1 Introduction

Herbs used in traditional medicine are known to contain potent active principles and may represent a valuable alternative notably for the treatment of chronic disease. It is however also known that it is quite common that the compositions of medicinal plants could notably change along with varieties, seasons, habitats, species misidentifications, etc. (Kaiser et al. 2016; Pan et al. 2016; Soro et al. 2016). The existence of different chemotypes could lead to differences in efficacy, or even safety of derived drugs (Bhatia et al. 2013). Because of the inherent complexity of plant composition, metabolomics is well suited for assessing globally changes in metabolites that can be of importance in assessing quality of herbs (Yang et al. 2006; Ye et al. 2007; Bhatia et al. 2013).

Bitter melon (Momordica charantia Linn.), also known as bitter gourd, is a world widely cultivated edible and medicinal vine from the Cucurbitaceae family. It has been used as a botanical dietary supplement in middle and far east Asia for treatment of diabetes (Leung et al. 2009), obesity (Popovich et al. 2011; Bao et al. 2013; Xu et al. 2014) and a number of reported diseases (Singh et al. 1998a, b; Sathishsekar and Subramanian 2005). In depth investigation of this plant have revealed over 70 triterpenoid saponins whose structures are categorized in two types of triterpene skeletons, known as cucurbitane-type triterpene glycosides (CTGs) and oleanane-type triterpene glycosides (OTGs) (Tomita et al. 2005; Nakamura et al. 2006; Tan et al. 2008). In previous bioactivity studies, CTGs, an exclusive type of tetracyclic triterpenoids in Cucurbitaceae family, have demonstrated anti-diabetic and hypoglycaemic effects (Harinantenaina et al. 2006; Keller et al. 2011). Our team has proved that CTGs play a critical role in bitter melon’s anti-diabetic activity through stimulation of GLUT4 translocation and activation of AMPK signaling pathway. CTGs further enhance fatty acid oxidation and glucose disposal in mice (Tan et al. 2008; Iseli et al. 2013). Meanwhile, Oleanolic acid, one of the OTGs aglycones, is widely distributed in plants and its anti-diabetic activities have been documented (Ayeleso et al. 2017). OTGs from other species than bitter melon are known to exhibit hypoglycemic properties as well (Duan et al. 2016; Luyen et al. 2018). Moreover, we demonstrated that oleanolic acid showed anti-hyperglycemia effect on type 2 diabetic mice (Zeng et al. 2012). Those discoveries confirmed ancient records and recent pharmacology results of bitter melon’s efficacy at the molecular level.

Our previous phytochemical investigations also demonstrated the complex composition of such plants. Bioactivity variations and changes in concentrations of the bioactive metabolites along with the varieties or habitats were noticed in sample to sample comparison. In order to obtain an overview of the metabolome variations that are likely to occur, and to lay the chemical foundation in view of a future possible evidenced based therapeutic usage of bitter melon, 55 bitter melon varieties were collected in different parts of China and all samples were submitted to a MS-based metabolomics investigation.

Our strategy consisted in the profiling of all extracts with an optimized UHPLC-HRMS for discrimination by PCA and HCA followed by pooling of samples for extensive metabolite annotation by molecular networking (MN) (Pan et al. 2016; Wang et al. 2016; Mohimani et al. 2017; Wolfender et al. 2019b).

2 Materials and methods

2.1 Chemicals and reagents

Reference compounds goyasaponins I, II, momordicosides A, B, S and T from bitter melon were isolated and identified in Prof. Yang Ye’s lab at Shanghai Institute of Materia Medica, Chinese Academy of Sciences (Tan et al. 2008).

2.2 Plant material, extraction and sample preparation

55 bitter melon seed samples were purchased from different local seed companies in China (Table S1). 10 g of each dried sample were used for extraction. The remaining material was kept as voucher specimen samples deposited in the Herbarium at the Shanghai Institute of Materia Medica, Chinese Academy of Sciences. The seeds were grounded and extracted with 70% ethanol (1:10 w/v) using ultra-sonication for 45 min. Process was repeated three times. Each solution was centrifuged at 4000 rpm for 10 min. The supernatant was filtered and evaporated under reduced pressure to remove solvents and yield dried extract. Ten mg of extract were dissolved in 85% methanol–water as a 1 mg/ml solution. Filtration on a 0.22 μm membrane was applied before UHPLC analysis. In addition, a pool of all extracts solutions was prepared and used as pooled sample mixture.

