2.1 Compositions of mulberry leaves

Mulberry leaves were sourced from a local traditional Chinese medicine market in Bo-Zhou, Anhui Province, China. The plant was identified by Prof. Dao-Quan Tang from the School of Pharmacy at Xuzhou Medical University and the specimen was deposited in the Laboratory of Pharmaceutical Analysis at Xuzhou Medical University. 2 kg well-dried mulberry leaves were extracted twice with 10 kg of 70% ethanol for 90 min, which was then concentrated at 60 ℃ under vacuum condition (-0.05Mpa) to generate a total of 135 g powder through spray-drying. The powder of mulberry leaves was detected by Agilent 1100 high-performance liquid chromatography (HPLC) system (Agilent Technologies Inc., California, USA), together with Waters Xevo G2XS QT of mass spectrometry (MS) system (Waters Corporation, Milford, MA, USA). A Waters BEH C18 column (2.1 mm × 50 mm, 1.7um) was used for the separation. The mobile phase consisted of 0.1% formic acid in water (solvent A) and acetonitrile (solvent B). The gradient program started with 95% A-5% B:0–1 min, followed by 0% A-100% B:1–8 min, 0% A-100% B:8–11 min, 95% A-5% B:11–12 min. The column temperature was at 40 °C. The sample injection volume was 2 μL and the flows rate was set at 0.3 mL/min. The analytes were determined in a positive ion mode. Mass spectrometer electrospray capillary voltage was 3.0 kV, sample cone 40 V, source temperature 100 °C, desolation temperature 400 °C, cone gas 50 L/h, and desolation gas 800 L/h. A total of 248 chemical compounds were identified. A complete form of all the compounds was available in Supplementary Table S1. The corresponding total ion chromatogram was present in Supplementary Figure S1.

2.2 Compound target of EEML

We first searched traditional Chinese medicine systems pharmacology database (TCMSP, http://tcmspw.com/tcmsp.php) by using chemical name, InChIKey, and CAS Number of all the compounds identified in the ethanol extract of mulberry leaves via LC-MS. Oral bioavailability (OB, > 30%) and drug likeness (DL, > 0.18) were used as selection criteria for active ingredients and the corresponding targets of each active compound were recorded. A total of 18 compounds, together with 96 target genes were collected. Due to the limitations of the TCMSP database, we then searched the PubChem database by using chemical names in order to obtain the two-dimensional structure of each compound, which were further searched in the online database PharmMapper http://lilab-ecust.cn/Pharmmapper/ in terms of diabetes-related targets. Top 30 predicted targets for each compound were kept for further analysis. In sum, a total of 58 compounds and 327 target genes were collected.

2.3 Targets of T2DM

Target genes related to T2DM were collected through searching GeneCards and OMIM database by using the keyword Type 2 Diabetes Mellitus. 9952 candidate genes were identified. All the gene names were matched to corresponding UniProt IDs, gene names, and protein names by using ID mapping tool in the protein database UniProt (http://www.uniprot.org/). In order to construct the network between genes related with T2DM and the compounds in the ethanol extract of mulberry leaves, we pooled all the target genes and used the online tool venny (https://bioinfogp.cnb.csic.es/tools/venny/) to generate the Venn diagram (Supplementary Figure S2), which revealed that a total of 37 targets were directly associated with T2DM and targeted by 63 compounds in the ethanol extract of mulberry leaves (Table 1). Among the 63 compounds, 14 were identified from TCMSP with confirmed functions in T2DM (Table 2) while other 49 were sourced from sourced from PharmMapper (Supplementary Table S2). According to the number of interactions for the 14 components, we identified the top 5 compounds that had the highest number of interactive targets (denoted as n in parentheses for each compound) that were responsible for T2DM, which were Kaempferol (n = 23), Licoricone (n = 7), Herbacetin (n = 6), Morin (n = 5), and Lobelanidine (n = 4).

