Rapid multi-component quantification and quality assessment of Relinqing® granule (Polygonum capitatum) by near-infrared spectroscopy and chemometrics

Summary

Relinqing® granule (RLQ), derived from Polygonum capitatum (P. capitatum), is traditionally used to treat urinary system disorders, but its quality control remains limited to gallic acid content in the 2020 Chinese Pharmacopeia. This narrow standard restricts comprehensive quality evaluation and broader use in functional foods and nutraceuticals. To address this issue, this study identified additional bioactive compounds and established a multi-marker quality control strategy. Twenty compounds were isolated from P. capitatum, among which gallic acid, ellagic acid, and quercetin showed significant anti-inflammatory activity through IL-6 inhibition. A near-infrared spectroscopy (NIRS) method combined with partial least squares (PLSs) regression was developed to predict the contents of these active compounds, using ultra-performance liquid chromatography (UPLC) as the reference. The resulting NIRS models showed high predictive accuracy, with correlation coefficients of 0.9000–0.9999, supporting their use for rapid, non-destructive, and cost-effective quality assessment of RLQ and related plant-derived products.

Graphical abstract

Highlights

• •NIRS-PLS models enable rapid multi-component quantification of Relinqing® Granule
• •Gallic acid, ellagic acid, and quercitrin are quality markers with anti-inflammatory activity
• •Twenty compounds were identified from Polygonum capitatum
• •NIRS provides rapid non-destructive quality control for TCM preparations

Spectroscopy; Analytical chemistry instrumentation; Molecular modeling

Introduction

Relinqing® granule (RLQ) is a widely used Chinese patent medicine for treating urinary system diseases, including urinary tract infections, pyelonephritis, and prostatitis.¹ Listed in the 2015 edition of the Chinese Pharmacopeia,² RLQ has shown notable clinical efficacy, a good safety profile, and broad-spectrum applications.^1,3,4 In traditional Chinese medicine (TCM), RLQ is prescribed to relieve symptoms such as frequent, urgent, and painful urination—typically associated with damp-heat syndromes in the lower jiao.³ Moreover, it is often combined with antibiotics to improve therapeutic outcomes.⁵ The granules are derived from a water-soluble extract of Polygonum capitatum (P. capitatum), known in Chinese as Tou-hua-liao and Latin name as Polygonum capitatum Buch.-Ham. ex D. Don).^6,7 P. capitatum holds considerable medicinal value within Miao culture in China and possesses a long-standing history of both economic significance and medicinal application. This plant is primarily distributed across provinces such as Guizhou, Yunnan, Sichuan, Tibet, and Guangxi, among others.⁸ According to the “Guangxi Traditional Chinese Medicine Annals,” P. capitatum is known for its ability to dispel wind, alleviate blood stasis, and relieve pain.⁹ This herb has been documented in the esteemed work “Yunnan Chinese Herbal Medicine,” where it is noted for its pharmacological effects of clearing heat and promoting diuresis. Clinical studies have demonstrated that P. capitatum exhibits significant therapeutic efficacy in treating various systemic diseases, particularly those affecting the lower urinary tract.¹⁰ Beyond its pharmaceutical use, P. capitatum is a natural source of bioactive compounds such as phenolic acids and flavonoids, which are increasingly recognized for their relevance to functional foods and nutraceuticals aimed at supporting urinary tract health and overall well-being. Modern pharmacological research has revealed that this herb possesses anti-inflammatory, antibacterial, antioxidant, and hypoglycemic properties.^11,12,13

Ensuring the quality, stability, and therapeutic efficacy of TCM preparations is critical for their clinical use.¹⁴ Pharmacological studies have revealed that phenolic acids and flavonoids are the principal bioactive constituents in RLQ.^15,16 However, the current 2020 edition of the Chinese Pharmacopeia only mandates the quantification of gallic acid (GA) for quality control, overlooking other significant compounds such as ellagic acid (EA) and quercitrin.¹⁷ Such a single-marker approach not only limits comprehensive quality evaluation in pharmaceutical contexts but also poses challenges for developing RLQ and related herbal products into standardized health supplements or functional foods. Ultra-performance liquid chromatography (UPLC) and high-performance liquid chromatography (HPLC) are the primary methods commonly employed for quality control of RLQ.^18,19 Despite their accuracy, these methods are time-consuming, labor-intensive, and require extensive sample preparation. Therefore, establishing a rapid, reliable, and comprehensive analytical approach remains an urgent challenge for the standardization of RLQ and similar herbal formulations. Near-infrared spectroscopy (NIRS) has emerged as a promising alternative to traditional chromatographic techniques, owing to its simplicity, speed, cost-effectiveness, and non-destructive nature.^20,21 Compared to conventional methods, NIRS enables rapid analysis, minimal sample preparation, and ease of operation.²² When combined with chemometrics, NIRS can simultaneously quantify multiple bioactive components in complex matrices, making it valuable not only for TCM but also for the quality control of botanical ingredients in the food and nutraceutical industries.²³ In recent years, NIRS has been widely applied across fields such as food safety, pharmaceuticals, and traditional medicine, especially for real-time monitoring and rapid quantification in complex mixtures.²⁴ When combined with chemometrics—the use of multivariate statistical and mathematical techniques to extract meaningful patterns from spectral data—NIRS becomes a powerful analytical platform for the quality evaluation of multi-component herbal medicines.^25,26,27,28

This study aims to systematically investigate the active constituents of RLQ, from extraction and purification to structural identification, and to develop a rapid, robust, and non-destructive quantitative model based on NIRS calibrated with UPLC reference data. By integrating spectroscopic and chromatographic techniques with chemometrics, we seek to establish a comprehensive quality control approach for RLQ that can be applied in pharmaceutical contexts, thus promoting the broader utilization of P. capitatum in health-related industries.

Material and methods

Chemicals and reagents

Acetonitrile and methyl alcohol (HPLC grade) were supplied by Fisher Chemicals (Pittsburgh, USA). Phosphoric acid (HPLC grade) was obtained from Chengdu Kelong Chemical Co., Ltd. (Chengdu, China). Distilled water (HPLC grade) was acquired by Beijing Watsons Distilled Water Co., Ltd (Beijing, China). All other reagents were of analytical grade. Reference standards (purity >98%), including GA (PS012548), EA (PS013592), and quercitrin (PS013787) were acquired from Chengdu Push Bio-technology Co., Ltd (Sichuan, China).

