Abstract

   Dendrobium officinale, a precious herbal medicine, has been used for a
   long time in Chinese history. The metabolites of D. officinale,
   regarded as its effective components to fight diseases, are
   significantly affected by cultivation substrates. In this study,
   ultra-performance liquid chromatography mass spectrometry (UPLC-MS/MS)
   was conducted to analyze D. officinale stems cultured in three
   different substrates: pine bark (PB), coconut coir (CC), and a pine
   bark: coconut coir 1:1 mix (PC). A total of 529 metabolites were
   identified. Multivariate statistical analysis methods were employed to
   analyze the difference in the content of metabolites extracted from
   different groups. By the criteria of variable importance in projection
   (VIP) value ≥1 and absolute log[2] (fold change) ≥1, there were a total
   of 68, 51, and 57 metabolites, with significant differences in content
   across groups being filtrated out between PB and PC, PB and CC, and PC
   and CC, respectively. The comparisons among the three groups revealed
   that flavonoids were the metabolites that fluctuated most. The results
   suggested the D. officinale stems from the PB group possessed a higher
   flavonoid content. Kyoto Encyclopedia of Genes and Genomes (KEGG)
   pathway enrichment analysis indicated that the significantly regulated
   metabolites were mainly connected with flavonoid biosynthesis. A
   comprehensive profile of the metabolic differentiation of D. officinale
   planted in different substrates was provided, which supports the
   selection of an optimum cultivation substrate for a higher biomass
   yield of D. officinale.

   Keywords: Dendrobium officinale, metabolomics, differential
   metabolites, cultivation substrates

1. Introduction

   Dendrobium is a genus of valuable medicinal herbs in the family
   Orchidacea [[38]1], containing 74 species and two varieties in China
   [[39]2] of which many species can be used as medicine [[40]3,[41]4].
   Dendrobium officinale is a precious species and famously known as one
   of the nine immortal herbs [[42]5], as it possesses many kinds of
   physiological functions [[43]6], including immune modulation [[44]7],
   antitumor activity [[45]8], and helping to reduce levels of blood
   glucose [[46]9,[47]10]. However, wild D. officinale is scarce because
   it requires a peculiar growth environment and takes a long time to
   grow. Moreover, wild D. officinale has been exploited excessively due
   to the large demand from the market [[48]11]. Recently, in order to
   satisfy market demand and save wild resources, researchers tried to
   plant D. officinale by means of artificial cultivation [[49]12,[50]13].
   The methods, which can be effectively applied to plant D. officinale,
   included a cultivation technique aiming to imitate the wild conditions
   and a plastic tunnel culture (PL). The PL technique in particular can
   simulate the growth environment of D. officinale in a greenhouse
   [[51]14]. Indeed, the yield substantially increased as a result.
   However, the quality of artificial D. officinale is evaluated not only
   by the yield but also by the effective ingredients. It is widely
   accepted that ingredients, such as polysaccharose, flavones, and
   phenolic acids [[52]15,[53]16,[54]17] are the main resource causing the
   benefits derived from D. officinale, and all of them fluctuate when the
   growth environment (cultivation substrate, illumination time, and
   temperature) of D. officinale changes. Thus, it is important to
   determine the optimal growth environment for D. officinale to gain a
   higher output and higher quality.

   Over recent years, many research studies have been carried out about
   evaluating the quality of D. officinale. Some [[55]18,[56]19] used
   growth indices consisting of survival rate, plant height, stem
   diameter, and number of roots and leaves, which can represent the yield
   of the plants to some degree. A series of studies [[57]20,[58]21] was
   carried out using effective ingredients, such as polysaccharides,
   alkaloids, and total flavonoids, which can represent the quality of the
   D. officinale, to estimate the quality of the medicinal plant.

   However, the efficacy of traditional Chinese medicine should be derived
   from the combination of several components rather than a single
   ingredient, such as polysaccharides, alkaloids, or flavonoids. Thus,
   the study of the detailed metabolite profile of a medicine sample,
   which could be used to guide the planting of the medicine, plays an
   important role in comprehensively evaluating medicinal efficacy. In
   this study, a widely targeted metabolomics analysis [[59]22],
   integrating the “universality” of a nontargeted metabolomic and the
   “accuracy” of a targeted metabolomics [[60]23], was employed to study
   the detailed metabolite profile of D. officinale. The widely targeted
   metabolomics analysis was implemented using liquid chromatography mass
   spectrometry (LC-MS), gas chromatography mass spectrometry (GC-MS), and
   nuclear magnetic resonance (NMR), combined with multiple reaction
   monitoring (MRM) [[61]24]. The whole process involved conducting the
   instruments to detect the contents of small molecule metabolites in
   cells, tissues, and organs or organisms in different conditions; using
   multivariate statistical analysis to screen the differential
   metabolites, and employing pathway analysis of differential metabolites
   and revealing the physiological mechanisms behind changes.

