The Drosophila antennal lobe (AL) has become an excellent model for studying early olfactory processing mechanisms. Local interneurons (LNs) connect a large number of glomeruli and are ideally positioned to increase computational capabilities of odor information processing in the AL. Although the neural circuit of the Drosophila AL has been intensively studied at both the input and the output level, the internal circuit is not yet well understood. An unambiguous characterization of LNs is essential to remedy this lack of knowledge. We used whole cell patch-clamp recordings and characterized four classes of LNs in detail using electrophysiological and morphological properties at the single neuron level. Each class of LN displayed unique characteristics in intrinsic electrophysiological properties, showing differences in firing patterns, degree of spike adaptation, and amplitude of spike afterhyperpolarization. Notably, one class of LNs had characteristic burst firing properties, whereas the others were tonically active. Morphologically, neurons from three classes innervated almost all glomeruli, while LNs from one class innervated a specific subpopulation of glomeruli. Three-dimensional reconstruction analyses revealed general characteristics of LN morphology and further differences in dendritic density and distribution within specific glomeruli between the different classes of LNs. Additionally, we found that LNs labeled by a specific enhancer trap line (GAL4-Krasavietz), which had previously been reported as cholinergic LNs, were mostly GABAergic. The current study provides a systematic characterization of olfactory LNs in Drosophila and demonstrates that a variety of inhibitory LNs, characterized by class-specific electrophysiological and morphological properties, construct the neural circuit of the AL.
The olfactory system has to extract information on a very large number of chemically highly diverse molecules and has to neuronally encode the identity and intensity of these molecules over a range of different odor concentrations. The selection pressures exerted by these demands have led to the evolution of a similar architecture in the primary olfactory centers of most animals across diverse phyla (Bargmann 2006; Hildebrand and Shepherd 1997; Su et al. 2009).
The insect antennal lobe (AL), analogous to the vertebrate olfactory bulb, is an important model for studying early olfactory processing. Especially the AL in the fruit fly, Drosophila melanogaster, with all its possibilities for genetic introduction of morphological markers and activity-dependent sensors, is at present a supreme model for olfactory neuroscientists (for review, see Olsen and Wilson 2008a; Vosshall and Stocker 2007). The AL of adult flies receives input from ∼1,300 olfactory sensory neurons (OSNs) housed in three types of sensilla (trichoid, basiconic, and coeloconic) (Couto et al. 2005; Vosshall and Stocker 2007). The OSNs are divided into ∼50 functional classes, dependent on the expression of different olfactory receptor proteins (Benton et al. 2009; Couto et al. 2005; Fishilevich and Vosshall 2005). All OSNs expressing a specific receptor protein converge into 1 of 50 olfactory glomeruli present in the AL (Fig. 1A). From the AL, the neural message is transmitted via ∼150 projection neurons (PNs), usually innervating a single glomerulus with dendritic arbors and sending an axon to higher brain areas such as the mushroom body and the lateral horn (Fig. 1A) (Marin et al. 2002; Stocker et al. 1990; Wong et al. 2002). The AL also contains ∼200 local interneurons (LNs), which connect many glomeruli and form a network providing transfer of information between glomeruli (Chou et al. 2010; Stocker 1994). Most LNs are inhibitory, γ-aminobutyric acid (GABA)ergic neurons (Ng et al. 2002; Okada et al. 2009; Wilson and Laurent 2005), but a small population of LNs was recently found to be cholinergic (Shang et al. 2007).
Although earlier studies reported a few variations of LNs (Christensen et al. 1993; Flanagan and Mercer 1989; Fonta et al. 1993; Matsumoto and Hildebrand 1981), until recently the heterogeneity of LNs had been overlooked in many insects. Recent studies revealed morphological diversity among LNs in a moth (Seki and Kanzaki 2008) and differences of intrinsic electrophysiological properties in the cockroach (Husch et al. 2009a,b), providing new insights into the variability of LNs. In Drosophila, facilitated by the use of LN-specific GAL4 lines, evidence regarding the diversity of LNs has gradually emerged, such as morphological diversity revealed by molecular labeling methods (Das et al. 2008; Lai et al. 2008; Okada et al. 2009) and diverse odor response profiles shown by optical recording techniques (Silbering et al. 2008). Notably, a recent study revealed functional causality between the generation of odor-elicited oscillation and a specific subpopulation of LNs (Tanaka et al. 2009). Furthermore, a highly comprehensive study revealing various morphological classes of LNs combined with neurotransmitter profiles, connectivity and physiological properties was recently published (Chou et al. 2010; see discussion). However, no studies have so far addressed the diversity of intrinsic electrophysiological LN properties combined with detailed morphological analysis of single neurons.
Here we present a detailed characterization of individual LNs in D. melanogaster. To determine the intrinsic physiological properties of the LNs, we used an isolated brain preparation. Using patch-clamp recording in isolated brain preparations followed by staining of single LNs, and three-dimensional (3D) reconstruction from confocal image stacks, physiological properties were paired with detailed morphological characteristics to augment our understanding of the role played by LNs in primary odor information processing.
Fly stocks and materials
All fly stocks were maintained on conventional cornmeal agar medium under a 12 h light: 12 h dark cycle at 25°C. All experimental animals were adult females, 2–12 days after eclosion. In choline acetyltransferase (ChAT) immunostaining, adult females 1–2 days old were used for technical reasons (see Immunocytochemistry). The GAL4-Krasavietz (Shang et al. 2007), GAL4-NP1227, GAL4-NP2426 lines (Okada et al. 2009; Sachse et al. 2007) were used to express GAL4 in specific LNs. UAS-GCaMP1.3, a Ca2+ sensitive fluorescent sensor (Nakai et al. 2001; Wang et al. 2003), was used to identify the GAL4-expressing neurons. A previous study showed that GCaMP1.3 does not alter the responses of the neurons that express it (Jayaraman and Laurent 2007). UAS-mCD8-GFP was used under control of GAL4-Krasavietz for some preparations for immunocytochemistry. All chemicals, unless stated otherwise, were provided by Sigma (Deisenhofen, Germany).
