Chimeric Antigen Receptor (CAR) T cell therapy has been shown to fight various diseases, including cancers. In addition, T cell manufacturing has become a large part of the biopharmaceutical industry due to recent advances in regenerative medicine to cure diseases such as cancer and HIV. Cell manufacturing can be optimized by monitoring cell types, their density, and morphology throughout the manufacturing process. At present, an offline flow cytometry analysis, which is labor-intensive, time consuming, and may introduce contamination, is the only way of characterizing cell samples during manufacturing. In a previous IR&D project, "Enabling In-Situ Label-Free Cell Feature Detection in Flow-Based Cell Expansion Bioreactors” (10-R8925), it was determined that, with an appropriate microscope setup, cell samples from a bioreactor can be classified as monocytes or lymphocytes. However, the internal structures of the cells could not be identified, and the types of lymphocytes (i.e., B cells, T cells, and NK cells) were not differentiable with a standard microscope setup. One technology that may offer the necessary capability is hyperspectral imaging (HSI).
While the eventual goal is to automate real-time cell classification during cell manufacturing, the focus of the current research is to use HSI to collect data on cell samples obtained from an SwRI-owned flow-based bioreactor. Performing exploratory data analysis will help us reveal the signatures, or “fingerprints,” of the cells to enable the detection, quantification, and classification of monocytes and lymphocytes within the sample without staining. Finding the “fingerprints” will allow tailoring the data collection method from hyperspectral to multispectral. With the unique signatures, a cell classification model will be developed to detect the cells, in addition to determining the class of the cell. Following model development, a low-cost camera and filter combination end solution will be created to provide a marketable and cost-effective system.
Using an HSI system, an initial hyperspectral dataset of cell samples has been captured with assistance from our bioreactor lab. In addition, we collaborated with an imaging spectroscopy expert from the Space Science and Engineering Division to build a modular hyperspectral spectrometer and video microscope system suitable for collecting data to perform analysis real-time and offline. We have also developed a software framework of multivariate analysis algorithms using non-negative matrix factorization (NMF). We are using this framework to perform exploratory data analysis on the captured dataset to search for the unique spectral signature(s) cell samples to aid in the classification and detection of the cells.