Image Analysis
Tracking of colony areas over time used three main steps using custom software in Matlab. First, the initial colony locations were detected. Then, the final colony areas were measured and merged colonies were separated. Finally, using the outlines of the final colony areas, the area of each colony was tracked through time. We measured the growth rate of colonies on Petri dishes via the diameter of a hypothetical flat circular colony with same area(Wimpenny, 1979). The initial growth rate was calculated by regressing diameter over time for the first four hours (12 frames) once growth was detected.
To detect initial colony locations, we applied a peak detection algorithm to a time-projection of the image stack: We worked with the “V” channel of the image in HSV colorspace, which we heuristically found to work best. First we determined a colonythreshold by subtracting a blurred first frame from a blurred final frame, and taking Otsu’s threshold on the result. Blurring was a Gaussian blur with a 1 pixel standard deviation applied twice. To create the time projection, each frame in the time series was blurred, first-frame subtracted, thresholded, and added to the projection resulting in an image where intensity peaks corresponded to locations where the plate had been colonized. Within this time projection, colony initial locations were detected by finding non-adjacent regional maxima. We found heuristically that this approach was more accurate than peak detection using just the final image, or by looking for increases in color early in the time lapses. At this point mis- or unidentified colonies could be manually removed or added, though this was rarely required.
The final colony areas were next calculated in the projection by finding pixels above the threshold in the final image that were contiguous with an initial colony location. If only a single initial colony location was within a contiguous colony region, the entire region was assigned to that colony. However, if more than one initial colony locations were within a colony region, we concluded that colonies had merged in the experiment. To find the boundary between two or more merged colonies, we applied a distance transformation to the multi-colony region then performed a marker-controlled watershed on this transformed image, with the colony locations as the markers.
Finally, tracking areas over time was done by counting the pixels above the threshold within a colony’s boundary at each time point, using the “L” channel of LAB colorspace on frames which had been blurred, first-frame subtracted, and thresholded with a value of 5. This low threshold (as opposed to Otsu’s threshold, used above) was used here because the higher sensitivity was heuristically determined to be better at identifying early stages of growth, even though the lower sensitivity Otsu’s threshold was better at generating a time-projection for finding initial colony locations.
Quantification of acidity was done using the pH indicator bromothymol blue, which is dark green at neutral pH and yellows as it acidifies. To quantify the relative acidity on plates with different glucose concentrations, we took images of the petri dishes after one week. We also imaged a no-cell control. We converted these images into CMYK colorspace and measured the mean Y (yellow) intensity over all pixels in each petri dish. We subtracted the no-cell control value from each treatment petri dish, and the resulting number is the acidity indicator.
References
Wimpenny JWT. (1979). The growth and form of bacterial colonies. J Gen Microbiol114: 483–486.