Combinatorial QM and MD in silico design of natural product-based DHFR inhibitors

Abstract

Cancer cells are distinguished from normal cells by their rapid rate of division. This high division rate can be explained by various factors, including unregulated cell cycle progression, which occurs when cancer cells bypass checkpoints; activation of growth signals in cancer cells; and override of growth suppressor factors, which generally regulate cell growth and division within physiological thresholds. Cancer cells bypass these controls due to genetic mutations or epigenetic changes. Additionally, changes in the tumor microenvironment (TME) can contribute to the accelerated growth and division of cancer cells. It is well established that cancer cells accumulate mutations more frequently than normal cells. This accumulation can, in part, explain their rapid rate of division. The cell cycle consists of four phases: G1, S, G2, and mitosis, with the first three phases categorized as interphase. When cancer cells experience division stress, their demand for nucleotide synthesis increases, which is essential for RNA and DNA synthesis. RNA synthesis primarily occurs during interphase, while DNA replication occurs in the S phase. One of the conserved enzymes involved in nucleotide synthesis is dihydrofolate reductase (DHFR). This enzyme’s role in purine and thymidylate synthesis is crucial in cancerous cells under conditions of division stress. Methotrexate, a well-known DHFR inhibitor, has been introduced for the treatment of cancers such as meningeal leukemia, lymphoma, and breast cancer. While effective in alleviating cancer symptoms, methotrexate can cause adverse effects, including hepatotoxicity, pulmonary complications, and renal impairment. Based on these considerations, we applied combinatorial studies, including molecular dynamics simulations alongside quantum mechanics, to design novel DHFR inhibitors for cancer cells using carbohydrate- and amino acid-based scaffolds. Additionally, our studies suggest that the designed inhibitors may exhibit fewer side effects than methotrexate.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-21262-5.

Introduction

Dihydrofolate reductase (DHFR) is a crucial enzyme in the synthesis of purines and thymidylate by converting dihydrofolate (DHF) to tetrahydrofolate (THF)¹. Purines and thymidylate are essential raw materials for cell proliferation, making them critical for synthesizing RNA and DNA^2,3. In rapidly dividing cancer cells, inhibiting the synthesis of pyrimidines and purines leads to cell death, underscoring the importance of targeting this pathway in anticancer therapy³. Consequently, DHFR has become one of the most attractive targets for developing novel anticancer drugs over the past two decades³.

The enzyme was first identified in 1958 as a target of the antifolate drug methotrexate in mammalian cells⁴. Moreover, DHFR inhibitors have broad therapeutic applications for treating malaria, as well as fungal, bacterial, and mycobacterial infections⁵. Notably, until 1980, DHFR was the only folate-dependent enzyme targeted in cancer chemotherapy^2,6–8.

The role of DHFR in thymidylate (TMP) synthesis is illustrated in Fig. 1. Folic acid, a water-soluble vitamin obtained through the diet, is an inactive precursor that undergoes reduction to tetrahydrofolic acid (THF) in two steps, facilitated by DHFR, consuming NADPH³. In the first step, DHFR reduces folic acid to dihydrofolic acid (DHF); in the second step, it further reduces DHF to THF. Subsequently, THF is converted to various forms, including 5,10-methylene-THF, 5,10-methenyl-THF, 5-formyl-THF, and 10-formyl-THF³.

Fig. 1.

The role of the Dihydrofolate reductase enzyme in thymidylate (TMP) synthesis can be observed.

As shown in Fig. 1, additional enzymes, such as serine hydroxymethyltransferase and thymidylate synthase, are involved in the synthesis of thymidylate. During this process, deoxyuridine monophosphate (dUMP) is methylated and converted to deoxythymidine monophosphate (dTMP) through the combined action of DHFR and thymidylate synthase^9,10. Furthermore, 5,10-methylene-THF, along with other forms such as 5,10-methenyl-THF, 5-formyl-THF, and 10-formyl-THF, participate in one-carbon unit transfer reactions essential for various biosynthetic processes³.

Since dihydrofolate reductase (DHFR) is a highly conserved enzyme, methotrexate (MTX) has been widely used as a competitive inhibitor targeting the DHFR substrate-binding site. The chemical structure of MTX is shown in Fig. 2a¹¹. Methotrexate is classified as an antifolate agent¹², and it functions by preventing DHFR from binding to folic acid, as illustrated in Fig. 1. MTX has been shown to induce apoptosis, particularly in rapidly proliferating cancer cells, by disrupting nucleotide biosynthesis¹².

Fig. 2.

(a) Methotrexate (MTX) can have function as the blocking agent of the substrate binding site of DHFR, (b) Our designed structures to inhibit DHFR: MNK.

Several other DHFR inhibitors have also been developed, including pemetrexed, which is used in cases of lung cancer and mesothelioma; raltitrexed, which is applied for colorectal cancer; and pralatrexate, which is specifically used for T-cell lymphoma. Other DHFR inhibitors, such as piritrexim, talotrexin, and nolatrexedm^3,13, progressed through clinical trials; however, due to their toxicity, the trials were halted. Moreover, compounds containing quinazoline, pyridopyrimidine, and triazine scaffolds are categorized as preclinical DHFR inhibitors and have demonstrated significant DHFR inhibition along with anticancer activity³.

However, this study focuses specifically on MTX for the following reasons:

• Established clinical efficacy: MTX is one of the most effective chemotherapeutic agents for a wide range of cancers, including meningeal leukemia¹⁴, lymphoma¹⁵, osteosarcoma¹⁶, non-Hodgkin’s lymphoma¹⁷, breast cancer¹⁸, and bladder cancer¹⁹.

• Broader therapeutic applications: In addition to its use in oncology, MTX has demonstrated efficacy in treating non-malignant conditions such as rheumatoid arthritis, psoriasis²⁰, and various other autoimmune disorders¹³.

• Activity against DHFR isoforms: MTX has been reported to inhibit multiple isoforms of DHFR, further supporting its versatility and effectiveness as a therapeutic agent^21,22.

