Tumors are composed of heterogeneous cancer cells. Individual cells have different (epi)genomic, transcriptomic and/or proteomic profiles, resulting in different functional phenotypes (static and/or dynamic features), like morphology, migratory speed, DNA damage response or cell division pattern. Some of the phenotypes can be correlated with tumorigenesis, tumor metastasis or therapy resistance.
Capturing the aggressive subpopulations of cancer cells requires an imaging platform that can image/visualize tens to hundreds of thousands of cells without losing spatial and temporal resolution. For this, we have developed an ultrawide field-of-view optical (UFO) microscope. UFO allows screening and imaging a large quantity of cells (10^4-5) from 2D cell cultures with subcellular resolution. We are also developing a similar setup with a light-sheet modality for imaging tissue slices or 3D samples.
(Machine learning-based) image analysis
To process (UFO-curated) big imaging datasets in a real-time fashion with high accuracy, we have developed a real-time cell segmentation and tracking algorithm, called FACT (fast and accurate cell tracking) algorithm. FACT combines modified, GPU-based WEKA-segmentation and real-time, Gaussian mixture model-based cell tracking; the algorithm can easily be applied to different cell types with detection and tracking accuracy above 90% and process speed of ~42,000 cells/min.
Our big, multidimensional imaging datasets can be further processed to extract quantifiable characteristics or phenotypic information of individual cells. We employ machine learning approaches to predict cell fate of aggressive cells and to further identify features of those aggressive cells.
Single cell profiling techniques
Single cell technology is suitable to analyze rare, heterogeneous and dynamically changing cells such as metastatic cancer cells, but standard cell identification and isolation protocols based on static fluorescence are often not compatible with the reality of cancer research, where we might lack clear biomarkers for cells of interest.
We develop microscopy-based functional single cell identification assays based on time-resolved, quantitative measurements of dynamic variables. We create isolation techniques that allow cell selection based on complex, time-varying and multidimensional parameters. We perform single cell, multi-omics profiling (single cell DNA/RNA sequencing and proteomic profiling) on cells of interest to relate phenotypic information to genetic, transcriptomic and proteomic profiles. Furthermore, we also develop multiplexing technologies that allow single cell profiling cells with many different phenotypes simultaneously.
Our multi-modal single cell sequencing datasets (linking to known phenotypes of interest) can be further integrated and processed to decipher phenotype-driving mechanisms or pathways. We further apply (machine learning-based) bioinformatic analysis to identify potential actionable targets (disease- or phenotype-associated biomarkers or druggable targets).