These observations were in agreement with a recent report in 3T3-L1 cells with uncertain mechanisms

ethod. These adjustments allowed for a better identification of the cell border, and were not required when the cytoplasm is strongly stained, as in the case of SG AGI-5198 price detection by immunofluorescence against eukaryotic initiation factor 3g shown below. We observed that a reduced number of cells were irregular and smaller than average, with an area smaller than 30 mm2.These include partially attached cells, due to the semi-adherence of the S2R+ cell line, as well as a small number of unhealthy cells, and were not included in subsequent analysis. Results Development of a MATLAB script for the identification of cells and SGs The accurate identification and characterization of cellular components using high-throughput microscopy, which retrieves a large amount of data, requires an automated methodology. To develop a MATLAB script for the computerized analysis of supramolecular aggregates we used SGs as a model system. These mRNA silencing foci are an ideal model since their size and number vary depending on the cellular physiology. Drosophila S2R+ cells were exposed or not to arsenite, a known inductor of oxidative stress that promotes SG formation. To visualize SGs and cell nuclei we co-stained the cells with oligodT-Cy3 and DAPI as indicated in Materials and Methods and 636 confocal images of 5126512 pixels were taken. As expected, in addition to SGs the FISH for polyadenylated RNA stained also the nucleus as Identification of SGs by correlation with prototype granules The next step was to identify the SGs present in the micrographs of the training set. This was done using a normalized 2-D cross-correlation between the matrix that correspond to the oligodT-Cy3 signal and several prototype SGs. Cells were considered positive when they contain two or more SGs. Average number of SGs in the SG-positive cells is indicated. These values were used to tune the MATLAB script and values computed after BUHO optimization with this training set are indicated. doi:10.1371/journal.pone.0051495.t001 3 A MATLAB Script for High-Throughput Image Analysis each prototype granule greatly improved their performance in identifying the SGs present in the original images. A preliminary analysis using Gaussian filters did not perform better than the prototypes selected from the micrographs. For each combination of prototype and micrograph, a matrix was obtained containing correlation coefficients with values from 21 to 1, according to the similarity between the local pixel pattern and the prototype granule. Next, the points with correlation coefficients lower than a threshold termed similarity threshold, which we empirically adjusted for each prototype were eliminated using the image to black and white function, and a value of 0 was assigned to them. Points with correlation coefficients above the ST were assigned a value of 1. We called these points “seeds”. As expected, this operation generated a significant number of seeds inside the nucleus, as the abundant nuclear polyadenylated RNA is recognized by the Cy3labelled probe. Seeds inside the nucleus were eliminated, together with a small amount of seeds generated in extracellular regions. We observed that a reduced number of SGs were weakly stained and were not recognized by the above operations. This was compensated by the application of the imfilter function with the fspecial: unsharp filter thus allowing faint or blurry SGs to be detected. These newly acquired SGs were added to the SGs identified in the unfiltere