Initially discovered in 1947²³, methotrexate was approved by both the U.S. Food and Drug Administration (FDA) and the European Medicines Agency (EMA) for cancer treatment in 1985¹³.

Despite its broad utility, MTX is associated with several notable side effects, including neurotoxicity²⁴, hematologic toxicity²⁵, pneumonitis²⁶, gastrointestinal disturbances²⁷, and renal impairment²⁸.

Due to the aforementioned side effects, our research group aimed to design novel DHFR inhibitors based on carbohydrates and amino acids with enhanced inhibitory efficacy and reduced toxicity.

Carbohydrates were selected as one of the primary building blocks of our designed structures because their presence can induce beneficial effects, such as improved metabolic stability²⁹, enhanced solubility, effective membrane permeability³⁰, acceptable biodistribution³¹, and more favourable ligand-receptor interactions³². In addition to the role of carbohydrates in drug discovery, peptides such as lactoferricin-derived peptides have exhibited significant antitumor activity³³. Furthermore, the ability of small molecules to efficiently permeate cell membranes and rapidly reach their targets³³ motivated us to base our designs on small molecules, incorporating monosaccharide building blocks and approximately three to four amino acids.

In the initial step, 20 structures incorporating amino acids and carbohydrates were designed based on their Electrostatic Potential (ESP) maps compared to MTX³⁴. Additional parameters considered in the design included surface area, volume, the number of aromatic rings, and the number of electronegative atoms, all of which were compared with those of the reference drug, MTX. Before proceeding to further studies, the selected structures from the previous step were optimized at the HF/6-31G* level of theory using the AutoDock/Vina plugin with scripts from the AutoDock Tools package^35,36.

For the docking study, we analysed the interactions between the designed structures and DHFR, comparing them to the interactions between MTX and DHFR. The protein structure used for docking was retrieved from the Protein Data Bank (PDB ID: 3EIG)³⁷.

After docking/Vina analyses, the structure shown in Fig. 2(b), referred to as MNK, exhibited the greatest similarity to MTX in the presence of the receptor compared to the other designed molecules and was therefore selected. In Table 1 (entry 14), MNK is identified as Trp-Tyr (Gluc)-Glu, where Trp, Tyr, Glu, and Gluc represent tryptophan, tyrosine, glutamic acid, and glucose, respectively.

Table 1.

Structural information and AutoDock/Vina results for methotrexate and designed amino acid-carbohydrate-based structures for DHFR inhibition.

Subsequently, the grid box searching procedure was performed for MNK. Two conformers of this molecule—one with the optimal pose and another with the most negative binding energy toward DHFR—were examined using molecular dynamics (MD) simulations in GROMACS 5.2³⁸. The results from the preceding steps were inputs for the MD studies.

The Discovery Studio modelling environment (Discovery Studio 4.5)³⁹ was used to analyze intermolecular interactions, including hydrogen bonding, electrostatic interactions, and van der Waals forces, between DHFR and MTX, as well as between the selected MNK conformers and the receptor.

Method and theoretical calculations

The selection of amino acids and monosaccharides for designing new anticancer drugs

This research selected glucose as the monosaccharide in the designed structures for two reasons. Most cancerous cells rely on increased glucose metabolism to produce cellular metabolites⁴⁰. Secondly, the overexpression of glucose transporters in cancer cells⁴¹ can enhance the selectivity of the designed structures toward cancerous cells (refer to Table 1, entries 2–21).

According to various studies, cyclic peptides exhibit better resistance to protease enzymes in the digestive system than linear peptides⁴². Therefore, cyclic peptides were incorporated into the design of specific structures (refer to Table 1, entries 7, 9, and 11–13). The number of amino acids had to exceed three to integrate the cyclic peptide moiety into the designed drug structures, resulting in larger structures than those containing three or fewer amino acids.

Additionally, tyrosine was included in the design of all compounds in this study, as Wadzinski et al. demonstrated the high potential of the phenolic group of tyrosine to play a crucial role in O-glycosylation processes⁴³ (see Table 1).

Figures 1 and 2 illustrate that folate and methotrexate contain three aromatic centers in their structures. Consequently, tryptophan and histidine were incorporated into most of the designed compounds alongside tyrosine.

The presence of glutamic acid in the structure of MTX (Fig. 2(a)) enables this drug to bind to reduced folate carrier 1, facilitating its entry into cells. Subsequently, folylpolyglutamate synthetase catalyzes the conversion of MTX into MTX polyglutamate (MTXPGs). These polyglutamate derivatives exhibit better retention in target cells and higher binding affinities for dihydrofolate reductase than MTX, thereby enhancing enzyme inhibition⁴⁴. As shown in Table 1, glutamic acid was incorporated into most of the compounds we designed.

Evaluation of volume, area, and electrostatic potential map of MTX and designed structures based on amino acids and carbohydrates

The initial geometrical optimization of the designed structures listed in Table 1 (entries 2–21) was performed using the MMFF force field (molecular mechanics) in Spartan 10 software³⁴. This was followed by a conformer search within a relative energy range of 0–10 kcal·mol⁻¹. Subsequently, the potential energy surfaces, volume, and area of the designed molecules were determined and compared with the corresponding parameters of the MTX molecule, extracted from the structure with the PDB ID 3EIG.

Specific systems were excluded from further analysis as their volumes, areas, and electronic properties deviated significantly from those of the reference drug (refer to Table 1, entries 6–13). As shown in Table 1, increasing the number of amino acids in the designed compounds led to more pronounced differences in critical parameters, including volume and area, than the reference drug. Moreover, structures containing cyclic peptides exhibited a lower area-to-volume ratio than those with linear peptides. The volume of the designed compounds incorporating cyclic peptides varied by approximately 50% relative to the reference drug. Based on these observations, these structures were not considered for further study (Table 1, entries 7, 9, 11, 12, and 13).

