DO leaves were collected from the Jianfengling area, Sanya City, Hainan Province, China. A voucher specimen (No. 20200123) was stored in the Key Laboratory of Modern Preparation of Traditional Chinese Medicine, Ministry of Education, Jiangxi University of Traditional Chinese Medicine, China.

The D. odorifera (500 g) extract (DLE) was prepared using air-dried leaves and 70% ethanol under a refluxing extraction (2 h, 5 L × twice) and filtered. The combined extracts were then rotary evaporated at 45°C and lyophilized (Shao et al., 2016). The ethanol extract was dissolved with 30% methanol and stored at 4°C until use. Rutin was accurately weighed to 5.41 mg and dissolved with 30% methanol to obtain the reference solution (0.541 mg/mL). The general flavonoid content of DLE was determined using the colorimetric method. Rutin was used as the standard. The absorbance of samples was measured at 500 nm.

The air-dried leaves of D. odorifera (100 g) were used to prepare 70% ethanol under refluxing extraction (2 h, 1 L × twice) and filtered. The combined extracts were then rotary evaporated at 45°C and lyophilized. The extract ethanol was dissolved with methanol and stored at 4°C until use. Separation was performed on a 250 × 4.6 mm Phenomenex C18 (5 μm, California, USA) at 30°C, and the UV absorbance was monitored at 260 nm on Agilent 1100. The flow rate was 1.0 mL/min. The solvent system used the improved method published earlier (Li et al., 2016): composed of acetonitrile (A) and 0.1% phosphoric acid (B) using an optimized gradient program: 0–10 min 35% A; 10–15 min 35%–48% A; 15–30 min 48%–49% A. The injection volume was 10 μL. The DLE in 70% ethanol was accurately weighed and dissolved in a 2 mL volumetric flask with methanol. The final solution was filtered through a 0.22 µm microfilter to obtain a reference solution.

C57BL/6J male mice (8~10 weeks old; weighing 18~22 g) were purchased from the College of Veterinary Medicine YanZhou University of Comparative Medicine (certificate No. SCXK (SU) 2022-0009) and placed in the laboratory of the First Affiliated Hospital of Nanchang University. All efforts were made to minimize animal suffering and the number of animals used. C57BL/6 mice were anesthetized with 1.2% isoflurane during the ischemia period. The environment temperature was maintained at 37°C during the ischemia and recovery period. A blunt tip 6–0 nylon monofilament was inserted into the neck vessel and advanced 8–10 mm forwards. After 1 h of occlusion, the monofilament was withdrawn, then 24 h reperfusion was immediately followed. Mice in the sham-operated group underwent the same surgical incision, but the nylon monofilament was not advanced into the internal carotid artery. Animals whose blood flow dropped below 30% of pre-ischemia levels were used for further study.

The administration dose was determined according to earlier studies (Mei et al., 2021), while the dose given to rats was converted corresponding to that of mice. The DLE treatment group was administered intragastrically with 0.6, 1.2, or 2.4 g/kg 1 h of the onset of cerebral ischemia and every 24 h during the experimental period, whereas the sham-operated group was administered intragastrically with a solution of equivalent volume NaCl. Mice were randomly categorized into six groups (8 mice for each group): (1) sham-operated mice (sham); (2) mice exposed to 1 h of ischemia and 24 h of reperfusion stroke (I/R); (3) mice exposed to 1 h of ischemia and 24 h of reperfusion stroke and treatment with a low dose of DLE (0.6 g/kg) (I/R+0.6 g/kg DLE); (4) mice exposed to 1 h of ischemia 24 h of reperfusion stroke, and treated with a middle dose of DLE (1.2 g/kg) (I/R+1.2 g/kg DLE); (5) mice exposed to 1 h of ischemia and 24 h of reperfusion stroke and treatment with a high dose of DLE (2.4 g/kg) (I/R+2.4 g/kgDLE); (6) 1 h of ischemia stroke, 24 h of reperfusion stroke and treatment with 10 mg/kg nimodipine (I/R+10 mg/kg NM) (Lv and Fu, 2018). At the end of the cerebral I/R period, the mice were examined for leakage of Evans blue, immunostaining, immunohistochemistry, and western blot analysis.

