Measuring the physical size of the cell is valuable in understanding cell growth control. Current singlecell volume measurement methods for mammalian cells are labor-intensive, inflexible, and can cause cell damage. We introduce CTRL: Cell Topography Reconstruction Learner, a label-free technique incorporating Deep Learning and Fluorescence Exclusion for reconstructing cell topography and estimating mammalian cell volume from DIC microscopy images alone. The method achieves quantitative accuracy, requires minimal sample preparation, and applies to extensive biological and experimental conditions. Using this method, we observe a noticeable reduction in cell size fluctuations during cell cycle, which is consistent with the presence of a cell size checkpoint.(https://GitHub.com/sxslabjhu/CTRL)
Main textCell size plays a critical role during cell growth, division, and proliferation 1-5 . Abnormalities in cell size regulation and growth control are thought to promote disease development 2,5-9 . Accurately measuring single-cell size remains a challenge for mammalian cells due to their irregular shape. Existing techniques require specialized hardware, fluorescent labeling 10,11 and/or cell suspension [12][13][14][15][16] . Fluorescent labeling or over-expression of a target marker can alter cell function. Cell suspension alters the cell shape and biochemical signaling from the extracellular matrix, and also potentially affecting cell size. None of these methods has been successfully applied to measure mammalian cell growth at the single-cell level.While sensitive and accurate methods have been developed to measure single-cell mass over time 17 , the relationship between cell size and mass is not always clear.An accurate and high throughput method of cell volume quantification is the Fluorescence Exclusion method (FXm), first proposed in 1983 14 and subsequently developed and refined by several groups [18][19][20] . Cells are seeded in a micro-fabricated chamber and a membrane-impermeable high molecular weight fluorescent dye (e.g. FITC-dextran) is injected into the microchamber (Fig. 1a). The cell excludes its volume in the microchamber, therefore the total fluorescence loss is proportional to the cell volume.The FXm method obtains the cell volume from a single epi-fluorescence image, and therefore is high throughput 3,18-20 . However, due to endocytosis 3,20 that is common in many cell types, the dye eventually enters the cytoplasm, and therefore FXm generally cannot accurately report cell volume in time-lapse without careful controls. Fluorescent imaging also introduces photobleaching, which alters the signal during time-lapse measurements. Moreover, microfluidic fabrication is needed to perform the experiment and the confinement of the microchamber may alter cell physiological processes over long periods. These drawbacks limit the use and applicability of FXm for studying cell growth.Convolutional Neural Networks (CNN) have been applied to microscopy images for both phenotype classification 21,22 and image segmentati...
