Non-obese diabetic (NOD) mice and Nrf2 ^−/− mice were obtained from the Model Animal Research Center of Nanjing University (Nanjing, China). C57BL/6 mice were from Comparative Medical Center of Yangzhou University (Yangzhou, China). Experiments used 6–12-week-old and sex-matched mice, unless otherwise indicated. All animal care and handling procedures were conducted in accordance with the protocols approved by the Institutional Animal Care and Use Committee (IACUC) at Yangzhou University (Yangzhou, China).

The following reagents were used: Astilbin (HY-N0509); 1-aminobenzotriazole (ABT, HY-103389), a nonspecific and irreversible inhibitor of CYP enzymes; etomoxir (HY-50202), an inhibitor of CPT1α; SF1670 (HY-15842), an inhibitor of PTEN; N-Acetyl-L-cysteine (NAC, HY-B0215), a GSH-related antioxidant; GW9662 (HY-16578), a potent and selective PPARγ antagonist; and dorsomorphin (Compound C, HY-13418A), a selective and ATP-competitive AMPK inhibitor (MedChemExpress, NJ, United States).

CD3⁺ CD4⁺ cells were isolated from the spleen by magnetic bead sorting (Miltenyi Biotec GmbH, BG, Germany). Isolated CD4⁺ T cells and unlabeled cells were counted and then cultured in RPMI containing 10% FBS and penicillin/streptomycin in a humidified atmosphere containing 5% CO₂ at 37°C. Cells were further stimulated in vitro with astilbin, with or without IL-12 (20 U/ml), plate-bound anti-CD3 (2 μg/ml) and soluble anti-CD28 (1 μg/ml) antibodies, GW9662, NAC, SPF1670, dorsomorphin, ABT, and etomoxir according to the experimental design. After 12 h, 24 h, or 2 days, cells were harvested for flow cytometric analysis.

Cells were stained with the indicated antibodies for 30 min at 4°C. The following antibodies for flow cytometry were obtained from BioLegend, CA, United States: anti-mouse CD4 (GK1.5), anti-mouse CCR6 (29-2L17), anti-mouse CCR9 (9B1), anti-mouse CD36 (HM36), and anti-human CD4 (OKT4). For intracellular cytokine analysis, cells were stimulated with PMA and ionomycin for 30 min and treated with brefeldin A for 4 h before staining Fixation/Permeabilization Solution Kit (BD, NJ, United States) followed by anti-mouse TNF-α (MP6-XT22, BioLegend, CA, United States), anti-mouse IFN-γ (XMG1.2, BD, NJ, United States), anti-human TNF-α (Mab11, BioLegend, CA, United States), or anti-human IFN-γ (4S.B3, BioLegend, CA, United States). Cell apoptosis was detected with an Annexin V staining kit (BD, NJ, United States) according to the manufacturer’s instructions. For cell proliferation, cells were labeled with 5 μM CFDA-SE (CFSE) (Invitrogen, CA, United States). Seventy-two hours later, they were collected and analyzed by flow cytometry. Acquisition was carried out on FACSVerse (BD, NJ, United States), and analysis was performed with FlowJo v10 software.

NOD mice were randomly divided into three groups, and each group was intraperitoneal injected twice weekly with PBS or astilbin (50, 100 mg/kg) for 7 weeks. Body weight and blood glucose of NOD mice were monitored once a week until the end of the observation period. NOD mice were characterized as diabetic when blood glucose readings were >13.9 mmol/L for two consecutive measurements. The indexes of dead mice were shown as the last blood glucose or weight data. At the end of the observation period, mice were euthanized by cervical dislocation, and pancreatic CD4⁺ T cells were measured by flow cytometry.

