Flavonoid profiling and antioxidant activity of Passiflora setacea DC. leaves

The wild species Passiflora setacea DC., known as ‘maracujá-do-sono’ or ‘maracujá-sururuca’, has chemical and pharmacological properties that are important to the food and pharmaceutical industries. This study aimed to identify and quantify polyphenols and flavonoids in extracts from leaves of in vivo orchard-grown and greenhouse-derived P. setacea plants. The extracts were also evaluated for their antioxidant (AAO) and acetylcholinesterase (AChE) inhibitory activity. The polyphenols and flavonoids were quantified spectrophotometrically. Based on the results, the hydroalcoholic extract from the leaves of P. setacea (PSHF-01) plants grown in the orchard was selected for further studies. In this extract, the flavonoids were characterized and identified using spectrophotometric and chromatographic methods. In addition, its antioxidant potential was evaluated through DPPH free radical scavenging, the ß-carotene/linoleic acid lipid peroxidation method, and the liposome peroxidation assay (TBARS). Furthermore, in silico activity prediction for flavonoids from P. setacea was performed using the Active-IT system. The total polyphenol content of the extracts varied from 11.32 ± 0.33 to 40.59 ± 1.5 mg of gallic acid equivalents per gram of extract. For flavonoid quantification, the amounts ranged from 3.79 ± 1.10 to 33.65 ± 4.73 mg of total flavonoids per gram of extract, calculated as apigenin. The extracts were considered promising for antioxidant activity using the β-carotene/linoleic acid peroxidation method. Concerning AChE inhibition, the extracts did not show significant values (%I > 40%). In the selected extract, PSHF-01, isovitexin and isoorientin were identified by chromatographic techniques (TLC, HPLC-UV and UPLC-PDA-ESI-MS/MS). Furthermore, the compounds luteolin-(O -glucopyranosyl)- C -glucopyranoside, (iso)orientin-C-xyloside, vicenin II, (iso)vitexin-2”-O-xyloside, and (iso)orientin-C-rhamnoside were tentatively characterized by UPLC-PDA-ESI-MS/MS. HPLC-DAD method was developed and validated for the determination of isovitexin in PSHF-01. Antioxidant activity was detected only in the ß-carotene/linoleic acid assay, whereas no activity was observed in the DPPH or TBARS assays. These differences may be attributed to distinct mechanisms underlying each assay. In silico predictions indicated that the evaluated flavonoids have the potential to exhibit hepatic metabolic stability and antiparasitic, anti-inflammatory, antitumor, and immunomodulatory activities. These results enhance our understanding of the chemical composition and biological properties of P. setacea and open new avenues for future research that could support its pharmaceutical potential and agronomic applications.

Passiflora species have been largely used by populations of different countries (Almeida et al., 2021), and some are used for therapeutic purposes due to their anxiolytic, sedative, and antidepressant properties (Duarte et al., 2020). Furthermore, there are species of economic value for the ornamental plant market due to their exotic flowers, the capacity for flowering throughout the year, and foliage (Mikovski et al., 2019), and for culinary purposes, resulting in a wide network of use, cultivation, and commercialization (Faleiro et al., 2017; Amaral, 2018; Rocha et al., 2020; Almeida et al., 2021).

Within the Passifloraceae family, Brazil is an important center of genetic diversity for Passiflora species. Passiflora setacea DC. is a Brazilian endemic wild species, popularly known as ‘maracujá-do-sono’, ‘maracujá-sururuca’, ‘maracujá-do-Cerrado’, ‘maracujá-pérola’, ‘maracujá-cascavel’, and ‘maracujá-de-veado’ (Almeida et al., 2021). It is found in both the Cerrado and Caatinga biomes, as well as in transitional areas between the Caatinga and the Cerrado (Braga et al., 2016). In Brazil, its use has been associated with sleep modulation by the population (Carvalho et al., 2018; Duarte et al., 2020).

Given the medicinal and agronomic relevance of P. setacea (Figure 1A), research has investigated its properties. In terms of agronomic traits, the species exhibits drought tolerance, and previous studies have reported tolerance to specific pests and diseases, resistance to premature plant death, Fusarium infection, and aerial diseases such as anthracnose (Ataíde et al., 2012; Braga et al., 2016; Muniz et al., 2022).

The potential of P. setacea pulp for anti-inflammatory and anti-diabetic effects has been described (Duarte, 2015; Duarte et al., 2020; Duarte et al., 2022). According to studies by Vetore-Neto (2015), the hydroethanolic extract from its leaves provides significant protection against the formation of ulcerative lesions in an acute ulceration model in rats. In the same work, the hydroethanolic extract was neither mutagenic in the Ames test nor toxic in acute and sub-acute in vivo assays. Furthermore, the local population has associated the use of this species with sleep modulation in Brazil (Carvalho et al., 2018; Duarte et al., 2020).

Despite the medicinal and agronomic importance of P. setacea, there is little information regarding phytochemical studies of this species, although some secondary metabolites have been reported. The Passiflora species and its flavonoids are important natural products with a broad spectrum of biological activities and nutraceutical properties. Flavonoids are the main secondary metabolites found in Passiflora, although alkaloids, saponins, and cyanogenic compounds have also been reported (Almeida et al., 2021). The flavonoids have been associated with the pharmacological effects observed in Passiflora species and are considered chemical markers of these species (Wosch et al., 2017; Almeida et al., 2021; Santos-Tierno et al., 2022). The main flavonoids identified are C-glycosides of apigenin and luteolin (Almeida et al., 2021).

In P. setacea, the presence of the flavonoids (Figure 2) vicenin II (Apigenin-6,8-di- C -β-D-glucopyranoside), schaftoside (apigenin-6- C -glucoside-8- C -riboside), isovitexin (apigenin-6- C -glucoside), and isoorientin (luteolin-6- C -glucoside) has been reported in hydroethanolic leaf extracts from in vivo and in vitro plants (Wosch et al., 2017; Santos-Tierno et al., 2022). In the latter, the flavonoids luteolin-7- O -dirhamnoside-3′- O -glucoside, apigenin-6- C -glucoside-7-O-glucoside, and apigenin-8- C -glucoside-7- O -glucoside have also been reported (Santos-Tierno et al., 2022).

In general, environmental factors such as seasonality, temperature, UV radiation, and other abiotic stresses can modulate the production, accumulation, and distribution of secondary metabolites in plants. The plant age and developmental stage contribute to variations in phytochemical composition and concentration (Gobbo-Neto and Lopes, 2007; Mierziak et al., 2014). Flavonoid biosynthesis is particularly responsive to environmental stress conditions (Petrussa et al., 2013). These compounds accumulate predominantly in epidermal and subepidermal tissues, where they function as UV filters and free radical scavengers, thereby protecting plant cells from oxidative damage induced by environmental stressors (Gobbo-Neto and Lopes, 2007). Furthermore, flavonoids play an important role in cold acclimation and freezing tolerance, as their synthesis is often enhanced under these conditions (Petrussa et al., 2013). This increased production highlights their contribution to plant adaptation under adverse environmental conditions.

Carvalho et al. (2018) reported seasonal variations in the phytochemical composition of P. setacea pulp under different water conditions. They observed higher flavonoid concentrations during the rainy season, suggesting that increased rainfall may constitute a stress condition for P. setacea. However, the effects of this factor have not yet been fully elucidated by these authors.

Several factors, such as the plant material and sample pre-treatment, the type of solvent, and the extraction technique, influence the efficiency of the flavonoid extraction procedure. Sample preparation generally involves two or sometimes three steps: drying, grinding, and defatting. The solvent used for extracting flavonoids from plant material is generally methanol, ethanol, acetonitrile, acetone, or mixtures of these with water. The choice of extraction technique depends on the stability of the compounds, the selected solvent, and the available equipment (Tzanova et al., 2020).

