Mechanisms of Xiangxue decoction in treating COVID-19 via HPLC‒Q-TOF‒MS/MS, network pharmacology, molecular docking and molecular dynamics simulation

Abstract

To predict the active components of Xiangxue Decoction (XXD), and explore its mechanism in treatment of COVID-19. The chemical compositions of compounds were identified using HPLC‒Q-TOF‒MS/MS technology. Active components were screened using UNIFI, and potential targets were retrieved from TCMSP, GeneCards. The STRING database was used to construct PPI networks. Enrichment analysis was conducted using Metascape. Molecular docking and Molecular dynamics simulations were further used to evaluate the interactions between the active components and potential targets. 162 drug targets, 1458 disease targets and 64 intersection targets were identified. Quercetin, kaempferol, luteolin, nobiletin and β-carotene were identified as the key components. The core targets included TNF, AKT1, IL-6, IL1B, HIF1A, STAT3 and EGFR. GO and KEGG analyses revealed the primary signaling pathways, including the JAK-STAT, MAPK, PI3K / Akt, and NF-kappa B. Molecular dynamics simulations confirmed β-carotene-IL-6, β-carotene-EGFR and β-carotene-TNF revealed preferable stability. Finally, HPLC‒Q-TOF‒MS/MS analysis detected 158 low-abundance and 73 high-abundance compounds in the XXD extract. The mechanism of XXD is complex and may be related to its signaling pathway. These results may support the use of XXD for the treatment of COVID-19.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-45496-z.

Introduction

COVID-19 is a respiratory tract infection caused by the SARS-CoV-2 virus, which has a considerable impact on human central nervous system, digestive system, respiratory system and other systems. Since its discovery in 2019, it has triggered a worldwide pandemic and brought great disaster to people around the world¹. To date, more than 770 million people worldwide have suffered from COVID-19, and more than 7 million people have died according to World Trade Organization (WTO) data reports. During the pandemic years, although researchers have been working on COVID-19 treatments and have achieved significant progress, several challenges remain. At present, the main treatment methods for COVID-19 include monoclonal antibodies, neuropathy drugs, targeted drug therapy, immunomodulatory therapy, antifibrotic and anticoagulant therapy, metabolic modulators, recovery from cognitive impairment and other methods^2,3. However, drug therapy has limited efficacy, drug resistance, side effects and other problems⁴. As a key strategy for containing the current pandemic, numerous relevant studies on COVID-19 are well underway and have yielded promising results. Current research falls into two categories: one focuses on specific viral components, while the other examines the whole virus. However, they also have problems such as antibody dependence and even some short-term or long-term harm to the human body^5,6. Therefore, there is an urgent need for highly effective drugs with minimal side effects to treat COVID-19.

Traditional Chinese medicine (TCM) has demonstrated significant advantages in preventing and controlling COVID-19. According to the guidance of TCM theory and the compatibility of various TCMs, it has a good effect on the symptoms of COVID-19 and other chain clinical symptoms^7,8. The Chinese medicinal compound Xiangxue Decoction (XXD) originates from the ancient Chinese text Taiping Huimin Heji Ju Fang and comprises four primary ingredients. Glycyrrhiza uralensis Fisch, Citrus reticulata Blanco, and Cyperus rotundus L., which dredges wind, relieves the surface, disperses cold and relieves pain. Perilla frutescens (L.) Britt contains phenolic compounds, terpenes and other key active components, which have strong antiviral potential^9,10. Flavonoids and tannins show antigenic viral activity, and they also have free radical scavenging activity^11,12. Flavonoids such as pectin isolated from Citrus reticulata Blanco also have inhibitory effects on SARS-CoV-2¹³. Active ingredients such as liquiritin and liquiritigenin isolated from Glycyrrhiza uralensis Fisch have also been proven to be effective in treating pneumonia and lung injury, and can effectively downregulate inducible nitric oxide synthase (iNOS) levels^14,15. However, the interaction mechanism and common effects of several drugs have not been studied in depth, so further studies are needed to clarify the mechanism and explore the therapeutic effect of these drugs on COVID-19.

With the continuous advancement of science and technology, network pharmacology has emerged as a key approach for studying compound components and analyzing drug-disease interactions. Network pharmacology has also developed from the original single “one drug-one target-one disease” to the current mode of “multiple targets forming a network and multicomponent therapy“¹⁶. TCM also has multicomponent, multitarget and comprehensive efficacy in treatment and diagnosis, so the application of network pharmacology to explore the composition mechanism of TCM compounds has become widely popular¹⁷.

In recent years, the rapid development of computational (in silico) technologies has provided powerful tools for deciphering the complex systems of multi-component and muti-target¹⁸. Network pharmacology, by constructing "component-target-disease" interaction networks, enables systematic prediction of TCM mechanisms of action and potential active ingredients. Building on this foundation, molecular docking and molecular dynamics simulations can visualize and evaluate interaction patterns and biding stability between active ingredients and key target proteins at the atomic level, providing structural biological evidence for network prediction¹⁹. Furthermore, machine learning and artificial intelligence models have been employed to enhance the accuracy and efficiency of target prediction. These multi-level and complementary computational approaches have been successfully applied in mechanistic exploration studies of various TCM compound formulations against COVID-19, demonstrating their effectiveness and practicality in this field²⁰.

In this study, we employed network pharmacology to identify relevant targets and signaling pathways of the TCM compound Xiangxue Decoction (XXD), establishing a foundation for subsequent research and novel drug development. We further conducted molecular docking analysis and molecular dynamics simulations to assess binding affinities between active compounds and core targets, alongside HPLC-Q-TOF-MS/MS analysis to characterize XXD’s phytochemical composition and bioactive constituents.

Analysis materials and methods

HPLC‒Q-TOF‒MS/MS

Instruments and materials

Among the four herbal components, Citrus reticulata Blanco, Perilla frutescens (L.) Britt and Glycyrrhiza uralensis Fisch were sourced from Oriental National Medicine Co., Ltd. (Jiaxing, Zhejiang, China). Citrus reticulata Blanco was obtained from the Affiliated Hospital of Zhejiang Chinese Medical University (Hangzhou, Zhejiang, China). All botanical materials were authenticated by Associate Professor Jing Chen of the College of Life Sciences, Zhejiang Chinese Medical University. All remaining reagents were analytical grade. Instrumentation included a high-resolution quadrupole time-of-flight liquid chromatography-mass spectrometry system (Waters, USA) and a circulating water multi-purpose vacuum pump (Zhengzhou Great Wall Science, Industry & Trade Co., Ltd., China).

Sample preparation

By consulting the records of the ancient text “Taiping Huimin Heji Ju Fang”, we determined that the ratio of the four medicinal materials in the XXD, namely, Cyperus rotundus L.: Perilla frutescens (L.) Britt: Glycyrrhiza uralensis Fisch: Citrus reticulata Blanco, was 4:4:1:2. The drink (110 g) was cut into small sections, 10 volumes of 70% ethanol (1100 mL) was added, and the mixture was heated and refluxed at 80 °C for 1 h. The supernatant was filtered and collected, and 8 volumes of 70% ethanol (880 mL) was added to the residue. The supernatants were combined, concentrated in a rotary evaporator, concentrated into two samples with high concentrations (100 mg/mL) and low concentrations (10 mg/mL), and stored in a refrigerator at 4 °C.

Chromatographic conditions

The column was a Waters SYNAPT G2-Si column (2.1 mm * 100 mm, 1.6 μm); the flow rate was 0.3 ml/min; the column temperature was 35 °C; the sample chamber temperature was 10 °C; the sample size was 2 µl; the mobile phase was 0.1% formic acid water and pure acetonitrile; and gradient elution was performed for 35 min (0–2 min, 5% acetonitrile; 2–32 min, 5–100% acetonitrile; 32–33 min, 100% acetonitrile; 33.5 min, 5% acetonitrile; 33.5–35 min, 5% acetonitrile).

