Optical toolboxes for massive optophysical cytometry

Abstract

Despite its trivial advantages of bypassing the laborious and costly labelling protocols and minimizing the detrimental perturbation to the cells, characterization of the biophysical properties of cells have long been ignored in mainstream cellular assays. The available techniques are often limited by low measurement throughput, which makes them impractical in comprehensive census of large cell populations (scale of millions cells) in order to reveal cellular diversity and heterogeneity that is pertinent to understanding of health and disease. The existing tools also often lack the measurement content, which limits them to a handful of biophysical properties (typically 1-2 features). This is starkly in contrast to biochemical assay approaches such as gene expression analysis that profiles thousands of gene expressions in a single experiment. To meet these demands, this work focused on the development of new label-free optical imaging-based cytometric tools. They mainly relied on quantitative phase imaging – a powerful imaging modality that allows extraction of biophysical phenotypes of single-cells from the optical images obtained (or called optophysical phenotypes) at high resolution. The image-based approach allowed the high-dimensional feature extraction and phenotyping (> 80 optophysical phenotypes). Furthermore, the utilization of a high-throughput imaging technique, namely time-stretch imaging, enabled ultralarge-scale optophysical phenotyping of living cells (>2 millions). A microfluidic imaging flow cytometry design had been developed for suspended cells. Another spinning platform design which allowed parallelized and high-throughput imaging had also been engineered to tailor large-scale optophysical phenotyping of adherent cells. To the best of our knowledge, the throughput achieved (5.88 mm2/s or ~4700 cells/s) at this resolution (~1 μm) was the fastest so far for imaging adherent cells. To demonstrate their capability, several challenging biological applications were demonstrated. First, label-free rare cell detection and the diagnosis of lung cancer had been demonstrated. The optical system was able to capture extremely rare cancer cells in peripheral blood mononuclear cells (1 in 100,000) using optophysical phenotypes alone. In addition, combined with deep learning, it could classify 7 lung cancer cell lines according to their histopathological subtypes at an average accuracy as high as 90%. With these experiments, the possibility of using this optical tool in liquid biopsy and diagnosing the lung cancer subtypes from the circulating tumor cells (CTCs) had been proven. Another experiment had been done to show the sensitivity of these optophysical properties. A molecularly targeted drug was applied to the cancer cells at different concentrations and imaged using the developed optical tool. It was found that statistically significant difference could be identified using the optophysical phenotypes at early time points (6 hours after treatment) above the half maximal inhibitory concentration (IC50). An imaging based on fluorescently labelling intracellular structures had been used to correlate the optophysical properties with intracellular structures, which acted as first steps for biologically interpreting these optophysical phenotypes. Summarizing the above points, this thesis presents the optical tools that allow high-throughput, high-content biophysical phenotyping at single-cell resolution. This could potentially impact the new generation of cell-based assay for advancing biological discovery as well as impacting clinical diagnosis.