Total protein was extracted and boiled for 5 min, resolved by electrophoresis, and transferred to PVDF membranes. Blots were blocked for 1 h with 5% milk in TBST followed by incubation with antibodies (dilution 1:500 to 1:2000) at 4°C overnight. The following antibodies were used for western blot: ACC (Cat# 3676, RRID:AB_2219397), p-ACC (Cat# 11818, RRID:AB_2687505), HKII (Cat# 2867, RRID:AB_2232946), mTOR (Cat# 2983, RRID:AB_2105622), p-mTOR (Cat# 5536, RRID:AB_10691552), Akt (Cat# 4691, RRID:AB_915783), p-Akt (Cat# 4060, RRID:AB_2315049), AMPK (Cat# 5831, RRID:AB_10622186), p-AMPK (Cat# 50081, RRID:AB_2799368), LKB1 (Cat# 3047, RRID:AB_2198327), p-LKB1 (Cat# 3482, RRID:AB_2198321), PTEN (Cat# 9188, RRID:AB_2253290), PPAR-γ (Cat# 2443, RRID:AB_823598), PI3K p110α (Cat# 4249, RRID:AB_2165248), β-actin (Cat# 4970, RRID:AB_2223172), NF-κB p65 (Cat# 8242, RRID:AB_10859369), NF-κB p-p65 (Cat# 3033, RRID:AB_331284), p38 (Cat# 8690, RRID:AB_10999090), p-p38 (Cat# 4511, RRID:AB_2139682), JNK (Cat# 9252, RRID:AB_2250373), p-JNK (Cat# 4668, RRID:AB_823588), Erk (Cat# 4695, RRID:AB_390779), p-Erk (Cat# 4370, RRID:AB_2315112), Stat3 (Cat# 4904, RRID:AB_331269), p-Stat3 (Cat# 9145, RRID:AB_2491009), SOCS3 (Cat# 52113, RRID:AB_2799408) (Cell Signaling Technology, MA, United States), Glut1 (Cat# ab115730, RRID:AB_10903230), CPT1α (Cat# ab234111, RRID:AB_2864319), LPL (Cat# ab21356, RRID:AB_446221) (Abcam, Cambridge, Britain), CYP1B1 (Cat# 18505-1-A, RRID:AB_2878548) (Proteintech, IL, United States), CYP19A1 (Cat# YT1190, RRID:AB_2864736) (Immunoway, DE, United States) and P-PPARγ (abs130911, absin, Shanghai China). HRP-conjugated anti-rabbit (Cat# 7074, RRID:AB_2099233, Cell Signaling Technology, MA, United States) secondary antibodies were used at 1:5000 dilution for 1 h at room temperature. The bands were detected by developing with chemiluminescent HRP substrate (Thermo Scientific, MA, United States), and the intensity of bands was determined by imaging with a molecular imager (Bio-Rad, CA, United States). Bands quantified using the ImageJ software. All results were normalized to those of β-actin, which was used as a loading control, and phosphorylated proteins is compared to the total amount of proteins and housekeeping genes (β-actin).

To determine the mitochondrial mass and the levels of intracellular and mitochondrial ROS (mROS), the cells were incubated with 100 nM MitoTracker Green (Beyotime, Shanghai, China) (De Nicola et al., 2011), 10 μM DCFH-DA (Beyotime, Shanghai, China), and 10 μM MitoSOX Indicator (Thermo Fisher Scientific, MA, United States) for 15 min in a CO₂ incubator at 37°C. After washing twice, the data were acquired on a FACSVerse (BD, NJ, United States) and analyzed with FlowJo v10 software.

Total RNA was isolated from mouse CD4⁺ T cells with TRIzol (1 ml) reagent and purified using Quick-RNA Miniprep columns and RNase-free DNase digestion. RNA integrity was assessed with the Agilent 2100 Bioanalyzer (Agilent Technologies, CA, United States). RNA purity and quantity were evaluated with a NanoDrop 2000 spectrophotometer (Thermo Scientific, MA, United States). Only samples with RIN scores above nine and a minimum total RNA of 500 ng were used for library preparation. Then, the libraries were constructed using TruSeq Stranded mRNA LT Sample Prep Kit (Illumina, CA, United States) according to the manufacturer’s instructions. Subsequently, the transcriptome sequencing data were used for a series of analyses, including differential expression analysis, gene enrichment analysis, and expression level analysis.

