Neural Coding of Leg Proprioception in Drosophila.

Methods to record and map proprioceptive signals in Drosophila Positioned in the proximal femur of each Drosophila leg is a femoral chordotonal organ (FeCO) that contains ~135 cell bodies ( Figure 1A ). The dendrites of the FeCO neurons are connected to the cuticle and surrounding muscles by attachment cells and tendons ( Shanbhag et al., 1992 ), and the axons of FeCO neurons project through the leg nerve into the ventral nerve cord (VNC; Phillis et al., 1996 ; Smith and Shepherd, 1996 ). Most FeCO axons arborize within the VNC neuropil, with only 3–4 cells from each leg projecting directly to the brain ( Figure 1A , D ) ( Tsubouchi et al., 2017 ). To investigate proprioceptive signal encoding of the FeCO population, we recorded the activity of proprioceptor axons while manipulating the joint angle between the femur and tibia of the fly’s right front leg ( Figure 1B – C and Movie S1 ). We chose to control leg kinematics, rather than force, because biomechanical studies have found that chordotonal neurons directly monitor joint displacement ( Field and Matheson, 1998 ). For fast and accurate positioning of the leg, we designed a magnetic control system that allowed us to manipulate a pin glued to the fly’s tibia using a servo-actuated magnet ( Figure 1B ). This system allowed us to reproduce the range of tibia positions and speeds observed in walking flies (see Methods for details). The femur and proximal leg joints were fixed to the fly holder with UV-cured glue. We continuously recorded the position of the tibia using an IR-sensitive video camera ( Figure 1C ), and automatically tracked its orientation to calculate the femur-tibia joint angle. We used the Gal4-UAS system ( Brand and Perrimon, 1993 ) to express a genetically encoded calcium indicator, GCaMP6f ( Chen et al., 2013 ) in the majority of FeCO neurons ( iav-Gal4 ; Kwon et al., 2010 ) and recorded their calcium activity in vivo with two-photon calcium imaging ( Figure 1B ). To identify proprioceptor projections with shared functional tuning, we first recorded calcium activity of FeCO axons as we swung the tibia from flexion to extension at 360°/s ( Figure 1F – H ). To categorize calcium activity in an unbiased manner, we then calculated pairwise correlations between the calcium signal (∆F/F) in each pixel and performed k-means clustering on the resulting correlation matrix. Figure 1F shows an example of a pixel correlation matrix before and after clustering. In this trial, we identified four groups of pixels whose activity was highly correlated (for details about how the number of clusters was selected, see Figure S1A ). Although this correlation-based clustering does not impose any spatial restrictions, we found that the pixels that clustered together based on their activity were also grouped together spatially ( Figure 1G – H ). This example illustrates that clustering of pixel correlations is sufficient to identify groups of FeCO axons that encode distinct proprioceptive stimulus features.

STAR METHODS Contact for reagents and resource sharing

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact John Tuthill ( tuthill@uw.edu ).

Experimental model and subject details

Drosophila melanogaster were raised on a standard cornmeal and molasses medium and kept at 25˚C in a 12:12 h light dark cycle. We used female flies 1 to 5 days post-eclosion for all experiments, except for FlpOut functional imaging experiments. In the FlpOut imaging experiments, we heatshocked the flies (genotype shown in the table below) 1–3 days post eclosion for 16 to 25 min at 37˚ C and imaged the flies 5 to 7 days after the heat shock. The genotypes used for each experiment are included in Table S1 .

Experimental model and subject details

Drosophila melanogaster were raised on a standard cornmeal and molasses medium and kept at 25˚C in a 12:12 h light dark cycle. We used female flies 1 to 5 days post-eclosion for all experiments, except for FlpOut functional imaging experiments. In the FlpOut imaging experiments, we heatshocked the flies (genotype shown in the table below) 1–3 days post eclosion for 16 to 25 min at 37˚ C and imaged the flies 5 to 7 days after the heat shock. The genotypes used for each experiment are included in Table S1 .

