CHS-J specimens produced by five different years, i.e., 2014, 2015, 2016, 2018, and 2020 (n = 20) and 35% (v/v) of base liquor of CHS-J were kindly provided by Jing brand Co. Ltd (Hubei, China), and the company declined to disclose the detailed information, such as the amount of each herb and the extraction procedure, as a trade secret. All specimens were stored at room temperature and light- protected environment at the Institute of Chinese Medical Sciences, University of Macau, Macao. Thirteen reference compounds, including Z-ligustilide, Baohuoside Ⅰ, Sagittatoside B, Epimedin B, Echinacoside, 2″-O-rhamnosyl Icariside II and Magnoflorine, were purchased from Baoji herbest Bio-Tech Co. Ltd (Shaanxi, China). Their purities were over 98%, as detected by HPLC-UV. Acetonitrile (HPLC-grade) and formic acid (HPLC-grade) were obtained from Merck (Darmstadt, Germany) and Aladdin Industrial Inc. (Shanghai, China), respectively. Deionized water was prepared by a Millipore Milli-Q water purification system (Bedford, MA, United States). A drying and heating oven (ED260 model, BINDER GmbH, Tuttlingen, Germany) and constant climate chambers (KBF 240 model, BINDER GmbH) were used in temperature-accelerated and light-accelerated tests, respectively.

Approximate 1 ml of CHS-J samples was mixed with the equivalent volume of acetonitrile. The solution was vortexed for 1 min and then centrifuged at 4^°C (14,800 rpm, 20 min). The supernatant was filtered through a 0.22 μm polytetrafluoroethylene filter (Millipore, Bedford, MA, United States) prior to the UHPLC-MS analysis. An equal volume (10 μl) of each CHS-J sample was mixed to generate a pooled QC sample, which offers a representative “mean” sample consisting of all analytes. The QC sample was analyzed 6 times at the beginning and end of the sequence, and then every 10 tested CHS-J samples to evaluate the system stability and method reproducibility.

After sample preparation, CHS-J samples were analyzed by UHPLC-Q-TOFMS. Chromatographic separation was conducted on a Waters ACQUITY UPLC BEH C₁₈ column (100 × 2.1 mm, 1.7 μm) with column temperature at 35°C. The analytes were subjected to gradient elution chromatography with a mobile phase composed of 0.1% formic acid (v/v) in water (phase A) and 0.1% formic acid (v/v) in acetonitrile (phase B). The solvent gradient elution program was used as follows with the flow rate of 0.3 ml min⁻¹: linear gradient from 5 to 40% B (0–15 min), 40–70% B (15–18 min), 70-100% B (18–20 min),: isocratic 100% B (20–22 min), 100–5% B (22–22.1 min), 5% B (22.1–25 min). The sample injection volume was 2 μl, and the autosampler temperature was maintained at 4°C. The electrospray ionization (ESI) source with MS^E continuum acquisition mode was adopted to collect mass spectra in positive and negative ion modes. The key MS parameters were set as follows: MS range, m/z 50 to m/z 1200; capillary voltage, +3.0 kV in ESI+ and -2.5 kV in ESI-; cone voltage (CV), 40 V; cone gas flow, 50 L h⁻¹; desolvation gas flow, 600 L h⁻¹; source temperature, 100 °C; desolvation temperature, 250 °C. A lock-mass calibrant of leucine-enkephalin (200 ng/ml) was continuously infused at flow-rate of 50 μl/min via a lock spray interface, generating the reference ions ([M+H]⁺ = 556.2771, [M−H]⁻ = 554.2615) for mass correction during the analysis.

