The human acute myeloid leukemia cell lines Thp1 and HL60 were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). The cell lines were cultured under humidified conditions at 37°C with 95% air and 5% CO₂ in RPMI-1640 medium supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin. The experimental reagents used included: RPMI-1640 with FBS (Gibco, Grand Island, United States); anti-rabbit IgG-HRP and anti-mouse IgG-HRP (Shanghai, China); CCK8 assay kit (APE BIO, Houston, TX, United States); apoptosis detection kit (BD Biosciences, San Diego, CA, United States); Bradford assay kit (Shanghai, China); GAPDH and β-actin antibody (Proteintech, Wuhan, China), and antibodies against cleaved PARP (CST, Danvers, MA, United States). BALB/c-nu/nu female mice (4–5 weeks old) were obtained from Huafukang Biotechnology Company, Ltd. The 13 sweet potato varieties tested in this study are listed in Table 1 and sourced from the Chengdu Institute of Biology.

The name of the sweet potato cultivars, comprising alphanumeric characters, has been officially approved in China and subsequently translated from Chinese.

The SWFs were obtained following the methods outlined by Guo et al. (15). In brief, the edible parts of sweet potato (leaves, stem, peel, and flesh) were subjected to freeze-drying and subsequently ground into powders (100 mesh). One hundred grams of this powder were then mixed with 2 L of 70% methanol (1:20, w/v), the mixture was placed in an ultrasonic extractor at 40°C with an ultrasonic frequency of 40 kHz for 30 min. This step was repeated for three times. After extraction, the mixture was centrifuged at a speed of 12,000 rpm. The supernatant was filtered through a 0.22 μm PTFE membrane syringe filter, and injected into an HPLC system (AS3000 Autosampler, Thermo Electron Corp., United States) equipped with a UV6000 DAD detector (United States). Rutin was employed as the reference standard (16).

According to Chen et al. (17), leaves were freeze-dried and ground into powder. Then, 50 mg of the powder was dissolved in 1.2 mL of 70% methanol by vortex oscillation for 30 s every 30 min, repeated six times. Following centrifugation at 12,000 rpm for 3 min, the supernatant was filtered using a microporous filter membrane (0.22 μm, SCAA-104; ANPEL, Shanghai, China). Subsequently, a 4 μL aliquot of the sample was analyzed using an ExionLC^™ AD UPLC-MS/MS system (Applied Biosystems 4500 QTRAP), employing the refined methodology established by Fraga et al. (18). The analysis was conducted using an Agilent SB-C18 chromatographic column (1.8 μm, 2.1 mm × 100 mm) at 40°C. The mobile phase (0.35 mL/min) consisted of pure water containing 0.5% glacial acetic acid (A) and acetonitrile containing 0.1% formic acid (B). From 0 to 9 min, the proportion of phase B was linearly increased from 5 to 95%, held at 95% for 1 min, then reduced to 5% within 1.1 min, and maintained for 2.9 min. The UPLC system was connected to an ESI-QTRAP-MS, with an ionization source voltage of 5,500 V in positive mode and −4,500 V in negative mode, both at a temperature of 550°C. GSI, GSII, and CUR were set to 50, 60, and 25 psi, respectively; CAD was set to high. The QQQ scan was conducted as an MRM experiment, with the collision gas (nitrogen) set to medium. Data processing methods were based on the research by Eriksson et al. (19).

Qualitative analysis of primary and secondary MS data was conducted by comparing precursor ions (Q1), fragment ions (Q3) values, isolation windows (15 Da), dwell time (ms), cycle time (1 s), retention times (RT), and fragmentation patterns with those obtained from standard injections under identical conditions when standards were available (Sigma-Aldrich, United States; http://www.sigmaaldrich.com/united-states.html). If standards were unavailable, comparisons were made using a self-compiled database MWDB (MetWare Biological Science and Technology Co., Ltd., Wuhan, China) and publicly accessible metabolite databases. Repeated signals from K⁺, Na⁺, NH₄⁺, and other high molecular weight substances were filtered out during the identification process. Quantitative analysis of metabolites was performed in MRM mode. Characteristic ions for each metabolite were identified via QQQ mass spectrometry to measure signal intensities. Chromatographic peak integration and correction were carried out using MultiQuant version 3.0.2 (AB SCIEX, Concord, Ontario, Canada). Relative metabolite concentrations were expressed as chromatographic peak area integrals (20).

