Materials and Reagents

Herbarium specimens and/or fresh leaf material

Equipment

Computer that can run Microsoft^® Windows^® XP (or later) and/or Mac^® OS X^® 10.4 (or later)
Flatbed photo scanner (Epson Expression, model: 10000XL )

Software

Adobe^® Photoshop^® CS4 (or later)
Epson^® Scan Utility v3.4.9.6 (https://support.epson.com/)
Java^TM (https://java.com/)
ImageJ (https://imagej.nih.gov/ij/)
Microsoft^® Excel^® 2011 (or later)
SHAPE v1.3 (http://lbm.ab.a.u-tokyo.ac.jp/~iwata/shape/)
SHAPE is built for Windows but if using a Mac, Wine and Winebottler (http://winebottler.kronenberg.org/) are required
R (https://r-project.org/)
Packages: ‘shapes,’ ‘ggplot2,’ ‘devtools,’ ‘ellipse,’ and ‘roxygen2’
RStudio (https://rstudio.com/products/rstudio/)
RStudio is an optional user interface for R

Procedure

Note: Examples of R scripts and ImageJ macros referenced throughout the protocol can be freely downloaded from GitHub (link: https://github.com/htsvoboda/LeafGeometricMorphometrics.git; Note 1).

Scanning fresh leaves or herbarium specimens
1. For fresh leaves: Place leaves, with petioles removed, flat on the scanner bed. Multiple leaves can be placed on the scanning bed at once, so long as they do not overlap. If the background of the scanner is not already white, place a white piece of paper on top of the leaves; subsequent analyses work best with a solid white background (Note 2).
2. For herbarium specimens: carefully place the entire sheet face-down on the scanner bed. Any loose material should be placed in a fragment packet on the sheet.
3. Scans should be saved as .jpg or .tif files and named with a unique identifier (e.g., accession number).
  Note: It is best to scan at 400 dpi (or a higher resolution) as this improves the quality of the image.
Acquiring and exporting landmarks
1. Create a spreadsheet in Excel with the following column names: ‘order,’ ‘label,’ ‘x,’ and ‘y.’ This file will be referred to as the ‘master spreadsheet.’
2. Open ImageJ
  For the first use: Select ‘Analyze > Set Measurements…’ and check only the ‘Display label’ checkbox. ‘Redirect’ and decimal place parameters can be left at their defaults.
3. From the ImageJ menu bar, select the ‘Point selection’ tool (Figure 1; Note 3).
  
  Figure 1. The ImageJ menu bar. Here, the ‘Point selection’ tool button is selected.
4. Open the first image to be landmarked. In the menu bar, select ‘File > Open...’
5. Using the ‘Point selection’ tool in multi-point mode, begin placing landmarks on your predetermined landmark points (see Figure 2; Notes 4 and 5).
  
  Figure 2. Placement of landmarks on some representative leaves. A. Seventeen landmarks placed on two fresh leaves of Vitis riparia; B. Six landmarks placed on a leaf of Passiflora campanulata from a herbarium specimen.
6. Once all landmarks have been applied to a leaf (or to multiple leaves per image), view the landmark data. In the menu bar, select ‘Analyze > Measure.’ A new window, called ‘Results’ (Figure 3), will appear with the x, y coordinates for each landmark.
  
  Figure 3. Example of the ‘Results’ window produced by ImageJ. This is viewed by selecting ‘Analyze > Measure’ in the menu bar.
7. Copy (CTRL + C) and paste (CTRL + V) these data from the ‘Results’ window directly into the ‘master spreadsheet’ (see Figure 4).
  
  Figure 4. Example format of the ‘master spreadsheet’
8. Close the ‘Results’ and image windows and repeat this process with each image file until all leaves have been landmarked. ‘File > Open Next’ can be used if all of the images are in the same folder.
9. Landmark coordinates for every leaf will be pasted as-is into the ‘master spreadsheet’. The column ‘order’ may need to be adjusted at the end to reflect a continuous numerical set.
10. Before analyzing the landmark data, the ‘master spreadsheet’ will need to be converted from Excel format (.xlsx) to a Tab Delimited Text format (.txt).
11. Data within the .txt file will need to be reformatted such that the rows are single leaves and the adjacent columns represent the landmark data (x, y coordinates) in sequential order. This can be done in R using the code from GitHub (file: ‘Protocol_stepB-11.R’; R Core Team, 2017).
12. Import the reformatted file (referred to as ‘reformatted.txt’) into Excel to verify the data was properly written.
13. Supplemental columns and information (i.e., species, genotype, sex, etc.) can be added at this point to help with downstream analyses (Figure 5).
  
