Abstract

   The metabolic profiles of tobacco leaves of two differential Chinese
   cultivars from different growing regions were analysed using gas
   chromatography–mass spectrometry (GC–MS). The results of principal
   component analysis, partial least-squares discriminant analysis and
   hierarchical cluster analysis showed significant differences in
   metabolome among three groups, identified 24 differential metabolites,
   and analysed the metabolic pathway in which the metabolites were
   involved. Among them, 13 metabolites were associated with geographical
   regions, including seven organic and fatty acids, four carbohydrates
   and two secondary metabolites. Four amino acids and two monosaccharides
   were associated with cultivars and the remaining five metabolites were
   associated with both. The relationships among the differential
   metabolites and the distinct characteristics of environment and
   cultivar were further discussed. In addition, correlation analysis
   indicated that most of the differential carbohydrates were negatively
   correlated with the differential amino acids and organic acids. Taken
   together, this study demonstrates the metabolite differences between
   two cultivars in different regions, and highlights the effect of
   environment and cultivar on tobacco leaf metabolism.

   Keywords: metabolic profiles, GC–MS, flue-cured tobacco leaves, growing
   regions, cultivars

1. Introduction

   Plant metabolomics is a comprehensive method used to estimate global
   metabolite changes [[45]1–[46]3], and has been applied in several plant
   research fields, such as biomarker selection [[47]4,[48]5], gene
   function research [[49]6], assessment of genetically modified plants
   [[50]7], identification of differential metabolites and pathways among
   different cultivars [[51]8–[52]10], developmental stages [[53]11] and
   environments [[54]1,[55]2]. The common analytical platforms in plant
   metabolomics studies are gas chromatography–mass spectrometry (GC–MS),
   liquid chromatography–mass spectrometry (LC–MS), capillary
   electrophoresis–mass spectrometry (CE–MS) and nuclear magnetic
   resonance (NMR) [[56]12]. Today, GC–MS has become one of the most
   applicable methods in plant metabolomics, and has good separation
   efficiency for polar small molecular metabolites and identification via
   various mass spectral libraries [[57]12].

   Numerous studies have shown that growing regions and climatic
   conditions have a significant impact on metabolites in tobacco leaves.
   Flue-cured tobacco leaves grown in China are divided into clear
   flavour, full flavour and middle flavour types, and tobacco leaves from
   Yunnan, Henan and Guizhou Provinces have the typical characteristics of
   these three above flavour types, respectively [[58]13]. The remarkable
   metabolite differences in tobacco leaves from the three geographical
   origins have been analysed by using GC–MS, LC–MS and CE–MS alone or in
   combination, and differential metabolites were also identified
   [[59]1,[60]2,[61]14–[62]16]. Hubei Province is one of the important
   planting areas for Chinese tobacco, and is located in central China,
   between Guizhou Province and Henan Province. Similar to tobacco leaves
   from Guizhou Province, leaves from Hubei Province are also classified
   as middle flavour type [[63]13]. However, little is known about the
   metabolic profile of flue-cured tobacco leaves from Hubei Province.
   Therefore, we chose Hubei and Yunnan as the growing origins in our
   study.

   In addition to the geographical origin factor, the quality and
   metabolic profiles of tobacco leaves are also obviously influenced by
   the cultivar factor. Tobacco is an important cash crop distributed
   widely in China, and lots of flue-cured tobacco cultivars have been
   bred to adapt to various geographical growing locations with great
   environmental differences [[64]17]. The chemical components, such as
   alkaloids and aroma substances, were diffusely analysed and compared
   among different tobacco cultivars, and significant differences were
   observed [[65]18–[66]21]. In our study, two special and differential
   Chinese cultivars were selected and analysed. Cuibi-1, which was
   selected from the Sanming tobacco planting area in Fujian Province, has
   an elegant, graceful and pure aroma, and is susceptible to black shank
   and bacterial wilt infections [[67]22]. Zhongyan-100 is a cultivar that
   is highly resistant to brown spot and black shank, and has favourable
   chemical composition and good adaptability [[68]23]. In this study, we
   analysed the metabolic profile of these two Chinese tobacco middle
   leaves in different regions using GC–MS, and identified their
   differential metabolites via a series of multivariate analyses. Their
   involved metabolic pathways and correlation of differential metabolites
   were also analysed.