2.3 UHPLC-HRMS profiling

All extract samples were subjected in a random sequence for metabolite profiling on a Waters Acquity UPLC system coupled with a Waters/Micromass LCT Premier time of flight (TOF) mass spectrometer. A mixture of reference saponins was inserted into the sequence every 10 injections as reference sample. Goyasaponins I (labeled as A9) and II (A11), as well as momordicosides A (B5), B (B9), S (B2) and T (B3) were concomitantly analyzed. The structures of the reference compounds see Fig. 1.

Fig. 1
figure 1

Chemical structures of reference compounds isolated from M. charantia extracts

Full size image

The injection volume was 2 μl. Temperature of sample chamber was kept at 10 °C. Water (UPLC grade, Waters) with 0.1% formic acid and acetonitrile (UPLC grade, Waters) with 0.1% formic acid were used for phase A and B, respectively. An Acquity UPLC BEH RP18 column (1.7 µm, 100 × 2.1 mm; Waters) was used at a column temperature of 45 °C. The chromatographic method started from an isocratic elution with 20% B for 0.5 min, continued with a linear gradient from 23 to 43% B in 7.0 min then increased to 95% B in another 3 min. A washing procedure was followed using 95% B for 2 min. The column was then equilibrated using 20% B for 2.5 min before next injection. The flow rate was 0.8 ml/min. ESI negative full scan mode was used for its high ionization efficiency, the data were acquired from 200 to 2000 m/z with the following parameters: capillary voltage, 2.4 kV; sample cone voltage, 40.0 V; source temperature, 120.0 °C; desolvation gas temperature (N2), 300.0 °C; desolvation gas flow, 700 unit; aperture 1 voltage, 35.0; resolution, 7500. A real-time calibration was performed using 20 ng/ml leucine-enkephalin solution in 50/50 acetonitrile–water as lock spray (lock mass m/z 554.2615).

2.4 UHPLC-HRMS-HRMS/MS profiling

A pooled sample of all extracts and reference compounds were analyzed on a Thermo Dionex Ultimate 3000 UHPLC interfaced to a Thermo Q-Exactive Plus mass spectrometer with a heated electrospray ionization (HESI) source. The LC method used for detailed metabolite profiling with higher resolution was as follows: column, Acquity UPLC BEH RP18 column (1.7 µm, 150 × 2.1 mm; Waters), column temperature, 45 °C, mobile phase, water (A) and acetonitrile (B) both with 0.1% formic acid, flow rate 0.5 ml/min, injection volume, 2 μl. The gradient started from a 20% isocratic elution for 1.5 min then followed by a linear gradient of 20% to 43% B until 18 min, and continued increasing to 95% B in 7 min. The column was washed using 95% B for 5 min and equilibrated 10 min before next injection. The analyses were performed in negative mode with following parameters: source voltage, 2.5 kV; capillary temperature, 256.25 °C; sheath gas flow (N2), 47.50 units; auxiliary gas flow, 11.25; spare gas flow, 2.25; probe heater temperature, 412.50 °C; S-Lens RF Level, 50. Full MS resolution, 70,000; scan range, 300–2000 m/z. The data-dependent MS/MS experiments were performed on the four most intense ions and the resolution was 17,500. The stepped normalized collision energy (NCE) was set as 15–30–45 units, and the isolation window was 2.0 m/z. A dynamic exclusion was applied on parent ions for 2.0 s after being fragmented.

2.5 Molecular networking

Molecular networks were created by global natural products social (GNPS, http://gnps.ucsd.edu) online workflow (Wang et al. 2016) with the following parameters: precursor ion mass tolerance, 1.0 Da; MS/MS fragment ion mass tolerance, 0.8 Da. Filters were applied to remove all edges which cosine score was below 0.7 and match fragment peaks < 6. The maximum number of neighbor nodes for one single node was set to 10. The final results were visualized both online at GNPS and using Cytoscape 3.4.0. All self-loops were removed at the data import step.

