Metabolic fingerprinting of Leontopodium species (Asteraceae) by means of ¹H NMR and HPLC-ESI-MS.

The genus Leontopodium, mainly distributed in Central and Eastern Asia, consists of ca. 34-58 different species. The European Leontopodium alpinum, commonly known as Edelweiss, has a long tradition in folk medicine. Recent research has resulted in the identification of prior unknown secondary metabolites, some of them with interesting biological activities. Despite this, nearly nothing is known about the Asian species of the genus. In this study, we applied proton nuclear magnetic resonance (¹H NMR) spectroscopy and liquid chromatography-mass spectrometry (LC-MS) metabolic fingerprinting to reveal insights into the metabolic patterns of 11 different Leontopodium species, and to conclude on their taxonomic relationship. Principal component analysis (PCA) of ¹H NMR fingerprints revealed two species groups. Discriminators for these groups were identified as fatty acids and sucrose for group A, and ent-kaurenoic acid and derivatives thereof for group B. Five diterpenes together with one sesquiterpene were isolated from Leontopodium franchetii roots; the compounds were described for the first time for L. franchetii: ent-kaur-16-en-19-oic acid, methyl-15α-angeloyloxy-ent-kaur-16-en-19-oate, methyl-ent-kaur-16-en-19-oate, 8-acetoxymodhephene, 19-acetoxy-ent-kaur-16-ene, methyl-15β-angeloyloxy-16,17-epoxy-ent-kauran-19-oate. In addition, differences in the metabolic profile between collected and cultivated species could be observed using a partial least squares-discriminant analysis (PLS-DA). PCA of the LC-MS fingerprints revealed three groups. Discriminating signals were compared to literature data and identified as two bisabolane derivatives responsible for discrimination of group A and C, and one ent-kaurenoic acid derivative, discriminating group B. A taxonomic relationship between a previously unidentified species and L. franchetii and Leontopodium sinense could be determined by comparing NMR fingerprints. This finding supports recent molecular data. Furthermore, Leontopodium dedekensii and L. sinense, two closely related species in terms of morphology and DNA-fingerprints, could be distinguished clearly using ¹H NMR and LC-MS metabolic fingerprinting.

Introduction

The genus Leontopodium R.Br. ex Cassini comprises between 34 (Dickoré, unpublished), 41 (Handel-Mazzetti, 1927), and 58 (Wu et al., 1994) different species. The main distribution of the genus is in Central and Eastern Asia. The centre of diversity is south-western China, where 15 to 18 different species can be found. Two species also occur in Europe: The widespread Leontopodium alpinum Cass., and its endemic sister species, Leontopodium nivale (Ten.) Huet. ex Hand.-Mazz., which has a disjunct distribution in the Central Apennines in Italy and the Pirin Mountains in Bulgaria. For people living in the European Alps, especially L. alpinum, which is known as the Alpine Edelweiss, is a very important part of their cultural heritage.

The Alpine Edelweiss (L. alpinum Cass. or L. nivale subsp. alpinum (Cass.) Greuter) has a long tradition in folk medicine. References from the year 1582 already mentioned the use of Edelweiss for the treatment of diarrhoea and dysentery (Tabernaemontanus, 1582). Several other applications in traditional medicine for extracts and plant parts of Edelweiss were described throughout the years. Recent phytochemical research on L. alpinum has resulted in the detection of almost 50 different, partly uncommon secondary metabolites, including sesquiterpenes (Dobner et al., 2003a; Gray et al., 2000; Schwaiger et al., 2004; Stuppner et al., 2002), diterpenes (e.g., Schwaiger et al., 2004), lignanes (Dobner et al., 2003a; Schwaiger et al., 2004), benzofurans (e.g., Dobner et al., 2003a), and phenolic compounds, such as the novel described leontopodic acids (Schwaiger et al., 2005). Some of these compounds are highly bioactive, which was demonstrated in several different pharmacological models. Hence, antibacterial (Dobner et al., 2003b), antioxidative and DNA-protecting (Schwaiger et al., 2005), and anti-inflammatory (Schwaiger et al., 2004) properties were observed, as well as an enhancement of cholinergic transmission in the brain (Hornick et al., 2008) and an inhibition of intimal hyperplasia of venous bypass grafts (Reisinger et al., 2009). Despite these results, nearly nothing is known about bioactive compounds in other Leontopodium species. Until now, only a few phytochemical and pharmacological investigations have been conducted on these species (e.g., Leontopodium longifolium Ling: Li et al., 2006; Leontopodium leontopodioides Beauverd: Li et al., 1994; Leontopodium andersonii C.B. Clarke: Schwaiger et al., 2010; Leontopodium nanum Hand.-Mazz.: Wang et al., 2002), although many species were used in Traditional Asian Medicine, e.g., in Tibet (Kletter and Kriechbaum, 2001).