Main instruments

Semipreparative HPLC was executed on an Agilent 1200 Series system equipped with a C₁₈ reversed-phase column (Eclipse XDB-C₁₈,5 μm; 9.4 mm∗250 mm) (Agilent Technologies, Palo Alto, USA). UV spectra were measured on a Shimadzu UV-2401PC spectrophotometer (Shimadzu, Kyoto, Japan). 1D NMR spectra were obtained on a Bruker AV 600 spectrometer (Bruker, Bremerhaven, Germany) with tetramethylsilane (TMS) as the internal standard. Chemical shifts (δ) are expressed in ppm with reference to the solvent signals. ESI-MS was run on a Shimadzu LC-MS-IT-TOF mass spectrometer (Shimadzu, Kyoto, Japan), and HR-ESI-MS was recorded using an Agilent 6200 series TOF instrument (Agilent Technologies, Palo Alto, US). Column chromatography (CC) was performed on silica gel (200–300 mesh, 300–400 mesh, Qingdao Haiyang Chemical Co. Ltd., Qingdao, People’s Republic of China). Lichroprep RP-18 (40–63 μm, Merck, Darmstadt, Germany) and Sephadex LH-20 gel (40–70 μm, Amersham Pharmacia Biotech AB, Uppsala, Sweden) were used. Precoated silica gel GF254 plates (Qingdao Haiyang Chemical Plant). The NIR spectrometer for this study was purchased from PerkinElmer Frontier (PE Frontier, MA, USA). The UPLC was purchased from Waters ACQUITY UPLC® (Waters Corporation, Milford, USA).

Plant materials

P. capitatum was provided by Guizhou Weimen Pharmaceutical Co., LTD in December 2023. It was produced in Guiyang City, Guizhou Province. It was identified as dried whole P. capitatum by Prof. Yi Zhang of Chengdu University of TCM and stored in the State Key Laboratory of Chengdu University of TCM (P. capitatum: 20231230).

RLQ granules

A total of 50 batches of RLQ are provided by Guizhou Weimen Pharmaceutical Co., LTD. The batch number is: 240101; 240102; 240201; 240202; 240301; 240302; 240401; 240402; 240501; 240502; 240601; 240602; 24603; 24604; 24605; 24606; 24607; 24608; 24609; 240610; 240611; 240612; 240613; 240614; 240701; 240702; 240703; 240704; 240705; 240706; 240707; 240708; 240709; 240710; 240711; 240712; 240813; 240814; 240815; 240816; 240817; 240818; 240819; 240820; 240821; 240822; 24,902; 230103; 230908; 231002. According to the specifications outlined in the Chinese Pharmacopoeia (2025 edition), the preparation of RLQ follows a standardized protocol as described later in discussion: P. capitatum (1,250 g) was employed as the principal raw material. Following the extraction procedure consistent with the pharmacopoeial requirements, the resulting filtrate was concentrated to a suitable volume, subjected to further filtration for purification, and subsequently processed via spray drying to obtain the final RLQ.

Extraction and isolation

The sample of P. capitatum (10.0 kg) was cut into pieces and subjected to reflux extraction with 80% EtOH four times (1.5 h each), yielding EtOH extract (1.8 kg). The extract was then processed via macroporous adsorption resin CC, using sequential gradient elution with H₂O (889.5 g), 30% EtOH (189.9 g), 60% EtOH (143.5 g), and 90% EtOH (241.5 g). The 60% EtOH-eluted fraction was further separated by MCI CC with a H₂O/MeOH gradient (10%–70%), yielding 10 crude fractions (Fr.1–Fr.10). Fr.1 (2.3 g) was purified by silica gel CC (100–200 mesh) with a CH₃Cl₂/MeOH gradient (20:1–0:1), followed by Sephadex LH-20 chromatography (MeOH), to afford compounds 1 (6.4 mg), 3 (6.8 mg), and 5 (10.5 mg). Fr.2 (30.8 mg) was purified by Sephadex LH-20 chromatography (MeOH) and semi-preparative HPLC (ACN/H₂O 30:70, flow rate 3 mL/min, detection wavelength 300 nm), yielding compounds 2 (6.4 mg, t_R = 7 min), 4 (4.4 mg, t_R = 10 min), and 8 (4.5 mg, t_R = 12 min). Fr.3 (10.9 g) was fractionated by silica gel CC with a CH₃Cl₂/MeOH gradient (20:1–0:1), yielding four subfractions (Fr.3.1–Fr.3.4). Fr.3.3 was further purified by Sephadex LH-20 chromatography (MeOH) and semi-preparative HPLC (ACN/H₂O 35:65, flow rate 3 mL/min, detection wavelength 300 nm), yielding compounds 6 (8.9 mg), 7 (7.9 mg), and 10 (9.4 mg). Fr.3.4 was purified by Sephadex LH-20 chromatography (MeOH) to afford compounds 19 (5.8 mg) and 20 (4.8 mg). Fr.4 (8.7 g) was purified by Sephadex LH-20 chromatography (MeOH) and semi-preparative HPLC (ACN/H₂O 25:75, flow rate 3 mL/min, detection wavelength 300 nm), yielding compounds 9 (6.4 mg, t_R = 5 min) and 11 (4.5 mg, t_R = 10 min). Fr.5 (3.4 g) was purified by Sephadex LH-20 chromatography (MeOH) and semi-preparative HPLC (ACN/H₂O 35:65, flow rate 3 mL/min, detection wavelength 300 nm), yielding compounds 12 (5.4 mg, t_R = 6 min) and 13 (4.4 mg, t_R = 8 min). Fr.6 (10.4 g) was fractionated by silica gel CC (100–200 mesh) with a CH₃Cl₂/MeOH gradient (20:1–0:1), yielding five subfractions (Fr.6.1–Fr.6.5). Fr.6.1 was purified by Sephadex LH-20 chromatography (MeOH) to afford compounds 14 (10.9 mg) and 18 (8.4 mg). Fr.6.3 was purified by semi-preparative HPLC (ACN/H₂O 25:75, flow rate 3 mL/min, detection wavelength 230 nm), yielding compounds 16 (6.4 mg, t_R = 7 min) and 17 (4.5 mg, t_R = 10 min). Fr.6.4 was purified by Sephadex LH-20 chromatography (MeOH) to afford compound 15 (4.8 mg). The flowchart of extraction and isolation was shown in Figure 1.

