Synergistic lipase inhibition by Garcinia parvifolia metabolites: a 1H NMR–based metabolomic deciphering of ligand interactions Authors: Ahmad Iqbal Noordin, Fazleen Izzany Abu Bakar, Fiffy Hanisdah Saikim, Santad Wichienchot, Azizul Isha, Mohd Fadzelly Abu Bakar Published: 2026 ABSTRACT IntroductionThe rising prevalence of obesity and metabolic disorders has intensified the demand for sustainable, plant-based functional ingredients for weight management. Garcinia parvifolia, an underutilized tropical fruit native to Malaysian Borneo, represents a promising natural source of anti-obesity compounds.ObjectivesThis study aimed to decipher the complex, ligand interactions responsible for pancreatic lipase inhibition in G. parvifolia using a multi-disciplinary metabolomics-driven approach.MethodsExtracts of varying polarities (70% ethanol, acetone, chloroform, and hexane) were analyzed using ¹H NMR-based metabolomics, in vitro p-nitrophenyl butyrate (p-NPB) lipase assays, and in silico molecular docking.ResultsThe results demonstrated that solvent polarity critically dictates chemical diversity, with the 70% ethanol extract yielding the highest metabolite abundance and the strongest lipase inhibitory activity (IC₅₀ = 85.3 µg/mL). Multivariate data analyses, including Principal Component Analysis (PCA) and Partial Least Squares (PLS), revealed that this bioactivity is not driven by a single isolated compound, but rather by a synergistic combination of polar phenolic and aromatic metabolites, alongside supportive amino and organic acids. Molecular docking validated the inhibitory potential of key annotated metabolites, demonstrating that compounds such as catechin, β-sitosterol, and β-caryophyllene interact with the enzyme via reversible hydrogen bonding, van der Waals forces, and hydrophobic interactions, exhibiting binding affinities of − 6.09, − 6.34, and − 5.19 kcal/mol, respectively.ConclusionThese findings establish G. parvifolia as a viable, sustainable, and multi-targeted plant-based alternative to synthetic fat blockers. Its integration into functional foods and anti-obesity nutraceutical formulations offers significant potential for advancing preventive and personalized nutrition.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1007/s11306-026-02536-z. Introduction The global increase in obesity and metabolic illnesses statistics has influenced the demand for sustainable plant-based solutions to promote health while corresponding to food advancements (Galanakis, 2024; Kraak & Aschemann-Witzel, 2024). Obesity is the leading contributor to cardio-metabolic diseases, including coronary artery disease, stroke, diabetes, and heart-related diseases (Mohamed et al., 2023; Valenzuela et al., 2023). Thus, obesity prevention has become the priority in reducing the burden of the statistical growth of chronic diseases (Mohajan & Mohajan, 2023). One possible strategy is to suppress pancreatic lipase; to reduce the major enzyme responsible for dietary fat breakdown; to reduce fat absorption and regulate weight management (Subramaniyan & Hanim, 2025). FDA-approved synthetic lipase inhibitors, such as orlistat, are widely used to reduce fat absorption despite reported adverse events (Zhu et al., 2024). Garcinia species is distinguished among plant supplements due to its high concentration of bioactive compounds such as polyphenols and xanthones, which can help with metabolic health, especially Garcinia cambogia (Dong et al., 2023; Espirito Santo et al., 2020; John et al., 2019; Saikia et al., 2024). However, a lesser-known relative, Garcinia parvifolia (G. parvifolia), native to the tropical rainforests of Southeast Asia, is traditionally used in indigenous medicine and is known to contain diverse bioactive compounds, including flavonoids, and organic acids (Ahmad et al., 2021), which exhibit anti-diabetic (Ramadhan et al., 2023), larvicidal activity (Ee & Cheow, 2007), virucidal activity (Adnan et al., 2019; Ee et al., 2009), anti-bacterial (Muhammad et al., 2019), reducing atherosclerosis risk (Sani et al., 2016), antioxidant, and anti-Alzheimer’s effects (Ali Hassan et al., 2013). Despite its potential, systematic studies on its bioactivity, particularly its role in lipase inhibition, remain scarce. Correlated to these, metabolite profiling has yet to be established for in-depth analysis of metabolites responsible for influencing bioactivity (Wolfender et al., 2015), especially lipase inhibition. Metabolomic analysis is one of the omics studies, providing insight into the metabolome’s influence on metabolome influences towards drug response in the metabolic pathway (Pan et al., 2024). Current analytical techniques in metabolomics analysis include nuclear magnetic resonance (NMR) spectroscopy, high-performance liquid chromatography (HPLC), capillary electrophoresis-mass spectrometry, and gas chromatography-mass spectrometry (GC-MS) (Munjal et al., 2022). This study utilizes NMR-based profiling to characterize the metabolomic composition of G. parvifolia and establish correlations between specific compounds and their lipase inhibitory activity. Furthermore, multivariate statistical analysis was used to evaluate metabolic biomarkers that contribute to lipase inhibition activity. At present, no study has yet applied a multi-solvent extraction strategy combined with a