Isolated brain preparation
Flies were anesthetized on ice and pinned on a dissecting chamber. The brain was isolated by removal of the head capsule, muscles, trachea, the proboscis, antennae and compound eyes. The connective tissue sheath (neurolemma) surrounding the AL was carefully removed using fine forceps. The brain (Fig. 1B) was transferred to a glass slip, stuck on the surface of the glass and stored in a bath solution containing (in mM) 130 NaCl, 5 KCl, 2 MgCl2, 2 CaCl2, 36 sucrose, and 5 HEPES; pH = 7.3.
Electrophysiology and staining
Recordings were obtained from cell bodies under visual control using a BX51WI microscope (Olympus, Hamburg, Germany) with infrared-differential interference contrast optics and a ×40 water-immersion objective (LUMPlanFl, NA: 0.8, Olympus). A CCD camera (IMAGO VGA, TILL Photonics, Gräfelfing, Germany) was used to monitor the recording region. Patch-clamp electrodes (6–11 MΩ) were filled with internal solution containing (in mM) 140 potassium aspartate, 10 HEPES, 1 KCl, 4 Mg-ATP, 0.5 Na3GTP, 1 EGTA, and 1 Lucifer yellow CH; pH = 7.3; cf. (Wilson and Laurent 2005). Ten neurons from the GAL4-NP1227 line were recorded with the internal solution without Lucifer yellow CH. Measurements were performed in the whole cell configuration under current-clamp conditions using an EPC 10 patch-clamp amplifier (HEKA Elektronik, Lambrecht, Germany) controlled by the Patch-Master software. Voltage recordings were sampled at 10 kHz. After rupturing the patch, the membrane potential was adjusted to approximately −50 mV by appropriate current injection. Stimulating pulses were applied for 3 s with an 11 s interstimulus interval. The 11 s interstimulus interval was adopted to assure the recovery time from hyperpolarization following trains of spikes (Fig. 1E). Current steps ranged from 3 to 20 pA to induce ∼5–10 mV step changes of baseline membrane potential, depending on the input resistance of the neurons. The spike parameters given in Table 1 were determined for the first spike above the threshold. To plot the frequency increase against the membrane potential increase (e.g., Fig. 2A4), all data for the same class recorded with ∼5–10 mV step changes were interpolated to calculate mean frequencies at every 5 mV step ranging from −50 to −15 mV. Recording data were analyzed in Igor Pro (Wavemetrics, Lake Oswego, OR) using custom software. After recording, Lucifer yellow was injected by passing a hyperpolarizing current (−1 nA) for 10–60 min. The state of staining was monitored under the microscope. The current injection time was thus adjusted to assure that the neuron was completely stained. Morphological structures were subsequently analyzed from confocal image stacks. For some preparations, 3D reconstruction techniques were applied to analyze morphological characteristics in greater detail. All recordings were started within 20 min after dissecting out the brain. One neuron per brain was recorded and stained.
To visualize glomerular structures, nc82 immunostainings were performed. Brains with a Lucifer yellow-injected cell were fixed in 4% paraformaldehyde in phosphate-buffered solution (PB; 0.1M, pH 7.4) for 30 min on ice. Then the brains were washed with phosphate-buffered saline solution (PBS; 0.1 M, pH 7.4) containing 0.2% Triton X-100 (PBST) for 60 min (3 times for 20 min) at room temperature (RT). After being blocked in 5% normal goat serum (NGS) in PBST (PBST-NGS) for 60 min at RT, the brains were incubated in 1:30 mouse monoclonal nc82 antibody [the Developmental Studies Hybridoma Bank (DSHB)] in PBST-NGS for 2 days at 4°C. Then the brains were washed for 60 min at RT and incubated in 1:200 goat anti-mouse Alexa Fluor 633 (A21052, Invitrogen) for 2 days at 4°C. Afterwards, brains were washed again for 60 min and mounted in Vectashield (Vector, Burlingame, CA).
To identify GABAergic and cholinergic neurons in the specific GAL4 lines, GABA or ChAT immunostaining was performed, either combined with or without GFP immunostaining. As reported previously, staining of cell bodies was inconsistent with ChAT immunostaining (Yasuyama et al. 1995). We found that cell bodies were only consistently labeled when we used young flies, thus only adult female flies of 1–2 days old were used for ChAT immunostaining. Brains were dissected out and fixed in 4% paraformaldehyde in PB for 30 min on ice. Then the brains were washed with PBST for 60 min at RT. After blocking with PBST-NGS for 60 min at RT, the brains were incubated in 1:200 mouse monoclonal anti-Drosophila ChAT antibody (4B1, DSHB) with 1:1,000 rabbit anti-GFP antibody (A11122, Invitrogen) or 1:500 rabbit GABA antibody (A2052, Sigma) with 1:500 mouse anti-GFP antibody (A11120, Invitrogen) in PBST-NGS for 2–3 days at 4°C. Then the brains were washed for 60 min at RT and incubated in 1:200 goat anti-mouse Alexa Fluor 633 with 1:500 goat anti-rabbit Alexa Fluor 488 (A11008) for ChAT and GFP immunostaining, respectively. For GABA and GFP immunostaining, brains were incubated with 1:500 goat anti-rabbit Alexa Fluor 546 (A11010, Invitrogen) and 1:500 goat anti-mouse Alexa Fluor 488 (A11001, Invitrogen) in PBST-NGS for 2–3 days at 4°C. Afterward brains were washed again for 60 min and mounted in Vectashield. To estimate the percentage of ChAT and GABA-positive LNs, double-labeled somata located in the antero-dorsal and lateral cell cluster of the AL were counted. Some preparations were doubly stained with anti-ChAT (with Alexa 633 for secondary antibody) and anti-GABA (with Alexa 546 for secondary antibody), or singly stained with either of them without anti-GFP staining, to compare ChAT immunopositive and GABA immunopositive cells with cells having GCaMP expression.
