(131b) Development and Comparison of Algorithms for Analysis of Fluorescent Images for Studying the Dynamics of Signal Transduction Pathways

Mathematical modeling plays an important role for studying signal transduction pathways due to the complexity of the systems under investigation. While many models for parts of signal transduction pathways can be found in the literature (e.g., see [1, 2]), it is always necessary to validate and adapt a model for a specific stimulant and cell type. Model validation usually requires a comparison between prediction of the model and experimental data. One promising technique for taking dynamic measurements is to use a green fluorescent protein (GFP) reporter system [3]. This method is based upon the idea that expression of certain genes will also result in the formation of GFP for a cell line that has been modified accordingly. It is then possible to take images which show the fluorescence of the cells, where the degree of fluorescence can be correlated with the concentration of the transcription factor that is present in the nucleus of the cells. However, the analysis of fluorescent images is not a trivial task due to several reasons: (1) a different number of cells may be seen in different images; (2) not all cells will express GFP; (3) some cells may not be stationary; (4) some of the fluorescence seen in the images may be an artifact of the image. Image analysis algorithms are required in order to address these points. Two image analysis methods are presented in this work. The goal of these algorithms is to determine which areas of an image represent cells where fluorescence can be seen and to quantify the amount of fluorescence in a second step. The first method uses wavelets for removing noise from the image and searches the pixels in two directions to determine if a pixel corresponds to a cell, the background, or to an artifact of the image. The second technique is based on k-means clustering and uses principal component analysis (PCA) [4-5]. A series of images of hepatocytes stimulated by TNF-α with three different concentrations have been processed and a mathematical model has been built to illustrate the NFκB dynamics for different stimulation concentrations of TNF-α. The results show that both of the image analysis methods successfully detect the fluorescent cell regions. The method based on k-means clustering and PCA has the advantage, when compared with an existing image analysis technique [6], to provide the information for the pixels of different intensity levels, while the method based on wavelets requires less computational effort.

References

[1] P. C. Heinrich, I. Behrmann, S. Haan, H. M. Hermanns. 2003. Principles of interleukin (IL)-6-type cytokine signaling and its regulation. Biochem. 374: 1-20. [2] Singh A. K., Jayaraman A., Hahn J. 2006. Modeling regulatory mechanisms in IL-6 signal transduction in hepatocytes. Biotechnol. Bioeng., 95 (5): 850?862. [3] S. Subramanian and F. Srienc. 1996. Quantitative analysis of transient gene expression in mammalian cells using the green fluorescent protein. J. Biotechnol., 49:137-151. [4] F. Guillemin, M.F. Devaux and F. Guillon. 2004. Evaluation of plant histology by automatic clustering based on individual cell morphological features. Image Anal Stereol, 23: 13-22. [5] P. Geladi, H. Grahn. 1996. Multivariate Image Analysis. John Wiley & Sons, Chichester, U.K., [6] S. Venkataraman, J. L. Morrell-Falvey, M. J. Doktycz, and H. Qi. 2005. Automated image analysis of fluorescence microscopic images to identify protein-protein interactions. Proc. 27th Annu. Conf. IEEE Engineering in Medicine and Biology, Shanghai.