Metabolic profiling of Lantana camara L. using UPLC-MS/MS and revealing its inflammation-related targets using network pharmacology-based and molecular docking analyses

Lantana camara L. is widely used in folk medicine for alleviation of inflammatory disorders, but studies that proved this folk use and that revealed the molecular mechanism of action in inflammation mitigation are not enough. Therefore, this study aimed to identify L. camara phytoconstituents using UPLC-MS/MS and explain their multi-level mechanism of action in inflammation alleviation using network pharmacology analysis together with molecular docking and in vitro testing. Fifty-seven phytoconstituents were identified in L. camara extract, from which the top hit compounds related to inflammation were ferulic acid, catechin gallate, myricetin and iso-ferulic acid. Whereas the most enriched inflammation related genes were PRKCA, RELA, IL2, MAPK 14 and FOS. Furthermore, the most enriched inflammation-related pathways were PI3K-Akt and MAPK signaling pathways. Molecular docking revealed that catechin gallate possessed the lowest binding energy against PRKCA, RELA and IL2, while myricetin had the most stabilized interaction against MAPK14 and FOS. In vitro cytotoxicity and anti-inflammatory testing indicated that L. camara extract is safer than piroxicam and has a strong anti-inflammatory activity comparable to it. This study is a first step in proving the folk uses of L. camara in palliating inflammatory ailments and institutes the groundwork for future clinical studies.

www.nature.com/scientificreports/ were unveiled via constructing a constituent-target (C-T) network ( Supplementary Fig. S2). Out of the identified 57 compounds from UPLC-MS/MS analysis, only 39 compounds were potential candidates for inflammationrelated protein targets, and 35 inflammation-related target genes were eventually fished out based on screening results from STITCH public database. Regarding STITCH 5.0 database, "combined score" is the parameter utilized to evaluate the strength of interactions between the input compound and the genes. Compounds possessing high combined scores have accurate and strong interactions with their corresponding genes 13 . In this study, only compounds having interaction scores higher than 0.4 were retained 13 ( Table 2). The constructed C-T network ( Supplementary Fig. S2) comprised 74 nodes (39 constituents and 35 target genes) and 479 edges with an average of 3.043 targets for each constituent, indicating the multi-target properties of the L. camara phytoconstituents. As deduced from Fig. 2a, the highest percentages of interactions were demonstrated by ferulic acid, followed by catechin gallate, then myricetin and iso-ferulic acid. Inspection of the targeted genes (Fig. 2b, Table 2) indicated that the genes PRKCA, RELA, IL2, MAPK14 and FOS were the most enriched ones possessing the highest combined scores and interaction percentages with the constituents in the C-T network, proposing their possible key role in suppressing inflammation. In addition, protein-protein interactions were examined using STRING database then visualized through P-P network analysis. From this network, strong correlations between the identified potential anti-inflammatory target proteins were spotted suggesting that they probably regulate the functions of each other ( Supplementary Fig. S3).
Potential metabolic pathways of inflammation were explored by forwarding the target genes to the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis [21][22][23] where annotation was restricted to Homo sapiens. As shown in Supplementary Fig. S4 and Table 3, the target genes were involved in 47 inflammationrelated pathways (having P-values < 0.05). The most enriched pathways were observed to be PI3K-Akt signaling pathway exhibiting the largest number of gene count followed by MAPK signaling pathway, Rap1 signaling pathway, Ras signaling pathway and Phospholipase D signaling pathway. The constructed networks were merged to generate the compound-target-pathway network (Fig. 3) which implied strong co-relations between the studied compounds and inflammation-related targets and pathways.
