Natural product virtual-interact-phenotypic target characterization: A novel approach demonstrated with Salvia miltiorrhiza extract

Abstract

Natural products (NPs) have historically been a fundamental source for drug discovery. Yet the complex nature of NPs presents substantial challenges in pinpointing bioactive constituents, and corresponding targets. In the present study, an innovative natural product virtual screening-interaction-phenotype (NP-VIP) strategy that integrates virtual screening, chemical proteomics, and metabolomics to identify and validate the bioactive targets of NPs. This approach reduces false positive results and enhances the efficiency of target identification. Salvia miltiorrhiza (SM), a herb with recognized therapeutic potential against ischemic stroke (IS), was used to illustrate the workflow. Utilizing virtual screening, chemical proteomics, and metabolomics, potential therapeutic targets for SM in the IS treatment were identified, totaling 29, 100, and 78, respectively. Further analysis via the NP-VIP strategy highlighted five high-confidence targets, including poly [ADP-ribose] polymerase 1 (PARP1), signal transducer and activator of transcription 3 (STAT3), amyloid precursor protein (APP), glutamate-ammonia ligase (GLUL), and glutamate decarboxylase 67 (GAD67). These targets were subsequently validated and found to play critical roles in the neuroprotective effects of SM. The study not only underscores the importance of SM in treating IS but also sets a precedent for NP research, proposing a comprehensive approach that could be adapted for broader pharmacological explorations.

Graphical abstract

Highlights

• •A multi-target NP-VIP strategy was introduced for natural product research.
• •Five high-confidence targets were identified from Salvia miltiorrhiza extract.
• •A target identification methodological paradigm was established for NPs.
• •Simulations, proteomics, and metabolomics were combined to enhance the research depth.

1. Introduction

Natural products (NPs) are derived from plants, animals, or microorganisms [1]. Owing to the diverse chemical structures and potential medicinal potential, NPs remain integral to drug development [2,3]. Commonly incorporated into medical practice, NPs encompass Chinese herbal medicine, Japanese Kampo medicine, Indian Ayurvedic medicine, and herbal therapies employed in the Americas. Notable examples include artemisinin [4], vincristine [5] and penicillin [6], all of which have achieved widespread clinical application. Despite this, these isolated and purified single components constitute only a fraction of NPs [7,8], typically found as complex mixtures. Characterizing and discovering the active ingredients of NPs, particularly elucidation of their mechanisms of action, remains a significant bottleneck in this field [9]. Identifying target proteins or cellular signaling pathways is crucial for understanding the biological functions and the pharmacological mechanisms of NPs [10]. Consequently, developing target identification methods tailored to the complex nature of NPs and their potential multi-pathway effects is essential in the study of NPs.

Recent advancements in computational power and artificial intelligence have enabled the exploration of target groups comprising multiple components [11]. For instance, Lyu et al. [12] efficiently docked 170 million compounds against a receptor within just 1.2 days using 1500 cores, subsequently validating the activity of screened molecules. By leveraging the chemical profiling of NPs, reverse target identification is facilitated through open-source websites such as LeDock [13], STITCH [14] and DrugBank [15]. Moreover, specific biological tools predict action pathways based on ligand structural similarity, taking into account potential bioactivity [16]. Ligand-based biological similarity screening extensively evaluates the biological properties of compounds to predict their activity against specific targets [17]. An exemplary instance is Chemical Checker, which converts biological activity data into machine-learning-compatible formats, thereby improving data utilization in virtual screening (VS) [18]. These VS methods have significantly advanced the targets discovery of NPs.

However, due to the challenge of fully characterizing NP components, VS still yields numerous false positive results. Therefore, experimental exploration and validation are crucial for identifying targets of NPs. Chemical proteomics is an emerging discipline focused on identifying compound targets techniques such as activity-based protein profiling (ABPP) [19], and compound-centric chemical proteomics (CCCP) [20] has been proven successful in identifying the targets of individual compounds such as artemisinin [4,21] and ganoderic acid D [22]. However, these methods, requiring compound modification, are limited in their application due to the complexity of NP mixtures. Alternatively, unlabeled chemical proteomics methods, such as stability of proteins from rates of oxidation (SPROX) [23] and cellular thermal shift assay (CETSA) [24], employ oxidation reactions and thermal stability assessments to identify drug targets by evaluating the impact of drug-protein interaction. In principle, these unbiased methods are applicable for identifying the target groups in NP mixtures, providing a promising approach for target identification.

It is noteworthy that both VS and chemical proteomics focus on confirming the interaction between compounds and targets. However, as NPs exert effects in cells or organisms, some active ingredients interact through non-covalent interactions with specific catalytic proteins or alter their configurations, thereby rapidly inducing phenotypic changes [25]. Therefore, understanding the proteins impacted by these phenotypic changes is crucial for a comprehensive elucidation of NP mechanisms of action. Building on this understanding, our research group has already identified target ensembles through analysis of the NP components, elucidating the anti-cancer mechanisms via multi-dose metabolomics to pinpoint targets within NPs [26].

To address the challenges in the target identification of NP, this study proposed a natural product virtual screening-interaction-phenotype (NP-VIP) method that integrates the merits of VS, CETSA and metabolomics, thereby enhancing the accuracy and efficiency of identifying target ensembles for NPs (Fig. 1). Salvia miltiorrhiza (SM), also known as Danshen, is derived from the dried root and rhizome of Salvia miltiorrhiza Bunge, a perennial plant in the Labiatae family native to China [27]. This traditional Chinese medicine is renowned for its potent antioxidant, anticoagulant, and anti-inflammatory properties, which make it particularly effective in treating cardiovascular and cerebrovascular diseases, including ischemic stroke (IS) and angina pectoris [28,29]. The herb's medicinal value is well-documented, first appearing in "Shennong's Herbal Classic", the earliest known treatise on Chinese herbal medicine, with its significance further underscored by its inclusion in the Chinese Pharmacopoeia. In this study, SM was used to demonstrate the workflow. Despite formulations associated with SM advancing to Phase III clinical trials, comprehensive research focusing on SM's molecular targets remains notably limited. In this context, the proposed VIP method offers a holistic strategy tailored for targets identification of SM. Specifically, VS is initially employed to rapidly and efficiently screen for potential targets of small molecules present in SM. Subsequently, CETSA is employed to identify proteins that directly bind with the compounds in SM and metabolomics, linked to phenotypic changes, and is further used to discover the phenotypically relevant target proteins in SM. By integrating the mentioned three target identification methods, this research thoroughly investigated the targets and pathways influenced by SM in treating IS, providing crucial evidence for the subsequent clinical application of SM. Furthermore, the VIP method serves as a reference for identifying targets in other substances exhibiting mixture effects.

Fig. 1.

