Chinese Journal of Natural Medicines  2017, Vol. 15Issue (8): 631-640  DOI: 10.3724/SP.J.1009.2017.00631

Cite this article as: 

JIA Yan, ZHANG Zheng-Zheng, WEI Yu-Hai, Qin Xue-Mei, LI Zhen-Yu. Metabolomics coupled with similarity analysis advances the eluci-dation of the cold/hot properties of traditional Chinese medicines[J]. Chinese Journal of Natural Medicines, 2017, 15(8): 631-640.

Research funding

This study was supported by the Science and Technology Innovation Team of Shanxi Province (No.2013131015) and Program for the Outstanding Innovative Teams of Higher Learning Institutions of Shanxi (OIT)

Corresponding author

LI Zhen-Yu, Tel/Fax: 86-351-7018379, E-mail:

Article history

Received on: 19-Aug.-2016
Available online: 20 Aug., 2017
Metabolomics coupled with similarity analysis advances the eluci-dation of the cold/hot properties of traditional Chinese medicines
JIA Yan1,2 , ZHANG Zheng-Zheng1,2 , WEI Yu-Hai3 , Qin Xue-Mei1 , LI Zhen-Yu1     
1 Modern Research Center for Traditional Chinese Medicine of Shanxi University, Taiyuan 030006, China;
2 College of Chemistry and Chemical Engineering of Shanxi University, Taiyuan 030006, China;
3 Qinghai Entry-Exit Inspection and Quarantine Bureau, Xining 810000, China
[Abstract]: It recently becomes an important and urgent mission for modern scientific research to identify and explain the theory of traditional Chinese medicine (TCM), which has been utilized in China for more than four millennia. Since few works have been contributed to understanding the TCM theory, the mechanism of actions of drugs with cold/hot properties remains unclear. In the present study, six kinds of typical herbs with cold or hot properties were orally administered into mice, and serum and liver samples were analyzed using an untargeted nuclear magnetic resonance (NMR) based metabolomics approach coupled with similarity analysis. This approach was performed to identify and quantify changes in metabolic pathways to elucidate drug actions on the treated mice. Our results showed that those drugs with same property exerted similar effects on the metabolic alterations in mouse serum and liver samples, while drugs with different property showed different effects. The effects of herbal medicines with cold/hot properties were exerted by regulating the pathways linked to glycometabolism, lipid metabolism, amino acids metabolism and other metabolic pathways. The results elucidated the differences and similarities of drugs with cold/hot properties, providing useful information on the explanation of medicinal properties of these TCMs.
[Key words]: Cold/hot property     Metabolomics     Traditional Chinese medicine     NMR     Similarity analysis    

The theory of medicinal properties is always pivotal to elucidate the characteristics of traditional Chinese medicine (TCM), which is in a range of medical practices utilized in China for more than four millennia. It is recorded in Chinese medical classics "Shennong's Herbal" that there are four properties of TCMs--cold, hot, warm and cool, which can be summed up as cold and hot. The medicinal properties also correspond to philosophical frameworks such as the theory of Yin-Yang and Five Elements, the human body meridian systems, and the Zang Fu theory [1].

The cold or hot property, regarded as the own specific characteristics of each drug, is believed to mainly originate from the reactions of the body to a specific herbal medicine. Generally speaking, herbs with the cold property are deemed to clear away heat, eliminate toxic substances, nourish yin and remedy hot syndromes; in contrast, herbs with the hot property usually dispel cold, warm up the interior, support yang, and thus treat cold syndromes [2]. Many modern scientific researches have been undertaken to elucidate the cold/hot properties of TCMs. One finding has been documented that chemical materials underlie the changes that are correlated with cold/hot properties of herbal drugs, such as the predictive system of the cold/hot property of herbal medicines [3]. Liang et al. have characterized Chinese medicine with cold and hot properties based on molecular network and chemical fragments [1]. Liu et al. have pointed out that the investigation of the distinct abilities of Chinese herbs to regulate neural cell functions appear to be associated with their cold or hot properties [2]. Study of the impact of cold-and hot-natured herbs on organs and tissues have also been performed by testing the central nervous system, prostaglandin, and endocrine system [4-5].

Metabolomics, a technique to study metabolic responses of complex organisms to drug or other stimuli, provides complementary information for exploring whole-organism function by identification of potential biomarkers, which is consistent with holistic view of TCM theory [6]. Nuclear magnetic resonance (NMR) spectroscopy, with the nondestructive nature of the analysis, the robust and reproducible measurements and the minimal preparation requirement, is one of most commonly used means of metabolomics [7]. The analysis of metabolomics based on NMR has been successfully applied in diverse research fields, encompassing disease diagnosis and evaluation [8], pharmacology and toxicity [9], nutritional intervention [10], and environment biology [11].

In the present work, an untargeted NMR-based metabolomics approach was used to map the metabolic profiles in serum and liver of mice treated by six kinds of herbal drugs with cold/hot properties. Differential endogenous metabolites were identified and the changed metabolic pathways were used to explain cold/hot properties of drugs. By combination of similarity analysis, it was expected to provide a basic view on the theory of TCM.