Gene ontology (GO) enrichment analysis for targets. Gene ontology (GO) enrichment analysis was carried out to the identified targets via importing to DAVID bioinformatics resources with limiting annotations to Homo sapiens, thus revealing the most enriched pathways and GO terms which have the highest log P value and gene counts. As depicted in Fig. 4a, the identified targets are associated with numerous biological processes, the most enriched ones are inflammatory response, response to cAMP, activation of MAPK activity and response to cytokine. The most significant molecular cellular components were plasma membrane, integral component of plasma membrane, cytosol and extracellular region. It was also concluded that the most enriched molecular functions were lysophosphatidic acid receptor activity, protein heterodimerization activity, enzyme binding and protein kinase activity. Nevertheless, functional annotations using DAVID bioinformatics resources revealed 1 BBID pathway named 3.T-cell receptor, and 30 BIOCARTA pathways such as: oxidative stress induced gene expression via Nrf2, Toll like receptor pathway, keratinocyte differentiation and BCR signaling pathway. Additionally, 52 KEGG pathways involving PI3K-Akt signaling pathway, Rap1 signaling pathway, Ras signaling pathway and proteoglycans in cancer were identified (Fig. 4b). All these recognized pathways possessed P-value less than or equal to 0.05, implying their striking association with inflammation.
Molecular docking studies of L. camara hit compounds in the active sites of the most enriched inflammation-related target genes. The    Validation of molecular docking protocol. Validation procedures for each docking software were attained using two methods. First is the redocking procedure which evaluates the accuracy of the docking poses and during which, the co-crystallized ligands were docked back into the receptor binding cavity. The re-docked complex was superimposed on to the reference co-crystallized complex and the RMSD value between the initial conformation and the re-docked one is calculated. A cut off value of 2 Å was set; therefore, complexes encompassing above this value were considered incorrect 24 . For each of the studied proteins: 4RA4, 1M49 and 6HWU, the re-docked complex was superimposed on to the reference co-crystallized complex to a great extent (Supplementary Fig. S5). Moreover, the RMSD value between the initial conformation and the re-docked one was calculated and all the three crystallographic structures displayed good values of 1.172, 0.386, 0.558 respectively (supplementary Table S2) indicating the efficiency of the docking protocol. Second is the utilization of enrichment calculations that are crucial for evaluating the quality of scoring and eliminating random or by chance selection of actives 25 . A validation set comprising active compounds for each of the investigated proteins was seeded into 1000 built-in Schrodinger ® decoys. Decoys are compounds that are similar in physical properties with respect to the reference ligand that might not bind effectively to a protein 26 . Validation parameters such as receiver operating characteristic (ROC), AUC-ROC, BEDROC and enrichment factor (EF at 2%, 5% and 10%) were then estimated. From the ROC plots, the area under the curve (AUC) computed the probability of how highly a randomly selected active is ranked compared to a randomly chosen decoy. The ideal range of AUC is 0-1, a value near ≤ 0.5 indicates that the software randomly selects true actives and false actives, where a value close to 1 highlights greater possibility to identify true actives before false ones 25 . As depicted in (supplementary Table S2), it was observed that all proteins scored promising AUC-ROC values.
Comparing EF values revealed that the investigated proteins were able to extract actives from a seeded random set, when the top 2, 5 and 10% of the total set were considered, noting that the maximum attainable enrichment factors are 50, 20, and 10 for EF(2%), EF(5%), and EF(10%), respectively 27 . Using BEDROC as a criterion to assess early recognition of actives from decoys at different tuning parameter value α 28 , all the proteins recorded the high scores at all α values. To conclude, all the enrichment values obtained for each docking procedure suggested that GLIDE software was able to filter the enriched database efficiently. ADME filtration of L. camara top hit compounds. QikProp module was utilized to calculate the ADME characteristics of the L. camara hit compounds, in order to assess their drug-likeness. L. camara hit compounds were regarded as drug candidates as they conformed to Lipinski's rule of 5 29 , and Jorgensen's rule of 3 30 (Supplementary Table S3).  