2018 Volume 66 Issue 4 Pages 353-357
Paeoniae Radix is one of the crude drugs frequently used in traditional Japanese medicine (Kampo medicine). It takes abundant labor and time to cultivate Paeonia lactiflora for medicinal use; high production cost is one of the main reasons why the domestic production of Paeoniae Radix is decreasing in Japan. To promote the production of Paeoniae Radix, we focused on Paeonia cultivars that produce commercially valuable flowers and investigated their possibility for medicinal use. We prepared 28 batches of peony roots derived from P. lactiflora, which were cultivated in Japan; 4 batches were crude drug samples, and 24 batches were cultivar roots. The elements contained in these samples were measured using inductively coupled plasma (ICP)-MS. The obtained data were then analyzed by principal component analysis (PCA) and back propagation artificial neural network (BPANN) analysis. No significant differences were found between the profiles of elements contained in crude drugs and cultivar roots. However, PCA results indicated a high similarity of the multielement fingerprints of crude drugs. Using the PCA results, we also assessed visible cluster trends and found that 5 batches of cultivars also showed fingerprints related to those of crude drugs. We certified this classification by BPANN. From the perspective of metallomics, our findings suggest that these 5 batches of Paeonia cultivars could be alternatives to crude drugs.
Paeoniae Radix (Latin name for crude drugs) is the dried root of Paeonia lactiflora PALLAS (Paeoniaceae), which is called peony in English and Shakuyaku in Japanese and is used frequently in traditional Japanese medicine (Kampo medicine).
Peony was first described in China in 200 B.C. and plant products were used in ancient Greece for their medicinal benefits.1) In Europe, herbaceous peony (P. officinalis) cultivars were prized between the 15th and 18th centuries, and breeding for flower color and shape began in the 19th century when P. lactiflora, which is native to China was introduced. In China, ornamental cultivars were developed in the 6–7th centuries from medicinal cultivars of P. lactiflora. P. lactiflora was introduced in Japan before the 10th century, and nearly 100 cultivars were created until the Edo period (17–19th centuries).2,3) In Japan, medicinal cultivars of P. lactiflora have been selectively bred and cultivated as Japanese peony and they are considered to have high quality. However, recently, some domestic Paeoniae Radix products were found to have originated from an ornamental cultivar or a cross between medicinal and ornamental cultivars.4)
In addition, although the amount of Paeoniae Radix imported from China is increasing, its domestic production is decreasing. Nishizawa et al. reported in 1986 that 40–50% of 200–250 metric tons of Paeoniae Radix used in Japan was domestically produced at that time.4) However, the self-sufficiency rate of Paeoniae Radix for the fiscal year 2014 was only 1.2%.5) Considering the difficulty in the assurance of quality and the price increase for Paeoniae Radix in China, it is necessary to promote its domestic production for stable supply. However, some problems should be resolved before promotion; one of the biggest problems is the reduction in the production cost. In fact, it takes approximately 3–5 years to harvest peony roots for medicinal use, and in that period, flowers and buds have to be picked to encourage the growth of root.6) This process requires abundant labor and time, further increasing the production cost; this economical problem is one of main reasons why its domestic cultivation is little.7)
We next focused on Paeonia cultivars. Generally, the flowers of medicinal cultivars are generally not so gorgeous that they can be marketable products.8) If ornamental cultivars that produce beautiful and commercially valuable flowers are medicinally available, farmers would be able to earn money both from flowers and roots.
In this study, we prepared 28 batches of P. lactiflora, which were all cultivated in Japan, comprising 24 batches of cultivar roots and 4 batches of crude drugs. We measured the elements not affected by processing and preserving conditions contained in these samples using inductively coupled plasma (ICP)-MS. First, to show the metallomic characteristics of crude drugs, we statistically analyzed the semiquantitatively measured multiple data of elements by principal component analysis (PCA). Then, we classified samples based on PCA and attempted to construct a model using back propagation artificial neural network (BPANN) to confirm the classification. We then sought some batches that showed metallomic fingerprints similar to those of crude drugs to investigate their possibility for medicinal use.