Whereas the metabolome is clearly defined as the ‘complete complement of small molecules present in an organism’ (Hall, 2006), there are different approaches to detect and investigate the metabolome. Throughout the years, various terms were defined, such as metabonomics, metabolomics, metabolic profiling and metabolic fingerprinting. A metabolic fingerprinting approach is defined as a ‘high-throughput qualitative screening of an organism or tissue with the primary aim of sample comparison and discrimination analysis’ (Hall, 2006).

Commonly used techniques for metabolic fingerprinting are LC–MS (liquid chromatography–mass spectrometry) and 1H NMR (proton nuclear magnetic resonance) spectroscopy. LC–MS as a fingerprinting technique was applied successfully in various fields of plant research, such as chemotaxonomy (Urbain et al., 2009), plant biochemistry (Kim and Park, 2009), food chemistry (Pongsuwan et al., 2008), and for the quality control of medicinal plants (e.g., Tianniam et al., 2008). The main advantage of mass spectrometry is its high sensitivity, which allows the detection of low molecular weight compounds at concentrations below the nanogram per millilitre range if optimal MS conditions can be provided (Seger and Sturm, 2007).

On the other hand, 1H NMR spectroscopy in combination with multivariate statistics has become a frequently used technique for metabolic fingerprinting. NMR spectroscopy has a long history in the qualitative and quantitative assessment of secondary plant metabolites (Holmes et al., 2006). 1H NMR spectroscopy is also commonly applied for quality control in food science and technology (e.g., Ali et al., 2009; Belton et al., 1998; Lachenmeier et al., 2005). NMR techniques are reproducible with rich structure information. The only essential requirement for compound detection in 1H NMR experiments is the availability of observable protons in a molecule, thus resulting in the applicability of 1H NMR to a wide range of plant metabolites. In this regard, 1H NMR enables the detection of constituents that could otherwise not be detected in LC–MS experiments, e.g., as in case of insufficient ionisation. Another major advantage compared to other analytical techniques is the matchless reproducibility. In contrast to NMR-based analysis, day to day variations are often a problem for LC–MS-based systems. Nevertheless, one of the great disadvantages of NMR spectroscopy is its relatively low sensitivity compared to modern mass spectrometry instrumentations (Holmes et al., 2006). Low concentration compounds may not be detectable with NMR. In addition, signal overlapping is often a problem if more than one compound is present in an NMR sample, e.g., when analysing plant extracts.

In our study, we used both 1H NMR spectroscopy and LC–MS in combination with multivariate statistics as an approach to detect metabolic fingerprints of species within the genus Leontopodium. We investigated roots of 11 different species, which were collected in the field, and the roots of 12 different cultivated species (Table 1). The main aim of this study was to reveal information about similarities and differences between the species of the genus Leontopodium by comparing their metabolic fingerprints, and to conclude on their relationship to each other.

Table 1

Population number, species names, sample origin, and voucher informationa for the investigated species.

Populations	Species	Sample origin	Voucher informationa
Collected plants (Province of Yunnan, south-western China)
SSG-06	L. andersonii C.B. Clarke	China, Yunnan, Lijiang, Yuhu Village	WU 0044003
SSG-14	L. andersonii C.B. Clarke	China, Yunnan, Zhongdian, Haba Village	WU 0043998
SSG-27	L. andersonii C.B. Clarke	China, Yunnan, Luquan, JiaoZiShan	WU 0043958
SSG-13A	L. artemisiifolium Beauverd	China, Yunnan, Zhongdian, Haba Village	WU 0043997
SSG-13B	L. cf. artemisiifolium Beauverd	China, Yunnan, Zhongdian, Haba Village	WU 0043997
SSG-26	L. caespitosum Diels	China, Yunnan, Luquan, JiaoZiShan	WU 0043960
SSG-11	L. calocephalum Beauverd	China, Yunnan, Zhongdian, Haba Village	WU 0043999
SSG-17	L. dedekensii Beauverd	China, Yunnan, Benzilan, Dongzhulin Monastery	WU 0043994
SSG-19	L. dedekensii Beauverd	China, Yunnan, Deqen, Atuntze	WU 0044004
SSG-22	L. dedekensii Beauverd	China, Yunnan, Weixi, Langping-Laching	WU 0043954
SSG-24	L. dedekensii Beauverd	China, Yunnan, Bingzhongluo	WU 0043956
SSG-09	L. franchetii Beauverd	China, Yunnan, Lijiang-Zhongdian	WU 0044016
SSG-15	L. franchetii Beauverd	China, Yunnan, Zhongdian, Cuo Bu La Ka Mt.	WU 0044008
SSG-18	L. himalayanum DC.	China, Yunnan, Zhongdian-Deqen	WU 0043993
SSG-04	L. sinense Hemsl. ex Forb. & Hemsl.	China, Yunnan, Dali, CangShan	WU 0043975
SSG-05	L. sinense Hemsl. ex Forb. & Hemsl.	China, Yunnan, Heqing, Da Shi Village	WU 0044001
SSG-08	L. souliei Beauverd	China, Yunnan, Lijiang-Zhongdian	WU 0044013
SSG-12	L. souliei Beauverd	China, Yunnan, Zhongdian, Haba Village	WU 0044000
SSG-20	L. cf. stracheyi C.B. Clarke ex Hemsl.	China, Yunnan, Weixi, La Ba Di Village	WU 0043995
SSG-21	L. cf. stracheyi C.B. Clarke ex Hemsl.	China, Yunnan, Weixi, Langping-Laching	WU 0043996
SSG-25	L. cf. stracheyi C.B. Clarke ex Hemsl.	China, Yunnan, GongShan	WU 0043957
SSG-16	L. sp.	China, Yunnan, Zhongdian	WU 0044007
Cultivated plants (Botanical Garden Giessen, Germany)
J22	L. alpinum Cass.	Austria, Styria, Rax Alpe	07/J/700
J01	L. calocephalum Beauverd	China, Yunnan, Da Xue Shan	03-678
J10	L. cf. calocephalum Beauverd	China, Sichuan, Min Shan	07-524
J03	L. dedekensii Beauverd	China, Yunnan, Hengduan Shan	03-302
J03A	L. dedekensii Beauverd	China, Yunnan, Hengduan Shan	03-302
J07	L. dedekensii Beauverd	China, Xizang, E-Tibet	03-664
J16	L. discolor Beauverd	Russia, Shakalin, Tymorsky distr.	07-365
J11	L. cf. haplophylloides Hand.-Mazz.	China, Sichuan, Shaluli Shan	07-526
J13	L. cf. haplophylloides Hand.-Mazz.	China, Sichuan, Litang	07-531
J06	L. cf. himalayanum DC.	China, Sichuan, Litang	05-520
J15	L. cf. leontopodinum Hand.-Mazz.	Tadjikistan, Pamir	07-547
J20	L. pusillum Hand.-Mazz.	China, Tibet, Bamda	07-556
J03B	L. aff. sinense Hemsl. ex Forb. & Hemsl.	China, Yunnan, Hengduan Shan	03-302
J09	L. cf. souliei Beauverd	China, Sichuan, Shaluli Shan	07-523
J14	L. cf. souliei Beauverd	China, Sichuan, Litang	07-532
J08	L. stracheyi C.B. Clarke ex Hemsl.	China, Xizang, E-Tibet	03-665
J12A	L. sp.	China, Quinghai, Huashixia	07-530A
J12B	L. sp.	China, Quinghai, Huashixia	07-530B

