Discovery of anti-inflammatory compounds from the stem bark of Garcinia latissima through in vitro and in silico approaches

Abstract

Garcinia latissima Miq. has been traditionally used by local communities in Indonesia for wound healing and to relieve itching. A phytochemical investigation of the stem bark of G. latissima led to the isolation of 12 tocotrienol and triterpenoid derivatives, including two previously undescribed metabolites, δ-(E)-deoxy-amplexichromanoyl acetate (1) and (20S,24S)-20,24-epoxylanostane-3β,25-diol (2), whose structures were elucidated using HRESIMS and NMR spectroscopic analyses. Among the isolates, (20R)-eupha-8,25-diene-3β,24ξ-diol (11) suppressed nitric oxide production in LPS-stimulated RAW 264.7 cells by more than 60% at concentrations of 10 and 50 µM, without significant cytotoxicity (cell viability > 80%). It also reduced the release of several inflammatory mediators, particularly MCP-1, as determined by a membrane antibody array. ELISA confirmed a significant decrease in MCP-1 levels, up to 2.3- and 2.8-fold at 3 and 24 h, respectively, following pre-treatment with compound 11 at 50 µM. Molecular docking and molecular dynamics simulations indicated a strong binding affinity of compound 11 to MD2, TAK1, and NF-κB1, key proteins in the TLR-4 signaling pathway, with predicted affinities higher than those of reference ligands. ADMET analysis further suggested favorable drug-like properties, including high predicted oral bioavailability and minimal toxicity. These findings suggest that compound 11 has promising anti-inflammatory potential, warranting further experimental studies to elucidate its mechanism of action and to validate its effects in additional inflammation-related bioassays.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-30017-1.

Introduction

Inflammation is a crucial defense mechanism of the immune system, activated in response to harmful stimuli by such as pathogens, toxins, or tissue injury. It involves the production of various inflammatory mediators such as cytokines, prostaglandins, and nitric oxide (NO). These mediators are synthesized through the action of key enzymes, including cyclooxygenase (COX) and nitric oxide synthase (NOS). However, excessive or uncontrolled production of these signaling molecules can result in chronic inflammation, which plays a central role in the immunopathogenesis of various diseases, including cancer, atherosclerosis, neurodegenerative disorders, and metabolic syndromes¹. Recent evidence have shown that an overactive immune response in acute diseases. For example, an excessive cytokine response, often termed a “cytokine storm”, has been implicated in severe COVID-19 and is associated with increased risk of respiratory failure and poor clinical outcomes². These insights have reinforced the need for therapeutic interventions that can modulate the inflammatory response. Conventional anti-inflammatory therapies, including non-steroidal anti-inflammatory drugs (NSAIDs), glucocorticoids, and immunosuppressants, have shown clinical efficacy but are often associated with non-specific actions and undesirable side effects^3,4. These limitations have prompted growing interest in exploring safer and more effective alternatives derived from natural sources. Medicinal plants have long been used in traditional medicine to treat inflammatory conditions, owing to their lower side effects profiles and the presence of structurally diverse bioactive components that serve as pharmacophores with various modes of action⁵. For instance, salicylic acid, originally isolated from the bark of the willow tree, was one of the first analgesic and antipyretic agents to be introduced, and it was later marketed in its acetylated form, aspirin, as a COX blocker⁶. Similarly, plumericin, a spirolactone iridoid derivative isolated from Himatanthus sucuuba (Spruce) Woodson (Apocynaceae), has been shown to exhibit anti-inflammatory effects both in vitro and in vivo through inhibition of the NF-κB pathway⁷. These examples underscore the therapeutic potential of phytochemicals and highlight the importance of continued exploration of bioactive compounds from medicinal plant.

Garcinia latissima Miq. (Clusiaceae) is an evergreen tree that can grow up to forty meters tall and is native to the eastern islands of Indonesia (including Sulawesi, Maluku, and Papua) as well as Papua New Guinea. In Maluku, local communities refer to G. latissima as Dolo Magota and traditionally use various parts of the plant for wound healing and the treatment of skin irritation^8,9. Previous studies have reported that methanol and ethyl acetate extracts of the leaves, fruits, and stem bark exhibit antibacterial activity against Bacillus subtilis and Pseudomonas aeruginosa^9–11. Antioxidant activity has also been observed in the methanolic leaf extract, as assessed by the 2,2-diphenyl-1-picrylhydrazyl (DPPH) assay¹¹. Phytochemical investigations of G. latissima has led to the identification of structurally diverse metabolites, including biflavonoids, xanthones, chromones, polyprenylated benzophenones, and triterpenoids^12–16.

Despite its ethnomedicinal use in treating inflammation-related conditions, limited studies have examined the anti-inflammatory potential of specific compounds from G. latissima. To address this gap, we conducted a phytochemical investigation of the stem bark, resulting in the isolation of twelve compounds, seven tocotrienol and five triterpenoid derivatives. These compound classes were selected based on prior evidence of their anti-inflammatory properties, tocotrienols, a form of vitamin E, are known to mitigate oxidative stress caused by factors associated with tissue inflammation, such as endogenous free radicals (e.g., ROS, NO, and superoxide anions) and UV radiation. They also downregulate the gene expression of pro-inflammatory proteins such as NF-κB, TNF-α, and iNOS¹⁷. Similarly, triterpenoid molecules are well- established anti-inflammatory agents that exert immunomodulatory effects by inhibiting multiple cytokine targets, particularly Th1- associated pro-inflammatory proteins, thereby contributing to their immunomodulatory activity¹⁸. Accordingly, the compounds isolated in this study were screened for their anti-inflammatory potential using nitric oxide inhibition assays, and the results are discussed herein.

Building on the in vitro screening results, we further investigated the potential mechanism insights into the anti-inflammatory activity of the most active compounds by exploring their interactions with key inflammation-related proteins. These proteins, particularly those within the TLR4 signaling pathway, are critical regulators of immune responses and are known to influence the expression of chemokines such as MCP-1, which plays a key role in monocyte recruitment and inflammatory progression. Understanding how bioactive ligands engage with these targets is essential for rational drug development. Computational methodologies such as molecular docking and molecular dynamics (MD) simulations provide powerful tools to analyze these interactions at the atomic level, offering insights into binding affinity, structural stability, and dynamic behavior. In this study, we employed both approaches to examine the molecular interactions and dynamic properties of the active compounds with selected protein targets involved in TLR4-mediated signaling.

Results and discussion

Structural Elucidation of the isolated compounds

A series of chromatographic separation of the EtOAc extract from stem bark of G. latissima led to the isolation of of two previously undescribed compounds: a tocotrienol ester, δ-(E)-deoxy-amplexichromanoyl acetate (1), and a lanostane triterpenoid, (20S,24S)−20,24-epoxylanostane-3β,25-diol (2), along with ten known derivatives. The known compounds include δ-garcinoic acid (3)¹⁹, γ-garcinoic acid (4)²⁰, δ−3,4-dehydrogarcinoic acid (5)²¹, menzoquinone (6)²², δ,δ-bigarcinoic acid (7)²³, δ,δ-bi-O-garcinoic acid (8)²³, kansuinone (9)²⁴, rel (8R,9S,20R)-lanost-24-ene-3β,20-diol (10)²⁵, (20R)-eupha-8,25-diene-3β,24ξ-diol (11)²⁶, and eupha-8,24-diene-3β,11β-diol-7-one or euphorol B (12)²⁷ (Fig. 1). The structure of previously reported known metabolites was determined by comparison of their NMR spectroscopic data with the literature.

Fig. 1.

Structures of isolated tocotrienol and triterpenoid (1−12) from the stem bark of G. latissima.

