1.       Introduction

Plants are the major contributors of natural products and are usually rich in nutritional or medicinal properties, which are attributed to the complex secondary metabolite constituents of them [1-3]. Plants are an important source of novel pharmacologically active compounds with many pharmaceutical drugs have been derived directly or indirectly from plants, and have played a central role in human health-care since ancient times [4-6]. Many pharmaceutical drugs are derived from plants that were first used in traditional systems of medicine [6]. According to the World Health Organization, about 25% of medicines are plant-derived [2].

Discoveries of novel molecules and advances in production of plant-based products have revived interest in natural product research [7,8]. The number of traditionally used plant species worldwide is estimated to be between 10,000 and 53,000; however, only a small proportion have been screened for biological activity [9-11], and the plants from some regions are less studied than others. Moreover, the potential of plants to yield new valuable drugs is under threat due to the alarming bio-diversity loss, with recent estimates indicating that every fifth plant species on earth is threatened with extinction [12]. Therefore; there is an urgent need for a time-efficient and systematic approach for unlocking the potential of plants in drug discovery. 

A correlation between phylogeny and biosynthetic pathways could offer a predictive approach enabling more efficient selection of plants for drug discovery. Following the assumption that plant-derived chemicals are constrained to evolutionary plant lineages, phylogeny-guided approaches have been seen as one of the time-efficient and informed approaches to plant-based drug discovery [13,14]. A series of studies have been conducted and verified that phylogeny is an efficient tool to facilitate drug discovery for diverse genera across different regions or cultures [13-18]. However, most of these studies focused on only a small cluster of genera, which limits its practical application. This approach also faces the limitation of incomplete sequence data. Moreover, phylogenetic distance correlated to feature similarity of species will also be invalid once beyond a certain threshold [19]. Therefore, a special perspective different from sequence-based phylogeny is valuable for understanding the evolution of bioactive features and facilitating the prediction and discovery of medicinal properties in plants.

Besides molecular biology which is in the view of nucleotide sequence comparison, metabolite feature is also closely related to the evolution of pathways for both primary and secondary metabolites. Many researchers have begun to explore phylogenetic distance between species from the diversity of metabolite features, either alone or in combination with sequence features. Clemente et al. (2007) presented a method for assessing the structural similarity of metabolic pathways for several organisms and reconstructed phylogenies that were very similar to the National Center for Biotechnology Information (NCBI) taxonomy [20]. Borenstein et al. (2008) predicted the phylogenetic tree by comparing the “seed set” of metabolic networks [21]. Mano et al. (2010) considered the topology of pathways as chains and used a pathway-alignment method to classify species [22]. Chang et al. (2011) proposed an approach from the perspective of enzyme substrates and corresponding products in which each organism is represented as a vector of substrate-product pairs. The vectors were then compared to reconstruct a phylogenetic tree [23]. Ma et al. (2013) demonstrated the usefulness of the global alignment of multiple metabolic networks to infer the phylogenetic relationships between species [24]. A. A. Abdullah et al. (2015) classified microorganism species based on the volatile metabolites emitted by them, and the results have been well explained in terms of their pathogenicity [25]. However, most of these studies have focused on microorganisms such as archaea, and only a few studies have involved land plants and the bioactive compounds produced by them.

The systemization of plants on the basis of their chemical constituents, which is also known as plant chemosystematics, could be helpful in solving taxonomical problems and exploring nutritional and medicinal properties from plants. Traditional chemosystematics of plants is based on the presence of selected metabolites. The incomplete data of metabolite constituents of plants limits its ability to solve taxonomical problems and discover new natural products or medicinal properties from plants [26, 27]. To perform a holistic review on the metabolite features of a species, we propose the concept of metabolite-content. Metabolite-content refers to all small molecules which are the products or intermediates of metabolism within an organism. It differs from metabolome in that the metabolite-content focuses on the qualitative collection of small metabolites and ignores the quantitative differences, which is instable with different parts and stages of one organism.

