A Metabolomics Study of the Volatile Oil from Prunella vulgaris L.on Pelvic Inflammatory Disease

DENG Jing,SU Qian,LIN Xiu-Lian,LIN Yan,LI Ya-Mei,LIN Li-Mei,LIAO Duan-Fang,XIA Bo-Hou

Key Laboratory for Quality Evaluation of Bulk Herbs of Hunan Province,Hunan University of Chinese Medicine,Changsha,Hunan 410208,China

Keywords

Pelvic inflammatory disease(PID)

Metabolomics method

Gas chromatography-mass spectrometry(GCMS)

Prunella vulgaris L.oil

Random forests

ABSTRACT

Objective Pelvic inflammatory disease(PID)is one of the most common gynaecological diseases.Here,this thesis aims to investigate the therapeutic effects of Prunella vulgaris L.oil on the PID by using metabolomics based on gas chromatographymass spectrometry(GC-MS)to address this challenge.

Methods First,measurements of pro-inflammatory cytokines and histological analysis of the uterus were conducted to validate the successful generation of a PID rat model.Furthermore,the volatile oil from Prunella vulgaris L.was administered to treat PID rats.Serum samples were collected before and after treatment and analyzed by GC-MS to generate metabolite profiles for each sample.The information generated from the qualitative and quantitative analysis of these metabolites was applied to distinguish between the PID model and normal control groups.

Results Some metabolites,such as acetic acid,succinic acid,glyceric acid,(R*,S*)-3,4-dihydroxybutanoic acid,3-hydroxyphenylacetic acid,D-ribose and myo-inositol showed a higher contribution in the classification model; thus,they can be considered as potential biomarkers.Furthermore,the therapeutic effect of the volatile oil extracted from Prunella vulgaris L.could also be visualized using GC-MS-based metabolomics.

Conclusions The results show that metabolomics studies are invaluable for disease diagnosis and therapeutic effect estimation.

1 Introduction

Pelvic inflammatory disease(PID)is one of the most common gynecological disease,commonly occurs in women younger than 35.It is rare before menarche,after menopause and during pregnancy[1].Often,the disease exhibits no specific symptoms but manifests itself as an inflammatory disorder of the upper female genital tract,such as salpingitis,endometritis,tubo-ovarian abscess and pelvic peritonitis[2].Moreover,subclinical PID or a delay in the treatment of PID may lead to long-term sequelae,including chronic pelvic pain,ectopic pregnancy and tubal infertility[3,4].Currently,no single test exists having adequate sensitivity and specificity to diagnose PID[5].Therefore,a more convenient,effective and precise diagnostic evaluation is needed.

High-throughput metabolomics strategies have been widely used in the biomedical sciences and have been proven to be powerful approaches in the diagnosis of human diseases,physiological evaluations,elucidation of biomarkers and drug toxicity tests[6].Metabolomics can assess the amplified output of a biological system since metabolic alterations are the consequence of genetic changes in metabolic pathways.Specific metabolic changes during disease progression have been reported and used to further understand disease treatment and drug response.Metabolomics-based approaches may prove valuable in aiding the clinical diagnosis of breast cancer,finding biomarkers for outcome prediction,and measuring responses to drug therapies.

To date,many analytical techniques have been applied to the metabolomics analysis of urine,including gas chromatography-mass spectrometry(GC-MS),high-resolution nuclear magnetic(NMR)and ultra-performance liquid chromatography-mass spectrometry(UPLC-MS)[7-10].GC-MS is one of the most widely used techniques and it can generate metabolic data and simultaneously,provide quantitative,qualitative and structural information on a wide range of biological molecules[10-12].Because of its stability,reproducibility,sensitivity and better separation of compounds,GC-MS is considered one of the most effective tools to perform metabolomics research.