metabolomic workflow to investigate the lipase inhibitors in G. parvifolia. Therefore, this study evaluated the responsible metabolite in contributing to lipase inhibitory activity in G. parvifolia through 1H NMR based metabolomic analysis and molecular docking analysis. Materials and methods Chemicals The extraction process utilized analytical grade solvents, including ethanol, n-hexane, acetone, and chloroform. Ethanol (95%) were obtained from R&M chemicals, acetonitrile, chloroform, n-Hexane, methanol, deuterated methanol-d4 (CD3OD, 99.8%), nondeuterated potassium dihydrogen phosphate (KH2PO4, pH 6.0), deuterium oxide (D2O, 99.9%) and trimethylsilylpropionic acid-d4 sodium salt (TSP) were acquired from Merck (Darmstadt, German) while porcine pancreatic lipase Type II phosphate buffer saline pH7.4, Tween 80, p-nitrophenyl butyrate (p-NPB), DMSO and orlistat were purchased from Sigma–Aldrich Co. (St. Louis, MO, USA). All other reagents were of analytical and HPLC grades. Plant materials Fruits of Garcinia parvifolia were collected in July 2024 from Penampang, Sabah, Malaysian Borneo, during the peak fruiting season. The sampling was conducted under approval from the Sabah Biodiversity Centre [Ref. No.: JKM/MBS.1000-2/2 JLD.19 (42)]. During the collection period, environmental conditions were characteristic of a tropical climate, with average temperatures of approximately 27–30 °C, high relative humidity, and moderate rainfall. Sampling was performed under stable weather conditions to minimize short-term environmental variability known to influence secondary metabolite profiles. Fruits were harvested at the fully ripe stage, as determined by external colour development and physical characteristics. Based on visual assessment, the fruits exhibited a maturation gradient from green (immature stage) to yellow-orange and subsequently to a uniform pinkish-red to deep red coloration at full ripeness. Fully ripe fruits selected for this study displayed consistent red to pink-red skin colour, slight softness upon gentle pressure, and ease of detachment from the pedicel, indicating optimal physiological maturity. To account for biological variability, samples were collected from seven independent trees of comparable age (approximately 6 years) grown under similar agronomic conditions. From each tree, 20 fruits were randomly harvested and pooled to constitute one biological replicate, resulting in a total of seven biological replicates (n = 7). All specimens were thoroughly documented, including precise location, habitat, and collection details. Taxonomic identification was conducted based on morphological characteristics (leaf morphology, fruit size, and floral traits) and verified through comparison with established references for the Garcinia genus. Identification was further confirmed by botanical authorities and cross-referenced with authenticated specimens deposited at the Borneensis Herbarium, Universiti Malaysia Sabah. A voucher specimen was deposited under accession number BORH 20201. Following authentication, fruit samples were processed by separating the peel and flesh, then the peel was sliced into small pieces. The peel was oven-dried at 40 °C for 72 h, then ground into powder and sieved through a 500 μm mesh to obtain a uniform particle size. The resulting powder was stored in a desiccator until further extraction. Oven drying was employed as a controlled sample-preservation method to facilitate rapid moisture removal, improve sample stability, and ensure consistent sample preparation for subsequent phytochemical extraction and metabolomic analysis (Bakir et al., 2023; Eapen et al., 2025). Sample extraction The extraction method utilized in this study was ultrasonic-assisted extraction (UAE), which has an efficiency of 85 to 97% at 20–80 °C for 10–60 min (Kumar et al., 2023). Dried samples were extracted with 70% (v/v) ethanol, acetone, chloroform, and hexane, which had different polarities, to determine the probable phytochemical composition depending on polarity. 100 g of dry powder sample was extracted with 500mL solvent using a sonicator for 20 min at room temperature. The extracted solution was poured out, then re-extracted again with the same amount of solvent, and stopped when the extracted solution came out clear. Accumulated extract solution up to 2.5 L per each solvent, then filtered through Whatman filter paper Grade 3 using Buchner funnels. Filtered extract solution was evaporated using a rotary evaporator to yield crude extract and further dried in sample bottles under a fume hood. Yield percentage recorded by crude extract per 100 g dry powder sample. Dried crude extracts were kept at −20 °C and ready for analysis. For bioassays, 10 mg/mL of stock solutions would be prepared in dimethyl sulfoxide (DMSO). In vitro pancreatic lipase inhibition assay The pancreatic lipase inhibition assay was conducted with slight modifications (Iswantini et al., 2021). 1 mg/mL crude extract was prepared from a stock solution with a final concentration of 10% DMSO solvent (v/v) and diluted to tested concentration (25–100 µg/mL) along with freshly prepared 1 mg/mL pancreatic lipase (PL) enzyme in Phosphate Buffer Saline (PBS) solution and 0.1% Tween 80. A 10 µM p-nitrophenyl butyrate (pNPB) stock solution was made by dissolving 20.9 mg of pNPB in 10 mL of acetonitrile. 