The standard workflow of natural product virtual screening-interaction-phenotype (NP-VIP). The key steps involved extraction and characterization of the natural product, identification of the potential target groups using in silico, chemical proteomics and multi-dose metabolomics approaches. PISA: proteome integral solubility alteration; GNPS: global natural products social molecular networking.

2. Materials and methods

2.1. Antibodies and reagents

The primary antibody against β-actin (81115-1-RR) and the secondary antibodies conjugated with horseradish peroxidase (HRP) were obtained from Proteintech (Wuhan, China). The primary antibody against poly [ADP-ribose] polymerase 1 (PARP1; YP-Ab-10859), signal transducer and activator of transcription 3 (STAT3; YP-Ab-02054), amyloid precursor protein (APP; YP-Ab-12871), glutamate-ammonia ligase (GLUL; YP-Ab-12846) and glutamate decarboxylase 67 (GAD67; YP-Ab-12726) were sourced from UpingBio (Shenzhen, China). Dichloromethane (DCM), acetonitrile (ACN) and formic acid were procured from Thermo Fisher Scientific (Shanghai, China). The PC12 cell lines were supplied by Pricella (Wuhan, China). Roswell Park Memorial Institute (RPMI) 1640 culture medium and fetal bovine serum (FBS) were acquired from Biosharp. Bicinchoninic acid (BCA) protein assay kit, and dimethyl sulfoxide (DMSO) were sourced from Beyotime Biotechnology (Shanghai, China). Methanol, 2-chloroacetamide (CAA), triethylammonium bicarbonate (TEAB), tris (2-carboxyethyl) phosphine (TCEP), trifluoroacetic acid (TFA), ammonium formate, tris-(hydroxymethyl) aminomethane (Tris), protease and phosphatase inhibitor cocktail were sourced from Sigma-Aldrich (St. Louis, MO, USA). Pierce quantitative colorimetric peptide assay kit and Tandem Mass Tag™ (TMT™) 10-plex were acquired from Thermo Fisher Scientific (Shanghai, China).

2.2. Chemical profiling and analytical method of Salvia miltiorrhiza extract (SME)

A total of 34 g of SME was derived from 250 g of SM. This extraction process entailed two consecutive treatments with 2.5 L of 75% ethanol each, followed by concentration via rotary evaporation and subsequent lyophilization. After dissolving the lyophilized samples in methanol, they were followed by dry loading and subsequent subjection to silica gel column chromatography using a 1:30 (w/w) of SME to silica gel. Dichloromethane and methanol were mixed in various ratios (40:1, 20:1, 10:1, 5:1, 2:1, 1:1, 1:2, 1:5, 1:10, 1:20, 1:40, 0:100, v/v), then sequentially subjected to gradient elution, rotary evaporation, and finally dissolved in methanol to a concentration of 2 μg/mL. The components were separated using an ACQUITY UPLC® BEH C₁₈ column (2.1 mm × 100 mm, 1.7 μm; Waters, Milford, MA, USA), and mass spectrometry (MS)1 and MS2 spectra were acquired using Thermo Orbitrap Fusion high-resolution mass spectrometry. The detailed mass spectrometry detection methods can be found in Text S1 of Supplementary data.

2.3. Compounds identification and molecular-network analysis by global natural products social molecular networking (GNPS)

Before GNPS analysis, the raw MS data were converted to mzXL format using MSConvert. Molecular networks were created using the online workflow on the GNPS website (http://gnps.ucsd.edu, accessed on 22 February 2022). The precursor ion mass tolerance was set at 2.0 Da and the MS2 fragment ion tolerance at 0.5 Da. A network was then created where edges were filtered to have a cosine score above 0.7 and more than six peaks matched. The achieved GNPS molecular networking was further annotated with MolNetEnhancer. The generated molecular network was imported into Cytoscape v3.7.1 (The Cytoscape Consortium, New York, NY, USA) and displayed as nodes and edges.

Meanwhile, the MS1 and MS2 raw data of the mass spectra acquired by high-resolution mass spectrometry were imported into the Compound Discoverer analysis software. The compounds were identified and compared based on their precise molecular weights, secondary spectra, and compound names. Through these two identification methods, the chemical composition database of SM was constructed, detailing each component.

2.4. Cell culture and oxygen-glucose deprivation/recovery (OGD/R) model

PC12 cells were cultured in RPMI 1640 medium supplemented with 10% FBS in a biochemical incubator under a humidified atmosphere containing 5% CO₂ at 37 °C. To simulate in vitro IS-like conditions, PC12 cells underwent OGD/R induction. The cells were cultured and allowed to adhere for 24 h before being washed twice with phosphate-buffered saline (PBS). The cells were incubated in glucose-free RPMI 1640 medium in an anaerobic chamber (1% oxygen, 94% nitrogen, 5% carbon dioxide) at 37 °C for 120 min to induce OGD injury. The cells were cultured in normal RPMI 1640 medium under normoxic conditions for 24 h as reperfusion. Detailed methods for cell viability assays are provided in the Text S2 of Supplementary data.

2.5. In vitro fluorescence detection method

2.5.1. Calcein-AM/Propidium Iodide (PI) staining

PC12 cells were plated at a density of 5.0 × 10⁴ cells per well in six-well plates, each containing 2.0 mL of culture medium. Following exposure to diverse stimuli, the cells were washed with PBS. Resuspended in 1x assay buffer, the cells were then stained with 0.5 μg/mL calcein-AM and 5 μM propidium iodide per well, followed by incubation at 37 °C for 30 min. Cell images were captured immediately using a fluorescence microscope (Nikon ECLIPSE 50i, Nikon, Tokyo, Japan).

2.5.2. Fluo-4 acetoxymethyl ester (Flou-4 AM) staining

The Fluo-4 AM fluorescent probe was used to detect intracellular Ca²⁺ levels. PC12 cells were plated in a 6-well plate and incubated with 2 μM Fluo-4 AM for 30 min in the dark. After washing with PBS, the cells were observed under a fluorescence microscope (Nikon ECLIPSE 50i).

2.6. Transmission electron microscopy (TEM)

The prepared cell samples were washed to remove the culture medium, fixed in 2.5% glutaraldehyde PBS buffer at room temperature for 2 h, followed by overnight fixation at 4 °C. Dehydration was performed using a gradient of 50%, 70%, 90%, and 100% ethanol, with each concentration dehydrated for 15 min. Images were captured using a Tecnai T10 transmission electron microscope (Hillsboro, OR, USA).

2.7. Rat model of middle cerebral artery occlusion

The rat stroke model was established using a modified ischemia/reperfusion (I/R) method. Following surgery, the rats were placed in an incubator at 37 ± 0.5 °C, and neurobehavioral scoring was conducted after 24 h. The Laboratory Animal Welfare and Ethics Committee of Zhejiang University approved the animal study protocols (Approval number: 161 ZJU20220457). All experiments adhered to relevant regulatory standards. Detailed animal modeling and neurologic deficit scoring methods are provided in Text S3 of Supplementary data. Detailed 2,3,5-triphenyltetrazolium chloride staining, Nissl staining, hematoxylin-eosin (H&E) staining and terminal deoxynucleotidyl transferase-mediated dUTP nick end labeling (TUNEL)/neuronal nuclei (NeuN) immunofluorescent double-labeling staining are provided in Text S4 of Supplementary data.

