Introduction

The introduction of antiretroviral therapy has been transformative in the prognosis of HIV/AIDS, transitioning this pandemic into a manageable chronic illness1. Despite the successes of combined antiretroviral therapy (cART) in reducing viral load, the currently available drugs cannot eliminate latently integrated proviral particles. Thus, the virus reactivates upon treatment interruption1,2. In addition, the currently available cART drugs are slowly becoming ineffective due to emerging multidrug-resistant viral strains and severe side effects causing cytotoxicity to host cells and non-adherence to treatment3. Given these factors, there remains a need to search for novel alternative drugs with less toxicity that can escape these treatment barriers.

Microbial natural products from underexplored environments are gaining impetus as new sources of novel chemistries that cannot be produced through synthetic chemistry approaches. Microbial natural products are attractive in drug development programs due to their high rigidity, making them a suitable candidate for protein-protein interactions. This also fulfils the criteria for a new mode of action that is not susceptible to emerging drug resistance challenges4,5. Endophytic microorganisms associated with medicinal plants qualify as an unexplored source and have been shown to produce a similar arsenal of bioactive natural products to their plant hosts with higher therapeutic potential6,7. Many small metabolites produced by endophytic fungi have been reported to exhibit antiviral properties7. For example, Bringmann et al.8 identified novel sorbicillinoid alkaloids, sorbicillatones A and B, from P. rubens. Sorbicillatone A demonstrated significant antiviral activity against HIV viral proteins. These studies suggest that P. rubens could be the source of novel anti-HIV compounds.

The challenge limiting the exploration of microbial natural products is that many remain undetected in culture media because they are encoded by biosynthetic pathways that remain silent or are expressed at lower levels9. Even when expressed, the complexity of crude extracts makes the chemical approaches to isolating and purifying these bioactive compounds complex and costly, often leading to the re-isolation of already-known compounds. One of the successful approaches in triggering the expression of silent biosynthetic pathways in microbial cultures is the OSMAC (one strain many compounds) approach, which involves systemic cultivation of microbial culture under varying conditions to simulate varying environmental triggers10. This, however, leads to complex crude extracts with a vast chemical space, making it difficult to decipher the compound(s) responsible for the observed activity11. This can be addressed by following an untargeted metabolomics approach coupled with molecular networking, which clusters molecules with similar fragment ions into families, making further dereplication less complicated.

This study explores the power of combining the OSMAC approach with untargeted metabolomics in widening the chemical space in an endophytic fungus, P. rubens P03MB2. The resulting extracts were evaluated for antiviral activity. Furthermore, the compounds identified were analyzed in silico for binding to the target HIV protease, and molecular dynamics simulations were used to analyze the stability of the formed ligand-protein complexes.

Methods and materials

Production of secondary metabolites from Penicillium rubens

The fungal species explored in this study, P. rubens P03MB2 was isolated and stored as glycerol cultures from our previous study9. The fungal culture was revived on potato dextrose agar (Neogen corporation, United Kingdom) supplemented with 50 µg/mL chloramphenicol (Sigma Aldrich, SA), and incubated at 28⁰C for five days. For large-scale fermentation, 15 × 500 mL Erlenmeyer flasks were prepared and divided into five Erlenmeyer flasks per media for three different media: malt extract broth (90 mL) and 28 g each of oats media and rice media. For solid media fermentation, 100 mL of WOFT (K2HPO4: 0.5 g L−1; yeast extract: 1 g L−1; sodium tartrate: 0.5 g L−1) was added to 28 g of rice and oats before the media was autoclaved12. Mycelial plugs (6 mm2) were transferred under sterile conditions to all prepared media, and the flasks were incubated at 28 °C in the dark for 14 days.

Extraction of secondary metabolites from Penicillium rubens

After the incubation period, an equal amount of ethyl acetate (ACE chemicals, Johannesburg, South Africa) was added to each Erlenmeyer flask and placed on an orbital shaker at 24 °C with shaking at 150 rpm overnight before separating the organic layer from the aqueous layer. The recovered organic layer (ethyl acetate) was dried using a rotatory evaporator at 40 °C, and the solid extract was collected and resuspended in a 1:1 mixture of distilled water and methanol in 25 mL McCartney bottles. The crude extracts were partially purified using solid phase extraction (SPE) reverse phase chromatography to remove chromatography-impairing lipids and simplify the extracts into smaller fractions. Fractionation of the crude extracts was performed as described by Stoszko et al.13. Concisely, the dried material was reconstituted in 50% methanol (MeOH) to a concentration of 100 mg/mL. Subsequently, 1 mL of this suspension was spiked with 20 µL of phosphoric acid (Sigma-Aldrich, South Africa) and loaded onto MAX (mixed mode, strong anion-exchange) and MCX (mixed mode, strong cation-exchange) cartridges acquired from Waters Corporation (Prague, Czech Republic). The sorbent surfaces of the cartridges were desalted and eluted with an increasing stepwise gradient using methanol (5%, 45%, 95%) to extract the adsorbed compounds. The eluted fractions of organic compounds were evaporated to dryness at 40 °C through incubation. The dried fractions were reconstituted to 1 mg/mL in methanol (HPLC grade, Sigma Aldrich, South Africa) and stored at 4 °C until further use.

Evaluation of cytotoxicity of the crude extracts from Penicillium rubens using MTT viability assay

The crude extracts’ cytotoxicity effects and cell viability were assessed using the CYQUANT MTT proliferation kit, following the manufacturer’s instructions (Thermo Fisher, Waltham, USA). Briefly, a 96-well culture plate was seeded with 15,000 TZM-bl cells per well in 200 µL of DMEM and incubated at 37 °C with 5% CO₂. After cell attachment, 10 µL of Penicillium rubens crude extracts and the positive control drug Zidovudine (AZT) (NHI AIDS research and reagent program) at 300 µg/mL were serially diluted 10-fold from 300 µg/mL in DMEM containing 10% heat-inactivated fetal bovine serum, 50 µg/mL gentamycin, and 25 mM HEPES (LTC Biosciences, Gainesville, FL, USA) buffer in respective wells. The plate was then incubated at 37 °C with 5% CO₂ and 95% humidity for 48 h. Following incubation, the supernatant (treatment medium) was replaced, and each well received 10 µL of 12 mM MTT solution before incubation for an additional four hours at 37⁰C with 5% CO2. Afterward, the MTT solution was discarded, 50 µL of 0.2% DMSO (dimethyl sulfoxide) was added to each well to dissolve the formazan crystals, and the plate was further incubated for 10 min at 37⁰C. The absorbance of each well was measured at 540 nm using the Victor Nivo microplate reader (PerkinElmer, MA, Waltham, USA).