Method Details Fly preparation for in vivo two-photon calcium imaging of FeCO axons To gain optical access to the VNC while moving the tibia, we used a previously described fly holder ( Tuthill and Wilson, 2016b ; Figure 1B ), but replaced the metal sheet that holds the fly’s thorax with thin, translucent plastic. The plastic sheet served as a light diffuser and provided a bright background during the automated tracking of the tibia. To place the fly into the holder, we first anesthetized the fly by cooling in a plastic tube on ice, then put the fly’s head through the hole in the holder and glued the ventral side of the thorax onto the hole using UV-cured glue (Bondic). We glued the head to the upper side of the fly holder. On the bottom side of the holder, we glued down the femur section of the right prothoracic leg so that we could control the femur-tibia joint angle by moving the tibia. When gluing the femur, we held it at a position where the movement of the tibia during the rotation of the femur-tibia joint was parallel to the plane of the fly holder. To eliminate mechanical interference, we glued down all other legs. We also pushed the abdomen to the left side and glued it at that position, so that the abdomen did not block tibia movement during flexion. To position the tibia with the magnetic control system described below, we cut a small piece of insect pin (length ~1.0 mm, 0.1 mm diameter; Living Systems Instrumentation) and glued it onto the tibia and the tarsus of the right prothoracic leg. We painted the pin with black India ink (Super Black, Speedball Art Products) to enhance contrast and improve tracking of the tibia/pin position. After immersing the preparation in Drosophila saline, we removed the cuticle above the prothoracic segment of the VNC with fine forceps and took out the digestive tract to reduce the movements of the VNC. We also removed fat bodies and larger trachea to improve the optical access to the leg neuropil. The fly remained alive throughout the experiment, as indicated by the presence of spontaneous leg movements when the magnet was removed. Fly saline contained: 103 mM NaCl, 3 mM KCl, 5 mM TES, 8 mM trehalose, 10 mM glucose, 26 mM NaHCO3, 1 mM NaH2PO4, 1.5 mM CaCl2, and 4 mM MgCl2 (pH 7.1, osmolality adjusted to 270–275 mOsm). Recordings were performed at room temperature.

