The new genetic tools developed by the Rubin Lab and imaged by the FlyLight Project (GAL4 lines) allow very small collections of neurons in the fly brain to be reproducibly targeted with neuronal activators and inactivators. Thus, powerful behavioral experiments are enabled where neurons are acutely and reversibly activated or inactivated to assess the consequences on selective behaviors. Most labs using this behavioral neurogenetic approach focus on a particular behavior or a particular region of the brain. We believe in the power and efficiency of a comparative approach and so our strategy is different.
The Fly Olympiad uses a collection of high-quality and high-throughput primary behavioral assays to screen a collection of thousands of GAL4 lines driving expression of activators or inactivators in many brain regions and neuronal combinations. Our primary screens focus on locomotion, vision, coordination, aggression, reproductive success, sleep and circadian rhythms, social interactions, gravitaxis, and an open field observation assay. In secondary screens, the development of complex higher-level behavioral assays is underway to elucidate a more complete understanding of the brain regions and individual neurons that are involved in producing particular behaviors. The more detailed secondary assays are being carried out by Janelia Scientists, Post Doctoral Fellows and Visiting Scientists. We have the potential to discover that some well-studied brain regions are involved in the control of novel behaviors as well as to assign functions to anatomical “terra incognita.” Furthermore, the Rubin collection of GAL4 lines will allow us to generate structure-function maps to determine which parts of the brain are associated with and possibly govern particular behaviors.
The matrix we are populating includes the detailed anatomical characterization of the neural expression patterns of the GAL4 lines and the quantitative analysis of multiple behaviors, generated as part of the Fly Olympiad Team Project. We expect to be able to mine this matrix for correlations between anatomy and behavior using tools developed by our computational colleagues. Correlations between anatomy and function will be computationally assessed and the most promising hypotheses will be tested with lower-throughput secondary, or more refined, behavioral assays. It is likely that new strategies will be required to narrow the GLA4 expression patterns further to make these correlations convincing. We will pursue this type of more detailed analysis on the best candidates from the primary screen.
We believe that a large-scale organized, controlled screen of this scale will provide powerful insight into basic questions in neuroscience and will be an invaluable resource to the fly community.
Accomplishments and Progress
Since the Olympiad Team project began in earnest in the Winter of 2009, we have made significant progress towards our goal of creating and conducting a series of behavioral screens. Specifically we have:
- Established several high-quality behavioral assays (eight total) that broadly cover the locomotor and sensory modalities. All assays we develop are carefully documented with detailed engineering drawings and experimental protocols.
- Worked collaboratively with Janelia Shared Resources to develop the infrastructure necessary for a project of this scale.
- Prototyped and installed temperature- and humidity-controlled environmental rooms for behavioral experiments.
- Implemented a screening pipeline with a reasonable series of rapid quality control checks.
- Employed software systems for video compression, tracking the movement of flies, and behavior analyses that address current assays and can easily accommodate future ones.
- Developed a robust metadata interface to track environmental and nonexperimental data.
- Designed a statistical analysis strategy to identify candidates—lines with behavioral metrics that are different from controls.
- Completed several pilot experiments using Rubin Lab GAL4 lines as well as know mutant lines and used the resulting data sets to refine behavior protocols, assess screening strategies, and test methods for correlating anatomy and behavior.
- Initiated a large-scale screen capable of screening at a rate of 75 lines/week of the Rubin Lab GAL4 collection using “The Box”, DAM, Gap Crossing, Aggression, open-field locomotion screen (FlyBowl), an observational activation screen, and a sterility assay.
The first and most complex behavioral apparatus we chose to develop, known as “The Box”, assesses basic locomotion, response to a mechanical startle, motion vision, photototaxis, and UV-green color preference. This initial locomotion and visual response assay borrows salient features from published behavioral assays (Seymour Benzer, Ulrike Heberlein, Mark Frye, Tom Clendenin). The Fly Olympiad members worked closely with our colleagues in the Instrument Design and Fabrication Group at JFRC to produce this quantitative behavioral assay. The experimental protocol is controlled by PC software that communicates with a microprocessor controller that presents stimuli with millisecond precision to six nearly flat tubes containing 15 flies each. The protocol is initially run at 24°C (the permissive temperature), and then the temperature is increased to 34°C (the restrictive temperature for UAS-Shibirets1) and the same sequences are run on the same population of flies.