2. Experimental

2.1. Plant materials

   Two Chinese commercial flue-cure tobacco cultivars (Cuibi-1 and
   Zhongyan-100) were chosen and used in this study. Cuibi-1 cultivars
   were planted in Yuxi City, Yunnan Province (YXC) and Xuanen County,
   Hubei Province (XNC), respectively, while Zhongyan-100, which is
   unsuitable growing in Yunnan due to climatic reasons, was only planted
   in Xuanen County (XNZ) ([69]table 1). The annual average temperature,
   annual precipitation, annual sun exposure time and altitude of the
   planting region in Xuanen County were 15.8°C, 1490 mm, 1136 h and
   1000 m.a.s.l., respectively, while those in Yuxi City were 16°C,
   900 mm, 2200 h and 1800 m.s.l., respectively. The data were obtained
   from China Meteorological Administration. Seeds were germinated and
   planted in the greenhouse. Then 100 seedlings of each cultivar with 7–9
   true leaves were transplanted into an agricultural field with 120 cm
   between rows and 60 cm between plants within rows. The field design was
   completely random. Water, fertilizer and pesticides were applied, as
   and when required.

Table 1.

   The information for plant material sampling.
   index cultivar growing site sampling leaf position sampling date
   XNC Cuibi-1 Xuanen County, Hubei Province, China middle part (11–13) 12
   August 2012
   YXC Cuibi-1 Yuxi City, Yunnan Province, China middle part (11–13) 1
   August 2012
   XNZ Zhongyan-100 Xuanen County, Hubei Province, China middle part
   (11–13) 7 August 2012
   [70]Open in a new tab

   Plant materials without any insects and mechanical damage were sampled
   from middle leaf positions (no. 11–13) at the mature leaf stage. Leaves
   from five tobacco plants were collected as a replicate, and six
   independent replicates were taken for analysis. After harvest, the leaf
   samples were immediately frozen in liquid nitrogen, lyophilized to
   dryness, and ground to a fine powder for subsequent analyses in a
   laboratory of the Zhengzhou Tobacco Research Institute.

2.2. Sample preparation

   Metabolites were extracted from lyophilized tobacco samples and
   analysed as previously described with minor modification [[71]1]. A
   20 mg sample of leaf powder was added to 1.0 ml of extraction solution,
   which consisted of methanol, chloroform and ddH[2]O with a ratio of
   5/2/2 (v/v/v), and 0.2 ml of ribitol (40 µg ml^–1) as the internal
   standard, respectively. The solvent mixture was extracted by an
   ultrasonic method for 40 min at room temperature. After vortexing for
   30 s, the extraction solution was centrifuged at 12 000 r.p.m. for
   20 min at room temperature. Subsequently, 400 µl of the supernatant was
   transferred to a 2 ml Eppendorf tube and dried by nitrogen evaporator.
   For derivatization, 50 µl of methoxyamine hydrochloride dissolved in
   pyridine (20 mg ml^–1) was added, and vortexed for 1 min, and then
   incubated for 90 min at 37°C. For silylation, 80 µl of
   N-methyl-N-(trimethylsilyl)trifluoroacetamide (MSTFA) was added, and
   incubated for 30 min at 37°C, and then vortexed for 30 s. The solution
   was left standing for 1 h, and centrifuged at 12 000 r.p.m. for 10 min,
   and then 80 µl of the supernatant was transferred to a trace sample
   vial for further analysis.