After calculating the designed structures’ volume, area, and electrostatic potential maps and comparing these features with MTX, 12 compounds were selected for docking studies. The results are presented in the following section.

Molecular docking and grid box search for our proposed structures

We performed geometry optimizations and frequency calculations for the remaining 12 compounds using the HF/6-31+G* basis set. The absence of imaginary frequencies confirmed that the optimized structures corresponded to true minima. The MTX structure was obtained from X-ray crystallography data (PDB ID: 3EIG); therefore, further optimization was not required.

The MTX–protein complex was processed using the VMD package before the docking procedure⁴⁵. Water molecules, other solvents, and co-crystallized compounds were removed using Discovery Studio 4.5.

The docking process was carried out using a grid box with dimensions of 30 × 24 × 24 ų, centered at x = 10.33, y = –4.78, and z = –18.398.

To validate the docking procedure and grid box parameters, we re-docked the MTX structure extracted from PDB ID 3EIG, following the method described by Baammi et al.⁴⁶. The output structure of MTX from AutoDock/Vina was then overlaid with the original MTX structure from the X-ray crystallography data. The calculated RMSD between the docked pose and the crystallographic structure was 0.463 Å (see Fig. 3). Since this value is < 2 Å^46,47, the docking parameters used in this section were deemed acceptable.

Fig. 3.

The obtained structure of MTX after Autodock/vina was superimposed with the MTX structure extracted from the first-order X-ray crystallography. The RMSD obtained was calculated to be 0.463 Å, which is acceptable.

For the next step of this study, we aimed to select the optimal inhibitors from our designed DHFR inhibitors. The inhibitors in entries 9, 16, and 17 of Table 1 were excluded due to their relatively low binding affinities toward DHFR.

As discussed in the previous section, additional parameters for selecting the optimal inhibitors included the area-to-volume ratio and the polar surface area (PSA) values. PSA is defined as the surface area of a molecule associated with heteroatoms (such as oxygen, nitrogen, and phosphorus) and polar hydrogen atoms⁴⁸.

The structure in entry 15 was omitted due to its low area-to-volume ratio. Subsequently, we retained those with the highest PSA values among the remaining structures. Consequently, the structures from entries 2, 3, 5, 20, and 21 were excluded because their PSA values were relatively low compared to those of the other designed inhibitors.

The structures in entries 14 and 18 exhibited acceptable characteristics, such as binding affinity toward DHFR, area-to-volume ratio, and PSA values. Among the remaining designed inhibitors, the structures in entries 4 and 19 were found to be similar; therefore, one was selected for further studies.

To assess the similarity of these structures to MTX in their interactions with the DHFR protein, two-dimensional images illustrating the interactions of MTX and the 12 designed structures with DHFR were generated. Figure 4(a) shows that many non-polar residues in the protein’s active site, such as isoleucine 60, interact with the aromatic rings in the MTX structure. Additionally, the pteridine ring of MTX formed hydrogen bonds with glutamine 30, valine 115, and isoleucine 7. Arginine 70 and arginine 31, polar residues of DHFR, interacted with the acidic glutamic fragment of MTX.

Fig. 4.

(a) A two-dimensional image representing the interaction of MTX and DHFR protein was extracted from X-ray crystallography using the PDB ID of 3EIG. A two-dimensional image of the interactions between structures including: (b) Trp-His-Tyr (gluc) (Table1, entry4), (c) Trp-Tyr (gluc)-GLU (Table 1, entry 14), (d) Asp-His-Tyr (gluc) (Table 1, entry 18), and (e) GLU-Trp-Tyr (gluc) (Table 1, entry 19) and DHFR protein, that have been obtained from Autodock/vina strudies.

Interactions with the protein’s active site are critical criteria for selecting optimal structures for subsequent steps. Therefore, the compounds listed in entries 4, 14, 18, and 19 of Table 1, which exhibited favourable binding energies and interactions at the desired positions with the DHFR protein, were selected for the next phase. Additionally, among the remaining inhibitors, the structures in entries 4, 14, and 18 demonstrated larger PSA values and acceptable binding affinities toward the receptor in this study. The interactions of these selected structures with the key residues of the DHFR protein are illustrated in Fig. 4b–e, respectively.

Furthermore, the structures in entries 4 and 19 of Table 1 exhibited similar behaviour in their interactions with the DHFR receptor., we selected one of them—the structure in entry 4 of Table 1—to continue our studies. The other structures, including those in entries 14 and 18, were also chosen for further investigation.

To identify the optimal active site of the DHFR protein, we conducted a grid box searching process using AutoDock 4.2.2 for the three selected structures: Trp-His-Tyr (gluc) (Table 1, entry 4), Trp-Tyr (gluc)-GLU (Table 1, entry 14), and Asp-His-Tyr (gluc) (Table 1, entry 18)⁴⁹. The findings were compared with those for MTX. The spacing between grid points was set to 0.375 Å, and the grid box size was fixed at 126 × 126 × 126 ų, with its center located at x = 9.541 Å, y = − 3.560 Å, and z = − 10.567 Å for each calculation. Additionally, a grid box search was performed for MTX (the reference structure with PDB ID 3EIG) to validate our methodology.

Two thousand docking runs were performed for each ligand, including MTX, Trp-His-Tyr (gluc) (Table 1, entry 4), Trp-Tyr (gluc)-GLU (Table 1, entry 14), and Asp-His-Tyr (gluc) (Table 1, entry 18), using the local search genetic algorithm (LGA)^50,52. The distribution chart illustrating the positions of MTX conformers relative to the DHFR receptor is presented in Figure S1. According to this chart, one of the MTX conformers, MTX-1-1, exhibited the most negative binding energy towards DHFR (Fig. 5a). In contrast, another conformer, MTX-86-1, was the most populated (Fig. 5b). The binding energy of MTX-1-1 toward the receptor was calculated to be − 14.25 kcal/mol, whereas the binding energy of the most populated conformer, MTX-86-1, was estimated to be − 8.5 kcal/mol.