Materials and Methods Plant Materials

Three kinds of cold medicine (CMs), Phellodendri Chinensis Cortex, Coptidis Rhizoma, and Scutellariae Radix, and three kinds of hot medicine (HMs), Aconiti Lateralis Radix Praeparata, Zingiberris Rhizoma, and Cinnamomi Cortex, were purchased from Beijing Tong Ren Tang Group Co., Ltd. (Beijing, China). All the herbal drugs were authenticated by Prof. QIN Xue-Mei and the voucher specimens were deposited in the herbarium of the Modern Research Center for Traditional Chinese Medicine of Shanxi University, Taiyuan, China. For TCMs with typical cold or hot properties, the hot and cold properties were defined as previously described [1].

Analytical grade K2HPO4·3H2O and NaH2PO4·2H2O were obtained from Guangfu Fine Chemical Research Institute (Tianjin, China) and Zhiyuan Chemical Reagent Co., Ltd. (Tianjin, China), respectively. HPLC-grade acetonitrile was obtained from Fisher Scientific Worldwide Co., Ltd. (NJ, USA). Sodium 3-trimethylsilyl [2, 2, 3, 3-d4] propionate (TSP) was procured from Cambridge Isotope Laboratories Inc (Andover, MA, USA). Deuterium Oxide (D2O, 99.9%) was purchased from Norell (Landisville, PA, USA). Phosphate buffer was prepared with K2HPO4 and NaH2PO4 (0.1 mol·L-1, pH 7.4), containing 10% D2O and 0.01% TSP.

Sample preparation of herbal drugs

All the herbal drugs were smashed and soaked in distilled water at room temperature and extracted by boiled hot water(1 : 10, W/V) firstly for 3 h, and then (1 : 8, W/V) 2 h, respectively. The afforded decoction were combined, concentrated in a rotary evaporator, freeze-dried to powder under vacuum, and then stored at -20 ℃ until use.

Chemical analysis of the decoctions of herbal drugs

The freeze-dried decoction powder (about 30 mg) was redissolved in KH2PO4buffer in D2O (adjusted to pH 6.0 by 1 mol·L-1 NaOD) containing 0.05% TSP. The sample was then centrifuged for 10 min at 13 000 r·min-1. The supernatants (600 μL) of all the samples were transferred into 5-mm NMR tubes for NMR analysis. Six replicates were prepared for each drug. Then the 1H NMR spectrum was acquired using the noesygppr1d pulse sequence, which consisted of 64 scans requiring a 2.654s acquisition time with the following parameters: spectral width = 12 345.7 Hz, spectral size = 65 536 points, and a relaxation delay = 1.0 s.

Experimental animals and treatment protocol

Fifty six male Institute of Cancer Research (ICR) mice weighing 18 to 22 g were commercially obtained from Vital River Laboratory Animal Technology Co., Ltd. (Beijing, China) and housed in an environmentally controlled breeding room (temperature: 20 ± 2 ℃, humidity: 60% ± 5%) a week before the start of the experiments. The mice were starved for 12 h before scarifying. Eyeball blood samples were separated using refrigerated centrifugation at 3 500 r·min-1 for 15 min to afford the serum samples. The serum samples were snap-frozen in liquid nitrogen and stored at -80℃ for further analysis. All animal experiments [license number SCKX-2012-0001] were carried out in accordance with the National Guidelines for Experimental Animal Welfare (MOST, China, 2006) at the Center for Animal Experiments, Shanxi University, which has full accreditation from the Association for Assessment and Accreditation of Laboratory Animal Care International. Maximum effort was exerted to minimize animal suffering and the number of animals necessary for the attainment of reliable data.

The mice were weighed and randomly divided into seven groups (n = 8/group): normal control group (NC); three CM groups, including Phellodendri Chinensis Cortex group (C1, 2.5 g·kg-1), Coptidis Rhizoma group (C2, 2.5 g·kg-1), Scutellariae Radix group (C3, 2.5 g·kg-1); and three HM groups, including Aconitilateralis Radix group (H1, 3.75 g·kg-1), Praeparata Zingiberris Rhizoma group (H2, 2.5 g·kg-1), and Cinnamomi Cortex group (H3, 1.25 g·kg-1). The administration doses used in the present study (equivalent to the raw drug) were calculated according to recommended therapeutic doses [12]. The drugs were orally administered (8:00 am) into mice and same volume of distilled water was given to mice in NC group using a feeding atraumatic needle once per day for 7 days. Their body weights were recorded every three days. On the day 7, the animals were sacrificed. Orbital blood and livers were collected for further study. The blood was centrifuged at 4 ℃ and 13 000 r·min-1 for 15 min, and the supernatants were stored at -80 ℃ prior to use. The liver was quickly dissected and washed with ice-cold saline solution, and then frozen in liquid nitrogen before use.