To determine the levels of seven changed compounds highlighted in untargeted metabolomics, including Z-ligustilide, baohuoside Ⅰ, sagittatoside B, epimedin B, echinacoside, 2″-O-rhamnosyl icariside II and magnoflorine, an ACQUITY™ UPLC system coupled with a Xevo TQD triple-quadrupole tandem mass spectrometry (QQQ-MS/MS, Waters Co., Manchester, United Kingdom) were used to quantitatively determine these components in CHS-J samples. Chromatographic separation was achieved on the identical BEH C₁₈ column using the same mobile phases. The flow-rate was set at 0.3 ml min⁻¹ with the following solvent gradient: 5–40% B (0–12 min), 40–100% B (12–16 min), 100% B (16–18 min), and the re-equilibrated by 5% B for 3 min. Data acquisition was conducted on a Xevo TQD QQQ-MS equipped with an ESI using multiple reaction monitoring (MRM) mode. MS was operated in the positive mode to obtain satisfactory MS response for seven investigated analytes. The optimized MS parameters were depicted as follows: capillary voltage, +3.5 kV; source temperature, 140°C; desolvation gas flow and temperature, 650 L h⁻¹, and 350°C; cone flow, 50 L h⁻¹. The ion transitions, cone voltage (CV), and collision energy (CE) for each compound were optimized and shown in Supplementary Table S1.

The chemical difference in CHS-J samples produced in different years might be attributed to the unstable components and the chemical variation of herbal medicines used in different years. To further confirm the instability of components in CHS-J sample during storage, the accelerated tests of the newly produced CHS-J (the year 2020) were conducted. In the temperature-accelerated tests, 100 ml of CHS-J samples (n = 8) were placed in clean glass bottles with airtight rubber seal flip caps and maintained at constant 60 ± 2°C and shield from light. 0.5 ml of liquor from each bottle was collected every 2 weeks and continuously collected for 8 weeks. In light-accelerated tests, 100 ml of CHS-J samples (n = 8) were exposed at the light (5500 ± 500 Lux, 50% R.H., 25°C), 0.5 ml of liquor was collected every 3 until 12 weeks. All collected liquor samples were stored at −80°C until analysis.

To reveal the potential degradation pathways, the selected reference standards were also subjected to temperature-accelerated and light-accelerated tests as described above. The reference standards were accurately weighed and dissolved in 35% (v/v) of base liquor individually at the final concentration of 20 μg ml⁻¹ with 3 replicates. 0.5 ml of solution from each bottle was collected weekly. All collected samples were stored at −80°C until UHPLC-Q-TOFMS analysis.

Due to containing several herbal extracts and the base liqueur, the CHS-J is a highly complex matrix consisting of numerous organic compounds with a wide range of concentrations. Under the optimized chromatographic and MS conditions, the major components in CHS-J were well separated and detected in both positive and negative ion modes. The representative base peak intensity (BPI) chromatograms of the QC sample are illustrated in Figure 1A. A pooled QC sample was used for the method validation of UHPLC/Q-TOFMS. The raw data acquired was pre-treated using Progenesis QI for peak picking and alignment. After “80% rule” processing, 2775 and 1143 variables were generated from the QC samples detected in positive and negative ion modes, respectively. These ions are fairly evenly distributed in whole chromatograms (Figure 1B). Furthermore, the relative standard deviation (RSD) distribution of these ions in QC samples was commonly used to evaluate the stability and reproducibility of untargeted metabolomics analysis. As shown in Figure 1C, 91.7 and 96.8% of the variables with RSD less than 30% were observed in positive and negative ion modes, respectively, along with 71.2 and 79.9% of the ions with RSD ≤10%, which demonstrated the excellent stability and reliability of the UPLC/Q-TOFMS method for untargeted metabolomics.