Of 28 batches of P. lactiflora that were collected, 4 batches were crude drugs (Table 1a) while 24 batches were cultivar roots (Table 1b). All samples were cultivated in Japan and all the samples shown in Table 1b were cultivated for non-medicinal use in the Medicinal Plant Center of Toyama Prefecture, Toyama, Japan. The herbal samples used in this study were all authenticated, genetically analyzed and certified to have the same characteristic nucleotide sequences in the nuclear ribosomal DNA intergenic transcribed spacer (ITS) regions as those of P. lactiflora.9,10) All of those samples were also confirmed to meet the criteria of paeoniflorin (not less than 2.0%) prescribed in the Japanese Pharmacopoeia11) (data not shown).
| No. | Sample name | Production area | TMPW No.a) | Collection date |
|---|---|---|---|---|
| D50 | Shakuyaku (Bonten) | Toyama, Japan | 25834 | 2.15.2008 |
| D51 | Shakuyaku (Bonten) | Toyama, Japan | 25835 | 2.15.2008 |
| D52 | Shakuyaku (Bonten) | Toyama, Japan | 25835 | 2.15.2008 |
| D7 | Shakuyaku (Yamato Shakuyaku) | Nara, Japan | 25818 | 2.14.2008 |
a) Specimen reference numbers at the Museum of Materia Medica, Institute of Natural Medicine, University of Toyama (TMPW).
| No. | Sample name | No. | Sample name | No. | Sample name |
|---|---|---|---|---|---|
| 3 | Raspberry thunder | 13 | Kouga | 30 | Fujimusume |
| 4 | Haru-no-niji | 15 | Blue sapphire | 31 | Kitasaisho |
| 6 | Momoyama | 17 | Miyama-no-yuki | 34 | Bonten |
| 8 | Highlight | 19 | Harmony | 37 | Primavera |
| 9 | Bridal icing | 22 | Queen red white | 38 | Flora |
| 10 | Rosario | 24 | Rose glory | 41 | Madame purple |
| 11 | Meigetsu | 26 | Shirayuki | 43 | Ruisugeiji |
| 12 | Sweet 16 | 29 | Mine-no-yuki | 45 | Ohgon |
These samples were cultivated at the Medicinal Plant Center of Toyama Prefecture, Toyama, Japan, and collected on October 24, 2007.
All samples (10–50 g) were dried and ground to powder. To 4 mg of each sample, 1 mL of HNO3 (Nacalai Tesque, Kyoto, Japan) was added; samples were then vortexed and maintained overnight at room temperature. Then, 0.2 mL of samples were diluted with 9.8 mL of water and filtered through a 0.45 µm-pore size hydrophilic polytetrafluoroethylene membrane filter (Millipore, Billerica, MA, U.S.A.). ICP-MS analysis was performed on Agilent 7500 Series ICP-MS (Agilent Technologies, Inc., Santa Clara, CA, U.S.A.).
Data AnalysisThe profiles of the 45 elements detected were imported to Pirouette software (Infometrics, Inc., Bothell WA, U.S.A.) to perform PCA. Based on PCA results, samples were divided into some groups.
Using those groups as classes, genetic algorithm (GA) was applied to the data of elements to select the variables. In this study, the initial population was 30 chromosomes, and the maximum number of allowed variables in a solution was 10. All chromosomes had a mutation probability of 0.01, and two-point crossovers were imposed with a probability of 0.5. Cross-validation was also performed.
After GA application, the data of selected elements were used for BPANN. Both GA and BPANN were conducted using Chemish software.12)
Elements contained in each sample were measured semiquantitatively by ICP-MS. A total of 45 elements were detected. The maximum concentrations of boron (B), sodium (Na), magnesium (Mg), potassium (K), and calcium (Ca) detected in the sample solvents exceeded 80 µg/L (1 mg/g in the original samples), and those of aluminum (Al), manganese (Mn), iron (Fe), and selenium (Se) detected in the sample solvents exceeded 8 µg/L (0.1 mg/g in the original samples). The profiles of these 9 main elements present in each sample are shown in Fig. 1. The maximum concentrations of copper (Cu), strontium (Sr), bromine (Br), rubidium (Rb), barium (Ba), and lead (Pb) were 0.8–8 µg/L (0.01–0.1 mg/g in the original samples), and those of lithium (Li), scandium (Sc), titanium (Ti), vanadium (V), chromium (Cr), nickel (Ni), gallium (Ga), arsenic (As), zirconium (Zr), molybdenum (Mo) and iodine (I) were less than 0.8 µg/L (0.01 mg/g in the original samples). The concentrations of cobalt (Co), germanium (Ge), yttrium (Y), silver (Ag), cadmium (Cd), tin (Sn), antimony (Sb), cesium (Cs), lanthanum (La), cerium (Ce), praseodymium (Pr), neodymium (Nd), hafnium (Hf), tungsten (W), mercury (Hg), thallium (Tl), bismuth (Bi), thorium (Th), and uranium (U) were not greater than 0.08 µg/L (1 µg/g in the original samples). It seems that some individual differences occurred in each profile, but no significant differences were found between the profiles of P. lactiflora cultivars and crude drugs (Fig. 1). Then, we analyzed their profiles by PCA (Fig. 2). Samples were separated into several clusters, with crude drug samples being the smallest cluster. Five samples of cultivars (sample 3, 6, 9, 13, 15) were also included in that cluster.
