Rapid Profiling of Metabolites Combined with Network Pharmacology to Explore the Potential Mechanism of Sanguisorba officinalis L. against Thrombocytopenia

Sanguisorba officinalis L. (SO), a well-known herbal medicine, has been proven to show effect against thrombocytopenia. However, metabolites of SO in vivo are still unclear, and the underlying mechanism of SO against thrombocytopenia from the aspect of metabolites have not been well elucidated. In this study, an improved analytical method combined with UHPLC-QTOF MS and a molecular network was developed for the rapid characterization of metabolites in vivo based on fragmentation patterns. Then, network pharmacology (NP) was used to elucidate the potential mechanism of SO against thrombocytopenia. As a result, a total of 1678 exogenous metabolites were detected in urine, feces, plasma, and bone marrow, in which 104 metabolites were tentatively characterized. These characterized metabolites that originated from plasma, urine, and feces were then imported to the NP analysis. The results showed that the metabolites from plasma, urine, and feces could be responsible for the pharmacological activity against thrombocytopenia by regulating the PI3K-Akt, MAPK, JAK-STAT, VEGF, chemokine, actin cytoskeleton, HIF-1, and pluripotency of stem cells. This study provides a rapid method for metabolite characterization and a new perspective of underlying mechanism study from the aspect of active metabolites in vivo.


Introduction
Herbal medicine (HM) plays a vital role in modern biological science and demonstrates definite curative effects; however, its mechanism of action requires further study. The metabolic analysis of HM in vivo is a systematic task aimed at the qualitative and quantitative analysis of metabolites in the living body [1], which is significantly helpful for understanding potential mechanisms.
As far as we know, metabolites are structurally similar compounds of prototype ingredients that share the same product ions or neutral loss ions in a collision-induced dissociation. Huan et al. developed a core structure-based search algorithm based on a hypothetical neutral loss for spectral similarity analysis and an unknown metabolite annotation [25]. Hsu et al. developed a strategy based on multiple types of correlated ion information for metabolite identification [26]. To facilitate data analysis, Bandeira et al. developed a molecular network (MN) which was a visualization platform for the analysis of untargeted MS/MS data and it was also a web-based MS data ecosystem [27]. Based on structurally similar compounds or analogs that produce similar fragmentation patterns, MN automatically compares and clusters the spectra of each precursor ion, and the resulting networks are plotted. Compounds from different samples can be marked using different colors, and the relative content of each metabolite can be shown using a pie chart. The compounds are clustered into different families in MN, facilitating the simultaneous characterization of abundant and trace compounds. This method has been successfully employed to analyze antibiotics [28], peptides [29], marine organisms [30,31], and components of HMs [32]. However, the metabolic analysis of HMs using the multifunctional MN has not been reported.
Network pharmacology (NP) was first proposed by Hopkins [33]. Owing to the viewpoint of "multi-constituents, multi-targets, and synergistic effects" of HM, NP has been widely used to prediction the molecular mechanisms of HMs [34]. Many NP studies have focused on the reported compounds from various databases to explore pharmacological mechanisms. However, few studies have been based on the metabolites in vivo, which have been proven to be the potential active constituents [35]. In NP, the network of predicted associations among drugs, targets, pathways, and disease for a particular group of biological samples by experimental results or database information can be summarized and visualized. For example, the connected nodes represent diverse biological entities which including drugs, genes, pathways, and disease. The edges represent the predicted functional associations, such as coexpression evidence among genes. Relying on this clustered network with high-throughput technologies, the potential relationship between drugs and diseases can be predicted.
In this study, a rapid metabolite profiling method was established for the characterization and visualization of metabolites of Sanguisorba officinalis L. (SO) in vivo, which are commonly used as a HM against thrombocytopenia (TCP) [36]. Profiled metabolites in plasma, urine, and feces were subsequently imported to perform an NP analysis to explain the molecular mechanism of SO against TCP. As a result, 1678 metabolites (repeated metabolites were not included) were detected, of which 913, 601, 364, and 160 metabolites were detected in the urine, feces, plasma, and bone marrow, respectively. The ingredients of SO in mice were mainly excreted through urine and feces. A total of 104 metabolites with a cosine value >0.7 were tentatively characterized. The NP analysis revealed that metabolites from plasma, urine, and feces could be responsible for the pharmacological activity of thrombocytopenia by regulating the following pathways in cancer: PI3K-Akt, MAPK, JAK-STAT, VEGF, chemokine, actin cytoskeleton, HIF-1, and pluripotency of stem cells.

