Structure-based discovery of a new type of scaffold compound as binding competitors for protein-bound Uremic Toxins

Abstract

Protein-bound uremic toxins (PBUTs) are the main cause of uremia, but traditional hemodialysis is ineffective in removing them because of their strong ability to bind to human serum albumin (HSA), highlighting the need for new treatments. In this study, first, structure-based docking was used to screen a diverse library of 200,376 virtual compounds against the active sites I and II. After two rounds of docking screening, 3944 candidate molecules were obtained. Second, 23 candidate molecules were obtained after ADMET prediction and toxicity analysis. Five candidate molecules were finally obtained after visual analysis and MM-PBSA calculations. We subsequently assessed their competitive displacement efficiency through a microdialysis experiment, and the results revealed that ZINC000008791789, ZINC000012297018, and ZINC000012296493 are promising binding competitors for PBUTs, as they have higher dialysis efficiency than the optimal displacer LA, approximately double the dialysis efficiency. The other two molecules, ZINC000031161007 and ZINC000004090361, although less efficient than LA, still outperformed the control group. Notably, four of them shared the same molecular scaffold, and three of them contained a flavonoid group. These findings provide a foundation for the development of more effective PBUT binding competitors, potentially benefiting uremia patients in the future.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-78766-9.

Introduction

Uremia is a complex clinical syndrome that occurs in patients with advanced kidney disease and is accompanied by systemic functional and metabolic disorders¹. Protein-bound uremic toxins (PBUTs), comprising cationic and anionic urotoxins, are the primary causes of uremia. Notably, anionic urotoxins such as indoxyl sulfate (IS), p-cresol sulfate (pCS), 3-carboxyl-4-methyl-5-propyl-2-furanopropionic acid (CMPF), and indoleacetic acid (IAA) are pivotal in chronic renal failure². Human serum albumin (HSA) contains two primary drug-binding sites, sites I and II, IS, pCS, and IAA bind to site II, whereas CMPF binds to site I, collectively binding 90% of albumin³. IS and pCS are the two most harmful toxins in the later stage of uremia^4,5. Despite efforts to enhance dialysis materials and their frequency, conventional hemodialysis methods struggle to eliminate these toxins effectively^6,7.

At present, one of the most effective methods for removing PBUTs is the replacement method, which involves injecting drugs such as ibuprofen, salvianolic acid, and fatty acids (FFAs) into the bloodstream^8–10. These drugs bind to site I or site II of human serum albumin (Fig. 1A), increasing the release of PBUTs and facilitating their removal. For example, in vitro studies have shown that lithospermic acid (LA) at 400 µM can increase the dialysis efficiency for IS and pCS by approximately 197.23% and 198.31%, respectively, making it the most potent competitive replacement agent known⁹. Prior to their application as competitive replacement agents for PBUTs, these drugs were identified as robust protein-binding ligands^11–14, Furthermore, molecular docking and probe displacement assays demonstrated that LA can bind to sites I and II on HSA¹³. However, owing to the high costs associated with drug development, few advancements have been made in discovering novel PBUT displacement agents. To reduce costs and increase screening efficiency, we used virtual screening alongside microdialysis experiments to find better PBUT competitive displacement agents.

Fig. 1.

Interaction analysis of lithospermic acid (LA) and major PBUTs with HSA and redocking validation. (A) Binding site of ligand on HSA (PDB ID: 2bxh). (B-D) Key amino acid residues involved in the binding of IS, pCS and IAA to the site II of HSA, respectively. Ligands are represented by magenta-colored sticks, limegreen-colored sticks, and red-colored sticks, respectively. (E) Key amino acid residues involved in the binding of CMPF to the site I of HSA. Ligand is represented by lightpink-colored sticks. (F) IS redocking validation. Orange represents the top-ranked pose of IS, while cyan represents the experimental position of IS. (G) Structure of LA. (H, I) Interactions of HSA and LA at sites II and I, respectively. Ligand LA is represented by orange sticks. All hydrogen bonds are represented by green dotted lines, and salt bridge interactions are represented by yellow dotted lines.

