Plant growth
Arabidopsis seeds (Col-0 ecotype) were sown directly into 60 × 60 × 80 mm pots of damp, lightly compressed soil (3:1 Levington M3 compost:perlite) and stratified at 4 °C for 7 days before transfer into a controlled environment chamber (Conviron, Canada) under short day conditions (12 h light 22 °C/12 h dark 15 °C, 200 µmol m−2 s−1 PAR at rosette level, 60% humidity). Leaf discs were excised from the largest leaves for scanning 30 days after germination. Tomato (Solanum lycopersicum var. Ailsa Craig), pea (Pisum sativum var. Arvense), barley (Hordeum vulgare var. Tipple Fulbourn) and oat (Avena sativa), were sown in 20 × 20 × 30 cm pots of M3 compost and grown under long day conditions (16 h light 22 °C/8 h dark 15 °C, 200 µmol m−2 s−1 PAR, 60% humidity). Leaf discs were excised from the largest, mature leaves for scanning. Rice seeds (Oryza latifolia) were germinated on wet filter paper in 90 mm diameter, 20 mm deep petri plates, and transplanted into water-saturated soil (70% v/v Kettering Loam (Boughton, UK), 23% v/v Vitax John Innes No. 3 (Leicester, UK), 5% v/v silica sand and 2% v/v Osmocote Extract Standard 5–6 month slow release fertiliser (Ipswich, UK)) in 105 × 105 × 185 mm pots, 8 days after germination. Rice plants had a constant water supply from the pot base and were grown in a controlled environment chamber (Conviron, Canada) with 12 h, 30 °C days and 12 h 24 °C nights, 700 µmol m−2 s−1 PAR at canopy level and 60% relative humidity.
For rice and Arabidopsis, n = 5. For the other species, and for the 2D analysis of rice, leaf sections n = 4. To allow comparison of leaves of the same species (or mutants) we selected leaf 5 for analysis in our experimental studies so they are at same developmental growth stage.
Sample preparation for microCT
Single leaf discs (5 mm diameter) were excised from the mid-point (length-ways) of selected leaves using a stainless steel cork borer and avoiding the mid-vein (Fig. 1a). Leaf discs were mounted between low density polystyrene, at a 45° angle to reduce the number of angular projections through the maximum thickness of the sample, in 1.5 mL polypropylene micro centrifuge tubes, mounted on a 10 cm length of a plastic pipettes (Fig. 1b–d). Sample holder components were selected based on their rigidity, providing a tight fit to reduce sample movement, and low X-ray absorption, enabling good contrast with leaf material. Sample holders were sealed with Sellotape® to reduce desiccation and acclimatised for 5 min with the sample in the X-ray beam. Leaf discs from monocot species were positioned so that the veins were parallel to the X-ray source prior to scanning to aid alignment after reconstruction.
X-ray microCT scanning
Single microCT scans of leaf discs were performed using a GE Phoenix Nanotom S 180NF (GE Sensing and Inspection Technologies GmbH, Wunstorf, Germany) fitted with a tungsten transmission target and a 5 MP (2304 × 2304 pixel) CMOS digital detector (Hamamatsu Photonics KK, Shizuoka, Japan). A three-point detector calibration was performed, collecting an average of 100 images, with 10 skip images per gain point. Scans were obtained at a spatial resolution of 2.75 μm (2304 × 1400 pixel field of view), with an electron acceleration energy of 85 kV and a current of 100 μA (higher spatial resolutions are possible if a smaller diameter sample can be used). Detector exposure time was 500 ms, collecting 3600 projections in ‘fast scan’ mode (sample rotates continuously), with no averaging or skip images and no pixel binning (1 × 1), resulting in a scan duration of 30 min per sample.