The percentage of cell viability was calculated using the formula below2:

$$\% Cell\;viability = Sample\;absorbance - media\;control\;Mean\;media\;control\;absorbance\; \times 100$$

Antiviral activity of the crude extracts from Penicillium rubens.

The crude extracts and fractions were prepared for activity testing by making a 3-fold dilution for each extract, starting at 300 µg/mL in DMEM in a 96-well flat-bottom plate. After that, 50 µL of diluted HIV-1 NL4.3 (MOI = 0.05) virus was dispensed to all the wells except cell control wells and incubated for one hour at 37 °C, 5% CO₂. TZM-bl cells were prepared at a density of 1 × 10⁵ cells/mL in a growth media containing 37.5 mg/mL DEAE dextran and were seeded in a 96-well plate and incubated at 37 °C 5% CO₂ for 72 h. Zidovudine (AZT) at 300 µg/mL was used as a positive control, while the uninfected and untreated cells were used as negative controls. After incubation, the culture medium (100 µL) was replaced with new DMEM media, and a 100 µL of Bright Glo luciferase reagent (Promega, Madison, WI, USA) was added to each well under low light conditions. Following this, the cells were allowed to lyse before they were transferred to a 96-well bottom flat black plate (Costar, Corning Incorporated, USA), followed by incubation at 37 °C. The relative luminescence unit was measured immediately after incubation using a Victor 2 luminometer plate reader (Perkin-Elmer Life Sciences, Shelton, CT) at 540 nm. The percentage of viral inhibition was expressed following this equation: % HIV inhibition = [(Average Sample-Average Control)/(1-(Average Viral Control-Average Control)] × 100 (2) The half-maximal inhibitory concentration (IC₅₀) causing morphological changes in the HIV inhibition dose-response curve by 50% was calculated using GraphPad Prism Software (v.5).

High-resolution liquid chromatography-mass spectrometry analysis

The selected bioactive crude extracts and fractions were analyzed on a Nexera LC-40D XR liquid chromatography hyphenated to a quadrupole time-of-flight (Q-ToF) tandem mass spectrometer (LCMS-9030 qTOF, Shimadzu Corporation, Kyoto, Japan) following a method described by Ramabulana et al., (2021) with some modifications. The separation was performed using a 2.7 μm particle size Shim-pack Velox C18 column (150 mm, 2.1 mm, Shimadzu Corporation, Kyoto, Japan) at 40⁰C. All samples were injected at a volume of 20 µL with solvent A (0.1% formic acid, Merck, Darmstadt, Germany) and solvent B (acetonitrile: 0.1% formic acid, LC-MS grade, Merck, Darmstadt, Germany). The flow rate at 0.4 mL/min throughout the 25 min gradient. The chromatographic gradient was set as follows: linear gradient of 5–65% B over 20 min, the gradient was ramped to 90% B in 3 min (20–23 min) and held at that composition for 3 min (23–26 min), the composition was returned to 5% B at 26.1 min until 25 min to recondition the column.

The Q-Tof mass spectrometer was operated in data-dependent acquisition (DDA) mode using positive electrospray ionization with the following parameters: interface voltage of 4.0 kV, interface temperature of 300⁰C, nebulization, and dry gas flow of 3 L/minute, heat block temperature of 400⁰C, DL temperature of 280⁰C, detector voltage of 1.8 kV and a flight temperature of 42⁰C. The spectra, MS1 and MS2, were generated for all ions with an m/z range between 100 and 1200, surpassing an intensity threshold of 3000. Fragmentation of ions was performed using argon as a collision gas at a collision energy of 30 ± 17 eV.

MZmine data pre-processing and GNPS-feature-based molecular networking

The data obtained from the mass spectrometry was converted to an open format, mzML file using MS convert software (http://proteowizard.sourceforge.net/downloads.shtml; accessed 02 October 2024). The mass spectrometry data files were first processed using MZmine 314. Thereafter, the data was uploaded on the GNPS2-MN (Global Natural Products Social-Molecular Network, Version 28.2) platform (https://gnps2.org/) accessed on 06 May 2025, and the data was processed following the Feature-Based Molecular Networking (FBMN) workflow15,16. The data was filtered by removing all MS/MS fragment ions within +/- 17 Da of the precursor m/z and choosing only the top 6 fragment ions in the +/- 50 Da window throughout the spectrum. The precursor ion mass tolerance was set to 0.02 Da, and an MS/MS fragment ion tolerance of 0.02 Da. A network was created where edges were filtered to have a cosine score above 0.7 and more than 6 matched peaks. Further, edges between two nodes were kept in the network only if each node appeared in each other’s top 10 most similar nodes. Finally, the maximum size of a molecular family was set to 100, and the lowest-scoring edges were removed from molecular families until the molecular family size was below this threshold. The spectra in the network were then searched against GNPS spectral libraries, which were filtered using the same conditions as described for the spectra above15,17. All retained matches between the network and library spectra were required to score above 0.7 and at least 6 matched peaks. Further annotation of MS/MS spectra was done using the dereplicator18 and the Sirius platform, Version 5.8.5, using the built-in CSI fingerID and Canopus features19,20,21. The web interface for feature-based molecular networking (https://fbmn-statsguide.gnps2.org/; accessed 06 May 2025) was used to clean up the FBMN data and statistical analysis and integration into the Cytoscape visualization tool22 The molecular networks were visualized using Cytoscape software, Version 3.10.223.