The epoxy-activated sepharose™6B (EAS-6B) was from GE Healthcare, Delaware, USA. The affinity resin was prepared using the earlier published formula (Lv and Fu, 2018). The leaf extract was dissolved in DMSO, then added into EAS-6B, which was fully expanded before 300 mmoL, soaked at 25°C for 24 h, lightly filtered with double distilled water for 15 min, drained, and preserved at 4°C under dry conditions. Brain tissue homogenates were centrifuged at 9,500 rpm for 10 min and 50,000 rpm for 30 min and diluted in lysate buffer to a total protein concentration of about 7 mg/mL. The homogenate was shaken gently at 4°C for 4 h with an affinity medium of 60 mL. This step was repeated five times. Sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE; Bio-Rad, California, USA) was performed with the supernatant solution, and silver staining was performed using Quantity One software 4.6.2. The proteins were identified in the mass spectrometry database from the SEQUEST output file.

Limma and Impute package in R software (4.0.4) were installed, and each group of data in section ‘Materials and chemicals’ to matrix were processed, and the missing matrix filled. The log₂ value of the data set was taken and corrected to the same level. The up-regulated and down-regulated proteins compared between control-affinity proteins or DLE-affinity proteins were screened out, and with differentially expressed proteins, the truncated values were set to log₂F ≥ 1 and p-value < 0.05. The volcano map and heat map were used to display the differential expressed proteins. In R software, define log₂FC as the x-axis and the p-value as the y-axis to draw the volcano map.

The screened differential expressed proteins were processed in R Studio 4.0.4 software using the WGCNA package to construct co-expression networks. First, the scale-free topology model Fitsigned R^2 and mean connectivity were calculated to determine the soft threshold (power) for the protein co-expression network. According to the compatible soft threshold, the degree of fitting was considered between the chosen genes and putative proteins. Then, the protein expression similarity matrix was converted into an adjacency matrix; then, the adjacency matrix was converted into a topological overlap matrix (TOM). The TOM assay was applied to convert the protein expression similarity matrix to the adjacency matrix. Hierarchical clustering based on TOM was constructed using a dynamic tree-cutting algorithm, Connectivity Cut was set as 1,000, and the preset number of modules was set as height 0.8 and minsize 3. After determining the protein modules, the eigenvector values of each module were calculated for clustering analysis, and the modules that were closer were merged into new modules. Genes with similar expression patterns were considered specific branch modules marked with a unique color.

GO enrichment and KEGG signaling pathway analysis were performed on gene modules using the annotation and database DAVID (https://david.ncifcrf.gov/). Cut-off criteria were set as count ≥10, and p < 0.05 was chosen as the critical value, while false discovery rate (FDR) correction was applied for multiple test corrections. The significance level was set to 0.05, and classification information was set to functional classification and human annotation. The biological processes (BP), cellular components (CC), and molecular functions (MF) module were applied to analyze the biological function of proteins, and −Log₂FC was defined to make the biological function map for the y-axis. The protein function was enriched by Fisher’s rectified p-value ≤ 0.01, and the signaling pathways with similar functions were classified according to the enrichment score.

0.1 mL/10 g of 2% Evans blue (Sigma Aldrich Co., Saint Louis, USA) was administered through a caudal vein 2 h before euthanasia. To quantitatively measure the leakage of Evans blue, the ipsilateral hemisphere was homogenized and centrifuged at 12,000 rpm for 20 min. The absorbance of the supernatant was measured by spectrophotometer at 620 nm, and the concentration of Evans blue was determined.

The I/R brain tissue was soaked with 4% paraformaldehyde for 24 h and then embedded with paraffin after dehydration. Anti-occludin, claudin-5, and zonula occludens-1 (ZO-1; rabbit polyclonal IgG, 1:50, Zyemd, Wuhan, China) antibodies were incubated with brain tissues. The occludin, claudin-5, and ZO-1 positive boundaries were quantitatively measured in a 20 μm enlarged view. The data were expressed as a percentage of the number of positive boundaries in cerebral I/R mice.

Brain tissues in each group were cut into 5 mm pieces, washed with phosphate-buffered saline, immobilized with cold ethanol at 4°C for 10 min, and fixed with 10% normal goat serum, 3% bovine serum albumin, and 0.1% Triton X-100 in phosphate-buffered saline for 1 h. Then, the tissue pieces were incubated with the occludin (ab216327, Abcam, Cambridge, UK), claudin-5 (ab131259, Abcam, Cambridge, UK), and ZO-1 (ab96587, Abcam, Cambridge, UK) primary antibodies (IF for 1:100) for 48 h at 4°C. Brain tissues were incubated overnight with Alexa Fluor® 488 conjugated donkey anti-goat IgG (H+L) antibody (ab150129, Abcam, Cambridge, UK; 1:2000) at 4°C, while the nuclei of all cells were washed and incubated with 4′6-diamidino-2-phenylindole (DAPI, ab104139, Abcam, Cambridge, UK; 1:1000). Briefly, confocal images were used to observe the fluorescence images of each group (ChemiScope 6100BZ, Shanghai, China).