Table 2 Specific information of 14 identified anti-diabetic components in the ethanol extract of mulberry leaves.

Mol ID1	Compound	PubChem CID2	Molecular Formula	No.3	Target4	OB%5	BBB6	DL7	Structure8
MOL000074	1,7-Bis(4-hydroxyphenyl)-hepta-4E,6E-dien-3-one	10613719	C19H18O3	3	ADRB2, PPARG, PRKACA	67.92	−0.42	0.24
MOL008226	14-Deoxyandrographolide	11624161	C20H30O4	1	PGR	56.3	−0.55	0.31
MOL002787	6-Feruloyl catalpol	44566578	C24H30O12	3	CA2, F2, PRSS1	31.38	−2.09	0.84
MOL008219	Andrograpanin	11666871	C20H30O3	1	PGR	56.3	−0.55	0.31
MOL001490	Bis(2-ethylhexyl) Phthalate	8343	C24H38O4	2	ADRB2, CHRM3	43.59	0.68	0.35
MOL002823	Herbacetin	5280544	C15H10O7	6	ACHE, AR, F2, GABRA1, PPARG, PRSS1	36.07	−0.65	0.27
MOL001781	Indigotin	10215	C16H10N2O2	1	PRKACA	38.2	0.02	0.26
MOL000422	Kaempferol	5280863	C15H10O6	23	ACHE, AKT1, ALOX5, AR, BCL2, CYP1B1, F2, F7, GABRA1, HMOX1, ICAM1, IKBKB, MMP1, NOS2, NOS3, PGR, PPARG, PRKACA, PRSS1, SELE, SLC6A2, TNF, XDH	41.88	−0.55	0.24
MOL004855	Licoricone	5319013	C22H22O6	7	AR, F10, F2, KCNH2, KDR, NOS2, PRSS1	63.58	−0.14	0.47
MOL012207	Lobelanidine	442646	C22H29NO2	4	ADRB2, CHRM3, F2, SLC6A3	60.53	0.31	0.32
MOL000737	Morin	5281670	C15H10O7	5	ALOX5, AR, GSR, PPARG, TOP1	46.23	−0.77	0.27
MOL000483	N-cis-Feruloyl Typamine	6440659	C18H19NO4	2	ADRB2, PDE3A	118.35	−0.27	0.26
MOL001539	Sanjoinenine	14729078	C29H35N3O4	1	F10	67.28	−0.24	0.79
MOL011931	Vitetrifolin E	11143042	C22H36O4	1	PGR	31.41	−0.04	0.3

Note: ¹Mol ID: Molecule ID used in the TCMSP database https://tcmspw.com/browse.php?qc=herbs. ²PubChem CID: Chemical compound ID used in the PubChem database https://pubchem.ncbi.nlm.nih.gov/. ³No.: Number of interactions with targets. ⁴Target: names of target proteins. ⁵OB: Oral bioavailability. ⁶BBB: Blood-brain barrier. ⁷DL: Drug-likeness. ⁸Structure: all the compound structures were sourced from PubChem database.

2.4 Concentration quantification of representative compounds

In order to confirm the presence of chemical components detected via LC-MS, we measure three representative compounds via ultra-performance liquid chromatography-tandem mass spectrometry (UPLC-MS/MS), which were kaempferol, licoricone ad morin. All experiments were carried out on a Thermo Fisher TSQ LC-MS/MS system (Thermo Fisher Scientific Co., San Jose, CA, USA) consisting of an Accela Autosampler, an Accela pump, and a Quantum Access triple quadrupole mass spectrometer. All system control, data acquisition, mass spectral data evaluation and data processing were performed using XCalibur software version 2.0.7 (Thermo Fischer, San Jose, CA, USA) and Thermo LCquan software version 2.5.6 (Thermo Fischer, San Jose, CA, USA).