To characterize the chemical differences across CHS-J samples with different storage years, CHS-J samples produced by 2014, 2015, 2016, and 2018 were used for untargeted metabolomics analysis. After data preprocessing, the partial least squares discriminant analysis (PLS-DA), a supervised multivariate data analysis, was constructed. Sevenfold cross-validation was conducted to validate the quality of the PLS-DA model. The parameters R² and Q² represent the interpretability of the variable and the model’s predictability, respectively. The cumulative values of R²X, R²Y, and Q² were close to the optimal value of 1.0, indicating the established models with excellent predictive capability and fitness (Wan et al., 2013). As depicted in Figure 2A, all tested CHS-J samples from four different production years were clustered in PLS-DA score plots in both positive (R²X = 0.596, R²Y = 0.964, Q² = 0.943) and negative (R²X = 0.702, R²Y = 0.98, Q² = 0.963) ion modes, indicating the significant differences in the chemical profile of CHS-J samples produced in the different years. To reveal the discrepant components contributing most to the group separation, the differentiated compounds were picked by S-plot derived from the OPLS-DA model constructed by CHS-J samples produced between the year 2014 and year 2018 (Figures 2B,C). Variables with variable importance in the projection (VIP) > 2 were tentatively highlighted to be important for discrimination (Figures 2B,C). The variables were further filtered by p < 0.05 and fold-change FC > 2. As a result, 39 and 14 variables were highlighted in positive and negative ion modes, respectively. The structures were tentatively assigned by an in-house database and further validated by their reference standards. Finally, 7 compounds were accurately identified, as shown in Table 1. The intensity tendency of these compounds in CHS-J samples with different storage years was analyzed, and most of the identified compounds had a gradually decreasing trend along with the increasing storage years (data not shown).

To confirm the findings from untargeted metabolomics, these seven highlighted components in CHS-J samples produced in different years were quantitatively determined by UHPLC-QQQ-MS/MS in multiple reaction monitoring (MRM) mode as described previously (Hu et al., 2021). As shown in Figure 3, the concentrations of seven highlighted components across CHS-J samples produced in 2018 were higher than those in tested samples produced in other years, except that no significant difference in the levels of echinacoside or magnoflorine between samples 2018 and 2016.

The difference in these compound levels in CHS-J samples produced in different years might be attributed to the instability of components and the chemical variation of herbal medicines used in different years. To further confirm the instability of components in CHS-J sample, the accelerated tests of the newly produced CHS-J (the year 2020) were conducted. The CHS-J samples were continuously collected for 8 and 12 weeks in temperature-accelerated and light-accelerated tests, respectively. The samples were collected every 2 weeks (T₀, T₂, T₄, T₆, T₈) in temperature-accelerated tests and every 3 weeks (L₀, L₃, L₆, L₉, L₁₂) in light-accelerated tests were analyzed by the developed UHPLC-Q-TOFMS. To visualize the chemical profile across groups, the principal component analysis (PCA), an unsupervised data analysis, was conducted. As shown in Figure 4A, all observations were located in the Hotelling T2 (0.95) ellipse, PCA score plots derived from both positive (R²X = 0.727, Q² = 0.633) and negative (R²X = 0.899, Q² = 0.729) ion modes exhibited tight clusters of CHS-J samples from each time point in temperature-accelerated tests and discrimination across groups. Importantly, the longer treatment by high temperature, the farther the cluster of CHS-J samples departs from untreated CHS-J group (T₀), indicating the more remarkable changes in the component profile of CHS-J samples along with an increasing treatment time. Similar results were observed in the light-accelerated tests (Figure 4B). To reveal the components contributing most to the group separation, the supervised OPLS-DA of temperature- (T₀ vs. T₈) and light-accelerated group (L₀ vs. L₁₂) were further constructed. The variables with VIP > 2 and p < 0.05 were highlighted as important components for the discrimination. As a result, 10 out of 48 and 9 out of 54 differential variables were identified by their reference standards in temperature- (Supplementary Table S2) and light- (Supplementary Table S3) accelerated groups, respectively. Among them, 8 compounds, including Z-ligustilide, echinacoside, magnoflorine, sagittatoside B, epimedin B, icariin, acteoside, and epimedin C, were common features between temperature- and light-accelerated tests (Figure 5A), indicating that these compounds might be sensitive to both temperature and light. While crustecdysone and chikusetsu saponin iva were thermosensitive components, 2′′-O-rhamnosyl Icariside II was photosensitive (Figure 5A). Furthermore, a Venn diagram for CHS-J samples produced by different years and accelerated test datasets show 7 common features (Figure 5B), which indicated that the instability of these components might largely contribute to the differences in CHS-J samples with different storage periods.