   Stems were collected from D. officinale cultivated in three kinds of
   cultivation substrates. By means of the widely targeted metabolomics
   analysis, ultra-performance liquid chromatography mass spectrometry
   (UPLC-MS/MS) was employed to obtain data (including retention time and
   peak area) from the metabolites of D. officinale for quantitative and
   qualitative analysis. The data obtained by quantitative and qualitative
   analysis were further analyzed using multivariate statistical analysis
   to observe whether the difference of the metabolites was significant,
   removing the interfering signals, and screening the differential
   metabolites of D. officinale in three different substrates. Finally,
   metabolic pathway analysis was conducted to search the metabolic
   pathway involved. The results revealed the difference of D. officinale
   metabolism among the three cultivation substrates (pine bark (PB),
   coconut coir (CC), and a pine bark: coconut coir 1:1 mix (PC)) and
   could provide basic information to guide D. officinale planting.

2. Results

2.1. Data Evaluation

   To assess the repeatability of samples and stability of instruments,
   Pearson’s correlation coefficients were computed with the “cor”
   function in R (Ver. 3.5.0). The results of correlation analysis are
   given in [62]Table 1. Correlation coefficients between intragroup
   samples (PB vs. PB, CC vs. CC, PC vs. PC) were higher than intergroup
   samples (PB vs. PC, PB vs. CC, PC vs. CC), which illustrated that the
   reproductivity of samples and the stability of the instrument were
   good. Thus, the method was suitable to carry out qualitative and
   quantitative analysis. The comparisons of PB vs. PC, PB vs. CC, and PC
   vs. CC exhibited a high correlation coefficient (0.8638, 0.9078, and
   0.9108, respectively), showing that the metabolites screened in the
   three pairs of comparative groups were reliable.

Table 1.

   Average Pearson’s correlation coefficients for comparisons of
   intragroup and intergroup samples.
     Group   Average Pearson’s Correlation Coefficient
   PB vs. PB                  0.9403
   PC vs. PC                  0.9422
   CC vs. CC                  0.9519
   PB vs. PC                  0.8636
   PB vs. CC                  0.9078
   PC vs. CC                  0.9108
   [63]Open in a new tab

   Note: the reproducibility of samples and the stability of the
   instruments are defined as good when the correlation coefficients of
   intragroup comparisons are higher than intergroup comparisons. The data
   were considered reliable when the correlation coefficients of
   intragroup samples were above 0.8.

2.2. Widely Targeted Metabolomics Profiling Results

   In this study, the widely targeted metabolomics technique was used to
   detect the metabolic information of D. officinale from three
   substrates—PB, PC, and CC. Qualitative and quantitative analysis were
   carried out on the data obtained from UPLC-MS/MS. The total ion current
   (TIC) maps ([64]Figure S1) of mass-spectrometry analysis of quality
   control (QC) samples detected at different times were well-overlapped,
   showing the instrument had good stability. As results, 529 metabolites
   were detected in total, and they were divided into 12 classes including
   flavonoids, phenolic acids, lipids, nucleotides and derivatives,
   alkaloids, amino acids and derivatives, lignans and coumarins, organic
   acids, quinones, tannins, terpenoids, and others. Among them, the top
   three most represented metabolites were flavonoids, phenolic acids, and
   lipids, which contained 135, 75, and 72 metabolites, respectively.
   After annotating all metabolites to the Kyoto Encyclopedia of Genes and
   Genomes (KEGG) compound database, 203 metabolites were matched, among
   which there were 116 metabolites with significant differences.

2.3. Principal Component Analysis

   The metabolic distribution trend among different groups and the degree
   of discreteness of different molecules in each group are shown in
   [65]Figure 1. QC samples were gathered in a group and separated from
   other samples, which shows the QC samples had good repeatability. The
   samples from PB and PC were separated in the direction of PC2, and
   those from PC and CC were also separated in the direction of PC2.
   However, the samples from CC and PB overlapped a lot not only in the
   direction of PC1 but PC2 as well. The overall results of the PCA
   indicate the samples of PC were different compared with the samples of
   PB and CC, while the difference between PB and CC was not significant.

