Image analysis of anatomical traits in stalk transections of maize and other grasses
- 7.8k Downloads
Grass stalks architecturally support leaves and reproductive structures, functionally support the transport of water and nutrients, and are harvested for multiple agricultural uses. Research on these basic and applied aspects of grass stalks would benefit from improved capabilities for measuring internal anatomical features. In particular, methods suitable for phenotyping populations of plants are needed.
To meet the need for large-scale measurements of stalk anatomy features, we developed custom image processing software that utilized a variety of global thresholding, local filtering, and feature detection methods to measure rind thickness, pith area, vascular bundle counts, and individual vascular bundle size from digital images of hand-cut transections of stalks collected with a flatbed document scanner. The tool determined vascular bundle number with an average accuracy of 90% across maize genotypes that varied five-fold for this trait. The method is demonstrated on maize, sorghum, and Miscanthus stalks. The computer source code is staged for download.
Simplicity of sample preparation and semi-automated analyses enabled by this tool greatly increase measurement throughput relative to standard microscopy-based techniques while maintaining high accuracy. The tool is expected to be useful in genetic and physiological studies of the relationships between stalk anatomy and traits such as biofuel suitability, water use efficiency, or nutrient transport.
KeywordsVascular Bundle Anatomical Feature Minimal Sample Preparation Stalk Diameter Anisotropic Diffusion Filter
The stalks of widely cultivated grass species such as maize and sorghum support multiple architectural and physiological functions, while contributing the most to aerial non-grain biomass. Visible in transections of such stalks are the many vascular bundles scattered throughout the parenchymatous pith. Surrounding the pith is a layer rich in collenchyma that is usually visibly distinct and commonly called the rind. The developmental mechanisms that determine the number of vascular bundles, their distribution, and the proportion of rind to total stem tissue in graminaceous crops is currently an important topic of research . For example, water movement through the plant may relate to the number and size of xylem-bearing vascular bundles. Therefore, selection-based breeding for stem anatomy traits could be an effective strategy to improve water use efficiency [2,3]. Also, cell walls in the sclerenchyma surrounding the bundles or collenchyma in the rind are typically highly lignified, which limits digestibility when the stems are used for animal feedstock or industrial fermentation for ethanol production [4,5]. Plants better suited for ethanol production may be identified through surveys of natural variation for stem anatomy traits [6-8]. Another reason for studying anatomical features of grass stalks is their relationship to mechanical properties. The strength of the stalk determines the degree of wasteful crop lodging and pre-harvest breakage in the field [9-12].
These various motivations for measuring the anatomical features of grass stems create the need for a method that is efficient enough to measure hundreds if not thousands of individuals within defined populations for the purpose of mapping the genetic loci responsible for variation in the trait. Traditional methods for studying anatomy usually rely on sectioning chemically fixed tissue with a microtome followed by mounting the cut section on glass slides for examination with a microscope. These microscopic methods give superb cellular-level resolution, and have been used in large-scale studies of anatomical features [13-15], but typically their throughput is low. Relaxing the resolution criterion from cellular to tissue level increases the feasibility of automation. Higher throughput achieved by greater automation would improve the feasibility of acquiring the large data sets needed for some types of studies, such as statistical genetic trait mapping.
Computerized processing of digital images is an increasingly common means of quantifying plant structure. Sometimes sophisticated microscopy is called for [16-18] but simpler devices such as flatbed document scanners are perfectly adequate in many cases [19,20]. To quantify stem anatomical features as phenotypes across populations of plants suitable for statistical genetic analyses, for example, a simple imaging device and minimal sample preparation may provide the appropriate balance between resolution and throughput. The goal of the present work was to create an image analysis tool that could operate on images of hand-cut stem transections obtained on a flatbed scanner to measure anatomical features in high throughput.
Range of trait values observed in a sample of thirty maize genotypes
Stalk diameter (cm)
3.26 ± 0.23
1.60 ± 0.06
Rind thickness (cm)
0.42 ± 0.08
0.13 ± 0.04
Vascular bundle density (cm−2)
57.3 ± 3.8
36.8 ± 2.8
Vascular bundle size (cm2)
9.04 ± 1.1 × 10−4
4.31 ± 1.2 × 10−4
Vascular bundle number and density
Dividing the number of bundles by the pith area gives vascular bundle density. This trait also varied among the 30 genotypes sampled, though relatively less than the stalk or rind thickness traits. The highest average density, observed in the OS602 genotype, was 57% greater than the lowest density observed, in the W182BN genotype (Table 1).
Determining the size of the bundles
Application to other species
Overall, the set of design decisions and technical solutions produced an effective tool for high-throughput quantification of anatomical features in grass stalks. The tool, written in the Matlab computer language, is staged for download at http://phytomorph.wisc.edu/download/HeckwolfPlantMethods2015/ along with a composite test image representing a variety of transection phenotypes so that the performance of future tools for studying stalk anatomy can be benchmarked against that described here.
