Discriminant Analysis of Undaria pinnatifida Production Areas Using Trace Elemental Analysis
Increasingly, attention is being paid to declaring the origin of agricultural and marine products after the advent of the bovine spongiform encephalopathy (BSE; commonly known as mad-cow disease). The display of the production centers on U. pinnatifida has been required in Japan since 2006. As an example of testing in another marine product, near-infrared spectra (NIR) and trace elemental analysis of U. pinnatifida are proven effective methods for discriminating production centers by us and Food and Agricultural Materials Inspection Center (FAMIC). In the present study, we found that X-ray fluorescence analysis of Br was also effective for the discrimination of production centers. The results of our study suggest that a combination of NIR and X-ray fluorescence analysis is a convenient and efficient method for determination due simple sampling procedures and increased effectiveness.
Issues related to the safety of foods have gathered the attention of consumers following the advent of bovine spongiform encephalopathy (BSE). There have also been improper displays of production centers on some marine products such as clams (Undaria pinnatifida). Hence, there is an urgent need for establishment of convenient and efficient scientific methods for discriminating the production centers.
In our earlier studies [1, 2], we reported that near-infrared spectroscopy is a useful method for discriminating production areas . In addition, Food and Agricultural Materials Inspection Center (FAMIC) reported that the inorganic elemental analysis results were valid markers for discriminating the production areas between China, Korea, and Japan (Sanriku and Naruto) of U. pinnatifida by ICP-MS spectroscopy [4–7]. Analysis results of these studies showed that the error rates were 0%, 26%, and 6% for samples from China, Korea, and Japan, respectively. In our present study, we hypothesize that additional accuracy can be achieved by classification and regression trees (CART [8, 9]) as a discriminant method.
In addition we explored the possibility of a more convenient discrimination method using elemental analysis of Br. The main purpose of this paper is reporting the possibility of the screening test, which would be conveniently applicable without using expensive and large equipments like an ICP-MS and so on, for discriminating production areas. Some NIR or X-ray fluorescence equipments are portable and the usages are not so complicated.
2. Materials and Methods
Results from the study in  involving ICP-MS analysis for 22 elements, namely, Al, Ba, Ca, Fe, K, Mg, Mn, Sr, Li, Co, Ni, Cu, Zn, Rb, Y, Mo, Cd, La, Nd, Sm, Gd, and W in 95 U. pinnatifida samples (29 China, 19 Korea, and 47 Japan (21 Sanriku and 26 Naruto)) were taken in our study for comparison of linear discriminant analysis (Table 1) and CART (Table 2). The overview of the sample preparation is below as follows:(1)cleaning samples with the ion-exchanged pure water thoroughly for getting rid of attached salt and the others, (2)drying them in the shade for one day under 20 degrees centigrade,(3)drying them in a vacuum with the pressure, about 5 mm Hg, under 107 degrees centigrade for one hour,(4)milling the dried samples into a fine small powder of less than 125 micrometers using a food processor (Millser IFM-700 G, Iwatani Corp.).
Sample preparation and analysis conditions for ICP-MS of the study are described in .
A total of 10 samples were independently taken in the present study: 3 samples each from China and Korea and 4 samples from Japan (2 each from Sanriku and Naruto). Samples for X-ray fluorescence analysis were prepared according to the method described in . A Shimadzu XRF-1800 was used for detecting trace elements by fundamental parameter methods. Sample briquettes were formed under 20 ton/cm2 pressure for 30 seconds with the MP-35-02 press.
Because we supposed that there could be the possibility of finding more convenient methods with using an equipment such as an X-ray fluorescence analysis, we gathered independent 10 samples from Riken Food Company.
3. Results and Discussion
3.1. CART Results and Comparison to Linear Discriminant Analysis
At first, we introduce the overview of how CART works.
In case of CART, if the target variable is categorical, then a classification tree is generated. To predict the category of the target variable using a classification tree, use the values of the predictor variables to move through the tree until you reach a terminal (leaf) node, then predict the category shown for that node. An example of a classification tree is known well in case of the Fisher’s Iris data (from the UCI Machine Learning Repository: Iris Data Set ). The target variable is “Species”, the species of Iris. We can see from the tree that if the value of the predictor variable “petal length” is less than or equal to 2.45 the species is Setosa. If the petal length is greater than 2.45, then additional splits are required to classify the species. Because CART is a nonparametric classification method, it is necessary to validate the obtained model. However, CART is usually more potent than LDA.
On the other hand, Linear discriminant analysis is a linear and parametric method with discriminating character. LDA focuses on finding optimal boundaries between classes.
Classification results between China, Korea, and Japan by CART are shown in Figure 1. Parent and terminal nodes had 2 and 1 minimum cases, respectively. –In node 1, the discriminant condition is given by the following equation.–In node 2, the discriminant condition is given by Ba .–In node 3, the discriminant condition is given by Cu .–In node 4, the discriminant condition is given by Sr .
In the event of the inclusion of 1 case of terminal node 3 and terminal node 4 in China and Japan, respectively, the maximum rate of error was 0/952/95. The ranking of variable importance was arranged in descending order, Nd, La, Fe, Sm, Al, Y, Gd, Cd, Ba, Cu, Li, and Sr. Rare earth elements, Fe, and Al were ranked as important. The classification results between Sanriku (Japan), Naruto (Japan), China, and Korea are shown in Figure 2. –In node 1, the discriminant condition is given by the following equation:–In node 2, the discriminant condition is given by Ba .–In node 3, the discriminant condition is given by Y .–In node 4, the discriminant condition is given by Cu .–In node 5, the discriminant condition is given by Cu .In node 6, the discriminant condition is given by 0.974 Y 0.226 Cd .–In node 7, the discriminant condition is given by Ba 0.0164 Sr .
