor-specific patterns. Third, we show that Cl -mediated inhibition exhibits distinct features at different levels of olfactory processing pointing toward multiple roles of inhibition in the olfactory system.

Optical imaging
For imaging intracellular Cland Ca 2ϩ dynamics in the AL, flies were restrained in custom-built holders and a small window was cut into the head capsule. The hole was covered with physiologic saline solution, and imaging was performed using a water immersion objective directly positioned above the exposed brain. Pharmaca [GABA, potassium gluconate (KGlu), PTX, 5-nitro-2(-3-phenylpropylamine) benzoic acid (NPPB)] were applied by exchanging the saline drop on the brain by a drop of the approximate volume and the targeted concentration. For NPPB an additional ethanol application was conducted to control for the solvent effect (data not shown). For transcuticular antennal imaging, flies were restrained as for the in vivo dissection method without opening the head capsule. Or22aexpressing OSNs were imaged from the posterior side of the antenna, while the majority of OSNs using Orco-Gal4 were recorded from the anterior side.
Imaging experiments were performed using TillPhotonics imaging equipment (TILL imago, Till Photonics GmbH) with a CCD-camera (PCO imaging, Sensicam) and a fluorescence microscope (Olympus, BX51WI) equipped with a 20ϫ water immersion objective (NA 0.95, XLUM Plan FI, Olympus) for AL imaging and a 10ϫ air objective (NA 0.30, UPlan FLN, Olympus) for antennal imaging. A monochromator (Polychrome V, Till Photonics) provided light at 440 nm excitation wavelength which was guided through a 470-nm dichroic short pass filter. The beam-splitter (Optical Insights, DV-CC) separated YFP from CFP emission with a 505 DCXR and narrowed the emissions with bandpass filters of 535/30 nm for YFP and 465/30 nm for CFP. Images of both emitted wavelengths were projected side by side onto a single CCD camera chip (PCO Imaging, Sensicam). Fourfold binning on the CCD-camera chip resulted in an image size of 344 ϫ 260 pixels with one pixel corresponding to 1.25 ϫ 1.25 m. Each recording lasted for 20 s with an acquisition rate of 2 Hz. Since Clomeleon yielded a very low signal-to-noise ratio, we had to apply long exposure times which limited our recording frequency. We also performed experiments with the usually used frequency of 4 Hz, resulting in weaker signal intensities and a lower dynamic range. Since these signals did not reveal different temporal patterns in the odor-evoked responses as the slower recorded signals, we decided in favor of an increased signal-to-noise ratio and maintained a recording frequency of 2 Hz for the whole study. Odors were applied 2 s after experiment onset and lasted for 2 s. Individual flies were imaged for up to 1 h, with interstimulus time intervals of 1-3 min.

Odor stimulation
Pure odorants were diluted in mineral oil (BioChemika Ultra; odor CAS: ethyl-3-hydroxybuytrate: 5405-41-4, benzaldehyde: 100-52-7, acetic acid: 64-19-7, cis-vaccenylacetate: 6186-98-7, pentyl acetate: 628-63-7, 1-hexanol: 111-27-3, ethyl benzoate: 93-89-0). For use, 6 l of 1:10 diluted odor was pipetted onto a small piece of filter paper (100 mm 2 , Whatman), which was inserted into a glass Pasteur pipette. A stimulus controller (Syntech, Stimulus Controller CS-55) was used to apply the odor into a continuous airstream at 1 l/min, monitored by a flow meter (Cole Parmer). An acrylic glass tube guided the airflow to the fly antennae. Two additional air sources (airflow 0.5 l/min) were connected to the tube and the stimulus controller. One of them consisted of the glass pipette containing the odor on filter paper and was hooked up for odor application, whereas the other pipette was empty and added clean air to the continuous airstream forming an air equation which was switched off during odor application.