Mass spectrometry conditions

During the data acquisition process, both positive and negative electrospray ionization (ESI) modes were employed. This dual-mode scanning strategy is essential because complex mixtures contain compounds with diverse physicochemical properties. Some components are more efficiently ionized through protonation in the positive ion mode ([M + H]⁺), while others compounds perform better through deprotonation in the negative ion mode ([M-H]⁻). Data acquisition in both modes ensures a more comprehensive and unbiased analysis of the chemical composition of XXD. Full-scan mode was used, the scan time was 0.2 s, and the scan range was 50-1200. The collision energy used was MSE, with a low collision energy of 6 V and a high collision energy of 15–45 V. Mass spectrometer correction was performed via sodium formate, and real-time quality correction was performed for leucine enkephalin (positive ion mode, 556.2771 m/z; negative ion mode, 554.2615 m/z). The capillary voltage was as follows: positive ion, 3 kV; negative ion, 2.5 kV; sample cone, 40 V; source offset, 80 V; source temperature, 120 °C; desolvation temperature, 500 °C; negative ion, 400 °C; desolvation gas, 1000 L/h; negative ion, 800 L/h; nebulizer, 6.5 bar²¹.

Chemical composition analysis

Two samples with different concentrations were analyzed according to the chromatography and mass spectrometry conditions. With information on the peaks of the compounds, the main chemical components of the samples were identified. Through screening, the components with high responses at different concentrations and good matching degrees in the samples were selected²².

Identification and verification of compounds

The UNIFI software was first employed to perform automatic matching based on precise mass numbers and the TCM database, which the mass error window for primary mass spectrometry was set to 5 ppm, generating preliminary results. Subsequently, all preliminary matching results undergo validation according to the following criteria: (1) Precise mass: In the primary mass spectrum (MS¹), the absolute error between the measured mass number of the precursor ion and the theoretical mass number should be < 5ppm. (2) Fragment matching quantity: The experimental MS/MS spectrum must match at least 3 or more fragment ions against the reference spectra in the database. (3) Chromatographic rationality: The retention time of the compound must align with the expected polarity trend of its chemical category. Only compounds meeting all the above validation criteria will pass the screening.

Network pharmacology analysis

Screening of active ingredients and potential targets of XXD compounds

The compounds identified by HPLC–Q-TOF–MS/MS analysis were complementarily integrated with the retrieved from the Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP, https://old.tcmsp-e.com/tcmsp.php) to form an initial drug active ingredient library for subsequent screening²³. We acknowledge that while TCMSP serves as a foundational and widely utilized resource for network pharmacology studies in TCM, it may exhibit delays in updates and limitations in database coverage. To ensure the robustness of our target database within this framework, we conducted ADME pre-screening, retaining only compounds that met the common drug similarity criteria for target collection. Blood-brain barrier (BBB) permeability was used to evaluate the potential neuroinflammation-related properties of active compounds in COVID-19. Caco-2 permeability serves as an indicator of intestinal absorption potential for oral drugs. Therefore, we conducted BBB permeability and Caco-2 permeability assessments for the active compounds. Simultaneously, for each qualifying compound, we gathered all predicted human targets listed in TCMSP and merged duplicate target. For compounds for which no corresponding target information is available in TCMSP, they will be explicitly listed and excluded from subsequent network construction, ensuring that the network is entirely based on the recorded compound-target relationships. Drug Likeness (DL) can be defined as the sum of the molecular physicochemical properties unique to the chemical substances of a drug and is typically used to describe the pharmacokinetics and safety of the drug²⁴. Oral Bioavailability (OB) refers to the amount of nutrients or bioactive substances that reach organs and tissues from ingested drugs to exert their biological activity. High oral bioavailability reduces the amount of drug required to achieve the desired pharmacological effect, thereby lowering the risk of side effects and toxicity²⁵. Therefore, we used an OB ≥ 30% and a DL ≥ 0.18 to screen for drug ingredients, which identified the active ingredients. Compounds screened above will be prioritized if they meet the following criteria: (i) Protein targets predicted or experimentally validated in databases; (ii) Pharmacological activities, such as antiviral, anti-inflammatory, or immunomodulatory, documented in literature reviews and associated with COVID-19 pathology. Finally, to eliminate discrepancies caused by varying nomenclature, we used UniProt database (https://www.uniprot.org)²⁶ to standardize all target names in the drug and disease datasets in to official human gene symbols.

Identification of targets in COVID-19 treatment

Our study also provided a comprehensive collection of potential therapeutic targets for COVID-19 via the GeneCards database (https://www.genecards.org)²⁷. The resulting potential target screens were then removed in duplicate for the next construction of the protein‒protein interaction (PPI) network. Duplicate entries in each target set had been removed. For genes listed with multiple protein isoforms in the source database, only the primary canonical gene symbol was retained to avoid overrepresentation of a single gene entity. This step ensured that each target in our list represented a unique gene.

Construction of the interrelationship between XXD targets and COVID-19 targets

The selected drug-active ingredients and potential targets were imported into Cytoscape 3.10.1²⁸ for network visualization, and the “drug-active ingredient-target” network was constructed. By intersecting the predicted targets of active compounds in XXD with known COVID-19-related targets, overlapping targets were obtained and visualized using Venny (https://bioinfogp.cnb.csic.es) to get a Venn diagram.

Construction of protein interaction (PPI) network diagrams

The cross-target sites between XXD and COVID-19 obtained from Venny (https://bioinfogp.cnb.csic.es) were imported into the STRING database (https://cn.string-db.org/)²⁹. The STRING database (https://www.genecards.org) was used to analyze the network data obtained from protein interactions. We obtained specific organism information by setting “Homo sapiens” as a filtering condition, and the minimum interaction threshold was set to 0.400 (moderate confidence) to obtain a comprehensive set of potential interactions for subsequent analysis.

Cytoscape is often used to integrate the network of biomolecular interactions with high-throughput expression data and other molecular states into a unified conceptual framework so that we can analyze the relationships between active ingredients and core targets. It is widely used in the field of network pharmacology³⁰.

We introduced the active components of XXD and the corresponding targets of COVID-19 into Cytoscape 3.10.1 to construct an interrelationship. The data obtained from the PPI were constructed via Cytoscape 3.10.1, and the network parameters, including degree centrality (DC), closeness centrality (CC) and betweenness centrality (BC), were analyzed. To eliminate selection bias, we defined core targets as the top 20% of nodes ranked by degree centrality. This widely accepted criterion ensures the selection of the most closely connected targets while maintaining objectivity and comparability across studies. Nodes in the PPI network were ranked by DC values, and core targets with higher scores were selected for further study.

Kyoto encyclopedia of genes and genomes (KEGG) pathway and gene ontology (GO) enrichment analysis

To explore the biological function of PPIs in COVID-19, we performed an enrichment analysis of GO and KEGG data via Metascape (https://Metascape.org)³¹. The p value of the GO function was filtered as less than 0.01. KEGG enrichment analysis facilitated the identification and determination of important signaling pathways involved in biologically relevant processes. We subsequently used a bioinformatics platform (https://www.bioinformatics.com.cn) to visualize the results of the GO and KEGG analyses³².

Molecular docking

Molecular docking is a computational modeling method that uses computers to identify therapeutically significant compounds, predict molecular-level interactions between ligands and targets, and describe the relationship between molecular structure and efficacy. It is widely used in new drug development. We used molecular docking to verify that the core network pharmacology-derived proteins were associated with the active components of XXD. Molecular docking of seven key target proteins and the top five compounds was performed via AutoDock Vina 4 and AutoDockTools 1.5.6³³. Briefly, the SDF structure of the ligand was derived from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/)³⁴ and subsequently converted to PDB format. Later, we used the RCSB protein database (https://www.pdb.org/)³⁵ to provide the crystal structure of the target protein and then imported it into PyMOL 2.5³⁶ to remove water molecules and existing ligands before docking. The form of the docked grid box was then designed to contain proteins via AutoDockTools 1.5.6. Furthermore, molecular docking of the target proteins and compounds was performed via AutoDock Vina 4, and the free binding energy was calculated. Finally, PyMOL 2.5 was used to visualize and analyze the interactions and binding forms between proteins and compounds.

Molecular dynamic stimulation

The stability of the predicted β-carotene-target complexes (TNF, IL-6, TGFR) was evaluated using full-atom molecular dynamics (MD) simulations performed with GROMACS 2022.2³⁷. Each system was constructed based on the optimal docking conformations obtained from previous molecular docking analyses.

