Our sampling strategy was as follows: for herbal drugs derived from a typical original plant and other species of the same genus, we collected the typical original plant and another plant that is often used instead of the original plant. For herbal drugs derived from a typical original plant and its varieties, we collected only the typical original plant. For the remaining herbal drugs derived from unambiguous species, we collected their specific origins according to the Korean pharmacopeia monographs. Both leaf samples and medicinal materials were collected from the original plants and pharmaceutical medicine markets, respectively.

To provide a good representation of the intraspecific genetic variability and the true situation in traditional herbal drug markets, we gathered 1,054 specimens to provide coverage for 208 species of 153 herbal drugs, including 530 medicinal materials and 524 original plant leaf materials. To provide an example for authenticating real herbal drugs, we collected 27 herbal drug materials from the Seoul Yangnyeongsi herbal medicine market, which is the largest herbal medicine market in Korea. The above-mentioned specimens are preserved at the Institute of Medicinal Plant Development, Chinese Academy of Medicinal Sciences. Detailed information regarding the data collection and the voucher information is provided in Table S1 and Table 1.

Total genomic DNA was isolated from 20 to 30 mg of the leaf samples or 50~100 mg of the medicinal materials using a Hi-DNAsecure Plant Kit (Tiangen Biotech Co., Beijing, China). The PCR amplification was performed in a 25 μL reaction mixture that contained 2 μL genomic DNA, 12.5 μL PCR MasterMix (Aidlab Biotechnologies Co., Beijing, China), 8.5 μL ddH₂O, and 1 μL each of the forward and reverse primers (2.5 μM, synthesized by Sangon Co., China). The primers and the reaction conditions followed previously published methods for ITS2 and psbA-trnH (Chen et al., 2010; Group et al., 2011). The PCR products were purified using the QIAquick PCR purification kit (Tiangen Biotech, Beijing, China), and the bidirectional sequencing was accomplished with an ABI 3730XL sequencer using the original amplification primer.

Data analysis was accomplished with a general pipeline: First of all, DNA barcodes were assembled and preprocessed. Secondly, DNA barcodes, together with specimens and references, were mapped on to species levels and formed the DNA barcode reference library of Korean pharmacopeia. Thirdly, the Korean pharmacopeia identification engine (KP-IDE) were developed and optimized. Finally, an independent dataset was applied to validate the reference library and identification engine. Common methods and data analysis techniques were as follow. High-quality bidirectional sequences (contigs) were assembled using CodonCode Aligner 6.0.1 (CodonCode Corporation, Massachusetts, USA). The delimitation and identification of the ITS2 and psbA-trnH core regions were achieved using HMMER 3.0 (http://hmmer.org). The sequence alignment analysis was accomplished using Muscle 3.8.1 (Edgar, 2004) and was restricted to the sequence after the annotation. The sequence divergences for the DNA barcode ITS2 and psbA-trnH region were calculated using PAUP^* 4.0 (Swofford, 2002) and MEGA7 (Kumar et al., 2016) using the Kimura-2-Parameter (K2P) distance metric. Neighbor-joining (NJ) trees were constructed based on the Muscle alignment results according to the K2P distances using MEGA7. Unknown specimens were identified using the (KP-IDE, Data Sheet 1) and herbal drug barcode libraries. KP-IDE was developed using Python 2.7 and Biopython 1.68, and the NCBI BLAST software was used as a similarity search engine to collect the nearest neighbors from the DNA barcode reference library.

The DNA barcode reference library of the Korean pharmacopeia contains the following four types of records: taxonomy records, specimen records, sequence records, and reference records (Figure 1). The taxonomy records include a collection of standard nomenclature and classification repositories of the other three types of records. The taxonomy records include organism scientific names, synonyms, and taxonomic lineages for each of the other record types. All specimen records, sequence records, and reference records are closely linked to the original nomenclature of the herbal drugs, and all records must map onto the origin's taxonomy entry at or below the species level. Currently, there are 6,777 taxonomy records (Data Sheet 2). The specimen records assemble various collateral data, particularly taxonomic information regarding specific herbal drug specimen, such as the locations in which the specimen were captured and other useful information. Each specimen record is linked to a companion sequence record. Currently, there are 1,054 specimen records included in this library (Data Sheet 3). The sequence records are a collection of DNA barcodes related to Korean herbal drugs and their closely related species. Currently, there are 30,744 deposited sequences, including 22049 ITS2 (Data Sheet 4) sequences and 8695 psbA-trnH sequences (Data Sheet 5). The reference records store all the bibliographies, including the publicly available DNA barcodes for Korean herbal drugs, and currently, there are 285 records (Data Sheet 6).