The data show the concentrations detected in each sample solvent. Nine elements with the maximum concentrations of more than 0.1 mg/g sample are shown.
P. lactiflora cultivars are indicated by open diamonds, and crude drug samples are indicated by closed circles. Each point represents the mean value (n=3).
Based on PCA results, our samples were divided into 5 clusters (Fig. 2). The clusters were as follows: cluster 1 (samples 3, 6, 9, 13, 15, D7, D50, D51, D52); cluster 2 (samples 8, 10, 11, 12, 24); cluster 3 (samples 4, 19, 41, 43); cluster 4 (samples 22, 26, 29, 31, 34); and cluster 5 (samples 17, 30, 37, 38, 45). We applied these grouping into BPANN. First, we attempted to use the data of all the elements detected by ICP-MS, but the analysis was unsuccessful.
Then, we used GA to select the variables appropriate for classification. The classes shown above were applied to the analysis. As a result, 9 elements (Na, Ca, Cu, Ge, Se, Br, Sr, Mo, Ag) were selected (Q2=0.894).
The data of those 9 elements were used for BPANN. Cross-validation and optimization of the hidden layer were also performed. The optimized number of the hidden layer was 17, and the learning rate was 110. The model’s R2 and Q2 values were 0.95 and 0.822, respectively. The model was well established as shown in Fig. 3. Our results suggest that some samples (samples 3, 6, 9, 13, 15) have metallomic profiles similar to those of crude drug samples.
Experimental classifications were based on back propagation artificial neural network (BPANN).
One sample (No. 4) was comparably isolated in class 3 (Fig. 3). We compared the pattern of the major elements contained in sample 4 with the average profiles of elements contained in crude drugs and cultivars in class 1, class 2 and class 3 (Fig. 4). The overall pattern of major elements contained in sample 4 was partly similar to that of the elements contained in cultivars in classes 1 and 2. However, the concentration of Ca was considerably low compared with the other samples, and we suppose this is why sample 4 was not classified well.
a) Typical chart of inductively coupled plasma (ICP)-MS of sample 4 and the average profiles of cultivars from class 1 (class 1 cul), crude drugs from class 1 (class 1 cru) and cultivars from classes 2 and 3. Thirteen elements with the maximum concentrations of more than 1 mg/L in the sample solvent (0.0125 mg/g samples) are shown. The data are shown as a relative concentration, which was calculated by setting the maximum concentration of each element contained in samples as 1 (n=3). b) Elements detected in each sample by ICP-MS. The data show the concentrations detected in sample 4 or the average concentration of samples in suggested groups. Six major elements are shown.
In the present study, we compared the metallomic profiles between the crude drugs and cultivars of P. lactiflora. We found several cultivar samples with patterns of elements similar to those of crude drugs samples, which have homogenous profiles of elements.