Investigating the biological activity of phytochemicals, such as flavonoids, to identify new potential pharmacological uses through in vitro and in vivo tests requires significant time and financial resources. In this context, in silico evaluation tools are valuable for discovering and developing treatments, reducing costs, and expediting the process. Virtual screening is a key technique in drug discovery for selecting compounds with potential activity. Active-IT is an advanced, in-house approach that utilizes ligand-based virtual screening, comprising four modules: molecular descriptor generation, machine-learning modeling, a database of predictive models, and a prediction module for new compounds. This system has become a valuable tool in drug discovery and development in recent years. Within our research team, it has been used to predict new activities for various secondary metabolites, including lignans (Rocha M. P. et al., 2018), alkaloids (Rocha M. P. et al., 2018), triterpenes (Oliveira et al., 2025), and coumarins (Sudan et al., 2021).

The present study aimed to quantify polyphenols and flavonoids in extracts from leaves of in vivo orchard-grown and greenhouse-grown P. setacea plants. The extracts were also evaluated for their antioxidant (AAO) and acetylcholinesterase (AChE) inhibitory activity. Based on the obtained results, the hydroalcoholic extract from the leaves of P. setacea (PSHF-01) plants grown in the orchard was selected for further studies. These included an evaluation of the antioxidant activity using other methods as well as the identification and quantification of flavonoids by chromatographic techniques. In addition, in silico predictions of the biological activities of flavonoids characterized in the extract of P. setacea were performed using the Active-IT system.

Plant material and growth conditions

Leaves and stems of Passiflora setacea DC. were harvested from one-year-old plants cultivated in an open-pollinated orchard (PS-01) at the ‘Universidade Federal de Viçosa’ (UFV, Brasil), as well as from in vivo-grown plants in the greenhouse (PSF-02 and PSF-03). Seeds of P. setacea cv. BRS Pérola do Cerrado were supplied by Embrapa Cerrados (Planaltina, DF, Brasil) and grown in pots measuring 34 × 25 × 25 cm filled with Plantmax® substrate, in a climate-controlled greenhouse under an average temperature of 28 °C, average daytime and nighttime CO 2 concentrations of 324 ppm, and a relative humidity of 65%, on the UFV campus (elevation of 680 m; 20°45′29.3″S, 42°52′13.0″W). Leaf samples from the third, fourth, and fifth fully expanded leaves at the shoot apex were collected from plants that had been growing for 7 months (Figure 1B). The samples were immediately frozen in liquid nitrogen and then freeze-dried (SP Scientific VirTis Advantage Plus EL-85 Benchtop Bulk Tray Freeze Dryer).

The plant material was processed at the ‘Serviço de Prospecção Farmacêutica, Fundação Ezequiel Dias’ (Brasil), where the leaves and stems were separated manually and dried in a ventilated oven at 40 °C. After drying, the material was pulverized in a knife mill.

This study received authorization from the SISGEN (Registration number A2B8959) for access to and the remittance of genetic material for scientific research.

Preparation of the ethanolic, aqueous, and hydroalcoholic extracts of leaves

Three different preparation methods were used: ethanol/sonication, water/infusion, and ethanol:water percolation. The extracts were prepared and stored in a refrigerator until use.

Powdered leaf material (10.0 g) was extracted with 300 mL ethanol 98°GL for 10 min by sonication in a 40 kHz ultrasonic bath at room temperature. The extracts were vacuum-filtered, and the solvent was removed under reduced pressure on a rotary evaporator at 40 °C to yield crude ethanolic extracts (PSEF-01, PSEF-02F). The procedure was repeated twice using the same material, and the combined extracts were dried and stored at 4 °C. The yields (%, w/w) of the ethanolic extracts were recorded (Table 1).

The aqueous extracts of Passiflora samples (PSAF-01 and PSAF-02) were obtained by infusion. Boiled water (150 mL) was poured onto each sample (approximately 5.0 g), which was soaked, covered, and left at room temperature for 30 min. Aqueous extracts were vacuum-filtered, and the procedure was repeated 3 times using the same powder. The combined extracts were frozen at −20 °C for later freeze-drying. The individual yields (%, w/w) were recorded (Table 1).

The hydroethanolic extract was prepared from 60.0 g of pulverized leaf material by exhaustive percolation with a 50% ethanol:water solvent system. The plant material was placed in a percolator, and extraction was performed at room temperature (approximately 25 °C) for 26 days. During the initial 3 days, 1.0 L of fresh solvent mixture was added daily. From day 4 to day 25, extraction was continued using the solvent obtained by distilling the collected percolate on a rotatory evaporator, after alcoholate adjustment in a refractometer. The total daily volume of solvent was approximately 2.0 L. In total, 48 L of the solvent mixture was used throughout the extraction procedure. After completion of the extraction, the combined extracts (PSHF-01, PSHF-03) were partially concentrated under reduced pressure at 60 °C to remove ethanol. The remaining aqueous fraction was freeze-dried at −80 °C to ensure extract stability. The final lyophilized extracts were stored at 4 °C, and the yields (%, w/w) were recorded (Table 1). It was not possible to obtain the hydroethanolic extract of sample PSHF-02 due to the small amount of leaves harvested from greenhouse-grown plants.

Determination of total phenolic content

The total phenolic content (TPC) was determined according to Sartori et al. (2014) with adaptations. The calibration curve was prepared using pyrogallol (0.25–0.6 μg/mL). The absorbance of the mixture was measured at 760 nm using a UV–Vis spectrophotometer (Multiskan GO Thermo Scientific). Methanol was used as a blank. The extract solution (30 μg/mL) was evaluated using the same methodology. The assay was done in triplicate. The total phenolic content was expressed as mg of gallic acid equivalent per gram of dry plant extract (PE/g) by comparison with the standard curve: A = 0.094C–0.0528 (R 2 = 0.9991), where A is the absorbance at 760 nm, and C is the concentration (µg/mL).

Determination of total flavonoid content

The total flavonoid content (TFC) was determined using the spectrophotometric method, as described by Nurcholis et al. (2021). The calibration curve was plotted using quercetin in 50% methanol (20–100 μg/mL) as the standard. The absorbance was read at 420 nm with a UV-Vis spectrophotometer (Multiskan GO, Thermo Scientific). Methanol was used as a blank. The extracts were diluted in methanol to 3 mg/mL and evaluated using the same methodology. The assay was done in triplicate. The results were expressed in mg quercetin per g dry extract (QE/g), by comparison with the standard curve A = 40.86C - 0.02 (R 2 = 0.9983), where A is the absorbance at 420 nm, and C is the concentration (mg/mL).

Acetylcholinesterase inhibitory activity assay

The AChE inhibitory activity was quantitatively assessed using a colorimetric microplate assay based on the Ellman method (Ellman et al., 1961), with modifications. The test sample (400 μg/mL) was prepared in the Tris/HCl buffer (50 mM, pH 8.0). AChE (2,000 °U/mL, Sigma-Aldrich, USA) was diluted in 0.1% BSA-Tris/HCl buffer and stored at −20 °C. The assay was performed in a 96-well microplate with 25 μL of the test sample, 50 μL of 0.1% BSA-Tris/HCl buffer, 125 μL of DTNB (3 mM), and 25 μL of ATCI (14 mM). A blank was prepared by replacing the sample with its diluent. The reaction was initiated by adding 25 μL of AChE (0.2°U/mL), followed by incubation at 37 °C. Absorbance was measured at 405 nm every 5 min for 25 min using a Multiskan GO microplate reader (Thermo Fisher Scientific, USA). Physostigmine (Eserine, Sigma, St. Louis, MO, USA) was used as the positive control. The assay was performed in triplicate, and % of enzyme inhibition (% I) was calculated according to Equation 1: where a is the ΔA/min of control; b = ΔA/min of test sample; and ΔA = change in absorbance.

