|
Flow cytometry is a well-established technique that is widely applied in numerous fields, including pathology, pharmacology, immunology, marine biology, plant biology, and molecular biology. Conventional methods for flow cytometry fail to accurately detect cellular phenotypic characteristics due to a limited number of variants and the lack of spatial metrics. Imaging-based cell analysis methods, such as high-content screening and imaging flow cytometry, are advantageous over those univariate or few-variate methods because they offer the capability of acquiring multi-dimensional information of single cells, from which cellular characteristics can be detected with high accuracy and high specificity. However, currently available imaging flow cytometry methods suffer from low throughput which is mainly limited by the imaging techniques, or specifically, the frame rate of the commercial imaging sensors, such as CCD or CMOS sensors. In order to address these problems, here we present optofluidic time-stretch microscopy with extremely high throughput which is capable of acquiring bright-field images of large populations of cells with a high spatial resolution of 780 nm and a high throughput of >1 million cells/s, which is two orders of magnitude higher than conventional univariate or few-variate flow cytometry methods and three orders of magnitude higher than other imaging flow cytometry methods. This is made possible by integrating an optical time-stretch microscope with a hydrodynamic-focusing microfluidic device. In addition, we apply machine-learning algorithms to the acquired images to extract multiple morphological features from each cellular image to identify and classify the cells in a label-free manner with accuracy higher than 90%, which is comparable with the fluorescence-based methods. Specifically, we experimentally performed optofluidic time-stretch microscopy to detect K562 cells (leukemia cell line) spiked in whole blood samples which were treated with different concentrations of anti-cancer drugs. In the experiment, more than 10,000 high-quality images of K652 cells were acquired for each concentration of the drug. With machine-learning-based image processing and analysis, 548 morphological features were extracted from each image to comprehensively evaluate its cellular phenotypes and hence dose-dependent morphological changes of the cells caused by the drug treatment. We further confirmed the dose-dependent results in different experimental trials where the cells were treated with the drugs for different time spans. This is potentially applicable for research of cellular drug responses directly with whole blood, hence, beneficial to drug discovery and drug screening. With such high throughput, high performance and good compatibility with existing techniques, we believe that optofluidic time-stretch microscopy with extreme throughput will revolutionize the flow cytometry field and play an integral role in high-throughput, high-accuracy, and label-free cell screening in the future.
|