Virtual screening of secondary metabolites from P. rubens

The X-ray crystal structure of HIV-1 Protease NL4-3 (PDB code: 6PJG) was obtained from the RCSB Protein Data Bank for molecular docking studies. Protein preparation was performed using AutoDock Tools v.1.5.624, where water molecules, ions, and co-crystallized ligands were removed. Missing side chain atoms and polar hydrogen atoms were added, and Kollman charges were assigned to the protein structure to optimize it for docking. The ligand library was first prepared in 3D format with explicit hydrogens and appropriate protonation states at pH 7.4 using OpenBabel v.2.4.125. These structures were then processed with Meeko v.0.5.126 to assign rotatable bonds and convert them to PDBQT format compatible with AutoDock Vina27.

The docking protocol was validated by re-docking the co-crystallized ligand (ORV), with success criteria defined as a root mean square deviation (RMSD) of less than 2.0 Å from the crystallographic pose. The binding site was defined based on the position of the co-crystallized ligand. A two-step docking approach was employed using AutoDock Vina: initial screening was conducted with default parameters (exhaustiveness = 8) to identify promising candidates, followed by a more rigorous docking phase for the top 10% of compounds based on binding affinity scores, using refined parameters (exhaustiveness = 128, number of binding modes = 20, and energy range = 3 kcal/mol). To ensure reproducibility and adequate sampling of conformational space, each compound was docked in triplicate using different random seeds.

Docking results were ranked according to binding affinity scores, and the top three scoring compounds were selected for subsequent analysis. The ligand binding poses, and protein-ligand interactions were visualized using Discovery Studio Visualizer version 20.1.0.1929528 and PyMol v.3.029.

Following molecular docking, Density Functional Theory (DFT) calculations were performed using Gaussian 16 software to refine the top-scoring compounds’ geometries further. DFT calculations utilized M06-2X functional with the 6-311 + + G(d, p) basis set for all selected ligands. The geometries of the compounds were optimized to the nearest local minimum, and frequency calculations were performed to confirm that the optimized structures corresponded to true minima. The highest occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) energies, as well as the HOMO-LUMO energy gap (∆E,), were calculated to investigate the electronic properties of the ligands. Additionally, global reactivity descriptors, including chemical potential (µ), chemical hardness (η), softness (S), electrophilic index (ω), and electronegativity (χ) were derived from HOMO-LUMO energies using the following equations:

$$\mu = \:\frac{{E_{{HOMO}} + E_{{LUMO}} }}{2}$$
(1)
$$\eta = \:\frac{{E_{{LUMO}} - E_{{HOMO}} }}{2}$$
(2)
$$S = \frac{1}{\eta }$$
(3)
$$\chi = - \mu$$
(4)
$$\omega = \frac{{\mu ^{2} }}{{2\eta }}$$
(5)

Molecular dynamics simulations

Molecular dynamics (MD) simulations were conducted using AMBER 20, employing the FF14SB force field parameters for protein description30,31. The LEAP module was utilized to add hydrogen atoms to the protein and incorporate counter-ions32. The system was enclosed in a TIP3P water box, maintaining a distance of 10 Šbetween the system surface and the box edge. Periodic boundary conditions were applied, with long-range electrostatics managed using a 12 Švan der Waals cutoff33. Initial minimization involved a restrained potential of 500 kcal/mol/Ų over 1000 steepest descent steps followed by 1000 conjugate gradient steps on the solute34. This was succeeded by 1000 steps of unrestrained conjugate gradient minimization for the entire system. The system was gradually heated from 0 to 300 K over 1 ps, employing a Langevin thermostat with collision frequency and harmonic restraints set at 5 kcal/mol/Ų. An unrestrained equilibration of the system was performed using an NPT ensemble at constant pressure (1 bar) and temperature (300 K)35. The production run for MD simulation lasted 200 ns, utilizing an isothermal-isobaric (NPT) ensemble with a Berendsen barostat36. Coordinates were saved at intervals throughout the simulation, and trajectories were analyzed post-simulation.

Post molecular dynamics simulation analysis

Post-MD trajectory analysis was conducted using AMBER20’s PTRAJ and CPPTRAJ modules. Key analyses included root mean square deviation (RMSD) and root mean square fluctuations (RMSF) to assess system stability and flexibility, the radius of gyration (RoG) for compactness, solvent accessible surface area (SASA) to analyze binding site exposure and ligand interactions, and hydrogen bond analysis for insights into interactions.

Binding free energy profiles were calculated using the Molecular Mechanics/Generalized Born Surface Area (MM/GBSA)37,38,39 approach, as follows:

$$\Delta {\text{G}}_{{{\text{bind}}}} = {\text{ G}}_{{{\text{complex}}}} {-}{\text{ G}}_{{{\text{receptor}}}} {-}{\text{ G}}_{{{\text{ligand}}}}$$
(6)
$$\Delta {\text{G}}_{{{\text{bind}}}} = {\text{ E}}_{{{\text{gas}}}} + {\text{ G}}_{{{\text{sol}}}} {-}{\text{ T}}\Delta {\text{S}}$$
(7)
$${\text{E}}_{{{\text{gas}}}} = {\text{ E}}_{{{\text{int}}}} + {\text{ E}}_{{{\text{vdW}}}} + {\text{ E}}_{{{\text{ele}}}}$$
(8)
$${\text{G}}_{{{\text{sol}}}} = {\text{ G}}_{{{\text{GB}}}} + {\text{ G}}_{{{\text{SA}}}}$$
(9)

In these equations, ΔGbind​ denotes the total binding free energy, Egas ​ indicates gas-phase energy, and Gsol ​ represents solvation energy. T and ΔS refer to temperature and solute entropy, respectively. The terms GGB ​ and GSA​ denote polar and non-polar solvation contributions, while EvdW and Eele specify van der Waals and electrostatic interactions. Residue-wise energy contributions were also calculated to provide insights into the key residues influencing binding affinity.