Data analysis
Data were analyzed using custom-written IDL software (ITT Visual Information Solutions). First, a rigid registration was accomplished for all recordings separately to minimize movement artifacts throughout the time series. Second, the ratio of the two fluorescent signals was calculated as F YFP /F CFP for every time point. The ratio (R) represents the relative magnitude of the signal intensity. To achieve a comparable standard for the calculation of the relative fluorescence changes of the ratio (⌬R/R), the fluorescence background was subtracted from the averaged values of frames 0 -5 in each measurement, such that baseline fluorescence was normalized to zero. The false color-coded fluorescence changes in the raw-data images were calculated as the delta of frame 5 and 30 (Clomeleon) and frame 5 and 15 (Cameleon). Specific time traces of a measurement depict the mean of a 7 ϫ 7-pixel coordinate (i.e., 9 ϫ 9 m), which was positioned into an anatomically identified glomerulus and plotted as a function over time. Since GH146-Gal4 does not label glomeruli VM5d and VM5v, these could not be characterized at the PN level (Grabe et al., 2015). To generate schematic AL maps, the mean value of frames 10 -30 for Clomeleon and 10 -15 for Cameleon of a specific glomerulus and odor was averaged over all animals imaged. Although the chloride and calcium kinetics are clearly odor induced, they develop very slowly over time and show their maximal response change after odor offset. We therefore selected a delayed time window for our signal evaluation to capture the maximum/minimum of the odor-induced responses. One has to keep in mind that the monitored Ca 2ϩ and Cldynamics are also dependent on the kinetics and concentrations (i.e., expression levels) of the fluorescent sensors and might not reflect accurately the physiologic time traces. However, this issue is more relevant for fast stimulus dynamics (Martelli and Fiala, 2019), while with regard to slow recording frequencies, as used here, the resulting kinetics of Cland Ca 2ϩ binding are rather negligible.
Responses in each fly were normalized to the highest Clor Ca 2ϩ signal in each animal over all odors. To extract the temporal aspect of odor separation in the different neuronal populations, Euclidean distances (L2-Norm) were calculated. To compare the results, we always used the same set of 11 identified glomeruli in each data set. For a given stimulus a, the n-dimensional population vector (v a ) was constructed using the relative fluorescence changes over time. Then the population vectors of two stimuli, a and b, were used to calculated the distance for every single data point (time) in the 40 frames as follows: ) 2 ) 1/2 , where i is an index for the i-th glomerulus. In addition to the Euclidean distances, principal component analysis was used to visualize the population activity of OSNs and PNs depending on the imaged reporter protein. Population vectors of all odor stimulations were aligned, taking into account time as the source of sample points, and number of glomeruli as the dimension of the original component space using the MATLAB statistical toolbox. All statistical analyses were performed using GraphPad InStat 3 as specified in each figure legend.

Clomeleon as an indicator of intracellular Cldynamics in olfactory neurons
We generated flies carrying the genetically encoded Clsensor Clomeleon (Kuner and Augustine, 2000) to visualize in vivo Cl --mediated inhibitory responses in the olfactory system. Using the binary GAL4-UAS transcriptional system (Brand and Perrimon, 1993), we expressed Clomeleon in the majority of OSNs using Orco-Gal4 (Wang et al., 2003) and in PNs using GH146-Gal4 (Stocker et al., 1997;Fig. 1A,B). To test whether Clomeleon is functional in Drosophila olfactory neurons, we optically monitored fluorescence changes in OSNs and PNs in the AL, while we applied KGlu to induce neuronal excitation globally and, concomitantly, inhibition through LN input onto OSNs and PNs ( Fig. 1C-F; see network scheme in Fig.  2A). Applying KGlu increased CFP fluorescence, while YFP fluorescence was strongly decreased; thereby, the YFP/CFP ratio was reduced. To verify that this reflected inhibition, we applied the inhibitory transmitter GABA. GABA application immediately reduced the YFP/CFP ratio in both OSNs and PNs (Fig. 1G,H). Notably, we observed a second, strong emission decrease which was delayed by ϳ35 s. The source of this second decrease is yet unclear, but could be due to the slow diffusion rate of GABA as it is bath applied to the whole brain and not actively perfused. The gradually increasing GABA concentration might surpass a threshold that initiates a strong inhibition reflected by the second phase. In combination with a gradual desensitization toward GABA , this could explain the observed slow and biphasic GABA effect. To confirm that these ratio changes were dependent on Cl -, we removed Clfrom the saline solution covering the fly's brain. Odor application before Clremoval induced a clear ratio change, which was significantly reduced using Cl --free saline (Fig. 1I). To further verify that our reporter was reflecting the intracellular Clconcentration, we applied the chloride channel blocker NPPB to block Clchannels in Drosophila neurons (O'Donnell et al., 1998). As expected, application of NPPB strongly reduced the Clinflux which was partly reversibly (Fig. 1J).
Since the YFP fluorescence has been reported to be affected by the pH value at [Cl -] i above 50 mM (Kuner and Augustine, 2000), we verified that the fluorescence emission was not influenced by pH changes within the physiologically relevant range of 6.9 -7.3 (Fig. 1K). This is in accordance with the described [Cl -] i in OSNs, which is ϳ24 mM in moths (Steinbrecht, 1992) and ϳ20 mM in flies (Reinert et al., 2011). Therefore, a potential influence of pH changes on Clomeleon is negligible. Overall, our results confirm that Clomeleon functions reliably as a Clindicator in olfactory neurons of the Drosophila AL.