www.nature.com/scientificreports/ In vitro cytotoxicity and anti-inflammatory activity of L. camara extract. In order to assess the safety of the tested extract, the cell cytotoxicity 50 (CC50), which is the drug concentration required for reducing the cell viability by 50%, was determined for the extract and the standard anti-inflammatory drug (piroxicam) using MTT test. The tested extract showed higher CC50 value (382.5 µg/mL) than that of piroxicam (100 µg/ mL) indicating that the extract is safer than piroxicam (Fig. 5a). Afterwards, anti-inflammatory activities of the extract compared to piroxicam were estimated using lipopolysaccharides (LPS)-stimulated WBC cells (Fig. 5b). Both extract and piroxicam showed comparable effective anti-inflammatory concentrations (EAICs) (48.08 µg/ mL and 42.50 µg/mL, respectively), suggesting the promising activity of the extract as anti-inflammatory candidate. To determine the mechanism of anti-inflammatory activity at the genetic level, the gene expression of four pro-inflammatory markers (TNF-α, IL-1β, INF-γ, IL-6) was measured by real time polymerase chain reaction (PCR) in normal WBCs and lipopolysaccharide (LPS)-treated WBCs (Fig. 5c). Regarding TNF-α, lipopolysaccharide (LPS) upregulated the expression of this gene by 2.1-folds. Upon treatment of the WBCs with the tested extract this upregulation was abolished to 0.89-fold which was comparable to that exerted by piroxicam (0.73fold). Meanwhile, LPS upregulated the expression of IL-1β by 5.23-folds which was attenuated by the extract to 1.69-folds. This value was in close agreement to that obtained by piroxicam (1.53-folds). Interestingly, the upregulation of the gene expression of INF-γ and IL-6 was significantly decreased by the tested extract and piroxicam to a similar level (error bars were shown in Fig. 5 and p values for all experiments were less than 0.05). It can be concluded that L. camara extract can serve as potential anti-inflammatory natural product assigning to its noticeable inhibition of the upregulated TNF-α, IL-1β, INF-γ, IL-6 expression levels. These results were compatible with that obtained from network pharmacology and molecular docking analyses that revealed the multi-target and multi-pathways nature of the tested extract regarding anti-inflammatory activity.
Fatty acids. Three peaks (53, 54 and 55) represented unsaturated fatty acids. The mass fragmentation of unsaturated fatty acids is represented by two characteristic fragments due to loss of water and CO 2 61 along with their characteristic fragment at 54 m/z that result from double-bond transfer and α-cleavage 62,63 . Compounds 53, 54 and 55 were identified as myristoleic acid, linolenic acid and linoleic acid methyl ester, respectively 64 . Meanwhile, two peaks (56 and 57) represented saturated fatty acids, the mass fragmentation of saturated fatty acids is represented by two characteristic fragments result from loss of water and CO 2 65 along with the fragment of Mclafferty rearrangement that was detected at 59 Da 66 . Compounds 56 and 57 were tentatively identified as arachidic acid and behenic acid, respectively 64 .
PubMed literature review was implemented to validate the role of the hit compounds identified from network pharmacology analysis in alleviation of inflammation. As can be observed in Supplementary Table S4, ferulic acid precluded methotrexate-induced hepatotoxicity via inducing Nrf2/HO-1 signaling and PPARγ, as well as abolishing oxidative stress and inflammation 67 . Catechin gallate diminished the levels of cyclo-oxygenase and lipoxygenase inflammatory mediators thus alleviated UV radiation-induced erythema 68 . Another previous work confirmed the protective effect of myricetin against liver fibrosis in a diet-induced non-alcoholic steatohepatitis rat model through inhibiting the TREM-1-TLR2/4-MyD88 signaling molecules in macrophages 69 . Meanwhile, isoferulic acid attenuated the production of PI3K/Akt-dependent NF-κB activity, thus, could serve as a potential drug for treating neuritis and other neuronal ailments 70 71 , multiple sclerosis 72 and inhibited LPS-induced acute lung injury and inflammatory response 73 . Moreover, regulation of nuclear-cytoplasmic shuttling of RELA aids in attenuation of inflammation 74 . It was also proved that loss of epithelial RELA results in deregulated intestinal proliferative/apoptotic homeostasis and susceptibility to inflammation 75 . Furthermore, expression and induction of a pancreatitis-associated protein (PAP1) depended on RelA/p65 levels, suggesting its multidimensional roles in treating cerulein pancreatitis 76 . Also, allergic inflammation was claimed to be influenced by nuclear factor κB1/RelA expression in human lung epithelial cells 77 . Meanwhile, interleukin-2 (IL-2) is the canonical T-cell growth factor that stimulates clonal expansion of T cells following antigen stimulation, hence plays a critical role in orchestrating optimal immune and inflammatory responses 78 . Therefore, targeting such protein contributes to alleviate inflammatory bowel diseases as well as suppressing inflammation synergized by respiratory viral infections 79 . Additionally, P38α/ MAPK14 is intracellular signaling regulator involved in biosynthesis of inflammatory mediator cytokines as TNF-α, IL-1, IL-6, and IL-1β, which induced the production of inflammatory proteins such as iNOS, NF-kB, and COX-2 80 . Also, regulation of MAPK14 expression prevented aggravation of myocarditis 81 , multiple sclerosis 82 and inflammatory bowel diseases 83 . Other recent work confirmed the vital role of MAPK14 in relieving the inhibitory control by autophagy on inflammation in response to a stress signal 84 .
