Stereotyped Odor-Evoked Activity in the Mushroom Body of Drosophila Revealed by Green Fluorescent Protein-Based Ca²⁺ Imaging

Introduction

Growing evidence indicates a stereotypical anatomical and functional organization in early stages of the olfactory pathway of insects and vertebrates (Hildebrand and Shepherd, 1997; Buck, 2000;Bozza and Mombaerts, 2001; Kauer and White, 2001; Korsching, 2002). In both insects and vertebrates, each olfactory receptor neuron (ORNs) likely expresses only one specific olfactory receptor (OR) gene (Chess et al., 1994; Ressler et al., 1994; Clyne et al., 1999; Malnic et al., 1999; Vosshall et al., 1999, 2000), and ORNs expressing the same ORs project to spatially invariant glomeruli in the antennal lobe in insects (Gao et al., 2000; Vosshall et al., 2000) and olfactory bulb in vertebrates (Mombaerts et al., 1996). Each odor may activate a combination of ORs and therefore ORNs (Malnic et al., 1999; de Bruyne et al., 2001). The combinatorial activation of ORs and the precise anatomical convergence of ORN projections give rise to topographic activation patterns of glomeruli that are both odor specific and conserved in different individuals (Stocker et al., 1983; Rodrigues, 1988; Johnson et al., 1998; Galizia et al., 1999; Rubin and Katz, 1999; Wachowiak and Cohen, 2001; Wang et al., 2003).

How are neurons in higher olfactory centers organized to respond to odor stimulation? Is the topographic nature of the odor-specific glomerular activation retained in higher-order olfactory centers? Addressing such questions is critical for our understanding of the functional organization of high-order brain centers and olfactory coding. In Drosophila, single-cell tracing has revealed a stereotyped “axon map” of projection neurons. Projection neurons innervating the same glomeurlus in the antennal lobe exhibit a strikingly similar axonal branching pattern in the protocerebrum, whereas those corresponding to different glomeruli differ (Marin et al., 2002; Wong et al., 2002). Furthermore, transneuronal tracer experiments in mice show that ORNs that express the same OR are targeted to specific clusters of thousands of olfactory cortical neurons, and that the locations of these clusters are conserved in different individuals (Zou et al., 2001). These findings suggest that a stereotyped organization may exist in the higher-order olfactory centers reflecting those in the antennal lobe and olfactory bulb.

We previously reported odor-specific spatial patterns of activity in the Drosophila mushroom body (MB) (Wang et al., 2001). However, that study used bulk labeling with Ca²⁺-sensitive dyes and wide-field fluorescence imaging, which do not allow adequate spatial resolution to detect the activity in individual neurons. Furthermore, the variability of dye loading in the living fly brain complicates comparisons of the spatial activity patterns between individual flies. To overcome these limitations, we developed a transgenic fly with targeted expression of a green fluorescence protein (GFP)-based Ca²⁺sensor (G-CaMP) (Nakai et al., 2001) in the MB. Using two-photon imaging (Denk and Svoboda, 1997), we recorded odor-evoked G-CaMP fluorescence transients in the MB with single-cell resolution. Exposure of the fly to odors resulted in a large increase in G-CaMP fluorescence intensity in specific groups of Kenyon cells (KCs), reflecting intracellular Ca²⁺ elevations in these neurons. KCs showing fluorescence transients evoked by a specific odor resided at similar locations of the brain across individual flies.

Materials and Methods

G-CaMP construct and transformation. The insert from the pN1-G-CaMP1.3 plasmid (Nakai et al., 2001) was cut out with BglII and NotI and inserted into the pUAST vector (Brand and Perrimon, 1993). The resulting construct UAS-G-CaMP was injected into 2202u embryos together with pTurbo (encoding transposase) to generate germ line transformants (Spradling and Rubin, 1982). The resulting transformants were crossed to an MB-specific GAL4 driver line, OK107 (Connolly et al., 1996), to examine the expression of G-CaMP. OK107 has the strongest expression in the MB among all GAL4 drivers. Forty-seven transformation lines were shown to have fluorescence, and the one with the strongest expression was chosen for the study reported here.