Odor stimulation induces peripheral Clinflux in dendrites of OSNs
Next, we analyzed whether odor stimulation causes a Clincrease at the most peripheral level of sensory transduction and performed transcuticular Climaging in OSN dendrites located on the fly's antenna ( Fig. 2A,B). Odor stimulation induced an odor-specific, spatially confined increase in [Cl -] i . These spatially restricted signals correspond to distinct sensillum types, which have welldescribed, specific distributions on the third antennal segment (Shanbhag et al., 1999;Grabe et al., 2016;Fig. 2C,D). To determine which sensillum types were excited by the odors used, we performed Ca 2ϩ imaging in comparison using the ratiometric Ca 2ϩ indicator Cameleon 2.1 (Miyawaki et al., 1999). Clsignals are characterized by a reduction in the Clomeleon's YFP/CFP ratio (ϭ increase in [Cl -] i ), whereas Ca 2ϩ signals are indicated by a ratio increase in the Cameleon's YFP/CFP ratio (ϭ increase in [Ca 2ϩ ] i ; Fig. 2D). Notably, some odors, such as ethyl-3hydroxybutyrate, evoked both a Ca 2ϩ and a Clsignal in the same areas of the antennal surface, indicating a concomitant Ca 2ϩ and Clinflux in OSNs. Other odorants, e.g., benzaldehyde, induced spatially non-overlapping Cland Ca 2ϩ signals, indicating independent excitation and inhibition events in distinct sensilla (Fig. 2D). This separation of inhibition and excitation is underlined by the strong Clincrease induced by acetic acid in the tip region of the antenna without significant Ca 2ϩ responses (Fig. 2E). Acetic acid activates solely OSNs present in one type of coeloconic sensilla (Abuin et al., 2011), which is not labeled by the Orco-Gal4 line. To verify that the observed antennal Clsignals indeed reflect neuronal inhibition, we expressed Clomeleon selectively in OSNs expressing Or47b. OSNs expressing this receptor selectively respond to the pheromone methyl laurate and are mainly inhibited by other odors (Hallem and Carlson, 2006;Dweck et al., 2015). Application of the odor ethyl-3-hydroxybutyrate, which induces an inhibition of Or47b-expressing OSNs as shown via single-sensillum recordings (Hallem and Carlson, 2006), leads to a strong and long-lasting Clinflux in this OSN type (Fig. 2F).
We next wondered whether the odor-induced antennal Clincrease derives within the sensillum and can be attributed to OSN dendrites or whether these signals rather reflect a feedback inhibition from the AL. We therefore monitored Clsignals following odor stimulation while we abolished any feedback signaling from the AL by cutting the antennal nerve (Fig. 2G). Interestingly, this treatment significantly reduced Clsignals in the antenna, but did not abolish them. This result demonstrates Clconductivity in dendrites of OSNs, indicating that Clchannels are present in OSNs and localized to the most peripheral dendritic compartments in the fly antenna. However, at

Cl --dependent, inhibitory odor maps in OSN terminals in the AL
Within the AL, OSNs are presynaptically inhibited by GABAergic LNs (Olsen and Wilson, 2008;Root et al., 2008;Mohamed et al., 2019) with varying and glomerulusspecific GABA sensitivities . To visualize odor-evoked inhibition at the level of the axonal termini across multiple glomeruli, we performed Climaging in presynaptic OSN axons in the AL using an in vivo preparation Fig. 3A). Due to the stereotypy of the glomerular AL morphology, we could reliably identify individual glomeruli in each animal using digital AL atlases (Laissue et al., 1999;Grabe et al., 2015;Fig. 3B). Each odor stimulation induced a specific combinatorial pattern of inhibited glomeruli, which was stereotypic among individuals (Figs. 3C). The time courses of YFP/CFP ratio changes in selected glomeruli revealed an odor-and glomerulus-specific Clinflux (Fig. 3D). However, a time-resolved analysis across multiple glomeruli showed that Clsignals are detected in all glomeruli optically accessible during the imaging experiments (Fig.  3E). In conclusion, strong and odor-specific inhibition of distinct glomeruli is accompanied by less intense, global inhibition across the entire OSN population. The Clsignals that were optically monitored lasted until the end of each measurement, i.e., they strongly outlasted the 2-s odor stimulation. Therefore, we examined how much time was required before the Clomeleon signal returned to baseline ( Fig. 3F-H). Odor application with different interstimulus intervals revealed that although the fluorescence emission (⌬R/R) continued to drop after stimulation, repetitive odor stimulation still elicited further Clomeleon signals after 10 or 60 s (Fig. 3F,G). A complete recovery of the Clomeleon fluorescence was not observed before 120 s after odor stimulation had elapsed (Fig. 3H). The actual kinetics of any fluorescence sensor depend on multiple factors, e.g., the concentration of the sensor determined by the expression level, the affinity of the sensor to its ligand, or the dynamic range of the sensor. Therefore, it is difficult to conclude to what degree the dynamics of Cltransients quantitatively reflect the actual balance between Clinflux and intracellular Clremoval. However, a slow recovery of Clomeleon signals has also been observed in mammalian neurons (Kuner and Augustine, 2000;Berglund et al., 2006) and has been attributed to the slow removal of [Cl -] i by transporters rather than kinetic properties of the Clsensor (Staley and Proctor, 1999;Berglund et al., 2009Berglund et al., , 2016. It is therefore quite conceiv-able that the odor-evoked Cl-transients in OSNs indeed strongly outlast the actual stimulation.