For the collected species, voucher numbers are indicated, for the cultivated samples cultivar numbers.

Results and discussion

1H NMR spectroscopy

Extraction of plant material and acquisition of NMR spectra

Powdered plant material was extracted directly with DMSO-d6. This is adequate for qualitative purposes and simplifies the extraction process. DMSO was the most appropriate solvent for our samples; both apolar and polar compounds were extracted, resulting in a broad range of metabolites. The extraction method used was simple and convenient, requiring just a small amount of plant material suspended in the solvent and extracted on a flat-bed shaker for 24 h. After centrifugation, an aliquot of the supernatant was analysed directly by 1H NMR spectroscopy. Due the use of an auto sampler, NMR experiments were also accomplished overnight, which was additionally time-saving. Three sample replicates (and accordingly only two sample replicates for Leontopodium himalayanum due to a lack of plant material) were used to test the precision of the method.

Multivariate statistical analysis and pattern recognition

The spectra were imported into AMIX and pre-processed using the bucketing function. By generating a number of integrated regions for each dataset, complexity of the NMR spectra was reduced. Here, we found a bucket width of δ 0.04 suitable for our data. A table of bucket-integrated spectra was exported as a spreadsheet.

Principal component analysis (PCA) was performed in SIMCA-P. PCA is an unsupervised method and was used to reduce the dataset in order to obtain the maximum variation between the samples. Mean-centering was chosen for scaling. The method focuses on the fluctuating part of the data, and leaves only the relevant variation (i.e., the variation between the samples) for analysis (van den Berg et al., 2006). A ten-component model was calculated and explained 98.1% of the variation, with the first two components explaining 84.2%. Principal component (PC) 1 was the dominant factor for classification of the groups, whereas PCs 3–10 did not influence the results. Intragroup clustering for each group highlights the good method precision (Fig. 1A). Discriminating NMR signals are presented in the loadings plot (Fig. 1B), typical 1H NMR spectra of the different species are displayed in Fig. 2A.

Fig. 1

Scores plot (A) and loadings plot (B) of principal component analysis (PCA) results obtained from 1H NMR spectra of 11 collected Leontopodium species using PC 1 (68.8%) vs. PC 2 (15.4%). Discriminating NMR signals are highlighted in the loadings plot.

Fig. 2

Typical 1H NMR spectra of DMSO-d6 root extracts of the 11 investigated Leontopodium species in the range of δ 9.00–0.00 (A). Comparison of 1H NMR spectra of the closely related L. sinense, L. franchetii, L. dedekensii, and an unidentified species, L. sp. in the range of δ 9.00–4.00 (B).

Two groups can be identified. The main group A consists of eight species with a similar metabolic pattern. Signals at δ 1.26 and δ 1.22 were compared to literature data and assigned to the lipid region, corresponding to the major signal of fatty acids (Rasmussen et al., 2006). These signals are responsible for the discrimination of Leontopodium cf. stracheyi.