In the replacement method, drugs compete for a single albumin binding site. To increase the removal of PBUTs, additional binding competitors are necessary, although this can heighten drug side effects. Therefore, screening for drugs effective at both sites I and II is vital. Site II of HSA is the primary binding site for IS and pCS, followed by site I (Fig. 1A-E). Key amino acids such as R410, Y411, and L430 stabilize the IS, and pCS binds at site II (Fig. 1B-D)^15–17. Additionally, K414, S489, and N391 contribute to the binding energy (Fig. 1B-D). R410 primarily provides binding energy through electrostatic interactions, while the polar amino acid Y411 contributes energy through hydrogen bonding with ligands via its phenolic hydroxyl group, and L430 in the hydrophobic core forms hydrogen bonds with ligands^15–17. This information serves as a reference for the virtual screening of docking sites.

Therefore, in this study, we initially performed two rounds of molecular docking to pinpoint structures docking to sites II and I on HSA with binding free energies below − 9.0 kcal/mol from a pool of 200,376 compounds sourced from the ZINC and PubChem databases, yielding 3,944 structures. Subsequent ADMET predictions led to the identification of 23 candidate molecules (Supplementary Fig. S1) with minimal risk to vital organs such as the heart and liver (Supplementary Table S1). Further examination of interactions and empirical screening resulted in the selection of 8 candidates (Supplementary Table S2). These molecules underwent MD simulations and MM-PBSA binding energy analyses, producing 5 candidates with stable HSA binding and low energies, confirmed through microdialysis experiments, resulting in the identification of three efficient lead compounds, ZINC000008791789 (ligand 1), ZINC000012297018 (ligand 3), and ZINC000012296493 (ligand 5) (Fig. 2A-E). These compounds exhibited a greater competitive displacement efficiency for PBUTs (IS and PCS, two major types of PBUTs) than the existing optimal displacement molecule, lithospermic acid (LA), which is approximately double the dialysis efficiency. The other two molecules, ZINC000004090361 (ligand 6) and ZINC000031161007 (ligand 7) (Fig. 2A-B, F-G), although less efficient than LA, still outperformed the control group.

Fig. 2.

Results of in vitro microdialysis experiment at 200 µM concentration and scaffold analysis and optimization of hit molecules. (A) The in vitro microdialysis efficacy of IS. (B) The in vitro microdialysis efficacy of pCS. **P < 0.01. (C-G) The structure of ligands 1, 3, and 5–7. (H) New scaffold structural formula. (I) The molecular structure of the modified molecule was preliminarily optimized.

Further analysis revealed that four of them shared the same molecular scaffold (Fig. 2C-F, H). Owing to their strong binding affinity at both the II and I sites, they may also be effective in clearing PBUTs from site I. Therefore, our findings provide a theoretical foundation and reference for the future development of more efficient PBUT competitive displacers, aiming to alleviate various diseases caused by PBUTs and prolong the life of patients.

Results

Redocking validation

The search space consisted of two principal binding pockets of HSA, known as sites II and I. Figure 1A illustrates the complete structure of HSA (PDB ID: 2bxh), with the ligand (indoxyl sulfate, IS) and waters removed for clarity. It also indicates the positions of the two binding pockets. We chose to dock with the protein in the ligand-bound state for several reasons. Firstly, before starting our work, we studied the structural characteristics of sites I and II of HSA, comparing the crystal structures of the unliganded form and several liganded forms. We found that in the unliganded structure, the hydrophobic cavity of site II narrows due to changes in the positions of amino acids such as R410 and K414, leading to obstruction at the cavity opening. Furthermore, by docking known ligands like IS and ibuprofen, we observed that they could not successfully dock into the binding pocket of site II but instead clustered near the opening. Therefore, we speculate that the conformational state of site II in the unliganded structure of HSA is not suitable for our docking screening. The second reason is that the ligands we screened are primarily intended to competitively displace PBUTs (such as IS, PCS, and IAA) bound to albumin. Therefore, as long as a ligand can bind to albumin in the ligand-bound state, it can meet the conditions necessary for its functionality. To validate the docking method, redocking was performed using Autodock Vina 1.1.2 (https://vina.scripps.edu/downloads/). The best-ranked pose achieved a root-mean-square deviation (RMSD) of 0.29 Å compared to the ligand IS in its experimental position within the PDB structure 2BXH of HSA (Fig. 1F). The RMSD value of less than 2.0 Å confirms the reliability of the docking method, as recommended in the literature¹⁸.