Images were taken with a Zeiss LSM510 META confocal microscope (Carl Zeiss, Jena, Germany) using a ×63 water immersion objective (C-Apochromat, NA: 1.2, Carl Zeiss) for brains with single neuron staining and ×40 water immersion objective (C-Apochromat, NA: 1.2, Carl Zeiss) for brains with GABA and ChAT immunostaining. Images were obtained at 0.45–46 μm intervals for ×63 and at 0.49–0.55 μm intervals for ×40 at 1,024 × 1,024 pixel resolution. Images from the right AL were mirror imaged to match those from the left AL. Confocal images were adjusted for contrast and brightness by using LSM Image Browser 4.0 (Carl Zeiss) and Adobe Photoshop CS.
Glomerular innervation map
The dendritic arborizations within the 50 glomeruli identified were mapped to analyze the global innervation pattern, using confocal stack images (Table 2). The glomerular identities and sensilla types are adopted from (Couto et al. 2005; Fishilevich and Vosshall 2005). Because the glomeruli VP1, VP2, VP3, located in the posterior ventral part of the AL (Stocker et al. 1983), could not be labeled with the monoclonal antibody nc82 (Laissue et al. 1999), they were excluded from the analysis in this study.
3D reconstruction and morphological data analysis
The complex dendritic branches of LNs were reconstructed by using the 3D semi-automatic reconstruction module compiled in the Amira 4.1.1 software (Visage Imaging, Berlin) (Evers et al. 2005; Schmitt et al. 2004). This module allowed the tracing process and created vertex models in which neurites were approximated by cylinders of a particular diameter. Glomerular neuropil were segmented semi-automatically using the LabelField Editor in Amira. Metric measurements were made using the statistics module in Amira. For the analysis of fully reconstructed LNs, the origin of the neurons was defined as the tip end of the primary neurite at the point of entry into the AL. The neural distance for each glomerulus from the origin was calculated by averaging neural distances from the origin to all the vertexes in the glomerulus and then normalized by the distance of the farthest glomerulus. The air distance for each glomerulus from the origin was calculated by averaging linear distances from the origin to all the vertexes in the glomerulus and then normalized by the distance of the farthest glomerulus. We display the comparison of the neural distance and the air distance as an indicator for how the neural path is different from the straight line between the origin and each glomerulus (e.g., Fig. 4F). The proximal terminal glomerular groups (Fig. 4) were defined by setting a variable threshold at a certain distance from the origin at each major half branch of each fully reconstructed neuron analyzed. Downstream branches were divided into between five and six (for Krasavietz class_2, between 4 and 5, due to its smaller number of innervated glomeruli) separated groups. This resulted in a division of a total of between 10 and 12 groups (between 8 and 10 groups for Krasavietz_class2), so-called proximal terminal glomerular groups.
Hierarchical cluster analysis was used to classify the neurons by combining a subset of the variables described and defined in the legend of Table 1 (cf. Nowak et al. 2003). All variables were normalized to their z-scores. Five variables, namely threshold, spike amplitude, afterhyperpolarization amplitude, spike adaptation rate, and spike amplitude accommodation rate, were used for the cluster analysis (Ward's method with Euclidean distance). Two variables, R-input and spike half-width, were not used for the clustering because these parameters did not differ between the different classes of LNs (Table 1). The cluster analysis was performed using PAST (http://folk.uio.no/ohammer/past/).
As LNs connect and modulate OSN input to PN output, we first investigated the response of a given LN to a highly defined stimulus. We thus used an isolated brain preparation, thereby eliminating peripheral input (Fig. 1B). In whole cell current-clamp recordings from cell bodies of genetically labeled LN-specific GAL4 lines, cells were held at a subthreshold resting level adjusted to −50 mV and were stimulated by depolarizing current injection (Fig. 1, C–E). The characteristic parameters of the LN response are given in Table 1. To characterize different types of LNs, we used three LN-specific enhancer trap GAL4 lines that contain moderate numbers of labeled LNs. The neurotransmitters expressed by these LNs have been reported in previous studies: 1) GAL4-Krasavietz consisting of ∼15 LNs [reported as cholinergic (Shang et al. 2007), but see following text]; 2) GAL4-NP1227 (also named GAL4-LN1) consisting of ∼18 LNs reported as GABAergic; and 3) GAL4-NP2426 (also named GAL4-LN2) consisting of ∼37 LNs reported as GABAergic (Okada et al. 2009; Sachse et al. 2007). We successfully recorded 45 LNs from the GAL4 lines studied. Initially we provided a detailed description of four classes of LNs characterized by specific electrophysiological properties and global innervation patterns. Second, we examined common characteristics of LN morphology and differences in branching density and distribution within specific glomeruli between the different physiological classes of LNs. In addition, we performed immunocytochemistry to investigate which neurotransmitter is expressed in LNs of each GAL4 line in comparison with previous studies.
Physiology and morphology of GAL4-Krasavietz LNs
Neurons recorded from the GAL4-Krasavietz line (n = 11) were classified into two different classes, called Krasavietz_class1 and Krasavietz_class2. Physiologically these two classes showed similar tonic firing patterns in response to injection of depolarizing current but differed in action potential shape and spike accommodation properties. Morphologically the classes could be separated based on their global innervation patterns in the AL.
Five neurons were characterized as Krasavietz_class1 LNs. The firing threshold of these neurons was −39 mV (Table 1) and they were tonically active during the current pulse (Fig. 2A, 1 and 2, and Supplemental Fig. S1A1). The firing frequency rose linearly with increasing current injection and adapted moderately during the current pulse (Fig. 2A, 1, 2, and 4, and Supplemental Fig. S1A). The neurons revealed a small afterhyperpolarization amplitude (Fig. 2A, 1 and 3, and Supplemental Fig. S1A, Table 1), and showed an initial spike amplitude accommodation (Fig. 2A1 and Supplemental Fig. S1A, Table 1). As regards morphological aspects, the neurons extended their primary neurites into the central core region of the AL and then branched radially into the peripheral glomeruli (Fig. 3A; see following text). Neurons of this class innervated all glomeruli in the AL (99.2 ± 1.1% of all identified glomeruli; mean ± SD, n = 5; Fig. 3A, Table 2,).
