High speed video capture and key point detection tracking with Ramona’s Multi-Camera Array Microscope provides startle-response analysis of larval zebrafish over an entire well plate, lending a new perspective to fields such as anxiety, epilepsy, development and regeneration while accelerating throughput.

Summary

Zebrafish (Danio rerio) are commonly used as a model organism for behavioral studies aiming to elucidate patterns of movement that correlate to specific stimuli or developmental states. These studies tend to observe young fish (>5dpf) in multi-well plates. Traditionally this type of experiment is either laboriously executed using a conventional microscope to image one fish at a time, limiting throughput of experiments, or with a low resolution system, limiting the behavioral measurements that can be captured. Here, we propose and demonstrate how one can use a novel Multi-Camera-Array-Microscope (MCAM™) system to record and stimulate zebrafish, through optical flashes or mechanical vibration, in standard well plate configurations including, 96-, 48-, and 24-well at both high spatial-temporal resolutions, and high throughput.

Introduction

Zebrafish are commonly used in toxicology screening, but their quick, tiny movements are easy to miss and even more challenging to record. With the capability to record video assays at up to 180fps, and 8-point skeletal tracking and analysis software, the Multi-Camera Array Microscope can clarify these movements and provide scientists with better data. This experiment was conducted using the workflows of our partners at Dr. Robin Tanguay’s lab at Oregon State University.

Challenges

Behavior and movement studies using zebrafish require observations across a large number of larvae with microscopes, typically at low magnifications (1x — 4x), to relate experimental conditions to the behavior of 3 to 7 day post-egg fertilization (dpf) larvae. This research is arduous and time-sensitive, with data collection alone being a multi-hour process of capturing one well at a time. This ordeal can over-expose the fish to light and radiation from the microscope, which can alter their behavior and ultimately skew results.

Additional challenges to machine learning include the wide range in shapes and styles of well plates currently being used.

Methods

Zebrafish were nurtured to 4dpf for use in this experiment and were then placed in a 48-well plate and loaded into the MCAM Kestrel to begin a synchronous video acquisition of the entire plate. Ramona’s MCAM Kestrel includes vibrational startle as a built-in feature, and the fish were observed unbothered until 2 seconds, when a 300 Hz vibrational startle occurred with 200 millisecond duration. A 5-second video of the experiment was recorded at 120 frames per second. Subsequent experiments routinely monitor zebrafish motion tracking at 30 to 180 fps for up to 4 hours depending on desired imaging parameters.

Videos of zebrafish larvae in these configurations can be captured, simultaneously, at up to 180 frames per second. The videos are then processed automatically using a custom convolutional key-point detection algorithm and yield high precision coordinates of eight key-points localized in cartesian space for each frame. Results are reported in spreadsheet format (CSV files) for synthesis and further analysis.

Utilizing convolutional key-point detection, we localized eight defined points in cartesian space with high accuracy and precision. These points are the snout, eyes, fish center, and four points along the tail, which we have found yield a complete representation of the fish movement.

Results

We were able to reach several experimental endpoints from this sample, including the keypoint tracking and derived skeletal visualization, region of interest analysis, thigmotaxis observation, total distance traveled over time. Key-point cartesian coordinates are reported to a CSV file output which is automatically analyzed and processed as well as provided to the user for further analysis.

In summation, we were able to perform end-to-end startle response tracking analysis for a 5 second video across an entire 24-well plate in less than 3 minutes including video acquisition, well extraction, convolutional key-point detection and analysis, increasing throughput and the kinematic information available to researchers. The tracking and analysis software is standardized and compatible with 24-, 48-, or 96-wells.

Watch the entire video below.

‍

‍

‍

‍

‍