Mechanism of Chinese Medicine Herbs Effects on Chronic Heart Failure Based on Metabolic Profiling

Chronic heart failure (CHF) is a major public health problem in huge population worldwide. The detailed understanding of CHF mechanism is still limited. Zheng (syndrome) is the criterion of diagnosis and therapeutic in Traditional Chinese Medicine (TCM). Syndrome prediction may be a better approach for understanding of CHF mechanism basis and its treatment. The authors studied disturbed metabolic biomarkers to construct a predicting mode to assess the diagnostic value of different syndrome of CHF and explore the Chinese herbal medicine (CHM) efficacy on CHF patients. A cohort of 110 patients from 11 independent centers was studied and all patients were divided into 3 groups according to TCM syndrome differentiation: group of Qi deficiency syndrome, group of Qi deficiency and Blood stasis syndrome, and group of Qi deficiency and Blood stasis and Water retention syndrome. Plasma metabolomic profiles were determined by UPLC-TOF/MS and analyzed by multivariate statistics. About 6 representative metabolites were highly possible to be associated with CHF, 4, 7, and 5 metabolites with Qi deficiency syndrome, Qi deficiency and Blood stasis syndrome, and Qi deficiency and Blood stasis and Water retention syndrome (VIP > 1, p < 0.05). The diagnostic model was further constructed based on the metabolites to diagnose other CHF patients with satisfying sensitivity and specificity (sensitivity and specificity are 97.1 and 80.6% for CHF group vs. NH group; 97.1 and 80.0% for QD group vs. NH group; 97.1 and 79.5% for QB group vs. NH group; 97.1 and 88.9% for QBW group vs. NH group), validating the robustness of plasma metabolic profiling to diagnostic strategy. By comparison of the metabolic profiles, 9 biomarkers, 2-arachidonoylglycerophosphocholine, LysoPE 16:0, PS 21:0, LysoPE 20:4, LysoPE 18:0, linoleic acid, LysoPE 18:2, 4-hydroxybenzenesulfonic acid, and LysoPE 22:6, may be especially for the effect of CHM granules. A predicting model was attempted to construct and predict patient based on the related symptoms of CHF and the potential biomarkers regulated by CHM were explored. This trial was registered with NCT01939236 (https://clinicaltrials.gov/).


INTRODUCTION
Chronic heart failure (CHF), as a complex clinical syndrome, results from any structural or functional cardiac disorder that impairs the ability of the ventricle to fill with or eject blood (Hunt et al., 2005). In 2000, an international cooperation research program on cardiovascular disease in Asia (InterASIA) was made, which included 15,518 urban or rural residents from 35 to 74 years old. The results showed that the prevalence of heart failure was 0.9% for the general population, 0.7% for the males, and 1.0% for the females, indicating that there were ∼4 million heart failure targets in China (Gu et al., 2003). CHF is the only cardiovascular disease with an increasing hospitalization burden and an ongoing drain on health care expenditures, for its complex molecular mechanisms cannot be easily deciphered (Ramani et al., 2010). As a complementary alternative, Traditional Chinese Medicine (TCM) may be to improve the state of CHF. Zheng (syndrome) is the key pathological principle of TCM. All diagnostic and therapeutic methods in TCM are based on the differentiations of syndrome, and this concept has been used for thousands of years in China (Gu, 1956;Li et al., 2007). Syndrome differentiation was based on information from traditional four diagnostic methods. For the complexity and multilevel relationships of four diagnostic methods, a mode which could distinguish syndrome differentiation and be verified is imperative. Many techniques of data mining are applied to syndromes in TCM (Lukman et al., 2007;Shi and Zhou, 2007;Gao et al., 2010;Yao et al., 2011). Chen et al. proposed a discovery algorithm based on revised mutual information to discover syndromes for chronic renal failure (Chen et al., 2007). In regards to coronary artery disease, Liu et al. designed standardization scale on inquiry diagnosis and constructed this diagnostic model by using the method of multi-label learning (Liu et al., 2010). With many achievements have been made in syndrome differentiation, there are still some problems left, deserving further discussion (Wang et al., 2009;Lu et al., 2012). Syndromes-identified techniques such as multi-label learning could identify syndrome information in TCM more effectively, and solve the multi-label problems of one sample with several syndromes. While there are many research challenges of multilabel learning to be addressed in the future, such as multi-label data usually suffers from inherent class imbalance and unequal misclassification costs. Metabolomics is distinctive for its overall concept and dynamic character, which may be an effective tool to reveal the scientific connotation of TCM. Metabolic alterations are measured in response to disease progression (Cheng et al., 2015). LC-MS based methods provide more compatible technique and high quality data for sensitive detection of smallmolecule metabolites with robust reliability and reproducibility (Gika et al., 2007). Abnormal metabolism may characterize syndrome differentiation. Our previous metabolomics study on blood stasis syndrome of CHF showed glucose metabolism and lipid metabolism disorders reinforce each other, which results a deterioration of coronary artery disease with blood stasis syndrome. These metabolites maybe used as indicators of clinical diagnosis and treatment of coronary artery disease . Wang et al. concluded that NMRbased metabolomics approach demonstrated alteration of energy metabolism and other potential biological mechanisms underlying CHF (Wang et al., 2013b). In this paper, we try to discover comprehensive metabolomic characteristics of CHF and its syndrome differentiation for accurate clinical diagnosis. A syndrome predictive method for CHF was to build with non-targeted metabolomics and potential biomarkers related to CHF syndrome differentiation were to explore. The predictive performance would be validated and these potential biomarkers would be used as predictors to diagnose the patients. Besides, special metabolic characteristic of CMH on CHF would also be elaborated.