Six neurons were characterized as Krasavietz_class2 LNs. These neurons started firing above a threshold of −28 mV (Table 1) and were tonically active during the current pulse (Fig. 2B, 1 and 2, and Supplemental Fig. S1B). The firing frequency rose linearly with increasing current injection and adapted moderately during the current pulse (Fig. 2B, 1, 2, and 4, and Supplemental Fig. S1B). No spike accommodation was visible and the spike amplitude remained constant during stimulation (Fig. 2B1 and Supplemental Fig. S1B, Table 1). These neurons typically displayed a significantly larger afterhyperpolarization amplitude than neurons of the Krasavietz_class1 LNs. (Fig. 2B, 1 and 3, Supplemental Fig. S1B, Table 1). The Krasavietz_class2 LNs arborized in only 70% of the AL glomeruli and lacked innervation in some ventral and dorsal glomeruli (70.0 ± 1.8%, n = 6; Fig. 3B, Table 2). Some neurons (4 of 6) extended small branches into the DM1 (also DM4 in one case) glomerulus of the contralateral AL (Table 2).
Physiology and morphology of GAL4-NP1227 (= LN1) LNs
All neurons (n = 18) recorded from the GAL4-NP1227 line displayed a burst firing pattern. This class was named NP1227_class1. Among them eight neurons were morphologically characterized revealing an extensive innervation pattern among the AL glomeruli.
The burst firing of NP1227_class1 LNs was induced at a threshold level of −39 mV (Fig. 2C1 and Supplemental Fig. S1C, Table 1). The spike number per burst was 7.4 ± 3.1 (n = 16) with a mean frequency of 43.7 ± 22.3 Hz (n = 16). The burst firing pattern was gradually transformed into a tonic pattern with increasing stimulus intensity (Fig. 2C, 1 and 2, and Supplemental Fig. S1C). The firing frequency rose abruptly even at low intensity of the electrical stimulus (Fig. 2C4). A slow depolarizing potential, which lasted a few hundred milliseconds, typically underlay several spikes in a bursting cycle (Fig. 2C3). The neurons innervated almost all glomeruli (93.5 ± 3.5%, n = 8; Fig. 3C, Table 2).
Physiology and morphology of GAL4-NP2426 ( = LN2) LNs
We morphologically and physiologically characterized 16 neurons from the GAL4-NP2426 line. One class was reliably characterized by its physiological and morphological properties and displayed a tonic firing pattern with fast spike adaptation and extensive arborization among the AL glomeruli. This class was named NP2426_class1. The other neurons could not be unambiguously characterized but are tentatively categorized into two subgroups.
Six neurons were characterized as NP2426_class1 LNs. These neurons displayed tonic activity with fast spike adaptation and terminated spiking after a few action potentials (Fig. 2D, 1 and 2, and Supplemental Fig. S1D). A remarkably high-threshold of −22 mV had to be passed to elicit a response (Table 1). These neurons could be forced to fire throughout the current pulse at ∼10 mV above threshold (Fig. 2D, 1 and 2, and Supplemental Fig. S1D). The rise of the spike frequency was linear (Fig. 2D4). Neurons of this class innervated almost all glomeruli (97.3 ± 3.0%, n = 6; Fig. 3D, Table 2).
OTHER VARIATIONS IN NP2426.
Ten neurons were recorded and stained but could not be characterized as a specific class. These neurons were divided into two groups. One group (group 1; n = 7) contained inactive neurons that showed only one spike or irregular spikes and had a high-threshold (Supplemental Fig. S2, B-G). These neurons innervated almost all glomeruli sparsely, which could also be observed for the NP2426_class1 LNs (Supplemental Fig. S2, A–C, E, and F). The other group (group 2; n = 3) revealed completely different morphological characteristics compared with the NP2426_class1 LNs and showed dense arborizations in almost all glomeruli (Supplemental Fig. S2, A and H–K). The activity pattern of these neurons was not uniform; one neuron displayed tonic firing and the other two neurons showed no spike firing (data not shown). Further analysis was not applied to these two groups.
Classification of different physiological classes
We describe physiological characteristics of each class of LNs and use statistical tests to compare specific properties (Table 1). Here we further evaluate how distinct these physiological classifications are.
There is a clear distinction between bursting (NP1227_class1) and nonbursting (Krasavietz_class1, Krasavietz_class2, and NP2426_class1) LNs (Fig. 2 and Supplemental Fig. S1). NP1227_class1 LNs always showed rapid spike firing with several spikes underlain by a slow depolarizing potential just above threshold (Fig. 2C and Supplemental Fig. S1C). To determine whether the three classes of nonbursting LNs could be distinctly segregated by the combination of physiological variables shown in Table 1, we used a cluster analysis. We chose five distinguishing parameters (threshold, spike amplitude, afterhyperpolarization amplitude, spike adaptation rate, and spike amplitude accommodation rate) for the clustering among the variables shown in Table 1. The hierarchical tree reveals that the three classes of nonbursting LNs indeed cluster separately (Supplemental Fig. S3). These results suggest that the three classes can be segregated by the combination of the five parameters, each of which represents a part of class-specific electrophysiological differences (Table 1).
