No cell is an island: characterising the leaf epidermis using epidermalmorph, a new R package

Summary The leaf epidermis is the interface between a plant and its environment. The epidermis is highly variable in morphology, with links to both phylogeny and environment, and this diversity is relevant to several fields, including physiology, functional traits, palaeobotany, taxonomy and developmental biology. Describing and measuring leaf epidermal traits remains challenging. Current approaches are either extremely labour‐intensive and not feasible for large studies or limited to measurements of individual cells. Here, we present a method to characterise individual cell size, shape (including the effect of neighbouring cells) and arrangement from light microscope images. We provide the first automated characterisation of cell arrangement (from traced images) as well as multiple new shape characteristics. We have implemented this method in an R package, epidermalmorph, and provide an example workflow using this package, which includes functions to evaluate trait reliability and optimal sampling effort for any given group of plants. We demonstrate that our new metrics of cell shape are independent of gross cell shape, unlike existing metrics. epidermalmorph provides a broadly applicable method for quantifying epidermal traits that we hope can be used to disentangle the fundamental relationships between form and function in the leaf epidermis.


Table S1
Reference list of all traits Table S2 Plants sampled for trait reliability analyses.

Figure S1
Graphical description of cell simulation algorithm.

Figure S2
Measured values of undulation index (UI; see Table 1) on simulated cells.

Figure S3
Measured values solidity (see Table 1) on simulated cells.     ______________________________________________ Methods S1 -Image preparation for trait reliability Fully expanded adult leaves were collected from healthy plants grown at the University of Tasmania and the Royal Botanic Gardens Edinburgh (Table S1). These plants were all growing in shaded, frost-free greenhouses. For species with a wide geographic distribution, we sampled leaves from individuals with multiple provenances (where this was possible).
For large leaves, pieces of approximately 1cm 2 were cut from either side of the midrib in the middle third of the leaf; for smaller leaves, the base, apex and, if possible, the margins were removed. These samples were soaked in commercial household bleach (50 gL -1 sodium hypochlorite and 13 gL -1 sodium hydroxide) until the cuticle separated from the mesophyll. Bleach was removed by thoroughly rinsing in water and remaining mesophyll tissue was removed using a fine paintbrush.
Sections were stained with 1% crystal violet or safranin solution for 1 minute, then mounted in phenol glycerine jelly. Several fields of view at ×10 magnification (field of view area, 0.56mm 2 ) were photographed from each section using a Nikon Digital Sight DS-L1 camera (Melville, NY, USA) mounted on a Leica DM 1000 microscope (Nussloch, Germany). The best 3-5 images (identified as the clearest, without damage or obvious distortion) for each individual plant were selected for analyses.
Image pre-processing and annotation was done in ImageJ. Images were converted to 8-bit grayscale, then Ridge Detection was performed (parameters manually set for each image to give best results) to binarize the image. For high-resolution images, we dilated the resulting image to better see the cell walls (1px wide → 3px wide). We then used the paintbrush tool to correct the tracing, and the flood fill tool to annotate cell types (we used a value of 85 for stomata, 170 for subsidiary cells and 50 for salt glands, but these are arbitrary and any values could be used). Images were then further analysed with EPIDERMALMORPH as described in the main text.  Figure S1. Simulated cell generation. We generated cells with 3,4,5 and 6 sides, with aspect ratios of 1,2 and 5. These formed the 'straight-walled' cells. Each of these cells was then undulated with a frequency of 1, 2 and 3, and an amplitude of 0.2, 1 and 2, although for some shapes the maximum amplitude was not reached because of dampening effects. Figure S2. Measured values of undulation index (UI; see Table 1) on simulated cells. On straight-walled cells (highlighted with white dotted line), UI increases with aspect ratio. Figure S3. Measured values of solidity (see Table 1) on simulated cells. Solidity values are inflated for non-convex cells (highlighted with white dotted lines).    Fig. 2), on simulated cells. Increases with undulation frequency. Note that for some cells (e.g. those highlighted by the dotted line), the frequency is lower than expectedthis is a quirk of the cell generation (specifically the rounding of the wavelength to fit an integer number of undulations along a side), not of the metric.