2.8. VS

Only those with a canonical SMILES structure, identified by ultra-performance liquid chromatography-mass spectrometry (UPLC-MS), were selected. Information regarding these compounds, including structure and canonical name, was retrieved using the “pubchempy" Python package. Targets were obtained from Swiss Target Prediction (probability >0.6, http://www.swisstargetprediction.ch/) and Similarity Ensemble Approach (Max Tanimoto coefficients >0.6, https://sea.bkslab.org/). Functional and cluster enrichment analyses were performed using the DAVID database (https://david.ncifcrf.gov/) and Metascape (https://metascape.org). Network analysis was performed using Cytoscape 3.9.1. Targets based on the biological activities were obtained from Chemical Checker, and batch searches of the compounds were performed through Compound ID (CID) to retrieve target information. The IS targets were obtained from the GeneCards database. The intersection of targets from structural, biological, and disease-based screenings was identified, and those targets were further subjected to docking using Pymol and Autodock.

2.9. Chemical proteomics analysis

Proteome integral solubility alteration (PISA) assay is designed to enhance the sensitivity and information content of chemoproteomics analysis by integrating methods such as CETSA with modern mass spectrometry techniques [30]. Initially, a sufficient quantity of OGD-treated PC12 cells is procured for protein extraction. The lysate is then evenly divided into two distinct experimental cohorts: a control group treated with DMSO and SME group at 1 mg/mL. Each group is incubated with the same protein solution at ambient temperature for 3 min. After incubation, a thermal gradient of 46–56 °C is applied. This thermal treatment is conducted using a polymerase chain reaction (PCR) instrument. After thermal processing, the proteins undergo a series of biochemical modifications including reduction, alkylation, and enzymatic digestion. This sequential treatment ensures comprehensive lysis, efficient desalting, and subsequent quantification of the resultant peptides.

The subsequent phase involves tandem mass tag (TMT) labeling of the peptides. After TMT labeling, the samples from each group are combined for a subsequent desalting procedure. In the final stage of the protocol, the samples undergo fractionation using a C₁₈ column. 2 μg of peptides from each fraction are separated using a Proxeon 1000 UHPLC system, coupled with an Orbitrap Exploris 480 mass spectrometer (Thermo Fisher Scientific, Shanghai, China). Raw MS data acquisition is performed in a data-dependent mode, achieving an MS resolution of 75,000 and an MS/MS resolution of 50,000. Protein identification and quantitative are conducted using the Proteome Discoverer 2.1 software (Thermo Fisher Scientific, Shanghai, China), in conjunction with the Mascot 2.6.0 search engine (Matrix Science, London, UK).

2.10. Untargeted metabolomics analysis

PC12 cells were plated in 6-well plates and incubated for 24 h, then exposed to various concentrations (0.3, 1.2, and 5.0 μg/mL) of SME for 24 h, with three replicate wells per concentration, before being subjected to OGD/R modeling. Cells were then harvested using 80% methanol and subjected to three freeze-thaw cycles in liquid nitrogen, followed by centrifugation at 20,000 rpm for 20 min at 4 °C. Protein concentration was measured with a BCA assay to normalize the re-solubilization volume; subsequently, the supernatant was freeze-dried for further analysis via low-temperature centrifugation. Quality control (QC) samples were prepared by combining 5 μL of the solutions obtained from each sample. The specific analysis methods and parameters were adapted from a previous study [26]. Multivariate statistical analysis, data visualization, TOXCMS analysis [31], volcano plot, diagram, and heat map were conducted using R. Functional enrichment analysis was performed using MetaboAnalyst 5.0 (https://www.metaboanalyst.ca/).

2.11. Drug affinity responsive target stability (DARTS) analysis

Aliquoted cell lysates, each containing 100 μg of protein, were incubated for 10 min at 25 °C with different concentrations of SME (0, 0.1, 0.5, 1.0 mg/mL) in equivalent volumes. Proteinase K (Sigma-Aldrich, St. Louis, MO, USA) was added simultaneously to all samples at a proteinase K to substrate mass ratio of 1:2000 and incubated at 25 °C for 5 min. Reactions were stopped by adding sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) loading buffer.

2.12. Western blot

Cells were cultured in 6-well plates, washed with cold PBS, and lysed with inhibitor-containing solution 24 h post-treatment. Harvested cells were centrifuged to obtain clear lysates for protein quantification via BCA assay. Protein samples were electrophoresed, blotted onto polyvinylidene fluoride (PVDF) membranes, blocked and incubated with primary antibodies overnight. Secondary antibodies were applied before protein detection using an enhanced chemiluminescent (ECL) substrate and imaging on the ChemiDoc MP system (Bio-Rad, USA).

2.13. Molecular docking and molecular dynamics

The protein's crystal structure was sourced from the Protein Data Bank (https://www.rcsb.org/), and the 3D structure of potentially active molecules was gathered from the PubChem database. Detailed information about methods of molecular docking and the optimization of parameters is noted in Text S5 of Supplementary data.

2.14. Statistical analysis

All results were presented as mean ± standard deviation (SD). Data analysis and graphics were produced employing R toolkit and statistical analysis was conducted by using Student's t-test and one-way analysis of variance (ANOVA). P < 0.05 indicates that the difference was statistically significant.

3. Results

3.1. Chemical profiling of SME

To comprehensively identify the chemical constituents of the SME, an orthogonal liquid phase separation method was utilized in this study. SME was first fractionated into 10 fractions according to polarity using normal phase silica column chromatography, and these fractions were further separated using reverse-phase C₁₈ column chromatography. The component analysis revealed that compounds with lower polarity were more abundant than those with higher polarity, potentially due to the use of 75% alcohol extraction (Fig. 2A). Subsequently, ultra-performance liquid chromatography-tandem mass spectrometry (UPLC-MS/MS) was utilized for component identification of each fraction. The chemical profile of SME was further analyzed through literature review and tandem mass spectrometry interpretation. To demonstrate the fragment similarity between the identified compounds, the GNPS molecular networking algorithm was employed, and the results were visualized using Cytoscape software, with 10 major clusters illustrated within the SME (Fig. 2B). For example, the phenylpropanoids and polyketides class is represented by 31 compounds, specifically by salvianolic acid B; the oligosaccharides compounds are represented by 60 nodes with stachyose as a core substance; the lipids and lipid-like molecules cluster contains 62 nodes, with typical compounds such as tanshinone IIA. These components are known active ingredients in SME with significant pharmacological effects on cardiovascular protection, anti-inflammatory, antioxidant, and other aspects [32]. A total of 151 compounds were identified in this comprehensive chemical analysis of SME, with their polarity confirmed by the fractionation patterns (Table S1). Fig. S1 enumerates the ion fragment matching information for selected representative compounds in SME (salvianolic acid C, salvianolic acid B, rosmarinic acid). To the best of our knowledge, this represents the most extensive characterization of the SME chemical composition reported to date in a single study.