The Agilent ZORBAX SB C18 column (2.1 mm × 150 mm, 3.5 μm) was used for separation. The mobile phase consisted of 0.1% formic acid in water (solvent A) and methanol (solvent B). Using a gradient program started with 60% A, a step to 90% A until 8 min, then with a isocratic elution of 90% A until 10 min, and re-equilibration from 10.01 min to 15 min with 60% A at a flow rate of 0.30 mL/min and column temperature at 40 °C. MS conditions were as follows: ESI in negative mode, vaporizer temperature at 280 °C, capillary temperature at 320 °C, sheath gas pressure at 40 psi, auxiliary gas pressure at 20 psi and spray voltage at 3000 V. Quantification was accomplished in multiple reaction monitoring (MRM) by monitoring the transition of m/z 301.2 → 151 for morin, 381.4 → 323 for licoricone and 285.2 → 241 for kaempferol. LC-MS/MS chromatogram with mixed surrogate standards for kaempferol, morin and licoricone was present in Supplementary Figure S3. This developed method was applied for the detection and quantification of the three compounds (Kaempferol, Morin, Licoricone) in the ethanol extract of mulberry leaves with calculated linear regression models (Supplementary Table S3).

2.5 Protein-protein interaction (PPI) network

The 37 identified T2DM targets were imported into the online platform STRING https://string-db.org/, and the organism was set to Homo sapiens (Supplementary Figure S4). High-confidence protein interactions with enrichment P-value<1e-16 were selected, which included 37 nodes and 138 edges with an average node degree of 7.46. A TSV file of PPI network was automatically generated via STRING, which included parameters such as nodes1, node2, and combined_score, etc. (Supplementary Table S4). The file was then input into CytoScape, an open-source software platform for visualizing complex networks https://cytoscape.org/, during which node1 column was set as Source Node, node2 column was set as Target Node, while the remaining columns were set as default parameters. Protein-protein interactions was present in CytoScape as an enrichment network (Fig. 2A). During the analysis, the NetworkAnalyzer plug-in was used to calculate the values of node degrees (Assenov et al., 2008), according to which, the higher the degree value was, the more important the node was in the network. Molecular Complex Detection (Mcode) Clustering ftp://ftp.mshri.on.ca/pub/BIND/Tools/MCODE of T2DM targets in the protein–protein interaction network was then performed with Mcode_Score set to greater than or equal to 6, from which the most highly interconnected regions in a network were identified. According to the analysis, only one sub-network was identified, which was presented in Fig. 2B.

Close

Fig. 2

Schematic illustration of the enrichment of protein–protein interactions between target proteins (P-value < 1e−16). Network nodes represent proteins, and the edges represent protein–protein associations. (A) PPI enrichment analysis through online STRING platform version 11.0. (B) Highly inter-connected region through Mcode analysis.

Export to PPT

2.6 Compound-target network construction

The interactional relationships between 63 active components in the ethanol extract of mulberry leaves and 37 corresponding targets were imported into CytoScape. Then, the active components and target genes were drawn into circle by using the function of Degree Circle Layout within the plug-in Layout. Moreover, components collected from TCMSP were coloured in blue while those predicted by PharmMapper were coloured in red. In addition, all target genes were in green colour. NetworkAnalyzer function in the plug-in Tools was used to generate the degree values of each node in the entire network graph, and then the Generate Style from Statistics function was used to draw the component-target network graph based on the degree values. That is, the sizes (diameters) of the nodes in the network were determined by degree values. The greater the degree value is, the larger the node diameter is, and the more important the node is in the whole network.