The assay was considered significant for samples with %I higher than 40% compared to physostigmine. Data were expressed as mean ± SD and analyzed using Microsoft Excel.

Evaluation of antioxidant activity of the extracts

Lipid peroxidation inhibition by β-carotene/linoleic acid

The antioxidant activity of the extracts was evaluated using the β-carotene/linoleic acid bleaching assay, adapted from Duarte-Almeida et al. (2006). An emulsion was prepared by mixing 25 mg of linoleic acid, 100 mg of Tween 40, and 1 mL of β-carotene solution in chloroform (1 mg/mL), then evaporating the solvent under reduced pressure and adding 50 mL of aerated distilled water with vigorous agitation. The samples were solubilized in methanol using sonication and serially diluted to obtain final concentrations ranging from 3.125 to 200 μg/mL. Quercetin (20 μg/mL; Sigma-Aldrich, St. Louis, MO, United States) and methanol were used as positive and negative controls, respectively. The reaction mixtures were incubated at 45 °C, and absorbance was recorded at 470 nm every 15 min for 120 min using a microplate reader (EL808IU, BioTek, Winooski, VT, United States). Antioxidant activity was expressed as IC 50 values. The assays were performed in triplicate, and results were presented as mean ± standard deviation (SD). IC 50 values were determined by nonlinear regression analysis using GraphPad Prism version 6.0. All samples were evaluated under the same experimental conditions and were considered promising when their IC 50 values were below 60 μg/mL.

2,2-Diphenyl-1-picrylhydrazyl radical (DPPH.)- scavenging activity

The antioxidant activity of the extract (PSHF-01) was evaluated using the DPPH (2,2-diphenyl-1-picrylhydrazyl) free radical-scavenging assay, adapted from the method of Brand-Williams et al. (1995). The sample was prepared by dissolving 1.5 mg of PSHF-01 in methanol by sonication, then quantitatively transferring it to a volumetric flask to prepare a stock solution. This solution was then serially diluted in a 96-well microplate to obtain a concentration range from 180 μg/mL to 8.7 μg/mL. The DPPH reagent was prepared at a final concentration of 120 μg/mL in methanol and stored in an amber volumetric flask to prevent photodegradation. Methanol served as the blank, while a pyrogallol solution (50 μg/mL) was utilized as the positive control. A calibration curve was also established to monitor the radical’s absorbance behavior under the test conditions. Absorbance readings for both samples and the calibration curve were recorded at lambda = 515 nm over a 50-min period at 25 °C, with measurements taken every 5 min. The antioxidant potential was determined by calculating the IC 50 (half-maximal inhibitory concentration), which represents the minimum concentration required to inhibit 50% of the initial radical concentration. According to the established criteria, the extract is classified as active if the IC 50 is <250 μg/mL and as promising if the IC 50 is <50 μg/mL. The assays were performed in triplicate.

Lipid peroxidation inhibition TBARS assay

Lipid peroxidation was assessed using an adapted methodology from Buritis and Bucar (2000). The extract (PSHF-01) was dissolved in HPLC-grade methanol at concentrations ranging from 0.0625 to 2 mg/mL. Bovine brain extract (type VII) liposomes in phosphate-buffered saline (PBS) were incubated with FeCl 3 (1 mM) and ascorbic acid (1 mM) for 20 min at 37 °C. The reaction was halted by adding 2% BHT, 1% thiobarbituric acid (TBA), and 1% HCl, followed by incubation in a water bath (85 °C–90 °C) for 30 min. To eliminate turbidity, the chromophore was extracted with n-butanol, vortexed, and centrifuged at 3,500 rpm for 10 min. Absorbance of the organic phase was measured at 532 nm in triplicate. The results were expressed as the effective concentration required to inhibit 50% of lipid peroxidation (EC 50), with samples classified as active (EC 50 < 200 μg/mL) or promising (EC 50 < 50 μg/mL).

Chemical profiling and flavonoid quantification in PSHF-01

Preliminary phytochemical screening (TLC) of flavonoids

Samples were prepared by dissolving 10 mg of PSHF-01 and 1.0 mg of each flavonoid standard (isovitexin, isoorientin, vitexin, orientin, apigenin, kaempferol, and rutin) in a 7:3 acetonitrile:water mixture, followed by 10 min of sonication to ensure complete solubilization. Flavonoid identification was conducted by Thin-Layer Chromatography (TLC) using silica gel F 254 plates, following an adapted protocol by Wagner and Bladt (2001). The stationary phase was eluted with a mobile phase of ethyl acetate, glacial acetic acid, formic acid, and water (100:11:11:27). Chromatographic profiles were visualized under UV light at 365 nm after using NP/PEG spray reagents.

Flavonoid identification using HPLC-UV

Sample preparation for high-performance liquid chromatography (HPLC) involved dissolving PSHF-01 in a (7:3) acetonitrile:water mixture. Chromatographic separation was performed on an Agilent 1100 equipped with a UV/vis detector system using an ODS Hypersil column (250 × 4 mm, 5 mm). The mobile phase consisted of water with 0.2% formic acid (Phase A) and acetonitrile with 0.1% formic acid (Phase B), employing a linear gradient from 10% to 90% B over 40 min. The analysis was conducted at a flow rate of 0.8 mL/min, with detection at 340 nm and an injection volume of 10 uL. Flavonoid identification was achieved by comparing the retention times of the extract with those of isolated standards (isovitexin, 1.0 mg/mL). To confirm these results, co-injections of the P. setacea sample (2.0 mg/mL) spiked with the standards were performed. This methodology, adapted from the Brazilian Pharmacopeia (Anvisa, 2019), enabled precise correlation of chromatographic peaks with their corresponding phenolic compounds.

HPLC-DAD analytical method validation and isovitexin quantification in PSHF-01

The analysis was performed on an Agilent Technologies 1200 series HPLC system equipped with a diode-array detector and an Agilent data handling system (Agilent Technologies, United States). Isovitexin standard (HWI pharma services GmbH, Germany) was solubilized in acetonitrile:water (1:1). The PSHF-01 sample was solubilized in HPLC-grade acetonitrile:water (1:1), centrifuged at 10,000 rpm for 10 min (Eppendorf, model 5415D), and filtered through 0.45 µm PTFE syringe filters.

The separation was achieved on a reversed-phase XDB-C18 column (4.6 × 250 mm, 5 μm), with a flow rate of 0.8 mL/min, detection at ƛ 340 nm, and an injection volume of 20 µL. The elution was carried out using water with 0.2% formic acid (Phase A) and acetonitrile with 0.1% formic acid (Phase B) in a linear gradient elution: 0–40 min 90% B, 40–45 min 90%–95% B.

System suitability

The system suitability parameters were obtained using the Agilent ChemStation software, as follows: retention factor (K), number of theoretical plates (N), tailing factor (Tf), and resolution (Rs). An isovitexin solution in acetonitrile (1 μg/mL) was used to evaluate the parameters.

Selectivity

The spectral purity of isovitexin in samples was evaluated using spectral data and peak purity metrics.

Calibration curve

The analytical curve for isovitexin was obtained using the method developed with the previously described chromatographic conditions. The calibration curve was obtained using isovitexin solution (10, 20, 30, 40, and 50 μg/mL) diluted in acetonitrile:water (1:1). The assay was done in triplicate, and all samples were injected in duplicate. The area values were plotted against the corresponding concentrations, and statistical analysis was performed.

Linearity

The analytical method used to quantify the compound isovitexin in P. setacea samples was validated according to the parameters recommended by the official guides (Brasil, 2011; FDA, 2015; Brasil, 2017).

