Outpatient prescriptions and inpatient orders containing RPM, PMP, and PMC in Longhua Hospital from December 1, 2016 to December 1, 2017 were chosen from the hospital information system (HIS) to analyze the information such as age, sex, and department by PA Software (self-developed, Chinese Software Copyright Registration No. 2017SR012498).

Gallic acid (#B20851), trans-SG (#B21757), emodin (#B20240), physcion (#B20242), and EG (#B20241) were bought from Yuanye Biotech (China). Cis-SG (#E-0261) and PG (#E-2497) were purchased from Tauto Biotech (China). Methanol and acetonitrile (HPLC grade) were provided by Merck (Germany). Deionized water was obtained using a Milli-Q water system (Millipore, United States). Phosphate acid was from Shanghai Wujin Chemicals (China). RPM, PMP, and PMC were collected from several provinces in China (Supplementary Table S1). These samples were authenticated by Chief Pharmacist Xiu-Feng Shi from Longhua Hospital. Voucher specimens were deposited in a light-absent and well-ventilated room.

From each batch, 25 g of RPM, PMP, and PMC samples were, respectively, boiled with 150 ml water for 1 h, filtered, then boiled with 100 ml water for 1 h, and combined together. For high-performance liquid chromatography (HPLC) analysis, the combined water extracts were concentrated to 0.5 g crude herb per mL. Then 1 ml of the concentrated extract was accurately measured, transferred to a 10-ml volumetric flask, and diluted with methanol. The mixture was sonicated for 30 min, restored overnight at 4°C, centrifuged for 15 min at 2,370 g, and filtered through a 0.22-µm microporous membrane.

Stock standard solutions of the accurately weighed reference compounds were prepared in methanol with appropriate concentrations. Then 200 μl of the standard mixture was centrifuged for 15 min at 13,680 g and filtered through a 0.22-µm microporous membrane before injection. The calibration curves were obtained by plotting the chromatographic peak areas versus the concentration of the analytes.

HPLC analysis was performed using an Agilent Technologies 1290 Infinity HPLC system (United States) coupled with a pro-shell 120 SB-C₁₈ column (2.7 μm, 100 × 2.1 mm) at 30°C. The mobile phase consisted of solvent A (0.1% phosphate acid in deionized water) and solvent B (acetonitrile). The gradient elution was as follows: 5–25% B at 0–10 min, 25%–50% B at 10–15 min, 50%–70% B at 15–20 min, 70%–90% B at 20–25 min, and 90% B at 25–30 min. The flow rate was 0.3 ml/min, the injection volume was 3 μl, and the UV wavelength was set at 280 nm.

The standard mixture solution was analyzed six times a day to estimate the precision. Five different sample solutions, prepared as previously described, were injected to check the repeatability. The stability test was carried out at different time points (0, 1, 2, 4, 8, 16, 20, and 24 h) with the same sample solution.

HPLC fingerprints of PM samples were established by using a similarity evaluation system for the chromatographic fingerprint of TCM (2004A), developed by the ChP Commission (China). Each established chromatographic fingerprint was recorded to present the chemical feature of RPM, PMP, and PMC. Common peaks of 15 batches of RPM, PMP, and PMC were confirmed, respectively, and merged together to compare the peak areas between RPM, PMP, and PMC in the same retention time. HCA was carried out in R to estimate chemical similarity among batches of PM based on Ward D2 linkage algorithm and Euclidean distance metric.

Female Balb/c mice (6 weeks old) were purchased from Shanghai Model Organisms Center (China). The mice were housed at a constant temperature and humidity on a 12-h light/dark cycle with free access to food and water and were allowed to accommodate for 7 days prior to the experiments. All procedures were approved by the Animal Ethics Committee of Longhua Hospital (Ethics Approval No. SCXK2020-1061) and were performed in accordance with the guidelines of the National Animal Welfare Law of China.

The mice were divided into 12 groups (7 mice/group): control, LPS, RPM_high, PMP_high, PMC_high, RPM_low + LPS, PMP_low + LPS, PMC_low + LPS, TSG_RPM, TSG_RPM + LPS, TSG_PMP + LPS, and TSG_PMC + LPS. The procedure of herbal water extract preparation was carried out as mentioned earlier, and the extracts were concentrated to 30 g/kg for RPM_high, PMP_high, and PMC_high groups, and 6 g/kg for RPM_low + LPS, PMP_low + LPS, and PMC_low + LPS groups. The dosages of TSG were calculated using the ratio of cis-SG and trans-SG average contents in 30 g RPM, PMP, and PMC based on the aforementioned chemical analysis. cis-SG and trans-SG were dissolved in water and kept in light-shielded bottles. These groups were orally administrated with RPM, PMP, and PMC or TSG daily for 14 days, respectively. On day 14, the relevant groups were injected through the caudal vein with 0.5 mg/kg lipopolysaccharide (LPS, Sigma, #L2880, United States) dissolved in normal saline 2 h before oral administration, and all mice were euthanized for sample collection 6 h after gavage.