In the event of inclusion of 1 case each of terminal nodes 2, 3, 5, and 7 for samples from Korea, China, China, and Sanriku, respectively, the maximum rate of error was 0/955/95. The ranking of variable importance was arranged in descending order: La, Nd, Fe, Y, Sm, Al, Cd, Gd, and Ba. Rare earth elements, Fe, and Al were ranked as important.
Based on the above CART method, more on the production centers could be extracted by the ICP-MS analysis results of Kadowaki and Tatsuguchi  and the proceeding for the discrimination of production centers of U. pinnatifida, FAMIC , than from the previous linear discriminant analysis.
3.2. Analysis Results by the CART on the Elemental Analysis with an X-Ray Fluorescence Method
The classification results from U. pinnatifida collected in China, Korea, and Japan based on CART with analysis of 9 major detectable elements detected by X-ray fluorescence method (Figure 3)—Fe, I, Br, As, Zn, Mn, Cu, Ni, and Cr—are given as follows.–In node 1, the discriminant condition is given by the following equation: Br .–In node 2, the discriminant condition is given by the following equation: Fe .
Terminal nodes 1, 2, and 3 mapped Japan, Korea, and China, respectively. These results suggest that bromine is an important parameter for the discrimination of production centers (characteristic of Korea samples, see Figure 4). Significant differences between Br content for China and Korea samples and between Japan and Korea samples were observed for multiple comparison of means values of Br content (ppm.) between China, Korea, and Japanese Undaria pinnatifidas by the Ryan-Joiner testing method . On the other hand, no differences were observed in I content between China and Korea and between Japan and Korea samples by the Ryan method ().
Bromine could not be completely dissolved and, further, Br has a tendency to vaporize in the acid decomposition sample preparation process for the ICP-MS. However in the case of X-ray fluorescence method, which is characterized by the rapid and easy handling, the above problem can be avoided and, hence, this could be a better and more promising method than ICP-MS. We earlier reported the rapidity and convenience of a discrimination method by near-infrared spectroscopy . By combining the two different methods of NIR and X-ray fluorescence methods, which essentially give different organic and inorganic information, respectively, before the final method, ICP-MS method, a more convenient and rapid discrimination method can be developed. One of the main reasons for this is that NIR and X-ray fluorescence methods do not require that powder samples be prepared as solutions. We conclude that the above combination of methods could be used as a convenient discrimination method to meet the regulatory requirements as those of 2006 for the display of the production centers on U. pinnatifida in Japan.
4. Conclusion(1)When using CART for the discrimination analysis on the production area from the elemental analysis results from the point of error rates, we take better discrimination results comparing to the LDA.(2)As one of convenient discrimination methods, we found the feasibility of using the elemental analysis, especially, the quantity of Br with the X-ray fluorescence analysis.
The author is grateful to Riken Food Company and the Director of the Department of Quality Control, Junichi Satoh, for providing a variety of Undaria pinnatifida samples and information about them. She is also grateful to Shimadzu Company for helping her with the X-ray fluorescence analysis and the grant from Sanriku Research Fund by Iwate Prefectural Government from 2005 to 2006.
M. Kaihara, K. Inaba, N. Kikuchi, and M. Sato, “Discrimination analysis on production areas of Undaria pinnatifida by using NIR spectra,” in Proceedings of the Summaries of Nondestructive Analysis Symposium, vol. 20, p. 162, 2004.View at: Google Scholar
M. Kaihara, K. Inaba, N. Kikuchi, and M. Sato, “Feasibility study of discrimination analysis of production area on wakame seaweeds,” Sanrikusogokenkyu, vol. 28, pp. 67–69, 2006.View at: Google Scholar
M. Kaihara and S. Kikuchi, “Discriminant analysis of countries growing wakame seaweeds: a preliminary comparison of visible-near infrared spectra using soft independent modelling, Randomforests and classification and regression trees,” Journal of Near Infrared Spectroscopy, vol. 15, no. 6, pp. 371–377, 2007.View at: Publisher Site | Google Scholar
M. Kadowaki and H. Tatsuguchi, “The first report, study of discriminating the original production area of boiled and salted Undaria pinnatifida,” Tech. Rep., pp. 24–34, FAMIC, 2005.View at: Google Scholar
FAMIC, “Manual for the discrimination of production centers of Undaria pinnatifida,” FAMIC, 2006.View at: Google Scholar
M. Kadowaki, Y. Hombu, M. Uchino et al., “The second report, study of discriminating the original production area of boiled and salted Undaria pinnatifida,” Tech. Rep., pp. 35–42, FAMIC, 2006.View at: Google Scholar
E. Nazuka, M. Kadokura, M. Kamiya, and K. Ariyama, “The third report, study of discriminating the original production area of boiled and salted Undaria pinnatifida,” Tech. Rep., pp. 1–8, FAMIC, 2006.View at: Google Scholar
L. Breiman, J. Friedman, R. Olshen, and C. Stone, Classification and Regression Trees, Chapman & Hall/CRC, NewYork, NY, USA, 1984.
A. Ohotaki, Y. Horie, and D. Steinberg, Applied Tree Based Method by CART, JUSE Press, Tokyo, Japan, 1998.
S.-X. Bao, Z.-H. Wang, and J.-S. Liu, “X-ray fluorescence analysis of trace elements in fruit juice,” Spectrochimica Acta B, vol. 54, no. 13, pp. 1893–1897, 1999.View at: Google Scholar
R. A. Fisher, “The use of multiple measurements in taxonomic problems,” Annals of Eugenics, vol. 7, pp. 179–188, 1936.View at: Google Scholar
W. P. Gardiner, Statistical Analysis Methods for Chemists: A Software-Based Approach, Royal Society of Chemistry, London, UK, 1997.