The potential targets of astilbin were predicted using the PharmMapper database (http://www.lilab-ecust.cn/pharmmapper/index.php) by performing Druggable Pharmacophore Models and SwissTargetPrediction (http://www.swisstargetprediction.ch/) on the basis of the principle that chemicals with similar structures may have similar functions. All the screened targets were input into the UniProt database (http://www.uniprot.org/) to exclude the same targets and non-Homo sapiens targets. The corresponding predicted target information obtained was collected and sorted according to fit score and probability.

The chemical structure of astilbin was obtained from the PubChem compound database (http://pubchem.ncbi.nlm.nih.gov/) with ID 29838-67-3. The crystal structures of CYP1B1 and CYP19A1 proteins with PDB IDs 3PM0 and 3S79, respectively, were obtained from the Research Collaboratory for Structural Bioinformatics (RCSB) Protein Data Bank. The crystal water molecules, ligand atoms, and ions that were bound to the proteins were removed. Hydrogen atoms were subsequently added using AutoDock Tool program version 1.5.6. Molecular docking between astilbin and human cytochrome P450 1B1 (CYP1B1) was performed using AutoDock Vina on the basis of the Lamarckian genetic algorithm, which combines energy evaluation through grids of affinity potential to find a suitable binding position for a ligand on a given protein. The grid box for docking was positioned properly at the active binding site in the center. The genetic algorithm and its run were set to 1000, as the docking algorithms were set on default. Finally, the results were retrieved as binding energy, and docking with binding energies lower than −5 kcal/mol was selected as a significant binding event and was visualized using PyMOL version 1.5.0.3. software for obtaining hydrogen bond, hydrophobic, and electrostatic interactions.

SPR experiments were performed with a Biacore T200 (GE Healthcare, MA, United States). CYP1B1 coupling protein (ImmunoClone, NY, United States) was diluted in 10 mM sodium acetate (pH 5.0) and immobilized on the sensor chip CM7 by primary amine coupling kit (Biacore, GE Healthcare, MA, United States) to a final response level of approximately 6000 resonance units. A blank channel was used as the negative control. All data were collected at 25°C with HBS-EP running buffer (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, 0.05% surfactant polysorbate 20, pH 7.4) at a flow rate of 30 μl/min. Concentration series of the original astilbin (two-fold dilutions starting from a top concentration of 200 μM) was injected over immobilized proteins for a contact time of 180 s and dissociation time of 300 s. The resulting data were fit to a 1:1 binding model using Biacore Evaluation Software v3.2, and the Kd value was determined.

Silencing of CYP1B1 in CD4⁺ T cells was achieved using lentivirus. The specific lentivirus CYP1B1-shRNA (5′-GCC​TGA​CCA​TTA​AGC​CCA​AGT​CTC​GAG​ACT​TGG​GCT​TAA​TGG​TCA​G GCTTTTT-3′) and negative control lentivirus shRNA were designed and synthesized by Genecreate, Wuhan, China. Lentiviral plasmid vectors and packaging vectors were co-transfected in 293T cells with Lipofectamine 3000 Reagent (Thermo Fisher, MA, United States). For lentiviral infection, CD4⁺ T cells were mixed with the virus and 10 μg/ml of polybrene (Sigma-Aldrich, MS, United States). The knockdown effect of CYP1B1 in CD4⁺ T cells was confirmed by RT-PCR and Western blot analysis.

All data are presented as the means ± SD or means ± SEM. Comparisons between two groups were assessed using the unpaired two-tailed student’s t test. Comparison of multiple groups was performed using one-way ANOVA or two-way ANOVA. All statistical tests were two-sided. Statistical differences are significant for *, p ≤ 0.05; **, p ≤ 0.01; ***, p ≤ 0.001.

