Background: Liver 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), the in situ imaging methods and off-line analyses are further required for fenestra quantification.
Methods: Primary 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).
Results: Key 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.
Conclusions: Combining AFM imaging with FCN training is able to quantify the morphological distributions of LSEC fenestrae on various substrates.
Significance: AFM 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.
Keywords: Atomic force microscopy; Fenestrae; Fully convolutional networks; Imaging recognition; Liver sinusoidal endothelial cells; Substrate stiffness.
Copyright © 2020 Elsevier B.V. All rights reserved.