Due to comparison with 1D (1H NMR) and 2D (HSQC) NMR spectra of sucrose, signals at δ 3.82, δ 3.78, δ 3.66, and δ 3.62 could be assigned to sucrose, the signal at δ 5.18 (J = 3.7 Hz) to its anomer. Species like Leontopodium artemisiifolium and Leontopodium calocephalum are discriminated by these signals. Furthermore, intraspecific variations in terms of primary metabolites (i.e., sucrose) can be found for species with more than one population included (e.g., L. andersonii, Leontopodium dedekensii, Leontopodium souliei). Metabolic discrepancies between populations are responsible for these variations; as a consequence, different populations belonging to the same species are misaligned in PCA. This variation may be a result of environmental influences, and highlights that the metabolite pattern (i.e., mostly primary metabolites) is strongly affected by ecological factors.

The metabolic fingerprint of L. dedekensii is most similar to NMR spectra of species belonging to group B (see below and Fig. 2B), even though the signals in the aliphatic region are missing. Furthermore, the characteristic signals at δ 8.30, δ 7.90, and δ 7.00 could be assigned to an already described benzofuran (1-{(2R∗,3S∗)-3-(β-d-glucopyranosyloxy)-2,3-dihydro-2-[1-(hydroxymethyl)vinyl]-1-benzofuran-5-yl}ethanone; Dobner et al., 2003a; see online supplementary data, Fig. S1). These signals were also identified for other Leontopodium species (L. artemisiifolium and L. cf. stracheyi). Nevertheless, the corresponding benzofuran is irrelevant for discrimination of those species (Fig. 1B).

Group B comprises L. franchetii, L. sinense, and an unidentified species, L. sp. The discriminating signals for this group can be found mostly in the aliphatic region (δ 2.50–0.50), other important signals have a higher chemical shift (i.e., δ 4.78 and δ 4.70). Comparison of the metabolic fingerprints of these species (Fig. 2B) shows that the spectrum of L. sp. is a mixture of the spectra of two other species, L. sinense and L. franchetii. Whereas the signals at δ 4.78 and δ 4.70 can be found in L. franchetii and L. sp., these resonances are missing in L. sinense. On the other hand, the characteristic benzofuran signals at δ 8.30, δ 7.90 and δ 7.00 (Dobner et al., 2003a), are present in the spectra of L. sinense and L. sp., and nonexistent in the spectrum of L. franchetii. The unidentified species was first considered to be L. dedekensii or L. sinense, because several morphological characters are congruent. Recent phylogenetic research based on DNA-fingerprinting (AFLP; Amplified Fragment Length Polymorphism) has shown that the unidentified species is closely related and a sister species to L. dedekensii and L. sinense. Due to similar morphology, also hybridisation between L. dedekensii and L. sp. has been probably partly responsible for the classification obtained with AFLP (Safer et al., 2011). Our results show, that L. sp. might be closely related to L. franchetii as well. Shared morphological features can be found in both species, but results from AFLP were in this case not clear enough to tell if there is a close phylogenetic relationship (Safer et al., 2011). Nevertheless, due to similarities in morphology, genetic and metabolic profile, L. dedekensii, L. franchetii, L. sinense and L. sp. can be definitely assigned to a mutual group. In addition, hybridisation between L. sinense and L. franchetii could also be a possible explanation for the similarities in the metabolic fingerprints of the three species within group B.

Isolation and identification of discriminating compounds from L. franchetii

Discriminating compounds for group B were isolated from the roots of L. franchetii using standard procedures (i.e., silica gel column chromatography, Sephadex column chromatography, preparative HPLC, etc.) as described in section 4.5. Structures of the pure compounds were elucidated using 1D (1H, 13C) and 2D (HSQC, HMBC, COSY) NMR spectra, and identified by comparison with NMR literature data (all substances have already been described: Bohlmann et al., 1980a,b; Bohlmann and Zdero, 1977; Bohlmann et al., 1981; Gray et al., 2000; Hasan et al., 1982; Ohno et al., 1979). Five diterpenes and one sesquiterpene (Fig. 3) were isolated and described the first time for L. franchetii: ent-kaur-16-en-19-oic acid (1), methyl-15α-angeloyloxy-ent-kaur-16-en-19-oate, (2), methyl-ent-kaur-16-en-19-oate (3), 8-acetoxymodhephene (4), 19-acetoxy-ent-kaur-16-ene (5), methyl-15β-angeloyloxy-16,17-epoxy-ent-kauran-19-oate (6).

