Signal Inference: A Method to Locate and Track Calcium Waves with Statistical Confidence

Optical imaging studies of cell signaling pose a recurring challenge to track signals relative to the topology of the cell structures. Image analysis platforms require the user to visually assess which signals merit further study and define them as geometric regions of interest (ROIs) ‐ throwing away information outside the ROI. Tools for segmentation, the process of identifying cellular structures in an image, let the user subjectively identify biological structures with pixel resolution. Thus, the experimentalist is challenged to objectively control for observer bias. We have developed a method to avoid manual ROI generation and use the entire data set to extract the physiology of calcium (Ca 2+ ) signaling within and between vascular endothelial cells. We hypothesize that signal tracking and segmentation could be made rigorous and semi‐automatic by requiring signals to be continuous in time and contiguous in space. This reduces the analysis of the pixel covariance matrix to tracking contiguous 2D foci with high covariance. We test if this is sufficient to 1) identify cell structure; 2) track Ca 2+ signals; and 3) provide statistical confidence metrics. The algorithm was developed using data from confocal fluorescence imaging of cytosolic Ca 2+ release and reuptake in freshly isolated endothelial cell tubes following local stimulation with acetylcholine (ACh) from a micropipette (1 μm tip, 1–2s). Results were tested with biological replicates under varied technical (scan rate, gain, amplification) and physiological conditions (stimulus dosage, location) and compared to manual segmentation and tracking. We illustrate a novel method for robust signal tracking with a minimum of user settings under different dose and gain settings. The method segmented cells by identifying foci with high spatial correlation. Individual foci overlap 78.9±18.1% of the area of manually segmented cell (n=5). Foci shape was also reproducible across different conditions (high/low gain for two ACh dosages). The foci boundaries for one preparation under four acquisition conditions were within 2.2 μm along 70.8% of the boundary (5 cells). This approach provided multiple metrics for each time course including the number, size, shape, rate and behavior of detected optical signals and cell structures. Typical run time for a single dataset (256 × 1024 pixels and 1054 frames) was less than 30 minutes on a (16GB RAM) computer. Support or Funding Information NIH F32HL107050, K01AG041208, R37HL041026