2.6 Multivariate statistical analysis

The UHPLC-TOF/MS profiling dataset was processed using the open-source software MZmine 2 to extract and align peaks (Pluskal et al. 2010). The peak detection was performed using 2D Grid Mass algorithm using following parameters: minimum height 100, m/z tolerance 0.06, min–max width time 0.03–0.4. Isotopic peaks grouping was using m/z tolerance 0.05, max charge 2 and representative isotope with lowest m/z. RANSAC aligner was applied using 50% min number of points and threshold 0.05. Duplicate peak filter and peak finder gap filling were applied then and a peak list of 511 × 55 matrix was obtained.

The matrix was normalized and then imported to SIMCA 13.0 (Umetrics, Sweden) and pretreated with mean-centered and Pareto variance scaling. A PCA model was generated with the data matrix. A Hierarchical clustering analysis (HCA) was performed using PCA algorithm but with > 95% (R2X = 0.952, 20 PCs) information of original feature of samples.

Potential markers were selected using a PCA “group to group comparison” scores contribution plot. A threshold score > 1.2 was applied to select variables that contributed to the differences between the two groups. Isotopes and adducts were removed manually from the final marker list by checking UHPLC-MS raw data.

Heat-maps were created using MeV 4.9.0. The data set was processed following normalization, divided by SD and mean-centered. Hierarchical clustering analysis was performed based on Euclidean distance.

2.7 Quantitation of four marker compounds using UHPLC-MS/MS

Four compounds, goyasaponin I (A9), goyasaponin II (A11), momordicoside S (B2) and momordicoside A (B5), were quantitated against an external standard calibration. All stock solutions of reference compounds and extract were prepared in 85% MeOH (HPLC grade, Thermo) at a concentration of 1 mg/ml.

Quantitation was performed using an Agilent 1290 infinity LC-system interfaced to an Agilent 6460 triple quadrupole mass spectrometer equipped with ESI source. Two μl of each sample were injected in UHPLC under following conditions: column, Acquity UPLC BEH RP18 column (1.7 µm, 100 × 2.1 mm; Waters); column temperature, 40 °C; mobile phase, water (A) and acetonitrile (B) both with 0.1% formic acid, flow rate 0.5 ml/min. The elution was performed as a linear gradient from 30 to 50% B in 5 min, and followed by a 2 min washing procedure with 95% B. Then the column was equilibrated 3 min before next injection. Quantitation was performed in multiple reaction monitoring (MRM) in ESI negative ion mode using following parameters: capillary voltage, 3.5 kV; gas flow, 10 l/min; gas temperature, 320 °C; nebulizer, 50 psi; sheath gas temperature, 350 °C; sheath gas flow, 11 l/min; collision gas, helium. The MRM method was optimized by injection of a standard compound solution, as shown in Table S2. The dwell time for each monitored transition was 120 ms. Each sample was injected as three analytical replicates. All accuracies and precisions were within the range of ± 15%, which is in accordance with the bioanalytical method validation guidelines of the US Food and Drug Administration (FDA 2013) and recovery of spiked compounds were above 95%.

Mann–Whitney tests were performed using followed by Dunnett’s multiple comparisons test was performed using GraphPad Prism version 6.0c, USA.

3 Results

Fifty-five purchased bitter melon seed samples (Table S1) were grinded and extracted with 70% ethanol–water. MVA specimen differentiation metabolite profiling of these crude extracts was performed by full scan UHPLC-HRMS on a TOF/MS instrument with an optimized reversed phase gradient. In parallel, the pool of all samples was profiled on an Orbitrap with a data-dependent acquisition using a longer LC method with similar gradient to acquire MS/MS spectra automatically on all main triterpenoids for annotation purposes. For improved marker identification two OTGs, goyasaponins I and II, as well as four CTGs, momordicosides A, B, S and T, which were previously isolated from bitter melon, were concomitantly analyzed. Database searching and MS/MS spectra analyses were performed as well for marker compounds dereplication and identification.

3.1 Differentiation of bitter melon extracts based on UHPLC-HRMS profiling

Each bitter melon seed sample was profiled by UHPLC-HRMS to investigate trends in composition modifications across varieties and locations. All samples were preprocessed by MZmine and a matrix of 55 × 511 features was generated and normalized. To view the differences across this complex dataset, the data were visualized using a heat-map of the HCA results that provided an overview of all observed metabolites (Fig. 2a). As shown in the heat map, clear chemical constituent’s composition differences between different bitter melon samples were observed, which may suggest differences in bioactivities.