Molecular docking analysis revealed the strong binding of the top hit compounds on the active sites of the most enriched genes. For example; the 2D and 3D interaction diagrams of catechin gallate in the active site of protein kinase C alpha type (PBD ID 4RA4) (Fig. 6a) showed that the strong binding-as expressed by its XP G score-was attributed to the formation of two hydrogen bonds between 3 and 4′ hydroxyl groups and Glu418, two hydrogen bonds between 3 and 4″ hydroxyl groups and Asp424, In addition to hydrophobic interactions with Phe350, Ala480, Met417, Tyr419, Ala366, Val420, Met470, Leu345 and Val353. Moreover, polar interactions with Asn468 and Thr401 and charged negative interactions with Asp481, Glu418 and Asp424 were observed 85 (Supplementary Table S5).
However, binding of catechin gallate with interleukin-2 (PBD ID 1M49) showed two hydrogen bonds between the hydroxyl groups at C-5 and C-7 and Glu68 and Lys43, respectively, together with four hydrogen bonds between 3″, 4″ and 5″ hydroxyl groups and Arg38. Also, a pi-pi stacking interaction between the aromatic ring www.nature.com/scientificreports/ A of the flavone moiety and Phe42 and hydrophobic interactions with Pro65, Phe44, Phe42 and Met39 were unveiled. Two charged positive interactions with Lys43 and Arg38, one charged negative interaction with Glu 68, beside one polar interaction with Thr41 were also deduced 86 (Fig. 6c, Supplementary Table S5).
On the other hand, the interaction pattern of myricetin with mitogen-activated protein kinase 14 (PBD ID 6HWU) included the formation of four hydrogen bonds between the following pairs: 3′ hydroxyl group and Ala51; 4 carbonyl group, 5 hydroxyl group and Met109; and 7 hydroxyl group and Val30; in addition to hydrophobic interactions with Val30, Leu171, Met109, Leu108, Val105, Ala51, Val52, Leu104, Leu75, Phe169 and Val38. There were also polar interactions with Thr106 and Hie107 and a charged positive interaction with Lys53 87 (Fig. 7a, Supplementary Table S5). Furthermore, the binding mode of myricetin with proto-oncogene c-Fos (PBD ID 1FOS) revealed the presence of four hydrogen bonds between 4 carbonyl group and ArgE 158, C-5 hydroxyl group and SerE 154, C-3′, C-4′ hydroxyl groups and SerF 278. Myricetin also engaged in a pi-pi stacking interaction through its flavone aromatic ring A moiety and ArgE 155. Charged positive interactions with the backbone amino acid residues ArgE 155, ArgE 158, LysF 282, ArgF 279, polar interactions with SerE 154 and SerF 278, and a hydrophobic interaction with AlaE 151 were also considered 88 (Fig. 7b, Supplementary Table S5).
The drug-likeness of compounds can be predicted by applying Lipinski's rule of 5. As claimed by Lipinski's rule of 5, a compound of known pharmacological activity is regarded active (having good absorption and/or permeation) if it possesses less than 10 hydrogen-bond acceptors (acptHB), less than 5 hydrogen-bond donors (donorHB), a molecular weight (mol_MW) lower than 500 Da and a calculated QPlogPo/wvalue less than five 29 . Only compounds conformed to minimally three of the above characteristics were considered active. www.nature.com/scientificreports/ In addition, the oral bioavailability (OB) of the top hit phytoconstituents was evaluated using the descriptor Jorgensen's rule of 3 30 . Only compounds demonstrating OB ≥ 30% were considered active. The hit L. camara constituents obey the above criteria and hence, were considered as drug candidates (Supplementary Table S3).