Figure 1.

   [66]Figure 1
   [67]Open in a new tab

   Score plot of principal component analysis (PCA) for D. officinale
   samples planted in pine bark (PB), coconut coir (CC), pine bark:
   coconut coir 1:1 mix (PC), and quality control (QC) samples. PC1
   represents the first principal component and PC2 represents the second
   principal component.

2.4. Orthogonal Partial Least Squares Discriminant Analysis

   Principal component analysis (PCA) can extract the main information
   effectively, but it is not sensitive enough to exclude the variables
   with smaller correlations; orthogonal partial least squares
   discriminant analysis (OPLS-DA), however, can achieve this. The
   variables with significant differences could also be screened by
   OPLS-DA. The OPLS-DA score plots are shown in [68]Figure 2. Samples
   from the groups of PB vs. PC, PB vs. CC, and CC vs. PC had an obvious
   degree of separation, which illustrated that the metabolites generally
   exhibited significant differences among the D. officinale samples
   cultured in different substrates. To validate the OPLS-DA models,
   seven-fold cross-validation and 200 permutation tests [[69]25] were
   conducted in this study (the results of permutation tests are shown in
   [70]Figure S2). The R^2 and Q^2 values of the models of PB vs. PC, PB
   vs. CC, and PC vs. CC were 0.998 and 0.855, 0.995 and 0.73, 0.99 and
   0.721, respectively, indicating that the models were of high fitness
   and predictability. Meanwhile, it can be said that the variable
   importance in projection (VIP) values generated from the OPLS-DA model
   were also reliable.

Figure 2.

   [71]Figure 2
   [72]Open in a new tab

   Score plots of orthogonal partial least squares discriminant analysis
   (OPLS-DA): (a) PB vs. PC; (b) PB vs. CC; (c) PC vs. CC. R^2Y is the
   interpretation rate of the model to the Y matrix (Class label) and Q^2
   the predictive ability of the model. The model is valid when Q^2 > 0.5.
   The X-axis is the predictive principal component, and the Y-axis is the
   orthogonal principal component.

2.5. Identification of Differential Metabolites

   Metabolites that differed significantly between D. officinale samples
   cultured in different substrates were screened using univariate
   analysis and multivariate statistical analysis. In this study, VIP
   values (VIP ≥1) combined with fold-change (FC) values (|log[2] (FC)| ≥
   1) were used to screen these metabolites. In general, the metabolites
   with VIP values over 1 contributed a lot to the classification in the
   OPLS-DA model, and the FC value was used to screen metabolites with
   content that varied significantly by substrate.

   Volcano plots are shown in [73]Figure 3 to display the metabolites that
   differed significantly. The detailed information of differential
   metabolites among the groups can be seen in [74]Tables S2–S4.

Figure 3.

   [75]Figure 3
   [76]Open in a new tab

   Volcano plots for (a) PB vs. PC; (b) PB vs. CC; (c) PC vs. CC. The
   green dots in the plots illustrated that the differential metabolites
   were significant and down-regulated, while the red dots illustrated
   that the differential metabolites were significant but up-regulated,
   and the black dots illustrated that the metabolites could be detected
   in samples but did not have any significant difference.

   From [77]Figure 3a, it can be seen that there were 68 metabolites with
   significantly different content between the PB and PC groups. Compared
   with PB, PC contained 34 metabolites with an upregulated state and 34
   metabolites with a downregulated state. A total of 51 metabolites
   significantly different in content between the PB and CC groups
   ([78]Figure 3b), among which 21 metabolites increased and 30
   metabolites decreased in content in the CC group compared to the PB
   group. A total of 57 metabolites were recognized as significantly
   different between PC and CC ([79]Figure 3c). Among these metabolites,
   there were 36 metabolites with significantly higher content in PC than
   CC.