The example data shown in the figure is a summary of the results from a single run of a tube of 15 male flies from our experimental control line, pBDPGal4U (empty GAL4 at attP2)/w+;;UAS-Shibirets1 in black relative to a Rubin Lab GAL4 line (in red). Sequence 2 consists of a series of mechanical startles for which the average locomotor response is shown. Sequence 3 is summarized with a tuning curve for the flies’ optomotor response to visual motion at different temporal frequencies delivered using a linear array of LEDs along the sides of the tubes. Control flies respond most strongly to motion between 8 and 20 Hz, and have a significantly attenuated response to motion at 42 Hz. The GAL4 line shown is considered ‘motion blind’ because when a neural inactivator is driven in this line, they stop responding to motion. As seen in Sequence 4, however, these flies aren’t fully blind. This sequence plots the displacement of flies toward green or UV light at two intensities (a low and a high level) presented at one end of the tubes. Flies are strongly attracted to the light source and are slightly more attracted to the higher intensity of light. Sequence 5 is a UV vs. green preference test (a wavelength-dependent behavior, a simple form of color vision). The animals’ preference is revealed by the average displacement of the population of flies, with 0 indicating no preference. Control lines typically start to prefer UV around intensity 10 (arbitrary units, but on a linear intensity scale).
As part of the large-scale effort at Janelia to understand the function of the Drosophila melanogaster nervous system through correlation of high-throughput behavioral (Fly Olympiad) and neuroanatomical (FlyLight) studies, we have combined the tools of FlyBowl (Simon and Dickinson, 2010) and Ctrax (Branson, et al., 2009) to create a high-throughput behavioral screen of fruit fly locomotor and social behaviors. FlyBowl, a chamber design developed by Simon and Dickinson (2010) to enhance the automated tracking, has been modified to increase throughput and consistency of image quality, which allows unsupervised use of an updated Ctrax tracking algorithm. We recorded simultaneously from 8 bowls containing 20 flies each for 1000 secs. In order to reduce disk storage, we developed a MATLAB-based data capture system which reduces image file size by a factor of 80 during recording and is loss-less for the tracking algorithm. The data analysis pipeline moves the data from recording computers to a cluster and performs quality checks, tracking, spatial and temporal data registration, sex classification, and statistical behavior analysis. This allowed processing of 116 videos in the same 24-hour period as data collection. To provide oversight for such large data sets, we have been developing visualization tools for examining the stability of experimental conditions over time and the suite of statistics generated by the data analysis pipeline. We are currently performing a neural activation screen on lines from the Rubin GAL4 collection by driving a genetically encoded dTRPA1 channel, which depolarizes the neurons in response to high temperature (29 degrees C and 50% RH). A pilot screen demonstrated the feasibility of our statistical analysis. We saw significant differences in our metrics for locomotor and social behaviors, which recapitulate human annotation. The development of this assay pipeline from data collection through automated analysis allows for the rapid generation of quantitative descriptions of behavior changes due to sparse neural activation. In combination with other behavioral assays and the anatomical annotation of the Rubin GAL4 collection, our goal as part of the Fly Olympiad project is to derive a functional map of the fruit fly brain.
We have developed a high-throughput assay to detect aggression and are conducting a neuronal activation screen using Drosophila UAS-dTrpA1 (Hamada et al., 2008) to identify neurons that induce aggressive behaviors. The original automated aggression assay (Dankert et al., 2009) was optimized for assaying two pairs of flies at a time. The throughput of the screen is limited by the number of experiments that can be run per day and the processing time for automated behavior scoring. We have modified both the hardware and software to produce a system capable of assaying 12 pairs of flies at once, which utilizes the Janelia computer cluster for parallel processing of automated behavior scoring. Using four of these rigs, we can currently assay aggressive and courtship interactions for 240 pairs of males per day, with the total processing time for behavior analysis cut from 27 hours to 2 hours. Based on our preliminary screen, a minimum of 12 pairs of flies is required to test each GAL4 line; thus, our current throughput is 20 lines/day or 100 lines/week.
For flies to successfully move across the terrestrial environment when walking, they need to integrate mechanosensory and visual cues to navigate complex terrain. In additional to their impressive aerial abilities, many species of flies, including Drosophila, are agile climbers, able to traverse gaps wider than their body length. Flies estimate gap size visually and then execute a complex motor program to cross. The Gap Crossing setup is a population assay based on the successful single fly paradigm developed by Pick and Strauss (2005), designed to assess the ability of flies to cross obstacles. The higher-throughput experiment, designed by Roland Strauss in cooperation with Janelia’s Instrument Design group consists of a series of five concentric gaps, increasing in width from 2 mm to 4 mm, filled with water plus detergent to prevent the flies from walking through the gaps instead of climbing. The arena is covered with a coated (slippery) glass plate. About 15 anesthetized flies are introduced into a shallow bowl in the center of the arena. After the flies revive, a motor automatically raises the lid to a height allowing the flies to leave the center and explore the arena but preventing jumping or flying. To raise the activity of the flies, a startle with a pager motor is given immediately before the lid rises. The behavior is recorded for 10 min. A tracking system records the position of the flies once per second and the number of flies on each ring is calculated as well as the number of flies that drowned while unsuccessfully attempting to cross a gap.