The secondary metabolite constituents of a plant are highly related to its pathways which are constrained to evolutionary phylogeny, and also related to the bioactive compounds of the plant which determine the medicinal and nutritional features of it [26]. Comparative classification of plants based on their metabolite-content-similarity could be used to explore the evolutionary and bioactive relation between them [28]. Here, we investigate the phylogenetic value of metabolite-content data, especially the predictive power of metabolite-content data in exploration of medicinal and edible plants for bioprospecting, using the KNApSAcK Core DB.

The KNApSAcK Core DB is an extensive plant-metabolite relation database that can be applied in multifaceted researches of plants, such as identification of metabolites, construction of integrated databases, bioinformatics and systems biology [29-32], and can be considered an advanced source of metabolite-content data of plants. The KNApSAcK Core DB contains 111,199 species-metabolite relationships that encompass 25,658 species and 50,899 metabolites, and these numbers are still growing [30].

In this paper, we reconstructed the phylogenetic tree for a set of plants which are distributed in different genera and families by their metabolite-content data obtained from KNApSAcK Core DB. We used a network based approach to abstract structurally similar metabolite groups as features, and measure the phylogenetic distance by a binary method. We also reconstructed the phylogenetic tree based on common DNA barcodes for a subset of plants, to investigate the predictive power of these two approaches, sequence- and metabolite-content-based method, in guiding the prediction of medicinal/edible plants for bioprospecting.

        2.       Material and Methods

        2.1.  Dataset and Preliminaries

The input metabolite-content data are species-metabolite relationships obtained from the KNApSAcK Core DB, which is a part of the KNApSAcK Family DB [30]. The KNApSAcK Core DB contains most of the published information about species-metabolite relations, but this is obviously far from complete regarding plants and other living organisms. We removed the plants with inadequate plant-metabolite relations to guarantee that the amount of metabolite-content of selected plants is sufficient enough to reveal their interrelations. The KNApSAcK Core DB also provides MOL molecular structure files for the metabolite compounds. We used R package ChemmineR (v2.26.0) to generate atom pair fingerprints from molecular structure description files [33]. And these molecular fingerprints were used to measure the structural similarity for all the metabolite pairs.

In this study, we also reconstructed phylogenetic tree for the same plant samples we used before based on three common DNA barcodes: two chloroplast barcodes rbcL and matK, and one nuclear barcode ITS2. The DNA sequence data are collected from GenBank [34], and certainly there is lack of data for some plants. Here we select the plants with both abundant metabolite-content (no less than 30 metabolites) and corresponding DNA barcode data as samples. There are 190 plants in total belong to 51 different families, with 172 plants in rbcL group, 165 plants in matK group and 160 plants in ITS2 group (Figure 1).

        2.2.  Phylogenetic Hypothesis

In this study, we produce phylogenetic hypothesis for each groups of samples by compiling DNA sequence data from the plastid markers rbcL, matK and nuclear marker ITS2 respectively. The sequence data of rbcL, matK and ITS2 are aligned by Clustal X 2.0 to compensate the missing and gapping data. Bayesian analyses of each sample groups were performed with MrBayes v3.2 [35,36]. We produced Bayesian phylogenetic hypothesis using the  model (Parameters: lset NST = 6 RATES = gamma). For each group we perform the analysis with more than 1,000,000 generations. The average standard deviation of the split frequencies (i.e., the average of all standard deviations of all observed splits between two independent analyses from different random trees) is down to <0.05 after the analysis is finished.

        2.3.  Clustering of Plants Based on Metabolite-Content Similarity

For classifying plants based on currently available metabolite-content data, firstly we need an approach that can compensate for the limitations of missing data. Adjacent metabolites along a metabolic pathway are often related to similar substructures, and structurally similar metabolites are involved in the same or similar pathway. Therefore, plants that share highly structurally similar metabolites are likely to be within the same category and represent similar bioactivity. In this study, we linked plants to structurally similar metabolite groups instead of individual metabolites.