The above-mentioned types of records were embedded in the (KP-IDE) to facilitate the use of this reference library for identifying unknown specimens (Figure 1). The KP-IDE was written in Python and can operate very well on a Unix-like operating system. Unknown specimens were identified by pasting their barcode sequence into a fasta format file. The query sequence must satisfy the following criteria: appropriate length (ITS2 ≥ 150 bp; psbA-trnH ≥ 100 bp), confusion bases ≤ 1% and no invalid bases. The query file was transferred to KP-IDE by the parameter “–q” or “–query.” Then, a BLAST search tool was used to collect the nearest neighbors from the user specified reference sequence record database or the default species level barcode record database (SBD). The parameter used for the transmission of the database is “–d” or “–database.” Currently, the species level barcode record database contains 5,986 sequences for 1,006 species, including 3770 ITS2 sequences and 2216 psbA-trnH sequences. The other two databases are the public record barcodes database (PBD) and all record barcodes on BOKP database (ABD). The public record barcodes database is a collection of records from published articles related to Korean herbal drugs and currently include 12,045 sequences from 2,983 species, including 8006 ITS2 sequences and 4039 psbA-trnH sequences. The final type of database consists of all record barcodes related to Korean herbal drugs that are available on BOKP. Currently, there are 30744 sequences from 6,528 species, including 22049 ITS2 sequences and 8695 psbA-trnH sequences. When the identification engine retrieved a result, the KP-IDE provided the taxonomic assignment (Figure 2A), sequence similarity (Figure 2B) between unknown and top nearest reference sequences, NJ Tree (Figure 2C) showing top nearest reference sequences, and sequence alignment (Figure 2D) between unknown and top nearest reference sequences.

All information, such as the specimen records, sequence records, and reference records, is closely linked by the organism nomenclature and must map onto the origin's taxonomy. Each taxonomy entry in this library includes an authority name, a primary name (i.e., the origin's name in the Korean pharmacopeia) and any number of synonyms in the Chinese pharmacopeia, the Flora of China and the NCBI taxonomy databases (only maintain the scientific name). Of the 232 herbal drug origins in the Korean pharmacopeia, 171 origins have the same nomenclature in the Chinese pharmacopeia, 20 origins have synonyms in the Chinese pharmacopeia, and 41 origins are not recorded in the Chinese pharmacopeia. Moreover, 143 origins have nomenclatures that are the same as the flora of China, 61 origins have synonyms to the flora of China, and 28 origins are not recorded in the flora of China. In total, 168 origins have the same nomenclatures in the NCBI taxonomy database, 56 origins have synonyms in the NCBI taxonomy database and 8 origins are not recorded in the NCBI taxonomy database. Detailed information regarding the origin nomenclature is provided in Table S2.

Here, the following six words were used to describe the differences in the origins between the Korean pharmacopeia and the Chinese pharmacopeia: absence, identical, alternation, expansion, contraction, and repartition. For the 159 herbal drugs obtained from plants and fungi in the Korean pharmacopeia, the following 10 herbal drugs were not recorded in the Chinese pharmacopeia (absence, 6%): Aralia Continentalis Root, Cardamon, Condurango, Gentian, Ipecac, Kalopanax Bark, Scopolia Rhizome, Senega, Swertia Herb, Valerian Root, and Rhizome. For the 149 herbal drugs that were recorded in both the Chinese pharmacopeia and the Korean pharmacopeia, 109 herbal drugs had the same origin (identical, 69%). 17 herbal drugs in the Korean pharmacopeia had more materials from the origin than those in the Chinese pharmacopeia (expansion, 11%), and 13 herbal drugs had less materials from the origin than those in the Chinese pharmacopeia (contraction, 8%). Compared with those in the Chinese pharmacopeia, seven herbal drugs in the Korean pharmacopeia had alternative origin materials (alternation, 4%). Finally, the origins of three herbal drugs were divided into two different herbal drug categories in the Chinese pharmacopeia. Epimedium Herb is the aerial part of Epimedium koreanum, E. brevicornum, E. pubescens, E. wushanense, and E. sagittatum, whereas E. wushanense is the origin of an independent herbal drug, Epimedium wushanense Herb, in the Chinese pharmacopeia. Liriope Tuber is the tuber of Liriope platyphylla and Ophiopogon japonicas, while Ophiopogon japonicas is the origin of another herbal drug, Ophiopogon Tuber, in the Chinese pharmacopeia. Phellodendron Bark is the bark of Phellodendron amurense and P. chinense; however, the two species are the origin of Phellodendron Amurense Bark and Phellodendron Chinense Bark, respectively, in the Chinese pharmacopeia. Detailed information regarding the differences in the origins of the herbal drugs in the Korean pharmacopeia and the Chinese pharmacopeia is provided in Table S3.