Fig. 5.

(a) The optimal binding pose of MTX with DHFR, exhibiting the most negative binding energy after conducting 2000 docking searches using AutoDock. (b) The most populated conformer of MTX bound to DHFR, as determined by 2000 docking searches using AutoDock.

According to the grid box search, MTX-1-1's position with the receptor (Fig. 5a) aligns with the X-ray crystallography structure (PDB ID: 3EIG).

Following the grid box search for the structures Trp-His-Tyr (gluc) (Table 1, entry 4), Trp-Tyr (gluc)-GLU (Table 1, entry 14), and Asp-His-Tyr (gluc) (Table 1, entry 18), the distribution charts for the grid box search conformers of these structures are presented in Figures S2–S4. Our results indicate that the conformer of Trp-Tyr (gluc)-GLU with the most negative binding energy toward the receptor exhibited the closest positional similarity to MTX-1-1-1 on the receptor compared to the other structures, including Trp-His-Tyr (gluc) (Table 1, entry 4) and Asp-His-Tyr (gluc) (Table 1, entry 18).

The conformer of Trp-Tyr (gluc)-GLU with the most negative binding energy toward the receptor (-9.74 kcal/mol) is referred to as MNK and is depicted in Fig. 6a. According to the AutoDock grid box search results, MNK interacts with the identical DHFR residues as MTX-1-1-1. These residues include isoleucine 16, aspartic acid 21, leucine 22, phenylalanine 34, lysine 54, lysine 55, and serine 118.

Fig. 6.

(a). The conformer of Trp-Tyr (gluc)-GLU (MNK) with the most negative binding energy towards the receptor (− 9.74 kcal/mol). MNK interacts with the same residues of DHFR that MTX-1-1 interacts with. (b): The most populated conformer of Trp-Tyr (gluc)-GLU (named UNK in this study) obtained from the grid box search.

Additionally, the most populated conformer of Trp-Tyr (gluc)-GLU, obtained from the grid box search, is shown in Fig. 6b. This conformer is referred to as UNK throughout the remainder of this study. As depicted in Fig. 6b, Trp-Tyr (gluc)-GLU interacts with DHFR residues, including leucine 166, methionine 37, tyrosine 162, glutamic acid 35, glycine 164, and threonine 40.

MNK and UNK structures, which resulted from the AutoDock grid box search as conformers of Trp-Tyr (gluc)-GLU, were selected for molecular dynamics simulation studies. Their interactions with DHFR were compared to those of MTX.

QM-based solubility assessment of selected designed structures in aqueous and organic phases

One critical consideration in designing new DHFR inhibitors is their solubility in both organic and aqueous phases. Since these inhibitors target DHFR within the cytoplasm of cells, they must exhibit significant solubility in water. Additionally, because the drugs must permeate the cell membrane to reach their target, their solubility in the lipophilic phase is also crucial.

To determine the logP parameter for the designed structures, we must calculate the free energy changes associated with transferring them between water and the 1-octanol phase. Equations 1 and 2 provide the relevant equations.

1

2

The free energy changes contributing to the solubility of our structures in both water and 1-octanol were calculated as ΔG₂ and ΔG₁. We derived ΔG for the transfer process between the phases above by determining the difference between these free energy changes. Equation 2 indicates how the ΔG transferring process is associated with logP⁵¹, representing the lipophilicity and hydrophilicity of drug molecules. In Eq. 2, R and T denote the ideal gas constant and temperature (298 K).

We began this part of our research by optimizing the chemical structures in both aqueous and organic phases using density functional theory (DFT) calculations in conjunction with a solvation model based on density (SMD)^52,55.

To validate our QM-based calculations, we selected seven known chemical structures—phenol⁵³, tramadol⁵⁴, methotrexate⁵⁸, furfural⁵³, phencyclidine⁵⁵, cytarabine⁵⁶, (2S,3R,4R,5S,6R)-2,4,5-trihydroxy-6-(hydroxymethyl) tetrahydro-2H-pyran-3-yl-L-tryptophan, abbreviated as D-fructose-L-Trp⁶¹, and (((2R,3R,4S)-2,3,4trihydroxytetrahydrofuran-2-yl) methyl)-Ltryptophan which is abbreviated as Xyl-Trp⁵⁷ with experimentally reported logP values. Using the approach above, we calculated their logP values (please refer to Table 2 and Figures S5 and S6 for the optimized structures in water and 1-octanol, respectively).

Table 2.

LogP values from experimental measurements and those calculated at the rM062X/6-31+G(d) level of theory.

(a). 56, (b). 59, (c) 57, (d) 60, (e) 61, and (f) 62, (g) 58

Subsequently, we plotted the logP values of these known structures against the QM-calculated logP values obtained in our study. Figure 7 illustrates the correlation between the experimental logP values of the selected structures and the calculated logP values.

Fig. 7.

Plot of the experimental log P versus the predicted clog P values.

As shown in Fig. 7, the experimentally measured logP values of the structures above exhibit a linear correlation with their theoretically calculated values at the rM062X/6–31 + G(d) level.

The graph in Fig. 7 can be described by the equation with . Here, y represents the experimentally reported logP values of the known structures listed in Table 3, while x corresponds to the QM-based calculated logP values of these structures.

Table 3.

Calculated solubilities in water and the organic phase, along with expected log P values from the plot in Fig. 7, for selected Designed molecules, including Trp-His-Tyr (gluc), MNK, and Asp-His-Tyr (gluc).

Additionally, we selected three structures from our designed molecules to assess their logP values using the approach mentioned above and to predict their expected logP values from the plot in Fig. 7. These structures, along with their calculated solubilities in both the water and organic phases, and their expected logP values from the graph in Fig. 7, are presented in Table 3.

It should be noted that the gas-phase optimized structures of the compounds, including Trp-His-Tyr (gluc), MNK, and Asp-His-Tyr (gluc), along with their optimized structures in water and 1-octanol, can be found in Figures S7, S8, and S9, respectively.