2.3. Gas chromatography–mass spectrometry analysis

   GC–MS analysis of the metabolites in the tobacco leaves was carried out
   using an Agilent 7890A gas chromatograph (GC) interfaced to an Agilent
   5975C mass-selective detector (Agilent, USA), controlled by an Agilent
   G1701EA GC-MSD ChemStation. Samples were tested in random order with
   quality control (QC) samples being inserted with every six samples in
   the running sequence. QC samples were mixed with the same amount of
   each sample. Chromatographic separations were achieved in an DB-5
   (30 m × 0.250 mm, 0.25 µm film thickness) capillary column (Agilent
   Technologies Inc., USA). The column flow, using helium as the carrier
   gas, was held constant at 1.0 ml min^−1. The column temperature was set
   to 70°C for 4 min, and programmed to 310°C at 5°C min^−1 and kept at
   this temperature for 10 min. The injector temperature was 290°C, and
   the injector was set in split mode (10 : 1) with an injection volume of
   1 μl. The interface temperature was 230°C and the ion source
   temperature was 280°C. Mass spectra were recorded at 70 eV, and used
   both full scan and selected ion monitoring (SIM) mode with scanning
   from 40 to 510 amu. Ions were acquired with a solvent cut time of
   8.0 min [[72]1].

2.4. Data processing and statistical analysis

   Quantitative analysis, which used a pseudo-targeted method, was carried
   out in SIM mode [[73]12,[74]24]. The components eluting within the
   total ion chromatogram were extracted in the Agilent MSD ChemStation.
   Then the metabolites were identified in two principal ways: (1) the
   identification of metabolites was based on mass spectral matching with
   NIST and Fiehn mass spectra libraries and (2) 49 commercial standards
   were used to confirm unambiguously (electronic supplementary material,
   table S1). Subsequently, the peaks detected in less than 80% of the
   samples were discarded according to the ‘80% rule' [[75]1], and the
   peak area of metabolites was normalized to the internal standard for
   further data analysis. Principal component analysis (PCA) was performed
   by Simca-p 13.0 Demo software (Umetrics, Sweden) with Pareto scaling to
   understand the relationships among samples. To investigate the
   differences in metabolite levels among tobacco leaf samples, partial
   least-squares discriminant analysis (PLS-DA) was then carried out by
   Simca-p 13.0 Demo software with Pareto scaling [[76]1]. These
   metabolites, which had variable importance in the projection (VIP)
   value > 1 (PLS-DA) and p < 0.05 (Mann–Whitney U test by SPSS18.0
   software), were identified as significant differential metabolites
   [[77]2]. TIGR MeV software (Version 4.1) was used to cluster the
   significant differences in metabolite levels [[78]20]. Metabolic
   pathway enrichment analysis was performed to confirm the important
   pathways related to metabolic phenotype according to the KEGG website
   ([79]www.genome.jp/kegg/). The correlation between differential
   metabolites was visualized using Cytoscape v. 3.5.1 software.

3. Results

3.1. Pseudo-targeted gas chromatography–mass spectrometry selected ion
monitoring analysis

   To establish the pseudo-targeted GC–MS SIM method, a QC sample was
   analysed with full scan mode, and the Agilent MSD ChemStation was used
   to output the raw data. The distinction of overlapped peaks and
   selection of characteristic ions for metabolites were performed by
   Automatic Mass Spectral Deconvolution and Identification System (AMDIS)
   software. The data were divided into 27 groups based on retention time
   (RT) and characteristic ions. The total ion chromatograms (TICs) from
   the full scan and pseudo-targeted SIM mode of the QC sample is shown in
   [80]figure 1. The pseudo-targeted GC–MS SIM method established with 27
   groups and 313 quantitative ions was used to investigate the metabolic
   difference between fresh flue-cured tobacco middle leaves from
   different cultivars and different geographical origins.

Figure 1.