Force field and parametric protein structures

The AMBER ff14SB force field was employed, with the gmx pdb2gmx tool applied³⁸. For the ligand β-carotene, a complete parametric scheme was adopted: the initial geometry was optimized using Gaussian 16 at the HF/ 6-31G* level³⁹, followed by electrostatic potential (ESP) calculations. The Restricted Electrostatic Potential (RESP) charges were then generated using Multiwfn. The β-carotene parameters were assigned using the general AMBER force field (GAFF) parameters, and the final topological file compatible with GROMACS 2022.2 was generated using sobtop⁴⁰, ensuring complete consistency within the AMBER force field framework.

System setup and equilibration

Each docking complex was solvated in a cubic TIP3P water molecule box, with a minimum protein-to-box edge distance of 1.0 nm, yielding an approximate size of 8.0 nm × 8.0 nm × 8.0 nm. Sodium and chloride ions were added to neutralize the system, simulating a 0.15 M physiological salt concentration. Periodic boundary conditions were applied in all directions. The system first performs energy minimization using the steepest descent algorithm (50000 steps) to eliminate stereo collisions. Subsequently, a two-stage equilibrium process is conducted: NVT equilibrium, achieved using a V-scale rebalanced thermostat (coupling constant = 0.1 ps) at 298 K for 100 ps. NPT balance, used a C- weighted barometer, 100 ps at 1 bar (coupling constant = 0.5 ps). Throughout the simulation, the bond lengths involving hydrogen atoms were constrained by the LINCS algorithm, allowing a time step of 2 fs. Van der Waals interactions were truncated at 1.2 nm, and long-range electrostatic interactions were handled using the particle-mesh Ewald (PME) method.

Production simulation and convergence evaluation

A 100 ns production MD simulation was conducted under NPT conditions (298 K, 1 bar). To ensure the reliability of the simulation data and verify that the system had reached equilibrium, the following metrics were used for convergence assessment. Root Mean Square Deviation (RMSD) of the protein backbone and side chain atoms relative to the initial structure. Root Mean Square Fluctuation (RMSF) of individual protein residues. The time evolution of system potential energy and temperature.

The RMSD and energy curves stabilized on a plateau after approximately 20 ns, indicating that the system had reached equilibrium. Consequently, subsequent trajectories (20–100 ns) were used for all analyses, including Solvent-Accessible Surface Area (SASA), rotational radius (Rg), and residue-specific SASA calculations³⁹.

Results

Identification of the ingredients of XXD

We identified the low-concentration and high-concentration samples, and identified 73 compounds in the high-concentration samples and 158 compounds in the low-concentration samples, among which the main components were flavonoids, saponins, and coumarins, including most chromatographic peaks of UPLC–MS (see Fig. 3). High response values were screened by Unifi software, yielding 33 components at high concentrations and 17 components with high response values in low-concentration samples, as shown in Tables 1 and 2.

Fig. 3.

HPLC‒Q-TOF‒MS/MS analysis of XXD. (a) and (b) are obtained from high-concentration solutions in positive and negative ion modes, respectively. A total of 73 compounds were detected, 33 of which were well matched. (c) and (d) are obtained from low-concentration solutions in positive and negative ion modes, respectively. A total of 158 compounds were detected, among which 17 were well matched.

Table 1.

HPLC‒Q-TOF‒MS/MS results for high-concentration solutions (100 mg/mL).

Component name	Formula	Neutral mass (Da)	Mass error (ppm)	Observed RT (min)	Adducts	Category
1,3,6-Trihydroxy-2-methylanthracene-9,10-dione	C15H10O5	270.05282	-0.7	8.88	[M + H]+	Quinones
Gancaonin E	C25H28O6	424.18859	-2.3	20.75	[M-H]−	Flavonoids
Gancaonin G	C21H20O5	352.13107	-1.6	14.85	[M + H]+	Flavonoids
Gancaonin I	C21H22O5	354.14672	-3.9	20.50	[M-H]−	Flavonoids
Glepidotin A	C20H18O5	338.11542	-2.6	17.43	[M-H]−	Flavonoids
Glyasperin A	C25H26O6	422.17294	1.9	20.99	[M-H]−	Flavonoids
Glyasperin B	C21H22O6	370.14164	-1.0	14.50	[M + H]+	Flavonoids
Glyasperin C	C21H24O5	356.16237	-1.8	16.13	[M-H]−	Flavonoids
Glyasperin D	C22H26O5	370.17802	-0.8	19.56	[M-H]−, [M+HCOO]−	Flavonoids
Glyasperin K	C22H24O5	368.16237	-2.2	20.05	[M-H]−	Flavonoids
Glycyroside	C27H30O13	562.16864	0.9	8.69	[M+HCOO]−	Flavonoids
Glyeurysaponin	C42H62O16	822.40379	1.5	14.13	[M-H]−	Saponins
Glyinflanin A	C25H28O5	408.19367	3.7	19.79	[M-H]−	Flavonoids
Hesperidin	C28H34O15	610.18977	0.4	8.88	[M-H]−, [M+HCOO]−	Flavonoids
Isogosferol	C16H14O5	286.08412	-2.1	10.64	[M + H]+	Coumarin
Isoliquiritin apioside	C26H30O13	550.16864	-0.6	9.46	[M-H]−, [M+HCOO]−	Flavonoids
Isoononin	C22H22O9	430.12638	-0.2	9.69	[M + H]+	Flavonoids
Isoschaftoside	C26H28O14	564.14791	2.0	6.97	[M-H]−	Flavonoids
Jaranol	C17H14O6	314.07904	-2.2	14.35	[M + H]+	Flavonoids
Kaempferol	C15H10O6	286.04774	-1.4	7.95	[M + H]+	Flavonoids
Licoagroisoflavone	C20H16O5	336.09977	-1.5	19.21	[M-H]−	Flavonoids
Licocoumarone	C20H20O5	340.13107	-3.0	16.77	[M-H]−	Coumarin
Licoisoflavone B	C20H16O6	352.09469	-3.1	18.53	[M-H]−	Flavonoids
Licoricesaponin A3	C48H72O21	984.45661	3.0	11.93	[M-H]−	Saponins
Licoricesaponin B2	C42H64O15	808.42452	2.9	16.10	[M-H]−	Saponins
Licoricesaponine D3	C50H76O21	1012.48791	2.9	13.58	[M-H]−	Saponins
Limonin	C26H30O8	470.19407	0.2	14.38	[M-H]−, [M+HCOO]−	Limonin
Morusin	C25H24O6	420.15729	-0.2	23.65	[M-H]−	Flavonoids
Narcissoside	C28H32O16	624.16903	-0.1	6.59	[M-H]−, [M+HCOO]−	Flavonoids
Naringenin	C15H12O5	272.06847	-2.8	8.37	[M + H]+	Flavonoids
Nicotiflorin	C27H30O15	594.15847	0.4	6.35	[M-H]−	Flavonoids
Phaseol	C20H16O5	336.09977	-1.0	19.20	[M + H]+	Flavonoids
Rosmarinic acid	C18H16O8	360.08452	-2.8	9.10	[M-H]−	Phenolic acid

Table 2.

HPLC‒Q-TOF‒MS/MS results for low-concentration solutions (10 mg/mL).