The following six parameters were used to characterize the inter-specific divergence and intraspecific divergence for herbal drug origins and their closely related species in the species level barcode record database: (i) average inter-specific distance, (ii) theta prime, (iii) average minimum inter-specific distance, (iv) average intraspecific distance, (v) theta, and (vi) average coalescent depth. The six parameters used for the ITS2 sequences are as follows: 0.0484 ± 0.0468, 0.0637 ± 0.0506, 0.0237 ± 0.0364, 0.0018 ± 0.0035, 0.0029 ± 0.0042, 0.0049 ± 0.0065; the six parameters used for the psbA-trnH sequences are as follows: 0.0245 ± 0.0549, 0.0535 ± 0.0992, 0.0107 ± 0.0386, 0.0023 ± 0.0276, 0.0037 ± 0.0237, 0.0066 ± 0.0437. The distributions of average inter-specific distance, the minimum inter-specific distance, average intra-specific distance, and coalescent depth for each species are shown in Figure 3A (ITS2) and Figure 3B (psbA-trnH).

Twenty-seven herbal drug materials of 23 herbal drugs were collected from the Seoul Yangnyeongsi herbal medicine market to give an example for identifing real market samples. High-quality DNA was easily extracted from all 27 samples; 22 samples were amplified ITS2 barcodes, whereas the other five samples were amplified psbA-trnH barcodes. When performing the authentication using KP-IDE, 26 samples were matched with their labels. The herbal drug Rubus Fruit was an exception and was matched to the genus species Rubus crataegifolius rather than to its official origin Rubus coreanus. The identification results are shown in Table 1.

The DNA barcoding method built by Paul Hebert still has a major impact on species identification decades after its development (Hebert et al., 2003). The DNA barcode reference library has been shown to be a reliable resource that is essential for the identification of unknown specimens related to public health and safety concerns, such as food and herbal drugs (Stoeckle et al., 2011; Galimberti et al., 2013). The consortium for the barcode of life has initiated several campaigns for large groups of animals and provided an informatics workbench for the acquisition, storage, analysis, and publication of DNA barcode records (Ratnasingham and Hebert, 2007). The fish barcode of life (Fish-Bol, http://www.fishbol.org) may be the most influential campaign and has been widely employed to identify commercial seafood products (Ward et al., 2009). After more than 10 years of development, Fish-Bol assembled a standardized reference DNA sequence library for more than 8,000 fish species, which covered a vast majority of the world's most important commercial species (Becker et al., 2011). Subsequently, Hanner et al. procured 254 seafood samples from numerous retail establishments located in five Canadian metropolitan areas and demonstrated that 41% of the samples were mislabeled (Hanner et al., 2011). Wong et al. indicated that 25% of the 91 market seafood samples were potentially mislabeled in North America (Wong and Hanner, 2008). DNA barcoding has been demonstrated to be a powerful tool for seafood authentication and was adopted by the U.S. Food and Drug Administration to identify seafood products (Deeds et al., 2014).

Although there are several published DNA barcode reference libraries for herbal drugs, there is a lack of an informatics workbench, such as the BOLD system, which would allow the establishment of projects and private data repositories. The limitation of a private data repository and sharing hamper the ability of data increment and the flexible application of herbal drug authentication. In 2010, the first medicinal materials DNA barcode database (MMDBD, www.cuhk.edu.hk/icm/mmdbd.htm) was published by Lou et al. (2010). The latest update of the MMDBD was completed in 2014, and 51,375 sequences from 1,661 species were deposited. In 2014, Chen et al. established a preliminary system for the DNA barcoding of herbal materials (www.tcmbarcode.cn), which contained 78,847 sequences belonging to 23,262 species (Chen et al., 2014). In 2016, Vassou et al. created a reference DNA barcode library for the authentication of Ayurvedic medicines (Vassou et al., 2016). This barcode library assembled 374 medicinal plants and revealed that only 79% of raw drugs were authentic. In 2017, Chen et al. sequenced ~95% of the species recorded in the JP and constructed an online DNA barcode identification system (http://www.jpbarcode.com; Chen et al., 2017). Here, we provide an open source identification engine that permits researchers interested in herbal drug authentication to use DNA barcoding to study, change, and distribute this software to anyone for any purpose. Users who are not familiar with bioinformatics may use the parameter “–d” followed with “SDD” or “selfDefinded” to use their private dataset.