After the IC₅₀ of astilbin on CD4⁺ T cells (181 μg/ml) was verified ex vivo (Supplementary Figure S1A), the effects of astilbin treatment at various doses on TNF-α and IFN-γ, two key inflammatory cytokines produced by effector CD4⁺ T cells, were checked. As shown in Figure 1A, TNF-α production could be substantially inhibited in a dose-dependent manner. Although IFN-γ secretion was also downregulated statistically, the extent was weak. CCR9 expression of lymphocytes indicates the gut-homing activity. Consistent with the effect of astilbin on NK1.1^- CD4⁺ NKG2D⁺ regulatory T cells (Han et al., 2020), the expression of CCR9 on CD4⁺ T cells was also enhanced with astilbin treatment in a dose-dependent manner. Unexpectedly, no changes in the expression of IL-10 and CCR6 in astilbin-treated CD4⁺ T cells were observed (Supplementary Figures S1B,C). Astilbin slightly decreased TGF-β1 secretion of CD4⁺ T cells (Supplementary Figure S1D). We also did not observe any effects of astilbin on the proliferation (Figures 1B,C) and apoptosis of CD4⁺ T cells at doses of 40–160 μg/ml (Figure 1D; Supplementary Figure S1E).

IFN-γ, IL-1β, and TNF-α production by autoreactive Th1 cells is involved in the destruction of insulin-secreting β-cells in type 1 diabetes (Kaufmann et al., 2019). Since we noted that astilbin could reduce IFN-γ and TNF-α in CD4⁺T cells, we then analyzed whether astilbin could repress the onset of type 1 diabetes in NOD mice by regulating CD4⁺T cells. Compared with control group, blood glucose levels and body weights were significantly controlled after the administration of astilbin (50 or 100 mg/kg) (Figures 1E,F). When mononuclear cells of the pancreas from mice were isolated, less CD4⁺ T cells infiltrated the pancreas of astilbin-administered mice (Figure 1G). These results demonstrated that astilbin ameliorated T1DM involved with the downregulation of CD4⁺ T cell activities. Finally, astilbin also efficiently decreased TNF-α and IFN-γ production of peripheral blood CD4⁺ T cells from human healthy individuals as displayed in Figures 1H,I. Therefore, we identified that astilbin exerted immune-suppressive activities via downregulating inflammatory cytokine secretions of CD4⁺ T cells.

Several studies have found that astilbin treatment increases the expression of SOCS3 (Wang et al., 2016; Han et al., 2020; Sharma et al., 2020). Variations of SOCS3 were first measured in freshly isolated, IL-12-activated, or α-CD3/α-CD28-activated CD4⁺ T cells treated by astilbin. Astilbin could increase SOCS3 expression of freshly isolated and activated CD4⁺ T cells. The critical role of SOCS3 is manifested by its binding to both the JAK kinase and the cytokine receptor, which can result in the inhibition of STAT3 phosphorylation. Consequently, as expected phosphorylated STAT3 (p-Y705) levels were downregulated in astilbin-treated CD4⁺ T cells (Figures 2A,B; Supplementary Figures S2A,B). However, when variations of SOCS3 were further detected after CD4⁺ T cells were treated by astilbin (80 μg/ml) at different time courses. SOCS3 expression was increased by 2-h coculture of astilbin and maintained high levels even at 48 h. In parallel, p-STAT3 level was decreased by 2-h treatment, maintained low levels at 8–16 h, and reversed to normal levels at 24 h (Figure 2C; Supplementary Figure S2C). This trend of p-STAT3 changes may indicate that STAT3 activation could be regulated by other factors, besides SOCS3.

In general, SOCS3 and STAT3 are mutually regulated in cells (Carow and Rottenberg, 2014). Considering the decreased activation of STAT3 and the upregulation of SOCS3, p-STAT3 might not have promoted SOCS3 transcription in astilbin-treated CD4⁺ T cells. SOCS3 transcription is also activated by PPARγ to prevent IL-17-derived cancer growth (Berger et al., 2013). We determined whether the increased SOCS3 of CD4⁺ T cells was promoted by PPARγ. As expected, total and phosphorylated PPARγ (p-PPARγ) were remarkably increased in astilbin-treated CD4⁺ T cells (Figure 2D; Supplementary Figure S2D). Moreover, astilbin treatment at different doses resulted in parallel variations of p-PPARγ and SOCS3 expression (Figure 2E; Supplementary Figure S2E) and the expression of SOCS3 decreased significantly after the addition of PPARγ inhibitor GW9662 in CD4⁺ T cells, which was similar to the changes of p-PPARγ (Figure 2F; Supplementary Figure S2F). Next, GW9662 was used simultaneously to analyze the effects on functional changes of CD4⁺ T cells, which were cotreated by astilbin. Treatment of GW9662 reversed TNF-α (Figure 2G) and IFN-γ (Figure 2H) production and decreased CCR9 expression (Figure 2I) of astilbin-treated CD4⁺ T cells (Supplementary Figure S2G).