Component name	Formula	Neutral mass (Da)	Mass error (ppm)	Observed RT (min)	Adducts	Category
1,3,6-Trihydroxy-2-methylanthracene-9,10-dione	C15H10O5	270.05282	-1.1	8.87	[M + H]+	Quinones
3,4,3’,4’-Tetrahydroxy-2-methoxychalcone	C16H14O6	302.07904	-2.0	8.86	[M + H]+	Flavonoids
8-Prenylated Eriodictyol	C20H20O6	356.12599	-0.6	15.90	[M-H]−	Flavonoids
Gancaonin E	C25H28O6	424.18859	-3.6	20.76	[M-H]−	Flavonoids
Gancaonin G	C21H20O5	352.13107	-0.7	14.83	[M + H]+	Flavonoids
Glyasperin A	C25H26O6	422.17294	-2.4	20.99	[M-H]−	Flavonoids
Glyasperin B	C21H22O6	370.14164	0.1	14.49	[M + H]+, [M + Na]+	Flavonoids
Glyasperin C	C21H24O5	356.16237	-3.3	16.13	[M-H]−	Flavonoids
Glyasperin D	C22H26O5	370.17802	-3.4	19.56	[M-H]−	Flavonoids
Glyasperin K	C22H24O5	368.16237	-2.0	20.05	[M-H]−	Flavonoids
Glyinflanin A	C25H28O5	408.19367	-1.7	20.45	[M-H]−	Flavonoids
Hesperidin	C28H34O15	610.18977	-0.1	8.87	[M-H]−, [M+HCOO]−	Flavonoids
Isogosferol	C16H14O5	286.08412	-4.2	10.65	[M-H]−	Coumarin
Isoliquiritin apioside	C26H30O13	550.16864	1.6	9.45	[M-H]−, [M+HCOO]−	Flavonoids
Isoononin	C22H22O9	430.12638	1.2	9.68	[M + H]+	Flavonoids
Jaranol	C17H14O6	314.07904	-1.6	14.33	[M + H]+	Flavonoids
Kaempferol	C15H10O6	286.04774	-1.3	7.95	[M + H]+	Flavonoids

Active compounds and potential targets of XXD

We identified 92 highly active ingredients from Glycyrrhiza uralensis Fisch, 5 from Citrus reticulata Blanco, 18 from Cyperus rotundus L., and 14 from Cyperus rotundus L. frutescens (L.) Britt. Table 3 shows the target details of the part composition.

Table 3.

Part of the active ingredients in XXD.

MOL ID	Compound	Molecular structure	OB	DL
Glycyrrhiza uralensis Fisch (Listed only the top ten active ingredients)
5,320,083	Glycyrol		90.78	0.67
122,851	Licopyranocoumarin		80.36	0.65
10,336,244	Shinpterocarpin		80.30	0.73
44,257,530	Phaseol		78.77	0.58
5,318,999	Licochalcone B		76.76	0.19
392,442	Glyasperin F		75.84	0.54
91,510	Inermine		75.18	0.54
177,149	Vestitol		74.66	0.21
101,664,572	Glyasperins M		72.67	0.59
928,837	(2R)-7-hydroxy-2-(4-hydroxyphenyl)chroman-4-one		71.12	0.18
Citrus reticulata Blanco
12,303,287	Citromitin		86.90	0.51
72,344	Nobiletin		61.67	0.52
439,246	Naringenin		59.29	0.21
3593	5,7-dihydroxy-2-(3-hydroxy-4-methoxyphenyl) chroman-4-one		47.74	0.27
222,284	Sitosterol		36.91	0.75
Cyperus rotundus L. (Listed only the top ten active ingredients)
11,723,309	Rosenonolactone		79.84	0.37
70,697,662	Khellol glucoside		74.96	0.72
72,301	Hyndarin		73.94	0.64
5,281,654	Isorhamnetin		49.60	0.30
5,280,343	Quercetin		46.43	0.28
78,384,888	1,4-Epoxy-16-hydroxyheneicos-1,3,12,14,18-pentaene		45.10	0.24
52,929,817	Sugeonyl acetate		45.08	0.20
N/A	Stigmasterol glucoside_qt		43.83	0.76
5,280,794	Stigmasterol		43.83	0.76
5,280,863	Kaempferol		41.88	0.24
Perilla frutescens (L.) Britt (Listed only the top ten active ingredients)
9064	(+)-Catechin		54.83	0.24
441,688	Cyanin		47.42	0.76
5,367,460	Linolenic acid ethyl ester		46.10	0.20
5,280,581	LAX		44.11	0.20
7,057,921	ZINC03860434		43.59	0.35
N/A	Eugenyl-β-D-glucopyranoside(cirtrusinc)		40.52	0.23
17,164	Methyl icosa-11,14-dienoate		39.67	0.23
5997	CLR		37.87	0.68
5,280,489	Beta-carotene		37.18	0.58
12,303,645	Beta-sitosterol		36.91	0.75

Abbreviations: XXD, Xiangxue Drink; OB, oral bioavailability; DL, drug likeness.

For the “compound-target-disease” network analysis, we selected 114 compounds and 227 crossover targets from the compounds with a large number of intersections. This study revealed that the quercetin and kaempferol of Glycyrrhiza uralensis Fisch and Cyperus rotundus L. and the luteolin, β-sitosterol, and naringenin of Citrus reticulata Blanco and Glycyrrhiza uralensis Fisch are able to regulate multiple targets. The characteristics of TCM compounds that can modulate the same target component indicate the multicomponent and multitarget functions of XXD, as shown in Fig. 1.

Fig. 1.

(a) Map of drug active ingredient targets. The yellow circle represents the compositions of Glycyrrhiza uralensis Fisch, the green diamond represents the compositions of Citrus reticulata Blanco, the gray V represents the compositions of Cyperus rotundus L., the purple square represents the compositions of Perilla frutescens (L.) Britt, and the orange circle represents the common compositions of four single Chinese medicines (Glycyrrhiza uralensis Fisch, Citrus reticulata Blanco, Cyperus rotundus L. and Perilla frutescens (L.) Britt). The blue square represents the target sites of the XXD. The component with the highest correlation with the target COVID-19 was Citrus reticulata Blanco, and the target with the highest correlation was PTGS2. (b) Venn diagram of XXD targets and COVID-19 targets. There are 162 unique targets for XXD, 1485 unique targets for COVID-19, and 64 intersection targets. (c) The PPI network involving XXD and COVID-19 targets. The circles represent the proteins targeted by COVID-19 and XXD, and the target protein with the highest correlation is TNF-α.

Construction and analysis of the protein‒protein interaction (PPI) network map

In total, 1458 disease targets of COVID-19 were identified by searching databases from GeneCards. Subsequently, we used the Venn tool to map the common targets and disease targets in XXD (Fig. 1) and obtained 64 intersection targets. These findings indicate that XXD is able to treat COVID-19 with multitarget intervention. We then constructed the PPI network by visualizing the intersection targets, as shown in Fig. 1. Whether the selected targets are core targets was analyzed by determining the DC values of the nodes in Fig. 1. We identified some of the core targets on the basis of the DC, BC and CC values (Table 4). To elucidate the biological relevance of these 63 overlapping targets, in addition to topological analysis, we classified them into four functional modules closely associated with the pathogenesis of COVID-19. Based on network topology (Table 4), core regulatory hubs were identified, including TNF, AKT1, IL6, TP53, and STAT3. Subsequently, the targets were broadly categorized as follows: (i) Inflammation or immune dysregulation (e.g., TNF, IL6, STAT3, MMPs, SERPINE1), directly involved in cytokine storms and immune thrombosis; (ii) Viral invasion and host interaction (e.g., DPP4, MAPK1), related to viral entry into cells and host response signaling; (iii) Cell death/survival and tissue repair (e.g., HMOX1, NFE2L2, CAT), participating in mitigating virus-induced oxidative damage. This structured classification demonstrates that the targets of XXD are not random but concentrated on key pathological processes, providing a foundation for further research.

Table 4.

Important targets and related indicators of the PPI network.

Name	Degree centrality	Closeness centrality	Betweenness centrality
TNF	58	0.95	0.05
AKT1	55	0.91	0.02
IL6	55	0.91	0.02
IL1B	54	0.90	0.02
TP53	53	0.88	0.02
HIF1A	51	0.86	0.01
TGFB1	51	0.86	0.02
STAT3	50	0.85	0.01
EGFR	50	0.85	0.02
PTGS2	50	0.85	0.02

Abbreviations: PPI, protein‒protein interaction.

GO analysis and KEGG enrichment

As shown in Fig. 2, we selected 10 signaling pathways via KEGG enrichment analysis to explore the signaling mechanism of XXD in the treatment of COVID-19. We identified pathways associated with cancer, lipids and atherosclerosis, as well as fluid shear and atherosclerosis. Among these, the most significantly enriched pathways were the “Pathways in cancer” (P = 1.32E-34, 34 genes), “Lipid and atherosclerosis” (P = 3.04E-28, 20 genes), and “Fluid shear stress and atherosclerosis” (P = 3.24E-32, 20 genes). The top 10 pathways are detailed in Table 5. The results of the GO analysis indicate that XXD plays an important role in the treatment of COVID-19, which is related mainly to other biological processes, such as cell migration, cytokine production, and vascular development.