  Figure 5. The reformatted ‘master spreadsheet’ (‘reformatted.txt’) as seen in Excel. The spreadsheet now displays each row as a leaf with its landmarks distributed across the columns. Additional information can be added to the reformatted spreadsheet that may be useful in downstream analyses (e.g., columns B-F).
14. Check landmarks for accuracy. This can be done by plotting the landmark coordinates in R with the package ‘ggplot2’ (Figure 6; Wickham, 2009). The corresponding R script can be downloaded from GitHub (file: ‘Protocol_stepB-14.R’).
  
  Figure 6. Landmarks are checked for accuracy using the R package ggplot. A. Example of specimen scan; B. The companion plotted landmark coordinates to the scan (A); C. Incorrectly placed landmarks will be immediately apparent visually. Note that (B) and (C) are inverted to that of (A), as pixel coordinates map inversely in a linear regression.
15. If there appears to be an obvious error in the position of the landmarks for any image (as in Figure 6C), redo landmark placement (step B5) for the affected leaf/file and paste the updated coordinates into the ‘master spreadsheet.’ Repeat steps B11-B12 to reformat and check landmark accuracy for the new spreadsheet.
Analysis of landmarks: Generalized Procrustes Analysis (GPA)
1. To perform GPA using the R package ‘shapes,’ (Dryden, 2017) the input file must consist of only landmark data.
2. The data can now be processed in R using the ‘shapes’ package.
Isolating leaves from scans
1. For scans of fresh leaves:
2. For scans of herbarium specimens:
3. The program SHAPE (Iwata and Ukai, 2002) uses binary leaf outline image files in BMP format. Leaf images can be converted in ImageJ by creating a batch macro.
  1. Open ImageJ.
  2. Create a macro to easily convert many images at once.
    1. From the ImageJ menu click ‘Process > Batch > Macro...’
    2. Select the appropriate ‘Input’ folder containing the images to be converted.
    3. Select an ‘Output’ folder to contain the new binary images.
    4. Choose ‘BMP’ from the ‘Output Format’ drop-down menu.
    5. Input the text from the script found on GitHub (file: ‘Protocol_stepD-3-b-v.txt’) into the large blank space provided.
  3. View the output folder to check that the .bmp files were properly converted (Note 9).
Analysis of outlines: Elliptical Fourier Descriptors (EFDs)
1. Use the SHAPE software to convert image outlines to chain code. We encourage users to become familiar with the SHAPE User Manual in order to better explore parameter choice.
  1. Open the ChainCoder program within SHAPE.
    1. Select ‘Files > Select Image File(s).’
    2. In the following window (Figure 11), select the folder of BMP files. The BMP files will then appear in the field ‘File(s).’
    3. Select all images and select the double arrow button to transfer the desired files into the ‘Selected File(s)’ field. Press ‘OK.’
      
      Figure 11. Selection of .bmp files for chain coding in the software SHAPE. Use the double arrow button to move all files to the ‘Selected File(s)’ field.
  2. Before beginning the analyses, select the ‘Config’ tab (Figure 12).
    1. Set ‘Object Color’ to ‘Dark (Black),’ and ‘Scale Included’ to ‘No.’ Leave the other fields at their defaults.
      