Another prominent regulator of antioxidant signaling pathways, that is, positively associated with PPARγ regulation is nuclear factor erythroid 2-related factor 2 (Nrf2) (Annie-Mathew et al., 2021). We then tested the expression of Nrf2 in astilbin-treated CD4⁺T cells. Interestingly, Nrf2 was also enhanced by astilbin treatment (Figure 2D). However, when CD4⁺ T cells derived from Nrf2 ^−/− mice were treated by astilbin, the decreased TNF-α production was not almost affected (Figure 2J; Supplementary Figure S2H), indicating that astilbin-mediated immunoregulatory effect was independent on Nrf2. Collectively, the above results demonstrated the key role of the PPARγ/SOCS3 pathway in astilbin-downregulated CD4⁺ T cells (Figure 2K).

Given that astilbin conducts anti-inflammatory activity via downregulating the phosphorylation of PI3K/Akt, NF-κB, and p38 MAPK signaling pathways, some molecules with diphosphatase activity might be stimulated by astilbin. Both phosphatase and tensin homolog (PTEN) was a phosphatase that efficiently inhibits the activation of PI3K/Akt, NF-κB, and p38 MAPK (Worby and Dixon, 2014). Through astilbin treatment, PTEN is remarkably induced in physiological CD4⁺ T cells and IL-12-stimulated CD4⁺ T cells. Given that PTEN is more increased in CD4⁺ T cells by the stimulation of α-CD3/α-CD28, the induction of PTEN by astilbin may be sheltered in α-CD3/α-CD28-stimulated CD4⁺ T cells. Correspondingly, phosphorylated Akt, NF-κB p65, and p38 decreased in activated CD4⁺T cells treated by astilbin (Figures 3A,B; Supplementary Figures S3A,B). In addition, 2-h astilbin treatment increased PTEN expression, and the increased expression lasted for 48 h. Similarly, induced PPARγ expression was observed after 2–24-h stimulation (Figure 3C; Supplementary Figure S3C).

When astilbin-treated CD4⁺ T cells were co-cultured with a PTEN inhibitor (SF1670), TNF-α (Figure 3D) production was partially reversed (Supplementary Figure S3D), and IFN-γ (Figure 3E) production was restored, confirming that decreased cytokine secretion of CD4⁺ T cells is induced by astilbin dependent on the induction of PTEN. On the other hand, incomplete restoration of TNF-α indicated that PTEN was only one of the signal molecules involved in astilbin-induced downregulation. Given that PTEN expression is transactivated by PPARγ (Worby and Dixon, 2014), we determined whether the inhibition of PPARγ affected PTEN and its downstream signaling molecules. As shown in Figure 3F, when CD4⁺ T cells were cotreated by astilbin and GW9662, increased PTEN expression was blocked, and the decreased expression of phosphorylated Akt, NF-κB p65, and p38 was recovered (Supplementary Figure S3E). Taken together, astilbin promoted the PPARγ/PTEN pathway to downregulate TNF-α and IFN-γ production of CD4⁺ T cells (Figure 3G).

