参见作者原研究论文

本实验方案简略版
Apr 2018

本文章节


 

An Image Analysis Pipeline to Quantify Emerging Cracks in Materials or Adhesion Defects in Living Tissues
分析活体组织中出现的材料裂缝或附着力缺陷的图像分析法   

引用 收藏 提问与回复 分享您的反馈 Cited by

Abstract

Microcracks in materials reflect their mechanical properties. The quantification of the number or orientation of such cracks is thus essential in many fields, including engineering and geology. In biology, cracks in soft tissues can reflect adhesion defects, and the analysis of their pattern can help to deduce the magnitude and orientation of tensions in organs and tissues. Here, we describe a semi-automatic method amenable to analyze cell separations occurring in the epidermis of Arabidopsis thaliana seedlings. Our protocol is applicable to any image exhibiting small cracks, and thus also adapted to the analysis of emerging cracks in animal tissues and materials.

Keywords: Cracks (裂缝), Cell adhesion (细胞粘附), Image analysis (图像分析), Mechanical properties (机械性能), Tension (张力), Github (Github)

Background

Microcracks are present in most materials; their number and extent generally increase when repeated stress is applied or when the temperature fluctuates, causing material fatigue and eventually, failure. Microcracks can reveal the magnitude and direction of the principal stresses that the material is experiencing. This property is widely used in mechanical engineering and geology (e.g., Kowallis and Wang, 1983; Kranz, 1983; Joseph et al., 2002). Microcracks are also present in biological structures, like bones. As in any material, they can lead to rupture (fatigue fracture). Such cracks also trigger signaling cascades involving osteoblasts and osteoclasts, resulting in bone remodeling (e.g., Mori and Burr, 1993). Cracks can also be observed in soft tissues, notably as a result of cell to cell adhesion defects. This is particularly obvious in plant tissues, where cells do not migrate or intercalate (e.g., Bouton et al., 2002). Although many tools have been developed to study cracks in material sciences, they are often adapted to analyze long and thin cracks (e.g., Griffiths et al., 2017); they have not been customized for the analysis of emerging cracks in soft biological tissues. Here we describe a pipeline that detects and analyzes such cracks.

Our pipeline is based on a script that segments regions exhibiting a clear-cut pixel intensity contrast corresponding to cell separations or cracks. In our original publication, we stained Arabidopsis thaliana seedlings with propidium iodide (see Verger et al., 2018 for the staining procedure), a fluorescent molecule that specifically binds to pectins in the cell wall. After washing the dye off, the cell contours are clearly marked. In a wild-type plant, this reveals a continuous epidermis where cells are fully attached to one another. However, when performing the same staining on a mutant with cell adhesion defects, holes between cells are revealed: a bright signal between cells marks emerging separations between these cells (Verger et al., 2018). The contrast between the cells and the cracks is in fact strong enough to detect and quantify cell separations. After segmenting these cracks, a principal component analysis is performed on each of these segmented areas, yielding various information: area of the crack as well as its principal orientation (angle of the crack) and the shape anisotropy (derived from the eigen values and vectors calculated in the principal component analysis of the crack shape). These data can then be compared for multiple samples and sample series. In principle, other staining method may be used on any types of tissues or materials, as long as the contrast is strong enough for the script to detect the cracks.

Software

  1. Fiji program (http://fiji.sc/)
    Open-source plugin-based image analysis software based on ImageJ (https://imagej.nih.gov/ij/
  2. Python 2.7 (Interpreted high-level programming language for general purpose programming) (https://www.python.org/)
  3. Python modules:
    1. Matplotlib (2D visualization and data plotting library) (https://matplotlib.org/)
      Nose (Python unit test framework) (http://nose.readthedocs.io/)
    2. Numpy (N-dimensional linear algebra library) (www.numpy.org/)
    3. Pandas (Data manipulation and analysis library) (https://pandas.pydata.org/)
    4. Pillow (Image manipulation library) (https://pillow.readthedocs.io/)
    5. Pycircstat (Circular statistics library) (https://github.com/circstat/pycircstat/)
    6. Scipy (Scientific computing library) (https://www.scipy.org/)
  4. Open source package management and environment management system
    1. Miniconda: https://conda.io/docs/
    2. LINUX: https://repo.continuum.io/miniconda/Miniconda2-latest-Linux-x86_64.sh
    3. MAC: https://repo.continuum.io/miniconda/Miniconda2-latest-MacOSX-x86_64.sh
    4. Windows: https://repo.continuum.io/miniconda/Miniconda2-latest-Windows-x86_64.exe
  5. Cell Separation Image Analysis Pipeline (Image analysis script described in this protocol) (https://github.com/sverger/Cell_separation_analysis)

Note: See "Installation procedure" in "Procedure" for the installation of Software 2 to 5.

Procedure

  1. Image acquisition
    Our image analysis pipeline was developed to detect and quantify cell separations in plant epidermis. Such images can be obtained using a confocal microscope and by either staining the cell wall or cell contour with a fluorescent dye, or by imaging plants expressing a fluorescent reporter of the cell contours (typically, a protein at the plasma membrane). Z-Stacks can be obtained and projected in 2D (e.g., using Fiji, max intensity). It is however crucial to obtain images in which there is a strong contrast between the cells and the “cracks” (i.e., the zone where cells are separated. In Figure 1C the regions corresponding to cells are of comparable pixel intensity as the gaps between cells, making it impossible to automatically distinguish them). Furthermore, because our pipeline works by segmenting the whole cracks, the cracks need to form a closed domain (For example, in Figure 1D, the joints marking the cracks (white zones), overlap with one another, making it impossible to distinguish them individually with our pipeline). Thus our pipeline can in principle work with any 2D grayscale image containing clear-cut closed cracks (see examples of suitable and non-suitable images in Figure 1).