   [81]Figure 1.
   [82]Open in a new tab

   GC–MS total ion chromatogram (TIC) of tobacco samples. (a,b) QC
   samples; (c,d) XNC samples; (e) and (f) YXC samples; (g,h) XNZ samples.
   (a,c,e,g) Were analysed in full scan mode, whereas (b,d,f,h) were
   analysed in SIM mode.

   The compound identification was performed based on matching with mass
   spectra libraries and commercial standards. Among the 313 peaks, 87
   compounds were identified totally, and 49 compounds were confirmed by
   standards (electronic supplementary material, table S1). Among the
   identified compounds, the primary metabolites, such as amino acids,
   sugars, and organic and fatty acids, were abundant in samples. Some
   kinds of secondary metabolites influencing tobacco leaf quality, such
   as alkaloids, chlorogenic acid and sterols were also detected.

   To inspect the reproducibility of the pseudo-targeted SIM method,
   relative standard deviation (RSD) of all 313 peaks were calculated
   based on their peak area in the QC samples. The results showed that 82%
   of the peaks, which were up to 98.62% of the total peak area, had an
   RSD of less than 30% ([83]figure 2), indicating that this method is
   reproducible and could be used in the subsequent experiment. It is hard
   to tell the differences among fresh tobacco leaf samples from different
   cultivars and geographical origins based on chromatograms ([84]figure
   1), so a multivariate statistical analysis was important for further
   data excavation.

Figure 2.

   [85]Figure 2.
   [86]Open in a new tab

   RSD distribution for all metabolites in three QC samples. The black
   columns represent the percentage of the peak number within the
   specified RSD% range, respectively. The line indicates the cumulative
   percentage of the peak area within the specified RSD% range.

3.2. Principal component analysis

   To get an overview of the effect of cultivar and growing site on
   tobacco leaf metabolomes, the metabolic data for Cuibi-1 and
   Zhongyan-100 middle leaves from two locations (Xuanen County and Yuxi
   City) were analysed using PCA analysis, which is an unsupervised
   multivariate analysis method. As shown in [87]figure 3a, five principal
   components of the PCA plot describe 94.4% of the total variance
   information, while the first principal component is as high as 66.6%.
   Moreover, all of the QC samples were clustered together and located at
   the centre of the PCA score plot ([88]figure 3b), indicating good
   analytical stability and reproducibility of this experiment. In the PCA
   model, three groups of tobacco leaf samples were well separated along
   the first principal component, and XNC samples were dispersed between
   the YXC and XNZ samples ([89]figure 3b), indicating that both cultivar
   and growing site had great impact on the metabolic profile of tobacco
   leaves.

Figure 3.

   Figure 3.
   [90]Open in a new tab

   Summary of the fit (a) and score plot (b) of PCA of tobacco samples
   from different cultivars and geographical origins. Green and blue
   columns in (a) represent the cumulative R2X and Q2, respectively.
   Yellow, red, green, blue circles in (b) represent the QC, XNZ, XNC, YXC
   samples, respectively.

3.3. Partial least-squares discriminant analysis

   To further investigate the differences among tobacco leaf samples, a
   PLS-DA model was established ([91]figure 4). Three groups of tobacco
   leaf samples were clearly separated from the score plot ([92]figure
   4a), and the distribution trend is similar to the PCA results. In this
   PLS-DA model, three components described 98.2% of the variation and
   predicted 95.3% according to cross-validation; R2Y and Q2 of the first
   two components were predominant among them, and the values were 92.3%
   and 89.0%, respectively. The results showed that the PLS-DA model had
   good robustness properties, and the selection of the first two
   components was large enough to examine this dataset. A permutation test
   could be used to evaluate the possible overfitting of the PLS-DA model,
   and a properly fitted model was identified as having values of
   R2-intercept less than 0.4 and a Q2-intercept less than 0.05 [[93]25].
   In this study, 200 permutation tests were performed; the R2-intercept
   was 0.249 and Q2-intercept was −0.327 ([94]figure 4b), indicating that
   the PLS-DA model had no overfitting and was credible.