Chemicals and Materials
Acetonitrile and formic acid were of LC-MS grade (Thermo Fisher, MA, USA). A water purification system (Milli-Q, Billerica, MA, USA) was used to generate ultrapure water. SO was obtained from Chengdu Jinhong Co., Ltd. (Chengdu, China) and confirmed using its morphological characteristics. All other reagents were of analytical grade.

Preparation of SO Extract
SO (100 g) was soaked in 800 mL of ultrapure water for 30 min and extracted with water at 100 • C thrice. The extracts were combined and concentrated using a rotary evaporator for intragastric administration.

Animal Experiments
Here, 12 mature Sprague Dawley (SD) rats (6 males and 6 females) weighing 30 ± 3 g (mean ± standard deviation) were provided by Southwest Medical University (Luzhou, Sichuan, China) and used. The rats were maintained within a temperature range of 20-25 • C and in a 12 h light/dark cycle. Prior to the drug administration, the rats were acclimated to the environment for seven days with free access to food and water. The rats were randomly divided into two groups (n = 6 per group, comprising three males and three females), i.e., the normal control group and the SO group (250 mg/kg, treatment). Thereafter, the rats were placed on a fast with free drinking water during the test. Plasma, urine, and fecal samples were collected 0.5, 1, and 2 h after oral administration. Immediately after the biosample collection, the rats were sacrificed, and the thighbones were stripped out to collect their bone marrows. All procedures were reviewed and approved by the Biomedical Ethical Committee of Southwest Medical University (no. 20211123-014) and were consistent with animal care guidelines.

Sample Preparation
Plasma samples (200 µL) were obtained by centrifugation at 3500 rpm (4 • C) for 15 min, and 600 µL of acetonitrile was added to 200 µL of plasma. The mixture was vortexed for 3 min, let stand for 10 min, and centrifuged at 10,000 rpm for 15 min at 4 • C. The supernatant was transferred to a sample vial for detection.
The urine samples were immediately stored at −70 • C until analysis. They were thawed on ice prior to injection and similarly processed as the plasma samples.
Freeze-dried feces (80 mg) were mixed with 800 µL of 80% methanol/water. The mixture was ultrasonicated in an ice bath for 30 min, and centrifuged at 10,000 rpm for 15 min at 4 • C. The supernatant (200 µL) was collected for further analyses.
The bone marrow was flushed using a 1.5 mL saline rinse via a syringe into a centrifuge tube, and the collected sample was added to methanol (600 µL). The sample was ultrasonicated in an ice bath for 30 min and centrifuged at 10,000 rpm (4 • C) for 15 min. The supernatant was collected for further analysis.
An X500R QTOF MS system was equipped with a Turbo V TM source and a Twin Sprayer probe for electrospray ionization (ESI) (Sciex, MA, USA). The source parameters for positive and negative polarity were as follows: temperature, 500 • C; full scan mass range, m/z 100-1500; ion source gases 1 and 2, 50 psi; curtain gas, 35 psi; and collisionally activated dissociation (CAD) gas, 7 psi.
The automated MS/MS data acquisition was performed in the information-dependent acquisition (IDA) mode. The specific parameters with dynamic background subtraction were as follows: maximum candidate ions, 10; intensity threshold exceedances, 200 cps; and former candidate ions were excluded for 5 s after 1 occurrence.

