Long-term Behavioral Tracking of Freely Swimming Weakly Electric Fish

Summary

We describe a set of techniques for studying spontaneous behavior of freely swimming weakly electric fish over an extended period of time, by synchronously measuring the animal's electric organ discharge timing, body position and posture both accurately and reliably in a specially designed aquarium tank inside a sensory isolation chamber.

Abstract

Long-term behavioral tracking can capture and quantify natural animal behaviors, including those occurring infrequently. Behaviors such as exploration and social interactions can be best studied by observing unrestrained, freely behaving animals. Weakly electric fish (WEF) display readily observable exploratory and social behaviors by emitting electric organ discharge (EOD). Here, we describe three effective techniques to synchronously measure the EOD, body position, and posture of a free-swimming WEF for an extended period of time. First, we describe the construction of an experimental tank inside of an isolation chamber designed to block external sources of sensory stimuli such as light, sound, and vibration. The aquarium was partitioned to accommodate four test specimens, and automated gates remotely control the animals' access to the central arena. Second, we describe a precise and reliable real-time EOD timing measurement method from freely swimming WEF. Signal distortions caused by the animal's body movements are corrected by spatial averaging and temporal processing stages. Third, we describe an underwater near-infrared imaging setup to observe unperturbed nocturnal animal behaviors. Infrared light pulses were used to synchronize the timing between the video and the physiological signal over a long recording duration. Our automated tracking software measures the animal's body position and posture reliably in an aquatic scene. In combination, these techniques enable long term observation of spontaneous behavior of freely swimming weakly electric fish in a reliable and precise manner. We believe our method can be similarly applied to the study of other aquatic animals by relating their physiological signals with exploratory or social behaviors.

Introduction

Background. Quantitative experiments on animal behavior (e.g. forced choice, shock avoidance, T-maze, etc.) are typically utilized to investigate specific hypotheses concerning sensory-motor skills, learning and memory formation. However, these restrictive experiments miss much of the richness of natural animal behavior and are likely to result in oversimplified models of the underlying neural basis of behavior. Experiments under more naturalistic conditions are therefore an important complement by which we can explore more fully a species behavioral repertoire. Experiments involving freely moving animals must, however, address unique technical challenges such as movement-induced recording artifacts. Unlike stimulus-evoked responses, spontaneously occurring exploratory behavior cannot be predicted; thus experimental subjects have to be constantly monitored and tracked over an extended period of time. Specific research questions can be best addressed by carefully selected organisms and available technical tools. For example, optical recording and stimulation techniques such as genetically-encoded calcium sensors¹ and optogenetics² have been successfully applied to freely moving genetic model organisms^3-5. Alternatively, miniaturized neural telemetry systems can record and stimulate freely moving small animals^6,7.

Electric fish. WEF species generate electric organ discharges (EODs), which allow them to sense their immediate surroundings or to communicate over greater distances. Temporal patterns of EODs vary under different conditions such as self-movements^8,9, sensory stimuli^10,11, and social interactions^12,13. Pulse-type WEF species produce a train of discrete pulses, as opposed to wave-type species which generate continuous quasi-sinusoidal waveforms. In general, pulse-type species exhibit more variable EOD rate compared to the wave-type species; and animals' EOD rates closely reflect novelty contents of their sensory surroundings^10,14. Pulse-type species can immediately shorten the inter-pulse interval (IPI) within a single pulse cycle in respond to a novel sensory perturbation (novelty response^10,11,14). The ongoing electric behavior of these fish can be perturbed by uncontrolled sensory stimuli from external sources; and different kinds of stimuli such as vibration, sound, electricity, and light are known trigger novelty responses. Therefore, special precautions must be taken to block or attenuate external sensory stimuli during a long-term observation of free-swimming WEF. In this way, changes in EOD rate and movement trajectories can be specifically attributed to stimuli presented by the experimenter.