Patients and Study Design
The study included 110 patients with chronic heart failure from 11 hospitals ( (Feinstein, 1964). Syndrome differentiation of CHF patients followed the TCM differentiation standard: according to the FIGURE 1 | Study Design. 6MWT, 6 Min Walking Test; CHF, Chronic heart failure; CHM, Chinese herbal medicine; LVEF, Left Ventricular Ejection Fraction; NH, Normal healthy group; QD, Qi deficiency group; QB, Qi deficiency and Blood stasis group; QBW, Qi deficiency and Blood stasis and Water retention group.
Guiding Principles for the Clinical Study of New Drugs in Traditional Chinese Medicine released in 2002 (Zheng, 2002).

Inclusion Criteria
Primary heart disease in this research was Coronary Heart Disease (CHD), which could be diagnosed by coronary computed tomography, coronary angiography, limb-salvage Q wave for electrocardiogram (ECG), history of acute myocardial infarction, ECG test, radionuclide examination support, etc. The included patients also had no history of taking antihypertensive drugs and exhibited a blood pressure under 160/100 mmHg or of hypertension. Following symptoms and signs were observed with a history of CHD: fatigue, difficulty breathing, fluid retention (edema), left ventricular enlargement, NYHA functional classification II or III, and end systolic volume of left ventricular increase and left ventricular ejection fraction (LVEF)<40. All subjects between 40 and 75 years old; If there was a "No" answer for any issue, the subject could not enter into this research.

Exclusion Criteria
Patients with one of the following diseases were excluded: (1) serious valvular heart disease; (2) cardiomyopathy; (3) pericardial disease; (4) congenital heart disease; (5) cardiac shock; (6) acute myocardial infarction (AMI) within 4 weeks; (7) acute myocarditis or serious arrhythmia with variation in hemodynamics. Patients who suffer from pulmonary artery hypertension caused by cor pulmonale, pulmonary embolism, or stroke within a half year were also excluded. Patients who suffer from serious hepatic or renal deficiency. Patients who suffer from diseases of blood system or malignant tumor. Patients who suffer from diabetes mellitus with serious complications, hyperthyrea, or hypothyrea. Patients who suffer from infection. Pregnant women or women in lactation. Patients with mental disorders. Patients with infectious disease or who join in other trials within 2 months of the present study.
All qualified participants had signed informed consent prior to the enrollment. This study was conducted according to the principles of the Declaration of Helsinki and the principles of Good Clinical Practice.
All patients formed the chronic heart failure group, and were further divided into 3 groups according to TCM syndrome differentiation, including group of Qi deficiency syndrome (QD), group of Qi deficiency and Blood stasis syndrome (QB), and group of Qi deficiency and Blood stasis and Water retention syndrome (QBW). The normal healthy participants composed the NH group.
Besides, according to previous research (Wang et al., 2013a), the most syndrome of CHF was Qi deficiency and Blood stasis syndrome, so 64 of the participants with CHF of Qi deficiency and Blood stasis syndrome (QB group) were enrolled in a randomized double-blind controlled clinical trial. According to Plasma samples were collected on morning of the first and the twenty-eighth day after enrollment and were immediately frozen at −80 • C. Six minutes of walking test (6 MWT) and echocardiography were also detected at the same time. Both the observer of the 6 MWT and the echo technician were blind to the allocation of the patients.
After several preliminary experiments by different mixtures of methanol, acetonitrile, and methanol/acetonitrile (1:1), finally the extraction solvent was acetonitrile. By mixing equal amounts of plasma (30 µL) from all samples, pooled quality control samples were prepared to ensure data quality for metabolic profiling. Detailed sample methods were shown in the Supplementary Material. This research was approved by the Institutional Committee on Human Research of Beijing University of Chinese Medicine (No. 200807007), and this trial is registered with NCT01939236 (Luo et al., 2015).