Results

Secondary metabolite production

The fungal secondary metabolites were produced following a systematic variation of growth media (MEA broth, rice media, oat media) to simulate a one-strain many compounds (OSMAC) approach. The crude extracts were then fractionated using solid phase extraction, which was performed using two mixed-mode reverse-phase matrices (MAX or MCX) to simplify the extracts. After the solid-phase extraction, the MEA crude extract partially purified using the MAX cartridge resulted in four fractions: MEA MAX crude extracts (0,004 g), MEA MAX 5% (0,003 g), MEA MAX 45% (0,003 g) and MEA MAX 95% (0,002 g). The MCX cartridge also resulted in MEA MCX 5% (0.001 g), MEA MCX 45% (0,001 g) and MEA MCX 95% (0,001 g) fractions. The oats crude extract purified using the MAX cartridge resulted in four fractions: oats MAX crude extracts (0,001 g), oats MAX 5% (0,001 g), oats MAX 45% (0,002 g) and oats MAX 95% (0,002 g). The MCX cartridge gave rise to four fractions: oats MCX crude extracts (0,001 g), oats MCX 5% (0,001 g), oats MCX 45% (0,001 g), oats MCX 95% (0,002 g). Rice crude extracts purified using the MAX cartridge resulted in four fractions: rice MAX crude extracts (0,001 g), rice MAX 5% (0,002 g), rice MAX 45% (0,001 g), and rice MAX 95% (0,001 g). On the other end, the MCX cartridge resulted in four fractions: rice MCX crude extracts (0,001 g), rice MCX 5% (0.001 g), rice MCX 45% (0,002 g), rice MCX 95% (0,002 g).

Luciferase-based antiviral activity assay

The anti-HIV-1 activities of the crude extracts and fractions from P. rubens P03MB2 were assessed using a luciferase-based antiviral reporter gene assay on the TZM-bl cell line. Antiviral activity was measured by calculating the percentage of viral inhibition of the extracts at varying 3-fold dilutions, and the results were reported as dose-response curves (Figs. 1, 2, 3, 4, 5 and 6). The pharmacological efficacy and potency of the extracts in this study were assessed relative to the control AZT. As expected, AZT demonstrated a strong pharmacological effect (high efficacy).

Fig. 1
figure 1

Dose-response curve of MEA media MAX crude extract’s anti-HIV activity tested in TZM-Bl cells for MEA media eluted in MAX cartridges. AZT was used as a positive control. MEA MAX crude extract (A), MEA MAX 95% fraction (B), MEA MAX 45% fraction (C) and MEA MAX 5% fraction (D).

Full size image

All the MEA MAX fractions exhibited weak efficacy (< 90% HIV inhibition). However, the crude extract and the 5% fraction displayed greater potency than AZT (0.002407 µg/mL), with values of 0.0008991 µg/mL (~ 0.27-fold) and 0.001445 µg/mL (~ 1.66-fold), respectively (Fig. 1). For the MEA MCX fractions, only the 5% fraction exhibited enhanced efficacy, achieving 97.5% HIV inhibition. Meanwhile, only the 95% fraction displayed promising potency (0.2810 µg/mL, Fig. 2). In the Oats MAX fractions, the 45% and 95% fractions showed enhanced efficacy (91% HIV inhibition). However, only the 45% fraction exhibited promising potency (0.7063 µg/mL, Fig. 3). Similarly, in the Oats MCX group, the 5% and 45% fractions demonstrated strong pharmacological effects, with 96.5% and 95.5% HIV inhibition, respectively. The 95% fraction stood out with a 0.0001169 µg/mL potency, which is more significant than that of AZT (0.002407 µg/mL). Other Oats MCX fractions also exhibited promising potency (0.02–0.04 µg/mL, Fig. 4). For the rice extracts, the 45% and 95% fractions demonstrated strong pharmacological effects, with 101% and 99% HIV inhibition, respectively (Fig. 5). The rice MAX crude extract (0.009807 µg/mL) and the 45% fraction (0.08032 µg/mL) exhibited notable potency. All rice MCX fractions displayed high efficacy (98–101% HIV inhibition), although only the 95% fraction showed promising potency (0.08168 µg/mL, Fig. 6). The HIV inhibition (pharmacological effects or efficacy) and potency (IC50) results for Penicillium rubens P03MB2 extracts are summarized in Table 1 (MEA media), Table 2 (Oat media), and Table 3 (Rice media), respectively.

Fig. 2
figure 2

Dose-response curve of MEA MCX crude extract’s anti-HIV activity tested in TZM-Bl cells for MEA media eluted in MCX cartridges. AZT was used as a positive control. MEA MCX crude extracts (A), MEA MCX 95% fraction (B), MEA MCX 45% fraction (C) and MEA 5% fraction (D).

Full size image
Fig. 3
figure 3

Dose-response curve of Penicillium rubens extract’s anti-HIV activity tested in TZM-Bl cells for Oats media eluted in MAX cartridges. AZT was used as a positive control. Oats MAX crude extracts (A), Oats MAX 95% fraction (B), Oats MAX 45% fraction (C) and Oats MAX 5% fraction (D).

Full size image
Fig. 4
figure 4

Dose-response curve of Penicillium rubens extract’s anti-HIV activity tested in TZM-Bl cells for Oats media eluted in MCX cartridges. AZT was used as a positive control. Oats MCX crude extracts (A), Oats MCX 95% fraction (B), Oats MCX 45% fraction (C) and Oats MCX 5% fraction (D).

Full size image
Fig. 5
figure 5

Dose-response curve of Penicillium rubens extract’s anti-HIV activity tested in TZM-Bl cells for Rice media eluted in MAX cartridges. AZT was used as a positive control. Rice MAX crude extracts (A), Rice MAX 95% fraction (B), Rice MAX 45% fraction (C) and Rice MAX 5% fraction (D).