Fig. 2
figure 2

Multivariate statistical analysis of UHPLC-HRMS profiling data acquired from M. charantia extracts. a Heat-map of M. charantia extracts corresponding to relative peak areas. Columns represent 55 varieties and rows represent 511 detected metabolites. b PCA score plot derived from profiling data of all samples, c HCA dendrogram, Group A colored in green and Group B in blue were cut off at first branch, d score plot of PCA marked CA and CB. e PCA “group to group comparison” contribution scores plot of CB-CA, weight = p1p2. Threshold score > |1.2| were used to select marker compounds (colored in red)

Full size image

Differentiation of the samples was performed by PCA and HCA after mean centering and pareto scaling. The PCA gave a model that represented 73.1% (PC1 + PC2 + PC3) of original dataset information. The parameters of this model (R2X = 0.731 and Q2 = 0.529) validated the quality of this PCA. The PCA score plot (Fig. 2b) provided a good visualization of the sample set relationship and further confirmed that the PCA model was reliable for representing bitter melon accessions.

To investigate the differentiation among bitter melon samples, the HCA was analyzed in more details. The HCA dendrogram (Fig. 2c) revealed two main branches corresponding to two groups of bitter melon samples with distinct chemical compositions. Based on this, samples were labeled with different colors to show the groups identified in the PCA (Fig. 2d). These combined PCA/HCA results provided a good discrimination of all samples and clearly evidenced the presence of two chemotypes of bitter melon varieties. Hereinafter, the samples colored in green in Fig. 2d are referred as chemotype A (CA) while the blue ones are described as chemotype B (CB). According to PCA/HCA, 24 bitter melon samples were assigned as CA while 31 were assigned as CB (see Table S1). The discovery of two chemotypes of bitter melon could lead finally in different biological activities.

3.2 Selection of marker compounds

To understand which specific metabolites cause the differences, the PCA contribution plot of scores was analyzed to identify the features significantly contributing to the chemotype differentiation.

In the contribution plot (Fig. 2e) a contribution score above 1.2 was chosen as a threshold and after filtering extra adducts and isotopes, a marker compound list was obtained as shown in Table 1. Compounds A1 to A13 were found to be potential markers of CA and B1 to B15 as potential markers of CB. These compounds represent the characteristics of two types of bitter melon, and identification of these marker compounds allows us to document at the molecular level the trends in composition that clearly differentiate those bitter melon chemotypes.

Table 1 Selected marker compounds of two bitter melon chemotypes
Full size table

3.3 Marker compounds annotation via molecular networking

By comparing UHPLC-TOF/MS retention time and HRMS1 data with available reference compounds from previous studies, (Okabe et al. 1980; Murakami et al. 2001; Tan et al. 2008) six marker compounds were unambiguously identified (LEVEL1) (Table 1). This included two OTGs (A9 goyasaponin I and A11 goyasaponin II) for CA and four CTGs (B2 momordicoside S, B3 momordicoside T, B5 momordicoside A and B9 momordicoside B) for CB. However, this was not sufficient for a comprehensive survey of all compounds responsible for the differences between bitter melon chemotypes. In order to annotate the largest possible number of marker compounds, MS/MS spectra of the pool of all samples and a mixture of references were recorded by Orbitrap with a similar UHPLC gradient and processed by molecular networking (MN) analysis (Wang et al. 2016).

A MN composed by 1405 nodes was generated from our dataset which included 24′724 MS/MS spectra (Fig. S1). In the first step of dereplication, six nodes were unambiguously annotated from the whole network by comparing the spectral data (retention time, MS and MS/MS) of the six reference compounds, goyasaponin I (A9), goyasaponin II (A11), momordicoside S (B2), momordicoside T (B3), momordicoside A (B5) and momordicoside B (B9). As highlighted in Fig. S1, this allowed the assignment of one cluster to OTG- and another to CTG-type triterpenoids. This was used to propagate the annotation to other triterpenoid derivatives in both clusters. The results indicated that these two clusters (referred as clusters A and B in the following text) indeed corresponded to OTGs or CTGs, respectively (Fig. 3).