Blood samples were collected, centrifuged to obtain serum, and stored at −80°C. The serum levels of ALT and AST were determined by using a Beckman Coulter AU5800 automatic biochemical analyzer (United States). Liver tissues were placed immediately in 4% paraformaldehyde to fix overnight, dehydrated, embedded in paraffin, sliced for 5-µm-thick sections, and stained with hematoxylin–eosin (HE). For enzyme-linked immunosorbent assay (ELISA), hepatic tissues were homogenized with RIPA lysis buffer containing protease and phosphatase inhibitor cocktail (Beyotime, China). Interleukin-6 (IL-6), interleukin-1β (IL-1β), and tumor necrosis factor-α (TNF-α) in hepatic homogenates were measured by using ELISA kits (Multi Sciences (LiankeBio), China) according to the manufacturer’s protocol.

Antigen retrieval and endogenous peroxidase elimination were performed in EDTA buffer (pH 9) for 5 min and in methanol containing 3% H₂O₂ for 10 min, respectively. The sections were then blocked in normal goat serum for 30 min, followed by incubation of primary antibodies against F4/80 (1:19200, CST, #70076, United States) and NF-κB p65 (1:400, CST, #8242, United States) at 4°C overnight and secondary antibody incubation for 30 min at room temperature. After washed with PBS, the sections were treated with DAB for 10 min, and restained with hematoxylin.

Of 77 outpatients, 115 prescriptions contained RPM, and 29,657 prescriptions of 12,769 outpatients contained PMP, while 1,44,025 prescriptions of 44,617 outpatients contained PMC (Table 1). Most outpatients were 40–80 years old (Figures 1A,C,E). Over 66% of outpatients using PM were female (Figure 1G). The departments prescribing PMC (Figure 1F) most frequently were oncology (25%), expert clinic (11%), and VIP clinic (9%). Expert clinic (23%), oncology (16%), and VIP clinic (8%) were the main departments prescribing PMP (Figure 1D). As for RPM (Figure 1B), the main departments were expert clinic (46%), rheumatology (17%), and VIP clinic (10%). (VIP clinic provides quality service for VIP patients. Expert clinic provides service by senior doctors.)

No RPM was used in inpatient wards. In total, 1,109 orders of 431 inpatients contained PMP, while 4,943 orders of 1,444 outpatients contained PMC (Table 1). Most inpatients were 50–80 years old (Figures 1J,L). Over 55% of inpatients using PM were female (Figure 1I). The departments prescribing PMC (Fig. 1M) and PMP (Figure 1K) most frequently were both oncology (65 and 48%). The dosages of PMC were 9 g (3%), 15 g (29%), and 30 g (68%), while the dosage ChP recommended was 9–15 g. As for PMP, 12 g (56%) was the most frequently used dosage, and 15 g accounted for 15% and 30 g was 3%, compared with the dosage recommended by ChP: 6–12 g (Figure 1H).

A good linearity of each marker ingredient was observed in a relatively wide concentration with a correlation coefficient above 0.999 (Supplementary Table S2). The relative standard deviation (RSD) of the relative peak area was calculated to verify the precision, repeatability, and stability of the method with 7 marked ingredients as the reference. Statistical analysis showed that the RSDs were lower than 1.19, 5.00, and 4.11%, respectively (Supplementary Table S2). All these demonstrated that the proposed methods were reliable for sample determination.