   Based on the differential metabolites that had significantly different
   content in the comparisons of PB vs. PC, PB vs. CC, and PC vs. CC, Venn
   diagram analysis was conducted to display the metabolites that existed
   commonly or uniquely in the group. As shown in [80]Figure 4, a total of
   116 metabolites ([81]Table S1) that existed in at least two groups were
   shown to obviously change in content. A further comparison for these
   metabolites’ contents among the three groups was conducted and the
   results are shown in [82]Table 2. Among the 116 differential
   metabolites, there were 53 flavonoids, accounting for nearly 40 percent
   of all identified flavonoids, and these showed a great variation of
   content during growth in different substrates. Compared to the PB
   group, there were 21 out of 32 flavonoids down-regulated and 17 out of
   21 flavonoids down-regulated in PC and CC, respectively. Besides
   flavonoids, 22 out of 72 lipids altered significantly, and compared
   with the PC group, the contents of most lipids (13 out of 16) decreased
   in PB, and nearly whole lipids (13 out of 14) decreased in PC. The
   distribution of all metabolites that differed significantly in content
   showed that flavonoids and lipids were the metabolites that fluctuated
   most.

Figure 4.

   [83]Figure 4
   [84]Open in a new tab

   Venn diagram illustrating shared or unique metabolites that differed
   significantly in terms of content among the different comparison.

Table 2.

   The number of metabolites that had significantly different contents in
   D. officinale cultured in PB, PC, and CC.
   Group PB vs. PC PB vs. CC PC vs. CC
   Up Down Up Down Up Down
   Flavonoids (53/135) 11 21 4 17 12 11
   Phenolic acid (16/75) 3 5 3 6 5 3
   Lipids (22/72) 13 3 4 2 13 1
   Amino acids and derivatives (3/64) 0 0 3 0 0 0
   Alkaloids (7/41) 4 0 0 3 4 0
   Nucleotides and derivatives (3/37) 1 1 2 0 0 0
   Organic acids (3/34) 1 0 2 1 1 1
   Lignans and Coumarins (6/17) 0 3 2 1 0 4
   Quinones (1/2) 0 1 0 0 0 1
   Others (2/49) 1 0 1 0 1 0
   Significantly differential metabolites 34 34 21 30 36 21
   Total metabolites that differed significantly in content (116) 68 51 57
   [85]Open in a new tab

   Note: the numbers in parentheses separated by (/) indicate the number
   of metabolites in that class that had significantly different contents
   across substrates metabolomic data. Up/Down represents the changed
   trend of the metabolite content in the comparison between the latter
   group and the former one.

   Hierarchical cluster analysis (HCA) was conducted to display the trend
   of metabolic variation based on the differential metabolites of
   flavonoids and lipids, respectively. Heatmaps of flavonoids and lipids
   are shown in [86]Figure 5. [87]Figure 5a shows that more than two
   thirds of the flavonoids that differed significantly in content
   exhibited the highest contents in the group of PB when compared with PC
   and CC. Compared to the group of PC, the samples in the CC group
   possess a higher content for a majority of flavonoids. Similarly, from
   [88]Figure 5b, we can see that a majority of lipids preferred
   accumulating in PB compared with PC and CC, while almost all of the
   lipids with significant variation in content had the lowest contents in
   CC. The above comparisons might suggest that the cultivation substrate
   of PB is more suitable for D. officinale to accumulate flavonoids and
   lipids. That is to say, PB may be the optimal cultivation substrate for
   D. officinale, with the highest biomass yield among PB, CC, and PC.

Figure 5.

   [89]Figure 5
   [90]Open in a new tab

   Heatmaps of hierarchical cluster analysis (HCA): (a) flavonoids; (b):
   lipids. The abscissa is used to display the names of samples, and the
   ordinate on the right is used to display the names of metabolites. The
   deeper the red color, the higher the content of the metabolites; the
   deeper the blue color, the lower the content of the metabolites.

2.6. Pathway Enrichment Analysis of Differential Metabolites

   A total of 203 metabolites in 543 identified metabolites were annotated
   in the KEGG compound database, of which only 17 metabolites showed
   significant differences in content in the comparisons of PB vs. PC,
   together with PB vs. CC and PC vs. CC. Furthermore, ten more
   differential metabolites, which have been detected in the samples but
   were not annotated on the KEGG pathway, were added to KEGG pathways
   according to their similar properties, such as chemical structures with
   known KEGG compounds. The results of the KEGG pathway enrichment
   analysis are shown in [91]Figure 6.

Figure 6.

   [92]Figure 6
   [93]Open in a new tab

   KEGG enrichment maps of differential metabolites: (a) PB vs. PC; (b) PB
   vs. CC; (c) PC vs. CC. The abscissa represents the enrichment factor of
   the pathway and the ordinate shows the names of pathways. The color of
   the dot represents the p-value, and the deeper the red of the dot, the
   stronger the enrichment effects. The size of points represents the
   number of metabolites enriched in the pathways.