Aquarium tank and isolation chamber. We therefore placed multiple layers of vibration absorbing materials under a large aquarium tank (2.1 m x 2.1 m x 0.3 m), and surrounded the tank with an insulated enclosure to block external sources of light, electrical noise, sound and heat flux. EOD rate depends on the surrounding temperature^15,16, thus the water temperature was tightly regulated at a tropical range (25±1 °C) for South American WEF species. We constructed a large and shallow (10 cm water depth) tank to observe spatial exploratory behaviors of WEF mainly restricted in two dimensions (Figure 1A). The tank was partitioned into a central arena to observe spatial behaviors, and four corner compartments to separately house individual fish (Figure 1B). Each compartment was built watertight to prevent electrical communication between individuals. Animals' access to the central arena was controlled from the outside by four motorized gates. The gates were placed between the compartments, and they became watertight when locked by nylon wing-nuts. No metallic parts were used underwater since WEF react sensitively to metals.

EOD recording. EODs are generated in a stereotyped manner by activation of single (in Mormyrids) or multiple spatially distributed electric organs (in Gymnotiforms)^17,18. Temporal modulations in the EOD rate can reveal higher-level neural activities, since the medullary pacemaker receives direct neural inputs from higher brain regions such as the diencephalic prepacemaker nucleus, which in turn receives axonal projections from the forebrain¹⁹. However, the EOD timing must be carefully extracted from a raw waveform recording and not biased by the animal's movement-induced distortions. The electric field generated by a WEF can be approximated as a dipole; thus EOD pulse amplitudes at recording electrodes depend on the relative distances and orientations between the animal and the electrodes^8,20. Animal's self-movements change the relative geometry between the animal and the electrodes, thus movements cause the EOD amplitudes at different electrodes to vary over time in a volatile manner (see Figure 2B in Jun et al.⁸). Furthermore, self-movements also change the shape of recorded EOD waveforms, because relative contributions from different set of the electric organs depend on their locations along the body length and their local curvatures introduced by tail bending. The movement-induced distortions in the EOD amplitudes and shapes can lead to inaccurate and unreliable EOD timing measurements. We overcame these problems by spatially averaging multiple EOD waveforms recorded at different locations, and by adding an envelope extraction filter to precisely determine the EOD timing from a free-swimming WEF. In addition, our technique also measures the EOD amplitudes, which indicate whether an animal is resting or actively moving based on the change of the EOD amplitudes over time (see Figures 2E and 2F). We recorded differentially amplified signals from the recording electrode pairs to reduce common-mode noise. Since the EOD pulses are generated at irregular time intervals, the EOD event time-series have a variable sampling rate. The EOD time-series can be converted to a constant sampling rate by interpolation if required by an analytic tool of choice.

Video recording. Although EOD recording can monitor a gross movement activity of an animal, video recording permits direct measurements of an animal's body position and posture. Near-infrared (NIR) illumination (λ = 800~900 nm) permits unperturbed visual observation of freely swimming fish^21,22, since WEFs are most active in darkness and their eyes are not sensitive to NIR spectrum^23,24. Most digital imaging sensors (e.g. CMOS or CCD) can capture NIR spectrum with the wavelength range between 800-900 nm, after removing an infrared (IR) blocking filter²⁵. Certain high-end consumer-grade webcams offer high-definition, wide viewing angle and good low-light sensitivity, which can produce an image quality comparable to, or superior to professional-grade IR cameras available at much greater costs. In addition, certain consumer-grade webcams are bundled with recording software that permits an extended recording duration by compressing video with no quality loss. Most professional-grade cameras offer time synchronization TTL pulse outputs or trigger TTL pulse inputs²⁶ for aligning the timing between the video with the digitized signals, but this feature is generally absent in consumer-grade webcams. However, the timing between a video recording and a signal digitizer can be accurately matched by concurrently capturing a periodically blinking IR LED with the camera and the signal digitizer. The initial and the final IR pulse timing can be used as two time calibration markers for converting the video frame numbers to the signal digitizer time unit and vice versa.