NMR measurements of the serum and liver samples

Serum samples were thawed and then prepared as follows: 450 μL of serum was mixed with 350 μL D2O and the mixture was centrifuged at 4 ℃ and 13 000 r·min-1 for 20 min. Six hundred μL of supernatants were transferred into 5-mm NMR tubes for 1H NMR analysis. Liver samples were thawed prior to use. Approximately 200 milligrams of the liver tissue was excised from mouse's liver and extracted with 1000 μL of Acetonitrile/H2O (1 : 1, V/V) via the ultrasonic cell homogenizer. After centrifugation at 4 ℃ and 3 500 r·min-1 for 15 min, the supernatants were transferred into 5-mL eppendorf (EP) collection tubes. After dried under a gentle steam of nitrogen, the sample was redissolved with 700 μL of phosphate buffer containing 0.01% TSP as the chemical shift reference (δ 0.00 ppm). The mixture was centrifuged at 4 ℃ and 3 500 r·min-1 for 15 min, and the supernatants were transferred into 5-mm NMR tubes for NMR analysis.

The one and two dimensional NMR spectra of the serum and liver samples were acquired on a Bruker 600 MHz AVANCE Ⅲ NMR spectrometer (Bruker, BioSpin, Germany) equipped with a Bruker 5-mm PA BBO probe, operating at 600.13 MHz 1H frequency and a temperature of 25 ℃. Serum samples were analyzed using Carr-Purcell-Meibom-Gill (CPMG) spin-echo pulse sequence to attenuate broad signals from proteins and lipoproteins due to their long transverse relation time, while liver samples were analyzed via noesygppr1d sequence. Each 1H NMR spectrum of serum samples was consisted of 64 scans with the following parameters: spectral width = 12 019.2 Hz, spectral size = 65 536, pulse width = 14 μs, and relaxation delay = 5.0 s; Each 1H NMR spectrum of liver was consisted of 64 scans with the following parameters: spectral width = 12 345.7 Hz, spectral size = 65 536, PW = 14 μs, and relaxation delay = 5.0 s. 1H-1H-correlated spectroscopy (COSY) was acquired using the cosygpprqf pulse sequence, which was consisted of 25 scans with 1.5 s RD and 6 602.1 and 6 601.5 Hz spectral widths in both dimensions. The heteronuclear single quantum coherence (HSQC) spectrum was obtained using the hmqcgpqf sequences with 110 scans. All the HSQC spectra were obtained with 1.5 s RD using spectral widths of 6 602.1 Hz in the 1H dimension and 36 219.4 Hz in the 13C dimension.

Data analysis

The 1D and 2D dimensional spectra were processed using MestReNova (version 8.0.1, Mestrelab Research, Santiago de Compostella, Spain), and all the spectra were manually phased and baseline corrected. The 1H NMR spectra of decoctions were calibrated to TSP at 0.00 ppm, and divided into integrated regions of equal width (0.01 ppm) corresponding to the region of δ 0.78-9.78. The region of δ 4.66-5.06 was excluded from the analysis because of the residual signals of H2O. The spectra of serum samples, which were referenced internally to the chemical shift of creatine at δ 3.04, were divided and the signal integral computed in 0.01 ppm intervals across the region δ 0.60-8.34. The region of δ 4.50-4.93 was removed to eliminate the effects of imperfect water saturation. The data were then normalized to the total sum of the spectra prior to analysis. The spectra of liver samples were referenced to the chemical shift of TSP at δ 0.00 and were manually phased and baseline corrected. The spectra were divided and the signal integral computed in 0.01 ppm intervals across the region δ 0.65-9.00. The region of δ 4.66-5.20 was removed to eliminate the influence of water signal. The data were then normalized to the tissue weights of the spectra prior to analysis.

Multivariate data analysis was performed with the software package SIMCA-P 13.0 (Umetrics, Sweden). Principal component analysis (PCA) was performed with the mean-centered data to generate an overview for group clustering and to search for possible outliers. Results were visualized in the form of the score plots, where each point represented an individual sample (or its metabolome). Orthogonal projection to latent structure with discriminant analysis (OPLS-DA) was further conducted with Pareto scaling, where s-plots, in which each coordinate represented one NMR spectral region (metabolite signal), was used to identify metabolites significantly contributing to the group separation. Qualities of the OPLS-DA models were assessed with R2X representing the explained variations, and Q2 for the model predictabilities [13]. For the quantitative data analysis of changes in the endogenous metabolites by the treatment of the six drugs, the relative content of metabolites were evaluated on the basis of the integrated regions (buckets) from the least-overlapping NMR signals of metabolites. These semi-quantitative data were expressed in the form of means ± standard error of the mean (SEM) and were also subjected to t-test to investigate the differences between the cold and hot drugs. Differences among the groups were considered to be statistically significant if P < 0.05. Similarity analysis is employed to evaluate the consistency of sample fingerprints using included angle cosine. The included angle cosine was calculated with the software of Excel 2003.

Results Chemical analysis of decoctions of herbal drugs

Firstly, 1H NMR fingerprints were used to give an overview of the chemical profile of each crude drug (Fig. S1); all of them were dominated by high concentrations of sugars, amino acids, and organic acids. Due to the high concentration of primary metabolites contained in the decoctions of each herbal drug, solvent partition was used to highlight the secondary metabolites contained in the decoctions, which were believed to be the bioactive compounds responsible for the action of herbal drugs. The results showed that not only the high level of primary metabolites, but also some secondary metabolites were present in the decoctions of each herbal drug (Tables S2-3 and Figs. S2-3). To find out whether the cold/hot properties were related to their chemical substances, multivariate data analysis was further applied to the NMR data of the decoctions of six herbs. The scores plotted from the principal component analysis (PCA) (Fig. S4) by PC1 (67.0 %) and PC2 (36.8 %) showed no clear separation among the six herbal drugs with cold and hot properties, suggesting that there were no obvious relationships between the chemical profile and the drug properties.