Comparison between odor-evoked Clsignaling in OSN dendrites and axons
As shown so far, odors induce a clear Clincrease at the level of the peripheral signal input, i.e., in the antenna (Fig. 2), and at the sites of synaptic transmission, i.e., in OSNs of the AL (Fig. 3). To examine the relationship between these two signal sources in more detail, we comparatively monitored odor-evoked [Cl -] i and [Ca 2ϩ ] i of a single OSN population at its dendrites and axonal termini. This was achieved by selective expression of Clomeleon or Cameleon, respectively, in OSNs expressing Or22a, which targets the glomerulus DM2 (Couto et al., 2005;Fishilevich and Vosshall, 2005). As described previously (Pelz et al., 2006), a strong Ca 2ϩ response was elicited by methyl hexanoate, while ethyl-3hydroxybutyrate induced an intermediate, and benzaldehyde no significant response (Fig. 4A,B). The relative intensities of odor-evoked Ca 2ϩ responses did not differ between antenna and AL (Fig. 4C). However, all three odors induced comparatively strong Clresponses in the fly antenna (Fig. 4A, lower panel), while only methyl hexanoate, one of the most potent activators of this OSN type, elicited a significant Clresponse at the AL level (Fig. 4B, lower  panel, D). Hence, the intensity of odor-evoked Clinflux at the level of OSN dendrites and somata is relatively independent of the actual intensity of the accompanying Ca 2ϩ influx. On the contrary Cl --mediated inhibition in the AL reflects more odor-specific inhibition.

Cl --dependent, inhibitory odor maps in PN dendrites in the AL
To analyze inhibitory patterns of output neurons in the AL, we performed Climaging at the dendrites of PNs using the enhancer trap line GH146-Gal4 that labels the majority of uniglomerular PNs (Stocker et al., 1997). Odor application induced clear spatially confined and odorantspecific patterns of inhibition that could be assigned to identified glomeruli (Fig. 5A,B). A time-resolved analysis across all glomeruli revealed a strongly pronounced Clinflux in a glomerulus-and odor-specific manner, and typically with some delay after odor onset (Fig. 5C). These odor-specific, inhibitory patterns evolve slowly over time and persist until the end of the measurement, as it is the case at the OSN level. Notably, we observed a concordance in the Clresponses between OSNs and PNs, in a way that a given odor inhibited the same glomeruli at the input and the output level of the AL (Figs. 3E, 5C). However, this correlation was only apparent for strongly inhib-   Since the spatiotemporal activity of PN ensembles is influenced by inhibitory, GABAergic LNs (Wilson and Laurent, 2005), we tested whether the Clomeleon signals were dependent on GABA receptors. Therefore, we performed Climaging experiments after silencing the inhibitory LN input by applying the GABA A -type antagonist picrotoxin (5 M) that blocks ionotropic Cl --ion channels. In addition to GABA A receptors, picrotoxin has been shown to block also glutamate-gated chloride channels (GluCl; Liu and Wilson, 2013). However, at the low concentration used in this study the antagonist mainly functions as a GABA A antagonist without affecting GluCl channels . Application of picrotoxin led to a significant reduction of the odor-induced Clsignals by on average 59% (Fig. 5D). This result indicates that the GABA A receptor contributes to the Cl --mediated inhibition at the AL output level.