Lighting & background. Image capturing through water can be technically challenging due to light reflections at the water surface. The water surface can act as a mirror to reflect a visual scene above water, and obscure visual features underwater; thus the scene above water must be rendered featureless to prevent visual interference. In order to image the whole aquarium, a camera needs to be placed directly above the water; and it should be hidden behind the ceiling over a small viewing hole to prevent its reflection on the water surface. Moreover, the water surface can produce glares and nonuniform illumination if light sources are incorrectly projected. Indirect illumination can achieve uniform brightness over the whole aquarium by aiming the light sources toward the ceiling, such that the ceiling and the surrounding walls can reflect and diffuse the light rays before reaching the water surface. Choose an IR illuminator that matches a spectral response of the camera (e.g. 850 nm peak wavelength). Electrical noise from the light sources can be minimized by using LED lights and placing their DC power supplies outside of the Faraday cage. Place a white background underneath the tank, since fish contrasts well in a white background at NIR wavelengths. Similarly, use of matte white color on the inner surfaces of the isolation chamber provides uniform and bright background illumination.

Video tracking. After a video recording, an automated image tracking algorithm can measure the animal's body positions and postures over time. The video tracking can be automatically performed by either ready-to-use software (Viewpoint or Ethovision), or user-programmable software (OpenCV or MATLAB Image processing toolbox). As the first step of image tracking, a valid tracking area needs to be defined by drawing a geometric shape to exclude the area outside (masking operation). Next, an animal's image needs to be isolated from the background by subtracting a background image from an image containing the animal. The subtracted image is converted to a binary format by applying an intensity threshold, such that the centroid and the orientation axis can be computed from binary morphological operations. In Gymnotiforms^27-29 and Mormyrids^30-32, the electroreceptor density is the highest near the head region; thus the head position at any moment indicates a location of the highest sensory acuity. The head and tail locations can be automatically determined by applying the image rotation and bounding-box operations. The head and tail ends could be distinguished from one another by manually defining them in the first frame, and by keeping track of their locations from comparing two successive frames.

Protocol

This procedure meets the requirements of the University of Ottawa Animal Care Committee. No conflict of interest is declared. Please refer to the Table of Materials and Reagents for the makes and models of the equipment and materials listed below. Custom written Spike2 and MATLAB scripts, and sample data are provided in the Supplemental File.