Fig. 2.

The components of Salvia miltiorrhiza extract (SME) exhibiting varied polarities were characterized by ultra high-performance liquid chromatography-tandem mass spectrometry (UHPLC-MS/MS) and subsequently clustered by global natural products social molecular networking (GNPS). (A) Total ion chromatography of different components of SME, with the different colors representing components with varying polarities. (B) The molecular network of SME. The edges between the nodes represent the relatedness of the compounds. Each node represents an individual ion, with nodes being connected by an edge when the cosine score is ≥ 0.7, and the thickness of each edge reflects the cosine score, which indicates the closeness of the relationship between the connected compounds.

3.2. Validation of the in vitro and in vivo efficacy of SME against ischemic stroke

To evaluate the efficacy of SME against IS, this study employed both in vitro and in vivo models. The in vitro experiments utilized an OGD/R model in PC12 cells, while the in vivo experiments employed a rat permanent middle cerebral artery occlusion (pMCAO) model, assessing SME's preventive and therapeutic effects on IS using multiple indicators. In the in vitro experiments, PC12 cells were pretreated with SME for 24 h, followed by 2 h of OGD modeling and 24 h of reoxygenation (Fig. 3A). The CCK-8 assay was then conducted to assess the proliferative effect of SME on OGD-induced PC12 cells. Compared to the control group, the cell survival rate of the OGD/R group decreased significantly (P < 0.001), while SME exhibited dose-dependent protective effects, significantly enhancing cell viability compared to the OGD/R group at 3 and 6 μg/mL (P < 0.001). Using the dual staining technique with calcein-AM and PI to differentiate live and dead cells, we observed a 55% cell mortality rate in the OGD model group. However, treatment with 6 μg/mL of SME notably reduced the cell mortality rate to 20% (Fig. 3B). Fluorescence microscopy analysis using the Fluo-4 calcium indicator revealed that OGD/R stimulation markedly increased the fluorescence intensity in PC12 cells, indicating a disruption of calcium homeostasis. SME pretreatment, however, effectively eliminated the Fluo-4 fluorescence signal, suggesting that SME helps maintain calcium balance in the nerve cells (Fig. 3C). Furthermore, ischemia and hypoxia impact mitochondrial function. TEM examination of the mitochondrial ultrastructure showed that cells under OGD/R conditions exhibited structural damage, such as swelling and blurred membrane boundaries. In contrast, SME-treated cells displayed significant protective effects, maintaining a relatively intact mitochondrial membrane structure. These findings suggest that SME exerts mitochondrial protective effects against hypoxic-ischemic injury, thereby providing neuroprotection (Fig. 3D).

Fig. 3.

The neuroprotective effects of Salvia miltiorrhiza extract (SME) were confirmed through in vitro and in vivo experiments using oxygen-glucose deprivation (OGD)-induced PC12 cells and middle cerebral artery occlusion (MCAO) rat model. (A) Cell viability of PC12 cells was protected by SME after oxygen-glucose deprivation/recovery (OGD/R) modeling. (B) Live/dead cell staining demonstrated that pretreatment with the extract significantly reduced OGD-induced cellular damage, as evidenced by the preservation of a higher proportion of live cells compared to controls. ^∗∗∗P < 0.0001 vs. Control group; ^###P < 0.001 vs. OGD group (one-way analysis of variance (ANOVA)). (C) Representative fluorescent microscopy images depicting intracellular calcium (Ca²⁺) concentrations. The cells were stained with Fluo-4 acetoxymethyl ester (Flou-4 AM) to visualize Ca²⁺ dynamics. The cells were stained with the Ca²⁺-sensitive fluorescent dye Fluo-4 AM to enable visualization of Ca²⁺ dynamics. (D) Ultrastructure of mitochondria visualized by transmission electron microscopy. (E) Schematic of the animal experimental design. (F) 2,3,5-Triphenyltetrazolium chloride staining (TTC) staining was performed to evaluate the cerebral infarct volume after 24 h of reperfusion (pale areas represent infarction tissues, and red areas represent normal tissues). (G) The effect of SME on behavioral scoring in MCAO rats. ^∗∗P < 0.01 vs. MCAO group by ANOVA with Student's t-test. (H) Results of Nissl Staining.. (I) Hematoxylin-eosin (H&E) staining of rat brain sections. (J) Terminal deoxynucleotidyl transferase-mediated dUTP nick end labeling (TUNEL) fluorescence in conjunction with neuronal nuclei (NeuN) immunostaining, where TUNEL staining (green) indicates DNA fragmentation, and the neuronal marker NeuN is shown in red.

In the in vivo experiments, rats were pretreated with SME (50 mg/kg/day) for 7 days before undergoing pMCAO surgery. The selection of in vivo dosage ensured that systemic exposure levels match the effective concentration of 5 μg/mL observed in PC12 cells (Fig. 3E). Twenty-four hours after surgery, TTC staining of brain sections revealed a 25% reduction in the cerebral infarct area in the SME-treated group compared to the MCAO group (P < 0.01) (Fig. 3F). Neurological deficit scores indicated significantly improved outcomes in the SME treatment group, highlighting better neurological prognosis compared to the MCAO group (Fig. 3G). Nissl staining of brain tissue sections showed enhanced preservation of neuronal structure in the SME-treated group (Fig. 3H). Histological evaluation with H&E staining (Fig. 3I) showed that SME pretreatment reduced ischemic injury, attenuated tissue necrosis, and maintained cellular integrity, further confirming the neuroprotective properties of SME. Co-localization images of TUNEL fluorescence combined with NeuN immunostaining results (Fig. 3J) showed overlapping green and red fluorescence in the MCAO group, indicating neuronal apoptosis. In contrast, SME pretreatment reduced ischemic injury, attenuated tissue necrosis, maintained cellular integrity, and decreased neuronal apoptosis in the rat brain. These comprehensive in vitro and in vivo experiments provide robust evidence for the neuroprotective effects of SME against IS injury, validating its potential therapeutic efficacy for the treatment of IS.