Figure 1.

The flowchart of extraction and isolation

Biological activity procedures

Cell viability using the CCK8 method

Through previous research and development, it was found that the main components in P. capitatum are flavonoids and phenolic acids, which are presumed to be the main anti-inflammatory active ingredients of P. capitatum.²⁹ Further investigations were carried out to assess the anti-inflammatory activities of the isolated flavonoids and phenolic acids. The cytotoxicity of nicotiflorin (compound 1), narcissoside (compound 2), isoquercitrin (compound 3), astragalin (compound 4), isovitexin (compound 5), luteolin (compound 6), quercetin (compound 7), quercitrin (compound 8), GA (compound 9), EA (compound 10) were investigated by using the cell counting kit 8 (CCK-8) assay. RAW264.7 cells in the logarithmic growth phase were utilized for this experiment. After counting with a cell counting plate, 100 μL of cell suspension containing 5 × 10⁴ cells was seeded onto a 96-well plate and cultured at 37°C and 5% CO₂ for 24 h. After 24 h of culturing at 37°C and 5% CO₂, ten compounds were added at concentrations of 10, 20, 40, 80, and 160 μM in DMEM for another 24 h. DMEM was added to the 96-well plate, the blank group, and the LPS group. After 4 h of continuous culturing, 100 μL of non-stimulated DMEM was added to the blank control group. The LPS group and the experimental group were induced with LPS solution at a concentration of 1 μg/mL for 24 h. After replacing the medium with fresh DMEM, 10 μL of CCK-8 reagent was added to each well, and the wells were incubated at 37°C for 4 h. Each group was set up with 3 replicate wells and blank wells (with DMEM and CCK-8 solution, but without cells). The absorbance was measured at a wavelength of 450 nm. The data are presented as means ± standard deviation. Cell viability was calculated as follows:

$$Cellviability(\%)=\left(ODoftreatmentwell-ODofblankwell\right)/\left(ODofblankgroup-ODofblankwell\right)\times 100\%$$

Based on the experimental results of CCK-8, the concentration groups of 10, 20, 40, 80, and 160 μM, 10 compounds were selected for further experiments. Accordingly, a structured experimental design was implemented, comprising a blank control group, an LPS-induced inflammatory model group, and treatment groups exposed to compound concentrations of 10, 20, 40, 80, and 160 μM. Each group was provided with 3 replicate wells. The experiment was performed in triplicate.

Measurement of nitric oxide in RAW264.7 cells

Nitric oxide (NO) levels were measured following the procedure described by Ying et al.³⁰ RAW264.7 cells were pretreated with compounds at a concentration range of 10–160 μM for 24 h; after that, the cells were stimulated with LPS (1 μg/mL) for 24 h. Then, Griess reagent was used to evaluate the NO production in the medium. For the blank groups, a sterile culture medium was used. The quantity of nitrite in the samples was measured by using the sodium nitrite standard curve as a reference point.

Enzyme-linked immunosorbent assay (ELISA) for detecting TNF-α, IL-1β, and IL-6 production

The ELISA assay for detecting the production of TNF-α, IL-1β, and IL-6 was performed by pretreating the cells with compounds for 24 h and then inducing them with LPS (100 ng/mL) for 24 h. The culture supernatant was evaluated using commercially available ELISA kits. All procedures were carried out by following the manufacturer’s instructions. The samples were analyzed in triplicate.

Content determination of quality control components

Chromatographic conditions

To measure the RLQ samples with the Waters ACQUITY UPLC® (Waters Corporation, Milford, USA) equipped with a binary pump, mobile phase degasser, temperature-controlled autosampler, column thermostat, and PDA detector. The chromatographic column was an ACQUITY UPLC BEH C₁₈ column (2.1 mm × 100 mm, 1.7 μm). The mobile phase consisted of water containing 0.4% phosphoric acid (solvent A) and methanol (solvent B). The gradient elution was as follows: 0%–2% B for 0–3 min, 2%–3% B for 3–5 min, 3∼3% B for 5–6 min, 3%–4% B for 6–7 min, 4%–14% B for 7–15 min,14%–16% B for 15–20 min, 16%–20% B for 20–25 min. The flow rate was kept at 0.3 mL/min, and the injection volume was 2 μL. The column was 30°C, and the UV detection wavelength was set at 254 nm.

Preparation for sample solution and method validation

Each sample powder (1.0 g) was ultrasonically extracted with 75 mL of 75% methanol for 30 min (300 W, 40 kHz). Following extraction, 75% methanol was added to compensate for the lost weight during extraction. The extract solution was filtered through a 0.22 μm membrane filter before aliquots were subjected to UPLC analysis. Reference standards of GA, EA, and quercitrin were accurately weighed and dissolved in methanol to get stock solutions, in which the concentrations of GA were 1.012 mg/mL, the concentrations of EA were 1.057 mg/mL, and the concentrations of quercitrin were 1.252 mg/mL. Then, this solution was successively diluted to appropriate concentrations to construct calibration curves, and validation of the method for determination using stability, precision, repeatability, and recovery rate tests.

Spectral information acquisition

NIR spectral acquisition

To measure the RLQ samples with the NIR spectrometer (PE Frontier, MA, USA) equipped with a diffuse reflection module. Take about 3 g of the sample powder into the quartz sample cup, lay it evenly and smoothly, and use the integrating sphere diffuse reflection method to collect the map according to the following conditions: the resolution is 8 cm⁻¹, the scanning wave number range is 10000–4000 cm⁻¹, and the signal is scanned 64 times in total. The room temperature was controlled to 22°C, and the relative humidity was 45%. Each sample was scanned three times, and its average spectrum was taken as the final sample spectrum.

Chemometrics analysis

The pre-processing of NIR spectra

In addition to valuable sample information, the original spectra obtained through NIR spectroscopy also encompass background and noise components that compromise the reliability and precision of the model. Consequently, it is imperative to perform spectral preprocessing on the constructed infrared spectral model. In this study, multiple spectral preprocessing techniques were applied to optimize the quality and interpretability of the NIR data. These methods included Savitzky-Golay derivative (SGD), multiplicative signal correction (MSC), standard normal variate (SNV), Savitzky-Golay filter (SGF), Gaussian filter (GF), and median filter (MF). Additionally, combinations of these techniques were also applied.