50 µL of extract solutions of varying concentrations (25, 50, 75, and 100 µg/mL) were mixed with 50 µL of pancreatic lipase solution. The mixtures were incubated at 37 °C for 30 min. To start the reaction, 20 µL of the PNPB solution was added, and the reaction mixtures were incubated for an additional 30 min at 37 °C. The enzymatic activity was quantified by measuring the hydrolysis of pNPB to p-nitrophenol at 405 nm using a UV-visible spectrophotometer. To ensure precise integration with the NMR metabolomics data, the lipase inhibition assay was performed in heptaplicate (n = 7 technical replicates) for each solvent extract, and the mean inhibition percentage was recorded at concentrations of 25, 50, 75, and 100 µg/mL. DMSO concentration in final reaction mixture was less than 1% (v/v). Orlistat was used as a positive control as a standard PL inhibitor. The equation below was used to compute the inhibition % of lipase activity: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{Inhibitory}}\,{\mathrm{activity}}\left( \% \right) = \left[ {\left( {{\mathrm{B}} - {\mathrm{b}}} \right)/\left( {{\mathrm{A}} - {\mathrm{a}}} \right) \times {\mathrm{1}}00} \right]$$\end{document} where, B = enzyme activity with inhibitor. b = negative control with inhibitor. A = enzyme activity without inhibitor. a = negative control without inhibitor. Nuclear magnetic resonance (NMR) and multivariate data analysis NMR analysis was evaluated using a slightly modified NMR protocol from (Kim et al., 2010). In a 2.0 mL Eppendorf tube containing 50 mg of each extract, 0.75 mL of a 1:1 (v/v) mixture of methanol-d4 and potassium dihydrogen phosphate buffer (pH 6.0) prepared in deuterium oxide, containing 0.01% (w/v) trimethylsilylpropionic acid-d4 sodium salt, was added. The mixture was vortexed for 1 minute, ultrasonicated for 15 min, and centrifuged at 10,000 rpm for 15 min. An aliquot of 0.6mL supernatant was transferred into 5 mm NMR tubes. Each extract was replicated seven times and subjected to 1H NMR analysis. The 1H NMR and 2D (J-resolved) NMR determinations were implemented using 500 MHz Varion INOVA NMR spectrometer (Varian Inc, California, USA), functioning at a frequency of 499.887 MHz at room temperature (25 °C). Deuterium oxide was used as an internal lock. Each 1H NMR spectrum was obtained with a presat acquisition time of 3.53 min, comprising 64 scans with a spectral width of 20 ppm. The obtained spectra were processed using Chenomx software (version 5.1, Alberta, Canada) with uniform settings applied to all spectra. Spectral intensities were segmented into bins of equal width (0.04 ppm) within the chemical shift range of 0.50–10.00 ppm. The regions between 4.70 and 4.90 ppm, corresponding to the residual water signal, and 3.23–3.36 ppm, corresponding to residual methanol, were excluded from the analysis. The annotation of metabolites was conducted by comparing the distinctive signals detected in the 1H NMR spectra of the extract produced from G. parvifolia with those documented in the literature, the Chenomx database (Chenomx NMR Suite 7.7 library), and the human metabolome database (www.hmdb.ca). Additionally, the 2D J-resolved was utilized to aid in the annotation of metabolites. SIMCA-P software (v. 13.0, Umetrics, Umeå, Sweden) was utilized to analyse multivariate data using the PCA and PLS analysis methods. The Pareto scaling is used in NMR-based metabolomics studies (Maulidiani et al., 2017) to process complex metabolic data prior to multivariate data analysis (MVDA) because it helps reduce the influence of noise while still maintaining the structure of the data. In the PLS model, the X-matrix consisted of the binned ¹H NMR chemical shifts from the 28 technical replicates (n = 7 per solvent extract). The Y-variable was defined as the percent pancreatic lipase inhibition at a standardized extract concentration of 100 µg/mL. To ensure statistical validity and capture analytical variance, the Y-data was structured as a 28 × 1 vector of individual, non-averaged inhibition percentages from the parallel assays, establishing a direct 1-to-1 pairwise regression against the 28 spectral profiles in the X-matrix. Model validation was subsequently performed using a 100-iteration permutation test to ensure the predictive robustness of the resulting correlation. Molecular docking In silico docking was performed using AutoDock Vina (via Webina) to investigate binding interactions with pancreatic lipase (PDB ID: 1LPB). The structures of orlistat (HMDB0015215) and catechin (HMDB0002780) were sourced from the Human Metabolome Database (HMDB 5.0). Water molecules and existing ligands were removed from the 1LPB crystal structure. A focused grid box (40 × 40 × 40 points, 1.0 Å spacing) was centered on the catalytic triad residues (Ser152, His263) at coordinates x: 4.448, y: 27.955, z: 49.675. The lowest energy or most favorable binding pose for each ligand was analyzed. Statistical analysis For the plant extraction yields, evaluations were performed using three independent biological replicates (n = 7, representing the pooled fruit samples from seven distinct trees), with each biological replicate analyzed in triplicate. For the in vitro pancreatic lipase inhibition assay and the NMR metabolomic profiling, each of the four solvent extracts was analyzed in heptaplicate (n = 7 technical replicates), yielding a total of 28 matched analytical datasets. Data are expressed as mean ± standard deviation (SD). Statistical differences among different solvent groups were determined by one-way Analysis of Variance (ANOVA) followed by Duncan’s post-hoc test (p 0.05) using RStudio software (R version 3.6). Results and discussion Effect of extraction solvent on yield of extraction The different solvent used in extraction showed significant effect on the extraction yield with different solvents (Supplementary Table S1). 