Fifteen batches of PMC, RPM, and PMP water extracts were analyzed by HPLC/UV under the aforementioned chromatographic conditions, and their chromatograms were recorded (Figure 2A). Thirty-six common peaks were observed between 1 and 22 min time intervals in all batches. Among them, 7 common peaks were identified as gallic acid, cis-SG, trans-SG, EG, PG, emodin, and physcion by comparing with the peaks of mark ingredients (Figure 2C). In general, the peaks in the established PMC fingerprint were lower than those in the established PMP and RPM fingerprints (Figure 2B). The peak of trans-SG in the established RPM fingerprint was the highest peak among all established fingerprints. HCA results were displayed on the edge of heat-map showing that the samples were properly sorted into three main clusters (Figure 2D). The left cluster consisted of S1-S15, the right cluster consisted of S16-S30, and the cluster in the middle consisted of S31-S45, corresponding the samples of PMC, RPM, and PMP, respectively. The middle and the right clusters were linked together, revealing that PMP was more similar to RPM than to PMC. In the heat-map, the blue–white–red color gradient indicated the peak area in each retention time from lowest to highest intensity. The heat-map inferred that roughly, with the retention time increase, the chemical constituents were successively concentrated in PMP, RPM, and PMC.

The contents of gallic acid were abundant in PMP, and the contents of PG were abundant in PMC, while the contents of trans-SG, EG, and physcion were abundant in RPM. In addition, the contents of cis-SG, trans-SG, and emodin in PMC were statistically lower than those in RPM and PMP. As for the dominant chemical component in PM, trans-SG, the contents were significantly different (p < 0.001), decreasing in the order of RPM, PMP, and PMC (Figure 3).

The dosages of TSG were calculated according to the ratio of cis-SG and trans-SG average contents in 30 g RPM, PMP, and PMC based on the results of content determination (Figure 3). Therein, 0.462 mg/kg cis-SG and 40 mg/kg trans-SG were administrated to TSG_RPM and TSG_RPM + LPS groups, 0.45 mg/kg cis-SG and 18 mg/kg trans-SG were administrated to the TSG_PMP + LPS group, and 0.18 mg/kg cis-SG and 4.8 mg/kg trans-SG were administrated to the TSG_PMC + LPS group.

Poor appetite and lethargy were observed in the PMC_high group, and one of them began losing weight since day 4, so we suspended its PMC gavage and gave remedy since day 6 for animal welfare considerations, but it still died on day 8 (Figure 4A). One of the remaining mice in the PMC_high group (n = 6) began losing weight since day 11 (Figure 4B). No death or weight loss was recorded in other groups (Figures 4A,B). There were no significant differences in the levels of serum ALT and AST between the control group and LPS group, indicating the mice were injected with LPS at a non-toxic dosage (Figures 4C,D). The ALT levels of RPM_low + LPS, PMP_low + LPS, PMC_low + LPS, TSG_RPM + LPS, and TSG_PMP + LPS groups significantly increased compared with the control group (Figure 4C), and the AST level of the RPM_low + LPS group significantly increased compared with the control group (Figure 4D). HE results showed that inflammatory lesions increased in the liver tissues of RPM_low + LPS group, while RPM_high, PMP_high, and PMC_high groups exhibited similar liver histology to that of the control group (Figure 4E). Macrophage is one of the effector cells involved in hepatic injury (Meng et al., 2017). Herein, we investigated the expressions of macrophage-specific marker F4/80 in the liver tissues of mice. IHC results illustrated that the F4/80 expressions of LPS and TSG_RPM groups did not change obviously compared with the control group, while the F4/80 expressions dramatically increased in the RPM_low + LPS group and slightly increased in RPM_high, PMP_high, PMC_high, PMP_low + LPS, PMC_low + LPS, TSG_RPM + LPS, TSG_PMP + LPS, and TSG_PMC + LPS groups (Figure 4F).

NF-κB signaling plays a critical role in inflammatory response which constitutes one of the main features of DILI (He et al., 2017; Giridharan and Srinivasan, 2018). Herein, we investigated the expressions of NF-κB p65 in the liver sections and the contents of inflammatory cytokines, IL-6, IL-1β, and TNF-α, in the liver homogenates of mice by IHC and ELISA. Compared with the control group, the contents of hepatic IL-6 were significantly higher in RPM_low + LPS, PMP_low + LPS, PMC_low + LPS, and TSG_RPM + LPS groups (Figure 5A). As for the contents of hepatic IL-1β and TNF-α, uptrends were found in the LPS-injected mice co-treated with PM or TSG when compared with the mice administrated with PM or TSG alone (Figures 5B,C). IHC results demonstrated that the expressions of p65 increased in RPM_low + LPS, PMP_low + LPS, PMC_low + LPS, TSG_RPM + LPS, TSG_PMP + LPS, and TSG_PMC + LPS groups, while the expressions of p65 in LPS, TSG_RPM, RPM_high, PMP_high, and PMC_high groups were not obviously altered compared with the control group (Figure 5D).