      Figure 12. The ‘Config’ tab of the ChainCoder program. This allows for configuration of parameters before beginning the chain coding process.
  3. Select the ‘Processing’ tab to begin processing photos.
    1. Select ‘Load Image’ (Note 10).
    2. Deselect the ‘Select Area’ box.
    3. Select ‘Gray Scale.’
    4. Select ‘Make Histogram.’
    5. Select ‘Binarize Image.’
    6. Check the ‘Ero Dil Filter’ and ‘Dil Ero Filter’ boxes and select to what degree to ‘erode’ and ‘dilate’ the outline; this option adds and subtracts the amount of pixels from the image to give a smoother border.
    7. Select the ‘Labeling Object’ button. A new window named ‘Chain Code Data’ will appear, allowing the user to indicate which objects, over a certain number of pixels, should be isolated for analysis.
    8. Select ‘Chain Coding.’ This will add the chain code to the user selection.
    9. Select ‘Save to File.’ For the first image, you will have to name the file. Subsequently, chain codes for further images will be saved as processed (Note 11).
  4. Repeat steps E1a-E1c for all further images. If many images need to be processed, hold down the ‘enter’ key (or put a weight on it until the image processing is finished).
2. Convert chain code to normalized EFDs.
  1. Open the CHC2NEF program within SHAPE. This program converts the chain code created above to normalized EFDs.
  2. A new window will appear (Figure 13). Select the chain code file produced in step E1c.ix (‘CHC File Name’), then create a name for the new NEF file that will be generated in the following steps (‘NEF File Name’).
    
    Figure 13. The CHC2NEF window. Select the chain code input file (.chc) and name the resulting normalized elliptical Fourier descriptors file (.nef).
  3. Set the ‘Max Harmonic No.’ Higher harmonic numbers lead to better shape approximations, but usually 20 is sufficient to recapitulate leaf shape accurately.
  4. Select the Normalization Method to be ‘based on the longest radius.’ This is the way the image will initially be oriented, and this option allows better subsequent manipulation to align properly.
  5. Click ‘OK.’
  6. A new window will appear. Click ‘Start !!’
  7. Orient the image so that all images are similarly aligned. This is an arbitrary designation, but needs to be consistent among all images.
    1. There are a number of buttons to assist in this process (see Figure 14). The image can be turned by any number of degrees in either direction, and arrow buttons can be adjusted by rotation degree. Alternatively, images can rotate by 180°.
      
      Figure 14. Orienting leaf image chain code to normalized elliptical Fourier descriptors. Utilize the three turning buttons at the right side of the screen to orient the leaf image appropriately. NEF code will reflect these changes (bottom of window).
    2. Once aligned, click ‘Save/Next Obj.’ and repeat until all images have been normalized.
3. Visualize Elliptical Fourier Descriptors (EFDs) using Principal Component Analysis (PCA). These analyses can be done within the SHAPE program itself (see below; Note 12).
4. The PrinPrint program can be used to view the ‘PC contours’ file at a later time.
  
  Figure 20. Eigenleaves visualized in the ‘PrinPrint’ program
Shape features: Aspect ratio and circularity
1. Open ImageJ.
2. Navigate to the menu bar and select ‘Analyze > Set Measurements...’
3. Check the ‘Area,’ ‘Shape Descriptors,’ and ‘Display label’ boxes.
4. Navigate to the menu bar and select ‘Process > Batch > Macro...’
5. The resulting data can be visualized using linear regression (Figure 21).
  
  Figure 21. Linear regression of Aspect Ratio (AR) and Circularity (Circ.) data. In this example, multiple leaves from several genotypes of Vitis riparia (purple) and V. rupestris (green) have been measured for AR and Circ., with the resulting data visualized in this linear regression. Low AR and Circ. values capture the deeper lobing and significant serration of V. riparia leaves compared to that of V. rupestris leaves.

Data analysis

Data analysis techniques are fundamental to the purpose of this protocol and are integrated within the procedure (i.e., C, E3, F5), as they are often challenging for new users to develop de novo. However, our methods represent a sampling of available methods for digital morphometric analysis. We encourage users to explore the literature and available programs to develop a method that is best suited for their material and the particular scientific inquiry.