Common morphological characteristics
Here we describe morphological features common to all four LN classes, revealed by full reconstruction analysis. We generated a full reconstruction of eight neurons (2 neurons of each class) to analyze how the glomeruli were wired by LN neurites. First, we hypothesized two simple wiring models of how LN neurites might connect glomeruli (schematic diagrams are shown in Fig. 4, A and B): a straight branching pattern (A) or a radial branching pattern (B). In the straight branching pattern, the neuron begins innervating proximal glomeruli from the point of entry (defined as “the origin”) into the AL and extends toward distant glomeruli (Fig. 4A, left). Therefore the neural distances to each glomerulus from the origin are variable in proportion to the linear distance from the origin. (Fig. 4A, right). In the radial branching pattern, the primary neurite branches several times at the center core of the AL and extends into peripheral glomeruli (Fig. 4B, left). Therefore the neural distances to each glomerulus from the origin are similar and biased to the terminal ends (Fig. 4B, right). We generated dendrograms of fully reconstructed neurons and compared them to these two models. Four neurons (1 neuron of each class) are shown in Fig. 4, E, G, I, and K. The arborizations of each glomerulus were labeled with different colors and the origin of the neurons was defined as the tip end of the primary neurite at the point of entry into the AL. The dendrograms revealed that the arborizations in each glomerulus were distributed in a parallel rather than in a serial manner (Fig. 4, E, G, I, and K), hence favoring the radial branching pattern. Furthermore, we calculated the neural distances to each glomerulus from the origin and normalized it by the maximum distance (i.e., the distance to the farthest glomerulus). The mean normalized distance for all glomeruli was 0.86 ± 0.06 for the Krasavietz_class1 LN (Fig. 4F), 0.74 ± 0.10 for the Krasavietz_class2 LN (H), 0.76 ± 0.10 for the NP1227_class1 LN (J), and 0.83 ± 0.09 for the NP2426_class1 LN (L). The mean distance of the eight fully reconstructed neurons was 0.81 ± 0.04 (n = 8). These results demonstrate that the distances to each glomerulus from the origin are biased toward the terminal branches (Fig. 4, F, H, J, and L; note the difference from the widely distributed normalized air distance). This feature matches well with the radial branching model rather than with the straight branching model.
Next, we asked how the neurites branch after the first branching point of the primary neurite. If subsequent branches emanate from the same location on the primary neurite, the distances of any combination of two glomeruli via LN neurites should be similar regardless of their locations within the AL (Fig. 4B). However, the dendrograms of all LN classes revealed that the neurites branched several times downstream of the first branching point, forming a hierarchical pattern (Fig. 4, E, G, I, and K; cf. schematic diagram shown in C.) In this hierarchical pattern, some glomeruli revealed shorter connections via LN neurites than others.
Finally, we asked whether glomeruli that are connected via short distance LN neurites belong to any functional and/or stereotypical subgroup. To define such a glomerular subgroup, we used appropriate criteria (see methods) to divide each neuron into 8–12 parts and identified the glomeruli in each subdivision (Fig. 4, E, G, I, and K). We did not find any stereotypical combination of glomeruli common to different classes (Fig. 4, E, G, I, and K) nor within the same LN class (data not shown). However, we found that a relatively large number of glomeruli could belong to more than two subdivisions (emphasized with black and red asterisks in Fig. 4, E, G, I, and K). This demonstrates that a certain number of glomeruli are innervated more than twice by different branches deriving from different glomerular subgroups, whose terminal glomeruli are not distinct but partly overlapping (cf. schematic diagram shown in Fig. 4D). In summary, our results revealed that all four classes of LNs followed a radial branching pattern with a hierarchical structure.
Density and distribution within a glomerulus
Considering the global innervation pattern within the AL, the four classes of LNs can be divided into two main types: Krasavietz_class2 LNs innervate a specific subpopulation of glomeruli, while the other three classes (Krasavietz_class1, NP1227_class1, and NP2426_class1) innervate almost all glomeruli (Table 2). To further examine morphological differences between the different classes of LNs, we compared the detailed dendritic density and distribution within six specific glomeruli (DM2, DL1, DA1, VA1d, VA1m+l, and V) including four or five neurons for each class (Fig. 5). First, we measured the neural volume per glomerular volume to evaluate the density of innervation for each glomerulus type. Krasavietz_class1 LNs showed denser branches than NP1227_class1 and NP2426_class1 LNs in all glomeruli analyzed with a significant difference for NP1227_class1 LNs in four glomeruli (VA1d, DM2, V, and DL1) and for NP2426_class1 LNs in two glomeruli (VA1d and V; Fig. 5E). Also Krasavietz_class2 LNs showed a significantly denser branches than NP1227_class1 and NP2426_class1 LNs in the three glomeruli DA1, VA1d, and DM2 (Fig. 5E). Interestingly, the difference between Krasavietz_class1 and Krasavietz_class2 LNs was glomerulus-specific. Krasavietz_class2 LNs showed for example a denser innervation pattern in DM2 than Krasavietz_class1 LNs, whereas DL1 was more densely innervated by Krasavietz_class1 LNs compared with Krasavietz_class2 LNs (Fig. 5E).
Second, we compared the spatial distribution of the dendritic branches within a glomerulus by performing a detailed analysis using the DM2 glomerulus. Taking advantage of 3D reconstruction analysis, we checked distribution patterns of dendrites within this glomerulus carefully and found that Krasavietz_class2 LNs innervated the whole volume of the glomerulus homogeneously (Fig. 5G), whereas the other three types, Krasavietz_class1, NP1227_class1, and NP2426_class1 LNs innervated only parts of it (Fig. 5F, H, I). This difference could also be observed in the two other glomeruli DA1 and VA1d (data not shown). In summary, Krasavietz_class2 LNs revealed a distinct innervation pattern within a glomerulus compared with the other three classes. Both Krasavietz class LNs showed denser branches than the NP1227_class1 and NP2426_class2 LNs (Fig. 5E).