Optical recording in living flies. Living flies were prepared for MB imaging as described previously (Wang et al., 2001). Briefly, a fly was held in a truncated pipette tip, exposing only the head. A tiny window was cut in the head to reveal the KC soma layer, and a drop of adult fly saline [115 mm NaCl, 5 mm KCl, 6 mm CaCl₂·2H₂O, 1 mm MgCl₂·6H₂O, 4 mm NaHCO₃, 1 mM NaH₂PO₄·1H₂O, 5 mm trehalose, 75 mm sucrose, and 5 mm N-Tris (hydroxymethyl) methyl-2-aminoethanesulfonic acid, pH 7.1, 356 mOsm] was placed immediately over the opening. The exposed brain region was made accessible to the water immersion lens [63×, 0.9 numerical aperture (NA); Olympus America, Melville, NY] by placing a coverslip with a small window matched to the head opening over the head.

Imaging with wide-field fluorescence microscope was as described previously (Wang et al., 2001). Fluorescence images (128 × 128 pixels; pixel size, 1.1 × 1.1 μm) were acquired using a CCD camera (20 frames per second).

For two-photon imaging, a custom-built two-photon laser scanning microscope was used as described previously (Lendvai et al., 2000). The objective was from Zeiss (40×, 0.8 NA; Zeiss, Thornwood, NY). Fluorescence was detected in whole-field detection mode. Microscope control was accomplished using the custom-written software package ScanImage (Pologruto et al., 2003). The light source was a Ti-Sapphire laser (Spectra-Physics, Mountain View, CA) running at a wavelength of ∼910 nm. Images (512 × 512 pixels; pixel size, 0.27 × 0.27 μm) were acquired at a rate of approximately two frames per second.

Odor delivery. Odors were delivered to the fly through a micropipette located ∼2 mm from the antennae by switching from a constant air stream bubbled through mineral oil to an odor stream, which was produced by bubbling air through mineral oil containing an odor. Odor was applied for 2 sec before switching back to the air stream. An air control was performed following the same procedure but with the odor stream replaced by a second air stream. Air controls were interleaved with odor stimulus trials. A specific concentration of an odor was achieved by diluting the pure chemical odorant in mineral oil. A concentration of 10^-3 corresponded to a 1000-fold dilution. We measured the final odor concentrations with a flame ionization detector (MicroFID; Pine Environmental, Cranbury, NJ) and found there was no attenuation of odor concentration from the outlet of the micropipette to the position of antennae. For ethyl acetate, a concentration of 10^-3 was ∼1.2‰ of the saturated vapor, and other concentrations were proportional to the dilutions made. There was a 4 min interval between successive odor stimulations to prevent habituation (Skoulakis and Acevedo, 2003). At such an interval, responses in the MB to odor stimuli were not affected by the preceding odor exposure.

Data analysis. Imaging data were analyzed using a custom program written in MatLab (MathWorks, Natick, MA). ΔF images were obtained by subtracting the mean of the background fluorescence, which was determined by averaging 10 frames before the stimulus, from the average of 3-4 frames at the peak of the response. We searched for regions of the MB showing odor-evoked G-CaMP fluorescence transients that were significantly above the control activity level. Two criteria were used. First, for individual pixels, the fluorescence intensity change had to exceed the background fluorescence fluctuations. Background fluorescence fluctuations were calculated by subtracting the average of 10 frames preceding the stimulus from each of the 10 frames. Then, the mean and SEM of the fluctuation were obtained for each pixel. A threshold was then applied to each pixel of the ΔF image. We found that a threshold corresponding to 3× SEM was optimal: this threshold excluded >80% of the fluorescence signals that were not distinguishable from air controls without exclusion of signals (<5%) as judged by other measures (supplemental Fig. 1 B, available at www.jneurosci.org/cgi/content/full/24/29/6507/DC1). However, not everything appearing in the ΔF image after such thresholding could be considered as odor-evoked activity. Occasionally, switching of the air flow could produce apparent fluorescence changes (supplemental Fig. 1C, available at www.jneurosci.org/cgi/content/full/24/29/6507/DC1). Therefore, a second criterion was established: fluorescence changes (ΔF/F) in regions corresponding to KC somata were scored only if they exceeded (Student's t test; p < 0.001) the air control level (air controls were performed after each odor stimulus trial by measuring fluorescence changes in response to an air stream).