70% (v/v) ethanol showed the highest extraction ability followed by acetone, chloroform and hexane respectively. Based on the percentage, main composition in Garcinia parvifolia peel was polar compound as compared to non-polar compound thus the most effective solvent in extracting compound was polar solvent. Lipase inhibition activity of G. parvifolia extracts Lipase inhibition assay exhibited increasing trends in inhibitory efficacy in all extracts across 25 to 100 µg/mL tested concentrations while orlistat was tested as positive control with potent inhibition activity of 83% at 25 µg/mL. 70% (v/v) ethanol showed significantly higher lipase inhibition activity with IC₅₀ of 85.3 µg/mL when compared to other extracts. Acetone extracts displayed IC₅₀ of 106.4 µg/mL while chloroform and hexane exhibited 40.18 and 30.06 µg/mL at highest tested concentrations. To compare, orlistat was used as positive control with IC₅₀ of 17.5 µg/mL to validate the assay condition. 70% (v/v) ethanol is an effective solvent for extracting both polar and non-polar bioactive compounds with lipase inhibitory properties which allows it to extract a broad range of phytochemicals such as polyphenols, flavonoids, and xanthones known to occur in Garcinia species. Table 1 displays lipase inhibition activity percentage and IC₅₀ (µg/mL) for orlistat, 70% (v/v) ethanol, acetone, chloroform, and hexane extracts. Table 1Lipase inhibition activity and IC50 based on samplesSample70% (v/v) EthanolAcetoneChloroformHexaneOrlistatConcentration (µg/mL)9.32 ± 1.15a5.30 ± 0.45b2.10 ± 0.82c1.37 ± 0.39c-2534.33 ± 0.69h26.33 ± 0.31k22.40 ± 0.76l6.46 ± 0.97n83.07 ± 0.185044.04 ± 0.32d32.17 ± 0.68h31.47 ± 0.47i16.71 ± 0.79m83.2 ± 0.417554.16 ± 0.98b42.48 ± 0.62e37.49 ± 0.76g21.49 ± 0.57l89.42 ± 0.2010061.18 ± 0.51a45.67 ± 0.73c40.18 ± 0.54f30.06 ± 0.77i93.41 ± 0.20IC50(µg/mL)85.3106.4NDND17.5Data are mean ± standard deviation of heptaplicate (n = 7) determination. Values with different superscripts are significantly different at p < 0.05 Table 1Lipase inhibition activity and IC50 based on samplesSample70% (v/v) EthanolAcetoneChloroformHexaneOrlistatConcentration (µg/mL)9.32 ± 1.15a5.30 ± 0.45b2.10 ± 0.82c1.37 ± 0.39c-2534.33 ± 0.69h26.33 ± 0.31k22.40 ± 0.76l6.46 ± 0.97n83.07 ± 0.185044.04 ± 0.32d32.17 ± 0.68h31.47 ± 0.47i16.71 ± 0.79m83.2 ± 0.417554.16 ± 0.98b42.48 ± 0.62e37.49 ± 0.76g21.49 ± 0.57l89.42 ± 0.2010061.18 ± 0.51a45.67 ± 0.73c40.18 ± 0.54f30.06 ± 0.77i93.41 ± 0.20IC50(µg/mL)85.3106.4NDND17.5Data are mean ± standard deviation of heptaplicate (n = 7) determination. Values with different superscripts are significantly different at p < 0.05 Lipase inhibition activity and IC50 based on samples Data are mean ± standard deviation of heptaplicate (n = 7) determination. Values with different superscripts are significantly different at p < 0.05 The greater efficacy of the 70% (v/v) ethanol extract suggests that 70% (v/v) ethanol is the most efficient solvent among those tested for isolating lipase inhibitors from G. parvifolia fruit. This results from the enhanced extraction of secondary metabolites by highly polar solvents (Wang et al., 2023). Both the extraction efficiency and the resulting lipase inhibitory activity are significantly influenced by the polarity of the solvent used (Yang et al., 2025). The natural origin of these extracts may offer a safe alternative or complementary therapeutic route using natural origin with possibly fewer side effects. 1H NMR spectra of G. parvifolia extract and metabolites annotation Metabolites present in G. parvifolia extracts prepared using various solvents, including 70% (v/v) ethanol, acetone, chloroform, and hexane, were annotated through 1H NMR analysis. The spectral profiles demonstrate pronounced variability in both signal intensity and diversity depending on solvent polarity. Supplementary Fig. S1 shows the stacked 1H NMR spectra of the different G. parvifolia extracts. Thirty-four metabolites were annotated in G. parvifolia, including sugars, amino acids, organic acids, flavonoids, and sterols. In contrast, acetone (red), chloroform (green), and hexane (yellow) extracts displayed minimal signal intensities across all regions, especially from δ 2.5–5.5 ppm, indicating a lower presence of hydrophilic metabolites. The reproducibility and distinctiveness of these signals suggest that ethanol and acetone are suitable solvents for extracting metabolite-rich fractions for bioactivity screening. These spectral differences confirm that solvent polarity critically dictates the chemical diversity of recovered metabolites. Multiple signals were observed across the chemical shift range of 0.9–9.6 ppm, indicating the presence of diverse metabolite classes such as sugars, amino acids, organic acids, flavonoids, and sterols. Prominent signals