SParticle, an algorithm for the analysis of filamentous microorganisms in submerged cultures

Streptomycetes are filamentous bacteria that produce a plethora of bioactive natural products and industrial enzymes. Their mycelial lifestyle typically results in high heterogeneity in bioreactors, with morphologies ranging from fragments and open mycelial mats to dense pellets. There is a strong correlation between morphology and production in submerged cultures, with small and open mycelia favouring enzyme production, while most antibiotics are produced mainly in pellets. Here we describe SParticle, a Streptomyces Particle analysis method that combines whole slide imaging with automated image analysis to characterize the morphology of submerged grown Streptomyces cultures. SParticle allows the analysis of over a thousand particles per hour, offering a high throughput method for the imaging and statistical analysis of mycelial morphologies. The software is available as a plugin for the open source software ImageJ and allows users to create custom filters for other microbes. Therefore, SParticle is a widely applicable tool for the analysis of filamentous microorganisms in submerged cultures. Electronic supplementary material The online version of this article (doi:10.1007/s10482-017-0939-y) contains supplementary material, which is available to authorized users.

Mean greater than 200. Highly bright areas. Pellets or mycelia, since being heterogeneous, should not reach these high levels of brightness on average. SDB greater than 55. Areas which are highly heterogeneous in brightness. No mycelia or pellet has that high a difference in between bright and dark pixels.
SDB less than 12. Very homogeneous areas. These are out of focus areas. Pellets or mycelia fragments are generally more heterogeneous because of phase contrast imaging.
Mean greater than 110 and SDB less than 25 OR Mean greater than 75 and SDB less than 22. These rules check for heterogeneity of brightness in images given that their brightness is greater than 75 or 110. As brightness increase the tolerance for heterogeneity increases.
Mean greater than 75 and Round less than 0.3. Average and above brightness areas that are not circular cannot be pellets. Because in phase contrast small areas are darker and big areas are on average brighter. Since big areas pellets and they have a close shape they should be more circular. Therefore bright areas that are not round are eliminated.
Mean greater than 100 and Circularity less than 0.09. Again checks for circularity on bright areas using another measurement of circularity other than roundness.
Round greater than 0.7 and SDB less than 25. Round and homogeneous areas. These are generally out of focus bubbles which cannot directly be identified via circularity and AR.
Area less than 10.000pix and Round greater than 0.8 and Mean greater than 100. Small areas which are highly round. These are generally bubbles that could not have been identified via AR or Circularity.

Area greater less than 2.500pix and Circularity greater than 0.1 and Mean greater than 100.
This is areas that are small and that have both high circularity and high roundness values.
This captures any bubble that is left from other rules as well as out of focus small circular areas.
Focus check. After previous rules are checked for all ROI's, remaining areas are looked at one more time using focus controlling. This is done by identifying the edges of the ROI and applying an "Edge Finding" transformation. A band on the edge of the area is taken and measured for average intensity. If the edge is well defined in the image, the Edge Finding 4 transformation should return a high value. By applying a threshold, any unfocused region that has been missed with previous filters was found and eliminated from measurements.

Data export
Data gathered in ImageJ were stored in the temporary result table but also in the folder from which the image that is analyzed originates. Therefore, the identified objects' measurements were sorted from small to large according to the area and written into a csv file. This csv file can be automatically opened in Microsoft Excel® along with a macro file which automatically draws important parameter graphs. These are interactive graphs linked to a small image file with the size of the bounding box of the ROI from the original image, which is created automatically during measurements. This represents a useful way to review the measured images, allowing a quick way to pull images representing a subpopulation or assessing the morphology of outliners and can also be used to select images for publications. Additionally, all discarded images are stored with specific details as to why they were discarded in the image name, allowing refinement of the filtering parameters.

Explanation of polar circularity
Pellets often have hyphae protruding outwards, and thus have a large perimeter length as compared to the surface area. Therefore, methods that calculate the circularity using the perimeter length would not give an accurate description of the shape. After Cartesian to Polar coordinate transformation, a perfect circle becomes a line in Radius vs. Angle graph, while any shape deviating from a circle in Cartesian coordinates deviates from a line in polar coordinates (Stojmenovic et al. 2013). For example, in a perfect circle the distance between the center and the border is equal for any angle. While in a square of 2 by 2 for example the distance from the center in relation to the angle will change from 1 (0 degrees) to the square root of 2 (45 degrees), and recede to 1 again for a 90 degree angle, and so on (More information in Fig. S2). Based on the standard deviation from the average radius we can distinguish different shapes. The standard deviations from the line calculated and divided by 5 the average distance value to correct for shape's size resulting in the polar circularity value.
The larger the polar circularity value the more the actual object radius deviates from the average radius, in other words, a shape that is less circular. While a perfect circle would have a value of 0, the value increases as the shape deviates from a circle. Squares of the same diameter have an increased deviation (standard deviation from r = 0.11), an oval with double the length of the width gives a larger deviation (standard deviation from r = 0.23). The figure of 8 deviates even more (standard deviation from r = 0.51). This demonstrates that shape deviations from a perfect circle can be identified using polar coordinates.