A comparative functional map of odor-evoked activation and inhibition in the AL
We next examined the overlap of the odor-evoked inhibitory patterns compared to the spatial patterns of glomerular Ca 2ϩ activities. To do so, we performed functional imaging experiments to a variety of different odors and monitored odor-evoked Ca 2ϩ as well as Clresponses by expressing Cameleon or Clomeleon in OSNs and PNs, respectively. Subsequently, we mapped the odor-induced responses to identified glomeruli to generate a functional AL atlas (Fig. 6). First, we observed that, in the majority of cases, the odorevoked maps of excitation and inhibition closely match at the input and the output level, i.e., those glomeruli which were excited were also often inhibited by a certain odor. Such a concordance suggests a gain control mechanism for odor-induced excitation as described for the OSN level (Olsen and Wilson, 2008), which should occur in all glomeruli receiving an excitatory input. Second, we observed that some glomeruli were inhibited without being excited. This finding suggests a second role of Cl --mediated inhibition in the Drosophila AL which could contribute to confining the spatiotemporal patterns, resulting in an enhanced contrast between different odor representations as shown for the honeybee AL . Notably, we never observed glomeruli, which were excited without being inhibited.

Input-output transformation
Last, we analyzed the difference between the odorevoked representations of input and output neurons for a subgroup of 11 glomeruli that could be unambiguously identified in each experiment (Fig. 7A). Since each fluorescent reporter protein exhibits different kinetics, one has to be careful when comparing temporal dynamics between different sensors. We therefore compared tem-poral aspects of odor-evoked responses of different processing levels for one reporter protein only. Quantification of the evoked mean responses to specific odors showed that excitatory as well as inhibitory odor responses were, on average, stronger at the PN level than at the OSN level (Fig. 7B) which is well in line with electrophysiological recordings (Wilson and Laurent, 2005;Bhandawat et al., 2007;Seki et al., 2017). To visualize how the odor-specific responses evolve over time, we applied principal component analyses to reduce the multidimensional, spatiotemporal activity/inhibition to three dimensions and illustrated the odor-evoked ensemble activity as trajectories over time (Fig. 7C). Independent of the reporter protein, different odors evoked distinct trajectories, which demonstrate an odor-specific separation of Ca 2ϩ as well as Clresponses at both processing levels, i.e., OSNs and PNs. To quantify how fast this odor separation evolved, we calculated Euclidean distances between the population vectors of the different odor representations for Cameleon and Clomeleon signals, respectively (Fig.  7D,E, upper panels). Interestingly, PN responses revealed in general lower Euclidean distances than OSN responses. Although PNs showed an increased level of inhibition, they also exhibited generally broader odorevoked responses compared to OSNs (Wilson et al., 2004;Seki et al., 2017). This broadening leads to wider odor tuning curves and a stronger overlap of odor representations at the PN level (Niewalda et al., 2011;Schubert et al., 2014;Seki et al., 2017), while PN responses show a higher degree of odor categorization according to behaviorally meaningful values (Niewalda et al., 2011;Knaden et al., 2012).
After normalizing all pair-wise Euclidean distances, we calculated the latencies to the half maximum odor separation (Fig. 7D,E, lower panel) and observed that it was reached significantly earlier in PNs than in OSNs. This finding is in accordance with electrophysiological recordings in Drosophila showing that PN responses have shorter latencies to reach 90% of their response peak than OSNs (Bhandawat et al., 2007) indicating that PNs act as high-pass filters that rapidly convey rising OSN responses to third-order neurons. When considering the Clomeleon responses, this latency shift is even more pronounced for Clsignals. This observation is most likely due to reciprocal inhibitory mechanisms that differently affect OSN and PN responses: PNs are inhibited by fast forward inhibition from OSNs via GABAergic LNs (Wilson and Laurent, 2005) before OSNs receive presynaptic feedback inhibition from PNs through, in turn, GABAergic LNs (Olsen and Wilson, 2008;Root et al., 2008). continued stimulation is marked in gray. E, False color pictures of averaged odor-evoked Clsignals for 14 glomeruli (42% of all glomeruli labeled by Orco-Gal4) over time across six to nine animals. Clomeleon responses were normalized to highest Clinflux in each animal over all odors before averaging. Black bar indicates odor application. F-H, Representative time courses of Clinflux to repeated stimulations of ethyl-3-hydroxybutyrate using interstimulus intervals of 10 s (F), 60 s (G), and 120 s (H). Odor stimulations are marked in gray.