The protein concentration was determined by the BCA method. The membranes separated by SDS-PAGE electrophoresis were transferred onto polyvinylidene difluoride membranes. The membranes were incubated for blocking with 5% nonfat dry milk at room temperature for 2 h, then incubated with anti-leucine-rich α-2 glycoprotein 1 (LRG1; ab178698, Abcam, Cambridge, UK), anti-transforming growth factor beta 1 (TGF-β1; ab235578, Abcam, Cambridge, UK), rho-associated protein kinase1 (ROCK1; ab134181, Abcam, Cambridge, UK), phosphorylated-mechanistic target of rapamycin (p-mTOR; ab109268, Abcam, Cambridge, UK), mTOR (ab13490, Abcam, Cambridge, UK), anti-phosphorylated-SMAD2 (p-SMAD2; ab280888, Abcam, Cambridge, UK), anti-SMAD2 (ab40855, Abcam, Cambridge, UK), anti-p-SMAD3 (ab52903, Abcam, Cambridge, UK), anti-SMAD3 (ab40854, Abcam, Cambridge, UK), anti-occludin (ab216327, Abcam, Cambridge, UK), anti-claudin-5 (ab131259, Abcam, Cambridge, UK), anti-ZO-1 (ab276131, Abcam, Cambridge, UK), and β-actin (ab8226, Abcam, Cambridge, UK) primary antibodies (1:1000), and then incubated with secondary antibody (HRP-conjugated secondary antibody, 1:2000, ab6728; Abcam, Cambridge, UK) for 50 min. Enhanced chemiluminescence was used for image capture through a bioimage analysis system (Bio-Rad, California, USA).

All values are expressed as mean ± SD. The data were analyzed based on a two-tailed student’s t-test (two groups) or a two-way ANOVA followed by Bonferroni’s test (three or more groups) through Graph Prism 5.2. The significance level was set at p < 0.05 and p < 0.01.

HPLC method was used for the detection of general flavones from D. odorifera leaves. At wavelengths of 211, 254, 270, and 360 nm, the absorption peak of fingerprint of general flavonoids of 70% ethanol extract of D. odorifera leaves was detected (Fig. 1). The mass proportion of general flavonoids was 140.74 ± 2.16 mg/g (Fig. 1).

Eight decreased expression protein bands were singled out from cerebral I/R mice treated with control-affinity resin or DLE (70% ethanol extract of D. odorifera leaves)-affinity resin via series affinity chromatography (Figs. 2A and 2B). The identified proteins from the control affinity resin were compared with those protein bands from DLE-affinity resin. In total, 2,190 proteins were identified (Suppl. Tables S1 and S2). For subsequent analysis, 952 differential expression proteins were screened (log₂Fold Change > 0.5, adj.p.Val < 0.05) (Suppl. Table S3 and Suppl. Fig. S1). The comparison revealed 644 up-regulated and 308 down-regulated differentially expressed proteins on proteins identified from the control affinity resin.

A total of 952 differential expressed proteins were screened by WGCNA. The parameter β was screened between 1 and 20, and the optimal soft threshold value was determined as 6. When the soft threshold power was set as 6, the degree of fitting was scale-free R^2 = 0.95 and trunc.R^2 = 0.95 (Figs. 3A and 3B). According to the soft threshold value, the network approximated the real biological network, and it amounted to the power law distribution. The relationship matrix was translated into an adjacency matrix, and finally, seven expression modules were constructed (Figs. 3D and 3E). Co-expression modules were constructed based on the TOM. The resulting protein tree and corresponding module colors are shown in Fig. 3B, of which the turquoise module had the strongest internal interaction (Fig. 3F).

Hub proteins identified via WGCNA were (protein name symbols name represented for protein name) AKT1, PIK3CA, NOS3, SMAD3, SMAD1, ATK1, MAPK1, TGFBR2, TGFBR1, MAPK3, IGF1R, LRG1, mTOR, ROCK1, TGFB1, TGF1, SMAD2, SMAD5 (Fig. 4A). Candidate proteins obtained from hub protein module via WGCNA were subjected to BP, CC, and MF analyses, and −Log₁₀FDR was defined to make the biological function map for the y-axis. BP was related to TGF-β receptor signaling pathway, common-partner SMAD protein phosphorylation, pathway-restricted SMAD protein phosphorylation, etc. CC was associated with positive regulation of pathway-restricted SMAD protein phosphorylation, SMAD protein complex, SMAD2-SMAD3 protein complex, etc. MF was related to the transcription factor complex, inhibitory-SMAD (I-SMAD) binding, SMAD binding, etc. (Fig. 4B). The hub proteins were associated with KEGG signaling pathways as TGF-β signaling pathway, mitogen-activated protein kinases (MAPK) signaling pathway, focal adhesion, adherens junction, etc. (Fig. 4C).