Compound 1 had a molecular formula of C₂₉H₄₂O₄ according to the HRESIMS data at m/z 472.3421 [M + NH₄]⁺ (calcd for C₂₉H₄₆NO₄⁺, 472.3421) and 395.2945 [M − CH₃COOH + H]⁺ (calcd for C₂₇H₃₉O₂⁺, 395.2945), which suggested 9 indices of hydrogen deficiency. The ¹H NMR spectrum displayed resonances of two meta-coupled aromatic protons at δ_H 6.48 (d, J = 3.0 Hz, H-7) and 6.38 (d, J = 3.0 Hz, H-5) and protons of two methylene groups at δ_H 2.69 (td, J = 7.0, 3.0 Hz, H-4) and 1.70−1.82 (m, H-3) (Table 1) whose latter pair were mutually coupled in the COSY correlation (Fig. 2a). The HMBC cross-peaks of H-5 and H-7 with two oxygenated sp² carbons C-6 (δ_C 148.0) and C-8a (δ_C 146.0), H-3 and H-4 with C-4a (δ_C 121.3) and oxygenated sp³ C-2 (δ_C 75.4), and H-4 with C-5 (δ_C 112.7) and C-8a (δ_C 146.0) generated a 6-chromanol ring, while correlations of a deshielded methyl proton at δ_H 2.12 (s, H-26) with C-7 (δ_C 115.7), C-8 (δ_C 127.5), and C-8a were indicative of its C-8 methylation (Fig. 2a). The farnesyl moiety in 1 was identified by typical signals of three olefinic protons at δ_H 5.12 (t, J = 7.0 Hz, H-11), 5.09 (t, J = 7.0 Hz, H-15), and 5.44 (t, J = 7.0 Hz, H-19) in which each proton coupled with methylenes H-10/H-9, H-14/H-13, and H-18/H-17, respectively, based on the COSY experiment. The HMBC correlations of methyls H-24 (δ_H 1.59, s) with C-11 (δ_C 124.5), C-12 (δ_C 135.2), and C-13 (δ_C 39.8), H-23 (δ_H 1.58, s) with C-15 (δ_C 124.7), C-16 (δ_C 134.6), and C-17 (δ_C 39.2), and H-22 (δ_H 1.64, s) with C-19 (δ_C 129.9), C-20 (δ_C 130.0), and C-21 (δ_C 70.6) supported the presence of the side chain. The attachment of the farnesyl and a methyl group at C-2 were established from the HMBC correlations of H-9 (δ_H 1.51−1.66, m) and H-25 (δ_H 1.26, s) to that carbon.

Table 1.

¹H (500 MHz) and ¹³C (125 MHz) NMR spectroscopic data of compounds 1 and 2 in CDCl₃ (δ in ppm).

Position	1		Position	2
	δC	δH (J in Hz)		δC	δH (J in Hz)
O-1			1	39.2	1.68b
2	75.4				0.97b
3	31.5	1.70−1.82, m	2	27.4	1.56−1.62, m
4	22.6	2.69, td (7.0, 3.0)	3	79.1	3.20, dd (11.5, 5.0)
4a	121.3		4	39.1
5	112.7	6.38, d (3.0)	5	56.0	0.73, dd (12.0, 2.5)
6	148.0		6	18.4	1.52b
7	115.7	6.48, d (3.0)			1.42b
8	127.5		7	27.1	1.78b
8a	146.0				1.20b
9	39.6	1.51−1.66, m	8	43.0	1.63b
10	22.3	2.10, m	9	50.9	1.31b
11	124.5	5.12, t (7.0)	10	37.3
12	135.2		11	21.9	1.50b
13	39.8	1.97, m			1.18b
14	26.7	2.06, m	12	35.4	1.53b
15	124.7	5.09, t (7.0)			1.27, ddd (14.5, 13.5, 3.0)
16	134.6		13	40.5
17	39.2	1.99, m	14	50.2
18	26.5	2.11, m	15	31.6	1.47b
19	129.9	5.44, t (7.0)			1.05, ddd (11.5, 8.5, 1.5)
20	130.0		16	26.0	1.73b
21	70.6	4.45, s			1.29b
22	14.1	1.64, s	17	49.9	1.85, dd (10.5, 6.0)
23	16.0a	1.58, s	18	15.6	0.97, s
24	16.1a	1.59, s	19	16.4	0.84, s
25	24.2	1.26, s	20	86.7
26	16.2	2.12, s	21	27.5	1.14, s
27	171.4		22	34.8	1.87b
28	21.2	2.07, s			1.66b
			23	26.5	1.84b
					1.79b
			24	86.4	3.63, dd (10.0, 5.0)
			25	70.4
			26	28.0	1.19, s
			27	24.2	1.11, s
			28	28.1	0.97, s
			29	15.5	0.77, s
			30	16.5	0.87, s

^aInterchangeable chemical shifts; ^boverlapping signals.

Fig. 2.

(a) Key COSY (blue line), HMBC (red arrow), and NOESY (blue dash arrow) correlations of 1; (b) Key COSY (blue line) and HMBC (red arrow) correlations of 2; (c) Key NOESY (blue dash arrow) correlations of 2.

The NMR data analysis suggested that 1 closely resembled those of δ-(E)-deoxy-amplexichromanol, a tocotrienol derivative isolated from G. amplexicaulis Vieill. ex Pierre²⁸. The difference was that the signals corresponding to an oxymethylene group at the terminal isoprene unit were more deshielded in 1, appearing at δ_H 4.45 (s, H-21, Δδ_H + 0.45) and δ_C 70.6 (C-21, Δδ_C + 1.50), compared to those of the known compound. Additional resonances for a carbonyl carbon at δ_C 171.4 (C-27) and methyl group at δ_H/δ_C 2.07 (s, H-28)/21.6 were also observed (Table 1). The HMBC cross-peaks of H-21 and H-28 to C-27 was indicative of the O-acylated C-21 (Fig. 2a). The E configuration of double bonds at the farnesyl chain were determined based on NOESY correlations of H-10/H-24, H-14/H-23, H-18/H-22, and H-19/H-21. The absolute configuration at C-2 was assigned to be 2R on the basis of a negative specific rotation ‒20.2 and a positive Cotton effect at 208 nm in the ECD spectrum (Figure S8, Supporting Information) in comparison with δ-garcinoic acid^23,29. It is also supported by the fact that naturally occurring tocotrienols exclusively contain 2R,14E,16E-configuration²⁹. Therefore, the structure of 1 was concluded to be δ-(E)-deoxy-amplexichromanoyl acetate.