Supplementary Fig. S1 Representative 1H NMR spectra obtained from water extracts of the six herbs
Supplementary table S1 Solvent gradient program of UPLC analysis
Supplementary table S2 1H NMR spectral assignments of secondary metabolite identified in the water extracts of Phellodendri Chinensis Cortex, Coptidis Rhizoma, Scutellariae Radix, Zingiberris Rhizoma, and Cinnamomi Cortex
Supplementary table S3 Assignments of secondary metabolite identified in the water extracts of Aconitilateralis Ra-dix Praeparata using HPLC-MS/MS analysis
Supplementary Fig. S2 Representative 1H NMR spectra of secondary metabolites in water extracts of six herbs using different sample preparation procedures (a) petroleum ether fraction of Zingiberris Rhizoma; (b) petroleum ether fraction of Cinnamomi Cortex; (c) EtOAc fraction of Cinnamomi Cortex; (d) petroleum ether fraction of Scutellariae Radix; (e) Chloroform fraction of Coptidis Rhizoma; (f) Chloroform fraction of Phellodendri Chinensis Cortex; (g) standard compound of Wogonin. 1: 6-gingerol; 2: Cinnamaldehyde; 3: Cinnamic acid; 4:Wogonin; 5: berberine; 6: epiberberine; 7: Palmatine; 8: Phellodendrine; 9: berberine;
Supplementary Fig. S3 pectra of secondary metabolites in chloroform (1) fraction of water extracts of Aconitilateralis Radix Praeparata using UHPLC-MS analysis (a) HPLC-MS TIC chromatograms of chloroform (1) fraction of the water extracts of Aconitilateralis Radix Praeparata; (b) full mass spectra of Benzoylmesaconine; (c) MS/MS spectra of Benzoylmesaconine; (d) full mass spectra of Benzoylaconine; (e) MS/MS spectra of Benzoylaconine
Supplementary Fig. S4 epresentative 2D serum spectra obtained from one normal mouse (a) 1H-1H COSY spectrum; (b) HSQC spectrum. The numbers mentioned above correspond to those listed in Table S2
General observations

Table 1 depicts the growth performance of the mice. At the end of the experiment, the body weights of the animals in seven groups were compared using one-way ANOVA. The results showed that there were no significant differences among the seven groups (P > 0.05). However, average daily feed intakes of the hot medicine (HM) groups were higher than that of normal (NC) group (P < 0.05), while the cold medicine (CM) groups showed no difference from that of the NC group. The reasons may be that HMs can enhance the body's energy metabolism and further elevated the demand for food intake.

Table 1 Effects of drugs with different properties on growth performance in mice
Identification of metabolite by 1D and 2D NMR spectroscopy

The typical 1H NMR spectra of the serum and liver samples obtained from the NC group are shown in Fig. 1. A total of 32 metabolites in the serum samples and 34 metabolites in the liver samples were unambiguously assigned, according to spectral signals. The metabolites detected were elucidated by the analyses of the 1H NMR spectra as well as the comparison with the chemical shifts, splitting pattern, and coupling constant of standard compounds from the Chenomx NMR suite (evaluation version, Chenomx Inc., Edmonton, Canada), Human Metabolome Database (HMDB), Biological Magnetic Resonance Data Bank (BMRB), as well as the literature data [14-15]. In addition, 2D NMR techniques, including 1H-1H COSY, and HSQC, provided additional information for the confirmation of compound identification (Table S4 and Figs. S5-6).

Figure 1 Representative 600 MHz 1H-NMR spectra of mice serum (A) and liver (B) samples obtained from one normal mouse. 1: Lipids; 2: Valine; 3: Isoleucine; 4: Leucine; 5: β-Hydroxybutyrate; 6: Lactate; 7: Alanine; 8: Lysine; 9: Acetate; 10: N-Acetylated glycoproteins; 11: O-Acetylated glycoproteins; 12: Acetone; 13: Acetoacetate; 14: Pyruvate; 15: Succinate; 16: Glutamate; 17: Glutamine; 18: Citrate; 19: Dimethylglycine; 20: Choline; 21: Cysteine; 22: Glycerophosphocholine; 23: Trimethylamine oxide; 24: Creatine; 25: Glycerol; 26: Glycine; 27: Glucose; 28: Tyrosine; 29: Histidine; 30: Phenylalanine; 31: Formate; 32: Hypoxanthine; 33: Arginine; 34: Xanthine; 35: Methionine; 36: Glutathione; 37: Malate; 38: Uracil; 39: Aspartate; 40: Dimethylamine; 41:Trimethylamine; 42: Uridine; 43: Fumarate; 44: Taurine; 45: Betaine; 46: Scyllo-inositol; 47: Niacinamide; 48: Adenosine; 49:α-Mannose; 50: Glycogen
Supplementary table S4 1H -and 13C NMR Data for Metabolites in mice serum and urine
Supplementary Fig. S5 Representative 2D spectra of liver obtained from one normal mouse (a) and (b) 1H-1H COSY spectrum; (c) HSQC spectrum. The numbers mentioned above correspond to those listed in Table S2