Putative neurotransmitter types of the different LN classes
To confirm the neurotransmitter of the four classes of LNs, we revisited the information published in previous studies by using ChAT and GABA immunocytochemistry. As reported previously, ∼95% of each population of LNs belonging to GAL4-NP1227 and GAL4-NP2426 showed GABA immunoreactivity (Okada et al. 2009; Tanaka et al. 2009) (data not shown). For the GAL4-Krasavietz line, our results somewhat contradicted the results of previous studies. Shang et al. (2007) reported that 63.1% of these neurons were ChAT-positive, whereas 39.4% were GABA-positive using GFP immunostaining detection of Krasavietz neurons. In addition, using expression of GFP and flies with a direct fusion of ChAT-DsRed, Shang et al. (2007) showed that 86.7% of LNs labeled by the Krasavietz line were cholinergic. Our immunostaining data showed that 76.6 ± 6.0% (n = 8) of LNs marked by the Krasavietz line were GABA-positive, whereas only 6.9 ± 10.1% (n = 5) were ChAT-positive using GFP immunostaining detection of Krasavietz neurons (Supplemental Fig. S4, A–H). Immunolabeling of GABA and/or ChAT with expression of GCaMP revealed that 94.5 ± 11.4% (n = 9) were GABA-positive and only 0.8 ± 1.6% (n = 9) were ChAT-positive (Supplemental Fig. S4, I–M). A higher ratio of cells showed ChAT immunoreactivity with GFP immunostaining. This might be due to the amplification effect of GFP immunostaining that can detect neurons with weak labeling. In our patch-clamp experiments, we selected target cells from cells that clearly expressed GCaMP, therefore Krasavietz_class1 and Krasavietz_class2 LNs are most probably GABAergic LNs. There might be a possibility that other neurotransmitters or neuropeptides colocalize with GABA in these LNs but we have not addressed this issue here.
We performed a systematic characterization of LNs in the Drosophila AL using three LN-specific GAL4 enhancer trap lines. Four classes of LNs were characterized based on class-specific electrophysiological and morphological properties. These four classes of LNs are very likely all GABAergic. Differences in firing patterns and morphology between the four classes of LNs may lead to different functional roles in the neural circuit of the AL, suggesting the presence of complex inhibitory networks.
An isolated brain
In the present study, we chose to work with an isolated brain preparation, thereby removing all input to the AL system. This allowed us to characterize the LNs under study in “splendid isolation” and to determine their electrical properties without bias from the spontaneous activity of OSNs. On the other hand, we could not perform any odor stimulation. Future studies will, of course, have to include natural stimuli, but we see the present investigation as an important stepping stone toward a deeper understanding of the function and role of LNs involved in olfactory coding and processing.
Characterization of neurons using LN-specific GAL4 lines
Characterization of neurons is a prerequisite for deciphering the function of neural circuits. Inhibitory LNs in particular are generally prevalent in broad neural circuits but difficult to characterize reliably due to their wide diversity (Ascoli et al. 2008). The insect AL contains a variety of inhibitory LNs that constitute the local circuit (Seki and Kanzaki 2008). We present a characterization of physiological, morphological and neurochemical properties that are matched in each class of LNs. In addition, we take advantage of the GAL4-UAS system in Drosophila. Development of the variety of GAL4 enhancer trap lines in which a small group of neurons are under the control of UAS-genes facilitates investigations of the function of a specific neuronal group. However, this system does not guarantee that the group marked in a specific line is a functionally homogeneous neural population.
Three specific LN GAL4 lines were selected in the present study for the following reasons; these three lines express GAL4 exclusively in LNs in the AL (Okada et al. 2009; Shang et al. 2007), consist of moderate number of LNs, and have some relevant functions demonstrated in previous studies (Sachse et al. 2007; Shang et al. 2007; Tanaka et al. 2009). Detailed characterizations of LNs in these three lines are thus potentially useful to provide a link to previous functional studies and to future studies addressing the function of specific subpopulation of LNs separately from OSNs and PNs using a variety of UAS tools.
Using the GAL4-Krasvietz line, we characterized two distinct classes of LNs, Krasavietz_class1 and Krasavietz_class2 LNs (Figs. 2 and 3). A previous study reported that this line labels LNs that innervate all glomeruli homogeneously (Shang et al. 2007). Our LNs characterized as Krasavietz_class1 LNs could correspond to this type. Krasavietz_class2 LNs, however, innervating a subpopulation of glomeruli, are described here for the first time. For the GAL4-NP1227 line, one class of neurons, all sharing burst firing properties and extensive innervation patterns among the AL glomeruli, were characterized (NP1227_class1, Figs. 2 and 3). All neurons recorded from this line (n = 18) showed burst firing activity without exception, suggesting that this line is likely to consist of a homogeneous class of neurons. While characterizing the GAL4-NP2426 line, however, we found a few variations. The robustly identified NP2426_class1 LNs revealed tonic firing with fast spike adaptation and extensive innervation among the AL glomeruli (Figs. 2 and 3). Other variations of GAL4-NP2426 could not be unambiguously categorized. Okada et al. (2009) classified the LNs in this line into LN2L, which have their cell bodies in the lateral cluster, and LN2V, which have their cell bodies in the ventral cluster. In the present study, the neurons from the GAL4-NP2426 line were only obtained from the lateral cluster. The LN2L might thus include a few variations. It is still possible that we were unable to find all subclasses in the three GAL4 lines under study. However, our results provide detailed characteristics of the four classes of LNs identified as well as the component structure of the three LN-specific GAL4 lines. This information is essential to interpret data obtained from experiments manipulating these GAL4 populations and to design experiments for further investigations.
Implications of distinct firing properties
Our investigations revealed class-specific differences in intrinsic electrophysiological properties of the four different LN classes. These kinds of current-induced variation in basic firing were often found in the vertebrate brain system and were used to classify different types of neurons (e.g., Kawaguchi 1993; Yang et al. 1996).