Test runs of the Gap Crossing assay have established that this system automatically detects (1) fly strains that get stuck at a certain gap width; (2) fly strains with motor defects, which attempt to climb but drown; and (3) fly strains whose movement patterns differed from wild type. Pilot screens have identified GAL4 lines that produce expected and novel phenotypic classes.
Circadian Rhythm and Sleep Assay
The commercially available Drosophila Activity Monitoring (DAM) system from TriKinetics is the field standard assay for circadian rhythm and sleep research. The apparatus consists of individual flies in small tubes with an infrared beam cross detection to monitor activity over many days. We set up an array and established a 12 day elevated temperature protocol where the flies are entrained to a light dark cycle for four days and then allowed to free-run in constant darkness for eight days. Commercial data analysis software was determined to be inadequate; an in-house version was written and will shortly be distributed under an open source license. We compare the total amount of sleep and sleep throughout the day-night cycle using the data from the last day of entrainment and the last day of free running. We also identify circadian periods in the activity levels. This is one of our primary assays and we are screening at a rate of as many as 75 GAL4 lines a week crossed to UAS-Shibirets1. This assay can be used to verify locomotor abnormalities detected in “The Box” and other locomotor assays in The Olympiad.
Courting is a complex behavioral ritual in Drosophila: males identify an appropriate partner, orient toward her, follow her, sing a courtship song, tap and lick her before attempting copulation (Hall, 1994). We currently perform a sterility assay to enrich for males that may be defective in aspects of this behavior. Five GAL4/ UAS-Shibirets1 males are preheated to the non-permissive temperature for more than 1 hour and then allowed to interact with control virgin females for two hours. The males are then discarded. The vials containing the females are returned to room temperature and later scored for larva and adult progeny. Lines identified as sterile will be evaluated in secondary assays to look at courtship and song production.
Because it is hard to find what we don’t look for, we have developed a human scored Observation Assay based on a pilot program developed by Clair McKeller in Julie Simpson’s laboratory. Multiple trained scientists observe fly behavior when neurons are activated with UAS-dTrpA1. We use a controlled vocabulary that captures up to 56 behavioral phenotypes as well as a video record for documentation and further analysis.
Data Analysis Pipeline
Janelia’s Scientific Computing group is working with us to implement a data analysis pipeline allowing continuous monitoring of the technical and biological performance of the assays. There are two sources of failure, apparatus and animal, which we need to monitor. Again we use “The Box” to illustrate how we currently manage these issues, and as an example for subsequent assays. In “The Box” we monitor a number of equipment status and performance metrics (red diamonds in figure) and use these to verify that an experiment was properly run. If it was not, we schedule the same cross to be re-run. If the flies are significantly inactive in an experiment (especially at the permissive temperature), it is possible, though rare, that this is an accurate reflection of the behavioral vigor of a particular cross, although it is more likely that these flies are simply from a bad batch. These flies will also be re-scheduled to run, but from a new cross. The advantage of using UAS-Shibirets1 as our initial inactivating effector is that the behavior of the flies at the permissive temperate serves as a control and alerts us to problematic animals. Should the following tests with flies from an independent cross appear to be inactive when tested in “The Box”, we label these flies as ‘inactive’ and remove them from the list of lines to be run in subsequent assays. Time permitting, we will examine these lines with an alternative effector. We aim for a re-test rate of 10%, although currently this is approximately 20%. We have used what we learned developing “The Box” to improve subsequent assay pipelines. The video compression, tracking, data analysis, and database insertion pipeline established for “The Box” is sufficiently generic to accommodate most other assays used in the Olympiad.
How the Fly Olympiad Operates
The Fly Olympiad operates as a large-scale team project run by a project scientist Wyatt Korff that is modeled as a biotech startup. The day-to-day work is accomplished by a core group of highly skilled research technicians and research specialists. Members of the team have complimentary research backgrounds in Neuroscience, Biomechanics, Engineering, Ethology, and Drosophila Molecular Biology. In addition, we have a number of post-doctoral fellows that are associated with the Janelia Visitor Program who have brought their own research program to Janelia and participate in the Fly Olympiad. We get substantial guidance from the Fly Olympiad Steering Committee consisting of Janelia Lab Heads whose research programs compliment the core mission of the Fly Olympiad as well as members with Operational and Process Engineering expertise that allows us to scale. In addition to these core members we work with an agile group of technicians, engineers, software engineers and scientists to prototype, run and scale large quantitative behavioral assays. We get considerable support from the Janelia Shared Resource and from the Scientific Computing groups. We feel that a project of this scale not only benefits from the synergy of such a diverse group, but would not be possible without such help.
1. Simon JC, Dickinson MH (2010) A new chamber for studying the behavior of Drosophila. PLoS ONE 5:e8793. doi:10.1371/journal.pone.0008793.
2. Branson K, Robie AA, Bender J, Perona P, Dickinson MH (2009) High-throughput ethomics in large groups of Drosophila. Nature Methods 6:451–457.