Reconstruction
Radiograph reconstruction was carried out using Phoenix Datos|x rec 2 reconstruction software (version 2.3.3; GE Sensing and Inspection Technologies GmbH, Wunstorf, Germany). Radiographs were assessed for sample movement using the autoscan|optimiser module, by comparing the difference between the first and last projection image (0° and 360° rotation) and applying an automatic directional and/or scale correction if movement and/or shrinkage were apparent. Any sample that required more than 3 pixel shift in x or y axis were either rescanned or disregarded as the image quality in these images was low. Beam hardening artefacts were mitigated using the multiple materials function in the BHC + module. A beam hardening correction of 7 was determined to be the most appropriate for plant leaves. Finally, radiographs were manually cropped i.e. resized to remove the scanned area beyond the leaf sample before being reconstructed into 3D volumes using a filtered back-projection algorithm.
Image analysis
An illustration of the image analysis workflow is provided in Fig. 1e–i.
Alignment and cropping
Grayscale volumes were aligned in 3D (adaxial leaf surface facing up), cropped to remove any damaged leaf material at the disc periphery, and converted to stacks of TIFF images in the Z dimension using VG StudioMAX (version 2.2.0; Volume Graphics GmbH, Heidelberg, Germany).
Mask creation
Leaf discs were segmented from the surrounding sample holder by creating material masks in Avizo Fire software (version 6.0.0 Fire; Thermo Fisher Scientific, USA), using the ‘Label Field’ function and then binarising the selection.
Thresholding
Individual grayscale TIFF stacks were thresholded using the ‘Threshold’ function in the open source software package ImageJ (version 1.48; [36]) and saved as binary TIFF stacks, differentiating solid material from airspace. The automated thresholding algorithm was selected based on comparison between the binarised and the greyscale images, to account for small differences between scans in sample/background contrast, leaf water content and polystyrene elements. Previous research by our group has shown that the IJ Iso-data algorithm proved effective for thresholding Arabidopsis [35]. However, it should be highlighted that a range of automated thresholding algorithms are available within ImageJ and will result in different outputs depending on the grayscale distributions of the image. This unfortunately, results in some level of manual selection of the most appropriate threshold algorithm. We would strongly recommend that the same threshold algorithm is used for all samples within the same study. For the rice and cereal leaves, the Li algorithm was used as they presented a finer pore structure. Material masks were thresholded using the automatic thresholding method ‘MaxEntropy’. All thresholded images were saved as binary TIFF stacks.
Intercellular airspace extraction
Binary material masks were combined with thresholded image stacks using the ‘Image Calculator’ function in ImageJ to create a composite image stack, isolating the extracellular airspace within each leaf disc.
Noise removal
Scans were de-noised using the ‘Remove Outliers’ function in ImageJ. Foreground and background particles < 3 × the spatial resolution were removed.
Region of interest selection
The inclusion of the mid-rib and/or major veins in images subjected to 3D analysis can artificially increase porosity measurements. In monocots, where vasculature is arranged in parallel cell files, regions of interest were selected between major veins. In rice in particular, which has dense vasculature, three 200 × 200 voxel regions were selected for analysis, and all 3D measurements were averaged across these technical replicates to provide representative data for the leaf disc as a whole. In all other species a region of interest (ROI) of ≥ 400 × 400 voxels was used. Due to the non-uniform structure and irregular vasculature of dicot leaves, it was not possible to entirely exclude vasculature, but the largest veins were avoided.
3D measurements
All 3D measurements were conducted using ImageJ (version 1.48; [36]). Leaf disc porosity, the number of individual air channels, the porosity distribution through the leaf disc depth, and the surface area of mesophyll cells exposed to intercellular airspace (Smes) were all calculated from data acquired using the ImageJ function ‘Analyze Particles’. Leaf porosity (%) was calculated using Eq. 1:
$$Porosity = \left( {\frac{{\sum A_{p} }}{{\sum A_{m} }}} \right) \times 100$$
(1)
where, ΣAp and ΣAm are the summation of the area (mm2) occupied by pores and the area of the mask for all slices within the entire z-stack. The distribution of porosity throughout the leaf disc was plotted by calculation of porosity on a slice-by-slice basis (increments equal to individual slice thickness, which is determined by the CT scan resolution) in the Z dimension, and plotted from the adaxial to abaxial surface.