Fig. 3
figure 3

a Clusters A corresponding to OTGs and b clusters B corresponding to CTGs in the molecular network of M. charantia extract. Dereplication was made using precursor ion mass against DNP

Full size image

In the second step, dereplication of all metabolites related to nodes connected in these two clusters was carried out against the dictionary of natural products (DNP) (Hall 2017) based on UHPLC-HRMS1 data restricted to compounds isolated from Curcurbitaceae family. For this a peak list was constructed after peak picking by retrieving chromatographic data regarding to parent ion masses and retention times of each nodes. The masses were compared with calculated exact masses of compounds in DNP with mass tolerance of ± 3 ppm. As a result, 15 nodes were characterized as OTGs within cluster A in addition to the references, including nine marker compounds of CA (Fig. 3a). Similarly, in cluster B, 25 nodes in total were annotated as CTGs containing 14 markers of CB (Fig. 3b). Most of the corresponding features appeared as [M − H]− ions while some of them also occurred as [M − HCOO]− (Table 1). The results of the DNP search confirm the classification by MN.

Combination of these results enabled the annotation of 6 triterpene glycosides at level 1, 36 at level 2 and 3 in cluster A and B (Sumner et al. 2007). This also revealed the presence of still unreported OTGs and CTGs according the molecular ions features detected (around 37% of all nodes detected on the two cluster assignments made). Most of the marker compounds of CA and CB were found and annotated in the MN except A4, A6, A8, A13 and B10 (Fig. 3). As observed in the MN, markers of CA were mainly found in cluster A, while almost all the markers of CB were included in cluster B. These results clearly indicate that bitter melon sample belonging to chemotype A represent OTGs rich varieties and those belonging to CB are CTGs rich. This information should provide the basis for an appropriate selection of bitter melon varieties based on their OTGs versus CTGs composition in future, once a clear involvement of the different constituents in the overall antidiabetic effects will be established.

The MN of bitter melon can be used in LC–MS guided isolation to find derivatives of bioactive compounds such as CTGs B2 and B3 (Tan et al. 2008; Iseli et al. 2013) in future studies. This analysis also showed that MN is capable of identifying a significant number of previously unknown triterpenoids analogues from M. charantia compared to those previously isolated from this species. Overall, MN navigation combined with LC–MS guided isolation will permit to efficiently obtain unreported OTGs or CTGS derivatives and thus expand our bitter melon compound library, for documenting their bioactivities.

3.4 Identification of new marker compounds in two bitter melon chemotypes

Markers not annotated in MN were investigated for further dereplication by searching their molecular formula against the DNP, A4, A6 and A8 were putatively identified as OTGs. A13 and B10 remained unknown as no dereplication hits was found in DNP. Moreover, A12 and B1 were annotated as OTG and CTG by MN but their molecular formula was still unreported. This indicated that A12 and B1 are probably new natural products (Table 1). Their structures were further investigated by interpretation of their HRMS-MS/MS spectra.

The MS/MS spectrum of A12 was compared with that of A11, which identified unambiguously as goyasaponin II (m/z 1509.7, cal. C70H109O35 [M − H]−). A11 showed major fragment ions m/z 807.4, 627.4, 583.4 and 469.3 in MS/MS spectra, which indicated its characteristic fragmentation pathway started with a neutral loss of saccharide moiety at C-28 (Fig. S2a) forming goyasaponin III (Shibuya et al. 1990), another OTG reported from M. charantia. The fragmentation continued with the loss of one or two glycosyl groups at C-3 acquired m/z 627.4 or 469.3 as described in Fig. S2a. Comparison of MS/MS spectra of A12 and A11 revealed that they had high similarity, except the [M − H]− mass m/z 1479. 6 and a high abundance fragment peak m/z 777.4 in A12 that were 30 Da less compared to A11 (1509.7 and 807.4). This suggested that they share a similar structure except for the glycoside moiety attached to the glucuronic acid group which is a pentose instead of a galactose in A12. The elemental compositions of all fragments were calculated based on exact masses with mass tolerance of ± 2 ppm. Thus, the most probable structure of compound A12 was proposed as gypsogenin-3-O-[pentosyl-glucuronopyranoside],28-O-[xylopyranosyl-xylopyranosyl-rhamnopyranosyl-(rhamnopyranosyl)-fucopyranosyl] ester (Fig. S2).