The data in Table 3 indicate that the Trp-Tyr (gluc)-GLU (MNK) structure has the most similar logP value to methotrexate among the other structures in Table 3. As a result, we expect that the permeation of MNK through the cell membrane will be similar to that of MTX.

As shown in Table 3, our designed structures exhibit acceptable solubility in both the organic and aqueous phases, indicating that they can interact well with the cell membrane and are soluble in the cytoplasm, where their target is located. However, their logP values are relatively low due to the incorporation of glucose into their structures. On the other hand, we anticipate that the glucose moiety in our designed structures will make them recognizable by glucose transporters (GLUTs), which facilitate glucose diffusion^58,59 into cells and are often overexpressed^59,60 in cancer cells due to the Warburg effect^62,63,69.

Furthermore, we expect that the incorporation of glucose into our designed structures will lead to their selective absorption by cancer cells, owing to their increased glucose consumption. Additionally, encapsulating our designed structures in nanoparticles or liposomes⁶⁹, which are expected to enter cells via endocytosis, could be another strategy for delivering our structures into cancer cells—an avenue we plan to explore in future research projects.

Based on the results obtained from various parts of our study, we selected two conformers of the structure Trp-Tyr (gluc)-GLU (MNK and UNK) for MD simulations. Furthermore, in the next section, the interactions between these two conformers and DHFR will be compared with the interaction between MTX (the main drug and our reference point) and DHFR.

Molecular dynamics simulation (MD)

To initiate the MD simulation, topology files for the molecules, including MTX, MNK, and UNK, were created according to the GROMOS 43a1 force field. The complex of these three molecules with the receptor was then considered separately. Each system generated receptor topology files based on the GROMOS 43a1 force field. Complexes of enzymes and ligands in each system were placed in a cubic box. A layer of SPC water molecules was added to each system as a solvent⁷¹.

Long-range Coulombic interactions were calculated using the PME method⁶⁵. The steepest descent algorithm was applied three times for each system in the initial steps to minimize the overall system. After minimization, the entire system was equilibrated in the NVT ensemble. The NVT equilibration step was performed at 100 K for 500 picoseconds. Following this, the temperature of the systems was increased from 100 to 300 K during the NPT equilibration steps⁷³.

NPT simulations were repeated five times for each system to achieve a temperature of 300 K, with each stage lasting one nanosecond. During the NPT step, both barostat and thermostat couplings were performed using the Berendsen algorithm^66,75. After NPT equilibration, the MD simulation for each system was carried out for approximately 360 ns. The Berendsen algorithm was applied for both barostat and thermostat couplings. Short-range electrostatic and van der Waals interactions were calculated within a cutoff distance of 1.2 nm.

Results and discussions

Comparison of the dynamic behaviour of DHFR in the presence of MTX, MNK, and UNK

After performing molecular dynamics (MD) simulations, we assessed and compared the dynamic behaviour of the DHFR enzyme in the presence of MTX and MNK, as both structures interacted with the same residues. The root-mean-square deviation (RMSD) values of DHFR in the presence of these structures were plotted over a 360-ns MD simulation (Fig. 8a and b, respectively). As shown in Fig. 8a, the average RMSD value of DHFR in the presence of MTX during 360 ns of the MD simulation fluctuated around 0.158 nm. In contrast, Fig. 8b shows that the average RMSD value of DHFR in the presence of MNK was 0.216 nm. Since the difference between these two average RMSD values is no greater than 0.058 nm, we conclude that the DHFR structure remains stable in the presence of both MTX and MNK.

Fig. 8.

(a) DHFR backbone RMSD values in the presence of MTX (series 1, blue plot). (b) DHFR backbone RMSD values in the presence of MNK (series 2, red plot), and (c) DHFR backbone RMSD values in the presence of UNK (series 3, pink plot).

Additionally, we investigated the interactions of another MNK conformer, designated UNK, with DHFR. As this conformer interacted with different residues compared to MTX and MNK, we obtained the RMSD plot of DHFR in the presence of UNK (Fig. 8c). The average RMSD value of DHFR in the presence of UNK was 0.164 nm. This analysis confirms the stability of the DHFR structure in the presence of UNK.

To recapitulate the RMSD analyses of the three systems, DHFR maintains a largely stable conformation in the presence of MTX and our designed structure (comprising two conformers: MNK and UNK) over 360 ns of MD simulation.

The root-mean-square fluctuation (RMSF) analysis, which measures the fluctuation of DHFR residues over 360 ns of MD simulation, was performed for DHFR in the above-mentioned systems. The RMSF values of DHFR in the presence of MTX and MNK were visualized in the same graph to facilitate comparison of DHFR’s behaviour in these two conditions. In the graph shown in Fig. 9a, the green series represents the RMSF values of DHFR in the presence of MTX, while the red series represents the RMSF values in the presence of MNK.

Fig. 9.

(a) RMSF plots of the protein backbone DHFR complexed with MTX (series 1, green plot) and MNK (series 2, red plot) during 360 ns MD simulation. (b) RMSF plots of the protein backbone DHFR complexed with UNK (series 3, purple plot).

As shown in Fig. 9a, the highest fluctuations were observed at residue 44 in both systems. In contrast, lower RMSF values were recorded for DHFR in the presence of MTX and MNK at specific residue positions, including 9, 53, and 100–120. The results obtained here are consistent with those in section "Molecular docking and grid box search for our proposed structures" (Fig. 6a).

Additionally, the overlap between the minima of the black and pink plots in Fig. 9a indicates that both structures, MTX and MNK, interact with similar residues of DHFR. To gain a better understanding of the dynamic behaviour of DHFR, the RMSF values of this protein were plotted in the presence of UNK (see Fig. 9b).

From the plot, it is evident that the lowest RMSF values occur within residue numbers 29–41 and 165–172. These results are consistent with Fig. 6b.