The detected metabolites in serum samples spectra included numerous organic acids, amino acids, carbohydrates and TCA cycle intermediates (lactate and citrate), together with some other metabolites of choline, ketone bodies (acetoacetate). Moreover, hepatic spectra contained peaks mainly from organic acids (phenylalanine, formate, fumarate and lactate, etc.), amine metabolites (trimethylamine (TMA), dimethylamine (DMA)) and other metabolites such as creatine, β-hydroxybutyrate, and betaine.

Metabolic changes of serum and liver in mice administered with cold/hot drugs

To obtain more detailed metabolic differences among NC, CM and HM groups, all the NMR data were subjected to multivariate data analysis. In the PLS-DA 3D score plot (Fig. 2a) of serum samples generated by PC1 (24.4%), PC2 (11.7%) and PC3 (14.4%), a clear separation among the groups could be seen, and the NC group was located in the middle, while the CM groups and the HM groups were located on the right and left sides, respectively. Similarly, the PLS-DA 3D score plot (Fig. 2b) of liver extracts was also generated by PC1 (13.4%), PC2 (11.0%) and PC3 (14.4%). Although some of the NC samples are overlapped with other groups, a tendency of separation could be seen among NC, CM, and HM groups. These clustering patterns suggested that metabolic changes had occurred in both CMs and HMs treated mice, and the CMs and HMs showed different effects on the mouse metabolome. Then, further analysis was performed to find the differential endogenous metabolites contributing to the separation of the HM and CM groups.

Figure 2 PLS-DA 3D score plots of serum (A) and liver (B) samples of mice. Red dots represent HMs-treated groups; Blue dots represent CMs-treated groups; Black dots represent normal group

In the PCA score plot (Fig. 3a) generated by PC1 (34.5%) and PC2 (21.3%), CM and HM groups could be separated clearly. PLS-DA model was validated using the response of the permutation test through 200 permutations, in which all R2 and Q2 values were lower than original ones. The good PLS-DA model (R2X = 0.778, R2Y = 0.966, Q2 = 0.801) indicated an excellent predictive power (Fig. 3b). Then OPLS-DA was further used to find the differential metabolites between the CM and HM groups (Fig. 3c). Further evaluation with CV-ANOVA approach confirmed the validity of these models with P-values of 2.83 × 10-7. The corresponding s-plot (Fig. 3d) indicated that the mice serum samples of HM groups showed higher levels of glucose, creatine, glutamate, pyruvate, dimethylglycine, acetone, phenylalanine and formate, and lower levels of lactate, lipids, leucine, valine, β-hydroxybutyrate, citrate, and glycerophosphocholine, compared with CM groups. They were the major contributors to the separation of serum profiles between CM and HM groups. The same strategy as described above was also performed to the liver extracts. The OPLS-DA score plot (Fig. 4c) and the corresponding s-plot (Fig. 4d) indicated that the levels of hepatic β-hydroxybutyrate, glycogen, phenylalanine, uridine, and glycine were lower (P < 0.05), and the levels of xanthine, glutathione and methionine were higher in HM groups than those in the CM groups.

Figure 3 PCA scores plots (a), permutation test model validation plots (b), OPLS-DA scores plots (c), and loading s-plot (d) of serum samples between HM groups (■) and CM groups (●). 1: Lipids; 4: Leucine; 5: β-Hydroxybutyrate; 6:Lactate; 8: Lysine; 12: Acetone; 14: Pyruvate; 16: Glutamate; 19: Dimethylglycine; 22: Glycerophosphocholine; 24: Creatine; 27: β-Glucose
Figure 4 PCA scores plots (a), permutation test model validation plots (b), OPLS-DA scores plots (c), and loading s-plot (d) of liver sample between HM groups (■) and CM groups (●). 5: β-Hydroxybutyrate; 26: Glycine; 30: Phenylalanine; 34: Xanthine; 35: Methionine; 36: Glutathione; 42: Uridine; 50: Glycogen
Quantitative analysis of metabolite levels

The normalized integral values of metabolites in serum and liver samples from the CM and HM mice are compared in Tables 2 and 3. Variations in metabolites levels were in agreement with the results of multivariate statistical analysis. HM groups showed the higher levels of serum glucose, creatine, lysine, glutamate, pyruvate, dimethylglycine, and acetone, whereas CM groups showed higher levels of lactate, lipids, leucine, valine, β-hydroxybutyrate, and glycerophosphocholine. In the liver tissues, higher amounts of xanthine, glutathione and methionine as well as lower amounts of glycogen, β-hydroxybutyrate, phenylalanine, uridine and glycine were observed in the HM groups, compared with the CM groups. The results were in agreement with that of the multivariate analysis.