The MS/MS spectrum of B1 was highly similar to that of B3, momordicoside T (Fig. S3). All major peaks in the spectrum of B3 including the deprotonated molecular ion could be found in that of B1, which revealed that B1 shared the same moiety of B3 after a characteristic glucosyl neutral loss of 162 Da (C6H10O5). B1 was thus and analogue of B3 with an additional hexose motley as shown in Fig. S3b. The elemental compositions of all fragments were calculated with mass tolerance within ± 2 ppm. Thus, two new compounds A12 and B1 were identified. Their structure types revealed by MS/MS analysis also verified that the chemotype differentiation of bitter melon and compounds classification model are reliable.

As a summary 12 out of 13 markers of CA were identified as OTGs while 13 out of 14 markers of CB were assigned to CTGs. This suggested that the two chemotypes of bitter melon mainly differ in in their richness either in cucurbitane or oleanane triterpenoids. It is particularly worth to mention that previous bioactivity studies (Harinantenaina et al. 2006; Tan et al. 2008; Keller et al. 2011; Iseli et al. 2013) indicated that CTGs are most probably involved in the anti-diabetic activity of bitter melon. The existence of two well distinct chemotypes revealed by this study may influence the pharmacological effect of the extract and such results need to be taken account into consideration for further quality control studies for the usage of bitter melon in diabetes management.

3.5 Content comparison of marker compounds between chemotypes A and B

To verify contribution of the marker compounds related to the differentiation, a heat-map with HCA in both rows and columns was generated based on UHPLC-TOF/MS peak area data (Fig. 4). As shown, the contents of markers in samples obviously matched with the chemotypes discrimination in Table S1, all CA varieties are OTGs rich while CB varieties are CTGs rich. As an example, “E05 Feicuilvzao” is a CA bitter melon and it has higher content of OTG markers than “E30 Xiangbaiyu” which is assigned as CB and is richer in CTG markers.

Fig. 4
figure 4

Heat-map of selected marker compounds in bitter melon extracts. Columns represent 55 varieties and rows represent 28 marker compounds of two chemotypes. Colors correspond to profiling peak areas

Full size image

To quantitatively inspect the differences between those bitter melon chemotypes, a quantitation of the four of the most abundant marker compounds was performed. Two OTGs A9 and A11, as well as two CTGs—B2 and B5 were chosen, and their concentrations in 55 bitter melon extracts were quantified through UHPLC-MS/MS in Multiple Reaction Monitoring (MRM) mode. The calibration curve range and equation of each compound was: 0.1, 0.2, 0.5, 1.0 and 2.0 μg/ml for A9 (y = 0.5456x + 13.8353); 0.2, 0.5, 1.0, 2.0 and 5.0 μg/ml for A11 (y = 0.2356x + 37.9698); 0.05, 0.1, 0.5, 1.0 and 5.0 μg/ml for B2 (y = 0.6543x − 3.9605); 0.05, 0.1, 0.5, 1.0 and 5.0 μg/ml for B5 (y = 1.2231x − 6.1481). The concentrations of compounds in each sample are shown in Table S1. Mann–Whitney tests were performed on the quantitation results. Figure 5 showed that the concentrations of A9 (p = 0.0423) and A11 (p = 0.0004) in CA were significantly higher than that in CB. In contrast, CTGs B2 and B5 were significantly more abundant in CB (p < 0.0001 for both compounds). Since B5 showed low p value and the most significant inter-group difference, B5 was selected as a good marker to distinguish those two chemotypes. Both heat-map and quantitation results proved that the selected marker compounds contribute to chemotype discrimination. Quantitation of B5 provided an efficient way to preliminary assign the bitter melon chemotypes.

Fig. 5
figure 5

Marker compounds’ concentrations in CA and CB dry samples: A9 goyasaponin I, A11 goyasaponin II, B2 momordicoside S and B5 momordicoside A. *P ≤ 0.05, **P ≤ 0.01

Full size image

Furthermore, by analyzing bitter melon samples across various regions in China, PCA score plot showed that there is no consistent correlation between geographical distributions and chemotype distributions. For example, bitter melons collected from Shandong, Hunan and Guangdong provinces belong to one of the chemotypes respectively. In many provinces however, the varieties from a given location do not belong to the same chemotypes, and significant chemical component differences between different provinces were observed (Fig. S4). The differences among these locations indicate that it is still difficult to correlate bitter melon chemotype in habitats and morphological traits. Therefore, identification of marker compounds in bitter melon varieties would indeed be necessary for chemotype assignment.