To assess how MTX and MNK affect the overall compactness of DHFR throughout the 360 ns of MD simulation, which reflects the stability of the DHFR-MTX and DHFR-MNK complexes, we plotted the protein’s radius of gyration (R_g) in the presence of these structures (Fig. 10a and b, respectively).

Fig. 10.

(a) R_g plot of DHFR in the presence of MTX. (b) R_g plot of DHFR in the presence of MNK. (c) R_g plot of DHFR in the presence of UNK.

As shown in Fig. 10a and b, the R_g values of DHFR fluctuated between 1.58 nm and 1.63 nm. Additionally, the average radius of gyration for DHFR in the presence of MTX and MNK was calculated to be 1.590 nm and 1.595 nm, respectively, indicating that the DHFR-MNK complex is as compact as the DHFR-MTX complex.

Also, to assess the compactness of the complex formed between DHFR and UNK, we plotted the Rg values of the DHFR protein in the presence of UNK during the MD simulation, as shown in Fig. 10c. The observable downward trend in the Rg values of DHFR, from 1.59 nm to 1.54 nm, indicates the potential formation of a strong complex between UNK and DHFR.

During the MD simulation, the number of hydrogen bonds between the designed ligands and MTX with DHFR was a key parameter in assessing their interactions. Figure 11a–c illustrate the number of H-bonds between the ligands—MTX, MNK, and UNK—and DHFR, respectively.

Fig. 11.

The intermolecular hydrogen-bonding distribution plot versus time for: (a) DHFR-MTX, (b) DHFR-MNK, and (c) DHFR-UNK.

As shown in Fig. 11b, MNK can form a maximum of 10 hydrogen bonds with DHFR, compared to the 8 H-bonds formed between MTX and DHFR (Fig. 11a). Additionally, UNK can form up to 8 H-bonds with DHFR. These results indicate that MNK forms a stronger complex with DHFR and may serve as a suitable alternative to MTX.

To better understand the dynamic behavior of the ligands MTX, MNK, and UNK in the presence of DHFR, we generated a graph representing the minimum distance between each ligand and DHFR. The corresponding graphs for the DHFR-MTX, DHFR-MNK, and DHFR-UNK complexes are presented in Fig. 12a–c, respectively.

Fig. 12.

Visualization of the average minimum distance between DHFR and (a) MTX, (b) MNK, and (c) UNK.

Additionally, the average minimum distances for these complexes were calculated, with values of 0.185 nm, 0.173 nm, and 0.200 nm for the MTX, MNK, and UNK complexes, respectively. These results, along with previous observations, suggest that the likelihood of forming a strong DHFR-MNK complex is greater than that of DHFR-MTX, indicating that MNK may serve as a suitable alternative to MTX.

Another parameter that can be used to evaluate the likelihood of complex formation between DHFR and the ligands, including MTX, MNK, and UNK, is the number of contacts between DHFR and these ligands. Figure 13a–c depict the number of contacts between the receptor and the ligands during 360 ns of MD simulation for MTX, MNK, and UNK, respectively.

Fig. 13.

Graph illustrating the number of contacts between DHFR and the ligands, including: (a) MTX, (b) MNK, and (c) UNK during 360 ns of MD simulation.

Figure 13a and c show that the number of contacts between DHFR and the ligands MTX and UNK fluctuates between 1500 and 2000. In contrast, the number of contacts between DHFR and MNK ranges from 1500 to 2500. The higher number of contacts between DHFR and MNK than MTX suggests that MNK may serve as a suitable alternative.

Binding free energy analysis of DHFR inhibitors, including MTX, MNK, and UNK, using the MM/PBSA method

In this section, we extracted 202 snapshots from a 360-ns MD simulation for each system: DHFR-MTX, DHFR-MNK, and DHFR-UNK. The binding free energies between DHFR and each of the ligands mentioned above were calculated using the Molecular Mechanics/Poisson–Boltzmann Surface Area (MM/PBSA) method⁶⁷. The g_mmpbsa package was used for these calculations. The energy components involved in each binding free energy calculation are summarized in Table 4.

Table 4.

Various energy components contributing to the determination of free binding energies.

MM/PBSA	DHFR and MTX	DHFR and MNK	DHFR and UNK
van der Waals energy (kJ/mol)	− 304.20	− 300.88	− 161.65
Electrostatic energy(kJ/mol)	− 83.631	− 92.28	− 57.71
Polar solvation energy(kJ/mol)	157.44	141.49	99.75
Non-polar solvation energy (kJ/mol)	− 24.83	− 25.65	− 13.88
Binding energy (kJ/mol)	− 255.21	− 277.32	− 133.48
Binding energy(kcal/mol)	− 61.00	− 66.28	− 31.90

According to the obtained results, the binding free energies for both conformers of our designed DHFR inhibitors, MNK and UNK, were − 66.28 kcal/mol and − 31.90 kcal/mol, respectively. These energies are similar to the binding free energy for the DHFR-MTX interaction (− 61.00 kcal/mol). These findings, combined with the analyses from previous sections, suggest that MNK could serve as a suitable alternative to MTX.

Additionally, to investigate whether the conformers of the designed ligand influence the secondary structure of DHFR, we plotted the changes in the secondary structure of DHFR in the presence of MNK, UNK, and MTX. The corresponding graphs are presented in Figure S10. This analysis, consistent with our other evaluations, indicates that DHFR exhibits similar dynamic behaviour in the presence of our designed inhibitors as it does in the presence of MTX.

To further illustrate the dynamic behaviour of the ligands, including MTX, MNK, and UNK, during MD simulations, we selected two snapshots from their respective simulations. From these snapshots, we generated 2D images representing the interactions of the ligands with DHFR residues. These images are shown in Figures S11–S13.

Moreover, we thoroughly analyzed the MD simulation trajectories to identify key amino acid residues that interact with the ligands—MTX, MNK, and UNK—throughout the simulation period. Using trajectory analysis tools, we monitored hydrogen bonds, hydrophobic contacts, and per-residue energy contributions across the entire MD simulation.