In Krasavietz_class1 and Krasavietz_class2 LNs, the input is tonically coded in proportion to the stimulus strength and duration. In contrast, NP1227_class1 and NP2426_class1 LNs do not process the input stimulus continuously but rather respond transiently, especially at low stimulus intensities. These variations of phase differences in firing activity might have different roles in the coding mechanisms of the AL. One of the important functions proposed for the inhibitory circuits in the AL is gain control. A wide range of odor concentrations have to be coded in PNs, and a possible role of LNs could be to prevent PNs from saturating their responses. The axon terminals of OSNs are proposed as targets of GABAergic inhibition for gain control (Olsen and Wilson 2008b; Root et al. 2008). As OSN responses to odors contain diverse temporal patterns with various (low to high) firing frequency ranges (Hallem and Carlson 2006), Krasavietz_class1 and Krasavietz_class2 LNs showing faithful transmission of the input, both in strength and duration, might have a favorable property for homogeneously accommodating the strength of OSN input and might contribute to the gain control mechanism.
Finding bursting neurons in the circuit is of particular interest in terms of considering the similarity between the insect AL and the vertebrate olfactory bulb neural circuits. The external tufted cells, which are a subpopulation of the juxtaglomerular cells (LNs at the glomerular layer, including periglomerular cells, external tufted cells and short axon cells) in the olfactory bulb system, exhibit intrinsically generated burst firing (Hayar et al. 2004b). The burst of external tufted cells is synchronous with other external tufted cells and entrained to olfactory nerve stimuli delivered over the range of sniffing frequencies (Hayar et al. 2004b). It is thus suggested that the external tufted cells serve in detection and amplification of the timing of the sensory input and coordinate the glomerular output (Hayar et al. 2004b). The external tufted cells are glutamatergic and thus potentially excitatory neurons (Hayar et al. 2004a), but they connect GABAergic periglomerular cells directly with mitral cells (equivalent of PNs in insects), making reciprocal synapses, and thus might provide inhibitory input to mitral cells. NP1227_class1 LNs could have a similar function to that of the external tufted cell-periglomerular cell circuit in the vertebrate olfactory system, enhancing input timing and coordinating PN activity. Circuit operation mediated by the bursting neurons and underlying mechanisms for the bursting properties require additional studies.
A recent study on Drosophila demonstrated that LNs labeled by the GAL4-NP2426 line are necessary for generating odor-elicited oscillatory activity (Tanaka et al. 2009). This study suggested that reciprocal interactions of LNs and PNs underlie the mechanisms for generating odor-evoked oscillations. Here we measured the intrinsic response of one of these neuronal classes but observed no oscillatory responses. However, we observed strong adaptation properties in NP2426_class1 LNs, which are remarkable at the low firing frequency range (around 10–20 Hz), a similar range to that at which oscillations were observed in Drosophila (∼10 Hz) (Tanaka et al. 2009). Thus NP2426_class1 LNs might have intrinsic properties favorable for generating oscillatory responses. We also found variations among the LNs from this GAL4 line. It will therefore be important to investigate the relationship between oscillatory mechanisms with a particular subclass of LNs in future studies.
Common and specific morphological characteristics in different classes of LNs
It is suggested that glomerular interactions play important roles in neural computation in the AL (Olsen and Wilson 2008b; Olsen et al. 2007). The substantial connections among glomeruli are mediated by LNs. The interaction between glomeruli, if any, should thus be evaluated not by the geometry of glomerular positions but rather by the neural distance of the LN neurites and its electrical consequences. As a first step toward understanding the role of LNs in glomerular interaction, we have to know the precise innervation pattern of LNs among AL glomeruli. Here we provide a complete innervation pattern of single LNs (Fig. 4). We found two common morphological features in all four classes of LNs: 1) the LNs branch radially and distances to all innervated glomeruli from the primary neurite distribute similarly around the terminal arbors of the LNs. This suggests that if action potentials are initiated in the primary neurite, they might propagate to all innervated glomeruli with nearly equal strength. 2) The LNs have a hierarchical branching structure where the tightly connected glomerular subgroups near the terminals partially overlap. Whether LNs operate locally using local circuits or globally through action potentials is a topic of debate (Christensen et al. 2001; Husch et al. 2009b). Local circuits formed by tightly connected glomeruli could contribute to local dendritic computations using subthreshold activity, such as lateral inhibition and input integration.
By investigating the glomerular innervation map and comparing dendritic density and spatial distribution within a glomerulus (Figs. 3 and 5, Table 2), we found a strong correlation between morphological and physiological characteristics. Recent studies suggested that modulation by lateral networks of information transfer from OSNs to PNs might not be homogeneous in all glomeruli but instead specific, depending on the identity of a glomerulus (Olsen et al. 2007; Schlief and Wilson 2007; Silbering and Galizia 2007). This assumption suggests the existence of a glomerulus-specific lateral network (Silbering et al. 2008). Our discovery of Krasavietz_class2 LNs provides direct evidence for a glomerulus-specific difference regarding lateral interactions. Krasavietz_class2 neurons innervate a specific subpopulation of glomeruli, and this innervation pattern is stereotypic (Table 2). The selection of glomeruli by Krasavietz_class2 neurons does not correlate with the specificity of olfactory receptors expressed by OSNs innervating these glomeruli (Table 2). Thus the functional significance is elusive. On the other hand, the other three LN classes analyzed here innervate almost all glomeruli more homogenously. A previous study also demonstrated that presynaptic sites of LNs are distributed homogeneously all over the AL in the neural population of GAL4-NP1227 and GAL4-NP2426 LNs (Okada et al. 2009). One means of accomplishing glomerulus-specific inhibitory effects was, however, proposed in a recent study, which showed heterogeneous GABAB receptor expression level at the terminals of OSN axons among glomeruli (Root et al. 2008).