Three-dimensional reconstruction. After recording odor responses, the MB was imaged from top to bottom (z-step = -1 μm), where the peduncle begins to emerge, and a three-dimensional MB structure was constructed with Amira (TGS, San Diego, CA). Neurons with odor-evoked G-CaMP fluorescence transients were mapped onto the MB structure. Wire frames were used to illustrate the reconstructed MBs to make the KCs inside visible to the viewer. The orientation of MBs from different flies were aligned based on the positions of the protrusion emanating from the calyx as well as the orientation of the peduncle and four axon bundles that emerge within the calyx (see Fig. 4A).

For quantitative comparison of the distributions of KCs displaying odor-evoked G-CaMP fluorescence transients across individual flies, each MB was scaled to a template with Amira. The template was an MB with dimensions closest to the average of all MBs being compared. The coordinates of KCs in the scaled MB were obtained with Amira and plotted in SigmaPlot (Systat Software, Richmond, CA). The mean distance between a cluster of KCs and the corresponding centroid was determined using Excel (Microsoft, Seattle, WA).

Results

We used the GAL4 enhancer trap system to target the expression of G-CaMP in MBs (Fig. 1A,B). OK107, a GAL4 line that has expression in most of the MB neurons, was used to drive G-CaMP expression (Connolly et al., 1996). In flies that had both OK107 and G-CaMP transgenes, the basal level of G-CaMP fluorescence allowed clear identification of the MB, which was accessible for imaging from the top of the head. The part of the MB examined consisted of the KC soma layer (30-35 μm thick), containing ∼2500 densely packed KCs per brain hemisphere (Technau and Heisenberg, 1982), and the calyx (∼30 μm thick), which lay below the soma layer. Delivery of an air-borne odor, ethyl acetate (EA), to a living fly elicited G-CaMP fluorescence transients in both KC soma layer and the calyx (Fig. 1C,F), reflecting activity-dependent increases in cytoplasmic Ca²⁺ concentration (Svoboda et al., 1997; Nakai et al., 2001; Wang et al., 2003; Yu et al., 2003). The increase in G-CaMP fluorescence could reach 100%. Odor-evoked fluorescence transients overlapped spatially with individual KC somata (Fig. 1C). Odor-evoked fluorescence transients were highly reproducible (Fig. 1C,F). The time courses of odor-evoked fluorescence transients were similar in different KC somata and different areas of the calyx (Fig. 1D,G) and across trials (Fig. 1C,F). Fluorescence transients reported in this study referred only to those significantly above the control level activity (see Materials and Methods) (Fig. 1E,H; supplemental Fig. 1 D, available at www.jneurosci.org/cgi/content/full/24/29/6507/DC1).

Different odorants evoked fluorescence transients in distinct subpopulations of KCs. Recordings at different depths in the soma layer showed that EA and benzaldehyde evoked fluorescence transients from different KCs at every optical section examined (n = 16 flies). Examples of recordings from three sections in a fly are shown in Figure 2A. The finding in the KC soma layer was consistent with recordings in the calyx, where EA and benzaldehyde induced distinct spatial patterns of fluorescence activity (Fig. 2B). Most benzaldehyde-evoked activity was in the center of the calyx, whereas EA-evoked activity was mostly in the medial and lateral regions.

The specificity of KC fluorescence transients in response to particular odors was further analyzed by recording responses to eight odors in the soma layer at a depth of -20 μm (Fig. 2C,D). Each odor evoked fluorescence transients in distinct groups of KCs. Although EA and benzaldehyde elicited fluorescence transients in several KCs, most other odors caused G-CaMP responses only in one KC, whereas no significant signal was detected for cineole at this location.