Fig. 2.

(a) Top 10 KEGG pathways associated with the hub genes. They were adapted from the following pathways in the KEGG database: hsa05200, hsa05417, hsa05418, hsa04668, hsa04657, hsa05161, hsa05142, hsa05215, hsa04151, and hsa05206. The pathway with the lowest p value was the pathway associated with cancer^44–46. (b) GO terms of the hub genes. The GO biological process with the lowest p value was the positive regulation of cell migration. The GO biological process with the highest p value was the positive regulation of angiogenesis. The GO cellular component with the highest p value was the endocytic vesicle, endosome membrane and endocytic vesicle membrane. The GO cellular component with the lowest p value was the membrane raft and membrane microdomain. The GO molecular function with the highest p value was growth factor activity. The GO molecular function with the lowest p value was cytokine receptor binding. GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Table 5.

Top 10 enriched KEGG pathways for XXD against COVID-19.

Term	Pathway Name	Count	Gene	P-Value
hsa05200	Pathways in cancer	34	AKT1, AR, BCL2, CASP8, CHUK, EGFR, ERBB2, GSTP1, HIF1A, HMOX1, IFNG, IKBKB, IL2, IL4, IL6, CXCL8, MET, MMP1, MMP2, MMP9, NFE2L2, NFKBIA, MAPK1, PTGS2, RELA, STAT1, STAT3, TGFB1, TP53, CREB1, MMP3, KDR, ABCC1, SIRT1	1.32E-34
hsa05417	Lipid and atherosclerosis	20	AKT1, BCL2, CASP8, CHUK, ICAM1, IKBKB, IL1B, IL6, CXCL8, MMP1, MMP3, MMP9, NFE2L2, NFKBIA, MAPK1, RELA, CCL2, STAT3, TNF, TP53	3.04E-28
hsa05418	Fluid shear stress and atherosclerosis	20	AKT1, BCL2, CAV1, CHUK, GSTP1, HMOX1, ICAM1, IFNG, IKBKB, IL1A, IL1B, KDR, MMP2, MMP9, NFE2L2, RELA, CCL2, THBD, TNF, TP53	3.24E-32
hsa04668	TNF signaling pathway	18	AKT1, CASP8, CHUK, CREB1, ICAM1, IKBKB, IL1B, IL6, CXCL10, IRF1, MMP3, MMP9, NFKBIA, MAPK1, PTGS2, RELA, CCL2, TNF	9.09E-30
hsa04657	IL-17 signaling pathway	44	CASP8, CHUK, IFNG, IKBKB, IL1B, IL4, IL6, CXCL8, CXCL10, MMP1, MMP3, MMP9, NFKBIA, MAPK1, PTGS2, RELA, CCL2, TNF, AKT1, CREB1, ICAM1, IRF1, IL2, IL10, SERPINE1, TGFB1, BCL2, NFE2L2, STAT3, TP53, STAT1, IL1A, AHR, HIF1A, EGFR, CTSD, ACACA, FASN, ADIPOQ, SIRT1, ERBB2, HMOX1, MET, ABCC1	2.10E-31
hsa05161	Hepatitis B	17	AKT1, BCL2, CASP8, CHUK, CREB1, IKBKB, IL6, CXCL8, MMP9, NFKBIA, MAPK1, RELA, STAT1, STAT3, TGFB1, TNF, TP53	6.11E-25
hsa05142	Chagas disease	17	AKT1, CASP8, CHUK, IFNG, IKBKB, IL1B, IL2, IL6, CXCL8, IL10, NFKBIA, SERPINE1, MAPK1, RELA, CCL2, TGFB1, TNF	1.48E-21
hsa05215	Prostate cancer	15	AKT1, AR, BCL2, CHUK, CREB1, EGFR, ERBB2, GSTP1, IKBKB, MMP3, MMP9, NFKBIA, MAPK1, RELA, TP53	1.01E-24
hsa04151	PI3K-Akt signaling pathway	15	AKT1, BCL2, CHUK, CREB1, EGFR, ERBB2, IKBKB, IL2, IL4, IL6, KDR, MET, MAPK1, RELA, TP53	5.37E-16
hsa05206	MicroRNAs in cancer	13	BCL2, EGFR, ERBB2, HMOX1, IKBKB, MET, MMP9, ABCC1, MAPK1, PTGS2, STAT3, TP53, SIRT1	5.58E-14

Molecular docking

On the basis of the results of network pharmacology, we selected the top five compounds, kaempferol, luteolin, nobiletin, β-carotene and quercetin, and coupled them to seven proteins to a large degree: IL6 (1IL6⁴¹, AKT1 (1UNP⁴², IL1B (5FB8), HIF1A (4H6J), STAT3 (6NJS), EGFR (3POZ) and TNF (7JRA⁴³. We determined the center coordinates of the docking grid box based on the geometric center of known active sites or co-crystallized ligand, with grid dimensions set to accommodate the ligand and allow moderate flexibility exploration. The specific protein structures and docking parameters used were detailed in Table 6. Additionally, the docking score between the protein and the compound was calculated. After visual analysis using PyMOL, the docking conformation with the lowest binding energy was selected as the optimal docking image for the receptor and ligand (Fig. 4). And residue within 4.0 Å of the ligand was defined as an interaction residue. A comprehensive summary of docking energy, hydrogen bonds, and interaction residues is presented in Table 7, with the complete list of residues available in Supplementary Table S12.

Table 6.

Protein structures and docking grid box parameters used in molecular docking.

Protein	PDB ID	Grid Box Center (x, y, z) (Å)	Grid Box Size (x, y, z) (Å)
IL6	1IL6	(-0.328, 0.341, 0.7518)	(55.875, 55.875, 55.875)
AKT1	1UNP	(12.974, -0.811, -2.825)	(58.000, 58.000, 58.000)
IL1B	5FB8	(8.084, -21.972, -33.116)	(80.875, 80.875, 80.875)
HIF1A	4H6J	(13.132, -11.215, -19.574)	(52.375, 52.375, 52.375)
STAT3	6NJS	(-2.065, 19.379, 24.188)	(117.750, 117.750, 117.750)
EGFR	3POZ	(19.504, 24.848, 15.430)	(67.750, 67.750, 67.750)
TNF	7JRA	(-3.449, -1.712, -28.023)	(63.837, 63.837, 63.837)

Fig. 4.

(a) Heatmap of binding energies. (b-i) Schematic diagram of docking between active compounds of XXD and IL6, AKT1, TNF, IL1B, HIF1A, STAT3 and EGFR. (b) Quercetin-TNF. (c) Nobiletin-TNF. (d) β-carotene-TNF. (e) β-carotene-EGFR. (f) β-carotene-IL6. (g) β-carotene-AKT1. (h) β-carotene-IL1B. (i) β-carotene-HIF1A. The green structure is the chemical composition of small molecules, the navy blue three-dimensional structure is the protein, and the wathet blue structure is the amino acid residues linked by hydrogen bonds. ((b) and (c) have hydrogen bond structures, whereas the rest do not. The binding of the group without conventional hydrogen bond may be mainly stabilized by hydrophobic interaction and van der Waals force.). In the protein molecules, and the names of the amino acid residues are shown. The yellow structure is the hydrogen bond corresponding to the protein in the small molecule, with the number above the length of the hydrogen bond. The three pairs with the lowest binding energy are β-carotene-IL6, β-carotene-TNF and β-carotene-EGFR.

Table 7.

Molecular docking results of XXD core components with COVID-19 key targets.