Figure 4.

   Figure 4.
   [95]Open in a new tab

   Score plot (a) and cross-validation plot (b) of the PLS-DA of tobacco
   samples from different cultivars and geographical origins. Red, green,
   blue circles in (a) represent the XNZ, XNC, YXC samples, respectively.
   Three components were fit by autofit model, and the parameters of the
   model were as follows: R2Y(cum) = 0.982, Q2(cum) = 0.953. (b)
   Cross-validation plot of PLS-DA mode with 200 permutation tests. The
   intercepts of R2 and Q2 were 0.249 and −0.327, respectively.

   The identification of differential metabolites between each two samples
   was performed using VIP values and further confirmed by the
   non-parametric Mann–Whitney U test. Among the three groups, a total of
   27 differential metabolites with VIP greater than 1 and p < 0.05 were
   screened out ([96]figure 5). Of these, 24 metabolites were verified via
   commercial standards and mass spectra libraries. A Venn plot showed
   that 18 metabolites in YXC samples and 11 metabolites in XNZ samples
   were significantly different from XNC samples, and 21 differential
   metabolites were observed between YXC samples and XNZ samples
   ([97]figure 6). In addition, five metabolites, sucrose, fructose,
   cellobiose, inositol and propanoic acid, were significantly different
   among the three pairwise comparisons. The relative concentration, VIP
   value, p-value and fold change of differential metabolites are shown in
   supporting information in electronic supplementary material, table S2.

Figure 5.

   [98]Figure 5.
   [99]Open in a new tab

   PLS-DA analysis of tobacco samples from different cultivars and
   geographical origins. (a) Score plot of XNC and YXC. (b) Loading plot
   of XNC and YXC. (c) Score plot of XNC and XNZ. (d) Loading plot of XNC
   and XNZ. (e) Score plot of YXC and XNZ. (f) Loading plot of YXC and
   XNZ. Red circles in loading plots represent the compounds whose VIP
   value was more than 1.

Figure 6.

   Figure 6.
   [100]Open in a new tab

   Venn diagram of the differential metabolites in a pair of samples.

3.4. Hierarchical cluster analysis

   The hierarchical cluster analysis (HCA) of identified metabolites was
   performed to visualize the clustering of 24 differential compounds
   ([101]figure 7). To determine the relationships and trends of the
   differential metabolites among the three groups of samples, the heat
   map was divided into four groups. The metabolites in group A had the
   highest number of the XNC samples and the lowest number of YXC samples.
   Significant differences in metabolite levels were observed between the
   YXC samples and two other samples, whereas no significant difference
   was found between the XNC samples and the XNZ samples. Group B is the
   largest group in [102]figure 7, and is shown as a cluster of 10
   differential metabolites, including four amino acids (proline, aspartic
   acid, glycine and glutamic acid), four organic acids (propanoic acid,
   malic acid, quinic acid and shikimic acid), and two disaccharides
   (cellobiose and sucrose). The levels of all the metabolites in group B
   were the highest in the XNZ samples, and most of these were
   significantly different from the two other samples, except for quinic
   acid and malic acid. Different trends of the metabolites in group B
   were observed between the XNC samples and the YXC samples. The levels
   of three amino acids (proline, aspartic acid and glycine) in the XNC
   samples were less than those in the YXC samples, but there were no
   notable differences between them. However, the levels of other
   metabolites in group B in the XNC samples were significantly higher
   than those in the YXC samples (electronic supplementary material, table
   S2). In group C, the metabolites included three monosaccharides
   (fructose, mannose and glucose), and their levels were the highest in
   the YXC samples and lowest in the XNZ samples, while significant
   differences were found between the XNZ samples and two other samples.
   Group D mainly contained some sugars (e.g. galactose, inositol, and
   lactulose), organic acid (citric acid) and phenolics (chlorogenic
   acid). The concentrations of the corresponding metabolites in group D
   were significantly higher in the YXC samples, and no significant
   differences were found between two other samples grown in Xuanen
   County, except for inositol.