We examined fluorescence transients in the MB in response to odors at different concentrations. KCs with odor-evoked fluorescence transients showed diverse degrees of dependency on odor concentration (Fig. 3). In some KCs (23 of 64 EA-responsive KCs recorded), G-CaMP fluorescence transients were elicited by a broad range of odor concentrations that spanned several orders of magnitude (Fig. 3A, white arrowheads point to KCs). The amplitudes of EA-evoked fluorescence transients in these KCs varied to different concentrations of EA (Fig. 3B). In other KCs (41 of 64 EA-responsive KCs recorded), fluorescence transients were only elicited by a narrow concentration range of odors (Fig. 3A, arrows and black-filled arrowheads point to KCs). The diverse concentration dependency was not the result of trial-to-trial variability. Repetitive stimulation with the same concentration series evoked consistent fluorescence transients in the same KCs (Fig. 3A, compare left and right panels).

The diverse concentration dependency was also observed in the KC dendritic region, the calyx. Calycal fluorescence transients were evoked either by broad (10 of 29 responsive areas) or narrow (19 of 29 responsive areas) odor concentration ranges. For example, in some calycal areas (Fig. 3C,D, area 1), fluorescence transients were evoked by EA at all four concentrations tested, spanning from 10^-5 to 10^-2. Meanwhile, there were areas in which fluorescence transients were detected only at a specific concentration (Fig. 3C,D, area 3) or at two different concentrations (Fig. 3C,D, areas 2 and 4).

To quantify the spatial distribution of KCs with odor-evoked fluorescence transients, we reconstructed the positions of the KCs in the MB. Typically, eight responses were recorded per fly and odor in the soma layer from optical sections located 2-4 μm apart. KCs with odor-evoked fluorescence transients were mapped to the corresponding reconstructed MB structure (Fig. 4A). Visual inspection revealed that EA and 3-octanol (OCT) evoked fluorescence transients in distinct sets of KCs that were located in different quadrants of the MB (Fig. 4B,C). KCs with OCT-evoked fluorescence transients were clustered in the central area, whereas KCs with EA-evoked fluorescence transients were localized primarily in two large clusters located laterally and medially.

To compare the distribution patterns of KCs with odor-evoked fluorescence transients across flies, we normalized the MBs to a template using a similar approach for constructing the standard Drosophila brain (Rein et al., 2002) and the coordinates of the KC somata along the dorsoventral, mediolateral, and anteroposterior axes were obtained. It was more evident in the normalized MB that EA and OCT evoked fluorescence transients in distinct sets of KCs, which spatial distributions were similar in different individuals. For example, KCs with fluorescence transients evoked by EA and OCT occupied different regions (Fig. 4D), and KCs with fluorescence transients evoked by the same odor from different flies were distributed in similar locations in the normalized MB (Fig. 4E,F).

The conserved odor-specific distribution patterns were further demonstrated by quantitative comparison of the distributions of KCs with EA- and OCT-evoked fluorescence transients along the mediolateral axis in different flies (Table 1). The normalized MB was evenly divided into six segments along the mediolateral axis, and distributions of KCs with EA- and OCT-evoked fluorescence transients in these segments from different flies were compared. The correlation coefficients were low for inter-odor comparisons (-0.20 ± 0.20, between KCs with EA- and OCT-evoked fluorescence transients) and were high for intra-odor comparisons (0.67 ± 0.19 and 0.84 ± 0.12, for KCs with fluorescence transients evoked by EA and OCT from five flies, respectively). This demonstrated that the overall distribution patterns of the KC somata with fluorescence transients evoked by EA and OCT were both odor specific and conserved across individual flies and therefore stereotyped.

We next asked whether this stereotypy might exist at the single-neuron level. However, the number of KCs involved in G-CaMP fluorescence responses to EA and OCT made it difficult to compare the distribution of single KCs from fly to fly. An ideal case would be if there were only a few neurons involved in response to an odor. Electrophysiological recordings have shown that several odors including cineole and 4-methyl cyclohexanol (MCH) fail to elicit any activity from 16 functional ORNs in the basiconic sensila on the antennae (de Bruyne et al., 2001), raising the possibility that they may activate few KCs. We therefore examined G-CaMP fluorescence in the MB in response to cineole and MCH.