Metabolite Data Acquisition and Preprocessing
The SCIEX OS 1.4 software (Sciex, MA, USA) was used to acquire and output the raw data files, which were converted into mzXML format using the MSConvert software. The transformed files were uploaded to the MN platform (http://gnps.ucsd.edu (accessed on 1 April 2022)) using the FileZilla software. Moreover, the minimum pair cosine score was set to 0.7, the minimum cluster size was set to 2, and the minimum matched fragment ions were set to 6 before MN was created. Furthermore, endogenous metabolites were eliminated by subtracting the data of normal control group. These parameters could be manually modified before generating the MN. The default values were used for the remaining parameters, and the MN was visualized using Cytoscape 3.7.1 (NIGMS, MD, USA).
The SO and TCP-related targets were input into the STRING database (http://stringdb.org/ (accessed on 1 June 2022)) to predict the possible protein-protein interaction (PPI) information. Then, the PPI data with degree values ≥28, ≥47, and ≥58 of plasma, urine, and feces were reserved. By using Metascape (http://metascape.org/gp/index.html# /main/step1 (accessed on 1 June 2022)), the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway and gene ontology (GO) enrichment evaluations were performed to clarify the pathways that were involved in metabolites acting on TCP.

Results and Discussion
Here, a rapid metabolite analysis method was developed, which included four steps: (1) the acquisition of MS/MS data by UHPLC-QTOF MS, (2) construction of a MN, (3) identification of metabolites originating from different tissues by library searching and MNguided annotation, (4) visualization of origins and relative quantification of metabolites in different tissues using pie charts in different colors, and (5) summarization of metabolic pathways. Thereafter, these metabolites from different tissues were used for the NP analysis.

Optimization of LC-MS Conditions
To characterize metabolites with different properties, generality should be adequately considered at the beginning of method development. A long-standing C 18 column and the general gradient used in this study started from 5% and ended at 100% acetonitrile within 20 min. This maintained the polar component in the column as much as possible and eluted the nonpolar component from the column during the running period since baseline resolution was not necessary for the MS/MS data acquisition and processing. ESI, which is regarded as the most popular soft ionization source, was used in this method because it was significantly less affected by the matrix effect [37]. SO, blank biosamples (plasma, urine, and feces), and rat's plasma, urine, and feces after administration of SO were analyzed and compared. As shown in Figure S1, the constituent of SO is extremely complex, and the chromatograms between blank biosamples and biosamples after administration are quite different. The mass spectrometer was programed to perform a full scan within m/z 100-1500 in positive and negative ion detection modes. For the nontargeted MS data acquisition, the IDA mode was fully competent to record practically all the MS/MS spectra, although the IDA strategy, i.e., sequential window acquisition of all theoretical fragment ions (SWATH), emerged as an unbiased acquisition method. The high-resolution mass analyzer ensured that the qualitative analysis was accurate. The UHPLC/QTOF/MS platform combined with the UHPLC and MS conditions is a generally analytical strategy that facilitates the rapid analysis of various metabolites.

Construction of the Molecular Network of SO
In a MN, the connected nodes with a cosine value represent compounds with similar structures and fragmentation patterns. The closer the cosine value gets to one, the higher the structural similarity. In addition, the metabolites from different origins can be distinguished using different colors. As shown in Figure 1, green, yellow, blue, purple, and gray nodes represent their origins, including feces, urine, plasma, bone marrow, and the prototype ingredients of SO, respectively. Therefore, the absorption and distribution (plasma, digestive system, and urine) of these metabolites can be easily understood using different colors. Based on this established network, the absorption, distribution, metabolism, and excretion of SO was intuitively demonstrated. The yellow and green nodes originating from the urine and feces, respectively, were intuitively observed to be the dominant nodes and were statistically calculated as 41.9% and 35.5% of the 1678 nodes. A few metabolites were observed in the plasma (blue nodes) and bone marrow (purple nodes). Therefore, urinary metabolism is the main excretion pathway of SO, followed by fecal metabolism. Furthermore, the relative abundance of each metabolite from different origins was intuitively demonstrated using pie charts. For example, the relative content of M7, a metabolite of ellagic acids (EAs), as shown in Figure 1, was 40 and 60% in the urine and feces pathways, respectively. In addition, M7 and M9 exhibited a cosine value of 0.95, demonstrating a high structural similarity to each other. The substituents of M9 were similar to that of M7, except for the occurrence of the sulfonic acidification of M9. M7 was detected in urine and feces, while M9 was only detected in urine.