    Figure 1. Suitable image type. A. Z-projection (maximal intensity) of a confocal image stack from a propidium iodide stained light-grown hypocotyl from the qua1-1 mutant with cell adhesion defects (see Verger et al., 2018). B. Cracks in rocks (credit photo: Pierre Thomas). In A and B the cracks are marked by a high pixel intensity contrast and form closed domains. C. Z-projection (maximal intensity) of a confocal image stack from a qua1-1 pPDF1::mCit:KA1 (plasma membrane reporter) cotyledon epidermis (see Verger et al., 2018). Although such fluorescent reporter can provide suitable images for our analysis, in this particular case the contrast between the cracks and the cell content is too low to allow a segmentation of the cracks with our pipeline. D. Picture of cracks in rocks (credit photo: Pierre Thomas). In this case the cracks do not form closed domains as most of them overlap. In addition, the pixel intensity is very variable throughout the picture such that some cracks do not exhibit a strong differential in pixel intensity. White arrowheads point to examples of cracks or cell separations in these images.

  2. Prerequisite: Image quality, preprocessing and threshold for “crack detection” in Fiji
    In order to determine if your images are suitable for this image analysis pipeline you need to make sure that the cracks will be properly segmented by pixel intensity:
    1. Load your 2D image in Fiji (Software 1).
    2. Change your image type to 8-bit. Image > Type > 8-bit (Figure 2B) and/or pick the right channel from a multichannel (e.g., RGB) image (Image > Color > Split channels).
    3. Optionally you may enhance the contrasts of your image using the “Enhance Contrast...” function (Process > Enhance Contrast…> Set “Saturated pixels” to 0.3%).
      Note: You may use a different approach (e.g., increase exposition time or laser intensity during image acquisition) to obtain enough contrast to segment the cracks.
    4. Smooth your image with a median filter to remove noise (Process > Filters > Median… [Figure 2C]). Set the radius to a suitable value to reduce the noise, without blurring the image too much. Here again, you may also use a different approach to reduce the noise of your image, as long as in the end, you obtain a smoother detection of the cracks.
    5. Using the “Threshold...” tool, determine the suitable threshold that best separates the cracks from the surrounding regions (Figures 2D and 2E).
      CRITICAL STEP: Depending on the nature and quality of your image, it may be difficult or impossible to segment the cracks based on a pixel intensity threshold. If most of your images are in this situation, this image analysis pipeline is not adapted to your study.
    6. If you are able to properly separate the cracks from the surrounding regions, you may proceed with the analysis. Save your image in .tif or .jpg (File > Save As > Tiff… or JPEG…) and add ”_XXXthld” at the end of the name (where XXX is the threshold value previously determined in Step B5 as suitable to segment the cracks (e.g., “sample_1_162thld.tif”. See Figures 2D-2F). The value before “thld” is the threshold value that will be used for the image segmentation later on (it has to be three digit long).
    7. You can pre-process as many images as you need before going further with the analysis. They will all be automatically processed if they are grouped in a folder. Note that in order to quantify and compare cracks orientation in multiple images, the images should be in a consistent orientation. The corresponding arborescence has to be organized as follows: A “main” directory (later on referred as “updir” in the script), containing subdirectories (e.g., different mutants or growth conditions), each containing all the corresponding images (Figure 2F).


      Figure 2. Image preprocessing in Fiji. A. Z-projection (maximal intensity) of a confocal image stack from a propidium iodide stained light-grown hypocotyl from a qua1-1 mutant with cell adhesion defects (see Verger et al., 2018). B-C. Images are preprocessed in imageJ: The image is converted to 8-bit (B) and a median blur with a radius of 3 is applied to reduce the noise and ease the segmentation (C). D-E. The threshold tool is used to determine the suitable threshold for segmentation in the pipeline. (D) Red zones will be segmented as cracks. (E) The threshold is adjusted in order to segment the cracks properly (e.g., here to a value of 162). F. Example of file arborescence required for the pipeline to process the image series. G. Z-projection (maximal intensity) of a confocal image stack from a propidium iodide stained light-grown hypocotyl from a wild-type seedling showing no cell adhesion defects and thus not suitable to detect cracks with our pipeline (see Verger et al., 2018). H-I. Because the differences in pixel intensity are locally small (unlike the qua1-1 mutant with bright cell separation signals in panel A-D), we are unable to segment properly the image either with a threshold value of 50 (H) or 45 (I) as an example.

  3. Installation procedure
    You will need to have python (Software 2) installed on your computer in order to run the script (Software 5). The script has been designed, and should thus run properly, with python 2.7. You can then either directly run the script if you already have all the required dependencies (Software 3) in your python environment or install all the required dependencies in your python environment. If you do not have Python installed, or do not wish to interfere with your current Python environment, proceed with the steps below for our recommended installation using miniconda (Software 4), which will install python and all the required dependencies.
    1. Download the miniconda installer from the official website (link below “Software 4”, or here LINUX, MAC and WINDOWS). Alternatively you can use wget to perform this download from a terminal (LINUX or MAC). Open a new terminal window and run the command line below:
      For LINUX:
      wget https://repo.continuum.io/miniconda/Miniconda2-latest-Linux-x86_64.sh
      For MAC:
      wget https://repo.continuum.io/miniconda/Miniconda2-latest-MacOSX-x86_64.sh
    2. Install miniconda by running the installer:
      For LINUX and MAC: Open a new terminal window, navigate to the directory where you downloaded the installer (most likely in your “Downloads” folder. e.g., cd path/to/Downlaods/.
      Note: Input the actual path that leads to the folder “/Downlaods/”
      ) and run:
      For LINUX:
      bash Miniconda2-latest-Linux-x86_64.sh
      rm Miniconda2-latest-Linux-x86_64.sh