were observed in the aliphatic region (δ 0.91–1.30 ppm), corresponding to terminal methyl and methylene protons. A strong multiplet region in the sugar group (δ 3.00 to 4.50) indicated the presence of various sugar moieties, including glucose (δ 3.23–4.20 ppm), xylose, and lactose, as well as choline (δ 3.20 ppm). These sugars exhibit overlapping multiplet patterns, which are typical of proton-rich saccharide structures. Distinct peaks in the region of δ 5.18–5.42 ppm were associated with the anomeric protons of sugar rings and possible polyols. The signals at δ 3.95 (m), δ 4.60 (d), and δ 6.99 (d) were ascribed to catechin. In the downfield aromatic region (δ 6.50–9.60 ppm), several low-intensity peaks were observed, consistent with phenolic compounds. Notably, a signal at δ ~ 6.66 ppm (d, J = 3.5 Hz) corresponds to catechin, a flavonoid known for its antioxidant properties (Nhu-Trang et al., 2023; Sheng et al., 2023). Other downfield signals at δ 7.30–8.20 ppm could be attributed to phenolic acids such as protocatechuic acid or gallic acid, which are common in Garcinia species (Kumatso & Sigge, 2024). Further supporting this, signals near δ 2.54–2.84 ppm were assigned to β-alanine and succinate, which are metabolites known to be involved in energy metabolism and reported to have moderate lipase inhibitory activity (Beyoğlu et al., 2024; Ding et al., 2022; Xue et al., 2020). In contrast, there were few unexpected metabolites including acetaminophen, caffeine and 3-Hydroxykynurenine. Several studies discuss the uptake of acetaminophen by plants from contaminated environments, such as soil or water containing pharmaceutical residues. For example, acetaminophen has been detected in plants like lettuce (Leitão et al., 2021), maize (Kudrna et al., 2023) and cucumber (Sun et al., 2019) when grown in contaminated conditions, but this is due to environmental exposure rather than natural production. The presence of caffeine in fruit peels such as dragon fruit has also been documented (Saputra et al., 2023).While direct reports for 3-hydroxykynurenine in fruit peels are limited, metabolites from the tryptophan-kynurenine pathway have been noted in mushroom (Sanjust et al., 2003). Supplementary Fig. S2 and Table 2 show the annotated metabolites along with their corresponding unique NMR signals obtained from various G. parvifolia extracts. Compound annotation was performed by comparing the 1H NMR spectra with those of reference metabolites available in the Chenomx database and the Human Metabolome Database (HMDB), both measured under conditions identical to those used for the extracts. Additionally, a 2D J-resolved experiment was employed to support the metabolite assignments Supplementary Fig. S3. Table 2The 1H-NMR signals for metabolites annotated in G. parvifolia extracted with different solvents corresponding multiplicity (s, singlet; d, doublet; t, triplet; dd, doublet of doublets; m, multiplet) and scalar coupling constant (J, Hz) valuesNo.Metabolite1H Chemical Shifts (multiplicity, J)Extract70% EtOHAcetoneCHCl₃Hexane1β-Sitosterol0.92 s++±±2Alanine1.50 d (J = 7.2)++±±3β-Caryophyllene1.62s±±±±4N-acetylcysteine2.05 s++±±5N-acetylglutamate2.09 s++±±6Acetaminophen2.12 s++±-7Acetoacetate2.26 s++±-8Methylamine2.61 s++±-9Aspartate2.82 dd (J = 5.5, 5.5)+++-10Succinate2.86 s+++-113-Hydromuconate2.98 d (J = 16.0)++++12Malate3.02 dd (J = 4.5, 9.0)±±--13β-alanine3.18 t (J = 8.5)++±±14Betaine3.26 s+++±15Caffeine3.34 s++±-16Choline3.38 s±±--17Trans-4-hydroxy-L-proline3.48 dd (J = 4.0, 8.0)++±-18Glucose3.50–3.66 m+++±193-Hydroxykynurenine3.70 d (J = 7.5)++±-20Glucose-6-phosphate3.78 t (J = 7.5)++±-21Anserine3.82 s±±--22Fucose3.86 dd (J = 3.2, 7.2)++±-23Polyols3.94 m±±--24Fructose3.98 dd (J = 3.2, 7.6)++±-25S-sulfocysteine4.02 dd (J = 4.5, 8.0)++±-26Isocitrate4.06 d (J = 7.0)++±-27Indole-3-lactate4.38 dd (J = 7.5, 7.5)++±-28Lactose4.46 d (J = 7.5)++±±29Xylose4.58–5.19 d (J = 3.2)++±±30Mandelate4.99 s++±±31Allantoin5.38 s++±-32Catechin6.66, 3.95 d (J = 3.5)++±-33Catechol6.86–6.93 m±±--344-Methylbenzaldehyde9.46 s±±--Symbols: + = detected, ± = weakly detected, - = not detected Table 2The 1H-NMR signals for metabolites annotated in G. parvifolia extracted with different solvents corresponding multiplicity (s, singlet; d, doublet; t, triplet; dd, doublet of doublets; m, multiplet) and scalar coupling constant (J, Hz) valuesNo.Metabolite1H Chemical Shifts (multiplicity, J)Extract70% EtOHAcetoneCHCl₃Hexane1β-Sitosterol0.92 s++±±2Alanine1.50 d (J = 7.2)++±±3β-Caryophyllene1.62s±±±±4N-acetylcysteine2.05 s++±±5N-acetylglutamate2.09 s++±±6Acetaminophen2.12 s++±-7Acetoacetate2.26 s++±-8Methylamine2.61 s++±-9Aspartate2.82 dd (J = 5.5, 5.5)+++-10Succinate2.86 s+++-113-Hydromuconate2.98 d (J = 16.0)++++12Malate3.02 dd (J = 4.5, 9.0)±±--13β-alanine3.18 t (J = 8.5)++±±14Betaine3.26 s+++±15Caffeine3.34 s++±-16Choline3.38 s±±--17Trans-4-hydroxy-L-proline3.48 dd (J = 4.0, 8.0)++±-18Glucose3.50–3.66 m+++±193-Hydroxykynurenine3.70 d (J = 7.5)++±-20Glucose-6-phosphate3.78 t (J = 7.5)++±-21Anserine3.82 s±±--22Fucose3.86 dd (J = 3.2, 7.2)++±-23Polyols3.94 m±±--24Fructose3.98 dd (J = 3.2, 7.6)++±-25S-sulfocysteine4.02 dd (J = 4.5, 8.0)++±-26Isocitrate4.06 d (J = 7.0)++±-27Indole-3-lactate4.38 dd (J = 7.5, 7.5)++±-28Lactose4.46 d (J = 7.5)++±±29Xylose4.58–5.19 d (J = 3.2)++±±30Mandelate4.99 