Therefore,in this study,we applied a metabolomics strategy on a chronic pelvic inflammation model to facilitate PID diagnosis,and to represent the therapeutic effects of the volatile oil fromPrunella vulgarisL..First,the metabolomics strategy was used to characterize the different metabolic patterns of different groups[13-15].The random forests algorithm was used to distinguish between normal and PID groups and to find potential biomarkers for facilitating PID diagnosis.Moreover,PID model rats were treated with the volatile oil fromPrunella vulgarisL.and the therapeutic effects were validated using the same metabolomics strategy[16,17].

2 Materials and Methods

2.1 Chemicals and reagents

Rat TNF-α,IL-1βand IL-6 enzyme-linked immunosorbent assay(ELISA)kits were purchased from Becton-Dickinson(San Diego,CA,USA).Dexamethasone acetate tablets were bought from Xianju Pharmaceutical Co.,Ltd.(Zhejiang,China).Sodium carboxymethylcellulose(CMC-Na)and sodium azide were obtained from the Sinopharm Chemical Reagent Co.,Ltd.(Shanghai,China).Ribitol,pyridine,urease,o-methylhydroxylamine hydrochloride,internal standard heptadecanoic acid(C17:0),derivatization reagents methoxamine,pyridine,Bis(trimethylsilyl)trifluoroacetamide(BSTFA)and N,O-bis(trimethylsilyl)trifluoroacetamide(BSTFA)containing 1% trimethylchlorosilane(TMCS)were all obtained from Sigma-Aldrich(St Louis,MO,USA).Distilled water was filtered using a Milli-Q®Water Purification System from Kertone Water Treatment Co.,Ltd.(Bedford,MA,USA).All other chemicals used were of analytical grade.

2.2 Equipments

Metabolic data were acquired with Shimadzu©QP-2010 gas chromatography mass spectrometer(Kyoto,Japan).Drying was carried out under nitrogen with HGC-24 apparatus(Tianjin Heng’ao Company,China).

2.3 Preparation of samples and standard solutions

Fruits ofPrunella vulgarisL.were obtained from the Gaoqiao natural herbal special market(Changsha,China).After the species of the samples was confirmed by Prof.WANG Zhi of Hunan University of Chinese Medicine,the material was dried in the shade and powdered by disintegrator(HX-200A,Yongkang Hardware and Medical Instrument Plant,China).The dried fruit powder(1 kg)was ground into particles,which were then passed through an 80 mesh and extracted by ultrasonication with methanol(10 L)at room temperature for 45 min,which was repeated three times.The filtrates were combined and concentrated using a rotary evaporator(Harbin Quefu Filter Material Co.,Ltd.,China).The methanol extract was then chromatographed on a Si(Silica)gel(100 - 200 mesh)column using petroleum ether(isocratic elution)to yield the oil fraction,which was concentrated and stored at - 4 °C for further pharmacology and metabolomics studies.

2.4 Animal experiment and sample collection

A surgically induced pelvic inflammation model was generated using the method described by CHANG et al.[18].SPF grade adult female Sprague-Dawley rats(140 - 160 g body weight,n=60)were used for the experiments.Ten rats were randomly selected to serve as the normal control group(NCG)and the remaining 50 rats were used to establish the pelvic inflammation model.From these 50 rats,10 rats were selected to serve as the sham group(SHG).The rest 40 rats were anesthetized with sodium pentobarbital(40 mg/kg,i.p.)and were fixed in a supine position.Then,the middle position of their lower abdomen was shaved and the skin in this area was sterilized using 10% povidone-iodine solution.Subsequently,a small midline incision was made in the sterile skin area,the uterus was carefully exposed,and 100 μL of a bacteria suspension(E.coli,Staphylococcus aureus,and Streptococcus hemolytic-βin a ratio of 2∶1∶1 in sterile normal saline,with a total of 3×109bacteria/mL)was injected into both the right and left sides of the uterus along the direction of the ovary(model group),using a 30-gauge needle placed at the uterine horn.For the SHG,the same amount of sterile normal saline was injected.After the injection,a 2% lidocaine solution was applied to the wound,which was then closed in layers.After 15 d,the disease model was established.Before model grouping,venous blood(500 μL)was collected from the suborbital vein from two rats from each of the NCG,SHG and model group.