Table 2 Comparison of normalized integral levels of metabolites in mice serum samples
Table 3 Comparison of normalized integral levels of metabolites in mice liver samples
Similarity analysis of serum and liver samples in mice administered with cold/hot drugs

Similarity analysis can be employed to evaluate the consistency of sample fingerprints obtained using Nuclear magnetic resonance (NMR) spectroscopy, which can objectively and comprehensively reflect the characteristics of traditional Chinese medicine. Every single spectrum can be viewed as a set of values (taken as Xi or Yi) of corresponding relative peak areas, which are considered a vector (taken as X or Y) in multidimensional space. Then the similarity analysis of two spectra was converted to the analysis of two vectors in multidimensional space, using the equation (1) to characterize the similarities of fingerprints. If cos θ is close to 1, it indicates a big similarity between the two fingerprints.

$\cos \theta =\frac{\sum\limits_{i=1}^{n}{XiYi}}{\sqrt{\sum\limits_{i=1}^{n}{X{{i}^{2}}}}\sqrt{\sum\limits_{i=1}^{n}{Y{{i}^{2}}}}}$ (1)

Using the relative peak areas of the NMR spectra, a representative NMR fingerprint was generated by the mean values of integral values (buckets) of the eight samples in each group. In order to compare the similarities between the six drugs, the similarity values (SV) of serum samples were calculated and are listed in Table 4. The results showed that the serum samples of three HMs were highly similar to each other with the value of exceeding 0.988. The high SV over the range of 0.975-0.994 were also found among the three CMs, which indicating small intragroup variations. Meanwhile, relatively lower SV at the range of 0.897-0.934 were observed between HMs and CMs. In addition, similarity analysis was also applied to the liver samples (Table 5). The SVs were more than 0.989 among the three cold medicines and among the three hot medicines, respectively. In addition, HMs and CMs demonstrated higher similarity values at the range of 0.956-0.978 than those of serum samples, suggesting that drugs exerted weaker effects on liver profile than those of serum samples. These results were consistent with those of multivariate analyses and suggested that drugs with same property exerted similar effects on endogenous metabolites, while drugs with different property showed different effects.

Table 4 Similarity values of mice serum NMR spectra obtained via included angle cosine method
Table 5 Similarity values of mice liver NMR spectra obtained via included angle cosine method

Metabolomics focuses on the systematic study of the full complement of metabolites in a range of biofluids including urine, plasma, serum, cerebrospinal fluid, synovial fluid, semen, and tissue homogenates [16]. Fifteen serum and eight hepatic differential metabolites suggested the changes in some metabolic pathways, including glycolysis, lipid, amino acids and other metabolism.

Glycogen, made and stored primarily in the hepatic cells, is a multibranched polysaccharide of glucose that serves as a form of energy storage in animals. Pyruvate can be made from glucose through glycolysis, and converted back to carbohydrates (such as glucose) via gluconeogenesis. Pyruvate plays a crucial role not only in glycolysis and amino acid metabolism, but also regarding the tricarboxylic acid (TCA) cycle [17-18]. Lactate, which is formed from pyruvate and NADH by lactate dehydrogenase in the cytoplasm, is closely associated with energy metabolism and cellular redox state [19]. In the present study, the HM group showed higher levels of serum pyruvate and glucose as well as lower level of serum lactate and hepatic glycogen than that of the CM groups, suggesting that hot medicines could inhibit anaerobic metabolism and boost the activity of gluconeogenesis. Citrate, which is also related to energy metabolism, was dominant intermediate of TCA. The higher serum citrate level in CM groups suggested that CM treatment could inhibit the activity of ATP citrate lyase and further inhibited the conversion of citrate to acetyl-CoA [20-21].

Fatty acids β-oxidation in the mitochondria is one of the ways to provide energy. β-hydroxybutyrate, one of three ketone bodies, is the intermediate of fatty acids β-oxidation in the mitochondria [20]. In the present study, the higher level of serum β-hydroxybutyrate in CM groups than the HM groups suggested that CMs could suppress β-oxidation of fatty acids. Besides, the level of serum lipids (mainly LDL-VLDL) was higher in CM groups than that in HM groups, suggesting that lipoprotein was decreased and lipolysis was enhanced upon HM treatment.

Amino acids are basic units for protein synthesis in organism. Phenylalanine is an essential amino acid, which cannot be synthesized by the body, and is highly concentrated in high protein foods. The higher level of phenylalanine in the HM mice was consistent with higher food intake in HM groups. Leucine and valine are branched chain essential amino acids particularly involved in energy metabolism [22-23]. Leucine can be degraded to form acetoacetate and acetyl-CoA, which play a role in the tricarboxylic acid (TCA) cycle. Similarly, within the TCA cycle, valine acts as an intermediary metabolite, forming succinyl-CoA. Therefore, compared with CM groups, the lower level of both leucine and valine in the serum of HM group might reflect enhancement of the TCA metabolic pathway generating energy.