Characterization of Metabolites
Previous studies have provided MS-based strategies [35,38,39] for metabolite profiling, combining a series of MS libraries [40] based on nontargeted metabolite profiling data. However, their applications have been limited by the number of spectra stored in libraries. Based on the method of a previous study [32], the MN established in this study clustered the structurally similar compounds into a network, greatly reducing the difficulties of characterization. In a network, a node represents a compound, and spectrum-to-spectrum alignments represent the edges between nodes. The connected nodes indicate compounds with similar fragmentation patterns and chemical structures. The obtained MS/MS spectra were searched against GNPS spectral libraries, in which the matched and unmatched compounds were illustrated. Ingredients undergo various metabolic transformations, such as hydroxylation, acetylation, and glucuronidation. The metabolic derivatives are a series of structurally similar compounds that possess the same molecular skeleton, resulting in similar fragmentation patterns during collision-induced dissociation. These similar fragmentation patterns can be recognized, clustered, and visualized using the MN, simultaneously facilitating the characterization of abundant and trace metabolites. Here, 1678 metabolites of SO were detected, in which 104 metabolites were tentatively characterized and are detailed in Table S1. For all cases, once one compound was annotated in a cluster or matched by the libraries, the structure of the related nodes could be annotated from this node according to the similarity of fragment ions/neutral loss and accurate mass differences between nodes. Two metabolites with a cosine value greater than 0.7 were characterized.

Feces
Fifty prototype components and 16 metabolites of SO from the feces were characterized; the main components were flavonoids and saponins (Table S1). M17 was detected in the feces and identified as catechin by a library search. As shown in Figure 2 Thus, they were identified as procyanidin B2 (M27) and procyanidin B2 3'-O-gallate (M31) [42]. Other flavonoid metabolites such as M19, M20, M21, and M26 originated from the parent drug 4 ,7-dihydroxyflavone conjugated with methyl, hydroxymethyl, and glycosyl under the action of intestinal microflora and were excreted in the feces. Other metabolic pathways including methylation, decarboxylation, glycosylation, and glucuronidation, which are common in intestinal metabolism [13,38,39,43], were detected in saponin metabolites, such as M58, M60, M83, and M85. Owing to the MS/MS collision, the M83 presence of fragmentation ions at m/z 797.4654, 599.3913, and 441.3704 demonstrated a continuous neutral loss of rhamnose, glucose, and glucuronic acid. Furthermore, EAs were converted via lactone ring-opening, dihydroxylation, decarboxylation, and further glucuronidation into urolithin B-O-glucuronide (M8), which is the main metabolic pathway of EA derivatives in the gastrointestinal tract [44].

Urine
Here, 19 parent drugs and 15 metabolites of SO were profiled in urine samples; the main components were EAs and glycosides (Table S1). EA metabolites in urine such as M6, M10, and M11 are sulfated, methylated, and glucuronide forms of EA, respectively. Methyl EA and EA glucuronide are common metabolites from urine metabolism [45]. Similarly, the lactone rings of EAs were hydrolyzed into urolithin through urinary excretion and underwent decarboxylation and conjugation with methyl groups (M2 and M3) [44,46]. Thus, the same neutral loss ions M-80 Da, which is common in natural products and represents the elimination of SO 3, were present in both M4 and M5. Furthermore, EA prototype ingredients and metabolites were practically present in urine, demonstrating that the pathway through the urinary system may be the main metabolic pathway for EAs. The urinary glycoside metabolites were mostly O-glycosides generated from the dehydration of gallic acid and monosaccharides, followed by conjugation with methyl, sulfonic acid, and glucuronic acid groups.

Plasma
Four prototype ingredients and six metabolites of SO from the plasma were characterized. The collision-induced fragmentation pathway of M4 exhibited continuous neutral losses of the sulfonic and ethyl groups with specific fragments observed at m/z 197.0457 and 166.9989, respectively. M36 exhibited fragment ion peaks at m/z 307.0830 and 277.0724, indicating the continuous neutral loss of the acetoxy and carbinol groups, respectively.

Bone Marrow
There have been few studies on the metabolite analysis of SO in the bone marrow, although active ingredients isolated from SO, such as ziyuglycosides I and II have been demonstrated to improve hematopoiesis in vitro [47]. Here, four metabolites (M87, M95, M98, and M99) of SO from the bone marrow were characterized.