Among these methods, MSC and SNV can mitigate the scattering effects caused by the uneven distribution and varying sizes of sample particles.³¹ SGF and SGD are effective in reducing baseline drift and background translation interference in NIR spectra, thereby enhancing resolution and sensitivity.³² GF and MF can efficiently eliminate noise signals within the spectra.³³

The quantitative model

In this study, a total of 50 batches of RLQ samples were collected and partitioned into calibration and verification sets using the Kennard-Stone (KS) method at a ratio of 8:2. Specifically, 40 batches were allocated to the calibration set, while 10 batches were designated for the validation set.

Screening feature variables can effectively eliminate irrelevant information, thereby enhancing both modeling speed and accuracy. This study further refined the characteristic spectral variables based on the aforementioned optimal pretreatment methods, primarily investigating the following variable screening techniques: moving window partial least squares (mwPLSs), competitive adaptive reweighted sampling (CARS), random forest (RF), and uninformative variable elimination (UVE). All methods were performed using MATLAB (Version R2023a; MathWorks, Natick) software.

Partial least squares (PLSs) is a regression method that is widely utilized in stoichiometry.³⁴ By employing potential variables to establish a correlation model between sets of observed variables, PLS can effectively capture the maximum correlation between sample spectral data and physicochemical values, thereby facilitating quantitative analysis.³⁵ This method adeptly addresses challenges associated with a large number of variables, significant data noise, and multicollinearity among variables. In this context, matrix (X) represents the NIR spectral data, while matrix (Y) denotes the content of three active ingredients in RLQ. Regression models were established for each respective dataset. The model performance was evaluated using the coefficient of determination (R²), the root-mean-square error of calibration (RMSEC), the root-mean-square error of cross-validation (RMSECV), and relative prediction deviation (RPD). A value of R² closer to 1, along with RMSEC and RMSECV approaching 0, indicates a higher accuracy of the constructed PLS model. The scoring criteria for RPD are as follows: RPD ≥2.5 signifies excellent performance; 2.5 > RPD ≥1.8 denotes good performance; 1.8 > RPD ≥1.4 reflects normal performance; and RPD <1.4 indicates poor performance.³⁶ PLS was performed by MATLAB (Version R2023a; MathWorks, Natick) software.

Results and discussion

Structural elucidation

Twenty compounds were isolated from P. capitatum (Figure 2), including nicotiflorin (1),³⁷ narcissoside (2),³⁸ isoquercetin (3),³⁹ astragalin (4),⁴⁰ isovitexin (5),⁴¹ luteolin (6),⁴¹ quercetin (7),⁴¹ quercitrin (8),⁴² GA (9),⁴³ EA (10),⁴⁴ protocatechuic acid (11),⁴⁵ caffeic acid (12),⁴⁶ trans-p-hydroxycinnamic acid (13),⁴⁷ 5-hydroxymethylfurfural (14),⁴⁸ isohematinic acid (15),⁴⁹ guanosine (16),⁵⁰ adenosine (17),⁵¹ α-D-glucopyranosyl-(1→1′)-3′-amino-3′-deoxy-β-D-glucopyranoside (18),⁵² byakangelicin (19),⁵³ and 1-p-hydroxy-cis-cinnamoyl cinnamic acid (20).⁵⁴ These compounds were identified by the comparison of their spectroscopic data with those reported in the literature.

Figure 2.

The chemical structures of compounds 1–20 isolated from P. capitatum

NMR data of compounds

NMR data of 1–20 compounds were included in the supplemental information.

The effect of 10 compounds on RAW 264.7 cell viability, NO release, and expression levels of inflammatory factors IL-6, IL-1β, and TNF-α

The effect of ten compounds on RAW264.7 cell viability was detected by CCK-8 assay (Figure 3). The results showed that compounds 1–10 exhibited cytotoxicity toward RAW264.7 cells at concentrations of 160, 80, 40, 20, and 10 μM. Quercitrin, GA, EA, nicotiflorin, and isoquercitrin showed no significant cytotoxicity to RAW264.7 cells within the 10–160 μM range. In contrast, narcissoside, quercetin, and isovitexin demonstrated significant cytotoxicity at 160 μM (p < 0.001). Luteolin exhibited marked cytotoxicity at 80 μM (p < 0.001). After LPS treatment, NO release and inflammatory cytokine levels in the model group were significantly elevated. Compared with the control group, the LPS group showed a significant increase in NO production (p < 0.001), confirming successful establishment of the inflammatory model (Figure 4). Compared to the LPS group, quercitrin, GA, EA, nicotiflorin, and isoquercitrin inhibited NO generation to varying degrees within the 10–160 μM range. Quercitrin (160 μM) reduced NO content by 90.76% (p < 0.001). GA (160 μM) reduced NO content by 66.51% (p < 0.001). EA (160 μM) reduced NO content by 55.02% (p < 0.001). Isoquercitrin (160 μM) reduced NO content by 57.68% (p < 0.001). Nicotiflorin (160 μM) reduced NO content by 67.66% (p < 0.001). Luteolin suppressed NO production in a concentration-dependent manner (0.5–40 μM), achieving an 84.47% reduction at 40 μM (p < 0.001). In terms of inflammatory cytokine levels (Figure 5), significant differences in cytokine levels were observed between the control and model groups (p < 0.001), further validating model success. Compared to the model group, quercitrin, GA, and EA significantly reduced IL-6 levels at 10–160 μM (p < 0.001), while IL-1β and TNF-α levels remained unchanged. The results indicated that quercitrin, GA, and EA mainly exerted anti-inflammatory effects by acting on IL-6.

Figure 3.

Effects on cell viability for RAW264.7 cells treated with PLS of different concentrations

Notes: #p < 0.05 and ##p < 0.01 vs. blank control group; ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001 vs. LPS, n = 3 (mean ± standard deviation). Statistical analysis was performed using one-way ANOVA followed by Dunnett’s post-hoc test.

Figure 4.

Effects on NO release amount for RAW264.7 cells with the ten compounds isolated from P. capitatum

Notes: ###p < 0.001 vs. blank control group; ∗∗∗p < 0.001 vs. LPS, n = 3 (mean ± standard deviation). Statistical analysis was performed using one-way ANOVA followed by Dunnett’s post-hoc test.