Glutathione is the major intracellular nonproteinthiol protecting cells against oxidative damage and harmful xenobiotics [24-25]. Glutathione is synthesized in tissues and cells (e.g., liver and intestine) from glutamate, cysteine and glycine, and hepatocytes are the major producer and exporter of glutathione [26-27]. In the present study, the HM groups showed higher hepatic glutathione and lower hepatic glycine levels than those of CM groups, indicating that hot medicines could promote glutathione synthesis in the liver [28]. Glycine is biosynthesized in the body from the serine, which is in turn derived from 3-phosphoglycerate. Glycine and arginine could be transformed into guanidinoacetate, which with methionine could change into creatine [29]. Glycine and creatine are important storage energy compounds. In the present study, higher levels of serum methionine and creatine as well as lower level of hepatic glycine were observed in the HM groups, indicating that hot medicines could enhance energy metabolism.

In summary, these potential biomarkers detected by 1H NMR provided the localized metabolome to understand the underlying molecular mechanisms of cold/hot medicine actions through the change of metabolic pathways.

In addition, the most relevant pathways were identified by the metabolic pathway analysis (MetPA) on MetaboAnalyst 2.0 [] [30]. HMs and CMs showed different effects on mice metabolome. The impact value of pathways analysis with MetPA was applied to evaluate the importance of the pathways on the differential metabolism between the CM groups and HM groups (Table S5 and Fig. S5). Six disturbed metabolic pathways were considered as the most relevant pathways involved in the mechanism (impact > 0.2). They were phenylalanine, tyrosine and tryptophan biosynthesis, glyoxylate and dicarboxylate metabolism, phenylalanine metabolism, valine, leucine and isoleucine biosynthesis, glycine, serine and threonine metabolism and glycerophospholipid metabolism. Citrate, phenylalanine, glycine, glycogen, formate, valine, dimethylglycine, and pyruvate involved in the six key pathways may denote their potential as targeted biomarkers related to the mechanism, which involves TCA, glycolysis, and amino acid metabolism referred to energy metabolism.

Figure 5 Summary of pathway analysis with MetPA. (a) Phenylalanine, tyrosine and tryptophan biosynthesis; (b) Glyoxylate and dicarboxylate metabolism; (c) Phenylalanine metabolism; (d) Valine, leunine and isoleucine biosynthesis; (e) Glycine, serine and threonine metabolism; (f) Glycerophospholipid metabolism
Supplementary table S5 Results of ingenuity pathway analysis with MetPAa

In the present study, a metabolomic approach based on 1H NMR coupled with multivariate statistical analysis was employed to study Chinese herbs regarding their cold/hot characteristics in a holistic manner. The results of multivariate analysis showed that metabolic changes had occurred in both HM and CM groups, and that HMs and CMs showed different effects on the mice metabolome. The similarity analysis showed that drugs with same property exerted similar effects on serum and liver in the mice, while drugs with different property showed different actions. Then differential metabolites of serum and liver between the HM and CM groups were identified to elucidate the mechanism of herbs with cold/hot properties exerted on the normal mice. The effect of herbal medicines with cold/hot properties was exerted by regulating pathways referring to glycometabolism, lipid metabolism, amino acid metabolism, and energy metabolism. The present study of Chinese herbs regarding their characteristics is expected to provide a basic view for the further study on the theory of TCM.