Fig. 3

Chemical structures of compounds 1–9 [ent-kaur-16-en-19-oic acid (1), methyl-15α-angeloyloxy-ent-kaur-16-en-19-oate (2), methyl-ent-kaur-16-en-19-oate (3), 8-acetoxymodhephene (4), 19-acetoxy-ent-kaur-16-ene (5), methyl-15β–angeloyloxy-16,17-epoxy-ent-kauran-19-oate (6), 3-methyl-1-{2-[(1R∗,2R∗,5R∗,6S∗)-2,5,6-tris(acetyloxy)-4-methylcyclohex-3-en-1-yl]propyl}but-2-enyl (2Z)-2-methylbut-2-enoate (7), methyl-ent-7α,9α-dihydroxy-15β-[(2Z)-2-methyl-but-2-enoyloxy]kaur-16-en-19-oate (8), (1R∗,5S∗,6S∗)-5-(acetyloxy)-6-[3-(acetyloxy)-1,5-dimethylhex-4-enyl]-3-methylcyclohex-2-en-4-on-1-yl (2Z)-2-methyl-but-2-enoate (9)].

The main compound in L. franchetii roots is compound 1 (ent-kaurenoic acid). To determine the influence of this compound on the discrimination of the samples, 1D and 2D NMR spectra of L. franchetii were compared with 1D and 2D NMR spectra of ent-kaurenoic acid (see online supplementary data, Fig. S2). The signals responsible for discrimination of L. franchetii (Fig. 1B) can be assigned to 1H resonances of ent-kaurenoic acid. Signals in the spectral region δ 1.90–0.90 are part of the basic structure of ent-kaurenoic acid, and the signals at δ 4.78 and δ 4.70 can be assigned to the exocyclic double bond at position C-16. Similar signals can also be found in the 1D NMR spectra of other diterpenes isolated in this study, indicating that ent-kaurenoic acid and its derivatives are the discriminators for L. franchetii and the other two species of group B.

Cultivated samples vs. collected samples

As observed in the PCA of the NMR fingerprints, intraspecific variations depending on ecological factors can be found within the genus Leontopodium. Therefore, 1H NMR spectra of collected species were compared to spectra of cultivated species using a partial least squares-discriminant analysis (PLS-DA). Contrary to PCA, which is an unsupervised method and can be applied without prior knowledge about samples, PLS-DA is a supervised method and uses information about the samples to maximise the differences between two or more a priori defined classes (Holmes et al., 2006).

Here, the samples were divided into two classes: class 1 represented the plants collected in China, whereas class 2 comprised all cultivated samples. A ten component model was calculated and explained 95.5% of the variation, with the first three PLS components explaining 79.5%. The result of the PLS-DA is displayed in a 3D-scores plot (Fig. 4). The scores plot showed a clear differentiation between the two classes, although some species (e.g., L. calocephalum, L. dedekensii, L. sinense, L. souliei; see Table 1) were present as both collected and cultivated samples.

Fig. 4

3D-scores plot of partial least squares-discriminant analysis (PLS-DA) results obtained from 1H NMR spectra of 11 collected and 12 cultivated Leontopodium species using PLS components 1–3. Class 1: collected plants, class 2: cultivated plants.

These findings suggest a correlation between metabolic patterns and ecological factors. Cultivated species were grown in the Botanical Garden of Giessen (Germany) and therefore not exposed to climatic conditions which can be found in the natural habitat (Qinghai-Tibetan plateau, QTP, south-western China), such as high UV radiation, high precipitation and similar. Environmental and climatic stress may cause changes in the production of primary and secondary metabolites. Furthermore, higher average temperatures in Giessen, Germany (compared to the QTP in China) could lead to increased plant growth and biochemical activity of the plants. Moreover, the plot clearly shows a tighter clustering within the cultivated sample group. These plants were grown under the exact same conditions and their metabolic profile is similar. Plants belonging to the class of collected species occupied different habitats and were exposed to unequal environmental conditions. As a consequence, the scattering within this group is more distinctive.

LC–MS

Acquisition of mass spectra

Extracts prepared for NMR fingerprinting (see Section 2.1.1) were also used for LC–MS analysis. Due to the broad and complex metabolite spectrum obtained by DMSO extraction, the LC method had to be rather long to enable separation of a large number of compounds (65 min). Both positive and negative ionisation modes were tested for mass detection. The positive mode was chosen for further analyses since ionisation in negative mode was not satisfying. LC–MS analyses were only performed for collected species. Triplicates of 22 different samples belonging to 11 species resulted in a total analyses time of nearly 5 days (for L. himalayanum only duplicates were analysed due to a lack of plant material).

For analysis of the acquired dataset with multivariate methods, LC–MS chromatograms were pre-processed using MZmine to compensate for variations in retention time and m/z value between the chromatographic runs. The pre-processed chromatograms were exported as a peak list table, with rows representing the individual samples, and columns representing the integrated and normalised peak areas.

A ten-component model was calculated and explained 93.9% of the variance, with the first two components explaining 46.7%. The remaining components contributed as follows: PC 3 (11.9%), PC 4 (10.3%), PC 5 (7.3%), PC 6 (5.3%), PC 7 (4.3%), PC 8 (3.6%), PC 9 (2.4%), and PC 10 (2.0%). Using PC 1 and PC 2, the species were found to be clustered into three groups (Fig. 5A). Intragroup clustering for each group indicates the good method precision. Discriminating m/z values are displayed in a loadings plot (Fig. 5B); typical total ion chromatograms (TICs) of the investigated species are presented in Fig. 6.