The current tool and future derivatives of it may be expected to enable systems-style projects attempting to discover links between the anatomical features and chemical or gene expression features, or large-studies of the genetic architecture affecting stalk architecture. More specifically, this tool may benefit biofuels researchers seeking a deeper understanding of the relationships between stalk anatomy, cell wall composition, and efficiency of carbohydrate conversion to ethanol because lignin content, which is correlated with features such as vascular bundle density quantified here (Table 1), limits the efficiency of biomass to ethanol conversion. Modeling of the size and number of vascular bundles in grass stalks for the purpose of achieving a quantitative understanding of stalk hydraulics is another area of research that may benefit from the tool.
This motivation for the development reported here, and therefore the resulting product, differs from a recent study that also used image analysis of maize stalk transections but to produce a 3D statistical model of vascular bundle distribution . The present work emphasizes measurement throughput, to address the need for quantifying the most salient anatomical features in thousands of stalk samples, whereas Legland et al.  created a normalized model of vascular bundle distributions that facilitates quantitative comparisons between categories of samples.
Although the tool was developed to be effective with minimal sample preparation, it may prove useful in studies of transections that have been stained to highlight features such as high lignin content. Incorporation of such labeling or staining steps may give the method reported here more power to resolve anatomical details, for example by differentiating between sclerenchyma and collenchyma within the rind [37,38].
Plant growth and sample collection
A set of 30 diverse maize inbred lines from the Wisconsin Diverse Association Panel , was grown at the Arlington Agricultural Research Station (University of Wisconsin-Madison) in 2013 in a field experiment using a randomized complete block design with two replications. Three representative plants per plot and field replication were harvested and brought to the lab 45 days after flowering for sample preparation.
Sample preparation and imaging
Transections of the third internode above the ground were cut by hand using razor blades into sections between 4 and 10 mm thick. Sections thicker than that resulted frequently in shadows or reflections on the rim of the stalk, producing a halo effect, which made the analysis of the stalks using standard parameter setting difficult or impossible. To produce an image for analysis, a total of 12 transections representing three individuals of four genotypes were placed on the horizontal imaging surface of an Epson Perfection V700 Photo Scanner and scanned at resolution of 800 dots per inch in red, green, blue color mode. Using these settings, we produced images with a height of 3800 to 3900 pixels and a width of 5000 to 5100 pixels and a resolution of 315 pixels per centimeter, with an error of at most 5 pixels. To enhance contrast between the samples and the background, the lid of the scanner was left open, resulting in a black background. The resulting files, typically larger than 30 megabytes, were saved in tagged image file format to a computer disk array.
This work was funded by grant IOS-1031416 from the National Science Foundation Plant Genome Research Program to E.P.S. and in part by the DOE Great Lakes Bioenergy Research Center (DOE BER Office of Science DE-FC02-07ER64494).
- 20.Smith HC, Niewohner DJ, Dewey GD, Longo AM, Guy TL, Higgins BR, et al. Using flatbed scanners to collect high-resolution time-lapsed images of the Arabidopsis root gravitropic response. J Visualized Experiments. 2014;83:e50878. doi:10.3791/50878.Google Scholar
- 24.Weikert J. Anisotropic Diffusion in Image Processing. Stuttgart: Teubner-Verlag; 1998.Google Scholar
- 25.Kovesi PD. MATLAB and Octave Functions for Computer Vision and Image Processing. 2002. http://www.csse.uwa.edu.au/~pk/research/matlabfns/#anisodiff. Accessed 18 Jan 2015.
- 26.Perona P, Malik J. Scale-space and edge detection using anisotropic diffusion. Proceedings of IEEE Computer Society Workshop on Computer Vision. 1987;16–22. http://www.cs.berkeley.edu/~malik/papers/MP-aniso.pdf. Accessed 18 Jan 2015.
- 28.Neubeck A, Van Gool L. Efficient Non-Maximum Suppression. 18th International Conference on Pattern Recognition (ICPR 2006); 20–24 August 2006; Hong Kong; China; 850–855. ftp://oenone-v06.ee.ethz.ch/publications/proceedings/eth_biwi_00446.pdf. Accessed 18 Jan 2015.
- 30.Eddins S. Homomorphic Filtering. 2013. http://blogs.mathworks.com/steve/2013/06/25/homomorphic-filtering-part-1/. Accessed 18 Jan 2015.
- 32.Beucher S, Lantuéjoul C. Use of watersheds in contour detection. International Workshop on image processing; real-time edge and motion detection/estimation; Rennes; France; 1979. http://cmm.ensmp.fr/~beucher/publi/watershed.pdf. Accessed 18 Jan 2015.
- 33.Nieniewski M, Serneels R. Segmentation of spinal cord images by means of watershed and region merging together with inhomogeneity correction. Machine Graphics Vision Intl. 2002;11:101–21.Google Scholar
- 38.Brauns F. The chemistry of lignin. New York: Academic; 1952.Google Scholar
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.