Image acquisition using a two-photon excitation microscope

We used a modified version of a custom two-photon microscope previously described in detail ( Euler et al., 2009 ). For the excitation source, we used a mode-locked Ti/sapphire laser (Mira 900-F, Coherent Inc) set at 930 nm and adjusted the laser power using a neutral density filter to keep the power at the back aperture of the objective (40x, 0.8 NA, 2.0 mm wd; Nikon Instruments Inc) below ~25 mW during the experiment. We controlled the galvo laser scanning mirrors and the image acquisition using ScanImage software (version 5.2) within Matlab (MathWorks). To detect GCaMP6f and Td-Tomato fluorescence, we used an ET510/80M (Chroma Technology Corporation) emission filter (GCaMP6f) or a 630 AF50/25R (Omega optical Inc) emission filter (Td-Tomato) and GaAsP photomultiplier tubes (H7422P-40 modified version without cooling; Hamamatsu Photonics). During the trials, we acquired images (256 × 120 pixels or 128 × 240 pixels) at 8.01 Hz. At the end of the experiment, we acquired a z-stack of the FeCO axons in the right hemisphere of the prothoracic leg neuropil to confirm the recording location. In initial pilot experiments, we sampled different imaging regions in the VNC, and selected 4 ROIs that captured all of the major response classes (shown in Figure 2 ). We used these 4 ROIs for all the experiments reported in this paper. These ROIs corresponded to the axon terminals of major axon bundles of the FeCO. Moving the tibia/pin using a magnetic control system We designed the magnetic control system to manipulate the variable most relevant for chordotonal neurons, namely joint displacement ( Field and Matheson, 1998 ). To move the tibia/pin to different positions, we attached a rare earth magnet (1 cm height x 5 mm diameter column) to a steel post (M3×20mm flat head machine screw) and controlled its position using a programmable servo motor (SilverMax QCI-X23C-1; Max speed 24,000 ˚/s, Position resolution 0.045˚; QuickSilver Controls Inc). We placed a piezoelectric crystal (PA3JEW; ThorLabs) between the post and the magnet in order to vibrate the tibia as described below. To move the magnet in a circular trajectory centered at the femur-tibia joint, we placed the motor on a micromanipulator (MP-285, Sutter Instruments) and adjusted its position while visually inspecting the movement of the magnet and the tibia using the tibia tracking camera described below. This brought the top edge of the magnet to the same height as the tibia/pin, and the inner edge of the magnet to be 1.5 mm from the center of the femur-tibia joint ( Figure 1C ). Because the pin was glued slightly off the center of the joint, the distance between the pin and the magnet was approximately 300 µm. For each trial, we controlled the speed and the position of the servo motor using QuickControl software (QuickSilver Controls, Inc). During all trials, we tracked the tibia position (as described below) to confirm the tibia movement during each trial. Because it was difficult to fully flex the femur-tibia joint without the tibia/pin and the magnet colliding with the abdomen, we only flexed the joint up to ~18˚. During swing motion trials ( Figures 2 and 4 ), we commanded the motor to move from a fully extended position (180˚) to a fully flexed position (18˚) at 360 ˚/s and move back to the original position with the same speed. (For velocity sensitivity experiments in Figure S5 , we also used 180, 720, and 1440 ˚/s movements.) We chose these range of speeds based on the available data on the kinematics of the leg movements in walking Drosophila and other insects. Although the exact kinematics of the femur-tibia joint during walking in Drosophila has not been described, quantification of gait parameters in freely walking flies show that the swing duration of the forelegs ranges from approximately 25 to 45 ms, while the stance duration ranges from 30 to 140 ms ( Mendes et al., 2013 ; Wosnitza et al., 2013 ). Data from other insects suggest that the femur-tibia joint does not move through the entire range of motion (0˚ to 180˚) during walking. For example, for the mesothoracic leg of the fast walking cockroach, the maximum femur-tibia joint excursion is approximately 60˚ ( Watson and Ritzmann, 1998 ) and our unpublished observations of tethered, walking Drosophila have found a maximum excursion of ~80⁰. These ranges would correspond to the maximum average speed of 2400–3200 ˚/s for the swing phase and approximately 2000–2670 ˚/s for the stance phase. This is comparable to the mean joint angular velocity reported for the mesothoracic leg of the cockroach during walking, which ranges from 0 to 800 ˚/s ( Watson and Ritzmann, 1998 ). For the data in Figure 2 , there was a 5-second interval between the flexion and the extension movement. We repeated this swing motion 3 times with a 5-second inter-trial interval. Responses to each repetition of the swing motion were similar, and thus we averaged the responses within each fly prior to averaging across flies. In the experiments using iav-Gal4 , the mean ratio of the amplitude of the response to the 2 nd stimulus compared to the 1 st , and 3 rd stimulus compared to the 2 nd , was 0.953±0.035 (mean±sem) and 0.9748±0.034, respectively, for bidirectional phasic neurons responding to flexion, 0.907±0.033 and 0.998±0.023, respectively, for bidirectional phasic neurons responding to extension, 0.987±0.023 and 1.001±0.025, respectively, for flexion selective phasic neurons, 0.984±0.025 and 1.024±0.039, respectively, for extension selective phasic neurons, 0.927±0.043 and 1.016±0.118, respectively, for flexion selective tonic neurons, and 1.099±0.036 and 1.176±0.054, respectively, for