Automated Fly Tracking System (AFT)

Caroline Linehan, Chris Rogers, Alan Koplin, & Isabelle Draper

A joint project between the Molecular Pharmacology Research Center at New England Medical Center and Tufts University.

Funded by The National Science Foundation

Goal:

The goal of this project is to develop and optimize a system enabling repeatable, high throughput screening of large collections of genetically and/or pharmacologically manipulated flies to identify genes regulating locomotor activity. The instrumentation allows flies to travel in two dimensions and reduces the restriction on fly motion.

Background:

1-D systems: Currently, most fly locomotion measurements are done by counting the number of times a fly crosses a light gate using simply a light source and photo-sensitive sensor. The fly is constrained to move in essentially a line in a small vial. This has the advantage that the measurements are simple and one can track numerous flies at a time (typically 64 flies). The disadvantage of these systems is that the fly’s motion is highly constrained and therefore may affect the measurements. Further, it is possible for the fly to walk a lot on only one side of the tube and appear to be stationary since it never crossed the light gate.

AFT System: Experimental Setup:

2-D systems: If the fly is instead constrained to an area, one can optically track the motion of the fly and get a better measure of the total distance covered by the fly during a given time period. Not only does this allow the fly more freedom of motion, but it also allows one to measure a larger number of parameters:

1. total distance traveled by the fly (cm),
2. fly velocity (average and instantaneous (cm/s)),
3. percent of time active
4. the amount of time the fly spends in the middle of the cylinder versus the edge

In order to transition to the 2-D system, the AFT system has been developed. A fly cassette was designed to isolate 24 flies and constrain them to walk in 2.5 cm diameter, 1 cm high cylinders (Figure 1). At the base of each cylinder there is a 1.5% Agar layer to keep the fly hydrated during the test run for 1-12 hours. Each cylinder is covered on the top by a fine mesh that allows the humidity from the Agar to escape but not the fly.

Figure 1 Fly Cassette

The fly cassette is loaded with flies (sedated with CO2) and placed on top of a Stocker Yale 12” x 18” high-frequency fluorescent light table, causing the flies to appear white on a black background in the camera view (Figure 2). The camera is mounted 18 cm above the cassette and light table with a custom fixture (Figure 3 ). The camera (a Kodak ES1) records a 1000x1000 pixel image (8 bit linear) once a second for 1-12 hours. These images are then post-processed to identify the individual fly locations.

Figure 2 Camera View

Figure 3 AFT System: the overall experimental setup including (1) a computer monitoring program, (2) 8-bit high resolution camera to track fly movement, (3) high frequency fluorescent light panel to illuminate fly cassette, and (4) 24-well fly cassette containing one fly per well.

The post-processing is fairly simple. First, the background image is created by erasing the area where the fly is replacing it with the average pixel value as the rest of the well. The background image is then subtracted, to isolate the fly, and then the image is broken up into 24 sub-images. Each sub-image is individually converted into a binary image with a custom threshold value set at the onset of the data reduction. From the binary sub-image, all possible fly areas are identified. Typically there is only one area identified and therefore that would be the fly – but in cases of multiple areas identified in the sub-image as potential flies, a weighting factor is applied to each area based on the area size (80%), distance from the position of the fly in the last image (10%), and the resulting fly velocity (10%). These weighting values were determined empirically for the least number of fly misidentifications.

Once the fly is identified, x- and y-positions are generated for each of the 24 flies with respect to time. With these positions, the four parameters mentioned before: total distance traveled, fly velocity, percent of time active, and the amount of time the fly spends in the middle of the cylinder versus the edge can be investigated. Further, fly locomotion as a function of genetic makeup can be measured.

Results:

Figure 2 Camera View

Future Work:

We are now working on a number of improvements to the system. First, we are developing an automated system to remove the background image. This will provide more accurate tracking and increase the speed of locomotion analysis. Second, the current racking system will slightly over-predict the fly motion because of small inconsistencies in the thresholding. If the lighting changes subtly or the fly turns on its side, the position of the fly can change by 1 pixel (.18mm). This will only affect the data if this digital jitter continues to accrue over thousands of images and can be removed through an anti-jitter filter that requires the fly to move more than one half of the fly length (~.15cm) before the distance is counted. Third, we will sometimes lose the fly at the edge of the enclosure. We are looking at ways of reducing the glare at the edge so that we can better detect the fly.

During the next 3 months we will be screening 100 different fly lines with the goal of identifying a hyperactive fly (one that travels at least twice the normal distance traveled). We will be performing an analysis with each fly line on the four parameters previously described to determine which parameters provide the most focused results. We also hope to improve the signal to noise ratio through optimizing the process for building the background image and determine if environmental stress from the AFT system leads to hyperactivity in older flies.

Related Literature:
1. Balch T., Khan Z., Veloso M.; Automatically Tracking and Analyzing Behavior of Live Insect Colonies. Carnegie Mellon University, Pittsburg PA.

2. Estivill-Castro V. et al.; Tracking Bees – A 3D, Outdoor Small Object Environment. Griffith University, Australia.

3. Noldus, L.P.J.J.; Spink, A.J.; Tegelenbosch, R.A.J. (2001). EthoVision: a versatile video tracking system for automation of behavioral experiments. Behavior Research Methods, Instruments & Computers, 33, 398-414.

4. Colazza, S.; Clemente, A.; Rosi, M. (1995). Locomotor activity in parasitoid insects measured by EthoVision: the key to better pest control. Noldus News, 2 (1).

5. Gilbert, P. et al.; Locomotor Performance of Drosophila Melanogaster: Interactions Among Developmental and Adult Temperatures, Age, and Geography. Evolution, 55(1) pp.205-209, 2001.

6. Cook, R.: The Courtship Tracking of Drosophila Melanogaster. Biol. Cybernetics 3491-106 (1979)

7. Groszmann, Daniel; Stereoscopic Measurements of Particle Dispersion in Microgravity Turbulent Flow. Tufts University 2001.

Associated Web Links:
http://www.nemc.org/mcri/MPRC/mprc.htm --Molecular Pharmacology Research Center
http://ase.tufts.edu/roboticsacademy/Projects/Fruit%20Fly/index.html --Robotics Academy associated research

This material is based upon work supported by the National Science Foundation under Grant No. 0212046. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation (NSF).