Automated Layer Analysis (ALAn): An Image Analysis Tool for the Unbiased Characterization of Mammalian Epithelial Architecture in Culture

Automated Layer Analysis (ALAn): An Image Analysis Tool for the Unbiased Characterization of Mammalian Epithelial Architecture in Culture 

Christian Cammarota^{, 1}, Dan T Bergstralh^{, 1,2, *, $}, and Tara M Finegan^{, 2,*, $}

¹ Department of Physics & Astronomy, University of Rochester, Rochester, NY 14627; 

² Department of Biology, University of Rochester, Rochester, NY 14627; 

^$ Current/Present address: Division of Biological Sciences, University of Missouri, Columbia, MO 65211.

*For correspondence: dan.bergstralh@missouri.edu and tara.finegan@missouri.edu. 

Abstract 

Cultured mammalian cells are a common model system for the study of epithelial biology and mechanics. Epithelia are often considered as ‘pseudo-two dimensional’ and thus imaged and analyzed with respect to the apical tissue surface. We found that the three-dimensional architecture of epithelial monolayers can vary widely even within small culture wells, and that layers that appear organized in plane of the tissue can show gross disorganization in the apical-basal plane. Here we present a detailed protocol for the use of our image analysis pipeline, Automated Layer Analysis (ALAn), developed to quantitatively characterize the architecture of cultured epithelial layers. ALAn is based on a set of rules that are applied to the spatial distributions of DNA and Actin signals in the apical-basal (depth) dimension of cultured layers obtained from imaging cultured cell layers using a confocal microscope. ALAn facilitates reproducibility across experiments, investigations, and labs, providing users with a quantitative, unbiased characterization of epithelial architecture and maturity.

Key Features

• This protocol was developed to spatially analyze epithelial monolayers in an automated and unbiased fashion, first published in Dawney and Cammarota et al. 2023 and advanced in Cammarota et al. 2023.

• ALAn requires 2 inputs: the spatial distributions of nuclei and Actin in cultured cells obtained using confocal fluorescence microscopy.

• This protocol is optimized for use in Marbin-Darby Canine Kidney (MDCK) cells and has been successfully applied to characterize human MCF-7 mammary gland derived and Caco-2 colon carcinoma cells.

• This protocol utilizes Imaris software to segment nuclei. Use of Imaris is not strictly necessary, and the protocol may be adapted such that an alternative method is used to segment nuclei. ALAn requires the centroid coordinates and volume of nuclei.

Graphical Overview            

Keywords: Epithelia, Cell culture, Image Analysis, 3D Architecture

Background

We developed a tool for the unbiased analysis of layer architecture (Automated Layer Analysis or ALAn) for two reasons. Firstly, cultured epithelial architecture can vary drastically between experiments, and we found this can be the case even across a single 1 mm² culture well (Dawney et al., 2023). Secondly, epithelial tissues are commonly considered to be ‘pseudo-two-dimensional,’ meaning that architecture is primarily considered with respect to the tissue plane/surface. We found that this perspective is not always predictive of architecture with respect to tissue depth (the apical-basal axis) (Dawney et al., 2023). We are interested in the mechanical and biological parameters that contribute to epithelial layer architecture and we developed this tool to quantify architecture development over time, and test the effect of plating density, culture time and presence of cadherin-mediated adhesion on architecture. Due to the variability in the shape of layers and the complexity of attempting to perform 3D analysis manually, it was important to us to develop an unbiased and automated tool which requires as little user input as possible. As a consequence of developing ALAn, we identified a developmental series of layer architectures that progress through the process of 'epithelialization', whereby individual cells form a tissue layer (Dawney et al., 2023, Cammarota et al., 2023). Our quantifications of cell height, cell density, and circularity show distinct morphological regimes for each architecture.

Several tools exist for the analysis of epithelial topology in the plane of the tissue (Etournay et al., 2016; Farrell, Weitz, Magnasco, and Zallen, 2017; Guirao and Bellaïche, 2017; Vicente-Munuera et al., 2020). As far as we are aware, ALAn is the first to characterize and quantify the apical-basal topology of cultured layers. To do so, the tool makes use of two pieces of spatial information: 1) a projection of actin signal intensity and 2) the positions of cell nuclei in the apical-basal dimension. Using this information ALAn outputs layer density and layer height and assigns each layer to one of four “organized” categories representing a developmental series (Immature, Intermediate A, Intermediate B and Mature) or a fifth category of exclusion (Disorganized) (Dawney et al., 2023 and Cammarota et al., 2023). ALAn also utilizes existing Python libraries to output in-plane quantifications of cell shape.