An ion peak at m/z 443.3883 [M − H₂O + H]⁺ (calcd for C₃₀H₅₁NO₂⁺, 443.3884) in the HRESIMS data of 2 indicated the molecular formula C₃₀H₅₂O₃, in accordance with the ¹H and ¹³C NMR data. Further 1D NMR and HSQC spectra analysis characterized 30 carbon signals in aliphatic regions and protons resonating at 0.73−3.63 ppm, including eight methyl singlets at δ_H 0.77 (H-29), 0.84 (H-19), 0.87 (H-30), 0.97 (H-18 and − 28), 1.11 (H-27), 1.14 (H-21), and 1.19 (H-26) and two oxymethines at δ_H 3.20 (dd, J = 11.5, 5.0 Hz, H-3) and 3.63 (dd, J = 10.0, 5.0 Hz, H-24), which were typical for a saturated triterpenoid compound (Table 1)²⁵. The tetracyclic-type structure of 2 was indicated by the HMBC correlations of five methyls to quaternary sp³ carbons, including H-18 with C-13 (δ_C 40.5), H-19 with C-10 (δ_C 37.3), H-30 with C-14 (δ_C 50.2), and H-28 and H-29 with C-4 (δ_C 39.1), while additional cross-peaks of the latter gem-dimethyl protons to oxymethine C-3 (δ_C 79.1) confirmed the hydroxylation of that carbon in the ring A (Fig. 2b)³⁰. The COSY cross-peaks of H-22/H-23 and H-23/H-24 and HMBC correlations of methyl H-21 with C-17 (δ_C 49.9), C-20 (δ_C 86.7), and C-22 (δ_C 34.8) suggested the side chain modification of 2 to form a furan ring linked to the tetracyclic skeleton at C-17, which was in agreement with its HRESIMS data providing 5 indices of hydrogen deficiency. The HMBC correlations of two methyls at H-26 and H-27 with oxygenated carbons C-24 (δ_C 86.4) and C-25 (δ_C 70.4) permitted the attachment of a 25-hydroxyisopropyl group at C-24. This assignment was further supported by the downfield shifts of the proton singlets for H-26 and H-27, attributable to the effect of the hydroxy group at C-25. Therefore, the NMR data of 2 resembled those of rel (3α,8R, 9S,20R,24S)−20,24-epoxytirucalla-3,25-diol²⁵.

The relative configuration of 2 was assigned based on its coupling constant and NOESY correlation analysis. The large J value of 11.5 Hz for H-3 indicated that this proton axially oriented. NOESY cross-peaks of H-3/H-5, H-5/H-9, and H-9/H-30 confirmed that these protons and CH₃−30 were cofacial and α-oriented, while the observed H-19/H-18, H-18/H-8, and H-30/H-17 correlations in 2 were consistent with those reported for lanostane triterpenoid structures (Fig. 2c). In the tetrahydrofuran moiety, the NOESY cross-peak of H-17/H-21 indicated that CH₃−21 was α-oriented. Although the orientation of H-24 could not be clearly assigned due to the absence of NOESY correlations, previous studies have shown that the relative configuration at C-20 and C-24 can be determined from differences in carbon chemical shifts. Compounds with a 20S,24S configuration, such as (20S,24S)-epoxydammarane-3α,25,28-triol, exhibited δ_C values of 27.0–27.2, 34.6–34.8, and 86.5 for C-21, C-22, and C-24, respectively. In contrast, those with a 20S,24R configuration, such as (20S,24R)-epoxydammarane-3α,25,28-triol, displayed δ_C values of 23.6–23.9 (C-21), 35.5–35.7 (C-22), and 83.2 (C-24)³¹. Based on this pattern, compound 2 was assigned an S configuration at both C-20 and C-24, as indicated by its carbon resonances for C-21 (δ_C 27.5), C-22 (δ_C 34.8), and C-24 (δ_C 86.4). Therefore, the structure of 2 was determined to be (20S,24S)−20,24-epoxylanostane-3β,25-diol. However, the absolute configuration remains undetermined due to the insufficient sample quantity for rigorous analysis.

The 1D and 2D NMR data of 10 were well-matched with those of a described tirucallane derivative, rel (8R,9S,20R)-tirucall-24-ene-3β,20-diol²⁵. However, after careful reanalysis of their NMR spectra (Table S1 and Figure S29), especially for HMBC and NOESY correlations, the positional misassignment of CH₃−18 and CH₃−30 was observed in the previous work leading to the incorrect conclusion of tirucallane triterpenoid structure. In both compounds, the HMBC cross-peaks of methyl protons at δ_H 0.96 with C-12 (δ_C 35.4), C-13 (δ_C 40.5), C-14 (δ_C 50.5), and C-17 (δ_C 50.0) and at δ_H 0.87 with C-8 (δ_C 42.4), C-13, C-14, and C-15 (δ_C 31.6) were observed, suggesting the two methyls to be located at C-18 and C-30, respectively (Figure S29a). The NOESY spectra were also found to be identical, including the correlations H-3/H-5, H-5/H-9, H-9/H-30, H-30/H-17, H-19/H-18, H-18/H-8, which indicated that the two compounds shared the same lanostane structure with the hydroxy group at C-3 being equatorially-oriented based on the ³J_H−2,H−3 value of 11.5 Hz (Figure S29b). The relative configuration at C-20 (20R) was determined by the comparable proton chemical shift of H-21 (δ_H 1.14) in compound 10 and in euphorfistrine C, a known compound possessing the same relative configuration in both the tetracyclic core and the C-20 side chain³². Additionally, an NMR study of an epimeric pair of 3β,6α-dihydroxy-5α-cholest-9(11)-en-23-one at C-20 demonstrated that the 20R isomer exhibited a downfield-shifted H-21 signal (δ_H 1.04) compared to its 20S counterpart (δ_H 0.93, H-21)³³. The NOESY correlations of H-21/H-12 eq and H-22/H-18 in 10 further corroborated the C-20 configuration³³. Therefore, the previous structure was thus revised to be (8R,9S,20R)-lanost-24-ene-3β,20-diol (10) (Figure S29c).

In vitro anti-inflammatory evaluation

Macrophages release inflammatory mediators, including inducible nitric oxide synthase (iNOS), which produces nitric oxide (NO) in response to infection. NO not only acts as a crucial defense molecule against pathogens but also plays a key role in regulating various cell-mediated immune processes during inflammation³⁴. Therefore, identifying compounds that can effectively suppress NO production is an key strategy for screening potential anti-inflammatory candidates. In this study, compounds isolated from G. latissima were tested at final concentrations of 10 and 50 µM, except for compound 6, which was excluded due to decomposition, to evaluate their NO inhibitory activities in LPS-treated RAW 264.7 cells. The results are presented in Fig. 3a.

Fig. 3.

In vitro anti-inflammatory activity of the isolated compounds in LPS-stimulated RAW 264.7 cells; (a) Inhibitory effects of the compounds on nitric oxide (NO) production at concentrations of 10 and 50 µM. Compounds showing cytotoxic at 50 µM (cell viability < 80%) were excluded from the graph. Bay 11–7082 (5 µM) was used as the positive control. (b) Effect of 11 (50 µM) on the production of inflammatory mediators as determined by a mouse inflammation antibody array; (c) Quantification of MCP-1 level by ELISA at 3 and 24 h following treatment of 11 at of 10 and 50 µM (*p < 0.05).

Among the garcinoic acids (GAs), only compound 5, which contains a 2H-chromene unit, inhibited NO production by more than 50% at both 10 and 50 µM without inducing cytotoxicity (cell viability > 80%, Figure S30). Sargaol, a derivative of compound 5 lacking the terminal carboxylic group, and geranylgeranyltoluquinol (GGT), which possesses a ring-opened pyran moiety and a decarboxylated side chain (Figure S28), were also tested for comparison. However, neither compound exhibited notable inhibitory effects at 10 µM, and both showed cytotoxicity at 50 µM. Within the triterpenoid series, euphane 11 markedly suppressed NO production by 63.4 ± 10.1% and 83.2 ± 2.8% at 10 and 50 µM, respectively, followed by the weaker inhibitory activity of compound 10. The remaining triterpenoids displayed cytotoxicity at 50 µM. Preliminary structure–activity relationship (SAR) analysis of the isolated compounds indicated that, among garcinoic acid derivatives, substitution of the 21-COOH group with methyl or acyl functionalities, as well as the ring-opening modification of tocotrienols, resulted in unfavorable cytotoxic effects. In contrast, the presence of an unsaturated chromane unit, as seen in compound 5, enhanced NO inhibitory activity. Although the major compound δ-GA (3) exhibited only weak NO inhibitory activity in this study, it has previously been reported to exert anti-inflammatory effects through other mechanisms. These include the suppression of pro-inflammatory mediators (IL-6, IL-1β, TNF-α, COX-2) in LPS-induced RAW 264.7 cells³⁵; inhibition of microsomal prostaglandin E synthase-1 (mPGES-1) in IL-1β-treated A549 cells²⁰; downregulation of phospho-p65 and phospho-IκBα in SARS-CoV-2 spike protein S1-induced human peripheral blood mononuclear cells (PBMCs)³⁶; and inhibition of 5-lipoxygenase in a cell-free assay³⁷. In the triterpenoid group, the active compound 11, the most active in this study, was previously reported to potently inhibit human 11β-HSD1, an enzyme that converts inactive glucocorticoid cortisone to active cortisol in the liver, adipose tissue, and brain, with an IC₅₀ value of 0.03 µM³⁸.

