Reverse engineering cell competition from a single-cell perspective using time-lapse microscopy and deep learning image analyses

细胞竞争是一种广泛保守的基本组织质量控制机制。MDCK野生型与MDCK突变体（Scribble-knockdown）的细胞竞争试验主要依赖于一种机械的竞争机制[Norman， 2012]，该机制假设组织的压缩应力驱动了竞争的结果。根据这一机制，增殖的野生型细胞胜过突变的Scribble细胞，导致这些凋亡的“失败者”细胞从上皮细胞顶部挤出。先前的研究表明，野生型细胞的分裂率增加[Bove 2017]，但这是“失败者”细胞群凋亡增加的原因还是结果仍然是一个谜。在我的研究中，我的目标是通过使用多模态延时显微镜连续几天进行图像竞争分析来回答这个问题。然后使用可以区分细胞类型的卷积神经网络（CNN）将图像分割为野生型或突变型对象，之后使用贝叶斯多目标跟踪器在整个延时过程中跟踪它们。利用荧光细胞周期指示器探针的共轭分析，可以确定细胞命运承诺的关键时间点，突出显示哪些细胞驱动竞争以及这些细胞响应的局部条件。

Cell competition is a widely conserved, fundamental tissue quality control mechanism. The cell competition assay of MDCK wild-type vs. MDCK mutants (Scribble-knockdown) relies on a mostly mechanical mechanism of competition [Norman, 2012] which posits that the compressive stresses of the tissue drive the outcome of the competition. According to this mechanism, proliferative wild-type cells outcompete mutant Scribble cells resulting in these apoptotic 'loser' cells being apically extruded from the epithelia. Previous studies show that there is an increased division rate of wild-type cells [Bove 2017], but what still remains a mystery is whether this is a cause or consequence of increased apoptosis in the 'loser' cell population. In my research, I aim to answer this question by using multi-modal, time-lapse microscopy to image competition assays continuously for several days. The images are then segmented into wild-type or mutant objects using a convolutional neural network (CNN) that can differentiate between the cell types, after which they are tracked throughout the time-lapse using a Bayesian multi-object tracker. Utilising a conjugate analysis of fluorescent cell-cycle indicator probes, key timepoints of cellular fate commitment can be identified that highlight which cells are driving the competition and what local conditions those cells are responding to.