2.7 Gene ontology (GO) enrichment analysis

In order to clarify the functions of target compounds in the ethanol extract of mulberry leaves and their role in signal transduction, we used the Database for Annotation, Visualization and Integrated Discovery (DAVID) to carry out GO functional enrichment of all the identified target proteins. According to the analysis, molecular function (MF), cellular component (CC) and biological process (BP) of target proteins were respectively described. GO analysis also returned a P-value for each GO term with a small P-value indicating that most different genes were enriched. According to the hypergeometric distribution relationship, a small P-value (<0.05) indicates enrichment of differential gene in the GO analysis. In addition, the smaller the P-value, the more significant the differential enrichment is. Since small P-value is not well present, we performed the -log10 (P-value) conversion. Thus, the larger the -log10 (P-value), the more significant the differential enrichment is, which is convenient for intuitive understanding of the meaning of the data during visualization.

2.8 KEGG pathway enrichment analysis

Target genes were imported into the Ensemble database (http://asia.ensembl.org/index.html). BioMart function in Ensemble was then used to convert the gene IDs into Entrez Gene IDs, which were entered into KOBAS (http://kobas.cbi.pku.edu.cn/kobas3/?T=1). Gene-List Enrichment function in KOBAS was used, Species was set to Homo Sapiens, Input Type was set to Entrez Gene ID, Input was set to Intersection Genes, while Database was selected as KEGG Pathway (K) in PATHWAY. To further verify that whether the biological processes related to the target protein are associated with the type 2 diabetes mellitus, a total of 30 pathways filtered with P-value < 0.05 were selected for further KEGG pathway analysis while the corresponding -log10(P-value) of each pathway was visualized. After the analysis of KEGG pathway enrichment, both bar plot and bubble plot were generated. It was noteworthy that -log10(P-value) had the same meaning as stated in Section 2.7.

2.9 Target pathway analysis

KEGG Mapper Tool was used to construct the pathway mapping in terms of the ethanol extract of mulberry leaves for the treatment of type 2 diabetes mellitus. The pathway map showed that ethanol extract of mulberry leaves was involved in the treatment of diabetes through pathways such as PI3K-AKT signaling pathway, NF-κB signaling pathway, and MAPK signaling pathway, etc.

3.1 Identification of anti-diabetic components and target genes

According to the LC-MS analysis, a total of 248 components were identified in the ethanol extract of mulberry leaves, which were all searched in the TCMSP database and a total of 86 compounds were found (Ru et al., 2014). Through using OB ≥ 30% and DL ≥ 0.18 as criteria for filtering, 18 compounds were selected, which were considered as active components and were linked with 96 target genes. As for other compounds (n = 162) that were not found in TCMSP database, we used PubChem to generate their two-dimensional structures (Kim et al., 2016). These structures were then imported into the PharmMapper server to predict whether they were associated with any disease (Liu et al., 2010). A total of 58 compounds were identified to be associated with 633 diseases via 328 target proteins, the names of which were then mapped through the Gene-Disease Association Database to get the names of target genes in TCM-Mesh server (Zhang et al., 2017). Moreover, the two databases, GeneCards and OMIM, were thoroughly searched for target genes that have been linked with diabetes by using the keyword diabetes-type 2. A total of 9948 target genes were collected. In order to find out the most relevant genes associated with both type 2 diabetes and components of mulberry leaves, the intersection of targets from TCMSP, PharmMapper, and GeneCards/OMIM were obtained. A total of 37 targets were identified, the details of which were recorded in Table 1. UniProt ID, gene names, protein names, and the interacted compounds of the targets were all included in the table.

Further analysis showed that the 37 targets interacted with 63 compounds of mulberry leaves. 14 compounds were identified from TCMSP with confirmed biological activity toward T2DM and their PubChem CID, molecular formula, oral bio-availability, blood–brain barrier, drug likeness, and 2-Dstructure were all present in Table 2. Among the 14 bioactive compounds, we quantified the concentration of three representative chemical components (Kaempferol, Licoricone and morin) via LC-MS/MS, according to which, content of morin was 1.58 ± 0.05 mg/g, kaempferol 7.28 ± 0.12 mg/g, and licoricone 0.31 ± 0.02 mg/g. The results were obtained with repetition in triplicates. As for the other 49 compounds, they were potentially bioactive in T2DM treatment and their information was collected from PharmMapper. The details of these compounds were recorded in Supplementary Table S2.