   The differential metabolites in the comparisons of PB vs. PC were
   annotated in 24 pathways, among which flavonoid biosynthesis and
   glycerophospholipid metabolism changed obviously. Similarly, the
   differential metabolites between PB and CC were annotated in 13
   pathways, and the flavonoid biosynthesis and glycerolipid metabolism
   altered prominently. The differential metabolites between PC and CC
   were involved in 17 pathways, totally. Among them, the pathways,
   including flavonoid biosynthesis, flavone and flavonol biosynthesis,
   and glycerophospholipid metabolism, changed significantly (p < 0.01).
   In the comparisons between PB and PC, PB and PC, and PC and CC, only
   flavonoid biosynthesis was significantly distinct.

3. Discussion

   Dendrobium officinale is a medicinal and edible plant of great economic
   value, exhibiting many physiological functions. However, most D.
   officinale on the market is planted artificially, and thus their
   quality is uneven. Here, we employed a widely targeted metabolomics
   technique to explore the metabolites with small molecules and evaluate
   the quality of D. officinale based on the contents of these
   metabolites.

   In this study, the obvious separation among the three groups in PCA
   results indicated that the metabolism of D. officinale collected from
   different cultivation substrates differ a lot. The three cultivation
   substrates (PB, PC, and CC) have different physical properties (e.g.,
   water-holding capacity) and chemical properties (e.g., content of
   nitrogen, potassium, and phosphorus). Therefore, it may be said that
   the difference of these properties in different cultivation substrates
   could affect the metabolism of D. officinale greatly. It has been
   reported [[94]26] that the physical and chemical properties of soil
   have a definite effect on the contents of the primary medicinal
   components of D. officinale under four cultivation modes.

   The metabolomic profiling results suggested that flavonoid compounds
   had maximum counts of metabolites in D. officinale stems. Flavonoids
   were also the metabolites with the greatest variation in the three
   cultivation substrates. For this reason, the flavonoids could be used
   to evaluate the quality of D. officinale. Flavonoids were also used to
   evaluate D. officinale quality based on medicinal effectiveness in a
   previous study [[95]27], because they perform many physiological
   functions. The results of differential metabolites analysis of PB vs.
   PC, PB vs. CC, and PC vs. CC (detailed information of significantly
   differential flavonoids is shown in [96]Tables S2–S4) suggested that PB
   was the optimal cultivation substrate. This supports a recommendation
   for employing certain cultivation substrates to increase the contents
   of some components. Our result is the same as a previous study [[97]24]
   conducted using a widely targeted metabolomic analysis to reveal
   metabolic alterations of Epimedium pubescens leaves at different growth
   phases.

   It should be noted that PB was only the optimal cultivation substrate
   when compared against PC and CC, and we only grew D. officinale in
   Haikou, China. Environmental factors have a great influence on D.
   officinale quality, so findings may not be generalizable.

4. Materials and Methods

4.1. Plant Samples and Chemicals

   A base, located in Haikou Hainan, China, was suitable to plant D.
   officinale because of its mild climate and abundant rainfall.
   Five-month-old D. officinale plantlets from Zhejiang province were
   transplanted in November 2017 to a porous mesh basin with a size of
   14.2 cm × 10.2 cm × 10.3 cm (length × width × height) in a greenhouse
   equipped with sunshade and sprinkling facilities. Each D. officinale
   plantlet was planted in a basin loaded with cultivation substrate to a
   volume of two thirds. The culture substrates included pine bark (PB),
   coconut coir (CC), and a pine bark: coconut coir 1:1 mix (PC), all of
   which had been sterilized and soaked in water. Each treatment
   replicated six units and every unit contained nine basins treated the
   same way. The plants were watered regularly, and the growth environment
   was monitored over time. D. officinale plantlets were shaded properly,
   cooled, and ventilated during summer and autumn. D. officinale stems
   were harvested almost two years later. The samples were packaged with a
   valve bag and then frozen using liquid nitrogen and finally place in
   storage at −80 °C.

   All reagents were HPLC-grade. Methanol, acetonitrile, and ethanol were
   purchased from Merck (Darmstadt, Germany) and standard compounds of
   dimethyl sulfoxide (DMSO) were obtained from BioBioPha (Yunnan, China).