For each ligand, we computed the interaction frequency of residues (i.e., the percentage of frames in which a residue-ligand contact occurred). We complemented this with MM-PBSA per-residue energy decomposition.

For MTX, the most frequent and energetically significant interactions involved Leu22, Phe34, Ile60, Asn64, Leu67, Val115, and Gly116. These residues consistently contributed to stable binding, confirming that MTX remained in the active site throughout the simulation (see Figure S14). This is consistent with the structural data presented in Fig. 4a of this manuscript.

For MNK, a similar interaction pattern was observed. Residues Ile7, Val8, Ile16, Asn19, Gly20, Leu22, Pro23, Trp24, Phe34, Tyr56, Ser59, Ile60, Val115, and Gly116 maintained frequent contact with MNK during the simulation, suggesting strong and persistent binding similar to MTX (see Figure S15).

For UNK, which represents a conformer of MNK interacting with an alternative site on DHFR, the dominant interacting residues were Arg28, Asn29, Arg31, Tyr33, Thr39, Thr40, Gln44, Lys63, Gln161, Leu166, and Phe179. These residues showed frequent interactions throughout the trajectory (see Figure S16), indicating a distinct but stable binding mode for this conformer.

These analyses confirm that the ligands remain stably bound throughout the simulation and that specific residues play dominant roles in mediating these interactions.

Theoretical toxicity assessment of selected designed structures and MTX

Another important aspect of our study was predicting the toxicity of our designed DHFR inhibitors and comparing their toxicity with that of the known DHFR inhibitor, methotrexate (MTX).

First, it is worth noting that numerous toxicity assessment methods have been developed, including in silico methods. In silico toxicity prediction methods were developed to complement other approaches, such as in vitro and in vivo toxicity tests. Computational methods are a preliminary step in toxicity assessment, as they are cost-effective, time-saving, and provide valuable insights when handling large datasets.

Our study employed the open-source software T.E.S.T. (Toxicity Estimation Software Tool, URL: https://www.epa.gov/comptox-tools/toxicity-estimation-software-tool-test)⁷⁷. We based our toxicity calculations on the oral rat LD₅₀, which is defined as the amount of a substance (in milligrams per kilogram of body weight, mg/kg) that, when administered orally to rats, causes the death of 50% of the test population⁶⁸.

We used the Nearest Neighbor method to predict the oral rat LD₅₀ for our designed inhibitors. The central principle of this method is that compounds with similar structures are likely to exhibit similar properties.

Table 5 presents the selected structures along with their predicted oral rat LD₅₀ values. The results for each structure are based on the dataset and figures provided in the supplementary information.

Table 5.

Comparison of the toxicity of MTX and designed DHFR inhibitors Using T.E.S.T. (Toxicity Estimation Software Tool).

(a).⁷⁹.

According to Table 5, the designed structure encoded as MNK has the lowest predicted toxicity among the tested structures, including Trp-His-Tyr (gluc) and Asp-His-Tyr (gluc), with a calculated LD₅₀ of 12,135.03 mg/kg. Additionally, our results indicate that all three selected structures have higher LD₅₀ values (indicating lower toxicity) compared to the experimentally reported LD₅₀ of MTX, which is 135 mg/kg.

Validation of our theoretical study alongside the prediction of IC₅₀ values for selected designed DHFR inhibitors

To validate our theoretical studies and predict the IC₅₀ values of our selected designed structures, including Trp-His-Tyr (gluc), MNK, and Asp-His-Tyr (gluc), we randomly collected 21 known DHFR inhibitors⁶⁹, for which average IC₅₀ values have been reported. The optimization of these structures was performed using the MMFF force field (molecular mechanics) in Spartan’14 software. Additionally, we conducted a conformer search for the selected compounds.

We then calculated the area and volume of these molecules (the data related to these known structures can be found in Table S4). By plotting the inverse log IC₅₀ values of these DHFR inhibitors against their calculated area parameters, we obtained a linear graph (Fig. 14) described by the equation and . This linear relationship was subsequently used to predict the IC₅₀ values for our selected designed DHFR inhibitors.

Fig. 14.

Experimental values of 21 collected known DHFR inhibitors versus their calculated area values ().

Additionally, we observed a linear relationship between the calculated volumes of the collected drugs and their 1/log IC₅₀ values (experimentally reported IC₅₀ values of these drugs). This relationship is described by the equation y = 0.002x − 0.33, with an R2 value of 0.73. The corresponding graph is shown in Figure S20.

Finally, we compared the predicted IC₅₀ values of our designed structures for DHFR inhibition with the mean IC₅₀ value of MTX, as reported by Heppler and coworkers using a DHFR enzymatic assay (0.12 ± 0.07 μM)⁷⁰.

In this section, we aimed to illustrate the potential relationship between the average binding energy values of known DHFR inhibitors and their reported IC₅₀ values. To this end, we selected 16 known DHFR inhibitors with experimentally measured IC₅₀ values. Binding energies of these inhibitors toward DHFR were calculated using AutoDock Vina. Subsequently, we plotted the experimental log(IC₅₀) values against the corresponding calculated binding energies⁷¹. The resulting linear regression yielded the equation y = 0.756x + 10.19, with a coefficient of determination (R²) of 0.72 (see Figure S21 and Table S5). These results indicate a significant correlation between the predicted binding affinities and the experimentally reported IC₅₀ values of DHFR inhibitors.

As shown in Table 6, the predicted IC₅₀ value for MTX aligns more closely with the experimental value when derived from Figure S17, suggesting better validation. Moreover, it should be noted that all designed compounds exhibit either lower IC₅₀ values than MTX or values comparable to those of MTX (according to all three methods), indicating their potential effectiveness at lower doses.

Table 6.

Predicted IC₅₀ Values of Selected Designed DHFR Inhibitors According to Fig. 14 and Figure S20, and Their Comparison with the Experimentally Reported Mean IC₅₀ Value of MTX.