Astilbin decreases lipid accumulation in high-fat-diet-induced obese mice (Perez-Najera et al., 2018) and inhibits proliferation and improves differentiation in HaCaT keratinocytes via activating AMPK (C. Zhang et al., 2017). The LKB1/AMPK/mTOR signaling pathway was also checked in astilbin-treated CD4⁺ T cells. Although no substantial changes of total LKB1, AMPK, and mTOR level were observed, phosphorylated LKB1 and AMPK were increased, and phosphorylated mTOR was decreased in CD4⁺ T cells (Figure 4A; Supplementary Figure S4A). Moreover, the increased expression of phosphorylated LKB1 and AMPK was observed given that CD4⁺ T cells were treated with astilbin for 2 h, accompanied by the decreased activation of mTOR (Figure 4B; Supplementary Figure S4B).

Given the metabolic role regulated by AMPK, key molecules associated with glycolipid metabolism in CD4⁺ T cells were further analyzed. Through astilbin treatment, glucose transporter of T cells (Glut1) decreased with no obvious changes of HK2, indicating decreased glucose intake. Meanwhile, ACC1 (a key enzyme in fatty acid synthesis) was decreased, while CPT1a and pACC1 representing fatty acid oxidation (FAO) were increased (Figure 4C; Supplementary Figure S4C). The CD36 receptor for the intake of fatty acid was also increased in a dose-dependent manner by astilbin treatment for 24 (Figures 4D,E) or 36 h (Supplementary Figure S4D). These data demonstrated that astilbin treatment induced metabolic reprogramming of CD4⁺ T cells, namely, the upregulation of lipid oxidation and downregulation of lipid synthesis and glucose catabolism.

PPARγ activation can transactivate CD36 expression (Marechal et al., 2018). When CD4⁺ T cells were cotreated with astilbin and GW9662, the upregulation of CD36 was remarkably inhibited (Figure 4F; Supplementary Figure S4E). Next, we determined whether decreased CPT1a and p-AMPK exerted inhibitory effects on CD4⁺ T cell function. When a CPT1a inhibitor (etomoxir) was added, IFN-γ production could be partially increased, but TNF-α downregulation could not be reversed at all. On the contrary, etomoxir synergized with astilbin to promote CD36 and CCR9 expression (Figure 4G; Supplementary Figure S4F). Simultaneously, etomoxir had no effects on PTEN expression (Figure 4I; Supplementary Figure S4H). Thus, the increased CPT1a expression in CD4⁺ T cells was only a stimulatory effect induced by astilbin, not a cause for the downregulation of CD4⁺ T cell function.

Unexpectedly, when an inhibitor of AMPK (dorsomorphin) was expected to block astilbin-induced activation of AMPK, decreased IFN-γ production could be reversed, whereas TNF-α production was more decreased (Figure 4H; Supplementary Figure S4G). Meanwhile, treatment with dorsomorphin had no effects on the increased expression of CD36 and CCR9 induced by astilbin (Figure 4H; Supplementary Figure S4G). Dorsomorphin even synergized astilbin to express PTEN (Figure 4J; Supplementary Figure S4I), which could explain the deeper inhibition of TNF-α production by cotreatment of astilbin and dorsomorphin. It could be inferred that although the expression of TNF-α, IFN-γ, CCR9, and CCR6 of CD4⁺ T cells could be regulated by metabolic intervention, functional changes are very different when using different small molecules with metabolic modulatory activity. Given that the CD36 and CCR9 expression of astilbin-treated CD4⁺T cells was dependent on PPARγ (Figures 2I, 4F), the expression of p-AMPK and CPT1a was inhibited in CD4⁺ T cells by the cotreatment of GW9662 (Figure 4K; Supplementary Figure S4J). Collectively, astilbin stimulated the PPARγ/AMPK axis to induce metabolic reprogramming and downregulate the activities of CD4⁺ T cells (Figure 4L).

The transcriptome profiles of CD4⁺ T cells were further compared before and after astilbin treatment. The heatmap of mRNA sequencing results is displayed in Supplementary Figure S5A. Based on KEGG pathway analysis, the top three pathways were in the immune system (108 genes), signal transduction (104 genes), and signaling molecules and interaction (76 genes) (Supplementary Figure S5B). The most changed biological processes were immune system process, inflammatory process, and leukocyte migration by gene ontology analysis. Of note, the NADPH oxidase complex and superoxide-generating NADPH oxidase activity were among the top changed pathways of cellular component and molecular function, respectively (Figures 5A,C).