1. Aquarium Tank and Isolation Chamber Setup

1. Anti-vibration floor. Construct an anti-vibration surface (2.1 m x 2.1 m) by stacking rubber pads, acoustic Styrofoam, marine plywood panel, and polyurethane foam pads from the bottom to the top (Figure 1A). Lay four wooden studs (5 cm x 10 cm) on the plywood panel to support the edges of the aquarium tank.
2. Floor heater. Lay an electrically shielded heating element over thermally graded foam padding (see Figure 1D bottom). Cover the heating element with a metallic mesh for electrical shielding.
3. Spatial tank. Construct a wide and shallow aquarium tank (1.8 m x 1.8 m x 30 cm) using 1.3 cm thick tempered glass panels, L-shaped aluminum frame and aquarium-grade silicone (see Figure 1A). Cover the underside of the tank with a large sheet of white background to provide high imaging contrast (see Protocol 3).
4. Divide the aquarium tank into a central arena (1.5 m diameter) and four corner compartments (see Figure 1B) by installing walls (22.5 cm tall) made of acrylic sheets (matte white, 0.64 cm thick).
   1. Bend four acrylic sheets (22.5 cm x 102.7 cm) by applying heat to create four curved wall sections, and attach them to the tank bottom using silicone caulk to separate the central arena from the four corner compartments. Leave 20 cm space between the curved sections for the gate installation.
   2. Separate neighboring corner compartments by installing four double walls with 15 cm gaps, which provide extra electrical isolation and places for underwater sensors such as a hydrophone.
5. Assemble four motorized gates, and install them between the corner compartments and the central arena.
   1. Assemble four door frames as shown in Figure 1C. Create six wells (0.64 cm deep) on each door frame, embed nylon acorn nuts (0.64 cm diameter thread) and secure them with epoxy.
   2. Cut four door panels from acrylic and rubber sheets, and create six holes (0.64 cm diameter) on the acrylic and rubber panels for the locking mechanism. Join the acrylic and rubber panels using silicone caulking.
   3. Install acrylic hinges to join the door panels with the door frames.
   4. Mount swinging arms on servomotors, and install them on the top of the door frames (see Figure 1C). Make loops with cable ties to link the swinging arms to the door panels.
   5. Position the gate assemblies on the gaps created between the curved wall sections, and secure them using silicone caulking.
   6. Connect all servomotors to a servo controller, and connect it to a power source and a computer via an active USB extension cable. Test the gates using control software supplied with the servo controller.
   7. After the silicone hardens, check for watertightness by locking all gates with nylon screws and filling one compartment at a time.
6. Isolation chamber. Construct an isolation chamber to surround the aquarium and block external sources of light, sound and electrical noise (see Figure 1D).
   1. Make three wall panels (2 m x 2 m x 5 cm) and four door panels (1.9 m x 0.95 m x 5 cm). For each panel, join aluminum moldings (5 cm x 2.5 cm) to create a rectangular frame; and rivet a white corrugated plastic panel on the aluminum frame. Fill acoustic fiberglass batts in the panels, and close with a black corrugated plastic panel.
   2. Install three wall panels on the anti-vibration floor, and install piano hinges to join the four door panels on the wall panels.
   3. Surround the isolation chamber with aluminum meshes, and ground meshes on all sides to create a Faraday cage.
7. Humidity control. Install a low-noise exhaust fan (Figure 1F top) to remove excess humidity build-up from heating. Place the exhaust fan at least 2 m away from the recording site, and install an air duct between the isolation chamber and the exhaust fan.
8. Routinely monitor and maintain the conditions of the tank water and animals.
   1. Maintain constant water conditions at 10 cm depth, 100 µS/cm conductivity and pH 7.0 by adding water or salt stock solution (refer to Knudsen³³ for the recipe). Add a bag of crushed coral if the pH drops below 6.5.
   2. Install vertical aquarium filters which can operate from shallow water for cleaning and aerating purposes (Figure 1F bottom). Disconnect the filters and take them out of the central arena during recording sessions.
   3. Deliver live mealworms on the bottom of the tank by attaching them on suction cups with elastics. Avoid free-floating preys such as blackworms to prevent uncontrolled feeding of stray preys during recording.

2. EOD Tracking

1. Electrodes installation. Assemble eight graphite electrodes, and space them equally on the curved wall of the central arena.
   1. Obtain drawing leads (15 cm in length; Mars Carbon 2 mm type HB) and shave off the outer coating of the leads.
   2. Cut eight 10 cm segments of coaxial cable (RG-174), wrap the cable core around one end of the graphite rods, and apply heat-shrink tubing over them for strong and stable electrical connection. Attach BNC jack connectors on the opposite ends (Figure 2A left).
   3. Position the electrodes on the wall by taping, and apply thin strips of masking tape on the electrode surfaces to protect from silicone. Apply silicone caulking to permanently hold the electrodes, and remove all tape before the silicone hardens (Figure 2A right).
2. Build eight cable assemblies by measuring the distance from each electrode to the amplifier unit, and cutting coaxial cables (RG-54) in lengths. Attach BNC plug connectors on both ends of the cables.
3. Use the cable assemblies to wire all electrodes to the amplifier unit. Differentially amplify by pairing two 90° oriented electrodes (see Figure 2B), and ground all coaxial shielding wires by connecting them to the Faraday cage.
4. Set the amplifier gain below the signal saturation limit, and apply a band-pass filter (200 Hz-5 kHz) to remove noise. Digitize the four recording electrode pairs at 40 kS/sec.
5. Online signal processing. The instructions are written for the Spike2 software, and the parameter settings are optimized for Gymnotus sp. (see Figure 2C for summary).
   1. Add a DC remove process (τ = 0.1 sec) to all recording channels.
   2. Add a rectify process to all recordings channels.
   3. Create a virtual channel by summing all four recording channels.
   4. Extract a unimodal envelope per EOD pulse by adding a RMS (root-mean-squared, ) process (τ = 0.25 msec) to the virtual channel, for generating a single peak per EOD cycle to unambiguously determine the pulse timing.
   5. Create a realmark channel from the virtual channel and record the time and values of the peak amplitudes, after setting an appropriate threshold to capture all EOD pulses without missing a pulse, while avoiding false positives.
   6. Monitor the instantaneous EOD rate in real-time by setting the channel display option of the realmark channel to an instantaneous frequency mode.
   7. Monitor the fish movement in real-time by duplicating the realmark channel, and set the display option to a waveform mode.
   8. Quantify an activity level from the RMS of the EOD amplitude slope by creating a virtual channel from the realmark channel (0.01 sec sampling period), and add slope (τ = 0.25 msec) and RMS (τ = 0.5 msec) processes.
   9. Export the realmark channel in the Spike2 software to the MATLAB format.