Protein	Compound	Docking Energy (kcal/mol)	Number of hydrogen bonds	Other Predominant Interactions	Core Interacting Residues (within 4.0 Å)*
TNF	quercetin	-7.54	8	/	SER-175, GLU-192, TYR-191, LYS-174, TYR-195
TNF	nobiletin	-8.14	4	/	TYR-191, LYS-174, GLN-225
TNF	β-carotene	-10.63	None	hydrophobic interactions	LEU-133, TYR-135, ARG-158, ASN-168, LEU-169, etc.
EGFR	β-carotene	-9.28	None	hydrophobic interactions	LEU-718, GLY-719, VAL-726, ALA-743, LYS-745, etc.
IL6	β-carotene	-9.03	None	hydrophobic interactions; van der Waals force	LEU-93, VAL-97, GLU-100, LYS-121, ILE-124, etc.
AKT1	β-carotene	-8.46	None	hydrophobic interactions; Static electricity	LYS-14, ARG-15, GLY-16, GLU-17, TYR-18, etc.
IL1B	β-carotene	-7.75	None	hydrophobic interactions	LEU-40, GLN-67, LYS-68, PRO-69, GLY-70, etc.
HIF1A	β-carotene	-7.31	None	hydrophobic interactions	ASN-326, CYS-358, PRO-360, THR-361, ALA-382, etc.

*The first five residues listed are representative. A complete list of all interaction residues (within 4.0 Å) for each complex is provided in Supplementary Table S12.

The calculation of binding energy is a common method for assessing the affinity between ligands and target proteins and can also be used as an indicator of the likelihood of receptor‒ligand interactions. It is generally accepted that when the binding energy between small molecules and proteins is negative (< 0), spontaneous binding can occur between them. The lower the binding energy between a small molecule and a protein is, the greater the stability of the resulting conformation, and the greater the affinity between the ligand and the receptor. To visualize the binding affinity between ligands and receptors, we generated a heatmap. The binding energies (kcal/mol) of the selected compounds and the target are shown in Fig. 4. The docking binding energies were almost all less than − 5 kcal/mol, indicating good binding between the ligand and the receptor. Notably, β-carotene had the highest binding energy to TNF (-10.63 kcal/mol). Finally, we ranked the docking binding energy and selected the first eight groups of molecules for ligand‒receptor visualization (Fig. 4), where quercetin‒TNF and nobiletin‒TNF were connected by hydrogen bonds. Based on these results, we speculate that all these compounds play important roles in mediating the therapeutic effect against COVID-19.

Molecular dynamic stimulation

The root mean square deviation of the molecular dynamics simulation can respond to the motion process of the complex; the larger the RMSD is, the more intense the fluctuation, and the more intense the movement; in contrast, the more stable the movement. According to Fig. 6, β-carotene-TNF showed some stability in the simulated system, with an RMSD value ranging between 0.07 and 0.32 (mean value: 0.27). Moreover, the RMSD values tended to increase within the initial 0 to 20 ns and tended to converge and remain stable after 20 ns, which was attributed mainly to the conformational change within the protein structure. Although β-carotene-IL-6 and β-carotene-EGFR were less stable in the simulated system than was β-carotene-TNF, the RMSD values were between 0.11 and 0.44 (mean 0.34) and 0.08 and 0.39 (mean 0.33), respectively but also tended to be relatively stable after 20 ns. Furthermore, the absence of breaks in the RMSD curve indicated that the compound had a strong affinity for TNF, IL-6 and EGFR, suggesting a consistent interaction with the protein pocket throughout the simulation.

Local structural fluctuations were assessed by determining the RMSF of individual residues within the molecular dynamics simulation trajectory. As shown in Fig. 6, the RMSF curves of the three complexes peak in parts of the protein, indicating significant fluctuations in these regions (0.44 for β-carotene-TNF, 0.44 for β-carotene-IL-6, and 0.54 for β-carotene-EGFR). However, the RMSF values of the other amino acid residues can still remain low, meaning that changes in the residues did not induce conformational changes in the entire system.

Rg characterizes the density of the protein structure, reflecting the changes in protein‒peptide chain relaxation during the simulation. A smaller Rg value indicates that the structure is more normal and stable. The Rg results for each system showed a downward trend from 0 to 30 ns and then stabilized, indicating that in the simulation (β-carotene-TNF: maximum 2.22, minimum 2.17, mean 2.19; β-carotene-IL-6: maximum 1.68, minimum 1.60, mean 1.63; β-carotene-EGFR: maximum 2.11, minimum 2.03, mean value 2.08). Overall, the low Rg and low Rg variation indicated that a stable protein structure and ligand binding had a relatively small effect on the protein structure.

The temporal variation in SASA can be used to quantify the exposed area of the protein in the solvent. The SASA curve for each system throughout the simulation is shown in Fig. 6. The SASA curves for the β-carotene-TNF complex, β-carotene-IL-6 complex and β-carotene-EGFR complex significantly decreased (β-carotene-TNF: maximum 217, minimum 191, mean 202; β-carotene-IL-6: maximum 111, minimum 92, mean 100; β-carotene-EGFR: maximum 172, minimum 151, mean 162). These results suggest the potential for small molecules to occupy the protein binding cavity, replace existing water molecules and thus form stable complexes (Fig. 5).

Fig. 5.

(a-i) Representative docking complexes of candidate components for molecular dynamics simulation. (a and d) β-carotene-TNF. (b and e) β-carotene-IL6. (c and f) β-carotene-EGFR. (g-i) Root mean square deviation of the candidate components. (g) β-carotene-TNF. (h) β-carotene-IL6. (i) β-carotene-EGFR. (j-l) Root mean square fluctuation values of the residues. (j) β-carotene-TNF. (k) β-carotene-IL6. (l) β-carotene-EGFR. (m-r) Changes in the solvent accessibility and surface area of the protein. (m and p) β-carotene-TNF. (n and q) β-carotene-IL6. (o and r) β-carotene-EGFR. (s-u) Gyration radius of the protein. (s) β-carotene-TNF. (t) β-carotene-IL6. (u) β-carotene-EGFR.

Discussion

This study used network pharmacology, molecular docking, and molecular dynamics simulations to elucidate the mechanisms of action and therapeutic effects of XXD against COVID-19. HPLC-Q-TOF-MS/MS analysis revealed that flavonoids, saponins, and coumarins were the primary chemical constituents of XXD. This composition is highly correlated with its hypothesized therapeutic effects against COVID-19. Flavonoids, such as quercetin, kaempferol, and luteolin are a major class of dietary polyphenols naturally found in herbs, plant-based foods, and beverages. Flavonoids are widely recognized for their broad-spectrum antiviral, and immunomodulatory properties. Studies have demonstrated that flavonoids exert direct antiviral effects by inhibiting various stages of the viral infection cycle and indirect effects by modulating the host’s response to viral infection and subsequent complications. Additionally, flavonoids exhibit antiviral activity by inhibiting viral proteases, RNA polymerase, mRNA, viral replication, and infectivity⁴⁷. Specifically, quercetin and luteolin can inhibit SARS-CoV-2 entry into cells by interfering with the binding of spike proteins to the ACE2 receptor and by suppressing key viral proteases (3CLpro and PLpro). Furthermore, their potent inhibitory effects on pro-inflammatory signaling pathways and cytokines make them important active components in mitigating severe COVID-19-associated cytokine storms⁴⁸. Saponins are renowned for their membrane-disrupting effects and immunoadjuvant activity⁴⁹. They can disrupt viral envelopes, inhibit viral binding sites, or delay viral genome replication by interfering with viral enzymes, thereby suppressing or blocking viral replication. Additionally, saponins enhance humoral and cellular immune responses, facilitating viral clearance. By modulating macrophage polarization and cytokine production, they exert anti-inflammatory effects, further contributing to potential pulmonary protection⁵⁰. These findings suggest that saponins in XXD may support host defense mechanisms, thereby playing a role in combating COVID-19. Coumarins exhibit significant anti-inflammatory, anticoagulant, and antiviral effects⁵¹. Given the thromboembolic complications and hypercoagulable state observed in severe COVID-19 patients, the potential anticoagulant activity of coumarins may provide additional therapeutic benefits. Some coumarin derivatives have also demonstrated inhibitory effects on viral replication⁵². Therefore, coumarins may broaden the therapeutic scope of XXD by addressing coagulation dysfunction in COVID-19. The coexistence of these bioactive components in XXD demonstrates a synergistic multi-mechanism strategy, which provides a robust material basis for the multi-target therapeutic effects predicted through network pharmacology and molecular docking studies. This reflects the holistic “multi-component, multi-target” concept inherent in traditional Chinese herbal formulations.

Through pharmacological analysis, we screened 92 compounds and then selected conditions for further screening, identifying major chemical components such as quercetin, luteolin, kaempferol, β-carotene, and nobiletin. These compounds intervene in and treat COVID-19 by modulating multiple signaling pathways. This study revealed that XXD could effectively treat COVID-19 through various active ingredients. Quercetin, a subclass of flavonoids, is widely found in vegetables and fruits and has excellent antiviral, anti-inflammatory, and antioxidant properties. Quercetin can effectively prevent infection by blocking viral particles from entering host cells, inhibiting proinflammatory signals, alleviating coagulation abnormalities, and binding to multiple viral protein targets to inhibit replication and transcription⁵³. Luteolin, a natural flavonoid, is widely present in vegetables, fruits, traditional Chinese herbs and spices and has the ability to inhibit viral activity⁵⁴. Studies have shown that the combination of luteolin and palmitoyl ethanolamide (PEA-LUT) can improve olfactory dysfunction caused by COVID-19⁵⁵. Additionally, kaempferol, a natural flavonoid commonly found in vegetables, fruits, and traditional Chinese herbs, exhibits antibacterial, anti-inflammatory, and antioxidant properties. The main protease (Mpro) of shenylphenol is often used as a core drug target for treating COVID-19. Research findings have shown that shenylphenol typically exerts its anti-COVID-19 effects by directly targeting the main protease (3CL pro) of SARS-CoV-2⁵⁶. β-Carotene, a type of carotenoid and a natural precursor to vitamin A, is commonly found in algae and traditional Chinese herbs. On the one hand, studies have shown that deficiencies in vitamins A and D are associated with COVID-19 infection, so patients infected with COVID-19 can supplement more vitamin-rich substances; on the other hand, owing to its antioxidant activity, β-carotene can stimulate lymphocyte responses, interleukin production, and natural killer cell activity, thereby enhancing the entire immune system to combat viral invasion⁵⁷. As a natural poly-methoxyflavone, nobiletin usually comes from the peel of citrus fruits and has preventive and therapeutic effects on various respiratory diseases. In the treatment of COVID-19, nobiletin can inhibit virus entry into cells, reduce inflammation, increase immunity, induce apoptosis, prevent oxidative stress, and restore cell viability⁵⁸. In summary, the active components identified in this study may play crucial roles in the therapeutic efficacy of XXD against COVID-19, suggesting potential research value.

After merging datasets to remove duplicate values, we identified cross-targets of XXD and COVID-19, including IL-6, AKT1, TNF, HIF1A, STAT3, and EGFR. Molecular docking revealed strong interactions between quercetin, luteolin, kaempferol, β-carotene, and nobiletin. Among these, the three pairs with the lowest binding energies were TNF / β-carotene, IL-6 / β-carotene, and EGFR / β-carotene.

Tumor necrosis factor (TNF) is an efficient cytokine primarily produced by monocytes and macrophages, but it can also be generated by B cells, T cells, and fibroblasts. It binds to its cognate receptors, TNF receptor 1 (TNFR1) and TNF receptor 2 (TNFR2), thereby modulating cellular processes like inflammation and cell death in specific physiological and pathological contexts. TNF signaling also plays a significant role in neuroinflammation and is crucial in neuroimmunological and neurodegenerative diseases. Excessive TNF-α is associated with autoimmune diseases and the onset and progression of cancer⁵⁹. Studies have shown that TNF-α can directly cause deterioration of the respiratory epithelium by producing inflammatory cytokines and can induce neutrophils to release MMP-9, leading to irreversible changes in pulmonary fibrosis⁶⁰. Researchers detected TNF-α in the blood and tissues of COVID-19 patients and reported that TNF-α, an inflammatory factor in patients’ blood, was upregulated. Experiments have also demonstrated that a single infusion of TNF-α antibodies can reduce some processes during COVID-19-related lung inflammation, decreasing the levels of TNF-α and other inflammatory mediators and exudates⁶¹.

In this study, IL-6 and AKT1 showed strong affinities to quercetin, nobiletin, and beta-carotene. As a key inflammatory cytokine, IL-6 plays a central role in acute inflammatory responses and cytokine storms. During COVID-19, activated monocytes, macrophages, and dendritic cells secrete IL-6 and other inflammatory cytokines. Elevated levels of IL-6 in the serum often accompany the development of acute respiratory distress syndrome, multiple organ failure, and clinical manifestations of COVID-19⁶². The expression of IL6 leads to an increase in C-reactive protein (CRP) levels, which are a significant biomarker for severe COVID-19. Therefore, studies have shown that inhibiting the IL-6 pathway may improve outcomes in critically ill patients with COVID-19⁶³. The PI3K / Akt / mTOR pathway is a crucial cellular signaling pathway that regulates various cellular functions and plays a vital role in catabolic processes, nutrient absorption, cell growth, differentiation, survival, proliferation, and movement. Research has revealed upregulated expression of this pathway in several diseases, including those related to tumors, autoimmunity, and viruses, suggesting its involvement in the pathogenesis of these conditions. Single or dual inhibitors may improve SARS-CoV-2 infection. Among them, AKT1 is a subtype of the Akt kinase family that plays a crucial role in cell survival and has positive impacts on metabolism and cancer treatment⁶⁴. Additionally, inhibition of Akt increases the number of regulatory T cells (Tregs) in the lungs of COVID-19 patients, thereby reducing inflammatory damage⁶⁵. Moreover, Akt inhibition may also suppress cytokine storms, fibrosis, and platelet activation in COVID-19 patients and prevent scar formation⁶⁶. Clearly, the common key targets of XXD and COVID-19 interact with each other, indicating their involvement in the pathogenesis and disease mechanisms of COVID-19, such as cell migration and cytokine production. These findings support the results of the GO and KEGG enrichment analyses.

Through GO enrichment analysis, we identified biological processes enriched in the targeted gene set, primarily positive regulation of cell migration, cell mobility, and cell movement. Among these processes, the positive regulation of cell migration plays a crucial role in the pathogenesis of COVID-19. Cell migration is essential for many physiological and pathophysiological processes, such as embryonic development, tissue morphogenesis, immune surveillance and inflammation, wound healing, and cancer metastasis, and serves as a hallmark of diseases such as cancer and malignant tumors⁶⁷. Cells migrate primarily in the body through four core biological processes: adhesion, cytoskeletal dynamics, nuclear dynamics, and matrix remodeling⁶⁸. When tissues or organs are damaged, the injured site releases biochemical signals (usually chemokines and cytokines) to gradient drive cell migration. Subsequently, different migratory pathways coordinate cell polarization, causing corresponding changes in proteins at the anterior and posterior parts of the polarity axis, which further leads to cell adhesion with the extracellular matrix (ECM) and subsequent movement⁶⁹. Therefore, the chemokine / receptor axis is critical in the pathogenesis of SARS-CoV-2 infection, mediating the transport of proinflammatory or anti-inflammatory monocytes to the infected area. After infection with COVID-19, monocytes migrate to tissues and partially differentiate into resident macrophages, while the remaining monocytes bind to macrophages and inhibit viral replication and spread by producing type I interferons⁷⁰. Studies have shown that chemokines play indispensable roles in monocyte migration. In inflamed lungs, macrophages and epithelial cells produce a wide range of chemokines that may serve as chemoattractants for monocytes / macrophages. Monocyte chemokine receptors can only undergo the correct migration process in the respiratory system when they recognize homologous chemokines expressed in inflamed airways⁷¹.