      For MAC:
      bash Miniconda2-latest-MacOSX-x86_64.sh
      rm Miniconda2-latest-MacOSX-x86_64.sh

      For WINDOWS: Execute the installer and follow the instructions.
      During the installation (LINUX, MAC and WINDOWS) you will be asked a number of choices. You can set the directory of your choice when asked (e.g., ~/.miniconda). Make sure to answer YES when asked to add conda to your PATH.
    3. At this point, you should have miniconda installed. Test your installation by closing your current terminal window and running conda in a new terminal to make sure the command is found:

      conda

    4. Download and extract the “cell_separation_analysis” repository from Github (following the link in Software 5, or here). On the Github page, click on the “Clone or download” green button at the right of the page. Then download and extract the zip.
      Note: This folder contains the script (“Cell_separation_analysis.py”), a dependencies installation file for miniconda (“cell-sep-env.yml”) and test sample images (“Test_files”).
    5. In a terminal, navigate to the “/cell_separation_analysis-master” folder that you have extracted.

      cd path/to/Cell_separation_analysis-master/

      Note: Input the actual path that leads to the folder “/Cell_separation_analysis-master”.
    6. In the same terminal, create a new conda environment using the provided YAML file that lists all the software dependencies:

      conda env create -f cell-sep-env.yml

      Note: This will install Python and all the required dependencies. It may take a few minutes to complete the installation.
    7. When the installation is complete, in the same terminal, activate the environment:

      source activate cell-sep-env

      Note: You will have to run this command every time you want to use the cell separation analysis program, just after opening a new terminal window.
    8. You can then check your installation and whether the script runs properly by running the script on our test images. In a terminal, run:

      ipython

      This will launch ipython. In ipython, navigate to the “/cell_separation_analysis-master” folder that you have downloaded:

      cd path/to/Cell_separation_analysis-master/

      In the python console, type:

      %run Cell_separation_analysis.py

      This should run the script and generate its output in the “/Test_files” folder of “/Cell_separation_analysis-master”.

  4. Running the script
    In order to run the script with your own images, you need to edit the “parameter” section of the script. To do so, open the “Cell_separation_analysis.py” file in a text editor. Scroll down to the section called “parameters” where there are 7 entries that you may modify (see Figure 3):


    Figure 3. Parameter settings. Cell_separation_analysis.py Python script opens in a text editor, displaying the “parameters” section. The parameters can be modified according to your own requirements and the file can be saved before running the script.

    1. Set the directory where you placed the images to analyze. Remember that your folder has to be organized in a specific way: A “main” directory corresponding here to “updir”, containing subdirectories (e.g., different mutants or growth conditions), each containing all the corresponding images (Figure 2F). You can either enter the full path to the directory (e.g., /Home/Path/to/updir/), or simply enter ./updir/ (where you replace “updir” by the actual name of your folder). In the latter case, you will have to navigate to the parent folder of your “updir” in Ipython before running the script (as described in Step C8).
    2. Set the pixel size. This is required in order to perform analyses of crack area. Usually, for confocal images this information can readily be found by loading the original image in Fiji and looking at its properties (Image > Properties… > Pixel width or height). The unit of length should be micron. For other types of images, you will need to determine the actual pixel size, for example in Fiji, using an internal scale (Analyze > Set Scale…).
    3. Determine the minimum and maximum area of crack. This step corresponds to a filter as it eliminates areas which are too small (“min_area_of_crack”, e.g., background noise) as well as the global background of the image (“max_area_of_crack”, e.g., the empty space around the tissue). These values are in pixel number. You may have to try several times with different values in order to determine the right parameters empirically.
    4. Set the threshold type. The cracks can be of either a higher or lower pixel intensity than the surrounding region. Thus the threshold can be set as “min” or max”. The “max” will detect and segment zones with lower signal intensity (i.e., the crack is darker than the surrounding region) and the “min” will detect and segment zones with higher signal intensity (i.e., the crack is lighter than the surrounding region).
    5. You may then decide to run more global analyses if you have multiple sample types (“Global_Output_Size”) and multiple images by sample type (Global_Polarhist_output). See “Data analysis” section to determine if you should run these analyses. These are by default set to “False” which mean they will not be performed. Replace “False” by “True” if you want them to be performed.
    6. Once all the parameters are correctly set, save the script and run it as described in Step C8.

Data analysis

  1. As explained in the background section, this script can generate different outputs. From the detected cracks in each image, the script will generate three images: a pixel intensity inverted version of the image, the same image with an overlay of the segmented areas, and the same image with an overlay of the anisotropy and principal angle of the area, directly saved as vectorial PDFs. It will also generate a .csv file containing, for each segmented cracks, the label number, center position, area in pixels and micrometer square, the main orientation (angle) of the crack, the shape anisotropy, the eigen values and vectors. Finally for each image, a polar histogram representing the distribution of crack orientations in the image is created and saved as a vectorial PDF (see output generated in the test of the script in the previous step and Figure 4).


    Figure 4. Output of the cell separation analysis pipeline. A. Pixel intensity inverted version of the image (Figure 2C). B. Same image as in (A) with an overlay of the segmented areas that are identified and labeled using different colors to ease visualization. C. Same image as in (B) with a representation of the vectors resulting from the principal component analysis of the crack shapes (Red crosses). To improve the visual output, the eigen vector that are mapped on the images are multiplied by a factor 2 and by the square root of the corresponding eigen value. D. A polar histogram representing the distribution of the crack orientations in an image. The square root of the principal eigen value for each plotted angle is added to normalize the angle value by its relative weight. A color map is used in the polar histograms representing the relative number of angles binned in each histogram bar (independently of their weight) where yellow is high and purple is low.