Recordings from optical sections separated by 2-4 μm revealed that very few KCs displayed fluorescence transients in response to both cineole (n = 17 recordings from 3 flies) and MCH (n = 16 recordings from 3 flies). The KCs with odor-evoked fluorescence transients appeared primarily in the top and middle portions of the soma layer for cineole and MCH, respectively. High-resolution measurements (sections 1-2 μm apart) in these subregions of the soma layer were performed to uncover KCs displaying fluorescence transients evoked by cineole and MCH. For cineole, four of six flies examined had three such KCs each, and the other two had four each (Fig. 5A-D). When reconstructed in the normalized MB for all six flies, three of the KCs with cineole-evoked fluorescence transients (Fig. 5A-D, a,b,c) formed three corresponding clusters (Fig. 5E). In each cluster, the mean distance of the KCs from the corresponding centroid was 3.4 ± 1.5, 4.3 ± 2.1, and 3.8 ± 1.8 μm for clusters a, b, and c, respectively. The fourth neuron, shown in two of the six flies (Fig. 5C,D), appeared in variable positions with respect to the other three neurons. For MCH, three flies examined each had two KCs displaying MCH-evoked fluorescence transients, and two others had one each (Fig. 5F-I). These neurons also formed tight clusters when plotted in the normalized MB (Fig. 5J), the mean distances of neurons in each cluster to the corresponding centroid were 3.8 ± 2.0 and 3.9 ± 2.8 μm. These results revealed a striking degree of stereotypy in the distribution of KCs with odor-evoked fluorescence transients at single-neuronal level: a KC displaying fluorescence transients evoked by cineole or MCH may be found in different individuals at a similar position in the MB with a spatial precision on the order of the size of one or two KC somata (∼3.9 μm).

The stereotyped distribution of KCs displaying odor-evoked fluorescence transients is consistent with stereotyped distribution of G-CaMP fluorescence activities in the calyx, revealed by recordings with wide-field fluorescence microscopy. The limited effective spatial resolution of this technique in thick tissues is a minor disadvantage in imaging the calyx, because the distribution of odor-evoked G-CaMP fluorescence activities is relatively diffuse (Figs. 1F, 2). Different odors evoked distribution patterns of fluorescence transients in the calyx that were odor specific and conserved across flies (Fig. 6, Table 2). The basic features of EA- and benzaldehyde-stimulated distribution patterns of G-CaMP fluorescence activity revealed by two-photon imaging were preserved when imaged with wide-field microscopy. EA-induced fluorescence activity was localized medially and laterally, whereas benzaldehyde-induced was distributed in the middle part of the calyx.

Discussion

The targeted expression of the GFP-based Ca²⁺ sensor G-CaMP, in combination with two-photon imaging, has allowed us to examine odor-evoked responses in the MB at the single-neuron level. We observed large fluorescence transients in the calyx and in individual KC somata in response to odor stimulation. Different odors evoked G-CaMP responses in different sets of KCs with diverse dependencies on odor concentrations. The spatial distribution of KCs displaying odor-evoked fluorescence transients (in somata and dendritic region) was stereotyped across flies.

Genetically encoded Ca²⁺ reporters have been used to map odor-evoked activities in the Drosophila antennal lobe (Wang et al., 2003) and to measure Ca²⁺ responses in KCs induced by extracellular application of high potassium or acetylcholine (Yu et al., 2003). In our studies, G-CaMP fluorescence transients in the calyx and somata of KCs likely reflect Ca²⁺ entry through synaptic nicotinic acetylcholine receptors (nAChRs) and voltage-gated Ca²⁺ channels. All projection neurons are cholinergic in Drosophila (Gorczyca and Hall, 1987; Yasuyama et al., 2002) and most intrinsic KCs express nAChRs (Goldberg et al., 1999; Oleskevich, 1999; Su and O'Dowd, 2003) and voltage-gated Ca²⁺ channels (Bicker and Kreissl, 1994; Wicher and Penzlin, 1997). Another possible source is Ca²⁺-induced Ca²⁺ release from intracellular stores.

In this study, we focused on KCs that showed large fluorescence transients that are significantly above the noise level. Under our experimental conditions, this corresponds to a small subset of KCs. Sampling from a population of ∼2800 KCs (four flies, random depths), we counted 58 and 39 KCs with fluorescence transients evoked by EA and BA, respectively. We estimate the overall number of KCs with large fluorescence transients in each brain hemisphere evoked by EA and BA to be 52 and 35, respectively, based on 2500 neurons in each half of the bilateral MB structures. In more extreme cases, cineole and MCH evoked fluorescence transients in only two to four KCs.