Full size image
Fig. 6
figure 6

Dose-response curve of Penicillium rubens extract’s anti-HIV activity tested on TZM-Bl cells for Rice media eluted in MCX cartridges. AZT was used as a positive control. Rice MCX crude extracts (A), Rice MCX 95% fraction (B), Rice MCX 45% fraction (C) and Rice MCX 5% fraction (D).

Full size image
Table 1 HIV-1 Inhibition concentration (IC50) on TZM-Bl cells for Penicillium Rubens P03MB2 extracts obtained from MEA media.
Full size table
Table 2 HIV-1 Inhibition concentration (IC50) on TZM-Bl cells for Penicillium Rubens P03MB2 extracts obtained from oat media.
Full size table
Table 3 HIV-1 Inhibition concentration (IC50) on TZM-Bl cells for Penicillium Rubens P03MB2 extracts obtained from rice media.
Full size table

Feature-based molecular network analysis

Nine fractions and three media controls (MEA, rice, and oat) were selected for LC-HRMS/MS analysis. These included two fractions from MEA (MEA_MCX 5% and MEA_MCX 95%), four fractions from rice media (Rice_MCX 5%, Rice_MCX 45%, Rice_MCX 95% and Rice_MAX 45%), and three fractions from oats media (Oats_MCX 5%, Oats_MCX 45%, and Oats_MCX 95%). Spectral analysis was performed using a feature-based molecular network (FBMN) to obtain a global view of the chemical space in P. rubens P03MB2 under various nutrient conditions and to facilitate metabolite annotation and dereplication. The constructed FBMN-molecular network comprised 6358 node clusters into 705 molecular families with 42 annotated nodes. FBMN data cleaning and statistical analysis were performed to identify true features and remove blanks before this was embedded into the Cytoscape visualization tool. After removing blanks, the final network consisted of true features arranged into 3576 nodes with 40 annotated nodes (Fig. 7).

Fig. 7
figure 7

Feature-based molecular network of P. rubens P03MB2. The color of each node corresponds to the compound media used to produce the extract.

Full size image

Virtual screening and molecular interaction analysis

The molecular docking approach was validated through an RMSD analysis, with the re-docked pose of ORV showing an RMSD value of 0.81 Å compared to its experimentally co-crystallized orientation. Following DFT geometry optimization, frequency calculations confirmed the absence of imaginary frequencies in the optimized geometries, ensuring our computational models’ structural stability and reliability. The electronic properties of the compounds revealed nuanced differences in molecular stability and reactivity (Table 4).

Table 4 Orbital energies of the identified compounds.
Full size table

Diosgenin exhibited the largest energy gap (∆E = 7.60 eV), which indicates minimal reactivity and high kinetic stability. In contrast, cyclopiazonic acid displayed the smallest energy gap (∆E = 6.25 eV), suggesting higher reactivity. Aflatoxin B1 displayed the most negative chemical potential (µ = −4.69 eV), reflecting a strong electron-withdrawing capacity, while its highest electrophilicity index (ω = 3.41) indicates a strong tendency to accept electrons. ORV and diosgenin exhibited moderate electrophilicity (ω = 2.42 and ω = 2.11, respectively), with ORV showing balanced stability and reactivity. These findings highlight aflatoxin B1’s electrophilic nature and diosgenin’s superior stability, which could influence its binding characteristics and efficacy as an inhibitor.

Complementing the electronic property analysis, we evaluated key physicochemical properties and molecular docking interactions (Table 5). The physicochemical properties of the studied compounds revealed variations that may influence their potential as HIV-1 protease inhibitors. Diosgenin, with the highest molecular weight (414.63 Da) and lipophilicity (LogP = 5.72), suggests enhanced membrane permeability. In contrast, aflatoxin B1, with the lowest molecular weight (312.28 Da) and lower lipophilicity (LogP = 2.28), may have reduced cellular penetration. Notable differences in topological polar surface area (TPSA) were observed, with aflatoxin B1 and cyclopiazonic acid exhibiting higher TPSA values (74.97 and 73.40 Ų, respectively), which could impact solubility and transport. Additionally, ORV’s high number of hydrogen bond acceptors (8) and rotatable bonds (4) suggests significant molecular flexibility, potentially enhancing its interaction capabilities. Diosgenin demonstrated the strongest binding affinity with a docking score of −8.8 kcal/mol. Cyclopiazonic acid followed with a docking score of −7.9 kcal/mol, while aflatoxin B1 showed a binding affinity of −7.8 kcal/mol. ORV, characterized by moderate electrophilicity (ω = 2.42), displayed a comparatively lower binding affinity of −6.8 kcal/mol.

Table 5 Virtual screening of the hit compounds against P. Rubens.
Full size table

Detailed molecular interaction analyses revealed distinct binding patterns within the enzyme’s active site for each inhibitor. Diosgenin (Fig. 8A) formed a dense network of interactions characterized by van der Waals contacts with residues LEU A:23, LEU B:23, ASP B:25, ALA A:28, GLY A:48, GLY A:49, and additional stabilizing interactions through hydrogen bonds with ARG B:8, ASP A:25, and GLY B:27. Carbon-hydrogen bonds with GLY A:27 and GLY A:49 further stabilized diosgenin’s position within the hydrophobic pocket. Additionally, alkyl and π-alkyl interactions involving ILE A:50, ILE B:50, ALA A:28, PRO B:81, and VAL B:82 also contributed to the stabilization of the ligand within the active site, indicating a high affinity for HIV-1 protease. Cyclopiazonic acid (Fig. 8B) exhibited multiple interactions within the active site, forming key van der Waals contacts with ARG A:8, LEU A:23, ILE A:84, ILE B:47, LYS B:45, ASP B:29, ASP B:30, ILE B:50, and MET B:46 in the hydrophobic region. Conventional hydrogen bonds formed with GLY B:27 complemented by carbon-hydrogen bonds with GLY B:48 enhanced its binding stability through alkyl and π-alkyl interactions with VAL A:82, PRO A:81, and PHE B:53.

Fig. 8
figure 8

Molecular docking analysis highlighting key binding interactions and residue contacts for diosgenin (A) and cyclopiazonic acid (B) within the HIV-1 protease active site.