s++±±31Allantoin5.38 s++±-32Catechin6.66, 3.95 d (J = 3.5)++±-33Catechol6.86–6.93 m±±--344-Methylbenzaldehyde9.46 s±±--Symbols: + = detected, ± = weakly detected, - = not detected The 1H-NMR signals for metabolites annotated in G. parvifolia extracted with different solvents corresponding multiplicity (s, singlet; d, doublet; t, triplet; dd, doublet of doublets; m, multiplet) and scalar coupling constant (J, Hz) values Symbols: + = detected, ± = weakly detected, - = not detected Discrimination of G. parvifolia extracts by PCA model The metabolite distribution presented in Table 2 provides a biochemical foundation for the clustering patterns observed in the PCA score and loading plots (Fig. 1). The PCA score plot (Fig. 1A) reveals a clear separation of G. parvifolia extracts based on solvent polarity, reflecting distinct differences in metabolite presence and abundance as determined by 1H NMR analysis. The first two principal components account for a cumulative 97.0% of the variance, with PC1 explaining 89.1% and PC2 accounting for 7.9%. Polar extracts includes 70% (v/v) ethanol and acetone form distinct clusters separate from the less polar extracts, in which chloroform and hexane are consistent with their broader metabolite profiles detailed in Table 2. Specifically, the 70% (v/v) ethanol and acetone extracts are positioned on the positive side of PC1, whereas chloroform and hexane extracts cluster closely on the negative side. This distribution suggests that PC1 primarily reflects the gradient from polar to non-polar chemical constituents. While chloroform and hexane extracts exhibit significant overlap, indicating similarity in their non-polar profiles, 70% ethanol and acetone extracts are further differentiated along PC2. Ethanol extracts localize in the lower-right quadrant, while acetone extracts occupy the upper-right quadrant, highlighting subtle chemical distinctions between these two polar solvents. All samples fall within the Hotelling’s T2 95% confidence ellipse, confirming the absence of significant outliers and supporting the statistical robustness of the model. Furthermore, the tight clustering of replicates within each solvent group demonstrates high reproducibility in extraction and analytical procedures. The PCA loading plot (Fig. 1B) identifies specific metabolites contributing to sample differentiation. Metabolites with strong positive loadings on PC1 predominantly include polar compounds such as glucose, fructose, succinate, betaine, choline, glucose-6-phosphate, and various amino acids. Their abundance in the 70% (v/v) ethanol and acetone extracts explains the positioning of these samples on the positive side of PC1. Conversely, metabolites such as β-sitosterol and β-caryophyllene, which exhibit weak detection or solvent-dependent presence, contribute less to PC1 and associate more closely with the less polar extracts. This observation aligns with the limited solubility of sterols and sesquiterpenes in aqueous alcohol solvents and their preferential extraction into organic solvents (Kumar et al., 2023; Simões et al., 2022). Fig. 1The PCA score (A) and loading (B) plots of different extracts of G. parvifolia The PCA score (A) and loading (B) plots of different extracts of G. parvifolia Phenolic compounds, including catechin and catechol, predominantly detected in the ethanol and acetone extracts, also influence PCA separation. Their aromatic proton signals (δ 6.6–6.9 ppm) serve as markers of polyphenolic content and are frequently reported as key contributors to antioxidant and enzyme-related bioactivities (Gao et al., 2025; Kopustinskiene et al., 2021). The PLS biplot (Fig. 2) integrates phytochemical composition with lipase inhibition activity, allowing direct correlation between metabolites listed in Table 2 and biological response. Extracts located closer to the lipase inhibition vector contain more metabolites that positively correlate with enzyme inhibition. In particular, the ethanol and acetone extracts are positioned nearer to this vector, which matches their higher diversity of metabolites. Several compounds found mainly in these extracts, such as catechin, catechol, 3-hydroxykynurenine, indole-3-lactate, and mandelate, are also aligned in the same direction as lipase inhibition on the biplot. This closeness suggests these metabolites likely contribute to the observed lipase inhibitory activity. Phenolic compounds annotated in G. parvifolia, especially catechin and catechol, are well-known inhibitors of digestive enzymes, including pancreatic lipase. These compounds can bind to the enzyme’s active site through hydrogen bonds and π–π interactions, reducing its efficiency (Sun et al., 2026; Wang et al., 2022). Their strong association with lipase inhibition in our PLS model supports earlier research showing that polyphenol-rich extracts have enhanced anti-obesity potential. Although detected in smaller amounts, aromatic metabolites like mandelate and 4-methylbenzaldehyde also contribute to the relationship with bioactivity. Aromatic acids and aldehydes have been reported to influence lipid digestion by inhibiting enzymes or competing with substrates (Huang et al., 2025; Yang et al., 2017). Besides phenolics, some amino acids and organic acids found in the ethanol and acetone extracts such as aspartate, succinate, isocitrate, and malate show