3.2 Network of target protein–protein interactions

All target proteins were imported into the online database STRING https://string-db.org/ version 11.0 to construct the protein–protein interaction network (Szklarczyk et al., 2019). According to the result, the network has significantly more interactions (n = 138) than expected (n = 41), which means that these proteins have more functional and physical interactions among themselves and could be biologically connected as a group (Supplementary Figure S4). In order to better illustrate the interactions between proteins, the network was then visualized in CytoScape (Fig. 2A), in which larger nodes means greater degrees (more connections with other nodes), hence pivotal roles in the network. As for the line thickness, it is calculated based on the combined score generated through STRING database. The thicker the line is, the stronger the connections are between the two nodes. Mcode clustering of the network in Fig. 2A was performed, from which the most highly inter-connected regions in a network were identified (Fig. 2B). According to the results, AKT1 is the core target gene in the network while 13 target genes (AKT1, TNF, MMP9, ICAM1, NOS3, MMP2, KDR, SELE, NMP1, NOS2, HMOXX1, PPARG, and ALOX5) were tightly linked together.

3.3 Interactions between anti-diabetic compounds and target genes

Generally speaking, a target can be modulated by multiple compounds while a compound could act on multiple targets. To better understand the interactions between 63 anti-diabetic compounds and 37 target genes, we constructed a network by using CytoScape 3.8.2, which included 100 nodes and 233 edges (Fig. 3). In the network, the degree of a node represents its connections with other nodes, which was reflected by the diameter of the node. That is, the nodes with more connections play pivotal roles in the entire interaction network, which indicates that the nodes could be key target genes during the treatment of type 2 diabetes or compounds that play important anti-diabetic roles in the ethanol extract of mulberry leaves. In specificity, the green circles represent 37 target genes (proteins), the blue circles represent 14 compounds collected from TCMSP database, while the pink circles represent 49 compounds predicted in the PharmMapper database. Grey lines connect circles together, indicating the interactions between the components and the targets. In addition, according to the diameters of the nodes, kaempferol (n = 23) and methyl lucidenate P (n = 11) are the two key important compounds from TCMSP and PharmMapper that have the highest number of interactions with target genes, respectively. In contrast, multiple target genes were modulated by the same number of compounds, which included acetylcholinesterase, polyunsaturated fatty acid 5-lipoxygenase, androgen receptor, vascular endothelial growth factor receptor 2, interstitial collagenase, 72 kDa type IV collagenase, matrix metalloproteinase 9, nitric oxide synthase, MORF4 family-associated protein 1, and peroxisome proliferator-activated receptor gamma.

Close

Fig. 3

The interaction network among components from the ethanol extract of mulberry leaves and T2DM-related target proteins. All the compounds were labelled with PubChem ID in order to avoid compound-name overlapping. The diameters of circles reflect the number of interactions. Green circles: target genes (proteins). Blue circles: 14 compounds collected from TCMSP database. Pink circles: 49 compounds collected from PharmMapper database. Grey lines: connections between circles.

Export to PPT

3.4 GO and KEGG pathway enrichment analysis

GO describes gene products with three independent categories, which include biological process, cellular component, and molecular function (Consortium, 2019). GO enrichment analysis was performed in order to find out which GO terms are over- or under-represented by using annotations for a specific gene set. In this study, 37 target proteins were analysed through DAVID (Dennis et al., 2003) and top 10 GO entries for each category were selected according to false discovery rate (FDR, < 0.05) as shown in Fig. 4.

Close

Fig. 4

GO enrichment analysis of target proteins. (A) Bar plot of the number of GO entries in the functional categories of biological process (BP, blue bar), cell composition (CC, green bar), and molecular function (MF, orange bar). (B) Bubble plot of GO entries filtered with -log₁₀(P-value < 0.05).