Verification of the interaction between LA and HSA, and the selection of ligands

Because analogs of compounds may have better effects than the compounds themselves¹⁹, we utilized LA, a known superior PBUT competitive displacement agent (Fig. 1G), as a reference molecule to screen for analogs in order to enhance the competitive displacement efficiency of PBUTs and analysed its interaction with HSA. The binding free energy of LA at sites I and II is -10.4 kcal/mol and − 9.5 kcal/mol, respectively. At site II, the carboxyl group of LA formed salt bridges with R410 and K414 (Fig. 1H). Additionally, a hydrogen bond was formed between the carboxyl oxygen of LA and the amide nitrogen of N391. However, it is important to note that these interactions may be influenced by the solvent, which could lead to a higher desolvation potential energy and consequently a higher actual binding energy. Furthermore, the phenolic hydroxyl oxygen atom of LA formed hydrogen bonds with the amide oxygen of L430 and the amide nitrogen of G434 at the bottom of the cavity, contributing to the stability of the complex. Site I, on the other hand, is a more extensive hydrophobic cavity. The carboxyl group of LA formed salt bridges with R222 and K199 at the opening of the cavity. Additionally, the phenolic hydroxyl group and the carboxyl group maintained a hydrogen bond and a salt bridge, respectively, with the buried alkaline residue R257 (Fig. 1I). The aromatic ring and alkyl carbon of LA interacted with multiple hydrophobic groups within the cavity. These findings provide evidence that LA can effectively dock with sites II and I, thereby enhancing the reliability of the docking system and highlighting the potential of LA as a reference molecule.

To identify effective alternatives to IS and pCS, we collected more than 0.2 million molecules from the PubChem database and the ZINC database, respectively, as detailed in Table 1. The molecules in the “PubChem-similarity to LA” subset with at least 85% similarity to LA and may exhibit biological activity similar to LA. The molecules in the “ZINC-world” subset consist of approved clinical drugs with well-established pharmacokinetic and toxicity empirical knowledge, making them suitable for drug repurposing. Furthermore, molecules in the “ZINC-natural products” subset possess diverse skeletal structures, which can be valuable for the discovery of novel molecular scaffolds.

Table 1.

Virtual screening of the number of molecules and the conditions calculated, at each step.

Compounds	Round 1 of docking	Round 2 of docking		Toxin analysis		Binding pose analysis	Stability analysis
Ligands subsets	Site II exhaustiveness = 4	Site I exhaustiveness = 20; Site II exhaustiveness = 20	Site I exhaustiveness = 4; Site II exhaustiveness = 20	ADMET prediction	Searched toxicity in ZINC-world	PLIP	MD/MM-PBSA
PubChem- similarity to LA (4178)	184	109	-	6	-	1	-
ZINC-world (5903)	121	25	-	0	1	-	-
ZINC-natural products (190295)	11,309	-	3810	17	-	7	5

Virtual screening based on molecular docking and ADMET prediction

More than 0.2 million structures mentioned earlier were successfully docked using AutoDock Vina 1.1.2 (https://vina.scripps.edu/downloads/). The top-ranking compound structures with a binding free energy ≤ -9 kcal/mol, made up approximately 5.3% of the initial ligand library, are selected for the second round of docking. During this round, 11,309 structures from “zinc-natural products” were docked to site I with an exhaustiveness value of 4, whereas all others were docked to sites II and I with an exhaustiveness value of 20. This screening process yielded a total of 3,944 top-ranking structures (Table 1). Among them, 3,919 structures performed well at both sites (binding energy ≤ -9.0 kcal/mol at least) and were subjected to ADMET prediction. The remaining 25 top-ranking structures were sourced from ZINC-world, and toxicity or side effects data were obtained from the DrugBank database. Eventually, a total of 23 low-toxicity molecules (Supplementary Table S1) with unique structures were identified from “ZINC-natural” and “PubChem-similarity to LA”. All of these candidates were visually inspected for their complementarity to the two binding sites of HSA. Each candidate was docked to HSA with an exhaustiveness value of 20 to search for multiple conformations.

Analysis of visual inspections

In site II, all 23 candidates formed hydrogen bonds or salt bridges with at least one of the residues of R410 and Y411. Furthermore, all of the compounds from the “ZINC-natural” subset formed hydrogen bonds or hydrophobic interactions with L430. They also formed hydrogen bonds or salt bridge interactions with other key amino acid residues, such as K414, S489, and N391. These residues have been identified as key sites for several toxins, including IS, pCS, and IAA (Fig. 1B-D)⁶. Therefore, all 23 candidate molecules meet the criteria by only considering whether they can interact with the key amino acids at site II.

