Date of Award
12-26-2023
Degree Name
Doctor of Philosophy (PhD)
Degree Type
Dissertation
Abstract
When studying biochemical processes, localizing the targets alone will not paint a complete picture. We also need techniques to probe how biomolecules interact with one another, including how molecules are organized into larger structures, how they are oriented with respect to surrounding molecules, and how local chemical parameters, like pH and hydrophobicity, vary spatially and temporally. My dissertation focuses on single-molecule orientation localization microscopy (SMOLM, Chapter 1). The objective is to develop imaging techniques for measuring how molecules are oriented with respect to their surrounding molecules, namely, 3D orientation. Achieving optimal imaging performance requires careful design of both the forward process (optical hardware) and the inverse process (estimation algorithm). In Chapter 2, I will first introduce various engineered microscopes I designed, including xy-polarized standard microscopy, pixOL, and cross DSFs. I will focus on pixOL, which I designed to optimally modulate the emission light collected from single molecules. The pixOL microscope encodes the 3D orientation and 3D position information of molecules into the shape of the dipole-spread functions, the vectorial analog of scalar point-spread functions, captured by a camera. In Chapter 3, I will extend the orientation measurement from single-molecule imaging to epifluorescence imaging. I will theoretically demonstrate that we can combine excitation modulation with emission modulation to image orientation spectrums for emitters with different orientations but spatially located at the same location. With the new pixOL microscope in hand, whose images change dramatically as emitters translate and rotate in 3D, in Chapter 4, I will next present three algorithms I designed for estimating the orientation, including iterative optimization-based algorithm, RoSEO3D, machine learning-based estimation algorithms, termed Deep-SMOLM and Deep-SMOLM3D, to decode information contained within the captured images. Deep-SMOLM is designed to deconvolve a set of intensity patterns, which we term basis images, from raw data rather than directly estimate 3D orientations. This architecture allows us to leverage the physical forward model to estimate robustly and simultaneously the 3D orientation and 2D position of single molecules. In Chapter 5, I will present our endeavor to map nanoscale structures inside biomolecular condensates. Using environmentally sensitive fluorogenic dyes, we visualized heterogeneous structures inside condensates called hubs, which arise from transient physical crosslinks formed through protein interactions. SMOLM maps the orientation of dyes at the interface of condensates. Our images show that proteins at the interface are organized with specific orientations rather than simply oriented randomly. Lastly, in Chapter 6, I will summarize this thesis and discuss the future directions for SMOLM.
Language
English (en)
Chair
Matthew Lew