4 Discussion

Bitter melon is a traditional Chinese medicine recorded in ancient documents. Nowadays it has also been studied as an anti-diabetic traditional medicine in clinical practice. Many previous studies have reported the anti-diabetic and hyperglycemic effects of CTGs. Our team has also previously conducted thorough chemical investigations, isolated various compounds and characterized their biological properties in vitro and in vivo assays. Several CTGs, represented by B2 and B3, showed promoting GLUT4 translocation effect associated with activation of AMPK, a key pathway leading to upregulation of fatty acid oxidation and glucose disposal. Those findings demonstrated that CTGs might played a therapeutic role in combating diabetes and obesity. In addition, OTGs from other plants have been reported to exhibit hyperglycemia effects (Duan et al. 2016; Luyen et al. 2018). To our knowledge no anti-diabetic activities for OTGs in bitter melon have been reported. We have however demonstrated that oleanolic acid, which is one of the aglycones of OTGs widely distributed in plants, showed anti-hyperglycemia effect on type 2 diabetic mice (Zeng et al. 2012). These studies indicated that through different mechanisms, both CTGs and OTGs may be involved at the molecular level in the effects of bitter melon. This also proposed the necessity of differentiation between CTGs and OTGs rich bitter melon varieties. In the present study, we successfully established a reliable and practical method of identifying bitter melon varieties with different chemotypes.

One of the issues related to the characterization of bitter melon bioactivity is their composition. This vegetable has been artificially cultivated all over the world, with nearly 100 varieties in China alone. To our knowledge our study establishes for the first time the presence of two chemotypes of bitter melon. At present however, no study explored the relation between the varieties and the impact on the possible health benefit of bitter melon. Based on our results, the impact of chemical differences between CA and CB on bitter melon’s bioactivities still requires further investigations. It is of great significance to reveal the relationship between CTGs/OTGs content and the activities of bitter melon extracts, and our research has laid the chemical foundation for future exploration of different therapeutic potential in bitter melon variations.

The PCA score plot also showed the chemotype distribution of bitter melon varieties in different regions of China (Fig. S4). Some geographical distribution trends could be found. For example, the varieties from Shandong, Hunan and Guangdong provinces are belong to the same chemotype, respectively. However, it is worth mentioning that in many cases the chemotype cannot be linked to a given province. These might be caused by the long-term artificial planting and cultivation that have randomized the distribution of various varieties, and it makes the relationship between bitter melon varieties and their activities more complicated and random. Thus, having marker to characterize chemotype will be key in future. The result also illustrated the importance of metabolomics to tackle such issue.

To the best of our knowledge this is the first comprehensive metabolomics study on bitter melon. The integration UHPLC-HRMS1 and MN provided a good strategy for highlighting composition traits across varieties and for the annotation of large numbers of triterpenoids. The MN that was established expanded the pool of triterpenoids in bitter melon, and it can be also used for tracing more bioactive derivatives from bitter melon extracts. Further study on mass targeted isolation of CTGs based on MN exploration will be carried to identify new bioactive derivatives and we have recently established efficient high resolution preparative strategies for this (Queiroz et al. 2019; Wolfender et al. 2019a).

The multivariate models and sample set we built in this study not only have provided the differentiation of bitter melon varieties, but also can be useful on classify more bitter melon varieties by PCA model fitting. In addition, B5 is a known marker compound with the most significant difference of content between CA and CB. Quantification of B5 and could be related to the CA and CB sample sets and can thus provide a simplified method for the preliminary chemotype assignment of different bitter melon varieties.

Our results should help to select plant sources with defined chemical composition and thus reproducible efficacy/safety profile. Such comprehensive chemical characterization is valuable contribution of modern research to traditional Chinese medicine.

5 Conclusion

In this study, it was found for the first time that bitter melon on the market could be roughly divided into two chemotypes. The identification of marker compounds showed that one of the bitter melon chemotype is richer in CTGs, while the other contains mainly OTGs. The results were further validated by UHPLC-MRM quantitation. We have proved in previous studies that CTGs have good anti-diabetic effects. The discovery of these two defined chemotypes is important for the selection of bitter melon varieties rich in bioactive components for the development of diabetes therapeutics or dietary supplements. Further studies on the bioactivities of both CTGs and OTGs are required to investigate in depth in order to verify if the final antidiabetic effects can be linked to a given chemotype.