The content of the Evans blue inside cerebral I/R brain tissue was examined to assess the effect of DLE on the blood-brain barrier permeability quantitatively. Quantitative content revealed DLE treatment significantly reduced Evans blue leakage (low, middle, high dose: 2.21 ± 0.24 mg/g, p < 0.01, 1.94 ± 0.21 mg/g, p < 0.01, 1.52 ± 0.24 mg/g, p < 0.01) compared with the cerebral I/R group (2.89 ± 0.26 mg/g, p < 0.01, sham: 0.25 ± 0.29), and 10 mg/kg nimodipine (1.54 ± 0.27 mg/g, p < 0.01) had the same effect (Fig. 5).

This linear shape along the tight junction proteins of blood vessels reduced with lower expression in response to I/R (Fig. 6A), with a weaker signal strength of occludin, claudin-5, and ZO-1 (occludin: 10.49 ± 8.77, p < 0.01; claudin-5: 7.49 ± 5.21, p < 0.01; ZO-1: 7.78 ± 10.15, p < 0.01; sham: occludin: 51.49 ± 9.47; claudin-5: 47.59 ± 6.42; ZO-1: 56.88 ± 8.49). The DLE dose-dependent improved the pathological injury of brain tissues in cerebral I/R mice by strengthening immunohistochemical expression intensity of tight junctions (low, middle, high dose DLE: occludin: 14.59 ± 7.44, p < 0.01, 18.48 ± 8.45, p < 0.01, 28.47 ± 5.49, p < 0.01; claudin-5: 14.59 ± 6.34, p < 0.01, 26.47 ± 4.17, p < 0.01, 37.09 ± 3.84, p < 0.01; ZO-1: 12.47 ± 8.55, p < 0.01, 19.09 ± 8.49, p < 0.01, 42.35 ± 9.78, p < 0.01, Figs. 6A and 6B), the integrity of vascular vessel expression was restored to some extent.

Tight junctions were continuous and well organized in the sham-operated blood vessel edge through immunofluorescent staining (Figs. 6C–6F). After stimulation with I/R, occludin, claudin-5, and ZO-1 revealed structural disturbance with reduced fluorescence compared with the sham-operated group. Moreover, immunocytochemical staining of tight junction proteins along the blood vessel edge was restored to a higher fluorescence expression intensity in response to the treatment of DLE. After treatment with DLE, the protein expression of LRG1, TGF-β1, p-SMAD2/SMAD2, p-SMAD3/SMAD3 were downregulated, while the expressions of ROCK1, p-mTOR/mTOR, occludin, claudin-5, ZO-1 were up-regulated (I/R+2.4 g/kg DLE: LRG1: 1.24 ± 0.02, p < 0.01; TGF-β1: 1.17 ± 0.05, p < 0.01; p-SMAD2/SMAD2: 1.24 ± 0.04, p < 0.01; p-SMAD3/SMAD3: 1.21 ± 0.03, p < 0.01, ROCK1: 1.44 ± 0.04, p < 0.01, p-mTOR/mTOR: 1.46 ± 0.07, p < 0.01, occludin: 1.45 ± 0.09, p < 0.01, claudin-5: 1.47 ± 0.04, p < 0.01, ZO-1: 1.45 ± 0.04, p < 0.01, Fig. 7) compared with those of the I/R group (LRG1: 1.72 ± 0.03, p < 0.01; TGF-β1: 1.66 ± 0.04, p < 0.01; p-SMAD2/SMAD2: 1.79 ± 0.07, p < 0.01; p-SMAD3/SMAD3: 1.74 ± 0.05, p < 0.01, ROCK1: 1.00 ± 0.05, p < 0.01, p-mTOR/mTOR: 1.00 ± 0.05, p < 0.01, occludin: 1.00 ± 0.03, p < 0.01, claudin-5: 1.02 ± 0.04, p < 0.01, ZO-1: 1.02 ± 0.07, p < 0.01, Fig. 7).