Arun Shivanandan
Single Molecule Localization Microscopy (SMLM) techniques, such as PhotoActivated Localization Microscopy (PALM) and STochastic Optical Reconstruction Microscopy(STORM), can get around the diffraction limited resolution of conventional fluorescent microscopy (FM). However, there are a number of possible sources of imaging errors in SMLM which can significantly impact such studies. They include labeling artifacts, a limited detection efficiency of label molecules(of about 40-60% in PALM, e.g.,) and an uncertainty in localization in the range of 20-50nm. This thesis describes a rigorous review of such sources of error. Also, the SMLM readout is different from FM, the tools used in FM may not be directly applicable to SMLM. Accurate and precise quantitative imaging with SMLM requires analytical, experimental and software tools that address such issues. We describe analytical methods that accounts for two major sources of errors in analysis of membrane protein organization with SMLM. We model limited detection efficiency as independent subsampling of the set of label molecules. We then use a theoretical property of commonly used second order properties in quantitative SMLM, such as Ripleyâs K-function, L(r) - r function and the Pair Correlation Function (PCF), to show that they are invariant to such subsampling. We derive expressions for their stochastic estimators due to subsampling, and characterize the errors. The results can be extended to co-localization analysis as well. We then describe a method that estimate the true locations of points given the observed ones in clusters. We characterize the relative Mean Squared Error of the combined approach, and find that it can significantly reduce the errors in quantification. We apply these methods on data on clustering due to photoblinking of individual fluorophores, and data with redundant labeling. We then study the theoretical properties of a function that has been proposed as an estimator of cluster size. We also describe a method to identify the cluster model from data. SMLM provides single molecule resolution images of specific molecular species. Atomic Force Microscopy ( AFM), on the other hand, provides nonspecific, high resolution spatial profile information. Correlative AFM-SMLM can provide not only validation of SMLM, but also the complementary information so obtained can be used to design innovative experiments. We describe in vitro imaging of actin filaments with an AFM-SMLM correlative tool, that could provide information about sources of imaging inhomogeneity in SMLM. The tool were also used to image live mammalian cells, and can be used to obtain nanoscale information about mechanical properties of cells and also as a tool for nanomanipulation. Co-localization is a common spatial interaction quantification method used in biological imaging. It is possible to use tools from spatial statistics to obtain better statistical power on tests of spatial interaction, compared to conventional co-localization measures. Such tools can also handle SMLM data better, since they work on point patterns rather than images. We describe an ImageJ/Fiji plugin that implements a spatial statistics framework that extends the concept of co-localization, by means of a model based on Gibbs function of interaction potentials. We describe the application of this software on both confocal microscopy data of virus-endosomes, and SMLM data on GPCR protein-clathrin.
EPFL2015