After discarding six infected rats,two dead rats and six rats whose vital signs was unstable,the rest of 24 model rats were randomly divided into three groups:model control group(MCG,n=8),dexamethasone group(DG,n=8)and oil group(OG,n=8).NCG,SHG and MCG rats were treated with 0.5%(w/v)CMC-Na,while DG rats were treated with 1.82 mg/kg dexamethasone,and OG rats were treated with 10 mL/kg(equal to 1 g/kg of crude drug)Prunella vulgarisL.oil.CMC-NA was administered intragastrically,whereas dexamethasone andPrunella vulgarisL.oil extract were orally administered.All drugs were administered once a day for 30 d.After treatment termination,venous blood was collected from the suborbital vein,the rats were sacrificed,and the uteri harvested and prepared for further histological study.

2.5 Serum preparation and collection

Following overnight fasting,venous blood samples(2 mL)were collected from each rat in blank tubes without anticoagulant or preservative.The fresh blood was allowed to clot at 4 °C for 1 h,and the serum was separated by centrifugation at 1 509.3×g to obtain serum and stored at - 80 °C until analysis.The collected serum was separated into two parts,one of which was used for metabolomics analysis and the other for the measurement of pro-inflammatory cytokine(TNF-α,IL-1βand IL-6)production by commercial ELISA kits,following the manufacturer’s instructions.The isolated uteri were fixed in buffered 10% formaldehyde for 24 h,embedded in paraffin,sectioned with a microtome,and stained with hematoxylin-eosin.

2.6 Serum pretreatment

Before analysis,the serum was thawed at 4 °C for 30 min.Subsequently,300 μL methanol(including 1 mg/mL of heptadecanoic acid/methanol as an internal standard)was added to 100 μL serum,mixed on a vortex for 15 s,and centrifuged for 15 min(25 155×g,4 °C).The supernatant was collected,dried,mixed with methoxamine/pyridine(20 mg/mL)for 15 s and incubated for 1 h(70 °C),followed by the addition of 100 μL BSTFA,mixing for 15 s,and an additional incubation for 1 h(70 °C).Following this pre-processing,all serum samples were analyzed by GC-MS in a random order.

2.7 GC-MS conditions

Chromatographic separation was performed on an Agilent HP-5MS instrument,equipped with a deactivated fused silica capillary column(30 m×0.25 mm×0.25 μm).The temperature was maintained at 70 °C for 4 min,and then increased to 300 °C at a rate of 8 °C/min,and held at that value for 3 min.The temperatures of the front injection port,ion source,and interface were set at 280 °C,280 °C and 230 °C,respectively.Helium was used as the carrier gas with a flow rate of 1.0 mL/min.One microliter of the sample was injected with a split ratio of 10 :1.The mass spectrometer was operated under the electron impact(EI)mode at an ionization energy of 70 eV and a detector voltage of 0.90 kV,and full scans were acquired at a rate of 0.2 s/scan.The mass spectrometer was operated with an m/z range from 35 to 600.

2.8 Data processing and analysis

Specific chromatography processing algorithms,including deconvolution,baseline removal and retention time shift corrections,were used to analyze metabolite fingerprints and compensate for the variations in the experimental procedures.Then,metabolites and internal standards were identified by searching the NIST 12 mass chromatography library with the GC-MS Postrun Analysis software(Shimadzu),according to retention time values for characteristic ions of authentic standards or as reported in the literature.For the quantitative analysis of chromatographic peaks,area calculations were performed directly by the Shimadzu GC-MS Postrun Analysis software.Accordingly,34 endogenous metabolites with high NIST match(＞ 90% similarity)were selected for detailed analysis.