Figure 5.

Effects on the inflammatory factor expression level of IL-6, IL-1β, and TNF-α in RAW 264.7 cells treated with compounds isolated from P. capitatum

Notes: ###p < 0.001 vs. blank control group; ∗∗∗p < 0.001 vs. LPS, n = 3 (mean ± standard deviation). Statistical analysis was performed using one-way ANOVA followed by Dunnett’s post-hoc test.

Simultaneous determination of three anti-inflammatory compounds in RLQ by UPLC

Sequentially obtain each control solution and inject it into the sample to quantify the peak area. Linear regression analysis was performed with the injection volume of each control as the horizontal coordinate and the peak area as the vertical coordinate. UPLC chromatograms of three active compounds in RLQ are shown in Figure 6A. The regression equation and linear range are shown in Table 1. The method was validated through assessments of stability, precision, recovery rates, and repeatability. The results were presented in Table 1. The relative standard deviation (RSD) for precision, stability, and repeatability of all three components was less than 3% (Table 1). Furthermore, a strong linear relationship existed between the injection volume of each component and its corresponding peak area. The recovery rates for each component fall within the range of 98%–101%. All these results indicated that the method was accurate and feasible. Compare the content of three components in different batches of RLQ using the boxplot (Figure 6B). According to the boxplot, the content of GA and quercitrin is higher than that of EA, and there are differences in content between batches. The differences in component content may be due to differences in the prescription herbs, such as differences in origin, harvesting time, processing methods, and so forth, which may lead to differences in the composition of the prepared prescription. Therefore, the results of the measurements demonstrate the need and importance of controlling the content of ingredients to ensure quality.

Figure 6.

Content of three chemical components in RLQ

(A) Liquid chromatogram of RLQ. (B) The boxplot of the content of three components in the fifty batches of RLQ, data are represented as mean ± SEM.

Table 1.

Examination of linear relationships

Component	Regression equation	r	Linear range (mg/mL)	Repeatability RSD (%)	Precision RSD (%)	Stability RSD (%)	Recovery
GA	y = 5,896,784.3137 x + 25,429.6667	0.9998	0.01590–0.51000	0.48	1.41	1.28	100.98
EA	y = 99,137,736.0815 x – 3,967.5423	0.9998	0.00009∼0.00300	0.99	1.03	1.27	98.94
Quercitrin	y = 8,326,008.1192 x + 979.3532	0.9998	0.00180∼0.05900	0.86	1.37	0.98	99.12

The quantitative models of three anti-inflammatory compounds in the RLQ based on NIRS

The NIR spectral characters of RLQ

RLQ has NIR spectra, such as Figure 7 in the 10000 ∼ 4000 cm⁻¹, it can be seen that the characteristic peaks of the NIR spectrum were mainly located at 8336 cm⁻¹, 6860 cm⁻¹, 5680 cm⁻¹, 5170 cm⁻¹, 4760 cm⁻¹, and 4340 cm⁻¹. The absorbance at 8336 cm⁻¹ is from the second overtone of C-H stretching. The 5170 cm⁻¹ result from the second overtone of C=O stretching, the peak at 4340 cm⁻¹ approaches the stretchable and curved frequency absorption bands of –CH₃ and O-H.⁵⁵

Figure 7.

The NIR spectroscopy spectra of RLQ samples

Sample division and selection of the spectral pretreatment method

The KS method was used to divide the sample into a calibration set and a validation set in an 8:2 ratio. The division of these sets is detailed in Table 2. The results indicated that the concentration range of the validation set was entirely encompassed within that of the calibration set, thereby satisfying the modeling requirements. SGD, MSC, SNV, SGF, GF, MF, and combinations of these techniques were used to explore the spectral pre-processing methods suitable for the establishment of a quantitative model. The differences among models obtained through various pretreatment methods and their combinations were summarized in Table 3. The preprocessed spectrum is shown in Figure 8. Based on the principle of the highest accuracy rate of the model, the best pre-processing method was selected. The results showed that GA had the best effect after the MSC+SGD treatment, with an R² value of 0.9996, RMSEC of 0.0075, and RMSECV of 0.1101. EA had the best effect without pretreatment, R² value was 0.9507, RMSEC was 0.0027, RMSECV was 0.0078. Quercetin had the best effect after the SGD pretreatment with an R² value of 0.9994, RMSEC of 0.0030, and RMSECV of 0.0285. Therefore, the experimental findings demonstrated that suboptimal spectral processing may compromise the integrity of spectral information, thereby adversely affecting the predictive accuracy of analytical models. These findings emphasized that careful selection of optimal preprocessing techniques was critical to enhancing the robustness and accuracy of multicomponent quantitative analysis models for the simultaneous determination of the three bioactive compounds in chemometric applications.

Table 2.

Data statistics of the calibration set and validation set

Active components	Calibration set				Validation set
	Maximum value (%)	Minimum value (%)	Average value (%)	Maximum value (%)	Minimum value (%)	Average value (%)
GA	8.087	5.545	7.249	7.477	6.349	7.240
EA	0.080	0.014	0.039	0.061	0.023	0.041
Quercitrin	1.121	0.513	0.740	1.005	0.516	0.738

Table 3.