extension selective tonic neurons. For the experiments using R73D10-Gal4 , they were 0.894±0.029 and 0.969±0.026, respectively, for flexion selective tonic neurons, and 1.179±0.036 and 1.061±0.031, respectively, for extension selective tonic neurons. For the club neurons (expressed using R64C04-Gal4 ), they were 1.058±0.038 and 0.997±0.021, respectively, for responses to flexion, and 1.015±0.036 and 0.975±0.046, respectively, for responses to extension. For the hook neurons (expressed using R21D12-Gal4 ), they were 0.971±0.017 and 0.984±0.019, respectively, for responses to flexion. For ramp-and-hold motion trials ( Figures 5 and 7 ), we programmed the motor to move in 18˚ steps between full extension and flexion at 240 ˚/s. There was a 3-second hold step between each ramp movement. We used two types of movements for a ramp-and-hold motion: one that started with a flexion of the joint and extended back to the original position and another that started with an extension of the joint and flexed back to the original position. We repeated each type of trial twice and averaged the responses within each fly before averaging across flies. We set the acceleration of the motor to 72000 ˚/s 2 for all movements. Movements of the tibia during each trial varied slightly due to several factors, including a small offset between the center of the motor rotation and the femur-tibia joint, and the acceleration and deceleration of the tibia movement in response to the magnet motion. Because these variations were relatively small (judging from how the responses changed with speed in Figure S5B ), we did not consider these differences in the initial summary of the responses to swing motion and ramp-and-hold motion ( Figures 2 , 4 , and 5A-C ). However, for quantifying the positional dependence of the responses, we plotted the response against the actual position of the tibia during each trial ( Figure 5D ). Because the actual position of the tibia differed slightly between preparations, we interpolated the data at 5˚ intervals when averaging position dependent responses across preparations. Tracking the femur-tibia joint angle To track the position of the tibia, we backlit the tibia/pin with an 850 nm IR LED (M850F2, ThorLabs) and recorded video using an IR sensitive high speed video camera (Basler Ace A800–510um, Basler AG) with a 1.0x InfiniStix lens (94 mm wd, Infinity) equipped with 900 nm short pass filter (Edmund optics) to filter out the two-photon laser light ( Figure 1B ). Because the servo motor was directly underneath the fly, we placed the camera to the side and used a prism to capture the view from below. We recorded the images at 180 Hz for the ramp-and-hold motion, and at 200 Hz for the swing motion (exposure time 2.5 ms for both types of motion). To synchronize the images taken by the camera with those taken by the two-photon microscope, we acquired both the camera exposure signal and the position of the galvo scanning mirrors at 20 kHz. After acquiring the images ( Figure 1C ), we identified the position of the dark tibia/pin against the bright background by thresholding the image. We then approximated the orientation of the leg as the long axis of an ellipse with the same normalized second central moments as the thresholded image ( Haralick and Shapiro, 1992 ). The spatial resolution of the image was 3.85 µm per pixel and assuming circular movement of the tibia/pin, 1-pixel movement at the edge of the tibia/pin (~1.2 mm from the center of the rotation) corresponded to 0.18˚. Vibrating the tibia using a piezoelectric crystal To vibrate the tibia at high frequencies, we moved the magnet using a piezoelectric crystal (PA3JEW, Max displacement 1.8 µm; ThorLabs) ( Figure 6A ). For controlling the movement of the piezo, we generated sine waves of different frequencies in Matlab (sampling frequency 10 kHz) and sent them to the piezo through a single channel open-loop piezo controller (Thorlabs). Because tibia movements induced by the piezo electric crystal were below the resolution of our tibia tracking system (3.85 µm/pixel), we first calibrated the piezo induced tibia movements using a separate tracking system equipped with a long working distance high magnification objective (50x, 0.45 NA, Nikon) connected to a high speed video camera (A800–510um, Basler) via InfiniTube FM-100 (Infinity) ( Figure S6 ). In this setup, we were able to record images at 4000 Hz (exposure 190 µs) with a spatial resolution of 0.106 µm/pixel. For better control of the tibia position during high-frequency vibration, we attached the magnet to the pin for all vibration experiments (magnet overlapped with the pin for ~200 µm). For each vibration frequency, we measured the amplitude of the tibia oscillation envelope ( Figure S6B , D ) and power spectrum of the tibia movement ( Figure S6C , E ; Thomson’s multitaper power spectral density estimate with time-half bandwidth product = 4). The power spectrum of tibia movement showed a large peak at the command frequency of the piezo electric crystal, suggesting that most of the vibrations were indeed at the target frequency ( Figure S6C ). Both the amplitude of the oscillation envelope and the power of the oscillation at the target frequency decreased greatly at around 2000 Hz ( Figure S6D , E ). Thus, we decided to use vibration stimuli up to 2000 Hz. For each stimulus, we presented 4 seconds of vibration twice with an inter-stimulus interval of 8 seconds. We averaged the responses within each fly before averaging across flies. For calculating ∆F/F maps and average response amplitudes, we used the activity level in a 1.25-second window starting 1.25 seconds after the vibration onset.

