Project Summary/Abstract
This project aims to develop a new optical imaging tool capable of dynamically profiling immune cells in live
tissue at a single-cell resolution. The immune system plays a vital role in defending the body against invaders
and combating diseases. However, uncontrolled or dysfunctional immune cell activity can worsen various
diseases, including cancer and autoimmune disorders. To gain a comprehensive understanding of the immune
system, we require tools that not only quantify different immune cells but also observe their functional state
changes and dynamic interactions with other cells in tissue. Recent advancements in immune profiling using
flow cytometry have provided valuable insights. However, tissue dissociation is necessary, resulting in the loss
of spatial information. This is a significant drawback because the spatial context and heterogeneity of immune
cell interactions are crucial for immune function. To address this issue, emerging techniques such as spatial
transcriptomics and cyclic immunofluorescence imaging have been developed. However, these techniques can
only capture static snapshots of cells and are not applicable to live samples. Additionally, they require complex
sample processing and time-consuming data acquisition, and are often limited to thin tissue sections. Observing
and characterizing the immune response in live tissue or animals at a single-cell resolution remains a major
challenge because of the highly dynamic nature of the immune system and its rapid response to external stimuli,
such as infection or drug treatment.
To overcome these challenges, this proposal seeks to leverage advances in high-speed Raman imaging and
deep learning to create a virtual staining tool for dynamic immunoprofiling. Stimulated Raman scattering (SRS)
microscopy is a label-free imaging technique that can rapidly examine intact live tissue based on the unique
Raman vibrational signatures of molecules. We will employ fast hyperspectral SRS imaging to acquire
comprehensive 5D datasets (spatial 3D + spectral + temporal) of tissue. Through iterative deep-learning training
using immunofluorescent labeling as the ground truth, we will construct a multiplex virtual staining model capable
of classifying multiple types of immune cells and their functional states. Different deep learning models will be
tested and compared to traditional machine learning approaches to evaluate the significance of spatial and
spectral features in distinguishing immune cells. The accuracy of classification will be quantitatively determined.
Validation of our virtual staining technique will involve diverse samples, including isolated immune cells from
mouse spleen, mouse lymph node tissue, and mouse tumor tissues. Successful implementation of multiplex
virtual staining of immune cells will enable dynamic profiling in live tissue or animals. We anticipate wide-ranging
applications of this technology in the study of infections, autoimmune diseases, neurodegenerative disorders,
and cancer immunotherapy.