3. Synchronized Video Tracking

1. Create a background scene.
   1. Hide any object that casts a reflection on the water surface by covering with matte white countertop film.
   2. Install a matte white corrugated plastic panel 15 cm below the ceiling to hide the camera and the air vent.
   3. Print grid patterns on a large sheet of white paper for calibrating a camera, and lay it underneath the tank to provide a high-contrast background.
2. Install light sources.
   1. Obtain IR LED lights and, remove built-in fans to reduce noise. Drive the LED with a current-regulated DC power supply placed outside of the Faraday cage.
   2. Install IR LED lights for imaging in darkness, and white LED lights for driving a diurnal light cycle in the test fish. Direct all light sources toward the ceiling to achieve indirect and uniform illumination (Figure 3A).
   3. Regulate the diurnal light cycle by driving the white LED lights with a timer-controlled switch (e.g. 12 hr on/12 hr off).
3. Install a camera directly above the aquarium.
   1. Obtain a NIR-sensitive camera, or remove an IR blocking filter by breaking a thin sheet of tinted glass at the back of the lens assembly. Make sure the viewing angle is wide enough to image the whole central arena.
   2. Make a small viewing hole in the middle of the ceiling panel, and place the camera directly above the hole.
   3. Install a white ring guard around the lens if the light sources generate glares.
4. Make a time-synchronized video recording.
   1. Place an IR LED at one of the four tank corners to generate time synchronization pulses (1 msec duration, 10 sec period). Add a load-limiting resistor (1 kΩ) in series, and drive the IR LED from a digital output port of the digitizer hardware.
   2. Use video recording software bundled with the camera if available. Select the highest recording quality (e.g. lossless compression) and the highest resolutions supported.
   3. Start the video recording immediately before starting the EOD recording, and stop the video recording immediately after the EOD recording.
   4. After the recording, convert the image frame numbers to the digitizer time unit by linearly interpolating between the first and the last light pulses captured by the signal digitizer and the video recording.
5. Automated image tracking
   The instructions are written for the MATLAB Image processing toolbox, and make use of its functions. A custom MATLAB script is provided with this submission for automated image tracking. 
   1. Import video. Import a video recording file directly to the MATLAB workspace using "Videoreader.read" function.
   2. Create a composite background image by combining two image frames. Replace the image region occupied by an animal with an unoccupied image of the same region from another frame (see Figure 3B).
   3. Specify an image region to track by drawing a circular mask around the central arena to exclude the area outside (Figure 3B bottom), and multiply by a constant (r_int) to set a minimum threshold for intensity difference. For example, setting rint = 0.85 will suppress the intensity fluctuations 15% = (1 - r_int) below the background.
   4. Image subtraction. Subtract an image frame (=IM_k) from the background image (=IM­₀) to obtain the difference image (=ΔIM_k). Use unsigned integer numerical precision to store the image intensity values as non-negative integers.
   5. Segment the difference image by applying an intensity threshold determined from the graythresh function. Clean the binary image using the bwmorph function, and select the largest blob corresponding to an animal after calculating all blob areas using the regionprops function.
   6. Determine the centroid and major orientation axis of the largest blob by applying the regionprops function, and rotate the image to align the major axis with the x-axis. Divide the image to the head and tail parts at the centroid (Figure 3D top).
   7. Determine the major axis of the head part, and rotate the entire image to align with the x-axis (Figure 3D bottom left). Fit bounding-boxes around the head and tail parts parallel to their major axes using the regionprops function.
   8. Determine the median y-coordinates of the blob at the left, center and right vertical edges of the bounding boxes (green dots in Figure 3D bottom); and assign them to five feature points (head-tip, mid-head, mid-body, mid-tail, tail-tip).
   9. Process successive frames after cropping an image frame centered at the animal's centroid determined from its previous frame.
   10. Manually assign the head orientation for the first frame, and use a dot-product between the orientation vectors from two successive frames to automatically determine the head orientation. Inspect the result, and manually flip the head orientation if incorrectly assigned.
6. Plot an animal trajectory by joining the head-tips, and smooth using median and average filters (n=3) if it has a jittery appearance. Superimpose the trajectory with a background image, and interpolate fish midlines using the five feature points (see Figure 2E).
7. Compute the average EOD rate at each image capture time by resampling the instantaneous EOD rate (100 Hz sampling rate) and averaging (0.0625 sec time window). Plot the trajectory in pseudo-colors determined from the time-matched EOD rate, and superimpose with a background image (see Figure 2F).