It is possible that there are KCs with small responses to odors that may have been missed because of limited signal-to-noise ratios. Furthermore, G-CaMP is a nonlinear sensor of Ca²⁺ concentration (Nakai et al., 2001) and neuronal activity and may not detect Ca²⁺ changes resulting from a single action potential. Using a synthetic Ca²⁺-sensitive dye, Calcium Green (K_Ds for Calcium Green-1 and -2; 190 and 550 nm, respectively), we have reported previously more broadly distributed activity in the KC soma layer (Wang et al., 2001). Such synthetic Ca²⁺ dyes are more sensitive than G-CaMP (T. Pologruto and K. Svoboda, unpublished observations) (Maravall et al., 2000). The large G-CaMP fluorescence transients that we reported here may therefore reveal only a subset of strongly responding units.

However, the sparseness of KCs with odor-evoked G-CaMP fluorescence transients correlates well with the sparseness of KC spiking responses to odor stimulation in the locust MB (Perez-Orive et al., 2002), where on average only 0.7% of KCs respond to an odor stimulation, and each responding KC fires 2.3 spikes on average (Perez-Orive et al., 2002). This correlation suggests that the fluorescence transients in sparse populations of KCs that we observed might represent odor-evoked suprathreshold KC spiking. Furthermore, intracellular recordings in the locust KCs have revealed odor-evoked EPSPs. KCs act as coincidence detectors on projection neuron input, and the conditions for appropriate EPSP summation that leads to KC firing are rarely met. We speculate that weak fluorescence signals that are possibly missed from our analysis might represent these subthreshold activities. However, the correlation between the G-CaMP fluorescence transients and KC electrical activities can only be established by combining imaging with electrophysiological recordings.

G-CaMP fluorescence transients in different KCs showed diverse degrees of dependency on odor concentration. In some KCs, fluorescence transients were evoked by odors at concentrations spanning several orders of magnitude, whereas in others only by specific concentrations or narrow ranges of concentrations. This finding is consistent with a recent study by Stopfer et al. (2003) in locust, in which they have examined intensity coding in projection neurons and KCs. They found that KCs respond to odors with diverse degrees of concentration invariance, with some units responding to broad ranges of concentration of an odor, whereas others responded more selectively to specific concentrations. The tuning of KCs to odor concentrations (i.e., selectivity of KC response to different concentrations of an odor) is interpreted as being a result of specific connections between projection neurons and KCs, oscillatory dynamics, intrinsic properties of KCs, and feed-forward inhibition (Perez-Orive et al., 2002; Stopfer et al., 2003).

For odors from different chemical classes such as EA, benzaldehyde, and OCT, there appears to be little overlap between corresponding KC populations with G-CaMP fluorescence transients. This may reflect the fact that these odors activate nonoverlapping sets of ORNs (de Bruyne et al., 2001). Recordings on other odors of dissimilar chemical structures also showed no overlapping KC populations with G-CaMP fluorescence transients. This seems to be different from olfactory cortical neurons in the pyriform cortex of mammals, which have been shown to be less selective to odors (Nemitz and Goldberg, 1983; Schoenbaum and Eichenbaum, 1995). However, different odors were used in these mammalian studies than those used here, complicating direct comparisons between these studies.

The locations of KCs with fluorescence transients evoked by a given odor in MBs are strikingly stereotyped across individuals. In the cases of cineole and MCH, where G-CaMP fluorescence transients were evoked in only a few KCs, individual responsive KCs could be located at similar positions of the MB in different flies with a spatial precision on the order of the size of one or two KC somata. This implies that there may be a genetically determined precise connection between KCs and projection neurons as that found between ORNs and projection neurons (Jefferis et al., 2001), despite the lack of stereotypy of projection neuron arborization patterns in the MB (Marin et al., 2002; Wong et al., 2002). A systematic examination of the spatial distribution of KC activities in relation to projection neuron activities will reveal the functional connections between projection neurons and KCs.