The cell counting kit-8 (CCK-8) (Beyotime, Shanghai, China) was used to assess cell viability, as described by Huang et al. (21). Thp1 and HL60 cells (2.0 × 10⁴ cells/well) were inoculated into a 96-well plate. Then, different concentrations of analytes were added and incubated for 24 or 48 h at 37°C. Subsequently, 10 μL of CCK-8 solution was added to each well for another 4 h incubation at 37°C under 5% CO₂. The absorbance was measured at 450 nm using a microplate reader, and the IC₅₀ values of analytes were calculated using SPSS version 21.0 (22).

The cells were seeded at a density of 5 × 10⁵ cells/well in a 6-well plate and subjected to analyte treatment with final concentrations of 0, 20, 40, and 60 μM for a duration of 24 h. Subsequently, the treated cells were harvested, washed twice with pre-cooled PBS buffer, and resuspended in 500 μL of Annexin V binding buffer. Then, the buffer was supplemented with Annexin V-FITC (5 μL) and propidium iodide (5 μL). Following 15 min incubation in darkness, flow cytometry analysis (BD FACSVerse flow cytometer, San Diego, CA) was conducted to evaluate the extent of apoptosis (23).

The treated cells were harvested, washed with PBS, and lysed using an ultrasonic cell disruptor. After centrifugation, the resulting supernatant was collected as protein extract. Protein concentration was determined by BCA assay kit and equal amounts of protein were loaded onto SDS-PAGE gel for separation and transferred onto a PVDF membrane. The membrane was blocked with 7.5% skim milk at room temperature for 1 h before incubating overnight at 4°C with specific primary antibodies. After washing with PBST, HRP-conjugated secondary antibodies were added and incubated at room temperature for 1 h. Protein bands were visualized via a touch imaging system (E-Blot Biotechnology Co., Ltd., Shanghai, China). Band intensities were quantified and normalized to a loading control such as GAPDH or β-actin (24).

The animal experiments were conducted in accordance with the guidelines approved by the Ethics Committee of Southwest Medical University (Protocol Number: 20211019-002). AML xenograft models were established using female BALB/c-nu/nu mice (4–6 weeks old). Mice were intraperitoneally (i.p.) treated with 2 mg of cyclophosphamide (CTX) for two consecutive days to induce immune suppression. After CTX treatment, 5 million HL60 cells were subcutaneously injected into the hind limb of each mouse. Tumor development was closely monitored, and once the subcutaneous tumor volume reached approximately 100 mm³, the mice are randomly divided into two groups (5 mice per group). The control group was treated with physiological saline (100 μL/mouse) by oral gavage daily, while the treatment group received 50 mg/kg cynaroside by oral gavage daily. Tumor volume and body weight are measured every 3 days to evaluate treatment efficacy and overall health status. When the largest tumor diameter approached 2 cm, the experiment is terminated. Tumors were weighed and imaged, and tumor volume was calculated using the following formula:

One-way ANOVA and unpaired t-tests were performed using SPSS software (version 21.0) and GraphPad Prism (version 8.3.0.538) to statistically analyze the differences between groups. A p-value <0.05 was considered statistically significant.