When the numbers of upregulated or downregulated genes associated with molecular function were checked, multiple genes were upregulated, but almost no genes related to the functions of antioxidant activity and chemoattraction were downregulated (Figure 5B). As expected, the JAK-STAT3, MAPK, and TNF signaling pathways which were associated with T cell activation, showed substantial changes (Figure 5D; Supplementary Table S1).

In addition, transcriptions of PTEN, CPT1a, and microsomal glutathione S transferase 1 with antioxidase activity were upregulated in astilbin-treated CD4⁺ T cells (Figure 5E), while those of Tnfsf11, Mapk11, IL-12, Mapk12, IL-12b, IL-2, STAT3, and IL-6 were decreased (Figure 5F). These results confirmed the downregulated expression of genes involved in inflammation and indicated the upregulation of genes with antioxidase activity in astilbin-treated CD4⁺ T cells.

Given that the transcriptions of NADPH oxidase complex and enzymes with antioxidant activity were increased in astilbin-treated CD4⁺ T cells, cytoplasmic and mROS of CD4⁺ T cells were measured after astilbin treatment. cROS were increased at astilbin doses of 40 and 80 μg/ml and peaked at 160 μg/ml (Figure 6A) after 24-h treatment. cROS also displayed similar patterns of variation after 36-h treatment (Supplementary Figure S6A). Unexpectedly, no obvious changes of mROS and mitochondrial weight were observed after 24-h treatment (Figure 6A); even mildly decreased mROS levels were observed after 36-h treatment (Supplementary Figure S6A). Human CD4⁺ T cells of peripheral blood from four healthy individuals also showed increased cROS production after astilbin treatment (Figure 6B).

Furthermore, Cellular ROS in astilbin-treated CD4⁺ T cells could be depleted by NAC (2 mM) (Figure 6C; Supplementary Figure S6B). When astilbin-treated CD4⁺ T cells were treated with NAC, decreased IFN-γ production and increased CCR9 and CD36 expression were completely restored (Figure 6D; Supplementary Figure S6C,D). PPAR expression is strongly induced and activated by ROS-oxidized lipids (Marion-Letellier et al., 2016). When NAC was added into CD4⁺ T cells, astilbin-increased expression of p-PPARγ, PTEN, p-AMPK, and SOCS3 was completely reversed (Figure 6E; Supplementary Figure S6E), confirming that PPARγ activation in astilbin-treated CD4⁺ T cells was dependent on cellular ROS production. Of note, TNF-α secretion was only partially reversed (Figure 6D; Supplementary Figure S6C,D), indicating that other factors were involved in this astilbin-induced downregulation.

Next, we determined whether inhibiting PPARγ, p-AMPK, or CPT1a would impact ROS production in astilbin-treated CD4⁺ T cells. When GW9662 was used to inhibit PPARγ activation, the increased ROS level induced by astilbin did not have any changes in CD4⁺ T cells (Figure 6F; Supplementary Figure S6F), indicating that the ROS production of CD4⁺ T cells was the upstream event of PPARγ activation as treated by astilbin. When FAO of CD4⁺ T cells was inhibited by CPT1a inhibitor (etomoxir) or AMPK inhibitor (dorsomorphin), ROS production induced by astilbin decreased (Figure 6G; Supplementary Figure S6F), indicating that the upregulated FAO induced by astilbin had a positive feedback for cellular ROS production. Taken together, astilbin induced cROS production in CD4⁺ T cells to activate the PPARγ signaling pathway.