Matrix curve

The matrix calibration curve was obtained using isovitexin (10, 20, 30, 40, and 50 μg/mL), with the addition of 100 µL of a PSH-01F sample diluted in acetonitrile:water (1:1). All samples were injected in triplicate. The area values were plotted against the corresponding concentrations, and statistical analysis was performed.

Limit of detection and quantification

The limits of detection (LOD) and quantitation (LOQ) were determined using the standard deviation of the response (SD) and the slope of the calibration curve (S) in the Equations 2, 3:

Accuracy (recovery)

Accuracy was verified by adding 100 μL of PSH-01F (10.12 mg/mL) to isovitexin at three concentration levels (Supplementary Table S2). All the samples were injected in triplicate. The area values were plotted against the corresponding concentrations, and statistical analysis was performed. Recovery was determined using the equation below:

Robustness

The robustness of the method was investigated by varying the following parameters: wavelengths (λ 335, 340, and 345 nm), flow rate (0.6, 0.8, and 1.0 mL/min), and oven temperature (23, 25, and 27 °C). The analyses were performed using three replicate samples of the isovitexin standard solution (10, 30, and 50 μg/mL). The concentrations of isovitexin were calculated, along with the relative standard deviation. The mean, SD, and RSD were calculated. It was considered acceptable when the RSD was less than or equal to 5% (Moura-Silva et al., 2023).

Analysis of isovitexin in PSHF-01

The developed method was applied to detect the concentration of isovitexin (mg isovitexin/100 mg of extract) in samples of P. setacea. The samples were analyzed in triplicate. The concentration of isovitexin was determined using the previously described analytical curve.

UPLC-PDA-ESI-MS

The chemical composition of PSHF-01 was evaluated by UPLC-PDA-ESI-MS/MS. The sample was prepared by dissolving 1 mg of PSH-01F in HPLC-grade methanol (1 mg/mL) with the aid of an ultrasonic bath for 10 min. The solution was centrifuged at 8480° g for 10° min, filtered through a 0.22 μm PVDF syringe filter membrane, and then 5 μL was injected into the Acquity UPLC system (Waters, USA). The chromatographic system was connected to a photodiode array detector (PDA) and a mass spectrometer (Xevo TQD) equipped with an electrospray ionization (ESI) source, operating in negative mode, and a quadrupole analyzer. The mobile phase solvents were water (A) and acetonitrile (B), each containing 0.1% v/v formic acid. The analysis was conducted at a flow rate of 0.3 mL/min using a convex gradient (curve 4) of 5%–22% B from 0–4.0°min, followed by a linear gradient of 22%–95% B from 4–5 min. A reverse linear gradient (5–6° min, 95%–5% B) was then employed to return to the initial conditions, followed by 2 min at 5% B to restore column equilibrium for the next injection. An Acquity UPLC HSS C18 column (1.7° μm; 50 × 2.1 mm i.d.; Waters) and a guard column (Acquity UPLC HSS VanGuard pre-column, 5 × 2.1 mm i.d., 1.7° μm; Waters) were used in the analysis. The column was maintained at 40 °C while the autosampler temperature was set at 19 °C. UV detection was performed in the wavelength range of 210–500 nm, while MS spectra were recorded in ESI negative mode using the following conditions: capillary voltage 3.50° kV; extractor voltage 3,0 V, RF lens 0,1 V, cone voltage ramping from 20 to 60 V; source temperature was set at 120 °C; desolvation temperature at 400 °C; nitrogen (N 2) was employed as the nebulizing and desolvation gas at a flow rate of 550 L/h, while argon was used as the collision gas. In the full-scan analysis mode, the mass range was set to m/z 100–1500 Da. MS/MS experiments were then conducted in product ion scan (daughter-scan) mode for the main peaks observed in the full-scan MS spectra (m/z 431, 447, 563, 579, 593, and 609). The source parameters were maintained as described for the MS1 analysis, except for the cone voltage, which was set to 30 V. We used collision energy ramping from 20 to 50 eV, with collision gas (argon) flow at 0.1 mL/min. The mass range for MS/MS experiments was 100–610 Da. Instrument control, data acquisition, and processing were performed using MassLynx™ V4.1.

Putative assignments for the major compounds in the extract were achieved by combined evaluation of their retention behavior, UV/Vis absorption spectra, Total ion count (TIC) spectra from the full-scan analysis, diagnostic product ions in MS/MS, along with comparison with fragmentation patterns reported for flavone C- and O-glycosides.

In silico target fishing using ligand-based virtual screening

In silico prediction of the biological activities of flavonoids characterized in P. setacea extract (Table 1S) was performed using the Active-IT system. 3D molecular structures of active and inactive compounds were obtained from PUBCHEM, using up to ten conformations, and potential pharmacophore fingerprints were generated using the 3D-Pharma module (Oliveira et al., 2025; Domingues et al., 2024). The fingerprints corresponding to multiple conformations were combined into a single modal vector representing dynamic conformational behavior, defined as a four-dimensional (4D) pharmacophore fingerprint. For each biological dataset (Santos et al., 2015), predictive models were constructed using the ExCVBA module with two supervised machine learning algorithms: support vector machines implemented with LibSVM (Chang et al., 2011) and the Naive Bayes classifier (Williams, 2017). Model development followed the Active-IT protocol, employing a recursive stratified random partitioning approach to ensure statistical robustness and reliability. In each interaction, compounds annotated as active or inactive were randomly divided into a training set (70%) and a validation set (30%). This procedure was repeated at least 30 times per dataset, and all generated models were retained for subsequent prediction (Figure 3).

Model performance was evaluated using the average score obtained by each compound across the models in which it appeared in the validation set. These scores were used to calculate the area under the receiver operating characteristic curve (AUC) and to support predictions of biological activity for new compounds. For activity prediction, the modal fingerprint of a query compound was generated and submitted to all available models associated with a given biological endpoint, yielding a set of raw scores. These raw scores were averaged and transformed into probabilities by comparison with the score distributions of known active and inactive compounds, thereby estimating the likelihood of belonging to each class.

The probability of activity (Pa) was calculated as the fraction of active compounds exhibiting raw scores lower than that of the query compound. In contrast, the probability of inactivity (Pi) was determined as the fraction of inactive compounds with higher raw scores. The difference between these probabilities (Pa − Pi) was used as the primary indicator of biological relevance. Based on previously established criteria, compounds presenting Pa − Pi ≥ 0.5 in support vector machine models or Pa − Pi ≥ 0.8 in Naïve Bayes models were considered to exhibit the most promising predicted activities and molecular targets. The results were analyzed and organized by main activity (Supplementary Table S2) using artificial intelligence–based tools, including ChatGPT (OpenAI, GPT-5.5), Microsoft Copilot, and Consensus AI. The prompt used in AI analysis is in the Supplementary Material (Prompt 1).

Statistical analysis

The results of the present study were processed and statistically analyzed using GraphPad PRISM (Version 9) and SPSS (Version 2.0). All values were expressed as mean ± standard deviation (SD).

Analysis of variance (one-way ANOVA) was performed, followed by Tukey’s multiple comparisons test. Differences were considered statistically significant at p < 0.05. Furthermore, the data referred to the linearity parameter were evaluated in the following tests: Jackknife Standardized Residual Test (evaluated the Outliers), Ryan-Joiner Test (evaluated the normality of residuals), Durbin-Watson Test (evaluated the independence of residuals), the Brown-Forsythe or Modified Levene Test (evaluated the homoscedasticity of residuals), and ANOVA (Test for deviation from linearity and regression significance).