  2. If you process multiple images from one sample series (e.g., technical or biological replicates), you can output a summary of their properties (see Step D5, "Global_Polarhist_Output = True"). It will generate a polar histogram representing the distribution of cracks orientation for all the images of a sample series pooled together, and save it as a vectorial PDF. It will also output a .txt file containing for each image the circular mean angle (between 0° and 180°), the resultant vector length (an estimation of the coordinated directionality of the cracks, between 0 and 1; a value of 0 means that the crack orientations are homogeneously distributed whereas, a value of 1 means that all the cracks have the same orientation) and the mean anisotropy of the crack shapes. At the end of the text file, a global analysis of all the images of a sample series is generated: circular mean angle, resultant vector length and shape anisotropy for the pooled values of all the images of the sample series. It will also output the result of an RAO's spacing test, which assesses whether the angles are uniformly distributed (statistically no preferential angle orientations), or if there is a significant angular bias.
  3. The script also offers the possibility to compare different type of samples. You can compare the total area of the cracks in two sets of samples (e.g., 10 images of mutants 1 compared with 10 images from mutants 2) (see Step D5, "Global_Output_Size = True"). This will run statistical tests on the compared samples to determine if the average area of cracks in images of one sample series is different from that of another sample series. The choice of statistical test depends on the normality and the variance of the data. First, a Shapiro's test for population normality is run on both sample series. If at least one of the sample series does not have a normally distributed population, a non-parametric Wilcoxon rank sum test is run to test whether the samples are statistically different. Otherwise, if both sample series are normally distributed, a Bartlett's test for equal variances is run. If both sample series have equal variance, a Student's t-test is run and otherwise a Welch's t-test is run to test whether the samples are statistically different. A summary of these tests is saved as a .txt file, and a boxplot of this comparison is saved as a vectorial PDF.
  4. You may then perform further analyses depending on your needs, using the different output files containing all the raw data.

Notes

  1. The most critical step in this protocol is to check the suitability of your images as described in “Procedure B”. If most of your images do not pass this test, this image analysis pipeline may not be suited for your study. Conversely, it is important to realize that it is also rare to be able to segment 100% of the objects that you would identify as cracks visually. You should then decide what is an acceptable yield in your case.
  2. Following our recommended installation procedure you should in principle be able to run our image analysis pipeline on any system (LINUX, MAC, and WINDOWS). However, the script has so far only been tested on a computer running Ubuntu 14.04 and Python 2.7. We recommend testing the macro with our sample images (Step C8) and check if the output images look similar to what is reported in our original publication (Verger et al., 2018).

Acknowledgments

This work was supported by the European Research Council (ERC-2013-CoG-615739 ‘‘MechanoDevo’’). We would like to thank Pierre Thomas (Professor at ENS de Lyon) for providing us the pictures in Figures 1B and 1D. This protocol was adapted from the published study (Verger et al., 2018).

Competing interests

The authors declare no conflict of interest or competing interests.

References

  1. Bouton, S., Leboeuf, E., Mouille, G., Leydecker, M. T., Talbotec, J., Granier, F., Lahaye, M., Hofte, H. and Truong, H. N. (2002). QUASIMODO1 encodes a putative membrane-bound glycosyltransferase required for normal pectin synthesis and cell adhesion in Arabidopsis. Plant Cell 14(10): 2577-2590.
  2. Griffiths, L., Heap, M. J., Baud, P. and Schmittbuhl, J. (2017). Quantification of microcrack characteristics and implications for stiffness and strength of granite. Int J Rock Mech Min 100: 138-150.
  3. Joseph, P. V., Rabello, M. S., Mattoso, L. H. C., Joseph, K. and Thomas, S. (2002). Environmental effects on the degradation behaviour of sisal fibre reinforced polypropylene composites. Compos Sci Technol 62(10-11): 1357-1372.
  4. Kowallis, B. J. and Wang, H. F. (1983). Microcrack study of granitic cores from Illinois deep borehole UPH 3. J Geophys Res-Sol Ea 88(B9): 7373-7380.
  5. Kranz, R. L. (1983). Microcracks in rocks: A review. Tectonophysics 100(1-3): 449-480.
  6. Mori, S. and Burr, D. B. (1993). Increased intracortical remodeling following fatigue damage. Bone 14(2): 103-109.
  7. Verger, S., Long, Y., Boudaoud, A. and Hamant, O. (2018). A tension-adhesion feedback loop in plant epidermis. eLife 7: e34460.

简介

材料中的微裂纹反映了它们的机械性能。 因此,在诸如工程和地质学的许多领域中,这种裂缝的数量或方向的量化是必不可少的。 在生物学中,软组织裂缝可以反映粘连缺陷,对其模式的分析有助于推断器官和组织中张力的大小和方向。 在这里,我们描述了一种半自动方法,适用于分析拟南芥拟南芥幼苗表皮中发生的细胞分离。 我们的协议适用于任何表现出小裂缝的图像,因此也适用于分析动物组织和材料中出现的裂缝。
【背景】大多数材料都存在微裂纹;当施加反复的应力或温度波动时,它们的数量和程度通常会增加,从而导致材料疲劳并最终导致失效。微裂纹可以揭示材料所经历的主应力的大小和方向。该性质广泛用于机械工程和地质学(例如,Kowallis和Wang,1983; Kranz,1983; Joseph et al。,2002)。微裂纹也存在于生物结构中,如骨骼。与任何材料一样,它们会导致破裂(疲劳断裂)。这种裂缝还引发涉及成骨细胞和破骨细胞的信号级联,导致骨重建(例如,Mori和Burr,1993)。在软组织中也可以观察到裂缝,特别是由于细胞间粘附缺陷。这在植物组织中尤其明显,其中细胞不迁移或嵌入(例如,Bouton et al。,2002)。虽然已经开发了许多工具来研究材料科学中的裂缝,但它们通常适用于分析细长裂缝(例如,Griffiths et al。,2017);它们还没有被定制用于分析软生物组织中出现的裂缝。在这里,我们描述了一个检测和分析这种裂缝的管道。