Subsequently, visualize whether the 23 candidate molecules mentioned earlier form hydrogen bonds or salt bridges with the amino acids Y150, R222, H242, R257, and K199 at site I, and whether the structures of their non-polar parts can be well enveloped by the hydrophobic amino acids inside sites II and I pockets, and form hydrophobic interactions with adjacent residues, etc., to prioritize meeting the screening criteria of site II. Secondly, molecules that meet the site I conditions to the greatest extent are selected as candidate molecules, because PBUTs mainly bind to site II, while site I mainly binds to CMPF. Molecules that satisfy these conditions may effectively clear PBUTs bound to both sites I and II, so considering the above conditions, 8 candidate molecules were ultimately selected (Fig. 3; Table 1). The binding free energy and main amino acid binding sites of these molecules in molecular docking are shown in Supplementary Table S2, and they all possess a short fatty acid chain and 4–8 rotatable bonds (meet Lipinski’s Rule of Five), allowing them to fit better into the two binding pockets and form more hydrophobic interactions within the cavity (Fig. 3). All of these candidates generated hydrogen bonds or salt bridges with at least three residues, namely, Y150, R222, H242, R257, and K199. These residues are the main amino acid residues responsible for the binding of 3-carboxy-4-methyl-5-propyl-2-furanpropanoic acid (CMPF), a PBUT that primarily binds to site I of HSA (Fig. 1E).

Fig. 3.

The interactions of the 8 candidates with site II of HSA were obtained through visual inspection. The interactions involving hydrogen bonds and salt bridges are depicted as green dashed lines and yellow dashed lines, respectively. The amino acid residues forming hydrogen bonds and salt bridges are represented as sticks, whereas hydrophobic interaction residues are illustrated as lines.

Analysis of the binding stability of the complexes based on MD simulation

To further analyse the stability of the complexes, MD simulations were conducted to calculate the energy, RMSD, and hydrogen bond interactions.

First, the total energies of the 8 complexes bound to site II were calculated and are shown in Fig. 4A. The results of low mean value and no abnormal fluctuations indicate that the complex simulation system is in a stable state. However, the energy changes during the protein-ligand binding process are relatively small compared to the total energy changes in the simulation system. These small changes in the protein-ligand binding state cannot be observed through the total energy. Therefore, we further evaluated and analysed the time-dependent changes in the RMSD values of the protein and ligand. Since the first 150 ns of the MD simulation were considered sufficient for stable conjugation between the ligands and site II of the protein, the mean and standard deviation of the RMSD values between 150 ns and 200 ns of the simulation were analysed and are shown as a bar chart in Fig. 4B. These values are expected to be small. The time-varying RMSD curves of the entire 200 ns are shown in Supplementary Fig. S2. These results indicate that most complexes exhibit good stabilization within site II of HSA. Furthermore, the radius of gyration (Rg) of the protein was analysed over time and is shown in Fig. 4C. Lower values indicate that the protein maintains a more compact and stable structure in all complexes.

Fig. 4.

Results of the MD simulations. (A) Mean values of energy for the complexes during the 200 ns MD simulations. (B) Mean and standard deviation values of the RMSD for the heavy atoms of the protein backbone and ligands between 150 ns and 200 ns of the MD simulation. (C) Protein Rg analysis during the 200 ns MD simulations after each ligand bound to site II. The mean and standard deviation values of HSA are: Apo (2.61 nm ± 0.019), LA (2.59 nm ± 0.011), Ligand 1 (2.61 nm ± 0.015), Ligand 2 (2.59 nm ± 0.016), Ligand 3 (2.58 nm ± 0.014), Ligand 4 (2.61 nm ± 0.012), Ligand 5 (2.58 nm ± 0.023), Ligand 6 (2.57 nm ± 0.019), Ligand 7 (2.60 nm ± 0.018), and Ligand 8 (2.58 nm ± 0.013). (D-L) Hydrogen bonds between each ligand and HSA residues over time. The interactions with residues of site I are shown in orange, with residues of site II shown in green, and with the principal residues of site II (R410, Y411, L430, K414, S489, N391) in cerulean.

The number of hydrogen bonds between each ligand and HSA was also analysed and is shown in Fig. 4D-L. The principal residues refer to the main residues involved in the binding of IS and pCS to site II (R410, Y411, L430, K414, S489, N391). As shown, ligands 1, 3, 5, 6, and 7 are the most promising, as they strongly interact with the principal residues of site II during most of the simulation time, indicating good binding to both the active and secondary sites of HSA. Additionally, ligands 1, 5, 6, and 7 also maintain a greater number of hydrogen bonds with the residues of site I.