In addition to nitric oxide, various cytokines act as key regulators of cell-to-cell communication, playing essential roles in initiating and modulating inflammatory responses¹. Understanding the inhibitory effects of natural compounds on these mediators is crucial for elucidating their mechanisms of action and for guiding the development of targeted therapies to manage inflammatory diseases more effectively. To further investigate the anti-inflammatory profile, compound 11, the most active NO inhibitor, was subjected to an antibody-pair-based cytokine assay, whereas compound 5 was excluded due to insufficient quantity. In LPS-stimulated RAW 264.7 cells (Fig. 3b), cytokine production was notably increased, particularly GM-CSF and RANTES, which were elevated by approximately 3.0-fold compared with the positive control. Treatment with compound 11 resulted in a pronounced downregulation of MCP-1, with levels reduced by up to 3.1-fold, followed by GM-CSF, SDF-1, IL-6, TNF-α, M-CSF, and G-CSF, each exhibiting at least a 1.1-fold decrease. Conversely, the expression of RANTES, sTNFR2, LIX, MIP-1α, TCA-3, and sTNFR1 was found to be elevated following treatment with compound 11.

To confirm the inhibitory effect of compound 11 on specific cytokine expression, the most strongly suppressed cytokine, MCP-1, was selected for further analysis using ELISA to determine its protein levels at 3 and 24 h after compound exposure. In LPS-induced RAW 264.7 cells, MCP-1 production reached approximately 400 pg/mL at 3 h and increased to ~ 62,000 pg/mL at 24 h (Fig. 3c). Notably, treatment with compound 11 at concentrations of 10 and 50 µM significantly reduced MCP-1 levels by 1.9- and 2.3-fold, respectively, at 3 h (p < 0.05), and maintained its inhibitory effect up to 2.8-fold at 24 h following pre-treatment at 50 µM. Overall, these results suggest that compound 11 exerts anti-inflammatory effects by suppressing NO production and modulating some pro-inflammatory cytokines, particularly MCP-1. Given that MCP-1 and other downregulated mediators (Fig. 3a and b) are regulated by NF-κB and MAPK pathways downstream of TLR-4³⁴, we hypothesize that compound 11 may target upstream signaling proteins within these cascades. This hypothesis is further supported by the early and significant decrease in MCP-1 levels detected as soon as 3 h after compound exposure (Fig. 3c).

Upon binding of LPS to the TLR-4/MD2 complex, intracellular signaling is initiated through the recruitment of adapter protein, including MyD88. The MyD88-dependent pathway is known to drive the production of pro-inflammatory cytokines by activating TAK1, which in turn stimulates the NF-κB and MAPK signal transduction pathways³⁴. Zhankuic acid A (ZAA), a tetracyclic triterpenoid structurally related to compound 11, has previously been reported to significantly block the phosphorylation of key transcription factors, including NF-κB p65, ERK1/2, c-Jun N-terminal kinase (JNK), and p38, suggesting its role in suppressing both NF-κB and MAPK pathways. Docking analysis further revealed that ZAA interacts with hydrophobic amino acid residues within MD2, the previously identified LPS-binding pocket. In vivo studies further demonstrated that ZAA suppressed TNF-α and IL-6 expression in C3H/HeN mice at a significantly greater extent than in TLR4 signaling-defective C3H/HeJ mice, confirming its selective blockade of the LPS/TLR-4/MD2 pathway³⁹. This evidence prompted us to further investigate, through computational methods, the potential inhibitory mechanisms of our active compound (11) within the TLR-4 signaling cascade.

A comparative assessment of molecular docking: compound 11 and inhibitors

The Protein Data Bank (PDB) is a valuable repository that houses the 3D structural information on a wide range of biological macromolecules, including proteins, nucleic acids, and ligands⁴⁰. Researchers widely use this resource to explore various aspects of molecular biology. In this study, compound 11 was found to exist as an inseparable diastereomeric mixture (24R- and 24S-isomers), as confirmed by 1D NMR data (Figures S26 and S27)²⁶. Therefore, both stereoisomers, designated as 11a (24R) and 11b (24S), were used for molecular docking studies against eight proteins involved in the TLR-4-mediated NF-κB signaling pathway, using the GOLD program. A redocking analysis was performed to validate the accuracy and consistency of the docking protocol. The re-docking of compounds 11a and 11b was compared to the reference ligands for each target protein (Figure S32). In most cases, the re-docked compounds (Figure S32, pink) showed a similar binding mode to that of the reference ligands (Figure S32, grey), indicating that the docking protocol was capable of reproducing the original ligand poses. The consistency in the docking results for both compounds 11a and 11b across multiple proteins suggests that the protocol effectively discriminates between potential binders and non-binders, supporting the reliability of the molecular docking process for further exploration of these compounds.

Molecular docking results suggested that MD2, TAK1, and NF-κB1 are potential targets for compounds 11a and 11b, with both isomers achieving higher GOLD fitness scores than their respective reference inhibitors, indicating favorable binding interactions^41–44. Accordingly, the crystal structures of MD2 (PDB ID: 2E56), TAK1 (PDB ID: 4GS6), and NF-κB1 (PDB ID: 1SVC) were retrieved from the PDB and used for molecular docking analyses. The results, summarized in Table 2 and S2, revealed that compound 11 exhibited strong binding affinities toward the active sites of MD2, TAK1, and NF-κB1. Specifically, GOLD fitness scores for compound 11a and 11b were 58.95 and 64.16 (MD2), 65.10 and 69.00 (TAK1), and 52.49 and 54.12 (NF-κB1), respectively (Table S2). These scores indicate that compound 11b, in particular, binds more strongly than the native inhibitors and may have a higher potential to interfere with TLR-4 related inflammatory signaling (Table 2).

Table 2.

Molecular Docking results of compound 11b with MD2, TAK1, and NF-κB1, based on GOLD fitness scores. Higher scores indicate stronger predicted binding affinity compared to native inhibitors.

Protein name	PDB ID	Ligand name	GOLD score
MD2	2E56	Myristic acid	53.57
		Compound 11b	64.16
TAK1	4GS6	5Z−7-oxozeaenol	58.00
		Compound 11b	69.00
NF-κB1	1SVC	Genistein	49.29
		Compound 11b	54.12

Furthermore, superimposition of the 3D binding poses revealed that compound 11b fits well within the active sites of the target proteins, aligning closely with known inhibitors such as myristic acid, 5Z-7-oxozeaenol, and genistein (Fig. 4)^41–44. These results support the hypothesis that compound 11 may serve as a promising modulator of the TLR-4-mediated NF-κB signaling pathway.

Fig. 4.

Comparative binding orientations of compound 11b and native inhibitors within the active sites of (a) MD2, (b) TAK1, and (c) NF-κB1 proteins, as determined via molecular docking utilizing GOLD software.