我们的管道基于一个脚本,该脚本将区域显示出与细胞分离或裂缝相对应的清晰像素强度对比度。在我们的原始出版物中,我们用碘化丙锭染色拟南芥幼苗(参见Verger et al。,2018进行染色程序),这是一种特异性结合果胶中的果胶的荧光分子。细胞壁。洗掉染料后,细胞轮廓清晰标记。在野生型植物中,这揭示了连续的表皮,其中细胞彼此完全附着。然而,当对具有细胞粘附缺陷的突变体进行相同染色时,细胞之间的孔被揭示:细胞之间的明亮信号标记这些细胞之间出现的分离(Verger 等,,2018)。事实上,细胞和裂缝之间的对比强度足以检测和量化细胞分离。在对这些裂缝进行分割之后,对这些分割区域中的每一个进行主成分分析,得到各种信息:裂缝的面积以及其主要取向(裂缝的角度)和形状各向异性(来自特征值和向量)在裂缝形状的主成分分析中计算得出)。然后可以比较多个样品和样品系列的这些数据。原则上,其他染色方法可以用于任何类型的组织或材料,只要对比度足够强以使脚本检测裂缝即可。

关键字:裂缝, 细胞粘附, 图像分析, 机械性能, 张力, Github

软件

  1. 斐济计划( http://fiji.sc/ )
    基于ImageJ的基于开源插件的图像分析软件( https://imagej.nih.gov/ij/ < / A>)&NBSP;
  2. Python 2.7(用于通用编程的解释高级编程语言)( https://www.python.org/)
  3. Python模块:
    1. Matplotlib(2D可视化和数据绘图库)( https://matplotlib.org/ )
      鼻子(Python单元测试框架)( http://nose.readthedocs.io/ )
    2. Numpy(N维线性代数库)( www.numpy.org/ )
    3. Pandas(数据处理和分析库)( https://pandas.pydata.org/ )
    4. 枕头(图像处理库)( https://pillow.readthedocs.io/ )
    5. Pycircstat(循环统计库)( https://github.com/circstat/pycircstat/ )
    6. Scipy(科学计算库)( https://www.scipy.org/ )
  4. 开源包管理和环境管理系统
    1. Miniconda: https://conda.io/docs/
    2. LINUX: https://repo.continuum.io/miniconda/Miniconda2-latest -Linux-x86_64.sh
    3. MAC: https://repo.continuum.io/miniconda/Miniconda2-latest -MacOSX-x86_64.sh
    4. Windows: https://repo.continuum.io/miniconda/Miniconda2-latest -Windows-x86_64.exe
  5. 细胞分离图像分析管道(本协议中描述的图像分析脚本)( https://github.com/sverger/Cell_separation_analysis < / a>)
注意:有关安装软件2至5的详细信息,请参阅“过程”中的“安装过程”。

程序

  1. 图像采集
    我们开发了图像分析管道,用于检测和量化植物表皮中的细胞分离。这样的图像可以使用共聚焦显微镜并通过用荧光染料染色细胞壁或细胞轮廓,或通过成像表达细胞轮廓的荧光报告物的植物(通常是质膜上的蛋白质)来获得。 Z-Stacks可以在2D中获得并投影(例如,使用斐济,最大强度)。然而,至关重要的是获得图像,其中细胞和“裂缝”之间存在强烈对比(即,即细胞分离的区域。在图1C中,对应于细胞的区域具有可比性像素强度作为细胞之间的间隙,使得无法自动区分它们)。此外,由于我们的管道通过分割整个裂缝来工作,裂缝需要形成一个封闭的区域(例如,在图1D中,标记裂缝的关节(白色区域),彼此重叠,使得无法单独区分它们与我们的管道)。因此,我们的管道原则上可以处理任何包含明确闭合裂缝的2D灰度图像(参见图1中合适和不合适图像的示例)。


    图1.合适的图像类型。 A.来自 qua1-1 突变体的碘化丙啶染色光生长下胚轴的共聚焦图像堆栈的Z投影(最大强度)细胞粘附缺陷(参见Verger et al。,2018)。 B.岩石裂缝(信用照片:Pierre Thomas)。在A和B中,裂缝由高像素强度对比度标记并形成闭合域。 C.来自 qua1-1 pPDF1 :: mCit:KA1 (质膜报告基因)子叶表皮的共聚焦图像堆栈的Z-投影(最大强度)(参见Verger 等人。,2018)。虽然这种荧光报告器可以为我们的分析提供合适的图像,但在这种特殊情况下,裂缝和细胞含量之间的对比度太低而不能用我们的管道分割裂缝。 D.岩石裂缝的图片(信用照片:Pierre Thomas)。在这种情况下,裂缝不会形成闭合区域,因为它们中的大多数重叠。另外,像素强度在整个图像中是非常可变的,使得一些裂缝在像素强度上不表现出强烈的差异。白色箭头指向这些图像中的裂缝或细胞分离的例子。