The samples selected for this study were leaves of P. setacea grown in the orchard (PS-01F) and in greenhouse-grown plants (PS-02F and PS-03F). Plants grown under field conditions are subject to a range of abiotic and biotic factors that influence the biosynthesis of secondary metabolites, particularly flavonoids, as reported in previous studies (Almeida et al., 2021). In contrast, plants cultivated in greenhouses are subjected to more controlled environmental conditions, which may affect the production and accumulation of these metabolites compared with their field-grown counterparts. The extractive solvents used in this work were water, ethanol:water (50% v/v), and ethanol.

The results obtained for the six leaf extracts of P. setacea are summarized in Table 1.

In this study, the extracts from leaves of orchard-grown and greenhouse-grown plants were prepared using three different extraction techniques (sonication, percolation, and infusion). The selection of the extraction methodology and solvent type is an important factor in optimizing both extract yield and associated biological activities. Ultrasonic extraction is a fast and economically viable method for extracting natural products without heating. It significantly reduces extraction time, enhances extraction efficiency, reduces solvent usage, and can be used with heat-sensitive substances. The solvent generally used is ethanol (Zhang et al., 2023). The proportion of dried plants and the solvent was standardized from 1.0 g of dried plant sample to 30 mL of ethanol. The most common method used by the population for preparing medicinal plant leaves is infusion (Azwanida, 2015). Although the solvent/sample ratio usually varies between 4:1 and 16:1 (Abubakar and Haque, 2020), in this study the proportion used was 1.0 g of dried leaves with 30 mL of water. The percolation method was conducted at room temperature (approximately 25 °C) over a total of 26 days. This procedure is suitable for extracting heat-sensitive substances (Zhang et al., 2023). The selection of solvent was based on the extraction conditions described for P. edulis in the Brazilian Pharmacopeia (Anvisa, 2019). As previously mentioned, ethanol, water, and their mixture are widely used solvents for the extraction of flavonoids (Tzanova et al., 2020).

The yields of the extraction methods are shown in Table 1. The sonication extraction technique with ethanol yielded 8.81% for sample PSEF-01% and 8.24% for PSEF-02. The extract obtained by infusion (aqueous extraction) showed a yield of 28.44% for PSAF-01% and 19.97% for PSAF-02. The yield was higher for the extract obtained from leaves using a percolation method with EtOH:H 2 O. The extract of leaves obtained from greenhouse plants (PSHF-03) showed a higher percentage yield (40.17%) than the leaves from orchard plants (PSHF-01) extracted with the same solvent (33.75%). The percolation with EtOH:H 2 O 50% was more efficient than the other two extraction methods. Although this result cannot be statistically validated because the extraction was performed only once using a pooled sample of individuals, the extraction yield serves as an indicator of the solvent’s efficiency and its capacity to selectively solubilize and recover specific metabolites from the plant matrix.

The total phenolic content (TPC) of the obtained extracts was evaluated using the Folin-Ciocalteu reagent, with gallic acid as the standard. This assay is a rapid and simple colorimetric method used to quantify the TPC (Airin and Islan, 2022). The method is based on an oxidation-reduction reaction in which electrons are transferred from phenolic compounds to phosphomolybdic/phosphotungstic acid complexes in an alkaline medium, forming blue complexes that are determined spectrophotometrically at approximately 760 nm (Ainsworth and Gillespie, 2007). The TPC of the evaluated extracts varied from 11.32 ± 0.33 (PSEF-03) mg of gallic acid equivalents per Gram of extract to 34.24 ± 1.55 (PSF-01) mg of gallic acid equivalents per Gram of extract (Table 1).

The total flavonoid content (TFC) is widely quantified using the aluminum chloride (AlCl 3) colorimetric assay, which is based on the formation of flavonoid–aluminum complexes. The resulting complexes are measured spectrophotometrically at 410–430 nm after adding an AlCl 3 solution (Pękal and Pyrzynska, 2014). The concentrations of total flavonoids in the samples varied from 3.79 ± 1.10 (PSAF-02) to 33.65 ± 4.73 (PSEF-01) mg of total flavonoids per Gram of extract, calculated as quercetin (Table 1).

According to the results obtained herein, the highest TPC and TFC contents were observed in extracts from leaves of orchard-grown plants (sample PSF-01) using the three different extraction methods. The polyphenol and flavonoid contents were higher in the orchard-grown samples when samples from orchard-grown and greenhouse-grown plants were compared using the same extraction method. This result was expected since UV light induces the synthesis of flavonoids and polyphenols (Agati et al., 2012; Petrussa et al., 2013; Zagoskina et al., 2023).

In the current study, the antioxidant activity of leaf extracts of P. setacea was evaluated using the β-carotene/linoleic acid peroxidation method, which assesses the sample’s ability to protect a lipid substrate from oxidation (Duarte-Almeida et al., 2006). All the evaluated extracts were considered active with IC 50 values lower than 60 μg/mL. The extract PSEF-02 showed the highest antioxidant activity in the β-carotene/linoleic acid system, with an IC 50 value of 39.71 ± 1.57 μg/mL.

This result was expected, since polyphenols can act as antioxidants due to their reducing activity (Zagoskina et al., 2023). A positive correlation between total phenolic content and antioxidant capacity has been described (Saeed et al., 2012; Muflihah et al., 2021; Martínez et al., 2022). Accordingly, an exploratory Pearson correlation analysis was performed between total phenolic content (TPC), total flavonoid content (TFC), and antioxidant activity (β-carotene/linoleic acid assay IC50 values). However, the results showed a weak and non-significant correlation between TFC and antioxidant activity (r = 0.2721, p = 0.6019), as well as a low-to-moderate but non-significant trend for TPC (r = 0.3014, p = 0.09087). Importantly, the lack of strong correlation is consistent with previous studies reporting that antioxidant activity in plant extracts cannot be solely explained by total phenolic or flavonoid contents, due to the complexity of phytochemical composition and the contribution of synergistic effects among multiple secondary metabolites (Huang and Prior, 2005; Floegel et al., 2011).

In our study, AChE inhibition and %I were investigated in vitro using a modified quantitative microplate assay. All P. setacea extracts showed %I below 40% compared to physostigmine (%I = 90.91 ± 0.68) and were considered not significant under the assay conditions used.

Considering that the leaves of Passiflora grown in the orchard are more easily accessible, have a greater amount of material available, and the results obtained (Table 1), the hydroalcoholic extract of P. setacea leaves grown in the orchard (PSHF-01) was selected for complementary studies.

The antioxidant activity of the selected sample was evaluated using two other methods: the DPPH and TBARS assays. Both methods present different and complementary mechanisms of action of the beta-carotene/linoleic acid assay (Danet, 2021). However, the extract was considered inactive in both the complementary assays. The different results obtained among the antioxidant assays can be explained by the different chemical principles and reaction environments involved in each method. The DPPH scavenging method involves the transfer of electrons from an antioxidant substance, such as the plant extract, to an oxidant (Gulcin and Alwasel, 2023). It is one of the most commonly used in vitro chemical antioxidant assays because it is sensitive, rapid, reliable, and the sample does not require any pre-treatment (Nwachukwu et al., 2021). However, its main limitation is that it does not reflect the complexity of antioxidant mechanisms occurring in biological systems because it uses a synthetic radical under non-physiological conditions. In addition, compounds with limited solubility or slow reactivity with DPPH may exhibit underestimated antioxidant activity (Wołosiak et al., 2022). The β-carotene/linoleic acid assay is mainly associated with the capacity of compounds to prevent the oxidative degradation of β-carotene induced by linoleic acid-derived radicals (Ghani et al., 2019). Factors such as emulsion stability, reaction conditions, and the polarity of the tested compounds may affect the reproducibility of results. Additionally, the complex emulsion system may lead to different responses depending on the partitioning behavior of antioxidants within the assay medium (Frankel and Meyer, 2000; Olszowy-Tomczyk and Wianowska, 2025).