Reagents

1. Dulbecco's Phosphate Buffered Saline (dPBS) (Sigma-Aldrich, SKU D1408).

2. Phosphate Buffered Saline (PBS) (such as tablets from ThermoFisher, Invitrogen Cat. No. 003002).

3. Paraformaldehyde, 37% (Fisher Scientific, Cat No. F79-500).

4. Tween 20 Detergent (Sigma, SKU 11332465001).

5. Vectashield antifade mountain medium with DAPI (Vector Laboratories, SKU H-1200).

6. Fluorescein Phalloidin (ThermoFisher, Invitrogen, Cat. No. F432).

Laboratory Supplies

1. Collagen IV coated 8-well µ-slide, #1.5 polymer coverslip, sterilized (Ibidi, Cat. No. 80822).

Equipment

1. Dragonfly 500 Spinning Disc Confocal Unit coupled to a Nikon Ti-E inverted microscope and Zyla 4.2 sCMOS camera. 405 nm and 488 nm laser lines. 40x NA 1.15 water immersion lens (Andor).

2. Computer workstation with a minimum 8 GB RAM, AMD or NVIDIA 2GB Graphics card and Monitor with 1280 x 1024 pixels.

Software 

1. Fusion (Version 2.3, Andor) – optional. Used for image acquisition.

2. Imaris (Versions 9.7, 9.8, 9.9 or 10.0, Oxford Instruments) – optional. Used for nuclei segmentation.

3. FIJI, ImageJ2 distribution (Version 2.14.0) (Rueden et al., 2017; Schindelin et al., 2012).

4. Microsoft Excel (Version independent).

5. Anaconda distribution (Version 3 - 2023.07-2).

Code

1. Code can be obtained from Github: https://github.com/Bergstralh-Lab/ALAn

   1. ‘Collation_Imaris2ALAn.ipynb’

   2. ‘ALAn_v3.ipynb’

Procedure

A. Cell culture and imaging (suggested approach).

1. Cell culture, fixation and staining.

1. Culture mammalian cells in Ibidi wells. 

2. After the desired period of culture time, remove the culture media and wash cells with dPBS.

3. Add a solution of 4% paraformaldehyde diluted in PBS with 2% Tween to the well and incubate for 10 minutes.

4. Wash the cells three times for 10 minutes with PBS with 0.2% Tween. 

5. Dilute Fluorescein Phalloidin 1:500 in PBS with 0.2% Tween and incubate for at least 12 hours at 4 ^oC.

6. Add approximately 500 µl (or at least enough to fully cover the cell layer) of Vectashield plus DAPI to the well.

2. Confocal imaging.

1. Cells can be imaged directly in the Ibidi wells through the #1.5 polymer coverslip bottom using an inverted confocal microscope.

   1. Image a region in XY with an area of at least 60 µm x 60 µm with the maximum possible resolution (minimum of 512 x 512 pixels).

2. Take confocal z-stacks at a spacing of at least 0.5 µm, preferably Nyquist sampling, starting from beneath the bottom of the chamber slide where no actin (FITC) signal can be detected and ending above the layer where no more actin signal can be detected.

3. Image Actin stained by FITC conjugated Phalloidin using the 488 nm line, and DNA stained by DAPI using the 405 lines. Use standard imaging technique to set acquisition parameters to ensure that you are taking advantage of the full dynamic imaging range and not saturating pixels. Note: Any saturation will invalidate the use of ALAn. 

B. Preparation of data for input into ALAn.

1. Segmentation of Nuclei using Imaris.

a. Save files using the following name format: ‘Identfier1_Identifier2_Identifier3_Identifier4.ims’ where the Identifiers are the Experimental parameters that you would like to log.

b. Load the files to be analyzed into the Imaris Arena using the ‘Observe Path’ function.

c. Select the image to be analyzed and in the Surpass view select the ‘Add new Cells’ function to begin segmentation.

d. Use the default settings to segment nuclei and cell borders (although we will skip the cell borders segmentation), and deselect ‘Segment only a Region of Interest’. 

e. Use the blue arrow to move to Nuclei Detection. Do not use the green arrow. 

f. Set your source channel to the DNA- stained channel (DAPI stained 405 channel using the above protocol).

g. Set the expected Nucleus Diameter to 7.3 µm. 