4.2. Sample Preparation and Extraction

   Sample preparation and extraction were performed following reference
   [[98]28]. Samples were freeze-dried in a lyophilizer (Scientz-100F, Xi
   Yu equipment Co. Ltd., Shanghai, China). Then, stems were ground with a
   zirconia bead for 1.5 min at 30 Hz using a mixer mill (MM 400, Retsch
   Technology, Haan, Germany). A total of 100 mg of powder was weighed and
   dissolved in aqueous methanol (0.6 mL, 70%) overnight at 4 °C. Then,
   the solution was whirled six times to improve the extraction rate.
   After centrifugation at 10,000× g for 10 min, the supernate was gained
   and filtrated employing the microfiltration membrane (SCAA-104, 0.22 μm
   pore size; ANPEL, Shanghai, China). The filtration was packed in a
   sample bottle for UPLC-MS/MS analysis. In addition, quality control
   (QC) samples were prepared by mixing three kinds of extractions from
   three different groups with the same weight.

4.3. UPLC-MS/MS Analysis

   The prepared sample extraction was analyzed by means of a
   UPLC–ESI-MS/MS system, including a UPLC system (Shim-pack UFLC SHIMADZU
   CBM30A, Shimadzu, Kyoto, Japan) and a ESI-MS/MS system (Applied
   Biosystems 4500 Q TRAP, AB SCIEX, Foster City, CA, USA). The separation
   operation was implemented using a 1.8 μm C18 column (100 mm × 2.1 mm
   ID) (ACQUITY UPLC HSS T3, Waters, Wexford, Ireland) equilibrated at 40
   °C. The mobile phases A and B were prepared by mixing ultrapure water
   and acetonitrile with 0.04% acetic acid, respectively. Chromatographic
   separation was completed by way of a gradient elution program: First,
   the mobile phase B rose from 5% to 95% within 10 min linearly. It was
   maintained at the level of 95% from 10 to 11 min, dropped to 5% from 11
   to 11.1 min and held steady until 14 min. The temperature of the column
   oven was 40 °C, the injection volume was 4 μL, and the flow rate was
   0.35 mL/min. The effluent was alternatively connected to an ESI–triple
   quadrupole-linear ion trap (QTRAP)-MS.

   Mass-spectrometry detection was conducted on a triple quadrupole-linear
   ion trap mass spectrometer, API 4500 Q TRAP UPLC-MS/MS system equipped
   with an ESI Turbo Ion-Spray interface. The main parameters were as
   follows: The temperature of electrospray ionization was set to 550 °C;
   the voltages of ion sprays were 5500 V and −4500 V in positive and
   negative ion mode, respectively; the curtain gas (CUR) was 30 psi, and
   the collision-activated dissociation (CAD) was high [[99]29]. Each ion
   pair was scanned at the conditions of the optimum declustering
   potential (DP) and collision energy (CE) in the triple quadruple pole
   (QQQ) system [[100]30].

   In the process of UPLC–ESI-MS/MS analysis, a QC sample was put behind
   each 10 test samples to monitor the stability of the instrument.

4.4. Qualitative and Quantitative Analysis of Metabolites

   On the basis of the self-built database (MetWare database, MWDB
   [[101]30,[102]31]) of Maiteville Biotechnology Co., Ltd. (Wuhan, China)
   and the public secondary mass-spectrometry database (e.g., MassBank,
   KNApSAcK, HMDB [[103]32], MoTo DB and METLIN [[104]31]), qualitative
   analysis of metabolites was carried out. During analysis, interference,
   which came from isotope signals and repeated signals of K^+, Na^+, and
   NH[4]^+ ions, as well as fragment ions derived from the metabolites
   which had higher molecular weight, was excluded by in-house software
   written in Perl.

   The quantification of metabolites was completed by the MRM of triple
   quadrupole mass spectrometry [[105]29]. In MRM mode, the precursor ions
   of the target substances were filtrated using a quadrupole rod to
   exclude the interference by removing the corresponding ions with
   different molecular weights. Next, many fragment ions, which were
   formed via the induced ionization in the collision chamber, were
   filtered by QQQ. A characteristic fragment ion was then selected to
   eliminate the interference of nongoal ions, rendering quantitative
   analysis more accurate and ensuring good repeatability. Finally, data
   gained from mass-spectrum analysis from different samples were used to
   calculate the integral results of the peak area. The retention time of
   the same kind of metabolite from different samples was calibrated by
   integral. The software Analyst 1.6.3 was used to process the
   mass-spectrometry data, and MultiaQuant was employed to complete the
   integration and calibration of chromatographic peaks.