In addition, the information of all hydrogen bonds formed between protein amino acid residues and ligand atoms during the simulation process was analysed, and the occupancy rates of these hydrogen bonds was calculated (Supplementary Fig. S3). By analysing the occupancy rates within the ranges of 25–49%, 50–69%, and greater than 70% of hydrogen bonds, it was found that the hydrogen bonds formed by these ligands with the residues R410, Y411, S489, N391, and L430 of site II all had an occupancy rate exceeding 50% during the simulation process. Ligand 1 formed hydrogen bonds with Y411, S489, N391, and L430, each with an occupancy rate exceeding 70%. Ligands 3, 4, 5, and 8 also formed hydrogen bonds with certain residues of R410, Y411, and N391, each with an occupancy rate exceeding 70%. This indicates that R410, Y411, S489, N391, and L430 play important roles in maintaining the stable binding of these 8 ligands with site II.

Furthermore, ligands 1, 2, 4, 6, 7, and 8 formed hydrogen bonds with some residues of Y150, K199, R257, H242, and R222 of site I, ligands 7 and 8 formed hydrogen bonds with Q196 and R218, each with an occupancy rate exceeding 70%. This suggests that Y150, K199, R257, H242, R222, Q196, and R218 residues also play important roles in the stable binding of these 8 ligands with site I.

The root mean square fluctuation (RMSF) was used to identify any changes or flexibility in the protein residues within the complexes. After binding to site II, all of the ligands significantly reduce the fluctuation of residues 410–440, suggesting that they can stably bind to this region and limit structural changes in this area. Additionally, the binding of ligand 1 resulted in an increase in the local RMSF values of the IIA region (residues 250–300), which is a section of the structure that makes up site I. This finding suggests that ligand 1 may affect the binding of toxins to site I by altering the structure of this region (Supplementary Fig. S4). Furthermore, the binding of LA, ligand 1, ligand 3, and ligand 8 to site I significantly induced local structural changes in the IIB and IIIA subdomains (Supplementary Fig. S6).

Moreover, the same analysis methods were applied to the MD simulation trajectory of the 8 candidates binding with site I of HSA. The RMSD and Rg curves are shown in Supplementary Fig. S5 and Supplementary Fig. S7A, respectively. These results indicate that all of the complexes in the studied systems remained stable during the MD simulation.

MM-PBSA binding free energy

From the 300 frames of MD trajectories, all complexes, including the control (LA), had similar and stable net binding energy values, which indicates the stability of LA and the candidates at the binding site. In the complex production simulation for site II, the binding energy of all of the candidates was lower than that of IS and the reference molecule LA, especially ligand 1, which was better than IS and LA (Fig. 5). Ligand 2 and ligand 8 are also slightly lower than IS and LA. Interestingly, for site I, ligand 7 is much better than both IS and LA (Supplementary Fig. S7B). This suggests that ligand 7 may have better competitive binding ability at both sites. The results of MM-PBSA once again suggests that these molecules are promising binding competitors for PBUTs.

Fig. 5.

Binding free energy results for the control and candidates at site II. The mean and standard deviation values were calculated by the MM-PBSA method for 300 frames of MD trajectories.

The five candidates improved the in vitro microdialysis efficacy of IS and pCS

We conducted microdialysis experiments on five candidate molecules. The dialysis efficiency of three candidates, namely, ligands 1, 3, and 5 for IS and PCS was greater than that of the reference molecule LA (Figs. 2 and 6). At a concentration of 400 µM (the optimal concentration of LA), these three molecules showed greater competitive binding ability than LA. Compared with that of the blank group, the dialysis efficiency of IS was improved by 89.3–114% (Table 2), whereas that of LA was improved by 65.9% (Table 2). The other two compounds, ligands 6 and 7, also exhibited some removal efficiency for IS and PCS, but their removal efficiency was lower than that of LA (Fig. 6). Moreover, at a concentration of 200 µM, the three candidate molecules achieved the same removal efficiency of IS and PCS as did 400 µM LA dialysis.

Fig. 6.

Results of in vitro microdialysis experiment at 400 µM. (A, C) The in vitro microdialysis efficacy of IS. (B, D) The in vitro microdialysis efficacy of pCS. **P < 0.01.

Table 2.

Improvement rates of the in vitro microdialysis efficacy of IS and pCS by 5 competitors.