The flavonoids content in the stem, leaves, flesh, and peel of 13 sweet potato cultivars was assessed. As presented in Table 2, the leaves exhibited the highest flavonoid content across all cultivars, followed by the stem and then the peel. Conversely, the flesh displayed the lowest flavonoids content in each cultivar. Notably, the highest flavonoid content was found in the leaves of Nanshu 017, reaching 2270.7 mg rutin/100 g dry weight (DW). Therefore, the leaves extract of Nanshu 017 was selected for both anti-tumor activity assay and component analysis.

We subsequently assessed the inhibitory efficacy of SWFs on the viability of AML cell lines. As depicted in Figures 1A,B, the SWFs exhibited significant growth inhibition against both AML cell types. The extract demonstrated a concentration-dependent inhibitory effect on HL60 and Thp1 AML cell lines. At 24 h, the IC₅₀ values for HL60 cells were determined to be 183.86 ± 14.26 mg/mL, while at 48 h they were found to be 170.75 ± 19.45 mg/mL (Figure 1C); for Thp1 cells, the corresponding IC₅₀ values were measured as 372.16 ± 30.88 mg/mL at 24 h and as 281.22 ± 30.19 mg/mL at 48 h (Figure 1C), respectively. These findings provide preliminary evidence supporting the anti-leukemic activity of SWFs.

To elucidate the specific flavonoid constituents responsible for the anti-leukemia activity in sweet potato extract (SPE), a metabolomics analysis was conducted. The results revealed the detection of 343 metabolites, categorized into nine distinct groups of flavonoids, in sweet potato leaves (Supplementary Table S1). Among them, flavonols exhibited the highest species diversity, comprising a total of 138 species. This was followed by flavones with 107 types and flavanones with 37 species. Subsequently, in descending order, were chalcones, flavanols, flavanonols, anthocyanins, tannin, and proanthocyanidins with 16, 13, 9, 8, 8, and 7 types, respectively (Figure 2A). As depicted in Figure 2B, a total of 32 components had a relative content greater than 1% among the 343 metabolites, 19 of which were classified as flavonols while 10 fell under the category of flavones. The top flavonol compound was identified as 6-hydroxykaempferol-7,6-O-diglucoside, with a relative content of 3.83%, while luteolin-7-O-gentiobioside (4.38%) emerged as the highest component among flavones (Figure 2C). Based on the types, relative content, and for which monomers can be readily obtained, nine components were selected for cell viability testing, including cynaroside (3.08%), lonicerin (2.44%), kaempferol-3-O-neohesperidoside (2.31%), nepitrin (2.02%), brassicin (1.22%), hesperetin-7-O-glucoside (1.25%), eriodictyol-7-O-glucoside (0.72%), taxifolin-3′-O-glucoside (0.22%), and yuanhuanin (0.16%).

The AML cell lines HL60 and Thp1 were exposed to various concentrations of the nine monomeric analytes for a duration of 24 h. Among these monomers, cynaroside, nepitrin, and yuanhuanin demonstrated significant dose-dependent inhibition on the growth of both cell lines (Figures 3A–C). The IC₅₀ values for cynaroside were determined as 31.77 and 18.46 μM for HL60 and Thp1 cells, respectively; while for nepitrin, they were found to be 39.93 and 41.93 μM; finally, yuanhuanin exhibited IC₅₀ values of 35.48 and 39.91 μM on HL60 and Thp1 cells, respectively. However, the other monomers, excluding hesperetin-7-O-glucoside, exhibited hardly inhibitory effect on these two cells (Figures 3D–I). At the concentration below 300 μM, kaempferol-3-O-neohesperidoside, to some extent, even enhanced the viability of AML cells. Therefore, we selected cynaroside, nepitrin, and yuanhuanin for further analysis. Based on these IC₅₀ values, the concentrations of 0, 20, 40, 60 μM for the three analytes were used in the subsequent experiments.