2.9 Random forests(RF)algorithm

RF is a supervised machine learning classifier introduced by Breiman that generates an ensemble of classification and regression trees.These many different decision trees are grown from various bootstrap samples.Each tree casts a vote for sample classification,and RF chooses the majority vote to decide the final classification result.Compared with other classification methods,RF is associated with lower bias,which make it applicable in many research fields.

Details of RF can be found in our previous study[19].Here,we briefly describe the methodology used for this study.First,ntreebootstrap samples were drawn from the original data.Then,each bootstrap sample was grown to the maximum extent by randomly selecting an mtryvariable from a subset of variables,which resulted in the best split without pruning at each node.Lastly,the prediction value was obtained by aggregating the votes of all trees.In RF,two important parameters need to be selected.One is the number of trees(ntree)and the other is the number of split variables(mtry).In the current study,ntreewas set to 500,andmtrywas equal to the square root of the number of variables in the X matrix.

The classifier performance was evaluated by the following measures:

TPis the number of true positives,TNis the number of true negatives,FPis the number of false positives,andFNis the number of false negatives.The area under the ROC curve(AUC)was also used as an evaluation performance index.The predictive ability of the model was assumed higher as the value of AUC approached 1.

2.10 Statistical analysis

Data are presented as means±SD.One-way ANOVA followed by LSDttest was used for the comparison of the levels of the pro-inflammatory cytokines.The data were analyzed using the SPSS 18.0(SPSS Inc.,Chicago,IL,USA)for Windows.P＜0.05 was considered to denote statistical significance.

3 Results

3.1 Assessment of pelvic inflammation model

As shown in Figure 1,after establishing the disease model for 15 d,inflammation of the endometrium was severe in the MCG,whereas almost nonexistent in the NCG and SHG.The MCG showed both severe endometrial inflammation and inflammatory infiltrate,indicating that the pelvic inflammation model was successfully established.

3.2 Assessment of the therapeutic effects of the Prunella vulgaris L.oil

As shown in Figure 2 A,after a 30-day treatment,the levels of IL-1β,TNF-αand IL-6 in the MCG were significantly higher than those in the SHG and NCG(P＜0.01).This increase was especially pronounced for IL-6.OG and DG dramatically suppressed this elevation of IL-6 levels,while a significant inhibition of the increase in the levels of IL-1βand TNF-αwas also observed(P＜0.05).

Similar results were obtained by histological analysis.As shown in Figure 2B - 2F,the structure of the uterine cavity was severely damaged in the MCG,which was accompanied by hyperemia and edema of the endometrium.In addition,the structure of muscularis externa and stratum mucosum appeared fuzzy and severe endometrial inflammation were observed(Figure 2B).On the contrary,the NCG and SHG exhibited less endometrial inflammation and inflammatory infiltrate,and an improved structural integrity of the uterine cavity and the glandular organ(Figure 2B).The therapeutic effect of OG is shown in Figure 2F,where little endometrial inflammation and inflammatory infiltrate can be observed,and the structure of the uterine cavity and of the glandular is intact.

3.3 Metabolomics variations in PID rats before and after drug treatment

3.3.1 RF classification of different groups based on metabolite informationThe results of the qualitative and quantitative analyses of metabolites found in the samples from different groups are listed in Table 1.All of these metabolite information were identified by standard substances.

The random forest method was used to classify different groups(Figure 3).The NCG clustered in the lower right part of the score plot,whereas the MCG clustered in the lower left part of the plot,demonstrating the different metabolic character induced by disease.These differences in metabolites can be used to diagnose different groups.Then,the RF classification model was used to classify the NCG and MCG based on their metabolic profiles.The classification accuracy,sensitivity,specificity and AUC values were 96.73%,95.23%,100% and 96.42%,respectively.Moreover,we also found that the OG was near to the NCG,suggesting that the metabolic character of the OG is similar to that of the NCG,resulting from the treatment with thePrunella vulgarisL.oil.