h. Ensure that the smooth feature should be auto-selected and set to a width of 1/10^th the nucleus diameter.

i. Select ‘Split Nuclei by Seed Points’. 

j. Use the blue arrow to move to Filter Nuclei Seed Points.

k. Set the minimum ‘Quality’ factor to 4. Note: This value is set to be relatively low because at this stage of the segmentation process it is best to find as many nuclei as possible. It is easy to remove nuclei in ALAn, but impossible to add more.

l. Use the blue arrow to move to Nuclei Threshold. 

m. Set the ‘Nucleus Threshold’ value to a suggested value of 120. Check to ensure that this value does not include the peak on the histogram that corresponds to background signal peak. If it does, then adjust the threshold up until it is past the peak.

n. Use the blue arrow to move to Filter Nuclei.

o. When the Thresholding ends, click the orange ‘X’. Do not do any further filtering at this point unless you disagree with the segmentation. 

p. Save the segmented summary statistics by first selecting the ‘Statistics’ option, then selecting the ‘Export All Statistics to File’ option. It is helpful to save all Imaris segmentation files in one directory, separate from the original .ims image files. The Imaris output creates a folder of files that contain the positional and volumetric information of segmented nuclei. 

q. Download ‘Collation_Imaris2ALAn.ipynb’ from the Bergstralh lab Github.

r. Open ‘Jupyter Notebook’ from Anaconda Navigator. This will open Jupyter Notebook in a new browser window of your default browser. A navigation menu of the files on your computer is shown.

s. Navigate to the location of the ‘Collation_Imaris2ALAn.ipynb’ file on your computer and open this notebook.

t. Run the code in Cell 1, which imports the necessary Python packages.

u. Define the input directory path, where the Imaris Statistics files are located, on Cell 3, Line 6.

v. Define the output directory name on Cell 3, Line 11 and the directory location of this directory on Cell 3, Lin4. This will be the location that .csv files containing the necessary nuclei positions will be saved.

w. Run through all code in the ‘Collation_Imaris2ALAn’ file.

2. Required data if you segment nuclei using a method not using Imaris as described in Section 1 

1. Segment nuclear positions and extract information about the volume and centroid of nuclei in the image file.

2. Generate a .csv file with 4 columns containing: Column 1- X coordinate of nucleus center; Column 2- Y coordinate of nucleus center; Column 3- Z coordinate of nucleus center; Column 4- Volume of Nucleus. Rows correspond to each segmented nucleus. 

3. Add Heading titles to row 1. A1 = Nucleus Position X; B1 = Nucleus Position Y; C1 = Nucleus Position Z; D1 = Nucleus Volume.

3. Scale XY confocal slices to 512 x 512 pixels. 

1. This can be done in Imaris in the Edit Menu by selecting ‘Resample 3D’ and resizing the X and Y values as 512. ‘Make sure to select Fixed Ratio X/Y’ not Fixed Ratio ‘X/Y/Z’.

4. Determine the spatial bounds of the image

1. Open the original image file using FIJI.

2. Select ‘Show Info’ from the Image menu.

3. Identify the X, Y and Z spatial bounds of the image file in the Show Info pane- values ExtMax0 through ExtMin2. ExtMax0 =X max; ExtMax1 = Y max; ExtMax2 = Z max; ExtMin0 = X min; ExtMin1 = Y min; ExtMin2 = Z min. 

4. Open the .csv file containing the segmented nuclei positions using Microsoft Excel.

5. Name Column E “Positions”.

6. Paste values ExtMax0 through ExtMin2 into cells E2-E7. 

5. Save confocal stacks as .tiff files in same folder as the corresponding .csv files containing the nuclei segmentation information. This can be done by opening proprietary file images in FIJI and Selecting ‘Save as’ > ‘Tiff’ in the File menu.

6. Ensure that the corresponding .csv files containing the nuclear and layer positional information, and the .tiff files containing the confocal stack images have the SAME NAME. ALAn will read in both files and perform analysis of these partner files. 

C. Use ALAn to batch process and analyze data.

1. Open ‘Jupyter Notebook’ from Anaconda Navigator. This will open Jupyter Notebook in a new browser window of your default browser. A navigation menu of the files on your computer is shown.

