Introduction
Morphological characterization and the subsequent diversity assessment of plant genetic resources rely on the accuracy and precision of descriptor lists. Bioversity International (2007) has detailed the process to develop descriptor lists; however, particularly for colour, psychophysical subjective assessment is used to discriminate the various descriptor states of the crop. With the advent of image-based phenotyping techniques, objective methods of characterizing colour may be developed, as in the case of tomato (Darrigues et al., Reference Darrigues, Hall, Van Der Knaap, Francis, Dujmovic and Gray2008) and strawberry (Yoshioka et al., Reference Yoshioka, Nakayama, Noguchi and Horie2013). On the other hand, pili (Canarium ovatum Engl.), an endemic crop in the Philippines, has no established descriptor list and use of image analysis may be useful to characterize the rich diversity of this tree species. Thus, the study aimed to systematically identify the colour descriptor states through image-based analysis using pili kernels as a model.
Experimental
Fifty-two (52) pili accessions from the core collection of the Institute of Crop Science, and the National Plant Genetic Resources Laboratory of the Institute of Plant Breeding, College of Agriculture and Food Science, University of the Philippines Los Baňos was used. Each accession was represented by 10 kernel samples as recommended by Bioversity International (2007). Images were captured using a calibrated VideometerLab 3 setup at a resolution of ~45 µm/pixel at the visible spectrum (Videometer A/S Hørsholm, Denmark). For better colour discrimination of images (Alata and Quintard, Reference Alata and Quintard2009), red (R), green (G) and blue (B) values per pixel were transformed to International Commission on Illumination (CIE) lightness (L*), green–red (a*) and blue–yellow (b*) values using the colour space converter plugin of ImageJ version 1.50i (Schneider et al., Reference Schneider, Rasband and Eliceiri2012). Per sample, the pixel matrices were separately extracted for CIE L*, a* and b* colour spaces as text images. The L*, a* and b* values per sample were averaged to get a representative colour value. From the mean values, cluster analysis was done using unweighted pair group method with arithmetic means and sequential, agglomerative, hierarchical and nested clustering parameters program of Numerical Taxonomy and Multivariate Analysis System-pc version 2.1 (Rohlf, Reference Rohlf2002). The dendrogram was cut at 3.5 Euclidian distance to ensure that the difference in colour is readily discernible by the observer; this was based from the Delta E value (ΔELab) that quantifies the distinguishable differences between two colours using the CIE1976 scheme (Sharma, Reference Sharma2003; Mokrzycki and Tatol, Reference Mokrzycki and Tatol2011). The averaged CIE L* a* b* values per cluster were used to represent a descriptor state. These values were visualized, named and converted to hex formats using Name that Colour (Mehta, Reference Mehta2001).
Discussion
Three clusters were observed from the dendrogram cut at 3.5 Euclidian distance (Fig. 1). The three clusters represent three descriptor states for kernel colour. A varying intensity of brown colour was observed between descriptor states. Descriptor states have corresponding hex and technical colour names (Table 1); however, the description is simplified to light brown, brown and dark brown to accommodate the wide range of users, as prescribed by Bioversity International (2007).
Pili kernel colour descriptor states of 53 accessions based on average CIE L* a* b* values. The dendrogram is cut at 3.5 Euclidian distance.
Colour profiles of pili kernel colour descriptor states
a Based on Mehta (Reference Mehta2001).
The light brown kernel colour was dominant in the germplasm collection with 33 accessions exhibiting the colour. This was followed by brown with 17 accessions. The dark brown coloured accessions, however, were only represented by two accessions indicating a gap in the diversity of the observed collection. It can also be observed that all accessions have varying average CIE L* a* and b* values which denote a degree of heterogeneity of colours within the cluster; however, the variations may result in non-perceivable differences due to low delta E values (Witzel et al., Reference Witzel, Burnham and Onley1973; Mokrzycki and Tatol, Reference Mokrzycki and Tatol2011). Cutting the cluster at a higher delta E of 3.5 will allow inexperienced observers to distinguish the colour variations (Mokrzycki and Tatol, Reference Mokrzycki and Tatol2011) in adherence to the standards of Bioversity International (2007). Furthermore, colour differences may be controlled by quantitative trait loci, such as in the case of the seed coat colour of chickpea (Hossain et al., Reference Hossain, Ford, McNeil, Pittock and Panozzo2011) and endosperm colour of sorghum grain (Fernandez et al., Reference Fernandez, Hamblin, Li, Rooney, Tuinstra and Kresovich2008), resulting in the continuous variation of brown colour in the accessions. Rather than viewing pili kernel colour as a discrete qualitative trait, the continuous variation in the intensity of brown colour may have a quantitative nature. Despite this, colour descriptor states, whether quantitative or qualitative in nature, are still vital in representing trait variations in the germplasm collection. On the other hand, charts and visuals of standards should be provided in order to distinguish continuous colour variations such as in the case of pili. Hex and CIE L*a*b* values, together with the technical colour name, may also be provided for reference.
Although pili was used as a model, the systematic method of establishing colour descriptor states can be applied to all homogenously-coloured parts of the plant of all crop species. It is also recommended to use a dynamic tree cut method in resolving clusters to ensure equal magnitudes of variations between clusters.
Acknowledgements
The authors would like to thank the Department of Science and Technology-Philippine Council for Agriculture, Aquatic and Natural Resources Research and Development (DOST-PCAARRD), Department of Science and Technology-Philippine Council for Health Research and Development (DOST-PCHRD), the T.T. Chang Genetic Resources Center-International Rice Research Institute (IRRI), University of Tsukuba, and the National Plant Genetic Resources Laboratory (NPGRL), College of Agriculture Food Science, University of the Philippines Los Baños for their support.