The results obtained by using different signal pre-treatment methods

Component		RAW	SGD	MSC	SNV	GF	MF	SGF	MSC+SGD	SNV+SGD	SNV+SGD+SGF	MSC+SGD+SGF	SNV+SGD+MSC	MSC+SGD+MF	SNV+SGD+GF	MSC+SGD+GF
GA	R2	0.8988	0.9942	0.9970	0.9870	0.9894	0.9903	0.9307	0.9996	0.9992	0.9912	0.9911	0.9951	0.9616	0.9968	0.9900
	RMSEC	0.0348	0.0312	0.0225	0.0468	0.0423	0.0405	0.1082	0.0075	0.0112	0.0386	0.0388	0.0287	0.0805	0.0233	0.0410
	RMSECV	0.1814	0.1048	0.0792	0.1137	0.2065	0.1413	0.2717	0.1102	0.0669	0.1397	0.1604	0.0625	0.1963	0.0810	0.1293
EA	R2	0.9889	0.9177	0.7204	0.7668	0.1323	0.6861	0.1172	0.9015	0.8829	0.2980	0.3088	0.9507	0.8914	0.2928	0.2892
	RMSEC	0.0040	0.0035	0.0065	0.0059	0.0115	0.0069	0.0116	0.0039	0.0042	0.0103	0.0102	0.0027	0.0041	0.0104	0.0104
	RMSECV	0.0024	0.0115	0.0113	0.0111	0.0118	0.0111	0.0119	0.0086	0.0118	0.0115	0.0118	0.0078	0.0101	0.0117	0.0118
Quercitrin	R2	0.9793	0.9994	0.9958	0.9797	0.9889	0.9980	0.9775	0.9983	0.9981	0.9957	0.9978	0.9952	0.9949	0.9848	0.9985
	RMSEC	0.0175	0.0030	0.0079	0.018	0.0129	0.0054	0.0183	0.0050	0.0053	0.0080	0.0056	0.0085	0.0087	0.0151	0.0047
	RMSECV	0.0406	0.0285	0.0248	0.0374	0.0358	0.0307	0.0574	0.0207	0.0309	0.0273	0.0139	0.0241	0.0256	0.0354	0.0166

Figure 8.

SGD convolution derivative (A), MSC + SGD convolution derivative (B) processed spectrum

Selection of a variable screening method

Four variable selection methods—mwPLS, CARS, RF, and UVE—were primarily investigated. The corresponding screening results are summarized in Table 4. The results indicated that the optimal wavelength selection method for the NIRS quantitative models of the three indicator components was CARS. The number of variables was reduced from 3,112 to 80, 45, and 93 variables, respectively, with this reduction leading to improved model parameters. During the CARS feature wavelength screening process, Monte Carlo sampling iterations were set to 50, and cross-validation folds to 5. As the sampling iterations increased, the CARS-PLS models established based on NIRS achieved the smallest RMSECV value at 20 sampling iterations, corresponding to the optimal selection of characteristic wavelength variables. The selected characteristic wavelengths in the NIR spectra were primarily distributed in the 6760–6970 nm, 4100–4800 nm, and 5750–5990 nm regions.

Table 4.

The results obtained by using different variable screening methods

Active components	Methods	Number of variables	R2	RMSEC	RMSECV
GA	mwPLS	125	0.9788	0.1492	0.2360
	CARS	80	0.9997	0.0075	0.1102
	UVE	285	0.9917	0.0375	0.4018
	random frog	70	0.9863	0.0480	0.2797
EA	mwPLS	220	0.9598	0.1125	0.2538
	CARS	45	0.9989	0.0004	0.0024
	UVE	291	0.9953	0.0008	0.0089
	random frog	49	0.9918	0.0011	0.0054
Quercitrin	mwPLS	210	0.9849	0.2387	0.3628
	CARS	93	0.9994	0.0038	0.0285
	UVE	322	0.9745	0.0195	0.0989
	random frog	67	0.9975	0.0061	0.0827

Establish and verify the NIR quantitative model

Through partitioning of the sample set for NIRS, determining preprocessing methods, and selecting variable screening approaches, quantitative models for three quality control components were ultimately established. The results demonstrated a significant increase in RPD, with RPD values for all three components exceeding 2.5, indicating excellent model performance. The R² values for the three components were all above 0.9, while RMSEC and RMSECV were below 0.1. These findings confirmed that the method effectively extracted relevant modeling variables and enhanced model performance. As shown in Figure 9, the NIR spectra of RLQ samples exhibited a linear relationship with content values measured by UPLC, demonstrating good model stability and predictive capability. Therefore, NIR spectroscopy enabled the rapid and accurate determination of three active ingredients in RLQ: GA, EA, and quercitrin.

Figure 9.

Quantitative calibration model of three components in RLQ

(A) GA. (B) EA. (C) Quercitrin.

To validate the accuracy of the established model, an external validation test was conducted by inputting the spectra of validation set samples (n = 10) into the quantitative model to obtain predicted values. The model’s accuracy was evaluated by calculating the relative deviations between predicted values and actual values. The results are shown in Figure 10. The average relative deviations between predicted and actual values for GA, EA, and quercitrin contents were 0.66%, 2.04%, and 1.65%, respectively. The predicted values of these three indicator components closely matched the actual values with consistent trends, demonstrating good accuracy of the established model. Therefore, NIRS combined with chemometrics can effectively predict the contents of three active components in RLQ samples.

Figure 10.

Quantitative analysis of three components through external validation

Discussion

This study systematically investigated the bioactive components of RLQ, covering the entire workflow from extraction, purification, and structural characterization to quantitative analysis using NIRS. By integrating NIRS with chromatographic techniques and combining TCM theory with modern chemometric modeling and analytical chemistry, a multidimensional collaborative analytical platform was constructed. The contents of three key active components in RLQ were determined using UPLC, while NIRS-PLS quantitative models were developed, achieving high predictive accuracy and robustness.

The applicability scope of the current model is established based on the specific sample and instrumental conditions of this study. Differences in instrument models and performance may pose challenges to the model prediction results, and the prediction accuracy is likely to be affected by factors such as the model of the NIRS or measurement parameters. In addition, the critical quality attributes (CQAs) of the samples to be tested (e.g., the content range of active ingredients) fall within the coverage of the calibration set of this model. Meanwhile, before extending the application of this model to other instruments or long-term monitoring, necessary external validation or model updating must be performed to ensure the reliability of its prediction results. Due to the limited sample size and the fact that all samples are derived from a single manufacturer, the model’s predictive capability for samples from different manufacturers, raw materials of diverse geographical origins, or various production batches remains to be verified, with the risk of extrapolation existing. The model has not yet been validated across different instruments, and its cross-platform robustness requires further investigation.

The value of this study lies in providing a proof-of-principle for a rapid analytical method for the quality control of this specific preparation. In the future, more samples will be further collected to expand and update the model, so as to cover a wider range of production variations. Subsequently, samples of RLQ produced by Guizhou Weimen Pharmaceutical Co., Ltd. will be systematically collected across different production batches, manufacturing years, and even raw material harvesting seasons. These newly collected samples will serve as an independent, time-series external validation set, which is employed to rigorously evaluate the long-term predictive robustness of the established model against natural production fluctuations, and further verify its practical applicability for the long-term online quality monitoring of this product. Furthermore, cross-instrument validation of the model will be performed using another NIRS of the same model or a different model. Efforts will be made to advance the model from laboratory-based static analysis to at-line or online real-time detection scenarios in production lines. The model is expected to be transformed into a practical analytical tool catering to specific production processes, thereby truly embodying its value for “broader practical applications.” We will further conduct systematic research on the impacts of inter-instrument differences and explore the use of spectral preprocessing, variable selection, or model updating algorithms to enhance the cross-platform and long-term robustness of the model. The ultimate goal is to establish a robust and universal NIRS quantitative model, laying a more solid foundation for its practical application in quality control across various links of the industrial chain.