  2. 先决条件:斐济“裂缝检测”的图像质量,预处理和阈值
    为了确定您的图像是否适合此图像分析管道,您需要确保裂缝按像素强度正确分割:
    1. 在斐济加载2D图像(软件1)。
    2. 将图像类型更改为8位。图像&gt;输入&gt; 8位(图2B)和/或从多通道(例如,RGB)图像中选择右通道(图像&gt;颜色&gt;分割通道)。
    3. 您可以选择使用“增强对比度...”功能增强图像的对比度(处理&gt;增强对比度...&gt;将“饱和像素”设置为0.3%)。
      注意:您可以使用不同的方法(例如,在图像采集期间增加曝光时间或激光强度)以获得足够的对比度来分割裂缝。
    4. 使用中值滤波器平滑图像以消除噪声(处理&gt;滤波器&gt;中位数... [图2C])。将半径设置为合适的值以减少噪声,而不会使图像模糊太多。在这里,您也可以使用不同的方法来降低图像的噪点,只要最后,您可以更平滑地检测裂缝。
    5. 使用“阈值...”工具,确定最佳区分裂缝与周围区域的合适阈值(图2D和2E)。
      关键步骤:根据图像的性质和质量,可能很难或不可能根据像素强度阈值对裂缝进行分割。如果您的大多数图像都处于这种情况,则此图像分析管道不适合您的研究。
    6. 如果能够正确地将裂缝与周围区域分开,则可以继续进行分析。将图像保存为.tif或.jpg(文件&gt;另存为&gt; Tiff ...或JPEG ...)并在名称末尾添加“_XXXthld”(其中XXX是先前在步骤B5中确定的适合分段的阈值裂缝(例如,“sample_1_162thld.tif”。见图2D-2F)。“thld”之前的值是稍后将用于图像分割的阈值(必须是三位数长)。
    7. 在进一步分析之前,您可以根据需要预处理任意数量的图像。如果它们被分组在一个文件夹中,它们都将被自动处理。请注意,为了量化和比较多个图像中的裂缝方向,图像应该处于一致的方向。相应的树状结构必须按如下方式组织:“主”目录(稍后在脚本中称为“updir”),包含子目录(例如,不同的突变体或生长条件),每个都包含所有相应的图像(图2F)。


      图2.斐济的图像预处理。 A.来自 qua1-1 的碘化丙啶染色的光生长下胚轴的共焦图像堆栈的Z投影(最大强度)具有细胞粘附缺陷的突变体(参见Verger et al。,2018)。公元前。图像在图像J中被预处理:图像被转换为8位(B)并且应用半径为3的中值模糊以减少噪声并且易于分割(C)。 d-E。阈值工具用于确定管道中分割的合适阈值。 (D)红色区域将被分割为裂缝。 (E)调整阈值以便适当地分割裂缝(例如,此处为值162)。 F.管道处理图像序列所需的文件树状结构的示例。 G.来自野生型幼苗的碘化丙啶染色的光生长下胚轴的共聚焦图像堆叠的Z-投影(最大强度),显示没有细胞粘附缺陷,因此不适合用我们的管道检测裂缝(参见Verger 等人,,2018)。 H-1。由于像素强度的差异局部较小(与面板AD中具有明亮细胞分离信号的 qua1-1 突变体不同),我们无法以阈值50(H)正确分割图像)或45(I)为例。

  3. 安装程序
    您需要在计算机上安装python(软件2)才能运行脚本(软件5)。该脚本已经设计好,因此应该使用python 2.7正常运行。如果已经在python环境中拥有所有必需的依赖项(软件3),或者在python环境中安装所有必需的依赖项,则可以直接运行该脚本。如果您没有安装Python,或者不希望干扰当前的Python环境,请使用miniconda(软件4)继续执行我们推荐的安装步骤,该软件将安装python和所有必需的依赖项。
    1. 从官方网站下载miniconda安装程序(链接在“软件4”下面,或在这里 LINUX , MAC 和 WINDOWS )。或者,您可以使用wget从终端(LINUX或MAC)执行此下载。打开一个新的终端窗口并运行以下命令行:
      对于LINUX:
      wget https://repo.continuum.io/miniconda/Miniconda2 -latest-Linux-x86_64.sh
      对于MAC:
      wget https://repo.continuum.io/miniconda/Miniconda2 -latest-MacOSX-x86_64.sh < / A>
    2. 通过运行安装程序安装miniconda:
      对于LINUX和MAC:打开一个新的终端窗口,导航到您下载安装程序的目录(很可能是在“Downloads”文件夹中。例如, cd path / to / Downlaods /。注意:输入指向文件夹“/ Downlaods /”的实际路径并运行:
      对于LINUX:
      bash Miniconda2-latest-Linux-x86_64.sh
      rm Miniconda2-latest-Linux-x86_64.sh

      对于MAC:
      bash Miniconda2-latest-MacOSX-x86_64.sh
      rm Miniconda2-latest-MacOSX-x86_64.sh

      对于WINDOWS:执行安装程序并按照说明操作。
      在安装过程中(LINUX,MAC和WINDOWS),您将被问到许多选择。您可以在询问时设置所选目录(例如,〜/ .miniconda)。当被要求将conda添加到PATH时,请务必回答“是”。
    3. 此时,你应该安装miniconda。通过关闭当前终端窗口并在新终端中运行conda来测试您的安装,以确保找到该命令:

      conda

    4. 从Github下载并提取“cell_separation_analysis”存储库(按照软件5中的链接,或此处)。在Github页面上,单击页面右侧的“克隆或下载”绿色按钮。然后下载并解压缩拉链。
      注意:此文件夹包含脚本(“Cell_separation_analysis.py”),miniconda的依赖项安装文件(“cell-sep-env.yml”)和测试样本图像(“Test_files”)。
    5. 在终端中,导航到已解压缩的“/ cell_separation_analysis-master”文件夹。