Figure 7.

   [103]Figure 7.
   [104]Open in a new tab

   Heat map of the differential metabolites. Green and red reflect the
   relative concentration of the metabolites.

3.5. Metabolic pathway analysis

   To deeply understand the differences in the metabolic networks among
   three samples, all the differential metabolites were submitted to the
   KEGG website for metabolic pathway enrichment analysis. The results
   showed that 21 of the differential metabolites were located in various
   pathways, whereas the other three carbohydrate compounds (lactulose,
   mannopyranose, 2-O-glycerol-galactopyranoside) were not found in any
   pathway. A generalization of differential metabolites among three
   groups of the tobacco leaf samples is seen in a metabolic pathway in
   [105]figure 8. Five primary metabolic pathways (glycolysis, sugar
   metabolism, amino acid metabolism, tricarboxylic acid cycle (TCA) and
   organic acid metabolism), and three secondary metabolic pathways
   (ascorbic acid metabolism, shikimic acid metabolism and alkaloid
   metabolism) were extracted and linked based on the KEGG pathway
   database. The above involved metabolic pathways were mostly similar to
   that observed in flue-cured tobacco among different planting regions
   and climate factors [[106]2].

Figure 8.

   [107]Figure 8.
   [108]Open in a new tab

   Metabolic pathway map of the different metabolites. Green, blue and red
   columns indicate the relative concentration of metabolites in XNC, YXC
   and XNZ, respectively. Asterisks * and ** indicate a significant
   difference between two samples at the 0.05 and 0.01 probability level,
   respectively. The metabolites are abbreviated as follows: G6P,
   glucose-6-phosphate; F6P, fructose-6-phosphate; 3PG,
   3-phosphoglycerate; PEP, phosphoenolpyruvic acid; inositol 3P,
   inositol-3-phosphate; xylulose 5P, xylulose-5-phosphate, ATP, adenosine
   triphosphate; ADP, adenosine diphosphate.

3.6. Correlation analysis

   To investigate the metabolite–metabolite correlation of the
   differential metabolites, the Pearson correlation coefficient values
   were filtered by the threshold (p > 0.65), and a network analysis was
   constructed with 24 differential metabolites and 146 edges ([109]figure
   9). The correlation of the differential metabolites indicated that 77
   were negative and 69 were positive. Most of the amino acids and organic
   acids in the differential metabolites were positively correlated with
   each other, but negatively correlated with the carbohydrates. Citric
   acid and two disaccharides (cellobiose and sucrose) were exceptions.
   The largest number of correlations was 18, and were related to shikimic
   acid, whereas the fewest correlations were related to threonate, which
   only correlated with phosphate. The correlation details among
   differential metabolites are shown in electronic supplementary
   material, table S3.

Figure 9.

   [110]Figure 9.
   [111]Open in a new tab

   Correlation analysis between the identified differential metabolites.
   The dashed lines between metabolites represent negative correlation,
   whereas solid lines represent positive correlation. All correlations in
   the figure reflect Pearson correlation coefficient values above the
   threshold (p > 0.65).

4. Discussion

   GC–MS has been used widely as a metabolomics method to investigate the
   metabolic profiles of plant materials and to identify the significantly
   differential metabolites between samples [[112]4,[113]5]. In the
   current study, significant differences in the metabolic profiles
   between two important Chinese cultivars of tobacco leaves grown in two
   locations were observed using a pseudo-targeted GC–MS SIM method, and
   24 differential metabolites were excavated and confirmed by
   multivariate analysis. Metabolic pathway enrichment and correlation
   analysis of differential metabolites were also conducted.