Results

EOD tracking results

The recorded EOD waveforms from different electrode pairs varied in amplitudes and shapes as expected from their unique positions and orientations (Figure 2C top). The use of multiple electrode pairs ensured strong signal reception at all possible positions and orientations of WEF within the tank. The envelope waveform (Figure 2C bottom, green trace) always contained a single peak per EOD cycle, which served as a reliable time marker for prec...

Discussion

Significance of our techniques. In summary, we first described the construction of a large aquarium tank and an isolation chamber to observe spontaneous exploratory behaviors produced by WEF. Next, we demonstrated the technique of recording and tracking the EOD rate and the movement states from unrestrained fish in real-time using multiple electrode pairs. Finally, we described the infrared video recording technique through water in a time-synchronized manner, and the image tracking algorithm to measure the body...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Aquarium Construction
Electrically shielded floor heater	ThermoSoft Corp., IL, USA	ThermoTile	www.thermosoft.com
Tempered glass panel	generic		0.5 in thick, used for the aquarium construction
Aquarium grade silicone	generic
Acrylic sheet	generic		0.25 in thick, matte white
Natural rubber sheet	generic		0.25 in thick
Servomotor	HTECHRCD Inc., Korea	HS-325HB, 180deg rotation	www.servocity.com
Servomotor arm mount	HITECHRCD Inc., Korea	56362 Large Spline	www.servocity.com
Servomotor controller (6 channels)	Sparkfun	ROB-09664	Micro Maestro 6-channel USB Servo Controller
Active USB extension cable	C2G	38990	12 m USB 2.0 A Male to A Female 4-Port Active Extension Cable
Exhaust fan	Nutone	ILFK120	www.homedepot.com
Vertical aquarium filter	Tetra, Germany	Whisper Internal Power Filter - 40i
Crushed coral			Used to increase the pH of the tank water
EOD Recording Setup
Graphite Electrodes	Staedtler, Germany	Mars Carbon 2-mm type HB	Shave the outer coating
Physiological Amplifier/Filter	Intronix, Canada	2015F
Coaxial Cable	generic	RG174	For electrodes assembly
Coaxial Cable	generic	RG54	For wiring use
BNC jack connector for RG-174	Amphenol Connex	112160	For electrodes assembly
BNC plug connector for RG-54	Amphenol Connex	112116	For wiring use
Signal digitizer hardware	Cambridge Electronic Design, UK	Power MKII 1401
Signal digitizer software	Cambridge Electronic Design, UK	Spike 2. ver 7
Visual Tracking Setup
White LED light	IKEA, Sweden	DIODER 201.194.18	www.ikea.com
Infrared LED light (850 nm)	Scene Electronics, China	S8100-60-B/C-IR	Remove built-in fan
USB webcam	Logitech Inc., CA, USA	C910	Remove Infrared blocking filter
Motorized camera	Logitech Inc., CA, USA	Quickcam Orbit	Remove Infrared blocking filter
Video recording software	Logitech Inc., CA, USA	Logitech Quickcam Software	Download from www.logitech.com
MATLAB	Mathworks, MA, USA	2012a	Image processing toolbox