Notes

In the scripts available from GitHub, any text preceded with ‘#’ should be considered user notes–these will not be read by the computer if the whole script is copied and pasted into the R or macro consoles.
Because Generalized Procrustes Analysis (GPA) and outline analysis do not require scaling, it is not necessary to include a ruler in scans for morphometric analyses; however, we recommend this as good practice.
If the point tool is in the single ‘Point’ setting, simply right-click on the button to change it to the ‘Multi-point’ setting.
Landmarks must be placed in the same order on all leaves.
If an error was made during placement, points can be deleted by holding the CTRL key and clicking on the point. It is also possible to move points by clicking on the imprecise point, then moving it to the desired location.
It may be necessary to adjust column headers or other information that might be erroneously askew from the conversion process.
If there is only one fresh leaf per scan, refer to the directions in step D1e-D1f. Repeat until all images have been converted to black and white.
For the first image file of the dataset, do not use the first line of code (e.g., run(“Open Next”)).
BMPs are large files, and it may be convenient to use an external hard drive or cloud storage client to store and use them from this point forward.
Viewing the first image in the viewing field may require using the ‘Zoom Out’ button in the top right corner to visualize properly.
This produces one file for all of the images’ chain code. If chain-coding cannot be completed in one sitting, the file can be updated to include the remaining files at a later date.
Alternatively, using the R package ‘Momocs’ (Bonhomme et al., 2014) converts NEFs to objects for a variety of graphical visualizations, including PCA.
The desired output will not be a folder of files, rather the measurement report that can be saved as a single file. Therefore, the output folder can be the same as the input folder. To minimize file size, select a small file format (we suggest .jpeg) in the ‘Output Format’ drop down menu.

Acknowledgments

This protocol was developed in part for the publications of Chitwood et al. (2016) and Klein et al. (2017). The authors are grateful to Dr. Dan Chitwood for his comments, guidance, and expertise in using and developing these methods. We would also like to thank Dr. Allison Miller lab undergraduate members for their comments and suggestions on the protocol, as well as three reviewers who helped improve this manuscript. A Saint Louis University Graduate Research Assistantship to LLK and an Ohio University Original Work Grant to HTS supported this work.

References

Bonhomme, V., Picq, S., Gaucherel, C. and Claude, J. (2014). Momocs: Outline analysis using R. J Stat Softw 56(1): 1-24.
Bookstein, F. L. (1991). Morphometric Tools for Landmark Data: Geometry and Biology. Cambridge University Press, pp: 435.
Chitwood, D. H., Klein, L. L., O’Hanlon, R., Chacko, S., Greg, M., Kitchen, C., Miller, A. J. and Londo, J. P. (2016). Latent developmental and evolutionary shapes embedded within the grapevine leaf. New Phytol 210(1): 343-355.
Chitwood, D. H. and Sinha, N. R. (2016). Evolutionary and environmental forces sculpting leaf development. Curr Biol 26(7): R297-306.
Cope, J. S., Corney, D., Clark, J. Y., Remagnino, P. and Wilkin, P. (2012). Plant species identification using digital morphometrics: A review. Expert Syst Appl 39(8): 7562–7573.
Dryden, I. L. (2017). Shapes: Statistical Shape Analysis. R package version 1.1-8.
Hearn, D. J. (2009). Shape analysis for the automated identification of plants from images of leaves. Taxon 58(3): 934-954.
Iwata, H. and Ukai, Y. (2002). SHAPE: a computer program package for quantitative evaluation of biological shapes based on elliptic Fourier descriptors. J Heredity 93: 384-385.
Klein, L. L., Caito, M., Chapnick, C., Kitchen, C., O’Hanlon Regan, Chitwood, D. H. and Miller, A. J. (2017). Digital morphometrics of two North American grapevines (Vitis: Vitaceae) quantifies leaf variation between species, within species, and among individuals. Front Plant Sci 8: 373.
Kuhl, F. P. and Giardina, C. R. (1982). Elliptic features of a closed contour. Comput Vision Graphs 18: 236-258.
McLellan, T. and Endler, J. A. (1998). The relative success of some methods for measuring and describing the shape of complex objects. Sys Biol 47(2): 264-281.
Murdoch, D., Chow, E. D. and Celayeta, J. F. (2007). ellipse: Functions for drawing ellipses and ellipse-like confidence regions. R package version 0.3-5.
R Core Team. (2017). R: a language and environment for statistical computing. R foundation for statistical computing. Vienna.
Wickham, H. (2009). ggplot2: Elegant Graphics for Data Analysis. Springer.
Wickham, H. and Chang, W. (2015). devtools: Tools to Make Developing R Packages Easier. R package version 1(0): 185.
Wickham, H., Danenberg, P. and Eugster, M. (2015). roxygen2: In-source documentation for R, 2015. R package version 5(1).
Zelditch, M. L., Swiderski, D. L., Sheets, H. D. and Fink, W. L. (2004). Geometric Morphometrics for Biologists: A Primer. Elsevier Academic Press, pp: 437.