GA, EA, and isoquercitrin are three natural polyphenolic compounds that demonstrate marked anti-inflammatory activity by modulating core inflammatory pathways such as NF-κB, though each exhibits distinct characteristics. As a fundamental phenolic acid, GA exerts direct and broad-spectrum effects by potently suppressing the NF-κB and MAPK signaling pathways, thereby rapidly downregulating multiple inflammatory mediators. EA is characterized by its dual-pathway approach: while inhibiting inflammatory signaling, it uniquely activates the Nrf2 antioxidant pathway to enhance cellular self-defense. Isoquercitrin, as a flavonoid glycoside, has the advantage of targeted immunomodulation. It is particularly effective in inhibiting the JAK-STAT and TLR4 pathways, demonstrating clear efficacy in anti-allergic responses and stabilizing immune cells, along with relatively high bioavailability. All three compounds demonstrate effective anti-inflammatory activity, with GA acting as a direct inhibitor, EA as a defense enhancer, and isoquercitrin as an immunomodulator. Together, they work in synergy to form a multi-target and multi-level anti-inflammatory network.

The in vitro activity observed in this study provides only preliminary and supportive pharmacodynamic evidence for considering these components as quality markers. Further validation through animal model experiments and in-depth investigations (such as chemical knockout or supplementation strategies) will be conducted to clarify their specific contributions and mechanisms of action in the overall therapeutic effect.

The interdisciplinary approach demonstrated here enables rapid identification and precise quantification of multiple bioactive markers, establishing a scalable analytical paradigm that can be applied beyond RLQ to other complex botanical products. The proposed methodology not only offers an integrated solution for the quality control of TCM preparations but also extends the application scope of NIRS to functional foods, nutraceuticals, and other plant-derived health products. By leveraging multimodal spectral data fusion and intelligent algorithm optimization, this work contributes to the international standardization and scientific advancement of quality evaluation systems for herbal medicines and related natural products.

Limitations of the study

Despite the contributions of this study to the systematic phytochemical research and quality control optimization of P. capitatum and its proprietary Chinese medicine RLQ, several limitations should be noted. In the phytochemical separation of P. capitatum, 20 structurally identified known compounds were obtained, but no novel compounds with unique skeletons were isolated, and only the 60% ethanol elution fraction was studied, leaving other fractions unexploited; meanwhile, the research on its pharmacodynamic material basis is insufficient, with gaps in structure-activity relationship and targeted structural modification. For in vitro anti-inflammatory activity evaluation, only 10 compounds were tested via a single NO detection method, without a full assessment of all isolates or in-depth mechanism exploration, and their antibacterial activity remains uninvestigated. Regarding the NIRS-UPLC quantitative model for RLQ, only three active components were quantified due to limited UPLC peaks and fixed 254 nm detection wavelength, which may miss potential active constituents, and future studies can combine more detection methods to establish comprehensive multi-component quantitative models for improved quality evaluation.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Prof. Dr. Yi Zhang (zhangyi@cdutcm.edu.cn).

Materials availability

All unique/stable materials generated in this study are available from the lead contact without restriction. Polygonum capitatum and Relinqing® Granule were obtained from Guizhou Weimen Pharmaceutical Co., Ltd., as described in the material and methods.

Data and code availability

• •Data: NMR data for compounds 1–20 are provided in the supplemental information.
• •Code: No custom code was generated in this study.
• •Other items: Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This work was funded by the National Key R&D Program of China (2023YFC3504400, 2023YFC3504402).

Author contributions

H.M.G.: literature search, visualization, and writing – draft manuscript; Y.Y.S. and X.Y.: resources and writing – draft manuscript; L.C.: literature search and data collection; S.F.S.: literature search; M.J.L., J.S.S., and Y.Z.: conceptualization, supervision writing – review and editing, and resources. All authors have read and approved the submitted manuscript.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
Polygonum capitatum	Guizhou Weimen Pharmaceutical Co., Ltd.	Voucher specimen: 20231230
Relinqing® granule (RLQ), 50 batches	Guizhou Weimen Pharmaceutical Co., Ltd.	Batch numbers listed in Methods
Chemicals, peptides, and recombinant proteins
Acetonitrile, HPLC grade	Fisher Chemicals	Analytical grade reagent
Phosphoric acid, HPLC grade	Chengdu Kelong Chemical Co., Ltd.	Analytical grade reagent
Gallic acid reference standard	Chengdu Push Bio-technology Co., Ltd.	PS012548; purity >98%
Ellagic acid reference standard	Chengdu Push Bio-technology Co., Ltd.	PS013592; purity >98%
Quercitrin reference standard	Chengdu Push Bio-technology Co., Ltd.	PS013787; purity >98%
Experimental models: Cell lines
RAW264.7 murine macrophage cells	Laboratory passage	Authenticated by STR profiling; mycoplasma negative
Critical commercial assays
CCK-8 assay	Chongqing Biospes Co., Ltd.	BG0025
Nitric oxide assay by Griess reaction	Shanghai Beyotime Biotechnology Co., Ltd.	S0021S
ELISA assay for TNF-α, IL-1β, and IL-6	Quanzhou Ruixin Biotechnology Co., Ltd.	RX202412M, RX203063M, RX203049M
Software and algorithms
MATLAB	MathWorks	Version R2023a

Experimental model and study participant details

No human participants or animal models were involved in this study. RAW264.7 murine macrophage cells were obtained from laboratory preservation. The cell line was authenticated by short tandem repeat profiling and tested negative for mycoplasma contamination. No human participants or animal models were involved in this study. The experimental system consisted of botanical materials, commercial Relinqing® granule (RLQ) samples, and an in vitro macrophage inflammatory model. Polygonum capitatum was provided by Guizhou Weimen Pharmaceutical Co., Ltd. in December 2023, produced in Guiyang, Guizhou Province, China, and authenticated by Prof. Yi Zhang. A voucher specimen was deposited in the State Key Laboratory of Chengdu University of Traditional Chinese Medicine (voucher no. 20231230). In addition, 50 batches of RLQ were supplied by the same manufacturer and used for quantitative analysis and model development. For the cellular experiments, RAW264.7 murine macrophages were used as the in vitro experimental model to evaluate the anti-inflammatory activities of compounds isolated from P. capitatum. Cells in the logarithmic growth phase were cultured under standard conditions at 37°C in 5% CO₂. An LPS-induced inflammatory model was established for functional evaluation, including cell viability, nitric oxide production, and cytokine secretion assays. No study participants were recruited, and no patient-derived materials, clinical specimens, or identifiable human data were included in this work.