Fig. 5

Scores plot (A) and loadings plot (B) of principal component analysis (PCA) results obtained from LC–MS fingerprints of 11 collected Leontopodium species using PC 1 (29.6%) vs. PC 2 (17.1%). Discriminating m/z values are highlighted in the loadings plot.

Fig. 6

Typical total ion chromatograms (TICs) of DMSO-d6 root extracts of the 11 investigated Leontopodium species, acquired in positive ionisation mode by electrospray ionisation (ESI).

Group A consists of four species. Species of this group are morphologically (Dickoré, unpublished) and genetically diverse (Safer et al., 2011), ranging from tall woody herbs like L. artemisiifolium to small shrubs like L. himalayanum. Discriminating compounds for group A are defined by m/z values of 501.1, 477.1, and 459.0. By comparison of retention times and mass spectra with literature data, the m/z value of 501.1 was assigned as [M+Na]+ to an already described bisabolane derivative (Fig. 3, compound 7: 3-methyl-1-{2-[(1R∗,2R∗,5R∗,6S∗)-2,5,6-tris(acetyloxy)-4-methylcyclohex-3-en-1-yl]propyl}but-2-enyl (2Z)-2-methylbut-2-enoate; i.e., an isomeric mixture;) with a calculated mass of 478 (Stuppner et al., 2002). The compound was detected as a dominant double peak at a retention time of 42 min (see TICs of L. himalayanum and L. artemisiifolium, Fig. 6; an extracted ion chromatogram (EIC) of the corresponding m/z value is provided as online supplementary data, Fig. S3). Unfortunately, m/z values of 477.1 and 459.0 could not be identified. Comparison with literature data suggested that the m/z value of 459 corresponds to a bisabolane derivative isolated from L. longifolium (= L. souliei; Li et al., 2006).

Leontopodium sinense and L. sp. are forming a clearly differentiated group B (mainly discriminated with PC 1). These species were already described as closely related (Safer et al., 2011), and this finding can be confirmed with both NMR and LC–MS fingerprinting. Unlike NMR spectra, TICs do not exhibit many differences between the two species. The group is discriminated by m/z values of 469.2, 329.3 and 311.4, which correspond to an ent-kaurenoic acid derivative described for L. alpinum (Fig. 3, compound 8: methyl-ent-7α,9α-dihydroxy-15β-[(2Z)-2-methyl-but-2-enoyloxy]kaur-16-en-19-oate, calculated mass 446; Schwaiger et al., 2004). Again, identification was done by comparing retention times and mass spectra: 469.2 [M+Na]+; 329.3 [M−C5H7O2−H2O]+; 311.4 [M−C5H7O2−2H2O]+. Furthermore, the m/z value of 915.1 could be assigned to [2 M+Na]+ (EICs of corresponding m/z values are provided as online supplementary data, Fig. S4).

Regarding the detection of ent-kaurenoic acid derviatives, the two approaches revealed different patterns. NMR metabolic fingerprinting resulted in a grouping of L. franchetii, L. sinense and L. sp (group B), and ent-kaurenoic acid (compound 1) and its derivatives (compounds 2, 3 and 6) could be determined as discriminating compounds. In LC–MS analysis, L. franchetii was not included within group B, and a previously isolated ent-kaurenoic acid derivative (compound 8; methyl-ent-7α,9α-dihydroxy-15β-[(2Z)-2-methyl-but-2-enoyloxy]kaur-16-en-19-oate, calculated mass 446; Schwaiger et al., 2004) was identified as discriminator. LC–MS peaks for compound 1 could not be identified for any of the three species, suggesting that ionisation of compound 1 is limited. On the other hand, LC–MS peaks for compound 8 with m/z 469, 329, and 311 (Schwaiger et al., 2004) could only be recognised for L. sinense and L. sp. This determines compound 8 with molecular weight 446 as main discriminator for the two species but not for L. franchetii.

Group C consists of L. andersonii, L. caespitosum, L. dedekensii, L. franchetii and L. cf. stracheyi. L. franchetii, which occupied a conspicuous position within group B in NMR fingerprinting, is classified within group C in LC–MS analysis. In terms of LC–MS, the main compound of L. franchetii, ent-kaurenoic acid, does not have any influence on the discrimination of this species (see above). L. andersonii is morphologically distinct and occupies a genetically unique position within the genus (Safer et al., 2011). NMR analysis did not reveal characteristic metabolic fingerprints for the species, placing L. andersonii within the large group A. In contrast, results of LC–MS fingerprinting showed a different pattern. Taking PC 3 (not shown) into account, the discrimination of L. andersonii was explicit. The species is discriminated by an m/z value of 457. Recent phytochemical investigations of L. andersonii discovered a novel bisabolone derivative (Fig. 3, compound 9; (1R∗,5S∗,6S∗)-5-(acetyloxy)-6-[3-(acetyloxy)-1,5-dimethylhex-4-enyl]-3-methylcyclohex-2-en-4-on-1-yl (2Z)-2-methyl-but-2-enoate, calculated mass 434; Schwaiger et al., 2010), which was not described for other Leontopodium species yet. Hence, the m/z value of 457 could be identified as [M+Na]+, and determines the bisabolone as the discriminating compound for L. andersonii (an EIC of the corresponding m/z value is provided as online supplementary data, Fig. S5).