4.5. Data Analysis

   In this study, the data gained from the above stages were highly
   dimensional and needed to be simplified. Thus, PCA and OPLS-DA were
   utilized. Data analysis was conducted mainly using R [[106]33]
   software. PCA analysis was performed by built-in “prcomp” function
   (Version 3.5.0, University of Auckland, Auckland, New Zealand), and
   data were processed by unit variance scaling (UV) to eliminate the
   influence of different measurement units before PCA. OPLS-DA was
   conducted on the raw data by the statistics function “MetaboAnalystR”
   [[107]34] (Version 1.0.1, McGill University, Quebec, Canada). In this
   process, data were analyzed in two groups: experimental group and
   control group, encoded 0 and 1, respectively. They were transformed
   with log[2] function and mean-centered. The VIP value was then
   extracted from the results of OPLS-DA. Finally, metabolites that
   exhibited statistically significant differences between each two groups
   of the three groups were screened by the conditions of VIP ≥1 and
   absolute log[2] (FC) ≥1 [[108]35]. Heatmap was conducted to exhibit the
   trend of the variation of differential metabolites by the pheatmap
   (Version 1.0.12) package of R software.

   Authenticated metabolites were annotated based on the KEGG compound
   database [[109]36] and annotated metabolites were then matched to the
   KEGG pathway database. Pathways that significantly regulated
   metabolites mapped to were fed into a metabolite sets enrichment
   analysis (MSEA), and their significance was calculated by the p-values
   from hypergeometric tests.

5. Conclusions

   Quantitative differences were explored among D. officinale stems from
   three different substrates. The UPLC-MS/MS-based widely targeted
   metabolomics method was conducted to investigate the variation of the
   extracted metabolites. The results showed that the chemical
   constitution of Dendrobium officinale cultured in three substrates were
   equal, but the contents were significantly different from one class to
   another. Data were analyzed by multivariate statistical methods
   containing PCA and OPLS-DA. The results illustrated that 68, 51, and 56
   differential metabolites were screened in PB vs. PC, PB vs. CC, and PC
   vs. CC, respectively. Further comparison with the HCA analysis for the
   116 significantly differential metabolites revealed that flavonoid-type
   compounds were the main bioactive metabolites that were regulated
   obviously due to the variation of cultivation substrates, and the
   heatmap suggested that most flavonoids preferred to accumulate in the
   PB group. Consequently, it could be concluded that PB was the best
   cultivation substrate to improve the contents of flavonoids in this
   study. Furthermore, it is indicated that only “flavonoid biosynthesis”
   was significantly affected by the fluctuated metabolites in the KEGG
   pathway enrichment analysis.

Supplementary Materials

   The following are available online at
   [110]https://www.mdpi.com/2218-1989/10/8/325/s1, Figure S1: The total
   ion chromatograms (TIC) of Dendrobium officinale QC samples: (a)
   positive ESI mode; (b) negative ESI mode., Figure S2: Plots of
   permutation tests on OPLS-DA models., Table S1: 116 differential
   metabolites screened by the group of PB vs PC, PB vs CC and PC vs CC,
   Table S2: Differential metabolites between PB and PC., Table S3:
   Differential metabolites between PB and CC., Table S4: Differential
   metabolites between PC and CC.
   [111]Click here for additional data file.^ (699.3KB, pdf)

Author Contributions

   Conceptualization, Y.-H.Y., H.C.; methodology, W.Z., Y.-H.Y.; software,
   S.-M.Z., H.-D.Y.; formal analysis, Q.Z., W.C. (Wenxue Chen);
   writing—original draft preparation, Y.-H.Y., S.-M.Z and H.-D.Y.;
   writing—review and editing, Y.-H.Y., S.-M.Z.; funding acquisition, W.C.
   (Weijun Chen) and H.C. All authors have read and agreed to the
   published version of the manuscript.

Funding

   This work is funded by the National Natural Science Foundation of China
   (grants no. 21705162), the Program of Hainan Association for Science
   and Technology Plans to Youth R and D Innovation (QCXM201924), and the
   Scientific Research Funds of Hainan University, China (KYQD (ZR)-1844).

Conflicts of Interest

   The authors declare no conflict of interest.

References