Finally, we aimed to investigate the molecular mechanisms of elevated ROS production in astilbin-treated CD4⁺ T cells. In general, cellular ROS is mainly produced by oxidative phosphorylation in mitochondria, cytochrome P450 (CYP) oxidases and endoplasmic reticulum oxidoreduction-1 in microsome, polyamine oxidases in cytoplasm, and cell membrane-coupled NADPH oxidases (Zhao et al., 2019). According to the integrated results predicted by the PharmMapper and UniProt databases, 33 candidate proteins of astilbin targets were determined, which are listed in Supplementary Table S2. Given that astilbin mainly induced cROS production of CD4⁺ T cells, we observed that two P450 enzymes (CYP1B1 and CYP19A1) were possibly docked by astilbin. Of note, the expression of CYP1B1 and CYP19A1 could be induced with astilbin (40 μg/ml), but downregulated (160 μg/ml) in CD4⁺ T cells (Figure 7A; Supplementary Figure S7A,B).

Next, ABT, a nonselective and irreversible inhibitor of CYP enzymes (Parrish et al., 2016), was used alone or in combination with astilbin to treat CD4⁺ T cells. As expected, ABT increased cROS production in CD4⁺ T cells. Compared with astilbin treatment (80 μg/ml) alone, a slightly lower level of cROS was observed in CD4⁺T cells treated by ABT (1 μM) alone. When low dose of ABT (0.3 or 1 μM) combined with astilbin (80 μg/ml) was used to treat CD4⁺ T cells, no enhancements of cROS were observed in the cytoplasm of CD4⁺ T cells compared with astilbin treatment (80 μg/ml) alone. However, cotreatment of high-dose ABT (2 μM) with astilbin (80 μg/ml) substantially increased cROS production (Figures 7B,C). This phenomenon indicated that astilbin promoted ROS production with similar molecular pattern to ABT.

As expected, ABT treatment (1 μM) alone reduced TNF-α and IFN-γ production to a lesser degree compared with the astilbin (80 μg/ml)-induced downregulation. In addition, there were obviously synergistic effects in TNF-α and IFN-γ production similar to cotreatment with astilbin (80 μg/ml) and ABT (1, 2 μM). Similar to astilbin, ABT (1 μM) alone increased CCR9 and CD36 expression. More increased CCR9 expression was only observed when ABT (2 μM) and astilbin (80 μg/ml) were both used. ABT and astilbin had no synergistic effects on CD36 expression, even if the ABT dose was 2 μM (Figure 7D; Supplementary Figure S7C). Thus, it was inferred that astilbin could use other molecular mechanisms to downregulate TNF-α and IFN-γ production besides ROS production; however, it induced CCR9 and CD36 expression dependent on ROS production.

The molecular docking study was conducted using AutoDock Vina 4.2.6. Binding affinity and presence of hydrogen or hydrophobic bonds between astilbin and CYP1B1 or CYP19A1 were predicted. Astilbin had higher binding affinity with CYP1B1 (−6.4 kcal/mol) than CYP19A1 (−3.3 kcal/mol). Then, we used computer-assisted modeling to further investigate whether astilbin interacted with CYP1B1 directly (Figure 7E). Three hydrogen bonds were formed in the astilbin–CYP1B1 complex of Glu229 and Gln340 residues (Figure 7F). In addition, six hydrophobic contacts contributed to the binding, namely, Pro195, Val198, Arg222, Arg194, Val198, and Lys512 (Supplementary Table S3). These interactions helped CYP1B1 and astilbin form stable complexes. Recombinant CYP1B1 and CYP19A1 proteins were used to detect the interaction with astilbin by SPR assay. CYP1B1 stably docked astilbin with a dissociation rate (K_D) of 18.2 (μM) (Figure 7G), but CYP19A1 could not bind astilbin efficiently (data not shown). When CYP1B1 was specifically knocked down by shRNA, the increase of cROS in astilbin-treated CD4⁺T cells disappeared with no variations of mROS production (Figure 7H; Supplementary Figure S7D). The decreased CYP1B1 expression in shRNA-transfected CD4⁺ T cells was confirmed in Supplementary Figure S7E,F. Consequently, the decrease of TNF-α production in CD4⁺ T cells induced by astilbin was reversed after being transfected with CYP1B1 shRNA (Figure 7I; Supplementary Figure S7G). Taken together, we identified that astilbin directly binds CYP1B1 to promote ROS production for the decreased secretion of inflammatory cytokines in CD4⁺ T cells.