Full size image

Aflatoxin B1 (Fig. 9A) demonstrated significant binding through van der Waals interactions with LEU A:23, ILE A:84, ILE B:84, GLY A:49, GLY B:27, ASP B:30, and GLY B:49 in the hydrophobic pocket. It formed conventional hydrogen bonds with ARG A:8 and ILE B:50, while benefiting from alkyl and π-alkyl interactions involving VAL A:82, ILE A:50, ALA B:28, VAL B:32, and ILE B:47 to enhance binding stability. The control ligand ORV (Fig. 9B) displayed comprehensive interactions characterized by van der Waals contacts with LEU A:23, LEU B:23, ALA B:28, ASP B:30, VAL B:32, and ILE B:84 in hydrophobic regions. Extensive hydrogen bonding was observed with ASP A:25, GLY A:27, ASP B:25 as well as carbon-hydrogen bonds involving GLY B:49 and other residues like ALA A/B28 and ILE B:84 further secured its position within the active site through additional alkyl and π-alkyl interactions.

Fig. 9
figure 9

Molecular docking analysis highlighting key binding interactions and residue contacts for aflatoxin B (A) and ORV (B) within the HIV-1 protease active site.

Full size image

Post-MD analysis evaluated the stability and dynamics of protein-ligand interactions across all complexes. Multiple parameters were analysed to assess binding stability and conformational changes comprehensively. RMSD analysis of the protein backbone and ligand positions revealed distinct stability profiles among the complexes (Fig. 10A). All systems achieved equilibrium after approximately 50 ns, with diosgenin demonstrating superior conformational stability (1.16 Å). Cyclopiazonic acid also maintained stable binding with RMSD fluctuations reaching 1.33 Å. In contrast, aflatoxin B1 and ORV showed higher conformational mobility, with RMSD values of 1.54 Å and 1.55 Å, respectively, suggesting less stable binding interactions. Local flexibility assessment through RMSF analysis highlighted significant differences in residue mobility (Fig. 10B). The catalytic regions, particularly around residues Asp25 and Ile50, showed minimal fluctuations in diosgenin (0.86 Å) and cyclopiazonic acid (0.88 Å) complexes. Aflatoxin B1 and ORV exhibited higher residue mobility (RMSF 0.99 Å and 0.90 Å, respectively), indicating reduced binding stability. Protein compactness, monitored through RoG, revealed differential structural behaviour (Fig. 11A). Diosgenin and aflatoxin B1 maintained more compact conformations, with RoG values of 8.47 Å and 7.79 Å, respectively, while cyclopiazonic acid and ORV exhibited slight structural expansion, with RoG values of 9.45 Å and 9.14 Å, respectively.

Fig. 10
figure 10

RMSD (A) and RMSF (B) trajectories of HIV-1 protease complexes with aflatoxin B1, cyclopiazonic acid, diosgenin, and ORV, illustrating structural stability and flexibility over 200 ns molecular dynamics simulations.

Full size image
Fig. 11
figure 11

RoG (A) and SASA (B) trajectories of HIV-1 protease complexes with aflatoxin B1, cyclopiazonic acid, diosgenin, and ORV, depicting protein compactness and solvent accessibility over 200 ns molecular dynamics simulations.

Full size image

SASA measurements indicated varying degrees of solvent exposure among the complexes (Fig. 11B). Diosgenin exhibited a SASA value of approximately 741 Ų, cyclopiazonic acid showed a SASA value of around 979 Ų, ORV had a SASA value of about 420 Ų, and aflatoxin B1 displayed a SASA value of approximately 572 Ų. The lower SASA values for ORV and aflatoxin B1 suggest better burial within the binding pocket than diosgenin and cyclopiazonic acid. Hydrogen bond analysis provided further insights into binding stability (Fig. 12A). Aflatoxin B1 formed the most stable hydrogen bonding network, with an average of 4.46 hydrogen bonds throughout the simulation, followed by diosgenin with approximately 2.93 bonds and cyclopiazonic acid with around 2.08 bonds. ORV formed the fewest hydrogen bonds, averaging 1.57 bonds, which may correlate with its relatively lower binding stability.

Fig. 12
figure 12

Hydrogen bond formation (A) and PCA analysis (B) trajectories of HIV-1 protease complexes with aflatoxin B1, cyclopiazonic acid, diosgenin, and ORV, illustrating binding stability and conformational variance over 200 ns molecular dynamics simulations.

Full size image

Principal component analysis (PCA) revealed distinct conformational sampling patterns (Fig. 12B). Aflatoxin B1 exhibited the most expansive conformational space, with standard deviations of ± 7.2986 in PC1 and ± 5.8549 in PC2, indicating significant structural variability and flexibility. Cyclopiazonic acid showed moderate conformational variability, with standard deviations of ± 6.4744 in PC1 and ± 3.7219 in PC2, suggesting some adaptability while maintaining a relatively compact distribution. Diosgenin presented the most constrained conformational exploration, with the lowest standard deviations of ± 4.5341 in PC1 and ± 4.2117 in PC2. This suggests a more rigid molecular structure with tightly controlled dynamics that may lead to more predictable binding characteristics. ORV displayed substantial conformational dispersion with standard deviations of ± 6.5256 in PC1 and ± 4.2574 in PC2, indicating notable structural fluctuations contributing to its complex molecular dynamics. The pairwise cluster centre distances further illuminated these differences, with the smallest distance of 0.0390 observed between Aflatoxin B1 and ORV, and the largest distance of 0.0722 between Aflatoxin B1 and Diosgenin. These variations in conformational sampling suggest differing degrees of molecular adaptability among the compounds.