Molecular insights into protein-ligand binding

To further investigate the stability of protein–ligand interactions, we conducted molecular dynamics (MD) simulations of the complex formed between compound 11b and the target proteins MD2, TAK1, and NF-κB1, as well as their respective native inhibitors: myristic acid, 5Z-7-oxozeaenol, and genistein. Each complex was simulated in triplicate (RUN1–RUN3) for 100 ns. Structural stability was assessed by calculating the root mean square deviation (RMSD) values over the course of the simulations, focusing on the geometric coordinates of residues within the active sites (Figure S33), For the MD2-myristic acid complex, RMSD values increased to approximately 3.0 Å within the first 5 ns, followed by sustained fluctuations between 2.0−3.0 Å with fluctuations of roughly 0.5 Å, persisting until the end of the simulation (Figure S33a). Conversely, the RMSD values for MD2-compound 11b complex continued to hover around 2.0−2.5 Å throughout the entire 100 ns simulation (Figure S33b), suggesting greater binding stability compared to the native inhibitor. In both the TAK1-5Z−7-oxozeaenol and TAK1-compound 11b systems, RMSD values initially increased to 2.0−2.5 Å within the first 5 ns, followed by relatively stable fluctuations approximately 0.5 Å until the end of the simulation (Figures S33c and S33d). For NF-κB1, which has its active site located in a loop region, the RMSD values for both genistein and compound 11b displayed showed greater variability. However, in the final 20 ns of the simulation, the RMSD values remained stable until the end of the simulation (Figures S33e and S33f).

The RMSF (Root Mean Square Fluctuation) provides additional insight into the flexibility and stability of protein-ligand complexes during molecular simulations, illustrating how the ligands influence the dynamics of these proteins. For example, the RMSF for MD2 show that myristic acid (Figures S34a) induces higher fluctuations compared to compound 11b (Figures S34b), suggesting that myristic acid leads to more flexibility in the protein. Similarly, the RMSF profile for TAK1 in complex with compound 11b (Figures S34d) differs from that of the natural ligand 5Z-7-oxozeaenol (Figures S34c), indicating potential differences in binding behavior or structural stability. Variations in RMSF values across multiple simulation runs (RUN1, RUN2, RUN3) reflect the inherent variability in molecular simulations, where tighter clustering of the curves suggests more stable interactions. Higher RMSF values at specific residues indicate regions within the protein that may undergo significant conformational changes upon ligand binding. Notably, in the NF-κB1 protein (Figures S34e and S34f), compound 11b demonstrates distinct RMSF patterns compared to genistein, possibly reflecting a stronger or more stable interaction with the protein. Overall, the RMSF data highlight the regions of the proteins most influenced by ligand binding, offering a deeper understanding of how these interactions impact protein function and stability.

The superimposition average structures from three MD simulation replicates for compound 11b and their native inhibitors within MD2, TAK11, and NF-κB1 are shown in Fig. 5. The MD2 binding site of MD2, being relatively rigid, allowed both myristic acid and compound 11b to align well. Similarly to TAK1 complex, both 5Z−7-oxozeaenol and compound 11b are demonstrated good compatiblity with the binding site. However due to the inherent flexibility of the loop region in NF-κB1 binding site, ligand mobility was greater, imparts mobility to the ligand, leading to imperfect alignment of genistein and compound 11b in the superimposed structures⁴⁴.

Fig. 5.

Superimposed of last snapshot from three replicates of MD simulations for (a) MD2-myristic acid, (b) MD2-compound 11b, (c) TAK1-5Z−7-oxozeaenol, (d) TAK1-compound 11b, (e) NF-κB1-genistein, and (f) NF-κB1-compound 11b.

To further evaluate the binding affinity of compound 11b to the target proteins, a total of 1,000 snapshots from the last 20 ns of each MD simulations, were used to calculate the binding affinity values of the protein-ligand complexes using the Solvated Interaction Energy (SIE) method (Table S3). For the MD2 system, the average ΔG_bind was −7.40 ± 1.15 kcal/mol for the MD2–myristic acid complex and −8.36 ± 1.24 kcal/mol for the MD2-compound 11b complexes. In the TAK1 system, TAK1-5Z−7-oxozeaenol and TAK1-compound 11b exhibited average ΔG_bind values of −6.93 ± 0.80 and −7.22 ± 0.96 kcal/mol, respectively. For NF-κB1, the genistein and compound 11b complexes showed average ΔG_bind values of −5.33 ± 1.37 kcal/mol and −6.31 ± 0.73 kcal/mol, respectively. These results demonstrate that compound 11b exhibits stronger binding affinity to all three protein targets, MD2, TAK1, and NF-κB1, compared to their respective native inhibitors. The negative ΔG_bind values, particularly in the MD2 and TAK1 systems, further support the potential of compound 11b as a promising modulator of TLR-4-mediated signaling pathways.

To complement the MD simulation and binding energy results, we analyzed the protein-ligand interaction profiles over the final 20 ns using interaction maps (Fig. 6). These maps illustrate the frequency and types of interactions, including hydrophobic interactions, hydrogen bond donor (HBD) and acceptor (HBA) interactions, and ionizable group participation. In the MD2 systems, particularly in the MD2-compound 11b system, the majority of observed interactions were hydrophobic, with a 50% probability of persistence across the simulation (Fig. 6a and b). This is consistent with the hydrophobic nature of the MD2 binding pocket, where nonpolar residues such as valine (V), leucine (L), isoleucine (I), and phenylalanine (F) tend to cluster in the protein core during folding, creating a hydrophobic environment conducive to such interactions. In the TAK1-5Z−7-oxozeaenol complexes, a distinct HBA interaction with Ala107 was observed, along with hydrophobic contacts with Val50 and Met104 (Fig. 6c). The TAK1-compound 11b system demonstrated a distinct HBA interaction with Asp175, in addition to several hydrophobic interactions (Fig. 6d). For the NF-κB1-genistein complex, two HBD interactions (Glu341 and Ser81) and one HBA interaction (Glu344) were identified (Fig. 6e). In contrast, the NF-κB1-compound 11b system, formed a single hydrophobic interaction with Tyr82 and an HBA interaction with Thr342 (Fig. 6f). As noted earlier, the flexibility of the loop region where the NF-κB1 binding site resides may contribute to the variability and mobility of these interactions.

Fig. 6.

Pharmacophore models and protein-ligand interaction maps for (a) MD2-myristic acid, (b) MD2-compound 11b, (c) TAK1-5Z−7-oxozeaenol, (d) TAK1-compound 11b, (e) NF-κB1-genistein, and (f) NF-κB1-compound 11b. Hydrophobic interactions are highlighted by yellow circles, hydrogen bond donors by green vectors, and hydrogen bond acceptors by red vectors, and negative ionizable groups by dark red vectors.

The in silico results highlight compound 11 as a promising therapeutic candidate for modulating the TLR-4-mediated NF-κB signaling pathway. Molecular docking and MD simulations demonstrated superior binding affinity and stability compared to native inhibitors, with consistent and favorable interactions observed in MD2, TAK1, and NF-κB1 protein complexes. These findings are in line with previous studies emphasizing the therapeutic relevance of targeting these proteins in inflammatory signaling^43,44. Notably, the ΔGbind value for compound 11 in the MD2 system (− 8.36 kcal/mol) reinforces its potential efficacy. Protein–ligand interaction mapping further confirmed compound 11’s compatibility with hydrophobic and hydrogen bond-forming residues, contributing to the stability of the complexes and supporting its role as a viable anti-inflammatory lead compound.