Discussion
Clomeleon-based Climaging in the Drosophila nervous system Hardly any optical imaging technique reaches the unmatched temporal precision of electrophysiological recordings as yet, and the determination of membrane potential changes represents the most accurate approach to determine how sensory stimuli are represented by single or small groups of neurons (Wilson et al., 2004;Wilson and Laurent, 2005;Seki et al., 2017). Optical imaging, on the contrary, offers the advantage of monitoring physiologic parameters that correlate with membrane potential changes across spatiotemporally distributed populations of neurons (Ahrens et al., 2013;Chen et al., 2013). Membrane depolarization is typically accompanied by increases in intracellular Ca 2ϩ from a variety of sources, and Ca 2ϩ imaging represents currently the "gold standard" for visualizing neuronal excitation in Drosophila . However, neuronal inhibition, most often mediated by Clinflux, is not directly captured using Ca 2ϩ imaging. Establishing Clomeleon as a tool for monitoring Cldynamics both in the peripheral and central nervous system provides an important step toward filling this gap. Its ratiometric nature as a FRET-based sensor demands some additional considerations in contrast to single chromophore sensors such as calcium reporters belonging to the GCaMP family (Tian et al., 2009). Especially the size, pH sensitivity, and slow response dynamics require the future development of a single chromophore chloride sensor, which hopefully eases its applicability. The development of the Clsensor SuperClomeleon, which still represents a FRET-based sensor, reveals an improved signal-to-noise ratio and needs to be established for the Drosophila olfactory system (Grimley et al., 2013).
It is important to consider that the monitored changes in intracellular Ca 2ϩ and Clderive from different cell processes within the neurons. Recorded changes in Ca 2ϩ and Clcan therefore depend on fluxes at the synapse or along the neuron, as well as release from intracellular calcium stores, which are mediated by ligand-gated as well as voltage-gated Ca 2ϩ and Clchannels in insects (Messina et al., 1996;Wicher et al., 2001;Fiala and Spall, 2003;Flores et al., 2006;Pézier et al., 2010). Since the temporal resolution in functional imaging recordings is rather low compared to electrophysiological recordings, the different dynamics of these ion channels are not visible in the fluorescence signal of the different sensors.