KEGG enrichment analysis revealed that XXD, when used to treat COVID-19, activated multiple signaling pathways. These included pathways related to cancer, lipids, and atherosclerosis, as well as the fluid shear stress and atherosclerosis pathway. Notably, the JAK-STAT, MAPK, PI3K/Akt, and NF-κB signaling pathways were identified as key pathways with significant therapeutic relevance. Studies indicate that cytokine receptors activate the JAK/STAT pathway by phosphorylating JAK (Janus kinase) and STAT (signal transducer and activator of transcription). Subsequently, STAT translocates to the nucleus, binds DNA, and promotes the expression of inflammatory mediators. This pathway critically regulates cytokine and chemokine signaling to modulate inflammation in organisms. The JAK-STAT signaling pathway can affect immune system function to a large extent by activating IL6 signaling processes, thus aggravating COVID-19. Therefore, inhibiting SAR-CoV-2 infection caused by cytokine storms is a coping strategy for the treatment of excessive inflammation⁷². The mitogen-activated protein kinases (MAPKs) constitute a highly conserved serine‒threonine protein kinase family that can transmit extracellular signals (including viral infection signals) to the nucleus through protein‒protein interactions and then transform into various output signals, which may affect cell apoptosis and host immune defense mechanisms⁷³. Extracellular signal-regulated kinase (ERK, also known as p42 / 44 MAPK), Janus kinase (JNK, also known as stress-activated protein kinase-1 (SAPK1)) and p38 MAP kinase (also known as SAPK 2 / RK) are the three major MAPK pathways in mammals. When the body is infected by viruses, the MAPK pathway is activated by the virus, and the conduction signals generated may also promote viral self-replication⁷⁴. The PI3K / Akt signaling pathway has become a novel therapeutic target for SARS-CoV-2 infection and participates in virus entry and the host immune response. It is well known that ACE 2 and CD147 are the major receptors for SARS-CoV-2 after entry, while other factors on the cell surface that access viral genomes in host cells are TMPRSS2 and Furin. Among them, the activation of the CD147 and Furin proteases facilitated the cascade of the PI3K / Akt signaling pathway. Moreover, when ACE2 bound to the virus, its endocytosis occurred simultaneously with the virus. Viral endocytosis is mediated through the clathrin pathway, which in turn is regulated by positive feedback from the PI3K / Akt signaling pathway⁷⁵. In addition, the PI3K / Akt pathway induces the phosphorylation of IFN regulator factor 3 (IRF 3) and type I interferon (IFN-I) to protect against viral invasion⁷⁶. The NF‒kappa B signaling pathway also plays a key role in the response to external stimuli such as cytokines, stress, UV, antigens, and heavy metals. Studies have shown that NF-kappa B signaling is involved in various physiological and pathological processes, including immune and inflammatory responses, cell survival and proliferation, metabolism, synaptic plasticity and memory-related activities⁷⁷. After infection with SARS-CoV-2, NF-kappa B signaling is activated in both classical and nonclassical ways. In the classical pathway, the proliferation of viruses in the host cell leads to the production and accumulation of dsRNA (transcription intermediate) in the cytoplasm, and after the combination of PKR (a dsRNA-dependent protein kinase, a threonine kinase), PKR is also activated, further activating the inhibitor of IKK (NF-kappa B kinase) and ultimately promoting the degradation of inhibitors and the activation of NF-kaapa B. When the signaling of transcription factors is transferred to the nucleus, the production and transcription of proinflammatory mediators can be initiated⁷⁸.

We evaluated the degree of protein‒compound binding via molecular docking and showed that most of the compounds had good binding affinity, with quercetin and nobiletin connected to TNF via hydrogen bonding. These results suggest that XXD may alleviate the symptoms of COVID-19 through a synergistic mechanism involving multiple targets and multiple pathways. Finally, we used molecular dynamics simulations to further verify the binding between β-carotene and TNF, IL-6, and EGFR. Among the three systems, the stability between β-carotene and TNF was better than that of the other two systems, but the three systems reached a steady state after 20 s and had small Rg values and better SASA values, indicating that β-carotene had a strong affinity for TNF, IL-6 and EGFR. The above analysis suggests that the molecular dynamics simulations provide preliminary structural insights into the potential stability of the complex.

UPLC-Q-TOF-MS/MS technology was used to collect high-resolution, high-precision mass spectrometry data for XXD. We were able to accurately identify the relevant chemical components according to the structure of the substance and the corresponding fragmentation pattern. This technique provides a reliable method for polysubstance identification in TCM compounds. We showed that the mass spectrometry data indicated abundant peaks in both positive and negative ion modes, laying the basis for the quasideterministic identification of chemicals in XXD. We prepared XXD solutions separately and ultimately identified a total of 159 chemicals at low concentrations of XXD, while we identified 74 compounds at high concentrations of XXD, including many flavonoids, saponins, and coumarins⁷⁹. Owing to its high speed, simplicity and high sensitivity, this method provides a solid foundation for the quality control and pharmacological studies of TCM compounds.

There are several limitations of this study that need to be noted. The mechanism analysis based on network pharmacology and molecular simulation is essentially predictive, and its final confirmation still requires validation through subsequent functional experiments.

In addition to predicting the molecular mechanisms of XXD in alleviating COVID-19, this study also provides a translational roadmap for modern drug development. Our prediction results indicate that specific flavonoids such as quercetin, luteolin as lead compounds serve as starting points for structural optimization. The multi-target network targeting viral infections and hyperinflammatory responses supports the adoption of multi-component pharmacological strategies for treating complex diseases. Furthermore, the integrated “prediction-validation-deepening” approach established in this study can serve as a replicable template to accelerate the discovery of active ingredients from traditional formulations, thereby reducing risks and guiding subsequent steps in the R&D pipeline, which from in vitro and animal testing to standardized botanical drug development.

Conclusion

In this study, we adopted an integrated strategy combining HPLC-Q-TOF-MS/MS analysis, network pharmacology, molecular docking, and molecular dynamics simulations to systematically investigate the potential mechanisms of XXD in the treatment of COVID-19. Chemical analysis identified multiple bioactive components in XXD, including quercetin, kaempferol, luteolin, nobiletin and β-carotene. Computational predictions suggested that these components may interact with core targets such as TNF, AKT1, IL-6, IL1B, HIF1A, STAT3, and EGFR, thereby regulating signaling pathways associated with viral infection and inflammatory responses. Molecular dynamics simulations further supported the stability of these predicted interactions. The primary contribution of this study lies in proposing a comprehensive hypothesis regarding the multi-component, muti-target, and muti-pathway action mechanism of XXD against COVID-19. This integrated approach of “chemical component identification-computational prediction-molecular simulation” provides a methodological framework for elucidating the complex pharmacological mechanisms of TCM compound formulations.

However, it is necessary to acknowledge the limitations of this study. The research findings are primarily computational and predictive in nature. Although the chemical composition has been experimentally identified and pathway regulation have not yet been validated through in vitro or in vivo functional experiments. Meanwhile, the target prediction in this study primarily relies on the TCMSP database. Alothough TCMSP is a pioneering resource, its static nature may not cover newly discovered targets or interactions, potential missing some potential target. Future work could achieve more comprehensive target information by integrating multiple complementary prediction platforms and combining experimental validation. Additionally, while our ADME screening includes common thresholds and additional parameters such as BBB permeability, it is based on predictive models. The pharmacokinetics of compounds in compound formulations may differ from single-molecule-based predictions. Furthermore, the clinical relevance of these predictions remains to be established, and the model does not fully account for the complex interactions among the components in the drug solution. So, future research should focus on experimental validation of the core components and targets for prediction. Furthermore, while molecular dynamics simulations provide valuable insights into structural dynamics, this study did not include quantitative binding free energy calculations, such as the molecular mechanics Poisson-Boltzmann surface area method (MM/ PBSA). These calculations can provide direct thermodynamic affinity data to supplement structural stability observations. We acknowledge this as a limitation and plan to conduct such analyses in future studies to further consolidate the energy basis of interactions.

In conclusion, this study provides a detailed mechanistic hypothesis and offers a novel research approach for future investigations into the potential role of XXD in improving COVID-19, thereby establishing a valuable computational foundation for understanding the scientific basis of this traditional formula.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Y.M. conceived the study, designed the experimental plan, wrote the manuscript, and performed data analysis. D.L. conducted data analysis, validated the study results, and participated in manuscript editing. Y.G. and H.K. carried out molecular dynamics simulations and processed the data. J.W. and J.C. were responsible for funding acquisition and manuscript review. All authors have read, reviewed, and approved the final version of the article.

Funding

This research was supported by The Project Supported by Zhejiang Provincial Natural Science Foundation of China (ZCLMS25H2703), the National Natural Science Foundation of China (82174272); Key projects of Jinhua Science and Technology Bureau (2024-3-038).

Data availability

All data generated or analysed during this study are included in this published article and its supplementary information files.

Declarations

Competing interests

The authors declare no competing interests.