3.3. VS and target prediction in SME

VS is a computational approach that leverages in silico simulations to rapidly and cost-effectively predict the binding interactions between a large number of compounds and biological targets, thereby accelerating the drug target discovery process. In this study, VS techniques were employed to obtain comprehensive target information for the bioactive compounds in SME, considering their structural and bioactive characteristics. Based on the known chemical profile of SME, VS was used to obtain comprehensive target information, including SwissTargetPrediction and similarity ensemble approach (SEA) (Fig. 4A). These tools identify potential targets based on the structural similarity of SME compounds to known bioactive molecules. After filtering, 126 SME compounds with PubChem CID numbers were subjected to target prediction using online platforms. This approach yielded 310 candidate target proteins with a probability and maximum Tanimoto coefficient both greater than 0.6, indicating high confidence in the predicted associations. Multidimensional analytical methods, including dimensionality reduction, signature-based representation, clustering, and predictive modeling, efficiently manage large volumes of chemical and biological data. The Chemical Checker platform, developed for drug development, offers a comprehensive data integration framework, categorizing small molecule information into five levels: Chemical (basic chemical properties), Targets (interaction with protein receptors), Network (focus on biological pathways and their perturbations), Cell (phenotypic outcomes in cell-based assays), and Clinical (clinical outcomes and patient treatments). To enhance its capability, Chemical Checker integrates several deep learning techniques, including latent semantic indexing (LSI) for handling categorical data, principal component analysis (PCA) for reducing the dimensionality of continuous data, and network embedding methods to map molecular signatures into vector spaces. These techniques allow Chemical Checker to effectively analyze and visualize bioactivity data, facilitating drug discovery by leveraging the principle of similarity where compounds with similar properties or therapeutic effects often share biological behaviors and mechanisms of action. This approach extends the concept of chemical similarity to include both local and global properties of molecules, supporting the platform's multidimensional algorithms. By applying the Chemical Checker's deep learning-based multidimensional similarity analysis to the SME compounds, the study focused on evaluating the influence of these components and their associated targets on various biological characteristics, such as gene expression profiles, protein levels, and responses in different cellular or physiological states (Fig. 4B). For instance, the analysis highlighted the structural and physicochemical similarities between salvianolic acid C and salvianolic acid B, two key compounds in SME, and their common targets following association with other relevant molecules (indicated by black dots). Salvianolic acid C exhibits greater congruence with salvianolic acid B in structure keys and physicochemical parameters, and one associated target has been found. Through the integration of the target information obtained from the structural similarity-based screening and the multidimensional biological similarity analysis, a total of 281 potential target proteins were identified for the 126 SME compounds. Additionally, the study cross-referenced the identified potential targets with the DisGeNET database, using “ischemic stroke” as a keyword, to pinpoint the overlapping targets that are relevant to the treatment of this disease. This comprehensive approach led to the identification of 29 common targets that are likely to be involved in the mechanisms of action of SME against IS (Fig. 4C).

Fig. 4.

Virtual screening was employed to identify the target groups of Salvia miltiorrhiza extract (SME). (A) A workflow diagram illustrates the virtual screening process. (B) The 2D projection of two characteristic monomers in SME extracted from Chemical Checker, highlighting the distribution and relationships between individual compounds within the extract. (C) A venn diagram depicts the results of the target identification, showing the overlap between the potential targets of SME and the therapeutic targets associated with ischemic stroke (IS). SM: Salvia miltiorrhiza.

3.4. Enrichment analysis and molecular docking of VS targets

To perform a comparative analysis of the targets identified based on the structural and biological similarity of SME components, the study undertook Gene Ontology (GO) enrichment analysis on the two sets of targets. This analysis encompassed biological pathway (BP), cellular component (CC), and molecular function (MF). The 63 targets identified through the structural similarity of SME components were primarily associated with the regulation of inflammatory response, membrane rafts, and nuclear receptor activity in BP, CC, and MF categories, respectively (Fig. S2). Conversely, the 74 targets identified through the biological similarity of SME components were primarily linked to the response to xenobiotic stimulus, membrane raft, and heme binding (Fig. S2). Comparison of the targets identified by the two methods revealed that neither method could fully capture all the IS-related target information. To address this, the study performed enrichment analysis on the 29 overlapped targets derived from the intersection of structurally and biologically similar targets with IS-related targets. The enrichment analysis indicated that SME may influence the progression of IS by regulating inflammatory response, extracellular matrix, and serine hydrolase activity (Fig. 5A). Additionally, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis was conducted on the 29 potential target proteins. The chord diagram illustrated the interactions between target proteins and important pathways, such as the enrichment of arachidonate 12-lipoxygenase (ALOX12), arachidonate 5-lipoxygenase (ALOX5), arachidonate 15-lipoxygenase (ALOX15), APP, and prostaglandin-endoperoxide synthase 2 (PTGS2) in the serotoninergic synapse function pathway, as well as the enrichment of PTGS2, ALOX5, ALOX15, and ALOX12 in the arachidonic acid metabolism (Fig. 5B).

Fig. 5.

Enrichment analysis and molecular docking were employed to further validate the targets identified through virtual screening for the treatment of ischemic stroke (IS) using Salvia miltiorrhiza extract (SME). (A) The results of Gene Ontology (GO) functional enrichment analysis for the 29 potential targets, categorized into biological processes (BP), cellular components (CC), and molecular functions (MF). (B) The chord diagram depicts the results of the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis for the potential targets, highlighting the top 6 enriched pathways and the interactions between the genes and these pathways. Each gene represented as an arc on the circle, and their connections (chords) indicating the number of interactions with KEGG pathways. (C) Distribution of target proteins identified by different screening methods. A total of 108 potential target proteins were recognized through in silico screening and subjected to the protein-protein interaction analysis, including structural similarity screening (63 proteins, yellow dots) and multidimensional biological property similarity screening (74 proteins, blue dots). The 29 overlapped targets selected by both methods are indicated by the green dots. (D) The protein-protein interaction (PPI) network constructed based on the 29 targets of the treatment of IS with SME. (E) The heat map illustrates the disparity between the actual docking score of the small molecule and the docking score of the native substrate.

Subsequently, to investigate the roles and interactions of these target proteins, protein-protein interaction (PPI) analysis was conducted. The structural similarity screening, multidimensional biological property similarity screening, and overlapping targets identified by both methods were color-coded in yellow, blue, and green, respectively. Most of these targets reside at the network center with high connectivity, indicating their diverse and significant biological functions (Fig. 5C). Further PPI analysis of the 29 target proteins showed that APP, STAT3, PTGS2, matrix metalloproteinase 9 (MMP9), and matrix metalloproteinase 2 (MMP2), with degrees of 10, 7, 6, 6, and 6, respectively, were the most central and highly connected proteins, indicating their critical biological functions (Fig. 5D). To further evaluate the binding interactions between the SME compounds and the 29 target proteins, batch molecular docking was performed following format conversion, hydrogenation, and torsion key checking. Proteins predicted by Alphafold2 were docked using a full docking method (Table S2), after which SME components were placed into the same docking pockets to assess their competitive interactions with the proteins' natural substrates. The heatmap displayed the docking energies between the compound-protein interactions (Fig. S3), revealing that acetylcholinesterase (ACHE), nitric oxide synthase 2 (NOS2), PARP1, and tankyrase (TNKS) had generally strong binding interactions with the SME components, with average binding energies of less than −8.2 kcal/mol. Among the 151 identified compounds in SME, 126 were assigned CID numbers. The docking scores for 22 proteins with natural substrates were calculated and used as a benchmark to adjust the docking scores for other ligands. The difference between the docking scores of small molecules and native substrates highlight the variations in binding capabilities between exogenous substances and natural substrates (Table S3). These differences, once processed, are illustrated in Fig. 5E as a heatmap, comparing the binding affinities between the protein, natural substrates, and small molecules in SME. Notably, the SME compound apigenin exhibited stronger binding to APP compared to its natural substrate, indicating potential competitive binding and modulation of protein function.