To further investigate the possible mechanisms underlying these chemical changes, accelerated tests of several reference standards were conducted. The individual reference standards with severe degradation, including chlorogenic acid, echinacoside, baohuoside I, 2″-O-rhamnosyl icariside II, Z-ligustilide dissolved in 35% (v/v) of base liquor were subjected to temperature-accelerated or light-accelerated tests, respectively. The solution before and after accelerated tests were analyzed by a UPLC/Q-TOFMS. The degradation products derived from these reference standards exposed to temperature and light were tentatively assigned by matching their MS and MS/MS data with that of the literatures (Xie et al., 2011; Dawidowicz and Typek, 2010; Zanoelo and Benincá, 2009; Zuo et al., 2013; Yan et al., 2018) and summarized in Supplementary Table S3. Taking chlorogenic acid as an example, the total ion chromatogram (TIC) of chlorogenic acid alcohol base solution (T₀) and chlorogenic acid alcohol base solution after 16 weeks (T₁₆) in temperature-accelerated tests was shown in Figure 6A. Chlorogenic acid was degraded seriously when exposed to high temperatures, and several new products were generated. A newly generated degradation peak at RT = 0.75 min in chlorogenic acid solution after 16 weeks of high-temperature treatment was identified as quinic acid. The presence of quinic acid in alcohol base solution is not surprising because chlorogenic acid is an ester comprised of caffeic and quinic acids. However, caffeic acid was not found in this solution. Two new induced peaks at RT = 1.48 and 1.94 min were identified as 3-O-caffeoylquinic acid and 4-O-caffeoylquinic acid, respectively. The proposed degradation mechanism was speculated and illustrated in Figure 6B. In the presence of hydroxide ion [OH]⁻, the nucleophile on the neighboring hydroxyl group of the quinic acid ring attacked the carbonyl carbon of 5-O-caffeoylquinic acid, which resulted in the internal transesterification of trans-5-caffeoylquinic acid. The caffeoyl moiety transferred from the C5- position to the 4- and 3-positions via ortho-cyclic ester intermediate structures, which further hydrolyzed into esters (Zanoelo and Benincá, 2009; Dawidowicz and Typek, 2010; Xie et al., 2011). But it is difficult to distinguish the two compounds due to their similar MS and MS/MS pattern. Furthermore, chlorogenic acid was also degraded seriously when exposed to light, and its degradation products mainly are the hydroxylated chlorogenic acids (Supplementary Figure S1). Our data indicated that the degradation mechanism of most compounds was hydrolysis and isomerization, which is related to the ethyl alcohol-water environment of CHS-J, while high temperature and light accelerated these degradation processes with varying degrees.

Compared to the control group, the exhaustive swimming time, hepatic glycogen level, muscle glycogen level, blood lactic acid level, urea nitrogen level, and lactic dehydrogenase level in both CHS-J samples produced in 2014 and 2020 groups showed a significant difference (p < 0.01), except hepatic glycogen and lactic dehydrogenase in CHS-J samples produced in 2020 group (p < 0.05), indicating that storage period had little influence on the anti-fatigue effect of CHS-J samples (Table 2). Moreover, compared with CHS-J samples produced in the 2020 group, CHS-J samples produced in the 2014 group had a decreased blood lactic acid and a higher level of lactic dehydrogenase. However, for the exhaustive swimming time, hepatic glycogen level, muscle glycogen level, and urea nitrogen level, there were no significant differences in these four anti-fatigue indexes between CHS-J samples produced in the 2014 group and 2020 group. Collectively, the chemical changes during the storage period had little influence on the anti-fatigue effect of CHS-J samples.