In silico ADMET prediction

Failures in drug development are often attributed to undesirable physicochemical properties, suboptimal pharmacokinetics, or unexpected toxicity of the lead compound during clinical studies. To address these challenges early in the discovery process, in silico screening approaches have been developed to predict the ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) properties of drug candidates at an early stage and to help minimize resource-intensive experimental assays^45,46. In this study, the drug-likeness of compound 11 was evaluated using the SwissADME webserver, based on Lipinski’s (Pfizer filter) and Veber’s (GlaxoSmithKline filter) rules, both widely used in the pharmaceutical industry⁴⁵. The results (Table 3) suggested that compound 11 met both criteria, indicating that it is likely to possess physicochemical properties compatible with oral bioavailability, despite one violation (Log P > 5). Its high lipophilicity, associated with poor solubility (Log S < 6), may hinder passive absorption through the gastrointestinal tract, as shown in the BOILED-Egg predictive model (Figure S35), which positioned compound 11 outside the white region. Additionally, the model indicated that compound 11 is unable to penetrate the blood–brain barrier (BBB; yolk region) (Table 3, Figure S35), preventing it from crossing into the central nervous system and reducing the risk of CNS-related side effects. This restriction also limits the drug’s activity to peripheral tissues and organs where inflammation occurs. Furthermore, the non-P-glycoprotein substrate property of compound 11 suggests it is not readily expelled from cells, potentially improving its bioavailability. Analysis using the pkCSM web tool (Table 3) indicated that compound 11 does not inhibit major CYP enzymes, heme-containing proteins that play critical roles in drug metabolism and detoxification, thus reducing the risk of metabolic drug–drug interactions⁴⁶. The compound exhibited moderate total clearance, primarily occurring in the liver and kidneys, with values in the range of 0.1–1.0 mL/min/kg. No significant toxicity was predicted in various organs, and the compound showed a relatively high LD₅₀ value of 2.405 mol/kg (LD₅₀ < 0.5 mol/kg = highly toxic), indicating low acute toxicity.

Table 3.

In Silico ADMET prediction of compound 11.

Parameter	Compound 11
I. Physicochemical properties
Molecular weight (MW)	442.72 g/mol
H-bond acceptors (HBA)	2
H-bond donors (HBD)	2
Rotatable bonds (RB)	5
Topological polar surface area (TPSA)	40.46 Å2
Consensus Log P	6.61
Log S (ESOL)c	−7.31
II. Pharmacokinetic properties
GI absorption	Low
BBB permeant	No
P-gp substrate	No
CYP inhibitorsd	None of them
Total clearance (logCL)e	0.485 ml/min/kg
III.Toxicity
Oral rat acute toxicity (LD50)	2.405 mol/kg
Mutagenicity	No
Hepatotoxicity	No
Skin sensitization	No

^clog S scale: insoluble < − 10; poor < − 6; moderate < − 4; soluble < − 2; very < 0 < high; ^dCytochrome 450 (CYP) predicted: CYP1A2, CYP2C19, CTP2C9, CYP2D6, and CYP3A4; ^elogCL: ≤ 0.1 mL/min/kg, moderate as > 0.1 to ≤ 1 mL/min/kg, and high as > 1 mL/min/kg.

Overall, these ADMET predictions suggest that compound 11 possesses promising drug-like properties, including good oral bioavailability, low toxicity, and limited BBB permeability^45,46. However, consistent with other triterpenoid-derived drug leads, compound 11 faces challenges related to poor water solubility, which may limit membrane permeability and reduce therapeutic efficacy. This common obstacle in drug development could be addressed through molecular modification, such as introducing hydrophilic functional groups (e.g., amino acids or sugar moieties), or by employing advanced drug delivery systems, including cyclodextrin inclusion complexes or nanoparticle encapsulation⁴⁷. These strategies hold promise for enhancing the solubility, bioavailability, and overall therapeutic performance of compound 11.

Conclusions

In this study, seven tocotrienol and five tetracyclic triterpenoid derivatives were isolated from the stem bark of Garcinia latissima, including two previously undescribed structures: δ-(E)-deoxy-amplexichromanoyl acetate (1) and (20S,24S)−20,24-epoxylanostane-3β,25-diol (2). Notably, this work constitutes the first report of tocotrienol derivatives from this species, expanding its known phytochemical profile. Among the isolated compounds, (20R)-eupha-8,25-diene-3β,24ξ-diol (11) demonstrated the most promising anti-inflammatory activity, by significantly inhibiting nitric oxide production and downregulating pro-inflammatory cytokines, particularly MCP-1. Computational analyses suggested that compound 11 may exert its effects by targeting key upstream regulators of the TLR-4 signaling pathway—namely MD2, TAK1, and NF-κB1—as indicated by strong binding affinities in molecular docking and molecular dynamics simulations. Additionally, in silico ADMET prediction indicated favorable drug-likeness, good oral bioavailability, and low toxicity, further supporting the potential of compound 11 as a candidate for further pharmacological and toxicological evaluation. Although these findings provide preliminary evidence of anti-inflammatory potential based on in vitro and computational data, they offer an insight for future experimental investigations aimed at validating the molecular mechanism of action and improving the pharmacokinetic properties of compound 11. From a chemical standpoint, future studies could focus on purifying plant extracts enriched in compounds structurally related to 11 for further chemical modification or stereoselective synthesis of the targeted bioactive molecule. Comprehensive biological evaluations, including target-based biochemical inhibition assays, solubility and permeability studies, and in vivo efficacy and toxicity assessments, will be essential to fully establish its therapeutic potential.

Materials and methods

Materials and general experimental procedure for phytochemical analysis

Optical rotations were recorded on a JASCO P-1010 polarimeter (Easton, MD, USA). The UV spectra were analyzed using a Shimadzu UV-2550 spectrometer (Kyoto, Japan), while IR data were obtained on a JASCO FT/IR-4200 spectrometer (Tokyo, Japan) by the attenuated total reflection (ATR) technique. The experimental ECD data were acquired on a JASCO J-815 circular dichroism spectrometer (Easton, MD, USA). The ¹H (500 MHz) and ¹³C (125 MHz) NMR spectra were measured on JEOL JNM-ECZ500R/S1 NMR spectrometer (Tokyo, Japan). The LC-QTOF-MS/MS analysis was performed using an Agilent HPLC 1260 series coupled with a QTOF 6540 UHD accurate mass (Waldbronn, Germany). Silica gel (70−230 mesh, Merck, Darmstadt, Germany), Sephadex LH-20 (25–100 mm, GE Healthcare Bio-Sciences AB, Uppsala, Sweden), and Chromatorex ODS/RP-C₁₈ (100–200 mesh; Fuji Silysia Chemical Ltd., Tokyo, Japan) were used for column chromatography. Radial chromatography was carried out using Chromatotron model 7924 T (Harrison Research, Palo Alto, CA, USA) with silica gel 60 GF254 containing gypsum (≤55 μm, Merck). Silica gel 60 F254 (0.2 mm, Merck) was used for TLC analysis.

Plant material

The stem bark of G. latissima Miq. (Clusiaceae) was collected from Somaetek forest, North Halmahera, Indonesia (1°25′22′′ LU 127°46′59′′ E) in October 2018. The plant material was identified by a botanist, Mr. Ikrar Supriyatna, and the voucher specimen (No. IV.C.338) was deposited at Bogor Botanical Garden, East Java, Indonesia. The plant is classified as not threatened and is cultivated in the Bogor and Cibodas Botanical Gardens. Necessary permissions from the local government were obtained for sample collection, and all procedures complied with the guidelines of the IUCN Policy Statement on Research Involving Species at Risk of Extinction and the Convention on International Trade in Endangered Species of Wild Fauna and Flora (CITES).