Is Clinflux part of the olfactory signal transduction in insects?
We observed an odor-evoked Clinflux in OSN dendrites of the Drosophila antenna. In vertebrates, Cl -conductance is an integral component of the canonical olfactory signal transduction cascade (Labarrera et al., 2013). Here, odor stimulation leads to a membranecurrent composed of a cationic and a delayed Clcomponent (Kurahashi and Yau, 1993). Although Cl -conductance is in most cases associated with neuronal inhibition, this Clcurrent amplifies the olfactory signal by Clefflux through a Ca 2ϩ -activated Clchannel which is most likely mediated by anoctamin-2 (ANO2; Lowe and Gold, 1993;Stephan et al., 2009;Delgado et al., 2016). The insect olfactory signal transduction is crucially different from that of vertebrates in two aspects: First, olfactory receptors of the OR and IR type are ionotropic receptors mediating excitatory cation influx (Sato et al., 2008;Wicher et al., 2008;Rytz et al., 2013). Metabotropic signaling cascades have been clearly described for insect OSNs, but their exact modulatory functions remain unclear as yet (Wicher et al., 2008). Second, the equilibrium potential of Cl -(E Cl ) in insect OSNs differs from that of vertebrates. Since [Cl -] i is lower than in the extracellular medium, as shown in moths (Steinbrecht, 1992), the electromotive force will lead to a Clinflux, if the membrane potential is shifted above E Cl (i.e., -36 mV). Hence, when OSNs become excited, a Clinflux through Ca 2ϩactivated Clchannels might result in hyperpolarization of the plasma membrane (Pézier et al., 2010).
Interestingly, dendrites of moth OSNs express an analogous Ca 2ϩ -activated Clchannel that functionally resembles ANO2 (Pézier et al., 2010). The Drosophila melanogaster genome contains two different ANO2 orthologues (CG6938, CG10353) whose molecular function has, however, not yet been studied. Additional experiments are needed to analyze the role of ANO2 in odor-evoked Cldynamics in Drosophila OSNs. The fact that the antennal Clinflux co-occurred frequently with a Ca 2ϩ influx further suggest the existence of Ca 2ϩ -activated Clchannels in the antenna. This type of Cl --mediated inhibition might reflect shunting inhibition as a mechanism for gain control leading to stabilization of odorevoked excitation (Wilson and Mainen, 2006).
In addition, we also observed Clinflux that was not directly correlated with the excitation of the respective OSNs, reflecting a second type of Cl --mediated inhibition. This finding suggests either again the existence of Cl -channels in OSN dendrites or, alternatively, a retrograde diffusion of Clfrom the AL. Interestingly, when we abolished any feedback signaling from the AL, we still ob-continued left, Schematic of the Drosophila AL indicating selective expression of Cameleon or Clomeleon in axonal termini of Or22a-expressing OSNs which converge to glomerulus DM2. Right, Averaged time courses of Ca 2ϩ (upper row) and Clinflux (lower row) in DM2 to three different odors. Odor stimulation is marked in gray. Lines represent means, color shading represents SEM (n ϭ 6). C, D, Quantification of Ca 2ϩ (C) and Cl -(D) influx in Or22a-expressing OSNs to three different odors and mineral oil. Data are shown as pair-wise comparisons between antenna (Or22a) and AL (DM2). Clomeleon and Cameleon responses have been normalized to highest Clor Ca 2ϩ influx in each animal over all odors, respectively. Clresponses to ethyl-3-hydoxybutyrate and benzaldehyde are significantly lower in the AL compared to the antenna ‫‪p‬ءء(‬ Ͻ 0.01, ‫‪p‬ءءء‬ Ͻ 0.001, Mann-Whitney test, n ϭ 6 -7). served Clinflux, supporting the first assumption. However, since the Clsignals were not identical but reduced, we assume that Cldynamics in OSN dendrites are partly influenced by Clinflux into OSN axonal termini in the AL. The latter assumption is further supported by our observation that applying GABA to the AL induced a significant Clinflux in the antenna (data not shown).
The comprehensive study by Hallem and Carlson on receptor-ligand interactions where a widespread inhibition below baseline firing rates among one third out of 24 selectively expressed ORs was observed (Hallem and Carlson, 2006), is well in line with our observation of inhibitory odor responses in the Drosophila antenna. Interestingly, OSNs expressing Or47b, known to selectively respond to the pheromone methyl laurate (Dweck et al., 2015), were never excited by the large odor set tested in the aforementioned study, but showed inhibitory responses to 34% of the odors. Those OSNs target glomerulus VA1d, and we indeed observed clear odor-evoked Clresponses in VA1d, while Ca 2ϩ -influx did never occur. In addition, Climaging of Or47b-expressing OSNs on the antenna confirms the odor-induced inhibition of this OSN type. As a second example, benzaldehyde elicited a strong Clinflux in OSNs expressing Or22a in the antenna without being accompanied by a Ca 2ϩ influx. This odor has already been characterized as an Or22a-inhibitor (Pelz et al., 2006;Wicher et al., 2008), which strongly suggests that our second type of Cl --mediated inhibition reflects hyperpolarization and thus odor-specific inhibition in the antenna. Therefore, our study demonstrates that inhibitory odor responses of OSNs are not only generated by a reduction in the intracellular cation concentration leading to a reduced firing rate as widely assumed, but that they are also carried by an influx of Cl -. It still remains to be investigated how Clchannels are integrated in the olfactory signal transduction machinery of insects.

Multiple roles of Clsignaling at the AL network level
Within the insect AL, odor representations are shaped by the inhibitory network of various types of GABAergic LNs Wilson and Laurent, 2005;Silbering and Galizia, 2007;Hong and Wilson, 2015;Mohamed et al., 2019). It has been shown that OSNs are presynaptically inhibited by LNs, mediated by both GABA A and GABA B receptors (Olsen and Wilson, 2008;Root et al., 2008). Since GABA A receptors are ligandactivated Clchannels, they provide a direct molecular substrate for the Clinflux in OSNs at the AL level. Likewise, PNs express both GABA A and GABA B receptors (Enell et al., 2007), and their odor responses are influenced by both receptor types (Wilson and Laurent, 2005;Silbering and Galizia, 2007). Here we confirm the contribution of GABA A receptors pharmacologically for Clinflux. In addition, our data provide evidence that the synaptic inhibition of PNs is stronger than that of OSNs, since we clearly see an increase in the number of inhibited glomeruli from the input to the output level. However, one has to keep in mind that the sensor dynamics might not reflect the potentially varying dynamics of the membrane potential in these different neuron types. Chloride ions themselves have their own dynamics, and potentially those dynamics reflect actual neuronal dynamics only loosely. Still, our data demonstrates a transformation of odor representations that is not accessible if only excitation-associated Ca 2ϩ is taken into account. Our findings suggest two distinct types of Clsignals in the AL, i.e., a global, moderate inhibition and a strong, cell-typespecific inhibition. This reflects the structural diversity of GABAergic LNs in the Drosophila AL (Chou et al., 2010;Seki et al., 2010;Hong and Wilson, 2015). The majority of LNs arborizes in most glomeruli, and therefore evenly distributes the input from most OSN types. Thus, we would expect that the level of inhibition in each glomerulus should mirror the level of activity in all glomeruli with varying sensitivities to the GABAergic input . This assumption provides a mechanism for global, inhibitory gain control at the cellular and network level to keep the olfactory circuitry in the operating state across odorant combinations and concentrations as shown for the zebrafish olfactory bulb (Zhu et al., 2013).
As a second type of Cl --mediated inhibition, we observed Clresponses that were not linked to any excitation, and most likely reflect local inhibition that specifically shapes neuronal information processing, analogous to the mammalian system (Mori et al., 1999;Urban, 2002). In fact, heterogeneous populations of LNs innervating only few glomeruli also exist (Chou et al., 2010;Seki et al., 2010), which might provide the neuronal substrate for such glomerulus-and odor-specific inhibition. Along that line, recent data provide evidence that patchy, but not global GABAergic LNs accomplish selective lateral inhibition between specific glomeruli processing odors with opposing hedonic valences (Mohamed et al., 2019).