In addition to the glomerular specificity of lateral networks, our results provide evidence for class differences in the formation of the specific intraglomerular organization. The extensive arborization of Krasavietz_class2 LNs, in contrast to the sparse arborization of the other three classes, suggests a different organization of the synaptic structures within the individual glomerulus. Recent studies reported that GAL4-NP1227 neurons innervate the core region of a glomerulus, while GAL4-NP2426 neurons innervate the entire glomerulus (Okada et al. 2009; Tanaka et al. 2009). Our results show that NP2426_class1 LNs only innervate parts of a glomerulus and do this only sparsely, whereas neurons from other variations of GAL4-NP2426 show extensive innervations covering the whole glomerular region (Figs. 3 and 5 and Supplemental Fig. S2). This suggests that GAL4-NP2426 consists of a heterogeneous LN population differing in dendritic distribution within a glomerulus. To elucidate the exact connections between AL neurons, further studies are needed to examine the synaptic connections between LNs, OSNs and PNs in detail, using electron microscopic and/or physiological approaches.
Diversity of LNs in the Drosophila AL
While this paper was under review, a study providing a comprehensive analysis of the LNs in the Drosophila AL was published (Chou et al. 2010). The authors analyze the morphology of ∼1,500 single LNs from 10 GAL4 lines using MARCM methodology. The odor response profiles of ∼100 LNs are also established using whole cell patch-clamp recordings. Of the lines studied in the present paper, Chou et al. included the GAL4-Krasavietz line, whereas GAL4-NP1227 and GAL4-NP2426 lines were not studied.
Chou et al. provide a highly comprehensive characterization of LNs in the Drosophila AL, whereas our study focuses on several subclasses of LNs and offers a detailed characterization of these. Overall, both studies show consistent results but from different point of views and are thus mutually informative and facilitate the understanding of LN function in olfactory coding. Chou et al. provides an unprecedented catalogue of morphological classes of LNs combining neurotransmitter profiles, connectivity, and physiological properties. Regarding the total number of LNs in the AL, Chou et al. reveals that there are at least ∼100 ipsilateral LNs and ∼100 bilateral LNs. We used three GAL4 lines; GAL4-Krasavietz [∼15 LNs, Shang et al. (2007)], GAL4-NP1227 [∼18 LNs, Okada et al. (2009) and Sachse et al. (2007)], and GAL4-NP2426 [∼37 LNs, Okada et al. (2009) and Sachse et al. (2007)]. A previous study reported that NP1227 and NP2426 lines express GAL4 mutually exclusively in different population of LNs (Okada et al. 2009). These two lines thus cover ∼55 LNs in total. The possibility of overlap between the Krasavietz line and the other two lines used still remains. However, the two classes of LNs characterized from GAL4-Krasavietz were distinct from the two classes of LNs characterized from GAL4-NP1227 and GAL4-NP2426, and the overlap should consequently be limited, if any. In line with this, the three GAL4 lines used in our study cover ∼70 LNs in total. Because not all LNs from these three lines were characterized in our study, we estimate that still roughly half remain uncharacterized among ∼100 ipsilateral LNs. Chou et al. also showed that most neurons in the Krasavietz line are GABAergic; this is consistent with our results but in conflict with the results of a previous study (Shang et al. 2007).
Morphologically Chou et al. categorize LNs into eight broad classes based on glomerular innervation patterns. The four classes of LNs presented in our study can be categorized into three of these. First, two of our classes, innervating almost all glomeruli (Krasavietz_class1 and NP2426_class1), can be categorized into two categories of Chou et al.: LNs innervating all glomeruli (pan-glomerular) or all but a few glomeruli (Table. 2). Second, NP1227_class1 LNs are included in “all but a few glomeruli” LNs (Table. 2). Finally, the Krasavietz_class2 LN type, which innervates a specific subpopulation of glomeruli, is highly similar to the “dumbbell neuron” of Chou et al., which is categorized into the LN class innervating a continuous region of the AL (Table. 2).
In the physiological analysis, Chou et al. used in vivo recordings to analyze odor response properties of LNs. In comparison, we used a different strategy to characterize fine details of LN electrophysiological properties using in situ recording. Functional subclasses could be segregated using the electrophysiological properties even though the global glomerular innervation patterns were similar, such as among Krasavietz_class1, NP1227_class1, and NP2426 class1 LNs.
Chou et al. find physiological differences in spontaneous firing rates and temporal firing patterns between the different genetic classes of LNs defined by the different GAL4 lines and between the different morphological classes. Interestingly, they show that LNs not innervating pheromone glomeruli tend to show transient burst excitation to odor stimuli. In our study, some NP1227_class1 LNs, which were characterized by burst firing, and some NP2426-class1 LNs, which showed fast adapting spikes, also did not innervate the pheromone glomeruli and might thus correspond to the nonpheromone glomeruli LNs of Chou et al. Although Chou et al. analyze physiological differences based on their different morphological classification, our results suggest that the two morphological classes showing pan-glomerular and “all but a few glomeruli” innervation patterns might be a continuum in some part, because we found both pan-glomerular and “all but a few glomeruli” innervation patterns within two classes of LNs (Krasavietz_class1 and NP2426_class1).
Here we characterized olfactory LNs in the AL by directly correlating physiological and morphological properties of these neurons. Our results provide essential information to understand the neural basis of olfactory processing in a model system, the Drosophila AL. This study offers stepping stones to future computational studies and to the design of future experimental approaches to deciphering odor coding mechanisms.
This study was supported by the Max Planck Society.
No conflicts of interest, financial or otherwise, are declared by the author(s).
We thank A. Baschwitz and V. Grabe for assisting with morphological reconstructions, S. Bisch-Knaden for support with statistical analyses, and M. Stensmyr for helpful comments on the manuscript. We are grateful to K. Ito for kindly providing the fly lines GAL4-NP1227 and GAL4-NP2426 and to G. Miesenböck for generously providing the fly line GAL4-Krasavietz:UAS-mCD8-GFP. The nc82-antibody and ChAT-antibody generated by E. Buchner and P. M. Salvaterra, respectively, were obtained from the Developmental Studies Hybridoma Bank, developed under the auspices of the National Institute of Child Health and Human Development and maintained by the University of Iowa.
↵1 The online version of this article contains supplemental data.
- Copyright © 2010 the American Physiological Society