The TBARS assay measures the inhibition of secondary products of lipid peroxidation, particularly malondialdehyde-like substances, usually formed at later stages of oxidative degradation. A limitation of this method is its low specificity, since compounds other than MDA can also react with thiobarbituric acid, overestimating the measured antioxidant effect. Accordingly, TBARS values should be interpreted with caution because they do not always reflect the true extent of lipid peroxidation (Janero, 1990; Moselhy et al., 2013). The absence of activity in this assay suggests that the extracts may not efficiently inhibit the formation of these secondary oxidation products under the tested conditions, or that the active constituents may act preferentially at earlier stages of lipid oxidation (De Leon and Borges, 2020).

The three methods aforementioned have different mechanisms of action and present advantages and disadvantages. However, they provide valuable and complementary information on different aspects of antioxidant activity, although they do not reflect physiological conditions and should be regarded as a preliminary screening. Therefore, the positive response observed only in the β-carotene/linoleic acid assay suggests that the extracts may exert antioxidant effects mainly by protecting lipid systems against radical-mediated oxidation, rather than by directly scavenging stable free radicals or inhibiting the formation of TBARS-reactive secondary oxidation products.

In these studies, different flavonoids were identified using the chromatographic techniques TLC, HPLC-UV, and UPLC-PDA-ESI-MS/MS.

For flavonoid characterization by TLC (Figure 4), the NP/PEG developer was used. Fluorescence in the colors yellow-orange, yellowish-green, and orange indicates the presence of flavonoids (Wagner and Bladt, 2001). Thus, the results obtained indicated the presence of different flavonoids in the hydroalcoholic extract of P. setacea leaves (PSHF-01). Of the seven flavonoid standards used, the results only indicated the presence of the flavonoids isovitexin (greenish fluorescence), retention factor (RF) = 0.56, and isoorientin (yellowish fluorescence), RF = 0.51, in the extract. Three other yellowish fluorescence spots (RFs = 0.33, 0.27, and 0.22) were observed in the sample on the developed TLC plate. These spots did not match any of the other patterns used in the analysis, indicating that they do not correspond to any of the patterns used.

For the characterization of flavonoids in the PSHF-01 extract by HPLC-UV, the retention times of authentic isovitexin (IV) and isoorientin (IO) standards were compared with those of the corresponding peaks detected in the sample chromatogram (Figure 5). In addition, co-injection experiments were performed by spiking the extract with the respective standards, resulting in increased peak intensities for isovitexin and isoorientin (Figure 5). Based on the chromatographic profiles, retention time matching, and co-injection analyses, the presence of the flavonoids isovitexin and isoorientin in the PSHF-01 was confirmed.

The analytical method using HPLC-DAD was performed to quantify isovitexin. To confirm the retention time and selectivity of the method, the isovitexin standard was added to the P. setacea samples, increasing the peak area relative to isovitexin. The spectral purity of isovitexin in samples was evaluated using spectral data and peak purity metrics (Figure 1S).

The analytical curve (Figure 6) was expressed as the area of the isovitexin peak vs. concentration of isovitexin (μg/mL): Y = 6.532X – 15.733. It was obtained with a correlation coefficient (R 2 = 0.9946, in accordance with the criteria of RDC n° 166/2017 (Brasil, 2017) which establishes as a minimum acceptable criterion that the analytical curve presents a correlation coefficient greater than or equal to 0.9900.

The limit of detection (LOD) represents the lowest concentration of isovitexin that can be detected. In contrast, the limit of quantification (LOQ) is the lowest concentration that can be accurately measured. Based on the response standard deviation and slope, the LOD was 2.34 μg/mL, and the LOQ was 7.107 μg/mL.

The matrix effect was investigated to assess possible interference caused by compounds in the vegetable extracts and was evaluated according to RDC 166/17 (Brasil, 2017). The matrix effect was not significant (Supplementary Figure S1; Supplementary Table S2), with a p-value <0.05. In this case, the lines are parallel.

In an ideal situation for the accuracy (recovery) evaluation, the matrix will be free of the target analyte. However, in phytochemical analysis, the latter occurs naturally in the matrix. Therefore, the most common technique for determining accuracy in natural product studies is the standard addition method. In terms of accuracy (Supplementary Table S3), the method provided recoveries ranging from 107.00% to 116.31% for isovitexin. The acceptable recovery followed these ranges for tests on complex matrices (80%–120%), such as natural products (Betz et al., 2011).

Considering robustness, the method was robust (Supplementary Table S7) to deliberate changes in the detection wavelengths and column temperature, with RSD below 5.0% in the conditions evaluated except for temperature evaluation at the lowest concentration tested (10 μg/mL). Thus, the column oven temperature must be precisely controlled when applying this method to quantify isovitexin at lower concentrations. The validation parameters showed satisfactory results. The method can be used to quantify the concentration of isovitexin in PSHF-01 (Supplementary Tables S1-S7).

Isovitexin quantity in the P. setacea sample (PSHF-01) was determined employing the validated HPLC-UV method. The isovitexin content was 0.223 ± 0.021 mg isovitexin/100 mg of extract.

The chemical composition of PSHF-01 was further investigated by UPLC-PDA-ESI-MS/MS (Figure 7). The chromatographic profile showed seven major peaks, eluted from 2.5 to 3.5 min, with a UV profile typical of flavones. The UV spectra associated with peaks 1, 2, three and six exhibited absorption maxima at 270 and 348 nm, whereas peaks 4, five and seven showed λ max at 269 and 334 nm, consistent with luteolin and apigenin glycosides, respectively. Peaks two and seven were identified as isoorientin and isovitexin, respectively, based on their chromatographic behavior, spectral data and comparison with authentic reference compounds. The identity of these compounds has already been confirmed by HPLC-UV analysis, through co-injection with authentic standards, as described in the previous section. The mass spectrum of the major chromatographic peak (peak 4) showed a deprotonated molecular ion at m/z 593. MS/MS analysis of this ion showed fragments at m/z 473 and 413, yielding a fragmentation pattern consistent with vicenin II (Figure 8). This finding suggests that vicenin II, or a structurally related isomeric flavone C-glycoside, may represent the major constituent of PSHF-01. Similar fragmentation patterns have been reported for vicenin II in Passiflora species. Nevertheless, the MS/MS data obtained in the present study are not sufficient for unequivocal identification, since isomeric compounds may exhibit comparable fragmentation behavior. Therefore, isolation of the compound followed by comprehensive spectroscopic characterization, particularly by NMR, and/or comparison with an authentic reference standard will be required to confirm its identity. The same analytical approach was applied to peaks 1, 2, five and 6, which were tentatively assigned as luteolin- O -glucopyranosyl-C-glucopyranoside (orientin or isorientin- O -glucoside), orientin or isoorientin- C -xyloside, vitexin or isovitexin-2”- O -xyloside and orientin or isoorientin-C-rhamnoside, respectively (Table 2). These putative assignments were based on full-scan and MS/MS fragmentation data, together with comparisons with previous reports of fragmentation patterns for apigenin and luteolin glycosides. The proposed fragmentation pathways for these compounds are presented in Figure 9. Overall, the UPLC-PDA-ESI-MS/MS analysis complemented the chemical characterization of PSHF-01 by enabling the tentative annotation of additional flavonoid glycosides that the previous analytical methods could not identify. However, except for isoorientin and isovitexin, all remaining compound assignments should be regarded as tentative. Isolation and complete structural elucidation, including NMR analysis and/or comparison with authentic reference compounds, will be necessary to achieve unequivocal identification of these constituents. Finally, the chemical profile proposed here for PSHF-01 consists mainly in C-glycosylflavones, which can explain the low to moderate antioxidant activity described in this study. Apigenin C-glycosides, including vitexin, isovitexin and vicenin-2, have been consistently associated with antioxidant activity in both chemical and cell-based assays. Although their radical scavenging capacity is generally lower than that of luteolin-derived C-glycosides because apigenin lacks the catechol group in the B ring, these compounds can neutralize reactive oxygen species through different mechanisms and may also enhance endogenous antioxidant defenses (Babaei et al., 2020). Their C-glycosidic linkage confers greater resistance to hydrolysis and improved chemical stability, which may contribute to their biological activity under physiological conditions.