Ultra Performance Liquid Chromatography/Time of Flight Mass Spectrometry (UPLC-TOF MS) Metabolomics Study Analysis
The liquid chromatographic separation of processed plasma was conducted on a 100 * 2.1-mm ACQUITY UPLC • R BEH C18 1.7-um column (Lot No. 0252350221) using an ACQUITY Ultra Performance LC (Waters Corporation, Milford Massachusetts, USA), whereas mass spectrometry was performed on a SYNAPT G2 Quadrupole-Time of Flight system (Waters Corporation, Milford Massachusetts, USA). With a random-number generator in Excel (Microsoft, Redmond, Washington), 3 sequences for samples in discovery phase, validation phase and placebo group were assigned by the study administrator. During analyses of the sample sequence, after each 20 injections, then 1 quality control sample was run to check the stability of system. The acquired mass spectrometry data were exported to data format (.centroid) files by Masslynx Software (version 4.0, Waters Corporation). Data pre-treatment procedures, such as nonlinear retention time alignment, peak discrimination, filtering, alignment, matching, and identification, were performed in Progenesis QI (34 Maple Street, Milford Massachusetts, 01757, USA), and retention time and the mass-to ratio data pairs as the parameters for each ion. The UPLC-QTOF-MS characteristic chromatogram of CHM granules (6 herbs And detailed method was shown in the Supplementary Material. And receiver operating characteristic (ROC) analysis were used for diagnosis of different CHF syndromes and validation of CHM treatment effects. To correct potential overfit of

OPLS-DA comparisons of Prior & Post-CHM group and
Prior & Post-Placebo group, five-fold Cross Validation was conducted to verify the specificity and sensitivity of the classifier. The data set was divided into 5 approximately equal bins and the models were evaluated 5 times. Statistical analyses were performed using python software version 2.7 (https:// www.python.org/) with scikit-learn version 0.18.2 (http://scikitlearn.org). An adjusted p < 0.05 was considered statistically significant.

RESULTS
A total of 110 patients with chronic heart failure from 11 hospitals and 54 normal healthy controls in China were enrolled in this research (Figure 1). The discovery phase included 34 normal healthy participants (NH) and 72 CHF participants, including the following groups: 15 QD, 39 QB, and 18 QBW patients.   obtained are shown in Supplementary Figures 1, 2. After peak alignment and removal of missing values, 442 negative-mode features were detected. After the OPLS-DA test, in SIMCA software, the ions with variable importance in the projection (VIP) values >1.0 and p-value <0.05 were considered the potential differential metabolites.