The binding free energies for each HIV-1 protease-inhibitor complex were computed using the MM/GBSA approach, with the results summarized in (Table 6). ORV demonstrated the most favourable binding free energy (ΔGbind = −47.75 ± 0.12 kcal/mol), primarily driven by strong van der Waals interactions (ΔEvdw = −56.85 ± 0.11 kcal/mol) and nonpolar solvation energies (ΔGnonpol = −7.84 ± 0.01 kcal/mol). This suggests that ORV forms robust hydrophobic interactions with HIV-1 protease, substantially contributing to its potential inhibitory mechanism. Diosgenin also exhibited a favourable binding free energy (ΔGbind = −34.59 ± 0.11 kcal/mol), characterized by substantial van der Waals interactions (ΔEvdw = −40.67 ± 0.12 kcal/mol). These hydrophobic interactions indicate a significant molecular recognition mechanism between diosgenin and the protease active site.

Table 6 Binding free energy contributions for HIV-1 protease ligand complexes.
Full size table

In contrast, cyclopiazonic acid (ΔGbind = −29.30 ± 0.44 kcal/mol) showed a distinctive binding profile with a notable contribution from electrostatic interactions (ΔEele = −9.52 ± 0.39 kcal/mol), which was counterbalanced by a higher polar solvation energy (ΔGpol = 21.03 ± 0.23 kcal/mol). The unfavourable polar solvation energy suggests that cyclopiazonic acid may experience less favourable interactions with the solvent, impacting its overall binding efficiency. Aflatoxin B1 displayed the least favourable binding energy (ΔGbind = −25.66 ± 0.21 kcal/mol), with relatively weaker contributions from van der Waals and electrostatic interactions. This suggests a comparatively less stable interaction with the HIV-1 protease.

Interaction energy decomposition analysis was performed further to elucidate each inhibitor’s contributions to binding affinity. The detailed analysis revealed distinct molecular interaction patterns across the studied compounds. Diosgenin exhibited a distinctive binding profile characterized by prominent van der Waals interactions (Fig. 13A). ARG B:8 (ΔGtotal = −1.213 kcal/mol) and PRO B:81 (ΔGtotal = −1.305 kcal/mol) were key contributors, suggesting effective molecular packing within the binding pocket. Additional stability was derived from electrostatic and non-polar solvation effects, particularly from hydrophobic residues such as ILE 50 (ΔGtotal = −1.243 kcal/mol) and LEU B:23 (ΔGtotal = −1.418 kcal/mol). ORV’s binding was predominantly characterized by significant van der Waals interactions, with key hydrophobic residues playing critical stabilization roles (Fig. 13B). Notably, ILE B:50 (ΔGtotal = −2.184 kcal/mol) and PRO A:81 (ΔGtotal = −1.106 kcal/mol) emerged as pivotal residues, establishing strong hydrophobic interactions. Complementary hydrophobic residues such as VAL A:82 and ILE A:84 further enhanced binding stability, underscoring the prominence of hydrophobic interactions in ORV’s binding mechanism.

Fig. 13
figure 13

Per-residue energy decomposition plots showing the contributions to the binding and stabilisation of (A) diosgenin and (B) ORV in the HIV-1 protease active site during 200 ns molecular dynamics simulations.

Full size image

Similarly, aflatoxin B1 demonstrated a complex interplay between van der Waals and electrostatic interactions. ARG A:8 (ΔGtotal = −2.386 kcal/mol) stood out as a key contributor, combining favourable van der Waals and electrostatic terms (Fig. 14A). GLY A:27 (ΔGtotal = −1.559 kcal/mol) also significantly stabilises the ligand-receptor complex, suggesting that hydrophobic and electrostatic interactions were crucial for aflatoxin’s binding affinity. Hydrophobic residues predominantly drove the binding of cyclopiazonic acid. ILE A:50 (ΔGtotal = −1.243 kcal/mol) and VAL B:82 (ΔGtotal = −1.233 kcal/mol) underscored the importance of hydrophobic interactions in enhancing binding stability (Fig. 14B). Interestingly, polar residues like ASP A:29 (ΔGtotal = 0.582 kcal/mol) contributed positively, indicating that polar solvation effects may modulate the overall binding dynamics.

Fig. 14
figure 14

Per-residue energy decomposition plots showing the contributions to the binding and stabilization of (A) aflatoxin B1 and (B) cyclopiazonic acid in the HIV-1 protease active site during 200 ns molecular dynamics simulations.

Full size image

Discussion

The P. rubens P03MB2 complex crude extracts and fractions were analyzed for anti-HIV-1 activity using the luciferase reporter gene assay. The MEA crude extract and 5% MAX fraction showed higher inhibition of HIV-1 virus with 2.67 and 1.66-fold more than the AZT control (0.00241 µg/mL), respectively. The most notable antiviral activity in oats fractions was observed in the MCX 95% fraction, which exhibited an IC50 value of 0,0001169 µg/mL, 21-fold higher than that of the positive reference AZT. Two fractions of rice (crude extracts and MAX 45% fractions) had good anti-HIV-1 activity profiles, reaching IC50 of 0.00981 and 0.08032 µg/mL, respectively. In the MCX fraction of oats media and rice media, we noted that antiviral activity (IC50) increased proportionally with the increasing strength of methanol used to elute the fraction. Increased strength of an organic solvent, such as methanol, has been shown to result in a higher yield in the extraction of secondary metabolites40. Although all media evaluated in this study produced extracts with anti-HIV-1 activity, good activity was observed from oats media.

We further conducted an untargeted metabolomics analysis followed by a feature-based molecular networking. Molecular networking resulted in 3576 metabolites with 40 annotated compounds, revealing that only 1.1% of compounds were known from this analysis. Only 24 of the annotated compounds were secondary metabolites produced from the fungal extracts. Most fungal secondary metabolites are produced from non-canonical biosynthetic pathways, making their prediction using bioinformatics pipelines challenging. Based on this fact, the low annotation observed in this study is not surprising. To prioritize compounds with potential anti-HIV-1 activity, we adopted an in silico approach to analyse the binding efficiencies of identified compounds from MS spectra to an HIV protease enzyme using molecular docking. Compounds including diosgenin, alpha-cyclopiazonic acid, alternariol, daidzein and aflatoxin showed a docking score between − 7 and − 8.8 kcal/mol. Most of these compounds are known to be mycotoxins produced by different fungal species.