Immunohistochemistry and anatomy

For confocal imaging of the FeCO neuron axons driven by each Gal4 line in the VNC ( Figure 3A ), we crossed flies carrying the Gal4 driver to flies carrying pJFRC7–20XUAS-IVS-mCD8::GFP and dissected the VNC out of the thorax in Drosophila saline. We first fixed the VNC in a 4% formaldehyde PBS solution for 15 minutes. After rinsing the VNC in PBS three times, we put it in blocking solution (5% normal goat serum in PBS with 0.2% Triton-X) for 20 minutes, then incubated it with a solution of primary antibody (anti-CD8 rat antibody 1:50 concentration; anti-brp mouse for neuropil staining; 1:50 concentration) in blocking solution for 24 hours at room temperature. At the end of the first incubation, we washed the VNC in PBS with 0.2% Triton-X (PBST) three times, then incubated the VNC in a solution of secondary antibody (anti-rat-Alexa 488 1:250 concentration; anti-mouse-Alexa 633 1:250 concentration) dissolved in blocking solution for 24 hours at room temperature. Finally, we washed the VNC in PBST three times and then mounted it on a slide with Vectashield (Vector Laboratories). We acquired a z-stack image of the slides on a confocal microscope (FV1000; Olympus) to capture the axonal projection pattern relative to the neuropil. For multicolor FlpOut experiments ( Figure 3B ), we crossed flies carrying the multicolor FlpOut cassettes and Flp recombinase drivers ( Nern et al., 2015 ) to flies carrying different Gal4 drivers, and dissected out the VNCs of resulting progeny. For temperature induced expression of Flp, we placed adult flies in a plastic tube and incubated them in a 37˚C water bath for 13 to 15 minutes (up to 1 hour for the R21D12-Gal4 flies). We dissected the VNC three days after the Flp induction and followed the procedure described in Nern et al. (2015) to detect HA (using anti-HA-rabbit antibody and anti-Rabbit-Alexa 594 secondary antibody), V5 (using DyLight 549-conjugated anti-V5), and FLAG (using anti-FLAG-rat antibody and anti-Rat-Alexa 647 secondary antibody) labels expressed due to Flp induction in individual neurons. VNCs were mounted in Vectashield and imaged on a confocal microscope (Leica SP8). Single neurons were manually traced using the Simple Neurite Tracer in Fiji ( Longair et al., 2011 ). For co-labeling of FeCO cell bodies ( Figure 3C – F ), we crossed flies carrying UAS-RedStinger , LexAop-nlsGFP , and ChAT-LexA to each of the three FeCO Gal4 lines. Legs from the resulting flies were removed with forceps, mounted in Vectashield, and imaged on a confocal microscope (FluoView 1000; Olympus). To improve image quality, we imaged each leg from both the dorsal and ventral sides and stitched the resulting images together using the pairwise stitching function in Fiji ( Preibisch et al., 2009 ). Cell bodies were then counted manually with Fiji. For in silico overlay of the expression patterns of specific Gal4 lines ( Figure 3G ), we used confocal stacks of each Gal4 line with neuropil counterstaining (from the Janelia FlyLight database; Jenett et al., 2012 ) and used the neuropil staining to align the expression pattern in the VNC using the Computational Morphometry Toolkit (CMTK; Jefferis et al., 2007 ).

Supplementary Material 1 Movie S1, related to Figure 1 . Controlling the fly femur-tibia joint during 2-photon imaging from proprioceptor axons. Top: A view of a fly from below during ramp-and-hold movement of the femur-tibia joint, recorded with an IR-sensitive high-speed video camera. To control joint angle, we glued an insect pin (painted black) to the tibia (indicated by a red line) and positioned it using a magnet mounted on a servo motor. The blue line indicates the femur. Bottom: Femur-tibia angle during the ramp-and-hold trial, automatically tracked from the video shown above. 2 Movie S2, related to Figure 2 . Functional organization of proprioceptor axons in the fly VNC. Example recordings of GCaMP6f signals from each region shown in Figure 2 . For each region, we first display the representative ROIs, calcium signals, and spatial layout as shown in Figure 2 , followed by an example recording during a swing stimulus. Red lines in the example recording represent the femur-tibia joint angle, with the red dot indicating the distal tip of the tibia. 3 Movie S3, related to Figure 5 . Functional subtypes of proprioceptor axons in the fly VNC. Left column: Example recordings of GCaMP6f signals from each region show in the left column of Figure 5A – C during a ramp-and-hold stimulus. Red lines in the example recording represent the femur-tibia joint angle during the trial. Right Column: same as the middle column of the Figure 5A – C . 4 Figure S1, related to Figures 1 and 2 . Example of cluster analysis and population responses for opposite swing stimuli. Figure S2, related to Figure 2 . Additional quantification of population responses. Figure S3, related to Figure 3 . Anatomy of Gal4 driver lines used to label FeCO neurons. Figure S4, related to Figure 4 . Responses of neurons labeled by specific Gal4 lines to swing stimuli. Figure S5, related to Figure 4 . Additional quantification of subclass-specific responses. Figure S6, related to Figure 6 . Calibration of vibration stimuli. Figure S7, related to Figure 7 . Examples responses of single club, claw, and extension selective phasic neurons. Table S1, Table of genotypes, related to STAR Methods experimental model and subject details.