Liang F, Li L, Wang ML, et al. Molecule network and chemical fragment-based characteristics of medicinal herbs with cold and hot properties from Chinese medicine[J]. J Ethnopharmocal, 2013, 148(3): 770-779. DOI:10.1016/j.jep.2013.04.055
Liu YQ, Cheng MC, Wang LX, et al. Functional analysis of cultured neural cells for evaluating cold/cool-and hot/warm-natured Chinese herbs[J]. Am J Chin Med, 2008, 36(4): 771-781. DOI:10.1142/S0192415X08006223
Long W, Liu PX, Xiang J, et al. A combination system for prediction of Chinese Materia Medica properties[J]. Comput Meth Prog Biomed, 2011, 101(3): 253-164. DOI:10.1016/j.cmpb.2011.01.006
Yang Y, Liang YH, Wang CZ, et al. Research of time phases property of neuroendcorine immune and hemorheology on deficiency of cold and hot syndrome rats[J]. Chin J Basic Med Tradit Pharm Chin Med, 2002, 8(2): 109-112.
Wang MQ, Yan SL, Li WH, et al. A study on Chinese herb of cold and hot on SD rats[J]. J Zhejiang Coll Tradit Chin Med, 2002, 26(6): 43-45.
Wang X, Sun H, Zhang A, et al. Potential role of metabolomics apporoaches in the area of traditional Chinese medicine: As pillars of the bridge between Chinese and Western medicine[J]. J Pharm Biomed Anal, 2011, 55(5): 859-868. DOI:10.1016/j.jpba.2011.01.042
Yang YX, Liu Y, Zheng LY, et al. Serum metabonomic analysis of apoE(-/-) mice reveals progression axes for atherosclerosis based on NMR spectroscopy[J]. Mol Biosyst, 2014, 10(12): 3170-3178. DOI:10.1039/C4MB00334A
Jian WX, Yuan ZK, Huang XP. Detaction and analysis on plasma metabolomics in patient with coronary heart disease of Xin-blood stasis syndrome pattern[J]. Chin J Integr Tradit West Med, 2010, 30(6): 579-584.
Nicholson JK, Lindon JC, Holmes E. 'Metabonomics': understanding the metabolic responses of living systems to pathophysiological stimuli via multivariate statistical analysis of biological NMR spectroscopic data[J]. Xenobiotica, 1990, 29(11): 1181-1189.
Liu GM, Xiao L, Cao W, et al. Changes in the metabolome of rats after exposure to arginine and N-carbamylglutamate in combination with diquat a compound that causes oxidative stress assessed by 1H NMR spectroscopy[J]. Food and Function, 2016, 7(2): 964-974. DOI:10.1039/C5FO01486G
Tiziana Cappello, Fátima Brandão, Sofia Guilherme, et al. Insights into the mechanisms underlying mercury-induced oxidative stress in gills of wild fish (Liza aurata) combining 1H NMR metabolomics and conventional biochemical assays[J]. Sci Total Environ, 2016, 548-549: 13-24. DOI:10.1016/j.scitotenv.2016.01.008
Pharmacopoeia of People Republic of China. [S] Vol 1. 2015: 136-137.
Eriksson L, Trygg J, Wold S. CV-ANOVA for significance testing of PLS and OPLS (R) models[J]. J Chemom, 2008, 22(11-12): 594-600. DOI:10.1002/cem.v22:11/12
Nicholson JK, Foxaii PJD. 750 MHz 1H and 1H-13C NMR Spectroscopy of human blood plasma[J]. Anal Chem, 1995, 67(5): 793-811. DOI:10.1021/ac00101a004
Shi XH, Xiao CN, Wang YL, et al. Gallic acid intake induces alterations to systems metabolism in rats[J]. J Proteome Res, 2013, 12(2): 991-1006. DOI:10.1021/pr301041k
Zhang AH, Sun H, Wang P, et al. Recent and potential developments of biofluid analyses in metabolomics[J]. J Proteomics, 2012, 75(4): 1079-1088. DOI:10.1016/j.jprot.2011.10.027
Kauppinen RA, Nicholls DG. Synaptosomal bioenergetics. The role of glycolysis, pyruvate oxidation and responses to hypoglycaemia[J]. Eur J Biochem, 1986, 158(1): 159-165. DOI:10.1111/ejb.1986.158.issue-1
Liu GM, Xiao L, Fang TT, et al. Pea fiber and wheat bran fiber show distinct metabolic profiles in rats as investigated by a 1H NMR-based metabollomic approach[J]. PLoS One, 2014, 9(12): e115561. DOI:10.1371/journal.pone.0115561
Liu GM, Wang Y, Wang ZS, et al. Nuclear magnetic resonance (NMR)-based metabolomic studies on urine and serum biochemical profiles after chronic cysteamine supplementation in rats[J]. J Agric Food Chem, 2011, 59(10): 5572-5578. DOI:10.1021/jf104129k
Sun TJ, Hayakawa K, Bateman KS, et al. Identification of the citrate-binding site of human ATP-citrate lyase using X-ray crystallography[J]. J Biol Chem, 2010, 285(35): 27418-27428. DOI:10.1074/jbc.M109.078667
Liu GM, Yang GJ, Fang TT, et al. NMR-based metabolomics studies reveal changes in biochemical profile of urine and plasma from rats fed with sweet potato fiber or sweet potato residue[J]. RSC Advances, 2014, 4(45): 23749-23758. DOI:10.1039/c4ra02421d
Kimball SR, Jefferson LS. Signaling pathways and mo-lecular mechanisms through which branched-chain amino acids mediate translational control of protein synthesis[J]. J Nutr, 2006, 136(S1): 227-231.
Liu GM, Fang TT, Yan T, et al. Metabolomic strategy for the detection of metabolic effects of spermine supplementation in weaned rats[J]. J Agr Food Chem, 2014, 62(36): 9035-9042. DOI:10.1021/jf500882t
Hayes JD, McLellan LI. Glutathione and glu-tathione-depen-dent enzymes represent a co-ordinately regulated defence against oxidative stress[J]. Free Radical Res, 1999, 31(4): 273-300. DOI:10.1080/10715769900300851
Schulz JB, Lindenau J, Seyfried J, et al. Glutathione, oxidative stress and neurodegeneration[J]. Eur J Biochem, 2000, 267(16): 4904-4911. DOI:10.1046/j.1432-1327.2000.01595.x
Wu GY, Fang YZ, Yang S., et al. Glutathione metabolism and its implications for health[J]. J Nutr, 2004, 134(3): 489-492.
Wu G. Functional amino acids in growth, reproduction and health[J]. Adv Nutr, 2010, 1(1): 31-37. DOI:10.3945/an.110.1008
Liu GM, Yan T, Fang TT, et al. Nutrimetabolomic analysis provides new insights into spermine-induced ileum-system alterations for suckling rats[J]. RSC Advances, 2015, 5(60): 48769-48778. DOI:10.1039/C5RA01507C
Van Milgen J. Modeling biochemical aspects of energy metabolism in mammals[J]. J Nutr, 2002, 132(10): 3195-3202.
Liu YT, Jia HM, Chang X, et al. The metabolic disturbances of isoproterenol induced myocardial infarction in rats based on a tissue targeted metabonomics[J]. Mol Biosyst, 2013, 9(11): 2823-2834. DOI:10.1039/c3mb70222g