Computer assisted analysis to improve throughput and precision of knockdown time assays

Physiological and behavioral assays in insects often rely on direct observation by investigators, which can reduce throughput and introduce bias. Knockdown time is commonly used in measurements of thermal stress and insecticide efficacy, but these measurements can be labor intensive for investigators. Here, we develop a novel computer assisted method to analyze results of a high-throughput time-to-knockdown (TKD) assay used to assess the heat tolerance of insects. Adult Drosophila melanogaster were held at 37.5°C in 96-well plates, and flies were video-recorded until movement ceased. To estimate knockdown time from the video files, video frames were transformed into a numerical response of pixel change rate using several custom scripts, and these results were compared with manual observation of video files by investigators. TKD estimates from computational methods presented reduced standard deviations relative to human inspected estimates, although computer-generated TKD estimates tended to be shorter than human-derived estimates. To reconcile this discrepancy, we examined videos in further detail and observed that individual flies had diverse activity profiles in the time immediately prior to knockdown, indicating that there is complex variation in individual responses to heat stress that can confound visual estimates of TKD. Despite the point estimate differences, rank order and significant differences between experimental groups were consistent among methods. However, whether the lack of consensus between human and computational methods reflects a true biological phenomenon or is an artifact requires further evaluation of diverse datasets.