2. Navigate to the location of the ‘ALAn_v3.ipynb’ file on your computer and open this Notebook.

3. Run the code in Cell 2, which imports the necessary Python packages.

4. Change the code on Cell 3, Line 7 to the path where your input partner .csv and .tiff files are located.
5. Run the code in Cell 3. The data will be imported into two lists. All of the information in the .csv files will be converted into a list of pandas DataFrames (list_of_dfs), and all of the .tiffs will be read in as a list of 4D numpy arrays (list_of_unshuffled_images). Any .tiff or .csv file that does not have a partner file with the same name will not be added to these lists.  A third list (list_of_names) will be generated to keep track of which image name is stored in each of the elements. 
6. Check that the file names, number of input images, and dimensions of the three lists are the same, shown in the output text below the cell.
7. Run the code in Cell 4. This will ‘load’ all of the functions of ALAn. Note: A description of each function and a description of its inputs and outputs are in the README.md file on the Bergstralh-Lab / ALAn Github page.
8. Change the input parameters of the batch_process function on code line Cell 5, line 4.
   1.  Firstly, define the channel that represents the image Actin signal in the .tiff file (this can be determined by opening the .tiff in FIJI)  (0 = first channel, 1 = second channel etc.) 
   2. Secondly, change the name that you would like to save you output .csv file as. 
9. Run the code in Cell 5. This will process all of the input images as a batch. A timer will output the time taken to run this function.
10. Open the output .csv file, which has the name as defined on Cell 5, line 4 as the last parameter of the batch_process function. The directory will be the same as your input files. Check that a valid output has been generated. The basic outputs of ALAn are:

Column A – Element number used by ALAn for each input.

Column B – Root name of input .tiff and .csv files.
Column C – Layer Determination – ALAn makes a categorization of the layer into the following categories: Disorganized, Immature, Intermediate A, Intermediate B and Mature. For a full explanation of these categories see (Dawney et al., 2023) and (Cammarota et al., 2023).

Column D – Number of cells identified in the basal tissue layer.

Column E  - Number of cells identified above the basal tissue layer.

Column F – Total number of cells (Sum of Column D and E).

Column G – Cell density of identified tissue layer per unit culture area.

Column H  - Height of identified tissue layer.

Column I – Percentage of total cells above the layer.

Column J – Average area of cells in the plane of the tissue.

Column K – Average cell perimeter in the plane of the tissue.

Column L – Average cell circularity in the plane of the tissue.

D. Use of ALAn to make diagnostic/informative plots and visuals. For any dataset, the functions of ALAn can be used to generate diagnostic and data outputs and figures. Here we provide a protocol for running the functions that provide the most informative figure outputs for the analysis of the data. An optional input of save_name = “string” allows you to save this plots as a .pdf.. Figures should be generated in an embedded figure in the Jupyter notebook (defined by code on Cell 2, Line 16: “%matplotlib notebook”).