The aerial parts of L. camara were collected from Antoniades Garden, Alexandria, Egypt with permission from the Agriculture Research Center, Giza, Egypt at "9 Cairo University Road, Giza District, Giza Governorate". The plant collection was accomplished in accordance with the national guidelines. The identity of the plant was confirmed by Dr. Therese Labib, specialist of plant identification in El Orman Garden, Cairo, Egypt. A voucher specimen (No. LC-250) was deposited at the herbarium of the Department of Pharmacognosy, Faculty of Pharmacy, Alexandria University.
Preparation of L. camara extract. Air-dried powdered leaves of L. camara (500 g) were extracted by sonication in 1 L of 95% ethanol in an ultrasonic bath apparatus 28 kHz/1100 W (3 L Alpha Plus, Japan) for 30 min at 35 °C. The obtained extract was filtered, and the procedure was repeated twice. The obtained extracts were combined and evaporated to dryness under reduced pressure using rotary evaporator at 45 °C to obtain 200 g dry residue. A portion of the dry residue of L. camara extract was dissolved in HPLC-grade methanol to obtain a sample solution of concentration 1 mg/mL. This sample solution was filtered using a membrane disc filter (0.2 μm), then degassed by sonication. After that, a full loop injection volume (10 μL) of this solution was injected into the chromatographic column.

Analysis of L. camara extract using UPLC-MS/MS technique. The chromatographic analysis was
accomplished using an UPLC XEVO TQD triple quadruple instrument Waters Corporation, Milford, MA01757 USA equipped with a Waters Acquity QSM pump, a LC-2040 autosampler, degasser in addition to Waters Acquity CM detector. The dimensions of Waters Acquity UPLC BEH C18 column was 50 mm (length), 2.1 mm (internal diameter) and 1.7 μm (particle size). The operation of the column was at a flow rate of 0.2 mL/ min and the system was thermostated at 30 °C. The mobile phase that used for analyses consisted of two phases; phase A: ultrapure water + 0.1% formic acid, and phase B: methanol + 0.1% formic acid. Elution was gradient one and its program was as following: 0. The mass spectrometric analysis and metabolites annotation were carried out according to the method described by Darwish et al. 89 as shown in the Supplementary data. Network pharmacology-based analysis. The 2D structures of the identified compounds yielded from UPLC-MS/MS analysis were converted to SMILES format using Schrodinger software (LLC, New York, NY, 2015), then furtherly subjected to network pharmacology-based analysis. The identification of the target genes linked to the selected constituents was performed using STITCH database (http:// stitch. embl. de/, ver.5.0) with the 'Homo sapiens' species settings. UniProt (http:// www. unipr ot. org/) 90,91 was utilized for retrieving gene information including name, gene ID and accession number. To retrieve information about functional annotation and the signaling pathways, bioprocesses, cellular components and molecular functions that were highly associated with inflammation target proteins, DAVID ver. 6.8 (https:// david. ncifc rf. gov/) and the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (http:// www. genome. jp/ kegg/ pathw ay. html) were employed. An adjusted p-value < 0.05 was set as a cut-off value for enriching the significance of contributing pathways to inflammation.
Three types of networks: constituent-target gene, gene-pathway, and constituent-gene-pathway networks were constructed and visualized by Cytoscape 3.8.2 (http:// www. cytos cape. org/) in order to visualize the interactions between compounds, target proteins and inflammation-related pathways. In the graphical network, each constituent, gene protein and pathway were described by node, and the interactions were encoded by edges. The network parameters were calculated using the network analyzer plug-in in Cytoscape where the weight of nodes in each constructed network was evaluated using Cytoscape combined score of interactions. Protein-protein interaction network (PPI network) was constructed using STRING database (https:// string-db. org/).

Molecular docking studies.