We used the Tanimoto coefficient to measure the structural similarity between two metabolites and constructed a network of metabolites based on chemical structure similarity [37]. The Tanimoto coefficient between two metabolites  ased on chemical structure similarity [37]. The Tanimoto coefficient between two metabolites

Results and Discussion

All of the sequence data were downloaded from GenBank (Table 1). It should be noted that not all samples have complete sequence data (Table 2). The ubiquitous missing and incomplete sequence data indicates that now the sequence data of plants included in GenBank are far from covering most of the plants, especially wild plants that not have been fully explored by human. The KNApSAcK species-metabolite relation database is also far from complete with a large amount of data fragments. However, the plants with abundant metabolite-content data included in KNApSAcK database are frequently inconsistent with plants with complete sequence data included in GenBank. The metabolite-content data of plants in KNApSAcK could be seen as a necessary supplement of sequence data in GenBank for facilitating the analysis of evolutionary relations between plants and guiding the prediction of medicinal/edible plants since the plants covered by these two databases are complementary to each other. The plant samples selected in our research are performing both adequate sequence and metabolite-content data with acceptable data missing. Thus, we could investigate the effect of these two types of data in extracting medicinal/edible patterns from the same plant samples. We reconstructed the phylogenetic trees for the three sample groups by corresponding sequence data and metabolite-content data respectively (Figure 2).

The uses information of plants was collected from published literature and online sources, and annotated as seven categories: edible plants, medicinal plants, medicinal/edible multi-useful plants, landscaping plants, timber plants, poisonous plants and wild plants (Table 3).

We investigated the strength in phylogenetic signal of medicinal and edible categories for each phylogenetic tree we obtained using the D statistic (Table 4). We found that plants with medicinal/edible uses are significantly clustered in metabolite-content-based phylogenetic trees of all the three sample groups. The rbcL- and matK-based trees also show moderate phylogenetic signal for medicinal/edible plants but much weaker than that in metabolite-content-based trees. The ITS2-based tree shows weak phylogenetic signal for both medicinal and edible plants.

Generally, the edible plants are more phylogenetically clustered than medicinal plants in all the three sample groups for both of the two approaches, with lower D estimate values and higher P(D>0) values. This suggests that comparing with edible plants, the distribution of medicinal plants across the lineages reveals some but less phylogenetic relations. The mechanism of medicinal plants is much subtler than edible plants and is related to the expression of small secondary metabolites which are sometimes randomly distributed along the clades. Moreover, the expressions of the secondary metabolites with medicinal bio-activity are more closely related to the overall metabolite features, i.e., metabolite-contents of the plants. The plants with similar metabolite-contents tend to have similar medicinal features, and such observations are more obvious comparing with sequence-based approach in our experiments. Thus we might found more phylogenetic patterns by skipping gene data and comparing metabolite-content data directly. Considering the gene data available from GenBank is usually incomplete, the metabolite-content data implies great potential applications in predicting medicinal properties.

As a tentative approach to narrow down the number of medicinal/edible plants selected for bioprospecting, we also identified the hot nodes that are significantly overrepresented by species of medicinal/edible uses (Table 5). We can observe that phylogenetic clustering was found for edible and medicinal plants in all of the tested phylogenetic trees except ITS2 sequence-based tree. The hot nodes in metabolite-content based phylogenetic trees tend to encompass more medicinal and edible plants than sequence-based phylogenetic trees. This suggests that comparing with sequence-based approach it is more effective to explore phylogenetic patterns for medicinal and edible plants with the metabolite-content-based approach. We also compare the observed patterns for edible and medicinal plants with those for random samples of the same size drawn from the phylogenies. For these hot nodes in each of the tested phylogenetic trees, we recorded the percentage of edible and medicinal plants included in them. We compared the observed number of medicinal/edible plants encompassed in the hot nodes to the one expected to be found randomly in the percentage of the plants encompassed in the hot nodes, and this was the gain in percentage of medicinal/edible hits compared with random.