1. Determine the element number for the dataset of interest from output .csv file of the batch_process function.
2. In a new cell in the Jupyter notebook, define your dataset of interest by typing “i=n’ where n is the element number of your chosen dataset.
3. In any subsequent cell, you can run the clear_debris function by typing “clear_debris(list_of_dfs[i], plot = True)” and run the cell. This function provides data on which objects identified by the nuclear segmentation input are considered as ‘real’ cells and used for analysis. A histogram showing the distribution of nuclear volumes in µm³ (x-axis) against frequency of these volumes is output (Fig 1A). The objects discarded as debris are shown in blue, and the nuclei identified as valid are orange. A pandas DataFrame containing the nuclei identified as true datapoints is output (df_cleared) and printed below the figure. 
4. In any subsequent cell, you can run the nuclear_centroid_actin_overlay function by typing “nuclear_centroid_actin_overlay(list_of_dfs[i], list_of_unshuffled_images[i])”. Use this function to test the input nuclear segmentation and clear_debris functions to ensure that the nuclei used by ALAn are reasonable. A figure showing the nuclear centroid positions in the tissue plane are overlayed with the actin channel to view the position of nuclei within each cell (Fig 1B). Black dots indicate ‘real’ nuclei, red dots indicate ‘false’ nuclei as defined by clear_debris. 
5. If you are not happy with the cutoff volume of ‘real’ nuclei, you can change the value that is used to define what a ‘real’ nucleus is in the clear_debris function. This is currently set to 1.5 standard deviations below the average nuclear volume (Defined on Cell 3, line 97).
6. In any subsequent cell, you can run the nuclei_distribution function by typing “nuclei_distribution(list_of_dfs[i], list_of_unshuffled_images[i], plot = True). This function determines if the layer is organized into a monolayer or not and returns a classification as ‘Organized’ or ‘Disorganized’. A plot showing the projected number of nuclei and actin intensity (double Y axis) across the depth of the layer, Z position in µm (X axis) (Fig 1C). This is a useful figure to interrogate the organization of cells across the depth of the tissue. 
7. In any subsequent cell, you can run the find_shoulders function (Fig 1D) by typing “find_shoulders(list_of_dfs[i], list_of_unshuffled_images[i], plot = True)”. This function determines whether the plot of actin intensity across the depth of the tissue has a ‘shoulder’. The output is a plot of Actin Intensity and the First Derivative of the Actin Profile (double Y-axis) against the depth of tissue, Z position(µm) (X-axis). These plots are used to identify the correct subcategory of ‘organized’ in terms of the developmental series of ‘epithelialization’. ‘Immature’ layers are defined by two rules that do not pertain to this function, all other organized layers are identified by this function. ‘Intermediate A’ layers have a one-peaked actin intensity derivative. ‘Intermediate B’ and ‘Mature’ layers have a two-peaked actin intensity derivative. By taking the ratio between the height of the apical peak to the lateral peak, we identify the difference between the two layer types. A ratio less than one is classified as an ‘Intermediate B’ layer, while a ratio of greater than one is classified as a ‘Mature’ layer. 
8. In any subsequent cell, you can run the terrain_map function (Fig 1E) by typing “terrain_map(list_of_dfs[i], list_of_unshuffled_images[i])”. This function plots all of the nuclei in an confocal stack on the XY plane, positioned by their XY position, sized based on nuclear size, and colored based on z position of the nucleus (Fig 1E). This allows for visualization of the position of extralayer cells or disorganized regions within the field of view. 
9. In any subsequent cell, you can run the sub_classify function (Fig 1F) by typing “sub_classify(list_of_dfs[i], list_of_unshuffled_images[i], plot = True)”. This function breaks the full XY field of view into the smallest classifiable sections (100px x 100 px), and outputs an ALAn classification of each of the subsections ‘all_sections_analyzed’ list. This function outputs a figure showing the full field of view divided into a grid of 25 squares colored by layer type (Fig 1F).
10. In any subsequent cell, you can run the xy_segmentation function by typing “xy_segmentation(list_of_dfs[i], list_of_unshuffled_images[i], plot = True)”. This only works for ‘Organized’ layer types. This function segments the basal actin of the underlying organized cell layer in the plane of the tissue to output shape parameters of cells in the plane of the tissue. This function uses region_props from the scikit-image package to extract shape parameters. The function will output 4 figures showing the stages of segmentation as compared to the projected actin signal. This output is useful to check that the actin signal has been correctly segmented for in-plane cell shape analysis. This function CAN output images for ‘Disorganized’ layers, it is imperative that you make sure to check that you are not running this function on a ‘Disorganized’ layer. 
     

Fig 1. Diagnostic and informative plots generated by ALAn.

A. The clear_debris function outputs a histogram showing the frequency of segmented objects with respect to their volume that was input into ALAn. The orange bars represent nuclei identified as ‘real’, and the blue bars represent ‘debris’ that is removed from the analysis. A dotted line indicates the threshold volume above which a nuclei is considered real. B) The nuclear_centroid_actin_overlay function outputs an image that displays the centroid position of the segmented nuclei overlaid onto the .tiff image of the actin signal projected on the XY plane of the tissue. C) The nuclei_distribution function outputs a plot showing both the projected signal of actin (red) and the nuclear distribution (blue), fit with a Gaussian, with respect to the depth of the tissue. The tissue layer top, bottom and the peak position of the actin signal are shown by dotted lines. D) The find_shoulders function outputs a plot of the actin signal (red), and the first derivative of this plot (blue) with respect to the tissue depth. The tissue layer top, bottom and the peak position of the actin signal are shown by dotted lines. E) The terrain_map function outputs a map of the position of the centroids of nuclei projected onto the XY plane of the tissue, displayed as dots. The size of the dots show realtive volumes of nuclei. Color of the dots represent relative positions of the nuclei with respect to the depth of the tissue. Color scale goes from Light orange = most basal to Purple = most apical. F) The sub_classify function outputs a map representing the XY field of view of the input 512 x 512 pixel .tiffs broken into 100 x 100 pixel sections, color-coded by the layer category as determined by the original version of ALAn. 'lawngreen' = Immature, ‘forestgreen’ = Intermediate, 'steelblue’ = Mature, 'orange’ = Disorganized.

