Advancing fluorescence microscopy in challenging live-cell environments

Abstract

Fluorescence microscopy techniques have become indispensable tools for cell biology because of their molecular specificity and non-invasive nature. As fluorescence microscopy pushes into increasingly complex cell environments, however, interpreting the resulting data quantitatively becomes significantly more challenging. This thesis describes developments to two fluorescence techniques that extend their applicability in challenging live-cell measurements. First, we introduce a reformulated theory of the autocorrelation function (ACF) in fluorescence correlation spectroscopy (FCS) that accounts for the length of the data. This reformulation of FCS allows experimental data to be analyzed in very short segments, which enables FCS data to be interpreted even in the presence of cell dynamics that would otherwise preclude successful analysis. We also discuss the statistics of fitting the ACF and introduce a simple method to analyze FCS data that enables proper model evaluation. Next, we implement a 3D localization microscopy scheme using the double helix point spread function (DHPSF). Complex structured background significantly hindered analysis of DHPSF image data of assembling human retroviral Gag, which led us to develop a deep learning (DL) method to recover localizations in these challenging conditions. We demonstrate the DL approach on assembling HIV-1 Gag puncta in live cells and apply it to tracking mobile Gag puncta in 3D.