Recently, metabolomic methods using GC-MS, LC-MS, or 1H-NMR have frequently been applied for assessing of the quality of crude drugs. Tarachiwin et al. and Tianniam et al. attempted to apply 1H-NMR and GC-MS, respectively, to evaluate the sensory quality of Angelicae Radix.13,14) However, both of their studies suggested that geological factors strongly affected the patterns of metabolite fingerprints rather than the sensory quality. On the other hand, the agricultural studies have revealed a relationship between cultivars and inorganic elements they contain. Galdón et al. and Chope and Terry used onion cultivars (Allium cepa L.) grown under the same agronomic, soil, and climatic conditions to demonstrate the presence of differences in mineral and trace element contents between onion cultivars.15,16) In this study, most samples we used were also cultivated in the same field. Moreover, in the former study, we successfully verified the complex origin of Paeoniae Radix not only by the production area but also the genetic type using the element profiles attained by ICP-MS.10) These results suggest that metallomic analysis has the possibility to be a simple method for distinguishing the quality of Paeoniae Radix owing to genetic characteristics.
In this study, we applied several statistical methods to analyze ICP-MS results. First, we used PCA to visualize the patterns of elements contained in samples as it is usually used for multivariate data of “omics” studies.14,17) Crude drug samples from a well-bred strain for medicinal use were plotted in smaller area compared with P. lactiflora cultivars. This indicated that patterns of elements contained in crude drug samples were considerably similar when compared with those of cultivars. Several strains used for crude drugs have considerably homogenous quality from the perspective of metallomics. However, several cultivar samples were also found in the same cluster as crude drug samples. If those cultivars are found to exhibit metallomic characteristics similar to those of crude drug samples, it is likely that the quality of those cultivars is similar to that of crude drugs. To confirm that classification, other statistical methods were used.
Based on PCA results, we divided samples into 5 groups. We then applied BPANN to construct a non-linear model to confirm the classification predicted by PCA results. BP is a supervised learning technique, and ANN is one of the most widely used methods for constructing a non-linear model and feedforward network, which is composed of 3 layers: input, hidden and output layers. It has been used in engineering and business applications, and is recently also adopted in biomedical research.18–20) Before BPANN analysis, we also applied GA, a stochastic method applying Darwin’s evolution hypothesis and used as the variable selection methods optimizing by the crossover and mutation operation.18,21,22) Using these statistical methods, we finally succeeded in constructing a good model (Q=0.822) according to the classification attained by PCA.
Recently, many studies have been conducted to identify or generate the strains of P. lactiflora that are both economically and medicinally superior to promote the domestic cultivation of Paeoniae Radix, and most of them were from the perspective of organic ingredients.4,8,23,24) In this study, we found that several cultivar samples (samples 3, 6, 9, 13, 15) had the similar metallomic profiles to those of crude drugs that were relatively homogenous. We demonstrated the possibility of usefulness of those cultivars for medicinal use and our metallomic analysis as a simple method for verifying the quality of Paeoniae Radix. However, in the model we constructed, we observed 1 isolated sample (No. 4) in class 3 (Fig. 3). We compared the patterns of the major elements contained in sample 4 and found several differences between the other samples. Sample 4, a cultivar strain known as “Haru-no-niji,” was recently indicated to have stronger antioxidative activity compared with other medicinal strain of P. lactiflora.25) Furthermore, this sample contained comparatively high concentration of paeoniflorin, one of the major active ingredients in Paeoniae Radix.26,27) We could not select this strain as a candidate, but it could be a candidate from another perspective. To identify the strain more suitable for medicinal use, it should be evaluated from various aspects using organic, pharmacological and metallomic analysis.
This study revealed that individual crude drug samples of Paeoniae Radix (P. lactiflora) have a homogenous quality when compared with P. lactiflora cultivars. We also found that the profiles of elements contained in some cultivar strains considerably resembled to those contained in crude drugs. These strains have possibility to be the alternatives to be used for crude drugs.
We thank Mr. Morikazu Murakami, Toyama Prefectural Medicinal Plants Center and Dr. Shu Zhu, Institute of Natural Medicine, University of Toyama, for the support with the materials. We also thank Dr. Tadashi Saito, Osaka University for the technical support with ICP-MS analysis. This study was supported by a Grant-in-Aid for Scientific Research (B), No. 25282071, 2013–2015 from the Japan Society for the Promotion of Science (JSPS). Partial support was also provided by Grant-in-Aid for Young Scientists (B), No. 15K19150, 2015-2017 from JSPS, Grant-in-Aid for Scientific Research (A), No. 17H00832, 2017-2019 from JSPS, and Ministry of Agriculture, Forestry and Fisheries (MAFF), Japan Grant.
The authors declare no conflict of interest.