Validation of Protocol

This work was developed to investigate the effects of cell plating density and time, and characterize the architectural development of MDCK monolayers in publications (Dawney et al., 2023) and (Cammarota et al., 2023). We developed an unbiased and thorough imaging strategy to image a comprehensive map of epithelial architecture, imaging the same 16 representative 300 x 300 µm regions from the 1 mm² wells. We developed ALAn using a dataset of over 19,000 cells across 351 imaged regions from 4 repeats of 3 plating densities and 2 repeats of the 16 regions. We further validated our strategy and broad application of the tool to analyze epithelial architecture using Caco-2 and MCF-7 cells.

General notes

• In principle, this tool can be applied to cultured layers imaged live. ALAn requires homogenous staining of actin across the layer. We have attempted to apply ALAn to MDCK layers imaged live using Hoechst or H2B- mCherry to segment nuclei and SiR-actin and LifeAct-GFP to visualize actin and had limited success. We found SiR-actin and LifeAct-GFP signal to be unpredictable in our hands and thus cell borders were not always detectable in our samples, particularly at lower densities. The SiR-actin profile generally reflects the layer architecture profiles distinguished from Phalloidin staining but background signal above the layer changes the projected actin profile. We anticipate that if alternative dyes, particularly those used for live imaging, the rules of layer category assignment will need to be modified and optimized. Due to our difficulties in obtaining high-quality movie data, we have not extended the ALAn code such that it can automatically analyze 4D movies. However, high quality frames from 4D movies can be readily input into ALAn to analyze architecture changes across time.

Acknowledgements

We thank the lab of Rene Marc Mege for requesting a detailed version of this protocol. We are grateful to past and present members of the Finegan-Bergstralh lab for their comments. This work was supported by a National Science Foundation CAREER award (PI: Bergstralh) and National Institutes of Health Grant R01GM125839 (PI: Bergstralh).

Competing interests

The authors declare no competing interests. 

References

Cammarota, C., Dawney, N. S., Bellomio, P. M., Jüng, M., Fletcher, A. G., Finegan, T. M., and Bergstralh, D. T. (2023, May 9). The Mechanical Influence of Densification on Initial Epithelial Architecture (p. 2023.05.07.539758). p. 2023.05.07.539758. bioRxiv. https://doi.org/10.1101/2023.05.07.539758

Dawney, N. S., Cammarota, C., Jia, Q., Shipley, A., Glichowski, J. A., Vasandani, M., … Bergstralh, D. T. (2023). A novel tool for the unbiased characterization of epithelial monolayer development in culture. Molecular Biology of the Cell, 34(4), ar25. https://doi.org/10.1091/mbc.E22-04-0121

Etournay, R., Merkel, M., Popović, M., Brandl, H., Dye, N. A., Aigouy, B., … Jülicher, F. (2016). TissueMiner: A multiscale analysis toolkit to quantify how cellular processes create tissue dynamics. ELife, 5, e14334. https://doi.org/10.7554/eLife.14334

Farrell, D. L., Weitz, O., Magnasco, M. O., and Zallen, J. A. (2017). SEGGA: A toolset for rapid automated analysis of epithelial cell polarity and dynamics. Development, 144(9), 1725–1734. https://doi.org/10.1242/dev.146837

Guirao, B., and Bellaïche, Y. (2017). Biomechanics of cell rearrangements in Drosophila. Current Opinion in Cell Biology, 48, 113–124. https://doi.org/10.1016/j.ceb.2017.06.004

Rueden, C. T., Schindelin, J., Hiner, M. C., DeZonia, B. E., Walter, A. E., Arena, E. T., and Eliceiri, K. W. (2017). ImageJ2: ImageJ for the next generation of scientific image data. BMC Bioinformatics, 18(1), 529. https://doi.org/10.1186/s12859-017-1934-z

Schindelin, J., Arganda-Carreras, I., Frise, E., Kaynig, V., Longair, M., Pietzsch, T., … Cardona, A. (2012). Fiji: An open-source platform for biological-image analysis. Nature Methods, 9(7), 676–682.

Vicente-Munuera, P., Gómez-Gálvez, P., Tetley, R. J., Forja, C., Tagua, A., Letrán, M., … Escudero, L. M. (2020). EpiGraph: An open-source platform to quantify epithelial organization. Bioinformatics, 36(4), 1314–1316. https://doi.org/10.1093/bioinformatics/btz683