Method details

Chemicals, reagents, instruments, plant materials, and RLQ samples

HPLC-grade acetonitrile and methanol were purchased from Fisher Chemicals. HPLC-grade phosphoric acid was obtained from Chengdu Kelong Chemical Co., Ltd., and HPLC-grade distilled water was obtained from Beijing Watsons Distilled Water Co., Ltd. Gallic acid (PS012548), ellagic acid (PS013592), and quercitrin (PS013787) reference standards with purity >98% were purchased from Chengdu Push Bio-technology Co., Ltd. Major analytical instruments included an Agilent 1200 Series semipreparative HPLC system, a Shimadzu UV-2401PC spectrophotometer, a Bruker AV 600 NMR spectrometer, a Shimadzu LC-MS-IT-TOF mass spectrometer, an Agilent 6200 series TOF instrument, a PerkinElmer Frontier NIR spectrometer, and a Waters ACQUITY UPLC® system. Silica gel, Lichroprep RP-18, Sephadex LH-20, and precoated silica gel GF254 plates were used for chromatographic separation and purification. Polygonum capitatum was provided by Guizhou Weimen Pharmaceutical Co., Ltd. in December 2023, authenticated by Prof. Yi Zhang, and deposited under voucher number 20231230. A total of 50 batches of RLQ from the same manufacturer were used for quantitative analysis and model development.

Extraction and isolation of compounds from Polygonum capitatum

Dried whole P. capitatum (10.0 kg) was cut into pieces and extracted four times with 80% ethanol under reflux for 1.5 h each, yielding 1.8 kg of crude ethanol extract. The extract was subjected to macroporous adsorption resin column chromatography and sequentially eluted with H₂O, 30% EtOH, 60% EtOH, and 90% EtOH. The 60% EtOH fraction was further separated by MCI column chromatography with an H₂O/MeOH gradient to afford 10 fractions (Fr.1–Fr.10). These fractions were purified by combinations of silica gel column chromatography, Sephadex LH-20 chromatography, and semipreparative HPLC using different ACN/H₂O systems, yielding 20 compounds in total.

Structural elucidation of isolated compounds

The isolated constituents were structurally characterized by UV, 1D NMR, ESI-MS, and HR-ESI-MS analyses. Chemical shifts were referenced to solvent signals, with tetramethylsilane used as the internal standard for NMR measurements. Final compound identification was achieved by comparison of spectroscopic data with literature reports.

Cell culture and in vitro anti-inflammatory activity assays

RAW264.7 murine macrophages were used as the in vitro inflammatory model. Cells were seeded and cultured at 37°C in 5% CO₂. Cells were treated with compounds at 10–160 μM for 24 h, then stimulated with LPS (1 μg/mL) for a further 24 h. Cell viability was measured using CCK-8. Nitric oxide production was measured using the Griess reaction. TNF-α, IL-1β, and IL-6 levels were measured using commercial ELISA kits.

UPLC-based quantification of quality control markers

Quantitative analysis was performed on a Waters ACQUITY UPLC® system with an ACQUITY UPLC BEH C18 column (2.1 mm × 100 mm, 1.7 μm). The mobile phase consisted of 0.4% phosphoric acid water and methanol with gradient elution. RLQ powder was extracted with 75% methanol under sonication, filtered, and subjected to UPLC analysis. Calibration curves were constructed using reference standards, and the method was validated for linearity, precision, repeatability, stability, and recovery.

NIR spectral acquisition, preprocessing, and model development

NIR spectra were collected using a PerkinElmer Frontier NIR spectrometer in diffuse reflectance mode (10000–4000 cm⁻¹, resolution 8 cm⁻¹). Various spectral preprocessing methods were tested, including SGD, MSC, SNV, SGF, GF, and MF. Samples were divided into calibration and validation sets using the Kennard–Stone method (8:2).

Variable selection and PLS regression analysis

Variable selection was performed using mwPLS, CARS, Random Frog, and UVE. Partial least squares (PLS) regression was used to establish quantitative models. Model performance was evaluated using R², RMSEC, RMSECV, and RPD. All calculations were performed in MATLAB R2023a.

Quantification and statistical analysis

All data are presented as mean ± standard deviation (SD). Statistical significance was determined by one-way ANOVA followed by Dunnett’s post-hoc test.

Asterisk definitions

p < 0.05, ##p < 0.01, ###p < 0.001 vs. blank control group; p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001 vs. LPS model group. All experimental data were expressed as mean ± standard deviation. For cell-based assays, each treatment group contained three replicate wells, and experiments were performed in triplicate. Cell viability was calculated from absorbance values at 450 nm using the CCK-8 assay. Nitric oxide production was quantified by the Griess method. TNF-α, IL-1β, and IL-6 levels were quantified using ELISA kits according to the manufacturer’s protocols.

For UPLC analysis, linear regression equations were constructed using the injection amount of each reference standard as the independent variable and peak area as the dependent variable. Method validation included linearity, precision, repeatability, stability, and recovery. RSD values were all below 3%, and recovery ranged from 98% to 101%.

For chemometric analysis, 50 RLQ batches were divided into calibration and validation sets by the Kennard–Stone algorithm (8:2). PLS regression models were evaluated using R², RMSEC, RMSECV, and RPD. External validation was performed using the validation set (n = 10). The average relative deviations between predicted and measured values were 0.66% for gallic acid, 2.04% for ellagic acid, and 1.65% for quercitrin. All chemometric calculations were carried out in MATLAB R2023a.

Published: June 26, 2026