Firstly, a total of 288 targets of cynaroside, nepitrin, and yuanhuanin were retrieved from SwissTargetPrediction and SuperPred using “Probability >0” as the criterion (Figure 4A). Subsequently, 15,913, 130, and 32 targets for AML were obtained through the GeneCards, OMIM, and PharmGkb databases, respectively. Furthermore, after filtration with a relevance score (>10) and duplicates removal, a total of 1,656 AML-related targets were obtained (Figure 4B). Finally, by intersecting the drugs’ target genes with the AML-related target genes, we identified 119 potential target genes for the drugs (Figure 4C).

The CytoNCA plugin was used to analyze the protein–protein interaction (PPI) network topology. Core genes TP53, TNF, EGFR, HIF1A, CASP3, HSP90AA1, BCL2, SRC, MMP9, KDR, NFKB1, PTGS2, HSP90AB1, MTOR, IL2, GSK3B, CXCR4, TLR4, ITGB1, and STAT1 were identified as shown in Figures 4D–G. The Degree values of the core genes are presented in Table 3. Additionally, EGFR, CASP3, SRC, and KDR among the core targets have been reported to have high expression levels in AML with plasma concentrations of 88 μg/L, 94 ng/L, 3.2 μg/L, and 11 μg/L, respectively (Supplementary Figure S1). These findings suggested that these four targets may serve as significant biomarkers for AML, therefore, we selected these four core genes for subsequent molecular docking validation.

The biological targets of SWFs in AML treatment were explored through GO and KEGG enrichment analyses. The GO analysis revealed critical biological processes, including the negative regulation of apoptotic processes, phosphorylation, and signal transduction, which play pivotal roles in controlling cell survival and proliferation in cancer. Moreover, the results of cellular components (CC) analysis indicated significant involvement of the cytoplasm, plasma membrane, nucleus, and extracellular exosomes. This suggests that these flavonoids target pathways affecting both intracellular and intercellular communication. Additionally, molecular function (MF) analysis highlighted essential activities such as ATP binding, protein kinase binding, and protein homodimerization that are crucial for modulating key enzymatic activities linked to cancer progression. Furthermore, KEGG pathway enrichment supported these observations by demonstrating that the drug target genes are predominantly enriched in cancer-related pathways, particularly the PI3K-Akt, MAPK, and Ras signaling pathways, which are well-known for regulating cell growth, survival, and apoptosis (see Figure 5).

Molecular docking was used to further confirm the interaction between these three compounds and the core targets. The docking results showed that the binding energy of each target with the active components was less than −5.0 kcal/mol, indicating that there is good binding activity between the receptor proteins and the ligand small molecules. The results were listed in Table 4 and Figure 6.

Based on the results of both network pharmacology and bioinformatics analyses, we hypothesized that the active components of SWFs possess the potential to induce apoptosis in AML cells. Therefore, flow cytometry was employed to quantify apoptosis, revealing a significant increase in apoptotic cells following treatment with these three analytes. Notably, as analyte concentrations increased, there was a gradual elevation in apoptotic rates (Figure 7). The highest apoptotic rates were 68.04, 37.72, and 50.20% for HL60 cells treated with 60 μM cynaroside, nepitrin, and yuanhuanin, respectively; while Thp1 cells exhibited the highest apoptotic rates of 73.45, 53.94, and 55.39% for the same treatments, respectively. Furthermore, Western blot analysis demonstrated an upregulation in cleaved PARP protein levels post-drug treatment, confirming their ability to induce apoptosis in HL60 and Thp1 cells (Supplementary Figure S3).

The growth of HL60-xenograft tumors was significantly inhibited by cynaroside, as demonstrated in Figures 8A,B. Notably, the treatment group exhibited noticeably smaller tumor sizes compared to the control group (Figure 8C). Furthermore, at the end of the experiment, tumor weights were measured as 0.82 ± 0.27 g in the treatment group and 1.99 ± 0.45 g in the control group, resulting in an inhibition rate of 58.79% (Figure 8D). These findings indicate that a dosage of 50 mg/kg cynaroside effectively suppresses HL60 cell growth in vivo while having minimal impact on mouse weight (Figure 8E).