Export to PPT

A total of 30 pathways filtered from KEGG analysis with P-value < 0.05 were selected and visualized as bar plot with the number of target proteins in each of the KEGG pathways (Fig. 5A) while the corresponding -log10(P-value) of each pathway was shown as bubble plot in Fig. 5B. Various inflammatory signaling pathways such as TNF signaling pathway, calcium signaling pathway, cGMP-PKG signaling pathway, cAMP signaling pathway, regulation of lipolysis in adipocytes, NF-κB signaling pathway, insulin resistance pathway, and sphingolipid signaling pathway were associated with type 2 diabetes mellitus. Thus, according to the GO and KEGG pathway enrichment analyses, components from the ethanol extract of mulberry leaves were involved in a variety of signaling, signal transduction, and lipolysis pathways associated with T2DM.

Close

Fig. 5

Target protein KEGG pathway analysis. (A) Bar plot of the number of target proteins in each of the KEGG pathways. (B) Bubble plot of KEGG pathways. All the KEGG pathways were filtered with P-value < 0.05.

Export to PPT

In order to elucidate the interactions of target proteins with all the identified KEGG pathways, CytoScape was used to construct a target-pathway network (Fig. 6). According to the diagram, the target proteins involved in the TNF signaling pathway included inhibitor of nuclear factor kappa-B kinase subunit beta (IKBKB), RAC-alpha serine/threonine-protein kinase (AKT1), E-selectin (SELE), tumour necrosis factor (TNF), matrix metalloproteinase 9 (MMP9), and intercellular adhesion molecule 1 (ICAM1). The target proteins involved in the calcium signaling pathway included muscarinic acetylcholine receptor M3 (CHRM3), nitric oxide synthase 2 (NOS2), nitric oxide synthase 3 (NOS3), Beta-1 adrenergic receptor (ADRB1), Beta-2 adrenergic receptor (ADRB2), and cAMP-dependent protein kinase catalytic subunit alpha (PRKACA). The target proteins involved in the cGMP-PKG signaling pathway included NOS3, cGMP-inhibited 3′ (PDE3A), AKT1, ADRB1, and ADRB2. The target proteins involved in the cAMP signaling pathway included PDE3A, AKT1, ADRB1, ADRB2, and PRKACA. The target proteins involved in the regulation of lipolysis in adipocytes included AKT1, ADRB1, ADRB2, and PRKACA. The target proteins involved in the NF-κB signaling pathway included IKBKB, apoptosis regulator Bcl-2 (BCL2), TNF, and ICAM1. The target proteins involved in the insulin resistance pathway included IKBKB, NOS3, AKT1, and TNF. The target proteins involved in the sphingolipid signaling pathway included NOS3, BCL2, AKT1, and TNF.

Close

Fig. 6

Schematic illustration of the network of pathway-target interactions. Blue-circle nodes represent the pathways involved in the target protein while pink-circle nodes represent target proteins. The diameter of the circle node represents the number of connections (interactions) with other nodes.

Export to PPT

3.5 Target path analysis

The 37 selected targeted genes were submitted to the Ensemble database by BioMart, through which the gene names were converted to Entrez Gene ID. KOBAS database was then used for KEGG pathway analysis, and target genes in the given analysis results were arranged in ascending order based on the P-values. The statistically significant target genes in pathways were highlighted in red and present in Fig. 7. The pathway map showed that ethanol extract of mulberry leaves was involved in the treatment of diabetes through regulating specific pathways, including PI3K-AKT signaling pathway, NF-κB signaling pathway, and MAPK signaling pathway, etc.

Close

Fig. 7

Pathway map of ethanol extract of mulberry leaves in the treatment of type 2 diabetes mellitus. The main target proteins that are involved in diabetes-related pathways are highlighted in red. Arrows represent the activation effect, T-arrows represent the inhibition effect, and segments show the activation effect or inhibition effect.

Export to PPT