Temporal aspects of odor-evoked chloride responses
The measured odor-induced Cland Ca 2ϩ responses reveal different temporal dynamics. However, the temporal differences between Ca 2ϩ -and Cl --evoked signals are difficult to interpret because it is not clear whether they derive from different reporter dynamics or indeed reflect physiologic properties. Hence when considering temporal dynamics, we restricted any comparison of data obtained to only one reporter protein and therefore compared dynamics of input and output neurons for Cameleon and Clomeleon separately.
Although the chloride influx is clearly odor-induced, it evolves slowly over time and outlasts the odor stimulation period. Such long-lasting chloride responses are consis-continued Clomeleon responses were normalized to highest Clinflux in each animal over all odors before averaging. Black bar indicates the odor application. D, Quantification of Clinflux to ethyl-3-hydroxybutyrate in PNs before, during and after applying of picrotoxin. The GABA A receptor blocker significantly reduces odor-evoked Clresponses ‫‪p‬ءء(‬ Ͻ 0.01, repeated measures ANOVA, n ϭ 9).

Figure 6.
A functional map of odor-evoked inhibition and excitation. A, B, Averaged odor-evoked Cl -(left, in blue) and Ca 2ϩ (right, in red) responses in OSNs (A) and PNs (B) are represented as schematic ALs for 11 odors according to the scales below. Responses were normalized to highest Clor Ca 2ϩ influx in each animal over all odors. Glomerular identities are indicated by AL maps at the top. AC, antennal commissure; AN, antennal nerve; ALT, AL tract. tent with observations in mammalian neurons (Kuner and Augustine, 2000;Berglund et al., 2006) and might reflect the relatively slow rate of Clremoval from the neurons (Staley and Proctor, 1999;Berglund et al., 2009Berglund et al., , 2016. This slow recovery in the Clresponse might affect the excitability of the neuron for a period significantly outlasting the stimulation. However, as mentioned above, the kinetics of fluorescence sensors depend on intrinsic parameters of the sensor itself and firm conclusions about the exact kinetics about the Clcurrents cannot be drawn as yet.
Determining Cland Ca 2؉ representations together provide a more accurate assessment of sensory processing The importance of synaptic inhibition for accurate behavioral responses to olfactory stimuli has been demonstrated in different species. Mice show accelerated discrimination ability when synaptic inhibition of mitral cells is increased by selectively altering granule cell function (Abraham et al., 2010). In locusts and flies, disruptive manipulation of the GABAergic AL network reduces the insects' ability to behaviorally discriminate between similar odors (Stopfer et al., 1997;Barth et al., 2014). The similarity between glomerular excitation patterns evoked by different odors often matches with the animals' ability to discriminate between the odors in behavioral tasks (Sachse and Galizia, 2003;Guerrieri et al., 2005;Niewalda et al., 2011;Barth et al., 2014;Carcaud et al., 2018). In Drosophila, the spatiotemporal, glomerular Ca 2ϩ activity patterns at the PN level reflect more accurately the animals' perception of similarities between odors than the patterns observed at the OSN level (Niewalda et al., 2011). This difference between OSNs and PNs could, at least partly, be due to the influence of GABA-mediated inhibition. Determining Cl --mediated inhibition across ensembles of neurons in addition to Ca 2ϩ -mediated excitation therefore enables us to more comprehensively and more accurately characterize sensory processing underlying the perception of olfactory or other sensory stimuli.