Previous validation of in silico prediction of known biological activities of the analyzed compounds

We predicted the biological activities of flavonoids using our Active-IT software. Before making these predictions, we validated our approach with known activities for these compounds. Of the four compounds included in this study, three appear in different datasets from the PubChem Bioassay and are classified as either active or inactive. These compounds are isoorientin, isovitexin, and vicenin II; each appears in four datasets within the Active-IT system. In total, there are 12 datasets from the PubChem Bioassay, and only three showed the compound isoorientin as active (See AIDs 602124, 602125, and 624351 in Supplementary Table S9). In all other bioassays, the tested compounds were inactive.

According to the Active-IT protocol, each bioactivity model is built from 30 individual models, with 70% of the dataset’s active and inactive compounds used as the modeling set and 30% as the validation set. The process is repeated 30 times by randomly re-sorting the compounds. For validation purposes, the compounds were submitted to Active-IT using only those 12 datasets. As expected, the activity predictions for the analyzed compounds using these models produced excellent results, with an Area Under the Curve (AUC) of 1.00 for SVM and 0.89 for Naïve Bayes. However, this is just an internal validation, since the compound sometimes appears in the modeling set. To avoid this kind of bias, all individual models (of the 30 original models) that included the compound in the modeling set were discarded, and only those that appeared in the validation set were retained (see Supplementary Tables S8 and S9 in the Supplementary Material). Again, the results were excellent for SVM models, with an AUC of 1.00, while Naïve Bayes performed slightly worse, with an AUC of 0.78. This validation can be considered external and will support the predictions made by the Active-IT system.

However, the Area Under the Receiver Operating Characteristic curve (AUC ROC) is a popular metric for binary classification, but it has major limitations. It can hide poor performance on imbalanced data and overlook the costs of false results (Mazumder, 2025). AUC is a single numerical value that shows how well a classifier works across all possible threshold levels. To better assess the predictability of machine learning approaches, we simulate a single threshold of 0.5 for SVM and 0.8 for Naïve Bayes to define a predicted active compound, as has been used in the predictions phase. Applying such thresholds, we obtain the classification errors (false positives and false negatives) and correct answers (true positives and true negatives), which help evaluate how well the approaches distinguish active compounds. The corresponding confusion matrices are present in Supplementary Table S10, and the statistical parameters extracted from the matrices are shown in Supplementary Table S11, allowing these results to be interpreted in terms of their significance. The definitions of the metrics can be obtained from several sources elsewhere (Mazumder, 2025; Lopes et al., 2017; De Winter and Lopes, 2018).

This comparison looks at internal and external validation of SVM and Naïve Bayes machine learning methods for prediction results. Both methods show perfect sensitivity (1.00) and a Negative Likelihood Ratio of zero, meaning all models could recover all active instances. In the internal testing, the SVM model performs perfectly, with an F1-score, precision, and Matthews Correlation Coefficient (MCC) all at 1.00, showing very reliable classification as expected. In comparison, the Naïve Bayes model performs worse, with a precision of 0.75 and an MCC of 0.82. External testing gives a better idea of real-world use. Here, the Naïve Bayes model’s precision drops to 0.60, leading to an F1-score of 0.75 and a moderate, but still good, MCC of 0.68. In external testing, the SVM model maintains very good performance, with a precision of 0.75, accuracy of 0.92, and MCC of 0.82, suggesting stronger practical significance for prediction. Overall, the SVM models perform better than Naïve Bayes, making them more reliable for prediction and more useful for identifying active cases.

In silico prediction of other biological activities of the identified compounds

Active-IT is a ligand-based virtual screening (LBVS) tool developed to predict the biological and pharmacological activities of small organic molecules. Its objective is to support and rationalize the selection of biological assays by prioritizing the most relevant predicted activities, thereby guiding the design of targeted bioassays. In addition, computational analysis enables the investigation of potential mechanisms of action, particularly when molecular targets or associated genes are known (Almeida et al., 2024).

More than 3700 bioactivity models were built, each comprising 30 SVM and 30 Naïve Bayes individual models, as well as 60 additional randomized models. Bioactivity models with low performance or discrimination between regular and randomized models are discarded. The prediction module uses the raw score distributions for active and inactive molecules to calculate the probability that the compound is active (Pa) or inactive (Pi). The in silico prediction order in Table 2S considers the likelihood of having the analyzed activity (expressed as Order), the AID (number from PubChem), the description of the AID assay, and the result. The analysis of predicted biological activities for flavonoids detected in P. setacea (Supplementary Tables S12 and S13) showed a probable pharmacological profile characterized by antimicrobial, antiparasitic, anti-inflammatory, and antitumor activity as well as immunomodulatory effects.

To facilitate interpretation of the predicted results, they were classified into the following categories: ADMET properties, antiparasitic targets, anti-tumor, antimicrobial, and anti-inflammatory activities. All classifications were subsequently verified and adjusted by the authors to ensure accuracy and consistency. An important observation is that many associated targets remain poorly explored. The predictions were then critically compared with published studies to identify putative biological activities that have not yet been experimentally reported for the investigated compounds.

Analysis of assays classified as ADMET revealed a pattern of genotoxic potential for the evaluated molecules. In the predictive analysis, the assays classified as ADMET revealed a pattern of genotoxic potential for the evaluated molecules. The compounds show metabolic stability. Really, flavonoids from P. alata have been shown to be genotoxic (Boeira et al., 2010). But other studies showed that flavonoids of the Passiflora genus are antigenotoxic (Lv et al., 2016). Consequently, more studies are needed to assess this activity. The prediction for Liver Microsomal (HLM) and Rat Liver Microsome Stability (RML) assays suggested t½ values higher than 30min (Supplementary Table S13), which was also confirmed by literature data. The half-life of orally administered vitexin/isovitexin in rats is approximately 6.3 h, while the intravenous route has T½ of less than 1 h (Mai et al., 2025). The activity of flavonoids (isovitexin, isoorientin, vicenin II, isovitexin 2″-O-xyloside) in DNA repair pathways has not been described in the consulted literature. This is a gap that can be explored.

The Active-IT system predicted multiple targets for viral diseases for the flavonoids analyzed. These include Zika, HIV-1 (3 targets), Hepatitis C, SARS-CoV-2, and Epstein-Barr viruses. These data could not be directly validated, but isovitexin and isoorientin have demonstrated in vitro antiviral activity against dengue virus serotype 2 (DENV-2), influenza A virus, and respiratory syncytial virus (RSV). In a work by Fonseca et al. (2022), isovitexin and isoorientin inhibited dengue virus replication. In additional studies, isovitexin exhibited in vitro activity against a mouse lung-adapted variant of influenza A virus (H1N1, strain FM1) and against RSV (Ye et al., 2011).