Differential Diagnosis of Metabolic Biomarkers in Different Syndromes
Presently clinical diagnosis of different TCM syndromes of CHF depends on personal experience of clinical doctors (Xu et al., 2016), so accurate diagnosis of the different stages of CHF, based on objective biomarkers in blood, is fundamental and significant for personalized treatment. So CHF and NH was compared to identify and characterize special metabolites of patients with CHF in discovery phase. QD, QB, and QBW were compared with NH to identify and characterize special metabolites of different Chinese syndromes in discovery phase. Clear differences were obtained for the following: CHF vs. NH, cumulative R2Y at 0.923 and Q2 at 0.892 (Figure 2A); QD vs. NH, R2Y at 0.976 and Q2 at 0.94 ( Figure 2B); QB vs. NH, R2Y at 0.967 and Q2 at 0.94 ( Figure 2C); and QBW vs. NH, R2Y at 0.957 and Q2 at 0.868 ( Figure 2D). With VIP values >1.0 and adjusted p-value < 0.05, the metabolites for each comparison appear in Table 2. Six specific metabolic biomarkers distinguished CHF from NH, 4 for QD from NH, 7 for QB from NH, and 5 for QBW from NH. PCA (Principal Component Analysis) result among QD, QB, QBW, and NH was shown in Supplementary Figure 3, and QD was the predominate Zheng contributing to CHF progression far away from NH.
With highest prediction specificity and sensitivity of the ROC, the optimal cut-off values were 0.1684 for CHF vs. NH, 0.2934 for QD vs. NH, 0.3403 for QB vs. NH, and 0.4763 for QBW vs. NH in the discovery phase. Different stages of CHF in the verification phase were predicted by the trained classifier and cut-off values. Predictive value was 85.0% for CHF vs. NH in the verification phase ( Figure 3E); 95.1% for QD vs. NH in the verification phase ( Figure 3F); for 99.1% QB vs. NH in the validation phase ( Figure 3G); and 90.1% for QBW vs. NH in the validation phase ( Figure 3H).

Special Metabolic Treatment Biomarkers of Chinese Medicine Herbs
There was no significant difference of 6 MWT and ejection fraction between Prior-CHM group and Prior-Placebo group, FIGURE 7 | Heat map of differential metabolites. NH, Normal healthy group; QD, Qi deficiency group; QB, Qi deficiency and Blood stasis group; QBW, Qi deficiency and Blood stasis and Water retention group. and 4 weeks later, both changed significantly between Post-CHM group and Post-Placebo group, as shown in Figure 4.
There were significant curative effects of CHM on different diseases, but usually with indistinct metabolic mechanisms (Lao et al., 2014). According to 2 PLS-DA comparisons, Prior-CHM vs. Prior-Placebo vs. NH, R2Y at 0.586, Q2 at 0.524 (Figure 6A), Post-CHM vs. Post-Placebo vs. NH, R2H at 0.514, Q2 at 0.39 (Figure 6C), and 2 OPLS-DA comparisons, Prior-CHM vs. Prior-Placebo, R2H at 0.505, Q2 at 0.0866 (Figure 6B), Post-CHM vs. Post-Placebo, R2H at 0.845, Q2 at 0.16 (Figure 6D), after the 4 weeks treatment, the signs of patients in CHM group left the Placebo group to the NH group. Then Post-CHM group and Prior-CHM group was compared to identify and characterize special metabolites of combined treatment. Post-Placebo and Prior-Placebo was compared to identify and characterize special metabolites of basic Western Medicine treatment. The metabolites with VIP values >1.0 and adjusted p-values <0.05 for 2 OPLS-DA comparisons appear in Table 3 and there were 11 metabolic biomarkers for distinguishing Post & Prior-CHM group, and 3 for Post & Prior-Placebo group. By comparison between the 11 and 3 metabolic biomarkers, there were 9 special metabolic biomarkers for Prior & Post-CHM group.
The ROC presentations, on the basis of Support Vector Classification (SVC) of Support Vector Machine (SVM) which trained by on the basis of the logistic regression of each biomarker panel, appear in Figure 6; the AUC, sensitivity, and specificity are 0.777, 64.5%, and 96.8% for Post-CHM vs. Prior-CHM (n = 62; Figure 6E); and 0.804, 63.6%, and 90.9% for Post-Placebo vs. Prior-Placebo (n = 66; Figure 6F), respectively.
At the same time, there existed a certain correlation between specific metabolites determined in CHF patients with QB Zheng (n = 7) and CHM efficacy in QB patients (n = 11), evaluated by Chi-squared test with P = 0.0557.