Molecular docking studies were performed using Glide module integrated in Schrodinger ® software. The Protein Data Bank (PDB) was utilized to retrieve the crystal structures of the most enriched target proteins recognized from network pharmacology analysis, named; protein kinase C alpha type (4RA4), transcription factor p65 (3QXY), interleukin-2 (1M49), mitogen-activated protein kinase 14 (6HWU) and proto-oncogene c-Fos (1FOS). These crystal structures were saved as pdb files for further preparation using the PrepWiz module. Location of the binding site for the docking experiments was determined using the receptor grid generation module. Some protein models have no co-crystallized ligands (ex: 3QXY and 1FOS), so the ligand was set as the centroid of specified selected residues retrieved from literature. Hence, the size of www.nature.com/scientificreports/ the receptor grid predetermined as (20 × 20 × 20 Å 3 ) was adjusted to accommodate ligands with size ≤ 20 Å to exclude large molecules with overestimated docking scores. For other models with co-crystallized ligands (ex: 4RA4, 1M49 and 6HWU) the boxes enclosing the centroids of co-crystallized ligands were set as the grids. 3D-structures of the top hit compounds recognized from network pharmacology analysis (ferulic acid, catechin gallate, myricetin and isoferulic acid) were imported as SDF files to be prepared using Ligprep module generating molecules with correct chiralities, ionization states, tautomers, stereochemistries and ring conformations. The generated compounds from the LigPrep file were flexibily docked using extra precision (XP) docking, and 2D and 3D ligand-target interactions were visualized in maestro interface. The docking protocol was validated using two methods: (i) redocking of the co-crystallized ligands into the binding sites of their corresponding proteins then the resulting complexes were superimposed on to the reference co-crystallized complexes and the root mean square deviation (RMSD) was calculated. This was done to ensure exact binding of the inhibitor to the active site where less deviation compared to the actual co-crystallized complex is more favorable. This method was exclusively performed for proteins bearing co-crystallized ligands (Protein kinase C alpha type, 4RA4; Interleukin-2, 1M49; and Mitogen-activated protein kinase 14, 6HWU). (ii) Enrichment calculations: for each of the investigated proteins, a validation set composed of known active ligands compiled from literature was constructed (Supplementary Table S6). The validation set compounds were seeded in 1000 Schrodinger ® built-in decoys then docked against the active site of target protein using XP mode. Protein-ligand complexes were validated using GLIDE enrichment calculator using numerous validation parameters such as receiver operating characteristic (ROC), AUC-ROC, BEDROC and enrichment factor (EF at 2%, 5% and 10%). These calculations aimed to enrich the docking procedure and to discriminate active compounds from non-active ones thus, avoiding false positive hits production. ADME and drug-likeness of top hit compounds. The top hit constituents related to inflammation were assessed for drug-likeness by calculating in-silico absorption, distribution, metabolism, and excretion (ADME) criteria and adopting Lipinski's rule of five 29 , by the aid of Qikprop module (Schrodinger suite 2017A). Only compounds with predicted oral bioavailability ≥ 30 and satisfying at least three criteria from Lipinski's rule of five were considered active.
In vitro cytotoxicity and anti-inflammatory activity testing. It was carried out according to the method described by Darwish et al. 92 as shown in the Supplementary data.

Conclusion
In this study, the phytoconstituents of L. camara extract were identified using UPLC-MS/MS analysis, then they were subjected to network pharmacology analysis that declared ferulic acid, catechin gallate, myricetin and isoferulic acid as the endogenous metabolites mostly associated to inflammation, and PRKCA, RELA, IL2, MAPK 14 and FOS as the main inflammation-related genes. The identified target genes were involved in 47 inflammation-related pathways, where the most enriched ones were PI3K-Akt signaling and MAPK signaling pathways. Molecular docking of top hit compounds on the active sites of the most enriched genes revealed that catechin gallate possessed the lowest binding energy against PRKCA, RELA and IL2, while myricetin exhibited the most stable interaction against MAPK14 and FOS. The extract was then forwarded to in vitro cytotoxicity and anti-inflammatory testing indicating comparable results to those of piroxicam. This study provides a profound explanation of the mechanism of the proposed anti-inflammatory activity of L. camara and recommends this plant as a source of potential anti-inflammatory agents. Further in vivo and clinical studies are recommended to affirm our outcomes.

Data availability
All data generated or analyzed during this study are included in this article (and its supplementary information files).