Korean herbal drugs following the Korean pharmacopeia quality standards are generally recognized as safe and efficacious. However, the inappropriate use of herbal drugs due to misidentification poses considerable health risks. To combat the health risks associated with herbal drug misidentification, various methods for herbal drugs identification have been developed worldwide, primarily including morphological, microscopic, and physicochemical identification (Choi et al., 2002; Kim et al., 2010; Sahoo et al., 2010). These techniques are useful in herbal drug authentication, but there are also some limitations. The accuracy of the morphological and microscopic identification results faces significant challenges due to a dwindling professional workforce (Miller, 2007), and the authenticity of the physicochemical identification results is compromised as closely related species may share similar plant ingredients (Chen et al., 2014). For example, the identification method for Akebia stem is described in the Korean pharmacopeia as follows: weigh 0.5 g of pulverized Akebia stem, add 10 mL of water, boil, allow to cool and shake vigorously, and a lasting fine foam is produced. This soap-like foaming phenomenon is produced by saponins that are present in the entire Akebia genus, Araliaceae and Liliaceae plants and not only in the diagnostic component of the origin plant Akebia quinata. Moreover, several other herbal drugs in the Korean pharmacopeia share these foam properties, such as Anemarrhena rhizome, Aster root and rhizome, and Codonopsis pilosula root.

DNA barcoding technology has been formally used for medicinal plant identification since 2010 (Chen et al., 2010) and has been shown to be a valuable addition to not only traditional identification methods but also practical field applications for the identification of medicinal plants and authentication of herbal drugs (Han J. et al., 2016). For example, the herbal drug Akebia stem has long been confused with Caulis aristolochiae manshuriensis in Asia and Europe. Akebia stem is the stem of A. quinata (Lardizabalaceae) whereas Caulis aristolochiae manshuriensis is the stem of Aristolochia manshuriensis (Aristolochiaceae). A. manshuriensis and other species of the same genus contain aristolochic acid (AA), which causes severe nephropathy. Although herbal drugs containing AA were rigorously administered in nearly all countries, the incidence of AA-contamination remains a worldwide problem (Debelle et al., 2008). To avoid the health risks due to AAs, Wu et al. (2015) provided a DNA barcoding-based authentication system that can efficiently and reliably distinguish Aristolochiaceous from non-Aristolochiaceous materials. The DNA barcode reference library established here contains nearly all herbal drugs listed in the Korean pharmacopeia, and the herbal drug authentication test using real market samples collected from the Seoul Yangnyeongsi herbal medicine market further validated the value of the BOKP in the routine authentication of herbal drugs.

BOKP is the primary system used for the authentication of herbal drugs in the Korean pharmacopeia. Notably, the assembly of incorrectly identified sequences, poor quality sequences or unauthorized sequences threatens to exasperate the herbal drug authentication crisis. Moreover, the validation of data records is a time-consuming and labor intensive process. Thus, an automatic data pre-processing software should be developed to minimize manual system maintenance and updating. As reported elsewhere the marker of psbA-trnH has several shortcomings, such as inversions, rps19 insertions, insertions/deletions, AT-rich direct repeats, and 2-3 tandem repeats (Pang et al., 2012). For example, a core ATGAAAAC/GTTTTCAT inversion exists among the origins of the herbal drug Polygonatum rhizome. A core AACAAAAC/GTTTTGTT inversion exists among the origins of the herbal drug Coptis rhizome. These inversions may lead to a substantial overestimation of the intra-/inter-specific distances and pose a challenge for species identification using both the blast and distance identification method (Whitlock et al., 2010). Moreover, the more frequent occurrence of the AT-rich direct repeats and tandem repeats in the origins of herbal drugs can lead to problems in obtaining bidirectional sequences and accurate identification results using blast-like methods (Group et al., 2009). Thus, a rigorously preprocessed protocol should be implemented automatically.

Economic development and increased international trade are leading to higher rates of inaccurate herbal drug identification, which risks patient safety and herbal efficacy. The failure in identification is undoubtedly due to nomenclature confusion, i.e., the phenomenon of synonymy. For example, Evodia Fruit is the fruit of Evodia rutaecarpa, E. rutaecarpa var. officinalis or E. rutaecarpa var. bodinieri (Rutaceae). This herb is slightly toxic due to the presence of many alkaloids and should not be overdosed or used for a long period. The authorized name for the three Evodia Fruit origins is integrated into Tetradium ruticarpum in the Flora of China and the NCBI taxonomy database. Therefore, if an unknown specimen is identified as T. ruticarpum, the identification engine should note the nomenclature transition to avoid confusing a Korean user. In total, 26% of the origins of Korean herbal drugs have synonyms in the flora of China, and 24% of the origins have synonyms in the NCBI taxonomy database. This synonymy is a general phenomenon for which the taxonomic issues (synonymies, misspellings, and alternate classifications) of nomenclature for Korean herbal drug origins should be resolved before the addition of new sequence data to the DNA barcode reference library.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.