Characterizing liver sinusoidal endothelial cell fenestrae on soft substrates upon AFM imaging and deep learning
Characterizing liver sinusoidal endothelial cell fenestrae on soft substrates upon AFM imaging and deep learning
复制标题
通过 AFM 成像和深度学习表征软基质上的肝窦内皮细胞窗孔
DOI：
10.1016/j.bbagen.2020.129702
复制
发表时间：
2020
影响因子：
3
通讯作者：
Mian Long
中科院分区：
文献类型：
--
作者：
Peiwen Li;Jin Zhou;Wang Li;Huan Wu;Jinrong Hu;Qihan Ding;Shouqin Lü;Jun Pan;Chunyu Zhang;Ning Li;Mian Long
肝窦内皮细胞（LSECs）表现出独特的开孔形态。窗的大小和数量的改变在各种肝脏疾病的进展中起着至关重要的作用。虽然它们的特征已经使用原子力显微镜（AFM）可视化，但它们的情境成像方法和离线分析还需要进一步用于窗的量化。方法小鼠原代LSECs分别在胶原i包被培养皿、聚二甲基硅氧烷（PDMS）或聚丙烯酰胺（PA）水凝胶底物上培养。采用AFM接触模式对单个固定LSECs上的开窗进行可视化。收集的图像使用内部开发的基于全卷积网络（FCN）的图像识别程序进行分析。结果首先优化了主要扫描参数，适用于不同水凝胶培养的LSECs。利用LSECs的中放大形态学图像开发基于fcn的窗窗识别程序。该程序使我们能够在软基材上经过两次训练，以81.6%的典型准确率从AFM图像中识别出绝大部分窗洞，并量化了孔隙率、窗洞数量和窗洞直径分布的统计数据。结论AFM成像与FCN训练相结合，可以定量表征LSEC孔在不同基质上的形态分布。本文获得和分析的afm图像提供了整个细胞表面超微结构的全局信息，这对于理解它们在不同刚度基质上个体细胞窗样进化的调控机制和病理生理相关性至关重要。
BackgroundLiver sinusoidal endothelial cells (LSECs) display unique fenestrated morphology. Alterations in the size and number of fenestrae play a crucial role in the progression of various liver diseases. While their features have been visualized using atomic force microscopy (AFM), thein situimaging methods and off-line analyses are further required for fenestra quantification.MethodsPrimary mouse LSECs were cultured on a collagen-I-coated culture dish, or a polydimethylsiloxane (PDMS) or polyacrylamide (PA) hydrogel substrate. An AFM contact mode was applied to visualize fenestrae on individual fixed LSECs. Collected images were analyzed using an in-house developed image recognition program based on fully convolutional networks (FCN).ResultsKey scanning parameters were first optimized for visualizing the fenestrae on LSECs on culture dish, which was also applicable for the LSECs cultured on various hydrogels. The intermediate-magnification morphology images of LSECs were used for developing the FCN-based, fenestra recognition program. This program enabled us to recognize the vast majority of fenestrae from AFM images after twice trainings at a typical accuracy of 81.6% on soft substrate and also quantify the statistics of porosity, number of fenestrae and distribution of fenestra diameter.ConclusionsCombining AFM imaging with FCN training is able to quantify the morphological distributions of LSEC fenestrae on various substrates.SignificanceAFM images acquired and analyzed here provided the global information of surface ultramicroscopic structures over an entire cell, which is fundamental in understanding their regulatory mechanisms and pathophysiological relevance in fenestra-like evolution of individual cells on stiffness-varied substrates.