The integrated analysis of the predictions demonstrates that the evaluated compounds can act on parasitic targets, exhibiting consistent antiparasitic activity against Leishmania, Trypanosoma, and Plasmodium. The analysis showed the probability of the compounds acting on Leishmania mexicana pyruvate kinase (LmPK). The predicted activity against Trypanosoma brucei is related to the possibility that the compounds can act as inhibitors of RNA Editing Ligase 1 (TbREL1). Another relevant putative target is the Plasmodium falciparum M17 leucine aminopeptidase (PFM17LAP) inhibitor (Bioassay AID 492977). To assess the reliability of the predicted activity, the available literature was examined, confirming that isovitexin has previously demonstrated antileishmanial activity against both promastigote and amastigote forms of Leishmania (L.) amazonensis (Ferreira et al., 2018). In addition, a study conducted by Atay et al. (2016) showed that isoorientin exhibited activity against P. falciparum, T. brucei rhodesiense, and Trypanosoma cruzi, but was inactive against Leishmania donovani. The in silico analysis proposes the targets that can be explored.

In the in silico studies, the following predicted targets correlated with cardiovascular disease include calcium sensitivity of cardiac Regulated Thin Filaments (RTF), Voltage-gated potassium channel subunit beta-2 (KCNAB2), estrogen receptor alpha signaling, and cAMP-regulated guanine nucleotide exchange factor 3 (EPAC1). Isovitexin has been more studied for its activity in cardiac disease. According to Azubuike-Osu et al. (2021), studies suggest that isovitexin has a cardioprotective role during ischemia-reperfusion injury and angiogenesis, as well as decreases perfusion pressure and increases the bioavailability of nitric oxide. The potential application of nutraceuticals in the prevention and treatment of cardiovascular diseases remains underexplored and deserves further investigation.

The targets associated with cancer-related processes predicted by the Active-IT system include DNA repair (RECQ1, PRKDC), tumor suppressor pathways (TP53), and epigenetic regulation (CBX1). These targets play central roles in tumor survival, genomic stability, and resistance to therapy. The involvement of metabolic enzymes such as PKM2 and ALDH1A3 further highlights the connection between flavonoids and metabolic reprogramming, a hallmark of cancer cells.

The antitumor activity of flavonoids has been widely discussed in several articles (Abotaleb et al., 2018; Chirumbolo et al., 2018; Rodriguez-Garcia et al., 2019; Kopustinskiene et al., 2020); however, the molecular mechanisms underlying this effect have not yet been fully elucidated and warrant further study.

Another important predicted activity concerns inflammatory and immune signaling, particularly involving NF-κB, Toll-like receptors, and AHR pathways. These findings are consistent with the extensive literature (Al-Khayri et al., 2022) describing flavonoids as modulators of chronic inflammation. The anti-inflammatory activity of isovitexin, isoorientin, and vicenin II has been studied in different targets. In the literature, isovitexin inhibited the expression of pro-inflammatory cytokines (TNF-α, IFN-γ, IL-2, and IL-17A) (Zhang et al., 2021), acute gouty arthritis via TLR4/MyD88/NF-κB (Sun et al., 2021), and acute lipopolysaccharide lung injury (ALI) through the impact of MAPK and NF-κB pathways and activation of HO-1/Nrf2 pathways (Lv et al., 2016). Isoorientin has been evaluated in different inflammatory processes (Anilkumar et al., 2017; Li et al., 2020; Liang et al., 2022; Hu et al., 2022; Jiang et al., 2026). Vicenin II reduces inflammatory mediators (Kang et al., 2015; Yin et al., 2019).

Another relevant aspect in the predictions is the strong representation of oxidative stress and redox-regulated pathways, particularly through NRF2 activation and modulation of mitochondrial function. Flavonoids are widely recognized for their antioxidant capacity, which involves both direct scavenging of reactive oxygen species (ROS) and indirect activation of endogenous defense systems. Search literature data show that the antioxidant and neuroprotective effects of flavonoids have been extensively investigated. Numerous studies have demonstrated that this class of compounds can modulate oxidative stress and protect against neuronal damage (Maher, 2019; Devi et al., 2021; Maan et al., 2020; Pietta, 2000; An et al., 2025).

Finally, the Platelet Dense Granule Release inhibition was predicted by flavonoids analyzed (AID 1889; order 18/96), which prevents thrombus formation. In the literature, 3′,4′-dihydroxyflavonol and quercetin have been reported to inhibit platelet-mediated arterial thrombosis and platelet granule exocytosis, which partially validates the prediction.

From a pharmacological perspective, the observed hypothetical multitarget profile shown in the in silico analysis is consistent with the structural characteristics of flavonoids, which include polyphenolic scaffolds that interact with diverse proteins and signaling systems. These characteristics of flavonoids may be advantageous for treating complex diseases, but may also pose challenges to selectivity and mechanistic specificity. Overall, the results support the hypothesis that flavonoids exhibit a broad-spectrum pharmacological potential, primarily driven by their ability to regulate oxidative stress, inflammation, and cellular metabolism. These mechanisms collectively contribute to their proposed roles as anti-cancer, anti-parasitic, chemopreventive, anti-inflammatory, and neuroprotective agents.

Passiflora setacea, an endemic wild species of the Brazilian Cerrado, has attracted considerable interest for its medicinal and agronomic properties. The present study shows the chemical profile and biological activity of P. setacea cv. BRS Pérola do Cerrado leaves from orchard-grown and greenhouse-grown plants. The data demonstrate that orchard-grown plants (PSF-01) consistently yield higher levels of total polyphenols and flavonoids than their greenhouse-derived counterparts, regardless of the extraction method. All evaluated extracts exhibited antioxidant activity in the β-carotene/linoleic acid peroxidation assay. However, the selected hydroalcoholic extract (PSHF-01) was not active in both the DPPH free radical scavenging and TBARS assays. In contrast, it showed antioxidant activity in the β-carotene/linoleic acid assay, suggesting that its antioxidant potential may be limited to protecting lipid systems against radical-mediated oxidation. This effect may be attributed to the specific mechanisms underlying the β-carotene/linoleic acid assay, which measures lipid peroxidation inhibition.

An HPLC method was developed to quantify isovitexin as a marker compound, enabling reliable future analysis. The isovitexin content in the P. setacea sample (PSHF-01) employing the validated HPLC-UV method. was 0.223 ± 0.021 mg isovitexin/100 mg of extract.

The phytochemical investigation of PSHF-01 identified C -glycosylflavonoids, including isovitexin, isoorientin, isovitexin-2″- O -xyloside, and vicenin II, the latter being the major compound. It might be interesting to further investigate the phytochemical profile of this species’ leaves under different experimental conditions, considering that some flavonoids previously reported in the literature were not identified in the present study.

The in silico studies for the flavonoids identified in this species showed a multitarget pharmacological profile. Future studies integrating omics approaches, network pharmacology, and experimental validation will be essential to further elucidate their mechanisms of action and therapeutic applicability.

The outcome of this study may contribute to increased understanding of the potential of P. setacea, whose favorable agronomic traits are complemented by promising prospects for alternative therapeutic applications.

Several challenges remain in the study of P. setacea. Future research should focus on identifying the compounds responsible for its sleep-modulating effects and establishing robust cell-based models to investigate intracellular reactive oxygen species (ROS) production, antioxidant mechanisms, and cytoprotective activity. Such approaches should include intracellular ROS quantification, assessment of lipid peroxidation, evaluation of endogenous antioxidant defense systems, and investigation of cytoprotective effects under oxidative stress conditions. These studies will provide a more physiologically relevant understanding of the antioxidant potential of P. setacea and help validate the effects previously observed in chemical assays. In addition, untargeted metabolomic profiling may reveal metabolic signatures associated with disease resistance and support the discovery of bioactive metabolites. Further investigations should also evaluate the influence of cultivation conditions on the accumulation of flavonoids and other phenolic compounds, apply in silico approaches and network pharmacology to predict molecular targets and signaling pathways, and experimentally characterize the molecular mechanisms underlying the biological activities attributed to P. setacea.