Correlation Network of Differential Metabolites
According to 4 OPLS-DA comparisons between Normal healthy group and different CHF groups, total 13 differential metabolites were identified. Average normalized quantities of the 13 differential metabolites were shown in the heat map Figure 7 among Normal healthy group and the CHF groups. Among them, 1 metabolites were elevated; 12 metabolites decreased in order according to the following groups: NH, CHF, QD, QB, and QBW.
Moreover, we constructed the metabolism-protein networks and identify 36 related proteins, such as Calcium-dependent phospholipase A2, Group IIF secretory phospholipase A2, and Cytosolic phospholipase A2 in the most important Arachidonic acid metabolism pathway (Figure 8). Up-regulated and downregulated metabolites determined were shown with red and blue points, and undetected metabolites are shown as gray points. Direct related enzymes were shown in green square frames, and others were with no color. Detailed information of points in

DISCUSSION
Coronary artery disease is the leading cause of death in the world (Lu et al., 2012). In TCM, coronary artery disease belongs to the scope of chest heartache and cardiodynia (Liu et al., 2010). A comprehensive metabolomic evaluation in our research described for 110 patients who underwent CHF in 11 independent centers and 54 normal healthy participants in the discovery and validation phase. LC-MS-based metabolomic approach was conducted to demonstrate metabolic differences between NH and CHF patients with different syndromes in this study and evaluate the effect of Chinese medicine herbs on CHF.
In placebo and CHM groups, the levels of arachidonic acid both increased, while arachidonic acid is the potential biomarkers of Qi deficiency and Blood stasis and Water retention syndrome. As one of the pivotal signaling molecules, arachidonic acid is associated with inflammation, pain and selfbalance function. Li et al. summarized the advances of giving insights into the arachidonic acid pathways in cardiovascular disease with metabolomic profiling (Ning et al., 2011). As a likely culprit for cardiovascular adverse effect associated with rofecoxib and NSAIDs, 20-hydroxyeicosatetraenoic acid (20-HETE) was identified during arachidonic acid metabolism in their study. On the side, it showed that epoxyeicosatrienoic acids (EETs) exhibited cardioprotective effects in a murine myocardial infarction model. It was reported that posphatidylcholine could produce proatherogenic species, choline, and trimethylamine oxide after metabolized by intestinal microbiota (Fan et al., 2016). It may be an implication for coronary artery disease with the decreasing level of phosphatidylcholine (Tang et al., 2013). Compared with patients in placebo group, patients in CHM groups had up-regulated more lysophosphatidyl ethanolamines. It was reported that the level of LysoPE 18:2 was down-regulated in patients with coronary atherosclerosis (Fan et al., 2016).
With the results of metabolites among each group, linoleic acid metabolism and arachidonic acid metabolism were identified as the main pathways according to the metabolismprotein network analysis and 63 potential related proteins included. Enzymes superfamily of Phospholipases A2 (PLA2s) are known to play multiple roles to maintain membrane phospholipid homeostasis and produce a variety of lipid mediators (Sun et al., 2010). It has been reported that increased or decreased expression of Ca2+-independent phospholipases A2 (iPLA2s) had profound effects on the metabolic state, CNS function, and cardiovascular performance, and dysregulation of iPLA2s can be a critical factor in the development of many diseases (Ramanadham et al., 2015). The superfamily of cytochrome p450 (CYP) takes part in the process of oxidation, peroxidation, and (or) reduction of vitamins, steroids, xenobiotics, and the majority of cardiovascular drugs in an oxygen-and NADPH-dependent manner. While during cardiovascular physiology and disease, the role of CYP is poorly understood. Evidence suggested that CYP plays an important role in the pathogenesis of cardiovascular diseases. The understanding was summarized as to the role that the receded CYP expression and (or) activity may play in the onset and progression of cardiovascular disease (Hunter et al., 2004).

CONCLUSION
In this paper, a predicting model was attempted to construct and predict patient based on the related symptoms. The Syndrome prediction process was implemented by the objective biomarkers in blood forecasting performance. It generated a better performance than 3 classifiers with ROC curve analyses. We concluded that the method in the study may be applied to predict the syndromes of CHF. The potential biomarkers regulated by CHM were explored and it may be essential for the further study of CHM additional efficacy.

AUTHOR CONTRIBUTIONS
KG, HZ, and JG: Designed the research, conducted performed the majority of the experiment, and revised the manuscript; BW, CJ, ZW, FZ, JiW, HX, and JuW: Assisted on supported several experimental performances and deal with the statistical data; JC and WW: Supervised the research and revised the manuscript.