Comparative ADMET analysis of MNK and MTX: pharmacokinetic properties and toxicity profiles

To support our toxicity assessments of MNK and MTX and to gain additional insight into their potential pharmacokinetic properties, we utilized the ADMETlab 3.0 online server⁷². The results of this analysis are presented in Table 7.

Table 7.

Predicted ADMET properties of MNK and MTX obtained using ADMETlab 3.0

Property	MNK	MTX
CYP450 interaction	No inhibition or substrate	No inhibition or substrate
HLM metabolic stability	Stable	Stable
hERG inhibition risk	Very low (0.004, 0.001)	Low (0.053, 0.04)
Plasma protein binding (PPB)	~ 43%	~ 45%
Fraction unbound (Fu)	53%	46%
P-gp inhibition	No	No
BCRP / MRP1 inhibition	No inhibition	MRP1 and OATP1B3 inhibited
Permeability (PAMPA)	High (1.0)	High (0.998)
Oral bioavailability (F20/30/50)	< 50%	> 50%
BBB penetration	No	No
Promiscuity (off-target binding)	0.359	0.628
Cytotoxicity (A549, HEK293, etc.)	Very low	Moderate
Neurotoxicity	0.069	0.61

As shown in Table 2, both compounds are predicted not to inhibit cytochrome P450 (CYP) enzymes, suggesting a low risk of metabolism-related side effects or drug-drug interactions⁷³. Additionally, MNK and MTX are predicted to be stable in human liver microsomes (HLMs), indicating that both compounds may exhibit favourable bioavailability and sufficient half-lives in the human body⁸⁵.

Besides, the research indicates that MNK has a lower protein binding affinity than MTX and a more significant proportion of the free form. Such a property would mean that MNK will exhibit more excellent pharmacological activity since a more significant percentage of the drug is in the active (free) form.

Finally, MNK is predicted to exhibit lower cytotoxicity, neurotoxicity, and cardiotoxicity than MTX, further supporting its potential as a safer and more effective therapeutic candidate.

According to Table 7, MTX is predicted to inhibit the MRP1 and OATP1B3 proteins, and their inhibitions are associated with nephron- and hepatotoxicity, respectively.

Both structures are predicted to diffuse passively along the cell membranes, which can be categorised as a favourable property. None of the structures is predicted to pass through the BBB, which shows that none of the structures may cause CNS toxicity.

For pharmacodynamic properties, one of the important parameters in Table 7 is promiscuity or off-target binding, which is 0.359 for MNK and 0.628 for MTX. This parameter suggests that MNK may be more selective for its intended target compared to MTX.

Conclusion

A distinguishing feature of cancer cells compared to normal cells is their altered metabolism. Cancer cells derive energy through a distinct process, unlike normal cells, often engaging in anaerobic glycolysis instead of relying on mitochondrial oxidative phosphorylation, even when oxygen is abundant and mitochondria are functional. This metabolic shift, known as the Warburg Effect^62,63,69, increases the expression of glucose transporters such as GLUT1^60,61, facilitating glucose uptake, the primary energy source for cancer cells. Additionally, the overexpression of dihydrofolate reductase (DHFR) is a notable trait in various cancers⁷⁴, including ovarian cancer. DHFR is essential for converting dihydrofolate into tetrahydrofolate (THF), a critical intermediate in pyrimidine biosynthesis⁸⁷.

With this in mind, our objective was to develop novel DHFR inhibitors derived from natural compounds, such as glucose and amino acids, to enhance uptake by cancer cells while suppressing DHFR’s role in DNA synthesis and cell division⁸⁷. Methotrexate (MTX), a well-established DHFR inhibitor, is widely used to treat cancers such as meningeal leukemia¹³ and lymphoma¹⁴. However, challenges such as drug resistance⁷⁵, side effects^24–28, and limited transport efficacy^76,77,89 reduce its therapeutic potential.

In our study, we initially designed 20 potential DHFR inhibitors. Many were excluded due to their large molecular volumes, unsuitable polar surface area (PSA) values, and weak binding affinities. Ultimately, three promising candidates were selected: Trp-His-Tyr (gluc), Trp-Tyr (gluc)-GLU, and Asp-His-Tyr (gluc). Using AutoDock 4.2.2 software⁴⁶, we performed a grid box search to identify the optimal binding poses of these structures, along with MTX, with DHFR. Among the tested candidates, the Trp-Tyr (gluc)-GLU conformer demonstrated the most negative binding energy and exhibited a binding orientation most similar to MTX. Following this analysis, two Trp-Tyr (gluc)-GLU conformers, labelled MNK and UNK, were selected for molecular dynamics (MD) simulations.

A quantum mechanics (QM)-based evaluation of the logP values for MTX and the three selected structures revealed that Trp-Tyr (gluc)-GLU had a logP value closest to that of MTX (refer to Table 3). Moreover, MD simulations showed that both MNK and UNK conformers of Trp-Tyr (gluc)-GLU had more negative free binding energies with DHFR than MTX, suggesting their potential as viable alternatives to MTX.

To further validate these findings, we assessed the toxicity profiles of the selected compounds using the open-source software T.E.S.T.⁷⁷. Consistent with other results, the toxicity analysis identified MNK as having the lowest toxicity among the evaluated compounds, which was also lower than that of MTX. Additionally, IC₅₀ predictions and validation studies confirmed that all designed compounds, including Trp-His-Tyr (gluc), Trp-Tyr (gluc)-GLU, and Asp-His-Tyr (gluc), exhibited lower IC₅₀ values than MTX.

In conclusion, MNK, with its superior interaction with DHFR, lower toxicity, and fewer side effects compared to MTX, holds promise as a natural-product-inspired alternative to MTX in cancer therapy.

Author contributions

S.K. and A. S. obtined the the results. S. K. prepared the manuscript, and A. F. conducted this study and revised its manuscript.

Data availability

All data generated or analysed during this study are included in this published article [and its supplementary information files].

Competing interests

The authors declare no competing interests.