Convergence of multimodal sensory pathways to the mushroom body calyx in Drosophila melanogaster

Detailed structural analyses of the mushroom body which plays critical roles in olfactory learning and memory revealed that it is directly connected with multiple primary sensory centers in Drosophila. Connectivity patterns between the mushroom body and primary sensory centers suggest that each mushroom body lobe processes information on different combinations of multiple sensory modalities. This finding provides a novel focus of research by Drosophila genetics for perception of the external world by integrating multisensory signals.


Results
We performed injections of dextran conjugated with fluorescent dye into the primary sensory centers to investigate whether the MB calyx in Drosophila is also connected with multiple primary sensory centers. The MB calyx can be divided into four parts: the main calyx (CA), dorsal, lateral, and ventral accessory calyces (d-, l-, and vACA) 8, [14][15][16][17][18][19] (Fig. 1h and Supplementary Movie S1). The dye injections revealed that the MB calyx is connected with not only the AL and OL, but also SEZ via three major antennal lobe tracts (ALTs), three OL-calycal tracts (OLCTs), and one subesophageal-calycal tract (SCT) (Fig. 1a-g). The neurons running through the OLCTs are also called visual projection neuron (VPN)-MBs 8 . The terminals of ALTs were observed not only in the CA 10,11 , but also in the dACA and lACA. The OLCTs terminated in both dACA and vACA, whereas the SCT in dACA.
We then screened for the Janelia Farm GAL4 strains 14 which visualize neurons running through these tracts and innervating the MB calyx except for the conventional ALT neurons which are known to supply branches into the main calyx 10,11,20 , subsequently identifying eleven strains, which labeled five OLCTs, two ALTs, and one SCT which send projections into the ACAs of the MB ( Table 1). The dACA is innervated by two OLCTs, the medial ALT (mALT), and SCT. The lACA is innervated by one OLCT and the mediolateral ALT (mlALT) and the vACA by four OLCTs, respectively (Table 1 and Supplementary Fig. S1). Dye injection and labeling with the GAL4 lines showed that the terminal areas of OLCT and ALT neurons are overlapping in the lACA and posterior area of the dACA, however, are distributed to the vACA and CA, respectively, in a segregated manner ( Fig. 1d-g).
We further visualized the morphologies of single cells of these tracts by performing the FLP-out experiments 21 using the GAL4 strains (Fig. 2, Supplementary Fig. S2, and Table 1). Most of the OLCTs emanated from the seventh layer of the medulla (ME7, serpentine layer), sixth layer of the lobula (LO6), or accessory medulla (AME). The ME7 and LO6 are the target layers of the chromatic Tm neurons 22,23 , suggesting that these layers are related to color processing. Only a few neurons innervate the layer entirely, however, most OLCT neurons have specific arborization areas in the OL. For example, one OLCT1 neuron innervates the dorsal three-quarters of the LO6 (Fig. 2a), whereas the innervation of one OLCT2 neuron is restricted to the ventralmost ME7 (Table 1). This result suggests that these neurons transfer information of specific visual fields. It should be noted that none of the OLCT neurons have the same morphologies with recently reported VPNs (VPN-MB1 and VPN-MB2) terminating in the vACA 8 . For the ALTs, we found two ALT neurons terminating in the ACA. One neuron, which innervates the VP1 glomerulus that receives chemosensory, hygrosensory, or thermosensory inputs from the sensory neurons in the sacculus of the antenna 19,24 , projects into the dACA through the mALT (Fig. 2d). The other, mlALT neuron, originating from the VP3 glomerulus which receives input from cold-sensing neurons 19,25 , runs through the mediolateral ALT and then pedunculus, and forms a cloud of terminals in the lACA, the structure of which is quite similar with the terminals of the tritocerebral tract in the orthopteran insects 26 (Fig. 2e). The morphology of this mlALT neuron appears quite similar with that of the t5ALT neuron 14,19 , though the neuron we found runs through the mlALT. The SCT neuron originating from the gnathal ganglia (GNG) or flange (FLA) joins the mALT and sends branches into the dACA (Fig. 2f).
Taken together, the dACA is connected with the LO6, ME7, AME, VP1 glomerulus, FLA, and GNG. On the other hand, the vACA is linked with the LO6, ME1-7, and AME, whereas the lACA is linked with the AME and VP3 glomerulus. This indicates that each calycal part receives sensory inputs of different combinations of modalities and that the MB in Drosophila may play a role in general learning and memory.
We furthermore investigated how each calycal part is governed by different classes of Kenyon cells. Kenyon cells are classified according to the lobes they terminate in: γ , α /β , or α '/β ' lobe neurons [15][16][17]27,28 . We analyzed the dendritic patterns of these neurons by using the GAL4 enhancer-trap strains we previously reported 28 . Since there were no available strains for labeling α '/β ' lobe neurons specifically, we compared staining patterns among the strains labeling Kenyon cells of multiple lobes to estimate the probable innervation areas of α '/β ' lobe neurons. Each population of γ , α /β , or α '/β ' lobe neurons arborizes in the CA, whereas the ACAs are innervated by specific populations ( Fig. 3 and Table 2). The dACA is innervated by a subpopulation of α /β lobe (α p/β p) Kenyon cells 28 , whereas the vACA is innervated by γ lobe neurons 8,[15][16][17] . However, the innervation into the lACA was not observed in those strains that specifically label γ or α /β lobe neurons, but strains which visualize α '/β ' lobe neurons, suggesting that the lACA is innervated by α '/β ' lobe neurons. Since each ACA is connected with different primary sensory centers, these anatomical findings suggest that each lobe may have different functions in multimodal sensory processing (Fig. 4).