Analysis of Metabolic Pathways
The UHPLC combined MN allowed for the rapid elucidation of the possible metabolic pathways of SO in vivo. As shown in Table S1, the exogenous metabolites were tentatively characterized. As shown in Figures S2-S5, the metabolites were classified into four classes (EAs, flavonoids, glycosides, and saponins) based on different skeletons.

Network Pharmacology Analysis
Totally, 273, 433, and 645 targets were revealed corresponding to the plasma, urine, and feces metabolites, respectively. The target genes related to TCP were searched and 142, 269, and 382 common targets among metabolites from plasma, urine and feces and disease were identified, respectively (shown in Figure 3). Furthermore, the statistical results showed that 49%, 73%, and 81% common targets were different among the plasma, urine, and feces metabolites. This also confirms the multi-component and multi-target characteristics of HM. Figure 3. The Venn diagram of targets between thrombocytopenia and metabolites. The targets of thrombocytopenia were marked in blue, and the targets of metabolites from plasma, urine, and feces were marked in red, yellow, and brown, respectively.
Meanwhile, the PPI network was constructed by using common targets, which were used to construct the metabolites-target-pathway network by using the Cytoscape software (shown in Figure 4). In the PPI network, the node with a higher score may be more important in the central correlation. According to the screening principle of degree value being larger than or equal to twice their respective median, 26, 57 and 82 key potential targets of plasma, urine, and feces metabolites with disease were screened out. . Targets/pathways/disease network of SO metabolites and thrombocytopenia. The targets of metabolites were marked in purple, the pathways were marked in blue, metabolites were marked in green, diseases were marked in red, and SO was marked in orange.
The KEGG database was utilized to note related pathways. A total of 168 pathways were obtained from key potential targets; nine pathways were probably the vital pathways, which possessed the count of significantly enriched (p value < 0.0001), comprising pathways in cancer, i.e., PI3K-Akt, MAPK, JAK-STAT, VEGF, chemokine, actin cytoskeleton, HIF-1, and pluripotency of stem cells (shown in Figure 5).
These enriched nine vital pathways were mainly associated with the hemopoietic system. PI3K-Akt was confirmed as a critical signaling pathway, which inhibits the aggregation and activation of platelets after allogeneic hematopoietic stem cell transplantation [48]. MAPK plays a vital role in the megakaryopoiesis differentiation process [49]. There was a study that demonstrated targeting JAK-STAT could lead to limiting treatment-emergent side effects, such as anemia and TCP [50]. In addition, the expression of HIF1A proteins in hematopoietic cells have been reported to cause lethal hemophagocytic lymphohistiocytosislike phenotypes: severe anemia, TCP, splenomegaly, and multi-organ failure upon HIF1A induction [51]. An experimental study has demonstrated that VEGF-A was crucial for the hematopoietic stem and progenitor cell response during the early phase of acute TCP [52]. Hao et al. [53] showed that increased chemokine in plasma, such as CXC, was implicated in the pathogenesis of immune TCP. The relationship between regulation of actin cytoskeleton signaling pathway and TCP has also been summarized [54]. Therefore, as shown in Figure 6, the functions involved in the above signal pathways are inferred to be closely related to the promotion of MK differentiation, which requires further verification.

Conclusions
In this study, a rapid metabolite analysis strategy was developed for the comprehensive metabolite profiling of SO in vivo. Metabolites in plasma, urine, and feces were then used for NP analysis to deduce the potential mechanism. Ultimately, a total of 1678 metabolites were detected, of which 913, 601, 364, and 160 were present in the urine, feces, plasma, and bone marrow, respectively. In addition, 104 metabolites were tentatively identified, including 13 EAs, 20 flavonoids, 20 glycosides, 32 saponins, and 19 other compounds. The metabolites of SO in mice were mainly excreted through the urine and feces. The metabolic pathways of methylation, glycosylation, glucuronidation, sulfonation, lactone opening, gallic acid conjugates, dehydration, and oxidation were identified. Based on these characterized metabolites in vivo, the NP analysis reveals that the mechanism of SO against TCP is related to the following pathways in cancer: PI3K-Akt, MAPK, JAK-STAT, VEGF, chemokine, actin cytoskeleton, HIF-1, and pluripotency of stem cells.