      cd path / to / Cell_separation_analysis-master /

      注意:输入指向文件夹“/ Cell_separation_analysis-master”的实际路径。
    6. 在同一个终端中,使用提供的YAML文件创建一个新的conda环境,该文件列出了所有软件依赖项:

      c onda env create -f cell-sep-env.yml

      注意:这将安装Python和所有必需的依赖项。可能需要几分钟才能完成安装。
    7. 安装完成后,在同一终端中激活环境:

      source activate cell-sep-env

      注意:每次打开新的终端窗口后,每次要使用单元格分离分析程序时都必须运行此命令。
    8. 然后,您可以通过在我们的测试映像上运行脚本来检查您的安装以及脚本是否正常运行。在终端中,运行:

      ipython

      这将启动ipython。在ipython中,导航到您下载的“/ cell_separation_analysis-master”文件夹:

      cd path / to / Cell_separation_analysis-master /

      在python控制台中,键入:

      %运行Cell_separation_analysis.py

      这应该运行脚本并在“/ Cell_separation_analysis-master”的“/ Test_files”文件夹中生成其输出。

  4. 运行脚本
    要使用您自己的图像运行脚本,您需要编辑脚本的“参数”部分。为此,请在文本编辑器中打开“Cell_separation_analysis.py”文件。向下滚动到名为“parameters”的部分,其中有7个条目可供修改(参见图3):


    图3.参数设置。 Cell_separation_analysis.py Python脚本在文本编辑器中打开,显示“参数”部分。可以根据您自己的要求修改参数,并在运行脚本之前保存文件。

    1. 设置放置要分析的图像的目录。请记住,您的文件夹必须以特定方式组织:“主”目录对应于“updir”,包含子目录(例如,不同的突变体或生长条件),每个都包含所有相应的图像(图2F)。您可以输入目录的完整路径(例如,/ Home / Path / to / updir /),或者只需输入./updir/(将“updir”替换为实际名称你的文件夹)。在后一种情况下,您必须在运行脚本之前导航到Ipython中“updir”的父文件夹(如步骤C8中所述)。
    2. 设置像素大小。这是为了进行裂缝区域分析所必需的。通常,对于共聚焦图像,通过在斐济加载原始图像并查看其属性(图像&gt;属性...&gt;像素宽度或高度),可以容易地找到该信息。长度单位应为微米。对于其他类型的图像,您需要使用内部比例(Analyze&gt; Set Scale ...)确定实际像素大小,例如在斐济。
    3. 确定最小和最大裂缝面积。此步骤对应于过滤器,因为它消除了太小的区域(“min_area_of_crack”,例如,背景噪声)以及图像的全局背景(“max_area_of_crack”,例如,组织周围的空白区域)。这些值以像素数表示。您可能需要尝试多次使用不同的值,以便根据经验确定正确的参数。
    4. 设置阈值类型。裂缝可以是比周围区域更高或更低的像素强度。因此,阈值可以设置为“min”或max“。 “max”将检测并分割信号强度较低的区域(即,裂缝比周围区域更暗),“min”将检测并分割信号强度较高的区域(即,裂缝比周围区域轻。)
    5. 如果您有多个样本类型(“Global_Output_Size”)和多个样本类型(Global_Polarhist_output),则可以决定运行更多全局分析。请参阅“数据分析”部分以确定是否应运行这些分析。这些默认设置为“False”,这意味着它们不会被执行。如果要执行“False”,请将“False”替换为“True”。
    6. 正确设置所有参数后,保存脚本并按步骤C8中所述运行。

数据分析

  1. 如背景部分所述,此脚本可以生成不同的输出。根据每个图像中检测到的裂缝,脚本将生成三个图像:图像的像素强度反转版本,具有分割区域的叠加的相同图像,以及具有各向异性和主角的叠加的相同图像。区域,直接保存为矢量PDF。它还将生成一个.csv文件,其中包含每个分段裂缝的标签号,中心位置,像素面积和微米平方,裂缝的主要方向(角度),形状各向异性,特征值和矢量。最后,对于每个图像,创建表示图像中裂缝方向分布的极坐标直方图,并将其保存为矢量PDF(参见上一步和图4中脚本测试中生成的输出)。


    图4.细胞分离分析管道的输出。 :一种。像素强度反转版图像(图2C)。 B.与(A)中的图像相同,具有使用不同颜色识别和标记的分割区域的叠加以便于可视化。 C.与(B)中的图像相同,具有由裂缝形状(红色十字形)的主成分分析得到的矢量的表示。为了改善视觉输出,将映射在图像上的特征向量乘以因子2和相应的特征值的平方根。 D.极坐标图,表示图像中裂缝方向的分布。将每个绘制角度的主特征值的平方根相加,以通过其相对权重对角度值进行归一化。在极坐标图中使用颜色图表示在每个直方图条中装箱的角度的相对数量(独立于它们的重量),其中黄色高且紫色低。

  2. 如果您处理来自一个样品系列(例如,技术或生物复制品)的多个图像,则可以输出其属性的摘要(请参阅步骤D5,“Global_Polarhist_Output = True”)。它将生成一个极坐标图,表示汇集在一起的样本序列的所有图像的裂缝方向分布,并将其保存为矢量PDF。它还将输出一个.txt文件,其中包含每个图像的圆形平均角度(介于0°和180°之间),合成矢量长度(裂缝协调方向性的估计,介于0和1之间;值为0表示裂缝取向均匀分布,而值1表示所有裂缝具有相同的取向)和裂缝形状的平均各向异性。在文本文件的末尾,生成样本系列的所有图像的全局分析:圆形平均角度,合成矢量长度和样本系列的所有图像的合并值的形状各向异性。它还将输出RAO间距测试的结果,该测试评估角度是否均匀分布(统计上没有优先角度方向),或者是否存在显着的角度偏差。
  3. 该脚本还提供了比较不同类型样品的可能性。您可以比较两组样品中裂缝的总面积(例如,突变体1的10幅图像与突变体2的10幅图像相比较)(参见步骤D5,“Global_Output_Size = True”)。这将对比较的样品进行统计测试,以确定一个样品系列的图像中的平均裂缝面积是否与另一个样品系列的不同。统计检验的选择取决于数据的正态性和方差。首先,Shapiro对种群正态性的检验在两个样本系列上进行。如果样品系列中的至少一个不具有正态分布的群体,则运行非参数Wilcoxon秩和检验以测试样品是否在统计学上不同。否则,如果两个样本系列都是正态分布的,则运行Bartlett对等方差的检验。如果两个样本系列具有相等的方差,则运行Student's t -test,否则运行Welch的 t -test来测试样本是否在统计上不同。这些测试的摘要保存为.txt文件,此比较的箱线图保存为矢量PDF。
  4. 然后,您可以根据需要使用包含所有原始数据的不同输出文件执行进一步分析。

笔记

  1. 该协议中最关键的步骤是检查图像的适用性,如“程序B”中所述。如果您的大多数图像未通过此测试,则此图像分析管道可能不适合您的研究。相反,重要的是要认识到,能够将100%的物体分割为视觉上的裂缝也是很少见的。然后,您应该确定您的案例中可接受的收益率。
  2. 按照我们推荐的安装程序,您原则上应该能够在任何系统(LINUX,MAC和WINDOWS)上运行我们的图像分析管道。但是,到目前为止,该脚本仅在运行Ubuntu 14.04和Python 2.7的计算机上进行了测试。我们建议使用我们的样本图像测试宏(步骤C8),并检查输出图像是否与我们的原始出版物(Verger et al。,2018)中报告的内容相似。

致谢

这项工作得到了欧洲研究理事会的支持(ERC-2013-CoG-615739“MechanoDevo”)。我们要感谢Pierre Thomas(ENS de Lyon教授)向我们提供了图1B和1D中的图片。该方案改编自已发表的研究(Verger et al。,2018)。

利益争夺

作者声明没有利益冲突或竞争利益。

参考

  1. Bouton,S.,Leboeuf,E.,Mouille,G.,Leydecker,M。T.,Talbotec,J.,Granier,F.,Lahaye,M.,Hofte,H。和Truong,H。N.(2002)。 QUASIMODO1编码拟南芥中正常果胶合成和细胞粘附所需的假定膜结合糖基转移酶。 植物细胞 14(10):2577-2590。
  2. Griffiths,L.,Heap,M。J.,Baud,P。和Schmittbuhl,J。(2017)。 微裂纹特征的量化以及对花岗岩刚度和强度的影响。 Int J Rock Mech Min 100:138-150。
  3. Joseph,P.V.,Rabello,M.S。,Mattoso,L.H.C.,Joseph,K。和Thomas,S。(2002)。 对剑麻纤维增强聚丙烯复合材料降解行为的环境影响。 复合材料Sci Technol 62(10-11):1357-1372。
  4. Kowallis,B。J.和Wang,H。F.(1983)。 伊利诺伊州深井UPH 3花岗岩岩心的微裂缝研究。 J Geophys Res-Sol Ea 88(B9):7373-7380。
  5. Kranz,R。L.(1983)。 岩石中的微裂纹:评论。 Tectonophysics 100( 1-3):449-480。
  6. Mori,S。和Burr,D.B。(1993)。 疲劳损伤后皮质内重塑增加。 Bone 14( 2):103-109。
  7. Verger,S.,Long,Y.,Boudaoud,A。和Hamant,O。(2018)。 植物表皮中的张力粘附反馈回路。 eLife 7:e34460。
  • English
  • 中文翻译
免责声明 × 为了向广大用户提供经翻译的内容,www.bio-protocol.org 采用人工翻译与计算机翻译结合的技术翻译了本文章。基于计算机的翻译质量再高,也不及 100% 的人工翻译的质量。为此,我们始终建议用户参考原始英文版本。 Bio-protocol., LLC对翻译版本的准确性不承担任何责任。
Copyright Verger et al. This article is distributed under the terms of the Creative Commons Attribution License (CC BY 4.0).
引用: Readers should cite both the Bio-protocol article and the original research article where this protocol was used:
  1. Verger, S., Cerutti, G. and Hamant, O. (2018). An Image Analysis Pipeline to Quantify Emerging Cracks in Materials or Adhesion Defects in Living Tissues. Bio-protocol 8(19): e3036. DOI: 10.21769/BioProtoc.3036.
  2. Verger, S., Long, Y., Boudaoud, A. and Hamant, O. (2018). A tension-adhesion feedback loop in plant epidermis. eLife 7: e34460.
提问与回复
提交问题/评论即表示您同意遵守我们的服务条款。如果您发现恶意或不符合我们的条款的言论,请联系我们:eb@bio-protocol.org。

如果您对本实验方案有任何疑问/意见, 强烈建议您发布在此处。我们将邀请本文作者以及部分用户回答您的问题/意见。为了作者与用户间沟通流畅(作者能准确理解您所遇到的问题并给与正确的建议),我们鼓励用户用图片的形式来说明遇到的问题。

如果您对本实验方案有任何疑问/意见, 强烈建议您发布在此处。我们将邀请本文作者以及部分用户回答您的问题/意见。为了作者与用户间沟通流畅(作者能准确理解您所遇到的问题并给与正确的建议),我们鼓励用户用图片的形式来说明遇到的问题。