3.5. Identification of direct interact target proteins for SME using the PISA approach

To efficiently screen for proteins that bind to the bioactive compounds in SME, the study employed a simplified CETSA method, PISA. The PISA procedure involves incubating cell lysates with SME, followed by heating treatment, and subsequent processing, including tandem mass tag (TMT) labeling, multiplex, fractionation, and analysis. Soluble proteins were first extracted from cells and incubated with SME at 1 mg/mL. The samples were then subjected to a temperature gradient ranging from 46 to 56 °C. After centrifugation to remove insoluble material, the remaining soluble proteins were digested into peptides, which are then labeled and analyzed (Fig. 6A). Following raw MS data, 2,411 proteins were identified, of which 1,468 (60.9%) contained more than two unique peptides, indicating high data quality (Fig. 6B). To identify potential SME target proteins, stringent criteria, including fold change (FC) thresholds (≤0.667 or ≥1.5) and P-value cutoffs (<0.05), were applied when compared with the SME-treated group to the control group. Ultimately, 100 differential proteins were obtained, 86 of which exhibited significantly increased thermal stability post-treatment (Fig. 6C). Notably, proteins such as Fev (FC = 2.24, P = 0.004), Utp3 (FC = 2.48, P = 0.03), and Calr (FC = 1.93, P = 9.38 × 10⁻⁷) demonstrated significantly increased thermal stability. Importantly, the previously identified target APP from the VS was also found to have increased thermal stability (FC = 1.545, P = 0.04) (Fig. 6D). Functional categorization of the identified target proteins revealed significant enrichment of proteins related to the autophagy pathway, as confirmed by KEGG pathway analysis (Fig. 6E). Furthermore, GO analysis highlighted the enrichment of postsynaptic membrane proteins among the identified targets (Fig. S4). These findings suggest that SME could positively affect neuronal survival and function by regulating neurotransmitter receptor expression and impacting mitochondrial bioenergetics, consistent with TEM analysis results. Electron microscopy images revealed that, following SME treatment, mitochondrial damage was partially repaired, and the resultant morphology more closely aligned with normal cellular appearance when compared to the OGD/R group, indicating potential therapeutic effects (Fig. 3D).

Fig. 6.

Quantitative proteomics employing tandem mass tag (TMT) labeling reveals the proteins that directly bind to Salvia miltiorrhiza extract (SME). (A) The workflow diagram illustrated the proteome integral solubility alteration (PISA) approach used in this study. (B) The identification process resulted in 2,411 proteins, each corresponding to a specific number of unique peptides. (C) The heatmap displays the expression changes of 101 differentially abundant proteins identified through the screening process. Blue indicates a decrease in protein levels, while red signifies an increase. Con: Control group; SME: SME (1 mg/mL). (D) The thermal stability distribution plot illustrates the profile of all proteins treated with SME. Red dots indicate proteins with higher thermal stability, whereas blue dots represent those with lower thermal stability (|FC| ≥ 1.5, P < 0.05). (E) The Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis results for the PISA screened proteins.

3.6. Identification of target proteins for SME via multi-dose metabolomics

Targets non-covalently bound to small molecules exhibit weak or transient interactions, making the identification challenging [33,34]. However, these interactions can be inferred through phenotypic changes [35]. In such cases, metabolomics offers direct and real-time insights by identifying the drug-induced metabolic changes, thereby elucidating the pharmacological mechanisms of action [36]. In the present study, the metabolomic principal component analysis (PCA) (Fig. S5) results showed significant separation between the OGD, control, as well as the SME treatment groups, indicating SME's substantial impact on the PC12 cell metabolism (Fig. 7A). The median effective dose (ED₅₀) for specific drug responses can be determined by screening metabolites that vary with SME dosage. For example, opposite trends were observed in the levels of pyroglutamic acid, l-glutamic acid, and γ-aminobutyric acid (GABA), with estimated ED₅₀ values of 1.97, 0.43, and 2.78 μg/mL, respectively (Fig. 7B). These findings suggest that SME effectively inhibits 5-oxoprolinase (OPLAH) enzyme activity at lower concentrations (0.43 μg/mL), leading to the accumulation of pyroglutamic acid and decreased production of l-glutamic acid. However, GABA production may be maintained due to increased GAD67) activity or expression at higher SME concentrations. Moreover, metabolites involved in reversible reactions mediated by GLUL and spermidine synthase (SMS), such as l-glutamine, l-glutamic acid, spermidine, and spermine, all exhibited similar EC₅₀ values, confirming the reversible nature of these metabolic processes.

Fig. 7.

Identification of dose-responsive metabolic enzymes through multi-dose metabolomics analysis. (A) The workflow diagram outlines the key steps involved in the multi-dose metabolomics approach, including sample preparation, metabolite detection and quantification, and data analysis. (B) The graph shows the concentration-related alterations in metabolite levels, highlighting key metabolic reactions and associated enzymes. Within the metabolic reactions, red and blue dots indicate increased or decreased metabolite concentrations, respectively, in response to the varying doses of the Salvia miltiorrhiza extract (SME) treatment. OPLAH: 5-oxoprolinase; GAD67: glutamate decarboxylase 67; GLUL: monia ligase; SMS: spermidine synthase; MS-DIAL: mass spectrometry data independent analysis for lipidomics.

3.7. Bioinformatics analysis of the dose-dependent metabolic enzymes modulated by SME

Through pre-normalization and dose-response trend metabolites plotting, 749 and 953 dose-altered ions were detected in positive and negative ion mode, respectively, with 82 ions were further identified as metabolites. The 82 dose-altered metabolites were further analyzed using the MetaboAnalyst platform, identifying associations with 78 metabolic enzyme targets. Among these, 16 enzymes exhibited dose-dependent changes in both upstream and downstream metabolites, while the remaining 62 enzymes showed dose-dependent changes in only one type of metabolite (Fig. 8A). Fig. 8B illustrates the enrichment analysis of 82 metabolic enzymes identified as subject to dose-dependent changes upon SME treatment, highlighting the top eight KEGG metabolic pathways. The chord diagram color-codes metabolic pathways and links them to specific enzymes via ribbon-like connecting lines, emphasizing the intricate network of interactions and significant pathways in which metabolic enzymes participate. For instance, the nucleotide metabolism pathway included 18 enzymes including cytidine deaminase (CDA), ctidine triphosphate synthase 1 (CTPS1), and deoxycytidine kinase (DCK), while the alanine, aspartate, and glutamate metabolism pathway comprised 13 enzymes such as GLUL, GAD67, and glutaminase (GLS), revealing the interplay and functional relationships between metabolites and enzymes in biological pathways.