   The stability and accuracy of detection methods are the prerequisites
   for a successful experiment. It is difficult to obtain too many
   standard compounds in metabolic profiling analysis [[114]25]. To solve
   the problem, QC samples were generally used to evaluate the stability
   and repeatability of the whole analytical process [[115]1,[116]12]. In
   the current study, all of QC samples were clustered together and
   located at the centre of the PCA score plot ([117]figure 3b), and most
   of the peaks had an RSD value less than 30% and represented almost all
   of the total responses ([118]figure 2), which is consistent with the
   results of previous research [[119]12,[120]24]. PLS-DA is a supervised
   method that is frequently used in multivariate analysis. The results of
   permutation test showed that the R2-intercept was 0.249 and the
   Q2-intercept was −0.327, indicating that our model was properly fitted.
   Based on the results, it was shown that the reproducibility and
   precision of this method were acceptable for the metabolomics analysis
   of tobacco leaves.

   Xuanen County was selected as the growing site to investigate the
   metabolic profile of tobacco leaves from Hubei Province, which is an
   important planting area of middle flavour flue-cured tobacco [[121]13].
   Yuxi City in Yunnan Province, which is the largest tobacco planting
   region in China, was also selected. In the present study, 13
   differential metabolites were observed between tobacco leaves from
   Xuanen County and those from Yuxi City, including seven organic and
   fatty acids (malic acid, citric acid, shikimic acid, quinic acid,
   phosphate, butanoic acid and threonate), four carbohydrates (galactose,
   2-O-glycerol-galactopyranoside, mannopyranose and lactulose), and two
   secondary metabolites (nicotine and chlorogenic acid) (figures [122]6
   and [123]7). Among them, the levels of differential acids (except for
   citric acid) in tobacco leaves from Xuanen County were significantly
   higher than in tobacco leaves from Yuxi City, whereas the carbohydrates
   levels were lower. Similar distributed trends of differential
   metabolites were reported between tobacco leaves from Yunnan and
   Guizhou in previous studies [[124]1,[125]2]. It has been reported that
   sufficient sun exposure time and higher altitude can enhance sugar
   content [[126]1,[127]2]. The total sun exposure time is shorter and the
   altitude is lower in Xuanen County than in Yuxi City, which might be
   the probable cause of our results from the two regions. The tobacco
   leaves from Hubei and Guizhou are classified as middle flavour type
   [[128]13,[129]19], but the levels of citric acid, which is an important
   metabolite in the TCA cycle, were different between the two regions.
   This might be due to the differences in environmental conditions
   between the two regions and the tobacco flavour complexity. Nicotine
   and chlorogenic acid, which are closely related to climatic factors,
   are two important secondary metabolites of tobacco leaves, and
   contribute greatly to tobacco flavour and quality
   [[130]1,[131]20,[132]26,[133]27]. Nicotine stimulation plays an
   important role in the physiological effects of smoking
   [[134]1,[135]27]. Numerous studies reported that the nicotine content
   of tobacco leaves could increase with the increased sun exposure time,
   rainfall, nitrogen fertilizer application amount, and so on [[136]28].
   In this study, the nicotine level in the tobacco leaves from Xuanen
   County was lower than the tobacco from Yuxi City, which is consistent
   with the previous report [[137]29]. However, no positive correlation
   was found between nicotine content and sun exposure time in our study.
   The reason for this discrepancy might be that the accumulation of
   nicotine is determined by a combination of many factors, such as
   climate, soil, cultivar and cultivation techniques, rather than one
   factor [[138]1,[139]28]. Chlorogenic acid is the main phenolic in
   tobacco leaves, and contributes to mild, sweet flavour and the scent of
   baking in tobacco products [[140]26]. The enhancement of chlorogenic
   acid was induced by increased altitude and long sun exposure time,
   which is consistent with our results, and might be related to
   near-ultraviolet light and the intensity of visible light
   [[141]30,[142]31]. Taking into account the different environments of
   the two regions, the higher levels of nicotine and lower levels of
   chlorogenic acid in the tobacco from Xuanen County than the tobacco
   from Yuxi City contributed to the flavour characteristics of each.