3.3.2 Biomarker discovery and pathway analysis based on metabolites informationAs can be seen from Figure 4A,some metabolites,such as propanoic acid,1,2-hydroquinone,L-threonine,L-proline,glucitol and octadecanoic acid,were significantly different from others.These metabolites could be considered as potential biomarkers for disease diagnosing and treatment,as they were associated with larger model weight values.A similar result was found in the OG,demonstrating the therapeutic effects of thePrunella vulgarisL.oil on the PID rat model(Figure 4B).The volatile oil mainly regulated the metabolism of propanoic acid,L-threonine,Lproline,glucitol,octadecanoic acid and cholesterol.Four of these metabolites were also identified in Figure 4A.Therefore,these common metabolites may be the therapeutic markers of the volatile oil fromPrunella vulgarisL..And the metabolic pathways results are displayed in Figure 5.As can be seen,some important metabolic pathways(citrate cycle metabolism,pyruvate metabolism,inositol phosphate metabolism)are disturbed in PID.

Table 1 Qualitative and quantitative metabolic profiles of animal models treated with different drugs

4 Discussion

Infection and inflammation of the uterus are common causes and manifestations of pelvic inflammation[20].Therefore,this study adopted the method of establishing mixed infection to simulate the occurrence of this disease.The index selected inPrunella vulgarisL.oil therapeutic effects evaluation are IL-1β,IL-6 and TNF-α.These are proinflammatory cytokines,which play an important role in the pathogenesis of PID in patients,as well as in experimental model rats with pelvic inflammation[21,22].These biomarkers,particularly IL-6,could be useful adjuncts for the clinical and experimental diagnosis of PID.Therefore,we can observe changes in the corresponding experimental results.

In metabolomics variations in PID rats before and after drug treatment section,we chose the RF method to analyze the metabolic profile changes between the different groups.Thus,samples from the same group always had larger similarities than those from different groups.To observe the sample distribution better,multidimensional scaling(MDS)was employed to map the proximity into a lower-dimensional space.

The discovery of novel metabolic biomarkers lies at the core of metabolomics surveys.In this experiment,four biomarkers were identified as the possible therapeutic markers of the volatile oil fromPrunella vulgarisL..To further explore the potential metabolic pathways affected in PID,biomarkers with large contributions were selected and submitted to the MetaboAnalyst server to construct the associated metabolic pathways.The results are displayed in Figure 5.As can be seen,some important metabolic pathways(citrate cycle metabolism,pyruvate metabolism,inositol phosphate metabolism)are disturbed in PID.

Some fatty acids had a higher contribution in the generation of the classification model.It is known that fatty acids,as energy sources and membrane components,have biological activities that influence cell and tissue metabolism(including lipid metabolism),function and responsiveness to hormonal and other signals.Indeed,our biological pathway analysis identified the citrate cycle(TAC cycle),which is the ultimate metabolic pathway of the three major nutrients(sugars,lipids and amino acids)and the central link among carbohydrate,lipid and amino acid metabolism.

The metabolites from the other major pathway identified,namely that of inositol phosphate metabolism,also had a higher contribution for model generation.Inositol phosphates are a group of mono-and poly-phosphorylated inositols,which play crucial roles in diverse cellular functions,such as cell growth,apoptosis,cell migration,endocytosis and cell differentiation.These inositol phosphate metabolites are also important energy sources,which are consistent with the hypothesis that both fatty acid and inositol phosphate pathways are involved in PID.

In the present study,we utilized a metabolomics method to accurately discriminate between the PID model and control groups and further represent the metabolic profile changes upon various drug treatments.The results demonstrate that our approach could effectively mine the metabolic characters hidden in the complex metabolomics data.Such a metabolomics method could be an attractive alternative technique for conducting research on disease pathogenesis and on pharmacodynamics,as well as for disease diagnosis.

Acknowledgements

We thank for the funding support from the National Natural Science Foundation of China(No.81503041),Natural Science Foundation of Hunan Province(No.2017JJ4045)and Changsha Science and Technology Project(No.kq1701073).

Competing Interests

The authors declare no conflict of interest.