Fig. 8.

Integrative bioinformatics analysis of dose-dependent metabolic enzymes identified through the multi-dose metabolomics study. (A) The number of the targeted metabolic enzymes. (B) The chord diagram showcases the top eight Kyoto Encyclopedia of Genes and Genomes (KEGG) metabolic pathways and their affiliated enzymes, ordered by their statistical significance (log P values). (C) Protein-protein interaction (PPI) analysis was performed on the 78 dose-sensitive metabolic enzymes using the Metascape tool to reveal the interactive dynamics within metabolic pathways. (D) The MCODE algorithm was employed to identify densely interconnected zones within the PPI network, aiding in the detection of functional protein modules that are critical for the dose-dependent metabolic regulation by the Salvia miltiorrhiza extract (SME).

The PPI network of the 78 metabolic enzymes, depicted in Fig. 8C, reveals intricate interactions among these enzymes, with nodes symbolizing enzymes and edges denoting interactions; larger nodes represent enzymes with a greater number of direct connections. These highly connected enzymes are more likely to serve as crucial targets for SME action, while the metabolic pathways they participate in are more likely to be the primary pathways regulated by SME. To identify densely connected regions or clusters within the network and recognize functionally relevant clusters of metabolic enzymes, the MCODE algorithm was applied for further analysis of the PPI network. The analysis highlighted a green cluster of 9 highly interconnected enzymes, with GLUL as the central node.

3.8. Verification of the interaction between SME and target proteins

The integration and selection of suitable targets for validation, following the identification of various targets by the three approaches, constitute a critical step in the VIP method. GO analysis revealed significant roles in the amino acid metabolic process and nucleoside phosphate metabolic process in BP, highlighting the essential involvement of amino acid transformation and nucleotide metabolism in the neuroprotective effects mediated by SME. Furthermore, the regulation of oxidative stress responses and metal ion binding underscores the adaptive mechanisms of neuronal cells in coping with oxidative stress and ion imbalance, common in the context of IS. This emphasizes the pharmacological effects of SME in mitigating cellular damage. MF analysis underscored the crucial importance of protein synthesis and repair in mediating the pharmacological action of SME (Fig. 9A). KEGG pathway enrichment analysis revealed several metabolic pathways closely associated with stroke pathology, including pyrimidine metabolism, glutathione metabolism, Biosynthesis of cofactors, and the hypoxia-inducible factor 1 (HIF-1) signaling pathway, among others (Fig. 9B). These findings suggest that these pathways may significantly contribute to the potential protective effects of SME against ischemic brain injury. Specifically, activation of the pyrimidine metabolism and glutathione metabolism pathways could affect intracellular energy metabolism and redox balance, while the regulation of cofactor biosynthesis helps maintain mitochondrial function and energy production, and the HIF-1 signaling pathway is directly involved in the cellular adaptive response to hypoxic environments. Proteins such as albumin (ALB), STAT3, estrogen receptor 1 (ESR1), GLUL, APP, GAD67, and PARP1 exhibited higher degrees in the PPI network, indicating a higher involvement in SME's regulation of IS (Fig. 9C).

Fig. 9.

The targets screened from the natural product virtual screening-interaction-phenotype (NP-VIP) methods were comprehensively analyzed and experimental verification of the interaction between Salvia miltiorrhiza extract (SME) and the target proteins. (A) Gene Ontology (GO) functional enrichment analysis results for potential targets of SME in treating ischemic stroke (IS), identified by three screening methods. (B) Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment chart shows the pathway analysis results for potential targets of SME in treating IS, identified via the above three screening processes. (C) The diagram depicts the 189 potential targets for SME in treating IS, as identified by the NP-VIP approach, along with the results from the associated protein-protein interaction (PPI) network analysis. (D) Drug affinity responsive target stability (DARTS) was utilized to validate the interactions between SME and the target proteins (poly [ADP-ribose] polymerase 1 (PARP1), amyloid precursor protein (APP), signal transducer and activator of transcription 3 (STAT3), glutamate decarboxylase 67 (GAD67), and glutamate-ammonia ligase (GLUL)). ^∗∗∗∗P < 0.0001 vs. Control group; ^#P < 0.05, ^##P < 0.01, ^###P < 0.001, ^####P < 0.0001 vs. pronase E group (n = 3, one-way analysis of variance (ANOVA)). (E) Root mean square deviation (RMSD) analysis of simulated structures: apigenin-APP, salvianolic acid C-STAT3, oleanolic acid-GAD67, salvianolic acid A-GLUL, salvianolic acid C-PARP1.

Using bioinformatics and systematic analysis, a collection of proteins associated with SME treatment for IS was selected. Five proteins, including STAT3, PARP1, APP, GAD67, and GLUL, identified through various target screening methods and central to the PPI network, were selected for experimental validation. Subsequently, DARTS technique combined with Western blot analysis was employed to validate and quantify the direct binding of SME components to potential target proteins. The principle of DARTS is that protein binding with small molecules induces a structural change towards a lower energy, more stable configuration, thereby enhancing resistance to proteolysis. Experimental results revealed that, with actin used as a negative control (not identified as a potential target of SME in NP-VIP screening), its expression levels remained unchanged before and after treatment. Conversely, at SME doses of 0.1, 0.5, and 1.0 mg/mL, the expression levels of GLUL, GAD67, STAT3, APP, and PARP1 increased dose-dependently (Fig. 9D), indicating that SME enhanced the ability of these proteins to resist pronase digestion, thereby confirming the specific binding interactions of SME's active components with these proteins.

Molecular docking experiments were conducted to investigate the interactions between 126 small molecules identified in SME and five target proteins (APP, GLUL, STAT3, GAD67, and PARP1) (Table S4). To validate the molecular docking results and assess protein-ligand binding stability, molecular dynamics simulations were performed on the protein-small molecule complexes with the lowest docking scores. Fig. 9E presents the root-mean-square deviation (RMSD) results of these complexes, illustrating the binding stability of APP with apigenin, GAD67 with oleanolic acid, GLUL with salvianolic acid A, PARP1 with salvianolic acid C, and STAT3 with oleanolic acid. RMSD fluctuations indicate conformational changes in the proteins during the simulation, with minor amplitude suggesting limited changes that eventually plateau, reflecting the structural stability of the protein-ligand complexes. Molecular dynamics simulations further confirm the binding capability between the proteins and small molecules, with this binding being relatively stable within a dynamic equilibrium. RMSD analysis indicates that despite some conformational adjustments during the simulation, the overall structural stability of the proteins was maintained, particularly after complex formation with small molecules. This stability suggests the presence of specific interaction forces between the proteins and small molecules.