   In the present study, six differential tobacco leaf metabolites were
   observed between Cuibi-1 and Zhongyan-100, including four amino acids
   (glycine, glutamic acid, aspartic acid and proline) and two
   monosaccharides (glucose and mannose) (figures [143]6 and [144]7). In
   addition, the levels of amino acids in Zhongyan-100 were significantly
   higher than in Cuibi-1, whereas the monosaccharide levels were lower.
   Among them, proline, which synthesized from glutamic acid, is a
   proteinogenic amino acid and is essential for primary metabolism.
   Numerous studies have reported that proline accumulation was found in
   plants under different environmental stresses, and that proline can act
   as a signalling molecule to participate in plant resistance
   [[145]32–[146]34]. Thus, high levels of amino acids in Zhongyan-100
   could partially contribute to the high resistance of this cultivar. As
   an important monosaccharide, glucose has fundamental and multiple
   effects on plant metabolism, and is reported to modulate biosynthesis
   of plant secondary metabolites that are involved in the quality and
   flavour of tobacco leaves [[147]35]. Therefore, high levels of
   monosaccharides in Cuibi-1, especially glucose, could explain the high
   aroma of the cultivar. In addition, the levels of propanoic acid and
   four carbohydrates (sucrose, fructose, cellobiose and inositol) were
   significantly different between the intersection of three samples,
   indicating that these differential metabolites were simultaneously
   influenced by both geographical origin and cultivar.

   The flavour and quality of plant edible organs were principally
   governed by the levels and ratios of sugar and organic acid
   [[148]36,[149]37]. Sugar and organic acid catabolism as the primary
   metabolism in the plants were also involved in the biosynthesis of
   secondary metabolites, including amino acids, vitamins and aroma
   volatiles, which then further influence the flavour and quality
   [[150]35,[151]37]. The results of correlation analysis in our study
   showed that most of the differential carbohydrates were negatively
   correlated with the differential amino acids and organic acids
   ([152]figure 9). This is probably related to the regulation of
   development and metabolism of tobacco leaves. Moreover, citric acid,
   sucrose and cellobiose were found as exceptions for each class, but the
   reason for this was not clear and still needs to be investigated in the
   future.

5. Conclusion

   The metabolic profiling of Cuibi-1 and Zhongyan-100 flue-cured tobacco
   middle leaves in different planting regions was analysed using GC–MS
   with the pseudo-targeted SIM method. Twenty-four differential
   metabolites were identified via multivariate analysis, including 13
   geographical region-related metabolites (malic acid, citric acid,
   shikimic acid, quinic acid, phosphate, butanoic acid, threonate,
   galactose, 2-O-glycerol-galactopyranoside, mannopyranose, lactulose,
   nicotine and chlorogenic acid), six cultivar-related metabolites
   (glycine, glutamic acid, aspartic acid, proline, glucose and mannose)
   and five metabolites (sucrose, fructose, cellobiose, inositol and
   propanoic acid) that are involved in both. The results highlight the
   effect of environment and cultivar on tobacco leaf metabolism.

Supplementary Material

   Supplementary Tables
   [153]rsos180261supp1.rar^ (46.5KB, rar)

Data accessibility

   All data used in this article are present in the article and its
   electronic supplementary material.

Authors' contributions

   J.Y., F.C.L. and H.R.T. conceived and designed the experiments; B.S.,
   K.S.W., and F.Z. performed the experiments; Q.C., Y.L., Y.Z. and X.R.W.
   analysed the data; B.S. and A.H.Z. wrote the paper. All authors gave
   final approval for publication.

Competing interests

   We declare we have no competing interests.

Funding

   This work was supported by National Natural Science Foundation of China
   (31400274), PhD Programs Foundation of Ministry of Education of China
   (2012M510149) and the Science and Technology Foundation of Guizhou
   Province (grant no. [2015]2099).

References