The phylogenetic distribution of medicinal and edible plants encompassed by hot nodes also shows that the edible plants perform more converge trends and gains in percentage of hits. This indicates that the edible features of plants are more closely associated with the phylogeny as well as the metabolite-content similarity, and also suggests that there may be many unexplored medicinal properties within the plant kingdoms. Moreover, we also investigated the coincidence rates of the medicinal/edible plants encompassed by hot nodes between the sequence-based and metabolite-content-based phylogenetic trees. We found that there is not significantly coincidence of medicinal plants encompassed by hot nodes of these two types of phylogenetic trees. In other words; the medicinal patterns identified by metabolite-content-based approach shows no significant similarity to the medicinal patterns identified by sequence-based approach. Our findings thus indicate that the metabolite-content-based approach might highlighted different group of medicinal plants with sequence-based approach, and might reflect more unexplored medicinal potential not associated with the sequence-similarity.

As a meaningful attempt, we imported more plant-metabolite relation data (28123 plant-metabolite relations associated with 1047 plants) and reconstructed phylogenetic tree by metabolite-content-similarity (Figure 3). We selected plants containing at least 14 metabolites to ensure data integrity. Plant uses information (edible or medicinal uses) was imported from KNApSAcK World Map DB. For the total 1047 tested plants, we found medicinal or edible uses information for 605 plants from World Map DB, with 543 plants having medicinal values, 345 plants having edible values. There are totally 303 plants with both medicinal and edible values. The remaining 442 plants which are lack of uses information are regarded as wild plants from which we may explore new medicinal properties. The hot nodes for medicinal plants encompass 288 plants, including 198 recorded medicinal plants. The remaining 90 wild plants encompassed by the hot nodes should be given priority for future screening for overall medicinal bioactivity because these plants perform highly metabolite-content-similarity with other 198 medicinal plants (Table 6).

   Conclusion

Many researchers have proved that edible and medicinal plants were derived mostly from some lineages, and tend to be clustered rather than scattered in the phylogenetic tree. Our study reveals that besides the sequence data, metabolite-content data is also closely associated with medicinal and edible bioactivity of plants and can explore the medicinal/edible patterns in a different perspective from DNA sequence-based plant phylogeny.

We found that comparing with DNA sequence-based approach, our metabolite-content-based approach performs fair even better predictive power of medicinal properties. Moreover, the hot nodes of metabolite-content-based approach highlight different medicinal/edible patterns comparing with DNA-sequence-based approach. This implies that metabolite-content-based approach could reflect unexplored medicinal/edible properties not recovered by the sequence-based approach.

Since sequence-based plant bioprospecting is frequently confined to the lack of DNA sequence data, it is rational to utilize metabolite-content data to extent the limitation of sequence-based bioprospecting. Metabolite-content-based plant phylogeny reconstruction could provide a new perspective in plant bioprospecting. With the improvement of metabolite-content database and the integration of various plant pharmacopoeia, such MC-guided bioprospecting approach can be further accelerated, and the predictive power for medicinal/edible plants will also be improved with the completeness of metabolite-content database in future.

Acknowledgements:

This work was supported by the National Bioscience Database Center in Japan; the Ministry of Education, Culture, Sports, Science, and Technology of Japan (16K07223 and 17K00406), Platform project for Supporting Drug Discovery and Life Science Research funded by Japan Agency for Medical Research and Development and NAIST Big Data Project.

Figure 1: Overview of 190 plants included in rbcL, matK and ITS2 sample groups.

Figure 3: MC-based phylogenetic tree for 1047 plants, with the hot nodes of medicinal/edible Plants.

Table 1: GenBank ID (rbcL, matK, ITS2) and use information of sample plants. Economic uses of plants are represented as following abbreviations: E (edible), M (medicinal), L (landscaping,), T (timber), P (poisonous), W (wild plant). Some plants are both medicinal and edible and are annotated as M/E. (*Partial sequence data).

Table 2: The amount of complete and partial sequences data of rbcL, matK and ITS2 sample groups.

Table 3: The amount of plants in each category of uses.

Table 4: Phylogenetic signal of medicinal/edible features in sequence-based and metabolite-content (MC) based trees.

Table 5: The number and proportion of medicinal/edible plants within the clades of hot nodes. Total plants included (%): The number (percentage) of the total plants included in the hot nodes of medicinal/edible uses. M/E Hits (%): The number (percentage) of the medicinal/edible plants included in the hot nodes of medicinal/edible uses. Gain in M/E hits: the percentage of gain in medicinal/edible plants included in hot nodes, compare with what would be expected by chance. Co-included plants (hits): the number of (medicinal/edible hits) plants included in the hot nodes of medicinal/edible uses for both of the sequence- and MC-based phylogenetic trees.

Table 6: The 90 plants with high priority for future screening for overall medicinal bioactivity.

1.       Dahanukar SA, Kulkarni RA, Rege N N (2000) Pharmacology of medicinal plants and natural products. Indian journal of pharmacology 32: S81-S118.

2.       Ciddi Veeresham (2012) Natural products derived from plants as a source of drugs 3: 200-201.

3.       Cseke LJ, Kirakosyan A, Kaufman PB, Warber S, Duke, JA, et al. (2016) Natural products from plants. CRC press.

4.       Newman DJ, Cragg GM, Snader KM (2000) The influence of natural products upon drug discovery. Natural product reports 17: 215-234.

5.       Cragg GM, Newman DJ (2013) Natural products: a continuing source of novel drug leads. Biochimica et Biophysica Acta (BBA)-General Subjects 1830: 3670-3695.

6.       DS Fabricant, NR Farnsworth (2001) The value of plants used in traditional medicine for drug discovery. Environmental health perspectives 109: 69-75.

7.       Ro DK, Paradise EM, Ouellet M, Fisher KJ, Newman KL, et al. (2006) Production of the antimalarial drug precursor artemisinic acid in engineered yeast. Nature 440: 940-943.

8.       Graham IA, Besser K, Blumer S, Branigan CA, Czechowski T, et al. (2010) The genetic map of Artemisia annua L. identifies loci affecting yield of the antimalarial drug artemisinin. science 327: 328-331.

9.       McChesney JD, Venkataraman SK, Henri JT (2007) Plant natural products: back to the future or into extinction?. Phytochemistry 68: 2015-2022.

10.    Soejarto DD, Fong HHS, Tan GT, Zhang HJ, Ma CY, et al. (2005) Ethnobotany/ethnopharmacology and mass bioprospecting: Issues on intellectual property and benefit-sharing. Journal of ethnopharmacology 100: 15-22.

11.    Gurib-Fakim A (2006) Medicinal plants: traditions of yesterday and drugs of tomorrow. Molecular aspects of Medicine 27: 1-93.

12.    Brummitt NA, Bachman SP (2010) Plants under pressure-a global assessment: the first report of the IUCN sampled red list index for plants. Kew, UK: Royal Botanic Gardens.

13.    Rønsted N, Symonds MR, Birkholm T, Christensen SB, Meerow AW, et al. (2012) Can phylogeny predict chemical diversity and potential medicinal activity of plants? A case study of Amaryllidaceae. BMC evolutionary biology 12: 182.

14.    Ernst M, Saslis-Lagoudakis CH, Grace OM, Nilsson N, Simonsen HT, et al. (2016) Evolutionary prediction of medicinal properties in the genus Euphorbia L. Scientific reports 6: 30531.

15.    Rønsted N, Savolainen V, Mølgaard P, Jäger AK (2008) Phylogenetic selection of Narcissus species for drug discovery. Biochemical Systematics and Ecology 36: 417-422.

16.    Saslis-Lagoudakis CH, Klitgaard BB, Forest F, Francis L, Savolainen V, et al. (2011) the use of phylogeny to interpret cross-cultural patterns in plant use and guide medicinal plant discovery: an example from Pterocarpus (Leguminosae). PloS one 6: e22275.

17.    Saslis-Lagoudakis CH, Savolainen V, Williamson EM, Forest F, Wagstaff SJ, et al. (2012) Phylogenies reveal predictive power of traditional medicine in bioprospecting. Proceedings of the National Academy of Sciences 109: 15835-15840.

18.    Yessoufou K, Daru BH, Muasya AM (2015) Phylogenetic exploration of commonly used medicinal plants in South Africa. Molecular ecology resources 15: 405-413.

19.    Kelly S, Grenyer R, Scotland RW (2014) Phylogenetic trees do not reliably predict feature diversity. Diversity and Distributions 20: 600-612.

20.    Clemente JC, Satou K, Valiente G (2007) Phylogenetic reconstruction from non-genomic data. Bioinformatics 23: e110-e115.

21.    Borenstein E, Kupiec M, Feldman MW, Ruppin E (2008) Large-scale reconstruction and phylogenetic analysis of metabolic environments. Proceedings of the National Academy of Sciences 105: 14482-14487.

22.    Mano A, Tuller T, Béjà O, Pinter RY (2010) Comparative classification of species and the study of pathway evolution based on the alignment of metabolic pathways. BMC bioinformatics 11: S38.

23.    Chang CW, Lyu PC, Arita M (2011) Reconstructing phylogeny from metabolic substrate-product relationships. BMC bioinformatics 12: S27.

24.    Ma CY, Lin SH, Lee CC, Tang CY, Berger B, et al. (2013) Reconstruction of phyletic trees by global alignment of multiple metabolic networks. BMC bioinformatics 14: S12.

25.    Abdullah AA, Altaf-Ul-Amin M, Ono N, Sato T, Sugiura T, et al. (2015) Development and mining of a volatile organic compound database. BioMed research international13: 1-13.

26.    Wink M (2003) Evolution of secondary metabolites from an ecological and molecular phylogenetic perspective. Phytochemistry 64: 3-19.

27.    Singh R (2016) Chemotaxonomy: a tool for plant classification. Journal of Medicinal Plants 4: 90-93.

28.    Liu K, Abdullah AA, Huang M, Nishioka T, Altaf-Ul-Amin M, et al. (2017) Novel Approach to Classify Plants Based on Metabolite-Content Similarity. BioMed research international.

29.    Shinbo Y, Nakamura Y, Altaf-Ul-Amin M, Asahi H, Kurokawa K, et al. (2006) KNApSAcK: a comprehensive species-metabolite relationship database. Plant metabolomics 165-181.

30.    Afendi FM, Okada T, Yamazaki M, Hirai-Morita A, Nakamura Y, et al. (2012) KNApSAcK family databases: integrated metabolite-plant species databases for multifaceted plant research. Plant and Cell Physiology 53: e1.

31.    Ikeda S, Abe T, Nakamura Y, Kibinge N, Hirai Morita A, et al. (2013) Systematization of the protein sequence diversity in enzymes related to secondary metabolic pathways in plants, in the context of big data biology inspired by the KNApSAcK Motorcycle database. Plant and Cell Physiology 54: 711-727.

32.    Nakamura Y, Mochamad Afendi F, Kawsar Parvin A, Ono N, Tanaka K, et al. (2014) KNApSAcK metabolite activity database for retrieving the relationships between metabolites and biological activities. Plant and Cell Physiology 55: e7.

33.    Cao Y, Charisi A, Cheng LC, Jiang T, Girke T (2008) ChemmineR: a compound mining framework for R. Bioinformatics 24: 1733-1734.

34.    Benson DA, Cavanaugh M, Clark K, et al. (2013) GenBank. Nucleic acids research 41: D36-D42.

35.    Larkin MA, Blackshields G, Brown NP, Chenna R, McGettigan PA, et al. (2007) Clustal W and Clustal X version 2.0. bioinformatics 23: 2947-2948.

36.    Huelsenbeck JP, Ronquist F (2001) MRBAYES: Bayesian inference of phylogenetic trees. Bioinformatics 17: 754-755.

37.    Godden JW, Xue L, Bajorath J (2000) Combinatorial preferences affect molecular similarity/diversity calculations using binary fingerprints and Tanimoto coefficients. Journal of Chemical Information and Computer Sciences 40: 163-166.