4. Discussion

NPs are pivotal sources for drug discovery, featuring compounds like artemisinin, ginkgolides, and paclitaxel that are already integrated into clinical therapies due to their demonstrated efficacy [21,37,38]. While contemporary research often emphasizes individual compound drugs, NP research fundamentally centers on the utilization of NP extracts. Certain NP extracts, such as Compound Danshen Dripping Pills [39] and Chinese Medicine Compound Tongxinluo [40], have shown considerable pharmacological activities, particularly in addressing conditions such as myocardial ischemia and ST-segment elevation myocardial infarction (STEMI)^,. However, researchers face significant challenges due to the complexity of NP constituents, variability in compound properties and target group. Traditional research methodologies, including isolation, purification, and activity validation, are time-consuming and may disrupt potential synergistic interactions within NPs. Therefore, the comprehensive examination of NPs and systematic exploration of their active constituents and pharmacological targets present pressing challenges in the field of natural medicine research. To address these challenges, our study introduces the NP-VIP strategy, an innovative approach integrating VS, chemical proteomics, and metabolomics. Unlike traditional methods, NP-VIP combines in silico simulation, PISA, and functional multi-dose metabolomics into a cohesive workflow that enhances target discovery accuracy and efficiency. This study aims to screen and identify targets of NPs from a holistic perspective, with the goal of establishing a research paradigm suited to the multi-component, multi-target composition of NPs. Aiming to establish a research paradigm suited to the multi-component, multi-target composition of NPs, this study focused on SME, developing the NP-VIP strategy to examine its therapeutic effects on IS.

Target screening for molecular-protein interactions can be accomplished through methods including in silico simulation prediction, chemical proteomics, and omics analysis. However, in-silico models might not fully encapsulate the complexities of the cellular environment and its dynamic changes, while chemical proteomics may overlook downstream metabolic alterations. To overcome these limitations and address the complex composition of natural extracts, the NP-VIP strategy integrates diverse informational sources to minimize the incidence of false positives and false negatives. In the proposed workflow, VS prioritizes target identification based on the chemical structure and biological activity of natural extracts; PISA experimentally identifies targets directly bound to small molecules; while multi-dose metabolomics provided real-time SME induced phenotypic information, facilitating the exploration of downstream metabolic changes. Applying this strategy to SME, multiple potential targets were successfully identified, and further validated (PARP1, STAT3, APP, GAD67, and GLUL) by using DARTS. Each of the five targets in SME exhibits distinct pharmacological effects. Research indicates that PARP1 plays a critical role in cell death under cerebral hypoxia-ischemia, particularly in programmed necrosis [41,42]. Inhibiting PARP1 through pharmacological suppression or gene deletion effectively reduces neural damage caused by cerebral hypoxia, highlighting PARP1 as a promising neuroprotective target [43]. SME reduced PARP1 expression after OGD/R conditions in PC12 cells (Fig. S6). In IS, STAT3 enhances cell survival pathways and upregulates anti-apoptotic genes, minimizing neuronal apoptosis and improving resilience [44]. Western blot analysis revealed decreased STAT3 expression in PC12 cells under OGD/R conditions, which is counteracted by SME (Fig. S6). APP has a pivotal physiological role in the mammalian brain. APP functions in routine physiological processes and exhibits neuroprotective effects under metabolic stress [45]. The role is particularly pronounced during both acute and chronic hypoxic-ischemic conditions, where APP expression is upregulated to counteract hypoxia and nutrient deprivation in brain cells [46]. Research shows that APP's neuroprotective function is enhanced by downregulating miR-130a-3p, protecting neurons from OGD-induced damage [47]. Additionally, SME modulates APP expression in PC12 cells under OGD/R conditions (Fig. S6). GAD67 which catalyzes the conversion of glutamate to GABA [48], is crucial for maintaining the balance of excitatory and inhibitory signals in the brain. GABA production regulates neuronal excitability and alleviating cellular excitatory damage during IS pathogenesis [49]. In MCAO rats, GAD67 content was reduced; interestingly, rats pretreated with SME showed a significant increase in GAD67 content (Fig. S6). This implies that SME could contribute to neuronal protection from excitatory injury by increasing GAD67 levels and promoting the conversion of glutamate to GABA. GLUL is a key enzyme in the brain responsible for converting the neurotransmitter glutamate to glutamine [50]. In IS, extracellular glutamate due to inadequate blood supply, exacerbating excitotoxicity. SME treatment modulates GLUL levels in PC12 cells under OGD/R conditions, potentially alleviating excitotoxic damage (Fig. S6).

This study demonstrated the efficacy of the NP-VIP methodology in identifying and validating multiple bioactive targets within SME for treating IS. However, NP-VIP has limitations. Initially, metabolomics and PISA treated SME as a holistic entity, which may obscure specific component-target relationships. To refine this, molecular docking was employed to match PISA-identified proteins with SME small molecules, prioritizing those with the lowest docking scores. Molecular dynamics simulations validated these interactions, enhancing our ability to link specific compounds to biological targets. Integrating advanced technologies like surface plasmon resonance (SPR) and ultrafiltration liquid chromatography-mass spectrometry (UF-LC-MS) could create a more comprehensive interaction map.

5. Conclusion

Overall, this study utilized a comprehensive NP-VIP methodology, highlighting the efficacy of SME in treating IS. This innovative approach integrated VS, PISA, and multi-dose metabolomics to successfully identify and validate multiple bioactive targets within SME, significantly enhancing the precision and reducing the error rates typically associated with traditional single-method approaches. The findings not only confirm the neuroprotective properties of SME but also establish a robust framework for NP research, promising enhanced drug discovery and therapeutic development. The NP-VIP method has shown promising applications in NP targets. Based on its mechanisms of action, this method can also be applied to the target identification of other monomeric compounds, making it a powerful tool for drug development. Furthermore, this approach can also be adopted in other disciplines, such as toxicology and environmental exposome target group analysis.

CRediT authorship contribution statement

Rui Xu: Writing – original draft, Conceptualization. Hengyuan Yu: Data curation, Writing – review & editing. Yichen Wang: Software. Boyu Li: Validation. Yong Chen: Supervision. Xuesong Liu: Funding acquisition. Tengfei Xu: Funding acquisition, Project administration, Supervision.

Declaration of competing interest

The authors declare that there are no conflicts of interest.

Appendix A. Supplementary data

The following are the Supplementary data to this article: