RMP was purchased from Tso-Ngon Tibetan Medicine Hospital. Other chemicals were purchased from Sigma Chemical Co unless otherwise specified.

Animal experiments were conducted in accordance with the Guidelines for Animal Experimentation of Beijing University of Chinese Medicine, and the protocol was approved by the Animal Ethics Committee of the Institution. A total of 48 male Wistar rats (220 ± 10 g) purchased from Sibeifu Laboratory Animal Technology Co., Ltd. (Beijing, China; Rodent license No. SCXK 2011-0004). The animals were kept under controlled lighting (12 h light/dark cycle), temperature (22 ± 2°C), humidity (50 ± 10%) and were allowed for a week to acclimate environment prior to group allocation.

Rats were randomly allocated into four groups (n = 12) as following: low dose group (LD), middle dose group (MD), and high dose group (HD), which were administrated with RMP at a dose of 83.33, 333.33, and 1333.33 (mg/kg/day) respectively, and control group (NC) was treated with approximately equal volume of 0.9% saline solution. The intragastric administration was implemented per day for a consecutive 15 days. Body weights were recorded once a day. The doses of RMP for rats were equivalent to 5, 20, and 80 times of the normal clinical dosage.

The experiment last for 30 days. Half of all rats (n = 6 for each group) were euthanized on the 15th day (after the last administration of RMP), and the remaining rats were euthanized on the 30th day (15 days after the recovery period). Blood samples were collected and stored in a heparinized tube on ice. Half of the blood samples were separated and frozen for further ICP-MS analysis, and the plasma samples were obtained by centrifugation and stored at −80°C until further ¹H NMR spectroscopic. The livers and kidneys were quickly removed from all rats at the time of death and were weighed up after rinsing with sterile 0.9% (w/v) sodium chloride solution. For liver and kidneys, half of tissues were immersed in 10% neutral-buffered formaldehyde for 24 h and embedded in paraffin to be stained with hematoxylin and eosin (H&E) for pathological analysis. And the residual tissues were placed in tubes and stored at −80°C until assessment.

The plasma samples were thawed only once in a biosafety fume hood, and then each 200 μL of plasma sample was mixed with 400 μL of deuterated phosphate buffer (NaH₂PO₄/K₂HPO₄, 0.045 mol/L, pH 7.47). The mixture was left to stand for a moment at room temperature and then centrifuged at 13,000 rpm for 15 min at 4°C for further purification. Aliquots of the supernatant (550 μL) of each sample were then transferred into a 5 mm NMR tube for NMR analysis (Feng et al., 2010).

Frozen kidney and liver tissues were homogenized in a mixture of an ice-cold extraction solvent (H₂O: CH₃CN 1:1) and centrifuged at 13,000 rpm at 4°C for 15 min. 1.00 mL volumes of supernatant (containing hydrophilic metabolites) was collected, concentrated under a stream of nitrogen and lyophilized and reconstituted in 600 μL 0.1 mol/L phosphate buffer (pH 7.47, 0.05% TSP, and 100% D2O), and 550 μL of the supernatant was decanted into a 5 mm NMR tube.

All ¹H NMR spectra were acquired at a ¹H frequency of 599.871 MHz using a Varian VNMRS 600 MHz NMR spectrometer, equipped with a 5-mm inverse-proton (HX) triple resonance probe. For plasma samples, the Carr–Purcell–Meiboom–Gill (CPMG) pulse sequence (RD-90°-(τ-180°-τ)_n-ACQ) with a fixed total spin-spin relaxation delay 2 nτ of 320 ms was applied to observe the signals of micromolecules. Aqueous tissues extract sample spectra were recorded using One-dimensional RESAT pulse sequence, with a water presaturation suppression applied for a recycle delay of 2 s and a mixing time of 100 ms. For each sample, the free induction decays (FIDs) were measured with 128 scans producing 64 k data points over spectral width of 12,019 Hz. Before Fourier transformation, the FIDs were zero-filled to 64 k points and processed with 0.5 Hz exponential line-broadenings (Jianxin et al., 2016). Standard two-dimensional (2D) NMR experiments such as ¹H-¹H COSY, ¹H-¹³C HSQC and ¹H-¹³C HMBC were also acquired to assist metabolite assignment.

The kidney and liver tissues were freeze-dried for 24 h at −80°C and homogenized individually with agate mortar prior to analysis.

Two tracer elements (⁷⁵As, ²⁰²Hg) in all samples (Blood, kidney and liver) were quantified by ICP-MS (Agilent Technologies, 7500 Ce) after a modified acid digestion with HNO₃+H₂O₂ (5:1) by microwave assisted digestion (CEM Co., MARS). The mercury and arsenic standard stock solution of 100 mg/L were purchased to measure the calibration curves, and were diluted to the standard solutions with 5% (w/w) HNO₃ in ultrapure deionized water. The external calibration procedures were employed for the quantification of the samples (Khan et al., 2015). Quantitation was achieved by a four-point calibration curve with a series of concentrations (0.1–100 ng mL⁻¹), using ⁷²Ge as internal standards for As and ²⁰⁹Bi for Hg.

The ¹H NMR spectra was processed using MestReNova software (Version 7.1.0, Mestrelab, Inc.), and the spectra was manually baseline- and phase-corrected and referenced to the TSP signal (δ0.00). The plasma spectra were referenced to resonance of lactate at δ1.33, and the tissues spectra were referenced to the chemical shift of TSP (δ0.00). The spectral region δ9.0–0.5 for each plasma sample and the spectral region δ9.5–0.5 for each tissue sample were all automatically data reduced to integrated segments of equal width (0.002 ppm). Spectral regions (δ5.20–4.70) were excluded to eliminate variations caused by imperfect water suppression. The integrated data of the remaining bins were normalized to the total sum of integrals for each spectrum to compensate for the effect of variation (Feng et al., 2011).

The NMR spectral data sets were examined with SIMCA-P+12.0 (Umetrics, Sweden) for multivariate statistical analysis. Principal component analysis (PCA) was usually implemented with mean-centered NMR data to identify general trends and outliers and to examine group clustering. Then, the supervised partial least squares discriminant analysis (PLS-DA) and orthogonal projections to the latentstructures discriminant analysis (OPLS-DA) at a unit variance-scaled approach were performed to specify the metabolic variations associated with the drug. The parameters of R2 and Q2 were used to assess the goodness of fitness and the predictive ability of the models, respectively. In order to reflect the variables contributing to the classification, the color-coded loadings plots with absolute value of coefficients (r) were used to identify significantly altered metabolites, and were generated with MATLAB R2012a (http://www.mathworks.com) with some in-house modifications (Jianxin et al., 2016).

In order to discriminate the variables contributing to sample clustering among the groups, the variable importance in the projection (VIP) values of all peaks from OPLS-DA models was analyzed. Moreover, an independent sample t-test was imported into detecting significant differences between the two groups by SPSS Statistics Base 17.0 (SPSS Inc., USA). In this study, we used an a priori cutoff VIP value > 1 of multivariate and p < 0.05 of univariate statistical significance to identify the distinguishing metabolites.

The histopathological changes in the liver and kidney were examined after HE staining. Compare to the NC group (Figure 1A), liver of HD rats on 15th days showed hydropic degeneration, inflammatory and severe epithelial necrosis (Figure 1B). And the pathologic injuries alleviated in liver tissue (Figure 1C) on 30th days. Kidney of the NC rat showed an normal structure in renal glomerulus and tubule (Figure 1D), while HD rats on 15th days showed slight pathologic transformations of renal glomerulus and tubule, such as, tubular epithelial cell degeneration and diaphanous tubular cast (Figure 1E). Fifteen days after regeneration phase showed markedly recovery trends for kidney injuries which induced by RMP (Figure 1F).

Representative 600 MHz ¹H NMR spectra of plasma and tissue extracts obtained from control and HD on 15th days were showed in Figure 2, with major metabolites in the integrated regions assigned. NMR spectral data was analyzed and specific metabolites were identified by the previously reported reference (Kim et al., 2014; Guo et al., 2015; Liao et al., 2016; Li et al., 2016) and in-house NMR database and the database of Chenomx NMR Suite 6.0 (Chenomx, Edmonton, Canada). Metabolites detected in plasma, tissue extracts including amino acids, glycolysis products, tricarboxylic acid cycle (TCA cycle) intermediates and choline metabolites, etc.

In order to display the overall metabolic trends and find the possible outliers, the NMR data of all samples was subjected to PCA analysis. The PCA score plots (Figure 3A) of plasma ¹H NMR spectra 15 days after administration with RMP displayed evident differentiation between NC and dosed groups along the first principal component, except a partial overlap between NC group and LD group. As for withdrawal period (Figure 3B), the discrimination between control and RMP groups was less clear compared with 15 days before. PLS-DA on 15th day (Figure S1A) highlighted the distinctive separations between controls and RMP groups but the overlapping between MD and HD groups still existed, which implied the similar metabolic phenotype between MD and HD groups in plasma, while the separation was less clear on 30th day (Figure S1B).

To screen metabolomics differences between control group and RMP groups and compare the significance of the identified metabolites contribution during the physiological alterations, pairwise OPLS-DA score plots and corresponding loading plots of plasma samples were performed in Figure 4 and Figure S2. The OPLS-DA score plots gave a clear separation between RMP groups and NC group. Color-coded coefficient plots showed detailed metabolites changes induced by RMP. Metabolites exhibiting significant changes (p < 0.05) were recognized as potential biomarkers according to the absolute cutoff value of correlation coefficients (|r|) and VIP value were summarized in Tables 1, 2. Results suggest little perturbation in low RPM group after 15-day oral administration. However, compared with control group, many metabolites in high dose RMP group, such as, glucose, betaine, and creatine, remarkable increased, while lipids, 3-hydroxybutate, pyruvate, citrate, branched-chain amino acids (BCAAs; valine, leucine, and isoleucine), glutamate and glutamine decreased at the same time. Most of the disturbed metabolites were ameliorated after the withdrawal of RMP, but there were still some significant metabolites in HD, including alanine, acetone, lactate, proline, glucose, and betaine.

Some separation of the liver samples from control and RMP groups were evident in PCA plot on 15th day (Figure 3C), but complete separation among RMP groups were not achieved. 52.2% of the variance was explained by PC1, while 17.4% was explained by PC2. It was observed that there was still a discrimination among the control and RMP groups on 30th day in PCA plot (Figure 3D), especially between the NC and HD group. The score plot was obtained with the first two PCs presenting 21.2 and 20.9% variance, respectively. The PLS-DA score plot of liver on 15th day showed in Figure S1C, and four groups displayed a distinct separation. While on day 30 (Figure S1D) the separation was less obvious.

Figure 5 and Figure S3 showed the OPLS-DA score plots and corresponding loading plots of liver samples. The ¹H NMR-identified relative metabolites in liver samples of dose groups are shown in Tables 1, 2. The metabolomics analysis of liver showed elevated concentrations of valine, leucine, proline, tyrosine, and tryptophan, as well as decreasing glucose, sarcosine and 3-hydroxybutyrate from RMP-administrated rats. The difference between control group and RMP groups became less significant on 15th day after the last administration. Metabolites regulation of the RMP group could be observed, which included leucine, valine, lactate, proline, glucose, and betaine.

According to the PCA score plots of the kidney tissue in Figure 3E, slight separation was observed on 15th day, with partial overlap between RMP groups and control group. However, the differences between control and RMP groups were observed on 30th say in PCA plot (Figure 3F), indicating that the toxicity of RMP in kidney may delay. From the PLS-DA score plot, the differentiation of RMP groups and control group is observed on 15th day (Figure S1E), while the separation was less apparent on 30th day (Figure S1F).

The separation between NC and each RMP group is exhibited in the score plots of OPLS-DA and corresponding loading plots of kidney samples were performed (Figure 6 and Figure S4). The color-coded co-efficient plots disclosed the increased levels of leucine, valine, isoleucine, and tyrosine, together with the decreased levels of taurine, betaine, choline, 3-hydroxybutyrate, and no evident recovery was displayed on 30th day. The identification of significant class-discriminating metabolites was summarized in Tables 1, 2.

Based on the identified biomarkers, MetaboAnalyst 3.0 (http://www.metaboanalyst.ca/MetaboAnalyst/) was performed to identify the relevant pathways involved in the study conditions. According to previous literature, pathways with the impact value above 0.1, which is calculated from pathway topology analysis, were screened out as potential target pathway. Finally, there were 9 potential target pathways identified in plasma samples, including valine, leucine and isoleucine biosynthesis, glycine, serine and threonine metabolism, methane metabolism, glyoxylate and dicarboxylate metabolism, pyruvate metabolism, alanine, aspartate and glutamate metabolism, aminoacyl-tRNA biosynthesis, glycolysis or gluconeogenesis and citrate cycle (TCA cycle) (Figure 7A). And there were 6 potential target pathways identified in liver (Figure 7B), including phenylalanine, tyrosine and tryptophan biosynthesis, valine, leucine and isoleucine biosynthesis, phenylalanine metabolism, tryptophan metabolism, alanine, aspartate and glutamate metabolism and tyrosine metabolism. There were 6 potential target pathways including Phenylalanine, tyrosine and tryptophan biosynthesis, valine, leucine and isoleucine biosynthesis, taurine and hypotaurine metabolism, phenylalanine metabolism, glycerolipid metabolism and tyrosine metabolism were identified in kidney samples (Figure 7C). The details of pathways were displayed in Tables S1–S3, Supporting Information.

RMP contains a great deal of mineral element including Hg and As (Zeng-cai-dan et al., 2016), which were besides the human body essential trace elements and may affect the metabolism in certain way. So it was necessary to measure the concentrations of heavy metal in tissues and biofluid. The results were summarized in Table S4 and Figure 8. The levels of As and Hg in rats plasma, liver and kidney tissues were dose-dependent. RMP groups, especially HD group, showed an increase of As concentrations after 15-day oral administration, and the concentration of As in plasma and kidney declined 15 days after the last administration, while the concentration of As showed a slightly elevation in liver. The concentration of Hg showed the same trend after the RMP administration. It was shown in Figure 8 that the Hg level in kidney was substantially higher than that in plasma and liver for all group. The level of Hg in kidney was substantially higher than that in plasma and liver of all RMP groups and even in the control group, which means that the accumulation of Hg was apparent in the kidney. Nevertheless, during the regeneration phase, a remarkable decrease of Hg was observed in both plasma and tissue samples.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.