Conclusions

We found both 1H NMR spectroscopy and LC–MS useful for metabolic fingerprinting of species of the genus Leontopodium. The combination of the two methods offered valuable insights about metabolic patterns of the different species. Whereas with NMR the total metabolic status could be recorded including primary and secondary metabolites, LC–MS fingerprinting exhibited details on specific secondary metabolites.

In NMR fingerprinting, the major compounds responsible for discrimination were identified as fatty acids, sucrose, and ent-kaurenoic acid and derivatives thereof. Ent-kaurenoic acid was identified as the main compound of L. franchetii. Altogether, five diterpenes and one sesquiterpene were isolated and described for the first time for L. franchetii. Furthermore, PLS-DA analysis between collected and cultivated species highlighted the influence of environmental and ecological factors on the production of metabolites as a result of modified biochemical activity.

With LC–MS fingerprinting, several discriminating compounds could be identified for the different groups, including two bisabolane derivatives and one ent-kaurenoic acid derivative. Since LC–MS did not offer much information on chemical structures of the compounds, comparison of the recorded mass spectra and retention times with literature data revealed attribution of the signals to the corresponding compounds. Furthermore, information about secondary metabolites of species not investigated yet could be obtained by checking their group assignment within PCA.

In addition, new insights concerning species relationships within the genus could be acquired with both fingerprinting approaches. The unidentified species, L. sp., which was considered to be closely related to L. sinense in molecular analysis, showed similarities in NMR fingerprints with L. franchetii as well. This indicates possible hybridisation events between L. sinense and L. franchetii. L. sinense and L. dedekensii are closely related to each other, and therefore often wrongly identified because they share several morphological characters. This close relationship was confirmed in a recent study (Safer et al., 2011) dealing with DNA-fingerprinting of Leontopodium species. Our results exhibited clear differences in the metabolic pattern of those two species, classifying L. sinense and L. dedekensii unambiguous into two groups. Where identification with morphological and molecular methods may be difficult, NMR and LC–MS fingerprinting approaches could offer additional information on species relationship and facilitate classification of the species.

Experimental

General experimental procedures

1D NMR spectra for metabolic fingerprinting were acquired on a Bruker Avance II spectrometer (Bruker BioSpin, Rheinstetten, Germany) equipped with an automated sample exchanger at a temperature of 300 K, operating at 600 MHz. 1D and 2D NMR spectra for structure elucidation were acquired on a Bruker DRX 300 spectrometer (Bruker BioSpin, Rheinstetten, Germany), operating at 300 MHz. LC–MS analyses were performed on an Agilent 1100 HPLC system (Agilent, Waldbronn, Germany) coupled with a Bruker Daltonics esquire 3000plus mass spectrometer (Bruker Daltonics, Bremen, Germany) equipped with an electrospray (ESI) interface. Semi-preparative HPLC was carried out on a Dionex preparative HPLC system (P580 pump, ASI 100 automated sampler, Ultimate 3000 column department, UVD 170 U detector; Dionex Softron, Germerling, Germany) equipped with a Gilson Abimed 206 fraction collector (Gilson International, Middleton, WI, USA). Column chromatographies were performed with Sephadex LH-20 (Pharmacia Biotech AB, Stockholm, Sweden) and silica gel 60 (0.040–0.063 mm; Merck, VWR, Darmstadt, Germany). CDCl3 and DMSO-d6 were obtained from Euriso-Top (Paris, France). All solvents used for HPLC analysis were gradient grade, all solvents for extraction technical grade.

Plant material

Whole plants were collected in south-western China in 2008 (Safer et al.). Vouchers are deposited in the herbaria of the University of Vienna, Austria (WU), and the Chinese Academy of Sciences in Beijing, China, (PE). Roots of cultivated plants were obtained from the Botanical Garden in Giessen (Germany). Only dried plant material was used for all analyses. Population numbers, species names and voucher information (WU) are listed in Table 1.

Extraction and sample preparation

Roots were frozen with liquid nitrogen and powdered using mortar and pestle. 100 mg of finely powdered plant material was weighed into a 1.5 ml Eppendorf tube. 1.2 ml of DMSO-d6 (containing 0.03% TMS) was added to each sample. The tubes were mixed thoroughly on a flat-bed shaker for 24 h. The samples were spun down in a micro-centrifuge at 14,000 rpm for 5 min. 700 μl of the supernatant was filtered through cotton wool into a 5 mm NMR tube. Triplicates were prepared for each sample (for Leontopodium himalayanum, only two samples were prepared due to a lack of plant material). The same samples were used for LC–MS analyses; for each sample, the extract was diluted with DMSO (1:5).