moderate alignment with lipase inhibition in the biplot. While these primary metabolites are not direct lipase inhibitors, their presence may indicate overall metabolic richness, which could enhance synergistic effects on bioactivity (Hayami & Yamamoto, 2021). Betaine and choline, strongly detected in polar extracts, are linked to lipid metabolism regulation and might indirectly affect lipase activity or lipid absorption (Adjoumani et al., 2019; Soleimani et al., 2023). Their positive correlation in the PLS model suggests they play a supportive, though secondary, role in lipase inhibition. Fig. 2The PLS biplot describing the correlation among the phytochemical constituents of different G. parvifolia extracts and their lipase inhibition activity The PLS biplot describing the correlation among the phytochemical constituents of different G. parvifolia extracts and their lipase inhibition activity In contrast, hexane and chloroform extracts are positioned away from the lipase inhibition vector in the PLS biplot. Even though compounds like β-sitosterol and β-caryophyllene are present, their weak correlation with lipase inhibition indicates that polar phenolic and aromatic metabolites contribute more significantly to the bioactivity than lipophilic compounds alone in this dataset. Overall, this chemometric-metabolite approach provides strong evidence that the lipase inhibitory effect of G. parvifolia is a result of multiple components working together, closely linked to its overall phytochemical complexity rather than any single compound. The multivariate analysis shows that bioactive extracts are characterized by coordinated changes across various metabolite classes, including phenolics, aromatics, amino acids, and organic acids, that together influence both chemical distinction and biological response. The PLS model particularly highlights that lipase inhibition is associated with a group of metabolites rather than one dominant marker, suggesting additive or synergistic interactions within the extract. These findings emphasize the value of holistic, metabolomics-driven approaches to understanding plant bioactivity and indicate that the therapeutic potential of G. parvifolia comes from integrated metabolic patterns instead of isolated compounds. To validate the predictability and robustness of the PCA and PLS model for lipase inhibition, a permutation test with 100 iterations was conducted. The original PLS model demonstrated exceptionally strong goodness-of-fit and predictability, yielding cumulative values of R2X (cum) = 0.97, R2Y (cum) = 0.967, and Q2 (cum) = 0.963. The permutation results confirmed that the model was highly statistically significant and free from overfitting (Yang et al., 2017), as all permuted R2 and Q2 values were lower than those of the original model (Supplementary Fig. S4). Specifically, the regression lines showed intercepts of R2 = (0.0, −0.00317) and Q2 = (0.0, −0.266). The negative intercept for Q2 is a strong indicator of the model’s high predictive validity, confirming that the chemical variations observed across the G. parvifolia extracts are directly correlated with their biological inhibitory effects on lipase. To further evaluate the quantitative contribution of individual metabolites to the predictive power of the PLS model, a Variable Importance in Projection (VIP) analysis was conducted (Supplementary Fig. S5). The VIP plot successfully identifies the most influential chemical markers that drive the differentiation among the varying solvent extracts and strongly correlate with lipase inhibitory activity. The VIP analysis revealed that the bioactivity of the extracts is heavily influenced by a diverse array of metabolites rather than a single dominant compound. Notably, key phenolic and aromatic compounds—such as catechin (δ 4.54, 6.66), catechol (δ 6.86), mandelate (δ 5.02), and indole-3-lactate (δ 4.34)—exhibited significant VIP values. This robustly supports the PLS biplot findings, confirming these compounds as the primary drivers of pancreatic lipase inhibition. Furthermore, the VIP plot highlights the critical supportive roles of various primary metabolites in distinguishing the highly active polar extracts (70% ethanol and acetone) from the less active non-polar extracts. Amino acids and derivatives, such as aspartate (δ 2.82), β-alanine (δ 3.18), trans-4-hydroxy-L-proline (δ 3.46), and 3-hydroxykynurenine (δ 3.7), along with organic acids like succinate (δ 2.34) and isocitrate (δ 4.06), ranked as highly important variables. The prominent VIP scores of this wide spectrum of hydrophilic and intermediate-polarity metabolites provide compelling statistical evidence that the anti-obesity potential of G. parvifolia is governed by a complex, multi-component synergy.” Molecular docking To explore the lipase inhibitory mechanism of G. parvifolia, representative metabolites annotated via metabolomics were subjected to molecular docking. Polar ligands were expected to demonstrate superior binding affinity compared to non-polar compounds due to enhanced hydrogen bonding. Catechin, β-sitosterol, and β-caryophyllene were analyzed due to different metabolite classes and the reference standard, orlistat (Supplementary Table S2). The interaction of orlistat is unique as it functions as an irreversible inhibitor to reduce fat absorption (Subramaniyan & Hanim, 2025). Its primary mechanism involves the formation of a stable covalent ester bond between the drug’s beta-lactone ring and the hydroxyl group of the catalytic residue Ser152 (Jagtap & Paul, 2025). Additionally, its long hydrocarbon tail extends into the enzyme’s hydrophobic groove, stabilizing the complex through extensive van der Waals interactions with lipophilic residues such as Phe77 and Phe215, effectively mimicking a triglyceride substrate before permanently disabling the enzyme (Tran et al., 2024). Thus, multi-ligand interaction of orlistat with lipase made it gold standard in clinical treatment for lipase inhibitor. Meanwhile, catechin acts as a polar inhibitor. Its multiple hydroxyl (-OH) groups act as hydrogen bond donors and acceptors. Docking simulations typically reveal a network of hydrogen bonds formed directly with the catalytic triad residues, specifically Ser152 and His263 (Ahmed et al., 2022). This suggests that catechin inhibits the enzyme by “capping” the active site or stabilizing a non-functional conformation of the protein (Cao et al., 2024). β-sitosterol relies almost exclusively on Van der Waals forces. It occupies the large, greasy substrate-binding channel of the pancreatic lipase (Sadino et al., 2024). The molecule wedges itself into this pocket, interacting with non-polar residues in which Asp72, Val76, Thr80, and Phe84, thereby physically blocking the entry of triglycerides to the active site via steric hindrance. Although β-sitosterol exhibited a slightly more favorable binding affinity than catechin (− 6.34 kcal/mol versus − 6.09 kcal/mol), catechin remains the principal candidate driving the extract’s bioactivity. This is supported by the poor in vitro performance of the lipophilic hexane extract (IC₅₀ > 100 µg/mL), where compounds like -sitosterol are prone to precipitation in the aqueous assay buffer. In contrast, the highly active polar extracts (70% ethanol and acetone) are rich in catechin; which is unlike the non-specific steric hindrance of β -sitosterol, directly neutralizes the enzyme by forming a stable network of hydrogen bonds with the crucial catalytic triad residues (Ser152, His263, and Asp176). β-caryophyllene, a sesquiterpene, exhibits a binding mode like β-sitosterol but with significantly lower affinity. Its interactions are purely hydrophobic, occurring within the lipophilic cleft. However, due to its smaller molecular volume compared to sterols or orlistat, it forms fewer contact points with the enzyme residues, resulting in weaker binding energy and limited stability within the active site. Even though limited study on lipase inhibition of β-caryophyllene towards pancreatic lipase, β-caryophyllene was found to inhibit monoacylglycerol lipase activity (Klawitter et al., 2024) and adipose triglyceride lipase (Hashiesh et al., 2020). In summary, multi-ligand interaction between lipase and secondary metabolites found in G. parvifolia support the experimental data of lipase inhibition assay in this research. Figure 3 displays 3D visual representations of in silico docking of orlistat, catechin, β-sitosterol, and β-caryophyllene. Fig. 3In silico docking of orlistat, catechin, β-sitosterol, and β-caryophyllene In silico docking of orlistat, catechin, β-sitosterol, and β-caryophyllene Limitations This study has several limitations. Bioactivity was assessed only through in vitro assays and in silico docking, without cellular or in vivo validation, so extrapolation to a physiological context should be made cautiously. Docking predictions for catechin, β-sitosterol, and β-caryophyllene were not confirmed with pure standards, and the docked complexes were not subjected to molecular dynamics simulations to assess pose stability over time. Docking was limited to three metabolites rather than the full top-VIP-ranked set, potentially excluding other contributing constituents. NMR-based metabolite identification was qualitative, based on chemical shifts rather than relative integration. Finally, samples were oven-dried rather than lyophilised, as a freeze-dryer was not routinely accessible to the research team. Future work should address these gaps. Conclusion In conclusion, this study provides valuable in vitro and in silico evidence of the pancreatic lipase inhibitory potential of G. parvifolia using a ¹H NMR-based metabolomics approach. The 70% (v/v) ethanol extract exhibited the strongest in vitro lipase inhibitory activity with an IC₅₀ of approximately 85.3 µg/mL compared to the acetone, chloroform, and hexane extracts. Chemometric profiling (PCA and PLS) successfully annotated and correlated key secondary metabolites—namely catechin, β-sitosterol, and β-caryophyllene—with this observed inhibitory effect. Furthermore, in silico molecular docking corroborated these findings, revealing favorable ligand binding interactions with pancreatic lipase via hydrogen bonding, van der Waals forces, and hydrophobic pocket capping. While these preliminary findings suggest that G. parvifolia holds potential as a sustainable natural source of lipase inhibitors, further in vivo animal studies and clinical trials are essential to validate its safety, bioavailability, and practical efficacy as a functional food ingredient or anti-obesity nutraceutical. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2