RESEARCH ARTICLE A computational approach to inferring cellular protein- binding affinities from quantitative fluorescence resonance energy transfer imaging Khamir Mehta 1 , Adam D. Hoppe 2 , Raghunandan Kainkaryam 1 , Peter J. Woolf 1,3 , and Jennifer J. Linderman 1,3 1 Department of Chemical Engineering, University of Michigan, Ann Arbor, MI, USA 2 Department of Chemistry and Biochemistry, South Dakota State University, Brookings, SD, USA 3 Program in Bioinformatics, University of Michigan, Ann Arbor, MI, USA Received: June 6, 2008 Revised: June 18, 2009 Accepted: September 1, 2009 Fluorescence resonance energy transfer (FRET) microscopy can measure the spatial distri- bution of protein interactions inside live cells. Such experiments give rise to complex data sets with many images of single cells, motivating data reduction and abstraction. In particular, determination of the value of the equilibrium dissociation constant (K d ) will provide a quantitative measure of protein–protein interactions, which is essential to reconstructing cellular signaling networks. Here, we investigate the feasibility of using quantitative FRET imaging of live cells to estimate the local value of K d for two interacting labeled molecules. An algorithm is developed to infer the values of K d using the intensity of individual voxels of 3-D FRET microscopy images. The performance of our algorithm is investigated using synthetic test data, both in the absence and in the presence of endogenous (unlabeled) proteins. The influence of optical blurring caused by the microscope (confocal or wide field) and detection noise on the accuracy of K d inference is studied. We show that deconvolution of images followed by analysis of intensity data at local level can improve the estimate of K d . Finally, the performance of this algorithm using cellular data on the interaction between yellow fluor- escent protein-Rac and cyan fluorescent protein-PBD in mammalian cells is shown. Keywords: Dissociation constant / Fluorescence resonance energy transfer / Image deconvolution / Optical blurring / Protein–protein interaction / Technology 1 Introduction Protein–protein interaction networks form a fundamental regulatory mechanism controlling the behavior of living cells. Characterization of these interactions, in particular the measurement of protein affinities, is of interest for various applications including tissue engineering, drug discovery and development of predictive models of cell behavior. Although many methods have been developed to measure the binding affinities of interacting proteins, including in vitro assays [1–6], methods for quantitative local characterization of protein– protein binding in live cells still require improvement. Fluorescence microscopy is the method of choice for direct visualization of proteins in native cellular environments [7–10], and recent developments in imaging techniques promise measurement of protein interactions with improved spatial and temporal resolution [5, 8, 11]. Protein–protein binding inside live cells can be visualized by fluorescence resonance energy transfer (FRET) [12]. FRET is the non- radiative transfer of fluorescence energy from an excited Abbreviations: CFP, cyan fluorescent protein; FRET, fluorescence (or forster) resonance energy transfer; PSF, point spread function; YFP, yellow fluorescent protein; 3-D FSR, 3-D FRET stoichiometry Correspondence: Professor Jennifer J. Linderman, Department of Chemical Engineering, University of Michigan, Ann Arbor, MI 481092136, USA E-mail: linderma@umich.edu Fax: 11-734-764-0459 & 2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.proteomics-journal.com Proteomics 2009, 9, 5371–5383 5371 DOI 10.1002/pmic.200800494