Smes (mm2 mm−2) was calculated using Eq. 2:
$$S_{\text{mes}} = \frac{{\sum P_{p} \times RES}}{{\sum A_{m} }}$$
(2)
where, ΣPp is the summation of the perimeters (mm) of each individual pore present within the entire z-stack and RES is the spatial resolution of the CT scan (mm). The number of individual pores, and their perimeters, were direct outputs of the ‘Analyze Particles’ function. The perimeter measure is implemented within the PolygonRoi class and is calculated by accounting for the straight and corner pixels of the boundary. In brief, straight edge pixels are measured as length 1, with corner pixels length \(\sqrt 2\).
Representative 3D renderings of plant material, with air channel diameters illustrated by heat map, were constructed in VG StudioMAX (version 2.2.0; Volume Graphics GmbH, Heidelberg, Germany) using the isosurface and Phong rendering tools. The heat map data was an output of the ‘Thickness’ function in the ImageJ plugin BoneJ (version 1.3.14; [37]) which also provides the mean and maximum channel diameter for each stack.
Sample preparation for 2D analysis of fixed tissue sections
Leaf discs of O. latifolia were fixed in 4% v/v formaldehyde in PEM buffer (1.5% w/v Pipes, 0.19% w/v EGTA, 0.124% w/v MgSO4, pH 7) immediately after CT scanning. After no more than 72 h, samples were rinsed in PEM buffer three times for 10 min each. Samples were dehydrated in an ascending ethanol series (10%, 30%, 50%, 70%, 90%, 100% v/v ethanol, 1 h each) then infiltrated with an ascending series of LR white resin (London Resin Company) in ethanol (10%, 20%, 30%, 50%, 70%, 90% v/v 1 h each then 3 × 8 + hours in 100% resin). Samples were kept at 4 °C throughout dehydration and infiltration. Finally samples were stood vertically in gelatine capsules filled with resin and left to polymerise for 5 days at 37 °C. 2 µm sections were cut with a Reichert-Jung Ultracut E ultramicrotome and dried onto vectabond-coated multi-well slides. 4–5 sections were imaged per biological replicate, each of which was at least a cell’s length apart. Sections were stained for 5 min in a 0.1 mg mL−1 solution of propidium iodide in water and rinsed in water before imaging. Samples were imaged using a Leica DM6 microscope and camera equipped with a CoolLED fluorescence system, and images were captured using LASX software. Samples were illuminated with the 535 nm LED line, and visualised through the Y3 filter.
2D measurements
The workflow for stereological analysis is illustrated in Additional file 2: Fig. S1. Masks representing total leaf area (Additional file 2: Fig. S1B) and individual airspaces (Additional file 2: Fig. S1C) were generated using ImageJ (FIJI v1.51u; [38] with the connection thresholding and edge detection plugins). Masks were smoothed using the Median filter, with a radius of 3 pixels. Airspace area was expressed as a percentage of total leaf area to give an estimate of porosity (the fraction of leaf volume occupied by intercellular airspace).
The perimeter of each individually segmented airspace was measured (Additional file 2: Fig. S1D) and summed to give the total perimeter of pores exposed to intercellular airspace (∑Pp, mm). The width of the microscope section analysed (W, mm) was measured (Additional file 2: Fig. S1A). The total cell surface area exposed to intercellular airspace per leaf surface area (Smes, mm2 mm−2) was calculated using the Eq. 3.
$$S_{mes} = \frac{{\sum P_{p} }}{W} \times F$$
(3)
where F is a stereological correction factor. In order to estimate 3D Smes from this data, airspaces were assumed to have a general prolate spheroid shape with the major axis being twice the length of the other two minor axes, as in Giuliani et al. [39], and accordingly, based on Thain [10], an F value of 1.42 was used.
Statistical analyses
All statistical analyses were conducted in Graphpad Prism software (version 7.03).