Cyclopiazonic acid (CPA) exhibited a binding affinity of −7.9 kcal/mol with a moderate energy gap (ΔE = 6.25 eV) and balanced electronic properties (ω = 2.36). It also showed to engage in stable van der Waals and π-alkyl interactions with key residues such as ARG A:8 and PRO A:81. CPA is an indole alkaloid produced by certain fungal species, including those from the genera Aspergillus and Penicillium41. CPA was produced abundantly in oats MCX 95%, corroborating the higher potency observed in the oats MCX 95% fraction. CPA has been shown in previous docking studies to exhibit antiviral activity against HIV protease (−7.9 kcal/mol), COVID-19 (−6.61 kcal/mol42and human respiratory syncytial virus (RSV). In other studies, it was shown to inhibit calcium ATPase in the endoplasmic reticulum, thereby increasing intracellular calcium in Hep2 cells43. Time of addition assays conducted in a study by Cui et al.44 showed that it inhibits the viral genome replication and transcription steps. Therefore, one can deduce that the inhibition mechanism occurs through increased intracellular calcium concentration, which is detrimental to viral replication. This might make CPA an interesting lead candidate for chemical modification since it presents a unique antiviral mechanism based on increasing intracellular calcium concentration in cells.

Another interesting molecule was diosgenin (spirost-5-en-3-ol), also identified in oats media fractions (MCX 45% and 95%). We observed that diosgenin had a significant binding affinity (−8.8 kcal/mol) and MM-GBSA ΔG score of −50.2 kcal/mol, reflecting its strong binding potential. Diosgenin’s large energy gap (ΔE = 7.60 eV) ensures high molecular stability and low reactivity. At the same time, its hydrophobic character (LogP = 5.72) and moderate electrophilicity index (ω = 2.11) contribute to robust interactions with the protease’s active site. Notably, diosgenin forms significant van der Waals interactions with LEU A:23 and LEU B:23 and hydrogen bonds with catalytic residues ARG B:8 and ASP A:25. These features allow for reduced molecular flexibility and enhanced stabilization within the active site, marking diosgenin as a standout candidate for further therapeutic exploration.

Diosgenin is a steroidal saponin used in pharmaceuticals as a precursor for steroidal drugs and is mainly found in plants. It is known for its bioactivity properties, including antibacterial, anti-inflammatory, anti-cardiovascular, hypoglycaemic, antioxidant, anticancer, and many other effects45. Also, the anti-HIV-1 effect of diosgenin is well documented, with the latest report showing that diosgenin isolated from Helicteres isora plant leaves inhibited two distinct HIV-1 subtype strains46. The results from our virtual screening and molecular interaction studies demonstrate the potential of diosgenin as the most effective protease inhibitor among the analysed compounds. The post-MD analysis comprehensively evaluated the binding characteristics of four ligands, with diosgenin emerging as the most promising candidate. Structural stability metrics provided the first indication of diosgenin’s superior performance. The compound demonstrated the lowest RMSD of 1.16 Å and RMSF of 0.86 Å, significantly lower than other studied compounds. These values directly translate to minimal conformational fluctuations and superior binding stability. The radius of gyration of 8.47 Å further validated these findings, revealing a tightly packed protein conformation that suggests robust interactions within the binding pocket.

Hydrogen bonding and solvent interactions offered additional insights into diosgenin’s molecular behaviour. The compound maintained a moderately stable hydrogen bond network of 2.93 bonds and displayed a relatively low SASA value of 741 Ų. These characteristics indicate optimal pocket occupancy and limited solvent exposure, further reinforcing its potential as an effective ligand. Binding energy calculations provided quantitative evidence of diosgenin’s efficacy. The compound demonstrated a strong affinity for the target protein with a binding free energy (ΔGbind) of −34.59 ± 0.11 kcal/mol, primarily driven by hydrophobic interactions. Principal component analysis complemented these findings, revealing constrained dynamics with minimal conformational variability.

In comparison, cyclopiazonic acid showed slightly less favourable characteristics. With marginally higher RMSD (1.33 Å) and RMSF (0.88 Å) values, and a less favourable binding free energy (ΔGbind) of −29.61 ± 0.09 kcal/mol, the compound exhibited fewer stable interactions compared to diosgenin. Aflatoxin B1 presented the most challenging molecular profile. Its high RMSD (1.54 Å) and RMSF (0.99 Å) values indicated significant structural flexibility and instability. The binding free energy (ΔGbind) of −25.66 ± 0.21 kcal/mol, combined with expansive conformational variability, suggested the weakest protein-ligand interactions among the studied compounds. Diosgenin outperformed the other compounds regarding stability, binding energy, and protein-ligand interaction patterns, making it a compelling candidate for further research and development. These findings offer a promising avenue for exploring natural product-derived compounds from endophytic fungi like P. rubens.

P. rubens has been established as an industrial producer of penicillin, a β-lactam antibiotic47. Its role as a penicillin producer has placed this prolific species in the limelight of biotechnological interest in producing secondary metabolites48. In addition, this has led to the species being the most extensively studied amongst its genus of more than 350 Penicillium species for novel secondary metabolites with medicinal potential48,49. However, very few studies have evaluated and reported on its production of secondary metabolites with anti-HIV properties, particularly secondary metabolites with anti-HIV-1 properties. For example, Bringmann et al.8 described a novel sorbicillinoid alkaloid (Sorbicillatone A) to demonstrate significant antiviral activity on the expression of HIV viral proteins.

Conclusion

The observed variances in the production of secondary metabolites with antiviral activities from different media indicated the role played by different stimuli in these media to mimic the conditions in the natural environment that awaken silent BGCs. The most promising compounds with anti-HIV activity were observed in oat media. The MM-GBSA ΔGbind calculations revealed diosgenin as the most promising compound, with a favourable binding free energy of −34.59 kcal/mol. This underscores its strong binding potential and stability within the protease active site. These findings emphasize the biosynthetic abilities of P. rubens cultivated in oats media as a good source of anti-HIV-1 secondary metabolites.