Extraction and isolation

The air-dried stem bark powder of G. latissima (3.0 kg) was macerated with EtOAc (3 × 15 L for 6 days). The solvent was evaporated under reduced pressure to obtain residue. The crude extract (69.0 g) was suspended in MeOH−H₂O (50:50) and partitioned with CH₂Cl₂ and EtOAc. The CH₂Cl₂-soluble fraction (50.8 g) was separated by silica gel column with gradient of hexanes: EtOAc (90:10−0:100) to afford fractions A‒I.

Fraction C (3.5 g) was chromatographed on a Sephadex LH-20 column with CH₂Cl₂:MeOH (50:50) to obtain subfractions C1–C4. Subfraction C1 (1.0 g) was loaded on silica gel column eluted with gradient of CH₂Cl₂:MeOH (100:0−95:5) to obtain subfractions C1.1–C1.3. Compound 1 (16.5 mg) was purified from fraction C1.2 (108.0 mg) through a separation using Chromatotron with eluent hexanes: acetone (87:13), while the same technique was also applied to subfraction C1.3 (350.0 mg) with gradient eluent of hexanes: EtOAc (85:15−70:30) to give compounds 2 (3.0 mg), 9 (4.0 mg), 10 (6.2 mg), and 11 (4.8 mg). Fraction D (4.0 g) was subjected on silica gel column with hexanes: acetone (80:20) to yield subfractions D1‒D4. Purification of subfraction D1 (47.5 mg) using Chromatotron with eluent hexanes: chloroform (0:100) afforded compound 6 (5.2 mg). Compound 4 (25.8 mg) was isolated from subfraction D2 (230.0 mg) using ODS column with eluent system of MeOH: H₂O (80:20), while compounds 3 (1.3 g), 7 (24.0 mg), and 8 (18.5 mg) were purified from subfraction D3 (2.2 g) using the same method with gradient of MeOH: H₂O (75:25−100:0). Fraction F (1.2 g) was chromatographed on Sephadex LH-20 column eluted with CH₂Cl₂:MeOH (50:50) to obtain subfractions F1−F4, in which compound 12 (3.2 mg) was precipitated from subfraction F1. Separation of fraction H (850.0 mg) using Sephadex LH-20 column with eluent CH₂Cl₂:MeOH (50:50) provided five subfractions (H1−H5), in which compound 5 (2.0 mg) was isolated from subfraction H2 (98.7 mg) after chromatographed using the same technique and solvent system.

δ-(E)-deoxy-amplexichromanoyl acetate (1): Pale brown gum; ‒20.2 (c 0.05, MeOH); UV (MeOH) λ_max: 296 and 216 nm; IR ν_max (ATR): 3405, 2923, 1714, 1466, 1376, and 1219 cm^− 1; ECD λ_max (c 0.05, MeOH) nm (log ɛ): 208 (+ 31.2); ¹H (500 MHz, CDCl₃) and ¹³C (125 MHz, CDCl₃) NMR spectroscopic data, see Table 1. HRESIMS at m/z 472.3421 [M + NH₄]⁺ (calcd for C₂₉H₄₆NO₄⁺: 472.3421) and 395.2945 [M − CH₃COOH + H]⁺ (calcd for C₂₇H₃₉O₂⁺: 395.2945).

(20S,24R)−20,24-epoxylanostane-3,25-diol (2): White powder; +13.5 (c 0.05, MeOH); IR ν_max (ATR): 3378, 1450, 1376, and 1042 cm⁻¹; ¹H (500 MHz, CDCl₃) and ¹³C (125 MHz, CDCl₃) NMR spectroscopic data, see Table 1. HRESIMS at m/z 443.3883 [M − H₂O + H]⁺ (calcd for C₃₀H₅₁O₂⁺: 443.3884).

Materials and cell culture for bioassay

Dimethyl sulfoxide (DMSO), 3-(4,5-dimethylthiazol-2-yl)−2,5-diphenyl tetrazolium bromide (MTT), lipopolysaccharide (LPS) from Escherichia coli O111:B4 (cat. no. L2630) and O55:B5 (cat. no. L2880), Griess reagent (cat. no. G4410), interferon-γ, Bay 11–7082, and etoposide were purchased from Sigma-Aldrich (Burlington, MA, USA). Dulbecco’s Modified Eagle Medium (DMEM), Penicillin-streptomycin (Pen-Strep), and Phosphate-Buffered Saline (PBS) were obtained from Gibco (Grand Island, NY, USA). Fetal bovine serum (FBS) was obtained from Hyclone (Logan, UT, USA). Mouse inflammation antibody array (ab133999) and MCP-1 ELISA kit (ab100721) were purchased from Abcam (Cambridge, UK). ELISA MAX™ Deluxe Set Mouse MCP-1 (cat. no. 432704) was obtained from Biolegend (San Diego, CA, USA). The stock solutions of the tested compounds (20 mM) were dissolved in DMSO, which the final concentration was controlled to be less than 1% (v/v). The stock solutions of the samples in DMSO were stored at ‒20 °C. RAW 264.7 mouse macrophage cell line (ATCC TIB-71) were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) supplemented with 10% fetal bovine serum (FBS) (v/v), penicillin (100 units/mL) and streptomycin (100 ug/mL) and incubated in an incubator at 37 °C and 5% CO₂.

In vitro cytotoxic evaluation using MTT assay

The isolated compounds were evaluated for cytotoxicity against the RAW 264.7 mouse macrophage cell line, which was performed using an MTT colorimetric method by incubating cells as described previously⁴⁸. Cells were seeded in 96-well plates at 1 × 10⁴ cells per well and were incubated at 37 °C in a humidification incubator with 5% CO₂ for 24 h prior to treatment. After incubation, the cells were treated with two different concentrations of the compounds (10 and 50 µM) and 5 µM Bay 11–7082, while 0.1% DMSO was used as the vehicle. They were incubated under the same conditions for 24 h. After which, MTT solution (5 mg/mL in PBS) was added to the wells and again incubated at 37 °C for an additional 3 h. The medium was then removed and mixed with DMSO (200 µL per well). The absorbance at 570 nm was measured with a PerkinElmer, EnSight Multimode Microplate Reader and the cytotoxicity was calculated. Etoposide was used as the positive control at the concentration of 50 µM to ensure that all of the cells were completely killed, confirming the validity of the assay.

In vitro nitric oxide (NO) inhibitory assay

The NO inhibitory activity of the isolated compounds was evaluated using a previously described protocol with slight modifications^48,49. The RAW 264.7 mouse macrophage (6 × 10⁴ cells/well) cell line were seeded into a 96-well plate in DMEM containing 10% FBS (v/v), penicillin (100 units/mL) and streptomycin (100 ug/mL). After incubation for 24 h at 37 °C in a 5% CO₂ humidified atmosphere, the cells were stimulated by the addition of 10 U/mL interferon-γ and 100 ng/mL LPS (from E. coli O111:B4) and treated with two different concentrations (10 and and 50 µM) of the tested compounds. The level of NO production in cell culture supernatants was determined by incubation of 100 µL of culture medium with 100 µL of Griess’ reagent at room temperature for 10 min and the absorbance was recorded at 540 nm using PerkinElmer EnSight multimode Plate Reader. Bay 11–7082 was used as the positive control at 5 µM.

