Our approach for double DNA staining and color deconvolution unambiguously identifies the kinetoplast and nucleus in micrographs of trypanosomatids. This allows the analysis of complex trypanosomatid phenotypes without the risk of bias; our approach avoids any manual identification of kinetoplasts and nuclei that would require a qualitative judgment based on morphology, quality of staining and location of organelle within the cell. This unambiguity allowed us to draw conclusions about kinetoplast and nucleus number and structure in an example cell line, the SCC1-mutAB mutant, which would not be possible from a single DAPI micrograph. For example identification of closely apposed kinetoplasts and nuclei (Figure ) and nuclear fragments (Figure ) became trivial following color deconvolution. Analysis of SCC1-mutAB showed that cells that resembled zoids often included a fragment of nucleus adjacent to the kinetoplast, indicating that in the absence of mitosis cytokinesis can still result in partial segregation of nuclear DNA. Notably, our technique also works on out-of-focus kinetoplasts and nuclei (Figure ), which enables accurate identification of organelles even in monstrous cells where accurate focusing on all cellular substructures may not be possible.
Separation of fluorescent signal from kinetoplasts and nuclei to two separate images by color deconvolution provides new opportunities for quantitative analysis of the DNA content of trypanosomatids. Automated measurement of DNA from these images was faster and less susceptible to experimentalist bias than quantification from manually segmented DAPI images and produced data of greater precision, approaching the precision of flow cytometry, while quantifying kinetoplast and nuclear DNA separately (Figure ). This was particularly evident for measurement of nuclear DNA content of L. mexicana
: manual segmentation of the images could not account for the overlap of kinetoplast and nucleus signal due to the close proximity of these organelles [5
] while automated analysis using color deconvolution could, improving precision.
We have also shown that automated quantification of the DNA content of kinetoplasts and nuclei from color-deconvolved images can be used for quantitative comparison of multiple samples (Figure ). As an example we reanalyzed the procyclic T. brucei SCC1-mutAB cell line where disruption of nucleus morphology makes analysis of nucleus number less relevant (Figure ). We have shown these cells can, following failure of mitosis, complete cytokinesis and re-enter S phase, reaching greater than four times normal nuclear DNA content.
We extended this quantitative analysis to build an automated tool capable of rapidly extracting data on kinetoplast and nucleus number, shape, position within the cell and DNA content along with cell length and width measurement from micrographs. Data collected by this method from procyclic T. brucei are consistent with both manual measurements and previously published data. Furthermore, we were able to quantify of growth in cell length and kinetoplast repositioning through the cell cycle (Figure ). This type of analysis is only possible with a data set such as ours where correlation of multiple morphological measurements can be made without relying on classifications based on kinetoplast and nucleus number.
Use of an automated analysis tool, such as the one described here, accelerates analysis and reduces the risk of bias. It also ensures data collection is perfectly replicable, that there is a permanent record of all the data collected for every cell and keeps the possibility of archiving, future analysis and manual curation of data open. This data can then be used in many ways; for example it may simply be directly analyzed, or may be used to guide more extensive manual analysis of abnormal cell classes; the data from any cell may be crossreferenced back to the original micrographs. Our tool for automated image analysis therefore has potential for supporting both small-scale work and high-throughput analysis of large numbers of samples such as RNAi library [13
] and drug candidate [14
] screening. We have made the software tools we developed to perform these automated analyses freely available for others to use and develop further at http://users.ox.ac.uk/~path0493/htiaot.html
Our tools for automated analysis of cell morphology did, however, have some limitations. Analysis of morphological features (length, width and kinetoplast and nucleus location) was limited to cells that did not have a branched skeleton. This would prevent partially out-of-focus cells and some mutant morphologies from being analyzed. Our approach to morphology analysis could be extended, by modifying the algorithm for analysis of the medial axis transform to take into account multiple skeleton branches, to analyze these cells. Other issues with accurate analysis of cells primarily arose from technical issues with sample preparation (debris and multiple cells lying in contact with each other) and image capture (out-of-focus cells). Potential improvements that could be made to reduce the impact of imperfections in the samples are focus stacking, to reduce the impact of out of focus cells, and taking into account image texture, from the phase contrast image, to differentiate debris and thresholding artifacts from cells.
There is clear potential to expand the capabilities of our DNA staining and image analysis tools and adapt them for related applications. Firstly the automated morphology analysis tools we have developed could be applied to other trypanosomatid organelles. The approach for measurement of kinetoplast and nucleus location could be used to measure the location of any fluorescently labeled structure in the cell. Similarly the MAT-based morphology analysis could be used to analyze morphology of the flagellum, if it were fluorescently labeled, giving automated measurement of flagellum length, width and curvature. Furthermore, the MAT cell shape analysis could be adapted to measure the distribution of a diffuse fluorescent stain through the cell. Secondly, the principle of using two DNA stains and color deconvolution for separate analysis of kinetoplast and nuclear DNA could be adapted to any system capable of analyzing fluorescence. For example, kinetoplast and nucleus DNA could be analyzed separately by flow cytometry using double DNA staining and zoids (kinetoplast DNA only) and dyskinetoplastic cells (nuclear DNA only) to determine the reference values for signal deconvolution. Finally, the approach of using two DNA stains and color deconvolution can be applied to any biological sample with regions of different DNA sequence bias. Examples include AT/GC rich banding in condensed chromosomes and AT or GC rich regions of the interphase eukaryotic nucleus arising from the underlying nuclear organization. The source code for our color deconvolution and morphology analysis tools are freely available for others to modify to enable these kinds of analyses.