Mechanical Adaptations of Epithelial Cells on Various Protruded Convex Geometries

The shape of epithelial tissue supports physiological functions of organs such as intestinal villi and corneal epithelium. Despite the mounting evidence showing the importance of geometry in tissue microenvironments, the current understanding on how it affects biophysical behaviors of cells is still elusive. Here, we cultured cells on various protruded convex structure such as triangle, square, and circle shape fabricated using two-photon laser lithography and quantitatively analyzed individual cells. Morphological data indicates that epithelial cells can sense the sharpness of the corner by showing the characteristic cell alignments, which was caused by actin contractility. Cell area was mainly influenced by surface convexity, and Rho-activation increased cell area on circle shape. Moreover, we found that intermediate filaments, vimentin, and cytokeratin 8/18, play important roles in growth and adaptation of epithelial cells by enhancing expression level on convex structure depending on the shape. In addition, microtubule building blocks, α-tubulin, was also responded on geometric structure, which indicates that intermediate filaments and microtubule can cooperatively secure mechanical stability of epithelial cells on convex surface. Altogether, the current study will expand our understanding of mechanical adaptations of cells on out-of-plane geometry.


Image analysis process
In the main text, we reported how the protruded structure influences the morphology and biomarker expression of the cells. Here we show how this information is quantitatively extracted from the raw image of the confocal microscopy using the imaging analysis codes.
As reported in the Methods section, 3D confocal fluorescence images of the cells on protruded structures were obtained and then the 3D confocal images were reconstructed in IMARIS software (Bitplane, Zürich, Switzerland). From that, we generated 2D projected-single plane image ( Figure SN1A) by applying 2D projection function where the reconstructed 3D objects can be orthogonally projected in XY plane direction with maximum intensity projection (MIP). This 2D projected image ( Figure SN1A) is then the starting point of our home-built IDL codes that digitalize the morphology and biomarker expression of every individual cells.
The first step is to segment the cells that are in confluent status using the watershed algorithm. As shown in Figure SN1A, the F-actin (Phalloidin-FITC, green signal) localizes well at the boundary of the cells. This feature allows us segmenting the whole image ( Figure SN1A) into individual patches ( Figure SN1B) and obtaining the boundary of each patch ( Figure SN1C). Note that each channel (green, blue and red) of a colored image is a 1000-pixel by 1000-pixel matrix and the magnitude of the matrix element is the intensity of the corresponding signal (e.g., green for F-actin and blue for nucleus). Therefore, after the watershed segmentation, we obtain the pixels' coordinate that consist of both the boundary and the region enclosed by the boundary of every cell. Two representative cells (the two cells in the colored boxes (red and purple) in Figure SN1C) extracted from the raw image are schematically shown in Figure  SN1D. The solid curves (red and purple) are the cell boundary. The green and blue area that fill the areas inside the solid curves represent the green and blue signals in the raw image ( Figure SN1A), respectively. In the following steps, we discuss how to quantify the biomarker expression and the cell morphology using the boundary (red and purple curves) and inner cell information (green and blue patches) as schematically presented in Figure SN1D.
To obtain the level of biomarker expression, we cropped the part of the raw image ( Figure  SN1E, F) that corresponds to the pixels that are identified to belong to the inner cell region (green areas in Figure SN 1D). For a certain type of marker, we further select the pixels whose brightness magnitude in this channel (e.g, green for F-actin, blue for nucleus) is greater than an empirical threshold value. The expression level of this biomarker is then the average value of these further selected pixels. The reason for setting a threshold value is to get rid of the white noises of the equipment like CCD.
For the morphology quantities, the calculation procedures are presented as below. The perimeter and area of the cells are obtained by simply counting the number of pixels that are identified as cell boundary (red and purple curves in Figure SN1D) and inner cell region (green areas in Figure SN1D), respectively. The mass center (red and purple crosses in Figure SN1D) of the cells are determined by taking average of the coordinate of the pixels that are enclosed by the cell boundary, i.e., the inner cell pixel. By fitting the inner cell points into an ellipsis (dotted black curves in Figure SN1D), we use the orientation of the long axis (red and purple dashed lines in Figure SN1D) of the fitted ellipsis as the orientation of the cell body. The relative angle is then the angle between the cell body orientation (red and purple dashed lines in Figure SN1D) and the orientation between the mass center of cell and the structure (origin lines in Figure SN1D).
Note that the cells are growing on the curved protruded structure instead of the flat surface. Therefore, the extracted morphology information needs to be corrected based the geometrical relationship between the area occupied by the cell and its xy-projection ( Figure SN1G). The formulism of this step has already been reported in the main text (2. Materials and Method/2.6. Cell morphological index and mathematical correction) and one can refer to that section for details.    for wall thickness of 60, 50 and 40 μm were 749, 580 and 399, respectively (N=3). Error bars in graph indicate S.E.M. One-way ANOVA with post-hoc Fisher's LSD was used. ****, p < 0.0001; *, p < 0.05; NS, not significant. for CN03 (resp. Lat-A) treatment on arm of the triangle and square structure were 340 (resp. 402) and 364 (resp. 343), respectively (N=3). Error bars in graph indicate S.E.M. One-way ANOVA test with post-hoc Fisher's LSD was used. NS, not significant.