Anti-inflammatory assay using mouse inflammation antibody array-membrane

The inhibitory activity of 11 towards 40 inflammatory targets in LPS-stimulated RAW 264.7 cell line was analyzed using a mouse inflammation antibody array following the manufacturer’s instructions. Cells were stimulated with LPS to induce inflammatory responses. Two membranes were used, including an inflammatory membrane containing supernatants of LPS-stimulated cells and a tested membrane containing LPS-stimulated cell culture medium treated with compound 11 at 50 µM. RAW 264.7 (1.25 × 10⁵ cells/well) were seeded into a 24-well tissue culture plate in DMEM containing 10% FBS (v/v), penicillin (100 units/mL) and streptomycin (100 ug/mL). After incubation for 24 h at 37 °C in a 5% CO₂ humidified atmosphere, the media were replaced with 0.5 mL of fresh culture medium containing LPS (100 ng/mL; from E. coli O55:B5) or LPS (100 ng/mL) + compound (50 µM). Finally, supernatants were taken after 24 h of exposure to compound and stored at − 20 °C until use. Membrane chemiluminescence was detected and quantified using UVITEC FireReader V1 machine and UVITec1D software. The assays were conducted in duplicates per membrane. Results are expressed as mean ± SD and values are relative to the positive control (biotin-conjugated IgG printed directly onto the array membrane) of each membrane, which is given an arbitrary value of 1.

Enzyme-linked immunosorbent assay (ELISA)

Mouse MCP-1 in 3 and 24 h–conditioned medium from LPS-stimulated RAW 264.7 cells was quantified using an MCP-1 ELISA kit following the manufacturer’s instructions. Briefly, cells were plated into 24-well plates (1.25 × 10⁵ cells/well/500 µL) in in DMEM containing 10% FBS (v/v), penicillin (100 units/mL) and streptomycin (100 ug/mL). After incubation for 24 h at 37 °C in a 5% CO₂ humidified atmosphere, cells were pre-treated with compound 11 at 10 and 50 µM for an hour followed by adding 100 ng/mL of LPS. Supernatants were taken after 3 and 24 h of compound and measured using the ELISA kit.

Statistical analysis

All experiments were conducted in triplicate (n = 3) unless specified otherwise. All data were presented as mean ± SEM (Standard Error of Mean). The significance in differences between group was determined by Duncan’s multiple range test (p < 0.05) using Graphad Prism software version 7.0 (GraphPad software Inc., San Diego, USA).

Exploring protein structure through molecular docking with GOLD software

The protein structures of the MD2, TAK1, and NF-κB1 proteins were obtained from the Protein Data Bank (PDB) database using the PDB IDs 2E56⁴¹, 4GS6⁴², and 1SVC⁴³, respectively. ChimeraX⁵⁰ was then used to visualize, separate ligands, and remove small molecules. The compounds 11a and 11b were created using the GaussView program⁵¹ and the structures were minimized by 100 steps of steepest descent, 10 conjugate gradient, step size 0.02 Å by Chimera program⁵². The ligands and receptors were both saved as mol2. The ligand was docked with the receptor 100 times using Genetic Optimisation for Ligand Docking (GOLD) Software⁵³. Noted that the grid box with 12 × 12 × 12 dimensions centered on the following XYZ coordinates for each model. For the model denoted as 2E56, the coordinates were − 1.99, 19.92, and 17.81. The model represented by 4GS6 was centered at coordinates 15.95, 16.98, and 53.97, while the model designated as 1SVC was centered at coordinates 34.30, 9.39, and 40.52. Subsequently, among the ensemble of docked conformations, those attaining the highest GOLD fitness score were carefully selected for further analysis and consideration.

Exploring complex structures through molecular dynamics simulation

Prior to MD simulations, the crystal structures were prepared by removing all crystallographic water molecules and counter ions. The protonation states of the protein-ligand complexes were calculated utilizing the PDB2PQR server⁵³ at a physiological pH of 7.4. Subsequently, all-atom molecular dynamics (MD) simulations were conducted on MD2, TAK1, and NF-κB1, along with their respective inhibitors and compound 11b. These simulations were carried out within a periodic boundary by AMBER20⁵³. For the force field parameters of the target proteins, the ff19SB force field⁵⁴ was applied, including both bonded and nonbonded interactions. Parameters for the inhibitors and compound 11b were generated using the leap module in conjunction with the general AMBER force field 2 (GAFF2)⁵⁵. Furthermore, restrained electrostatic potential (RESP) charges⁵⁶ for ligands were computed following established standard procedures. To complete the structural of each system, any missing hydrogen atoms were added using the LeaP module. Each protein–ligand complex was placed in a triclinic box solvated with TIP3P water molecules, with a 12 Å buffer from the solute to the box boundary. Additionally, Cl- ions were introduced into the systems to neutralize the charges. All newly added hydrogen atoms were subjected to energy minimization using a combination of steepest descent and conjugate gradient algorithms to resolve unfavorable contacts and optimize local geometry. Subsequently, the entire protein-ligand complex underwent full minimization using the same iterative approach to reach a stable energy state. Following minimization, a two-phase equilibration protocol was employed: a 1 ns simulation under the NVT ensemble using the Berendsen thermostat to maintain the system at 300 K, followed by a 1 ns NPT simulation using the Parrinello-Rahman barostat to stabilize pressure at 1 atm. The SHAKE algorithm was applied to constrain all bonds involving hydrogen atoms, allowing for a 2 fs integration time step in the production MD simulations. Each MD simulation was run in triplicate, each time with a different random seed value. These replicates were started from identical minimized structures, ensuring statistical reliability and robustness in our computational investigations.

After conducting the MD simulation, we used the cpptraj module of AmberTools23⁵⁷ to calculate the root-mean-square displacement (RMSD). ChimeraX⁵⁰ was used to visualize protein-ligand interactions. Furthermore, the solvated interaction energy (SIE) method⁵⁸ was used to calculate the binding affinities of protein-ligand complexes. The SIE binding free energy of the complex was calculated over 1000 snapshots from the last 20 ns for each individual simulation (a total of 1000 snapshots). The LigandScout program⁵⁹ was used to generate a map of protein-ligand interactions.

In silico ADMET parameter analysis

The physicochemical and pharmacokinetic properties of compound 11, including absorption, distribution, metabolism, excretion, and toxicity, (ADMET) were performed using SwissADME and pkCSM webservers by uploading the simplified molecular-input line-entry system (SMILES) string of the molecule^45,46. The Lipinski’s rule of five is in accordance to the following criteria: molecular weight ≤ 500, hydrogen bond donors (HBD) ≤ 5, hydrogen bond acceptors (HBA) ≤ 10, and octanol-water partition coefficient (Log P) ≤ 5, while Verber rule includes two extended parameters, the number of rotational bonds (RB) ≤ 10 and the topological polar surface area (TPSA) ≤ 140.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Edwin R. Sukandar: Investigation, Formal analysis, Writing - Original Draft. Nitchakan Darai: Investigation, Formal analysis, Writing - Original Draft. Jaruwan Chatwichien: Investigation. Sutthida Wongsuwan: Investigation. Sutin Kaennakam: Methodology, Validation. Thitiporn Pattarakankul: Investigation, Formal analysis. Retno Purbowati: Investigation. Taslim Ersam: Conceptualization, Resources. Kowit Hengphasatporn: Validation, Writing - Review & Editing. Yasuteru Shigeta: Methodology, Supervision. Patipark Kueanjinda: Validation, Writing - Review & Editing. Tanapat Palaga: Methodology, Validation, Writing - Review & Editing. Santi Tip-pyang: Conceptualization, Resources. Warinthorn Chavasiri: Supervision, Funding acquisition. Thanyada Rungrotmongkol: Supervision, Writing - Review & Editing. Chanat Aonbangkhen: Project administration, Funding acquisition, Writing - Review & Editing. All authors read and approved the final manuscript.

Data availability

The data underlying this study are available in Supporting Information.

Declarations

Competing interests

The authors declare that a patent application related to aspects of this work is currently under preparation for submission in Thailand. The authors may have potential future financial interests arising from the commercialization of this technology. No other conflicts of interest are declared.