Discussion
This study reveals multimodal sensory pathways into the MB calyx in Drosophila as has been reported in such insects as the honeybee, ant, butterfly and cockroach 12,13,29,30 . Recent work has also revealed that two other OLCT neurons, VPN-MB1 and MB2 are required for visual memories of color and brightness in Drosophila 8 . In the bumblebee and honeybee, the neurons of the anterior inferior optic tract and the lobula tract which arborize in the serpentine layer of the medulla and in LO5 or LO6 respectively and project to the MB calyx respond to color and motion stimuli 31 . Due to morphological similarities of the OLCT neurons found in this study to these neurons and VPN-MBs 8 , the Drosophila MB may also receive similar visual information and combine it with odor, gustatory, and temperature information in each lobe. Recent physiological works support this idea: the γ lobe neurons innervating the vACA respond to visual stimuli 8 , the ALT neuron terminating in the lACA shows calcium responses to temperature shifts 19 , and taste activity has been observed in the dACA 9 . Further physiological works revealing the sensory information transferred from the primary sensory centers to the MB calyx by the neurons found in this study need as yet to be determined.
Previous works have shown that each MB lobe contributes to a different temporal phase of memory 1,2 . This study suggests that each lobe functions in integrating a different combination of multiple sensory modalities. In the cockroach, the MB extrinsic neurons which innervate the MB lobe respond to multimodal sensory inputs in a context-specific manner 32 . Readily available Drosophila genetics 17,18,28 can provide opportunities to reveal novel neural mechanisms by which a rich variety of sensory cues are combined to represent coherent perception of the external world.   Dextran injection into the primary sensory centers. Flies were anesthetized in a vial on ice for less than a minute and were fixed to plastic chambers with wax and epoxy. The compound eye, top of the head, or proboscis was opened with forceps in Drosophila saline (in mM: NaCl 103, KCl 3, MgCl 2 4, CaCl 2 1.5, NaHCO 3 26, TES 5, trehalose 10, glucose 10, sucrose 7, NaH 2 PO 4 1, adjusted to pH 7.25 with HCl). After fat, air sacs, and sheathes around the OL, AL, or SEZ were gently removed with forceps, dextran conjugated with tetramethylrhodamine and biotin (3 kDa, D-3308, Life Technologies) was injected into one of the OL, AL and SEZ with forceps 20 . Within 20 minutes after dextran injection, the brains were dissected out and fixed with 4% formaldehyde or 4% paraformaldehyde (PFA) for 50 minutes at room temperature. The brains were then washed in phosphate  buffered saline (PBS, pH 7.4). When the GFP and tetramethylrhodamine signals needed to be enhanced, the brains were further incubated in the rabbit anti-GFP antibody and Alexa Fluor 568-conjugated streptavidin (S-11226, Life Technologies; 1 mg/l diluted in 0.2% Triton X-100/PBS (PBST)), respectively, followed by the antibody staining procedures (see below). The brains were finally mounted in 50 or 80% glycerol/PBS. Antibody staining. Brains were dissected out in PBS and fixed with 4% PFA for 50 minutes. After washing with PBS for 5 minutes three times, the brains were then shaken in the blocking solution: 10% goat serum in PBST for an hour. They were incubated in the blocking solution containing primary antibodies overnight at room temperature with shaking. After washing with PBST, the brains were then incubated in the blocking solution  Table 2. GAL4 enhancer-trap strains that label Kenyon cells. It should be noted that more lobes were labeled with NP1131 and NP2758 than those in our previous work 28 , possibly because the GFP expression is enhanced by the repeats of UAS in this work.  Labeling single neurons with GFP. Janelia Farm GAL4 strains labeled multiple neurons including our targets, which prevented us from analyzing the morphology of single cells. To reduce the number of neurons labeled, we used the FLP-out system 21 . In this system, we crossed each GAL4 strain with the flies carrying Hs-flp and UAS > CD2, y + > CD8::GFP and incubated one-day old adult offspring at 37 degrees Celcius for 1 hour to express the flippase gene to induce somatic recombination randomly, which resulted in GFP expression in a fewer numbers of neurons. We then maintained the flies for 6-8 days at 25 degrees Celcius before dissecting the brains out. The brains were then fixed with 4% PFA, and immunostained with the rabbit anti-GFP antibody (diluted at 1:1000) and the mouse anti-SYNAPSIN antibody (diluted at 1:20). For secondary antibodies, we used goat anti-rabbit antibody conjugated with Alexa Flour 488 (diluted at 1:1000) and goat anti-mouse antibody conjugated with Alexa Flour 568 (diluted at 1:200). The brains were washed with PBS and finally mounted in 80% glycerol/PBS. The brains expressing CD8::GFP in a single neuron of interest were imaged.
Confocal imaging and three-dimensional reconstructions of optical sections. Confocal serial optical images of the whole-mount brain were taken at 0.7-1 μ m z-intervals with a Carl Zeiss confocal microscope LSM 700 equipped with water-immersion C-Apochromat 40× . Three-dimensional reconstruction and surface rendering were performed with Zeiss ZEN 2012 and Avizo 5.1, respectively. The brightness, color, and contrast of images were adjusted with Photoshop CS 5.1. For the left panels of Fig. 2d-f, the antibody signal levels of the fibers we analyzed were adjusted to increase the signal/noise ratio by changing the levels manually in Photoshop before the reconstructions. For the right panels of Fig. 2f, antibody signals except for the SCT1, such as the noise backgrounds on the surface of the brain were manually removed to visualize the GAL4 positive neuron more clearly.
Terminology. The spatial definition and nomenclature of each lobe and axonal tract is based on ref. 39. The layer divisions in the medulla and lobula are referred to ref. 40.