Five hundred and twelve scans were accumulated, resulting in an acquisition time of 30 min per sample. A water suppression pulse sequence was used. The relaxation delay was 2.40 s, the acquisition time 1.36 s. Spectral width was δ 20.00, size of FID 32 k, and size of real spectrum 64 k. Fourier transformation and polynomial baseline correction were carried out automatically, phase correction was done manually using TOPSPIN 2.0 (Bruker Biospin). 1H NMR chemical shifts in the spectra were referenced to TMS at δ 0.00. To reduce the size of the spectra to a number of variables suitable for statistical analysis, 1H NMR spectra were imported into AMIX (Analysis of MIXtures software v.3.7.5, Bruker Biospin). Spectral intensities were bucket-integrated to equal width (δ 0.04). The regions between δ 3.60 and 3.00 (residual water) and δ 2.56–2.46 (residual DMSO) were removed prior to statistical analysis. Spectra were normalised to the total signal area. The pre-processed spectra were exported as a bucket table with rows representing the individual NMR spectra, and columns (comprising 220 variables) representing the integrated regions. Principal component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA) were performed with the programme SIMCA-P ver. 10.0 (Umetrics, Umeå, Sweden). The mean-centering scaling method (ctr) was applied to both PCA and PLS-DA.

Isolation of discriminating compounds from L. franchetii roots

Dried roots (115 g) were ground using a laboratory mill (IKA MF10 basic). The finely powdered roots were extracted with CH2Cl2 using an ultrasonic bath and repeated maceration. The solvent was evaporated under reduced pressure to obtain 10.7 g crude extract. The extract was subjected to silica gel column chromatography (50 × 5 cm) and eluted with a pet-ether-Me2CO gradient (9:1 to 4:6) yielding seven fractions (Lf1A–Lf1G). Lf1H and Lf1I were obtained by flushing the column with Me2CO and MeOH, respectively. Fraction Lf1B (3.35 g) was applied to a Sephadex LH-20 column (100 × 4 cm) and eluted with CH2Cl2–Me2CO (85:15), yielding five subfractions (Lf2A–Lf2E). Lf2E gave 1.50 g of compound 1. Fraction Lf2B was subjected to a silica gel column chromatography (90 × 3.5 cm) using a solvent system of pet-ether–CH2Cl2 by gradient elution (8:2 to 2:8). This resulted in ten subfractions (Lf3A–Lf3J), whereas Lf3H gave 40 mg of compound 2. Lf3K was obtained by flushing the column with pure CH2Cl2. Lf3E was subjected to a semi-preparative HPLC (column: Waters XTerra C18 5 μm, 100 × 7.80 mm; solvent system: H2O (A)–MeOH (B); gradient: 0 min 70% B, 15 min 98% B, 30 min 98% B), yielding 13 mg of compound 3. Lf3F was also separated with the semi-preparative HPLC system (isocratic H2O–MeOH 20:80), resulting in compound 4 (5 mg) and compound 5 (9 mg). Lf3K was purified with a silica gel column chromatography (38 × 3 cm) using CH2Cl2 with 2% Me2CO as a solvent system. The column was flushed with CH2Cl2–Me2CO (8:2) and pure Me2CO at the end, resulting in a total of 10 subfractions (Lf4A–Lf4J). Lf4C gave 67 mg of compound 6. Structures of the compounds were elucidated via 1D and 2D NMR spectroscopy using CDCl3 as an NMR-solvent; NMR experiments (1H, 13C, HSQC, HMBC and COSY) were carried out using Bruker standard acquisition parameters. Spectroscopic data for compounds 1, 2, 3, 5, and 6 are provided as online supplementary material (Tables S1 and S2).

The separation was carried out using a Phenomenex LUNA C18 column (3 μm, 150 × 2.00 mm) at 40 °C, with a mobile phase including H2O (A), and a mixture of MeOH and MeCN (1:10, v/v) containing 0.9% HCO2H and 0.1% HOAc (B). Analyses were performed at a flow rate of 0.2 ml/min using the following gradient: 0 min 15% B, 15 min 25% B, 25 min 45% B, 30 min 85% B, 55 min 95% B, 65 min 95% B. The injection volume was 10 μl. Detection was performed in both positive and negative ionisation mode in the m/z range of 100–1000. The following ESI conditions were used: Nebulizer 40.0 psi, dry gas 5.0 l/min, dry temperature 300 °C, and capillary voltage 1500 V. Acquired spectra were saved as total ion chromatograms (TICs) in NetCDF format.

TICs were pre-processed with the programme MZmine ver 1.97 (Katajamaa and Oresic, 2005). Mass peaks were detected, chromatograms were retention time normalised, deconvoluted, isotopic peaks were grouped, and the chromatograms were aligned. To avoid missing data, gaps were filled via the peak finder function. Duplicate peaks were filtered and a linear normalizer was applied. Pre-processed spectra were exported as a peak list table, with rows representing the individual mass spectra, and columns (comprising 199 variables) representing the integrated and normalised peak areas. The peak list table was imported into SIMCA-P 10.0 (Umetrics, Umeå, Sweden); a PCA was carried out using mean-centering (ctr) for scaling.