Competitors	Improvement rates of dialysis efficacy of IS		Improvement rates of dialysis efficacy of pCS
	Conc. 200 µM	Conc. 400 µM	Conc. 200 µM	Conc. 400 µM
LA	20.87%	65.88%	19.89%	73.32%
Ligand 1	111.17%	91%	121.81%	96.44%
Ligand 3	112.30%	114.0%	121.87%	122.78%
Ligand 5	114.78%	89.32%	117.67%	85.81%
Ligand 6	15.80%	38.56%	20.61%	47.41%
Ligand 7	51.87%	40.41%	62.99%	64.82%

We further analysed and found that ligands 1, 3, 5, and 6 have different molecular scaffolds from LA. All of them share a common scaffold, as shown in Fig. 2H, indicating that compounds containing this novel scaffold may play a role in the clearance of PBUTs. Moreover, ligands 3, 5 and 7 also have a flavonoid group at the R2 substituent (Fig. 2D-E, G, H), which may also have anti-inflammatory and antioxidant effects in addition to the clearance of PBUTs. They will serve as effective lead compounds for PBUTs competitive substitutes. The discovery of this new scaffold also provides support for the development of more efficient PBUT competitive substitutes in the future.

Discussion

It is challenging to remove PBUTs through traditional hemodialysis because they strongly bind to HSA. The native kidney achieves high clearance of protein-bound uremic toxins mainly through tubular secretion, a function that cannot be replicated by conventional dialysis²⁰. Therefore, to increase the hemodialysis efficiency of PBUTs, one effective method is to increase their free state by using competing displacing agents. However, in the past, binding competitors have mostly used ibuprofen⁸, salvianolic acids⁹, and FFA¹⁰, the removal efficiency of dialysis also needs to be improved.

In this study, we used the albumin site II as the main target and site I as the secondary target. We conducted a virtual screening of ZINC, and the PubChem library for the first time and identified five compounds that can stably bind to HSA. Furthermore, we proved that ZINC000008791789 (ligand 1), ZINC000012297018 (ligand 3), and ZINC000012296493 (ligand 5) could significantly improve the dialysis clearance rate of IS and pCS in in vitro microdialysis experiments. The dialysis efficiency of these three molecules at a concentration of 200 µM was higher than that of LA with optimal efficacy (400 µM), indicating their high efficiency. However, further verification through in vivo experiments and clinical trials is necessary to confirm their effectiveness. This finding also preliminarily confirms the feasibility of virtual screening with albumin as the target and provides a new approach and idea for searching for new efficient binding competitors in the future. Additionally, since the binding of site I was also considered during screening, these three compounds may have a high competitive displacement efficiency for CMPF, which is bound to site I. This hypothesis requires further verification through in vitro microdialysis experiments.

In addition, an analysis of the structural formula, revealed that ligands 1, 3, 5, and 6 have the same scaffold (Fig. 2C-F, H). We have filed a patent application proposing that small molecules modified with this new scaffold could increase the clearance of PBUTs. Moreover, when comparing ligands 1 and 6, the dialysis efficiency of 1 is better than that of 6, indicating that the effectiveness is enhanced when the R1 substituent is the indole group. When comparing ligand 7 with ligands 1, 3, 5, and 6, it is evident that ligand 7 also demonstrates some competitive displacement dialysis efficiency, although it is weak. By examining the structural formula, we can observe that ligand 7 does not possess the new scaffold but does contains a flavonoid group, indicating that flavonoids also exhibit a certain degree of clearance efficiency for PBUTs. Ligands 3 and 5 possess both the new scaffold and flavonoid groups, resulting in the highest dialysis efficiency. Ligand 1 was more efficient, possibly because of the combination of the new scaffold and the R1 group being an indole group. Therefore, we speculate that molecules containing both the new scaffold, an indole group at R1 and a flavonoid group at R2, the structure depicted in Fig. 2I may exhibit even higher clearance efficiency for PBUTs. Consequently, we identified a new class of scaffold compounds that may have the potential to be highly effective in clearing PBUTs, providing a reference for the development of new competitive displacers.

Furthermore, ligands 3 and 5, which we screened, belong to flavonoid derivatives. In recent years, flavonoids have played an increasingly important role in anti-cancer research²¹. And flavonoids are also considered to have beneficial effects on human health when consumed as part of the diet²². These health-promoting properties are associated with their antioxidant²³, anti-inflammatory²⁴, and anticancer properties²⁵. Therefore, these two flavonoid derivatives could be beneficial in alleviating renal fibrosis and inflammation, as they possess antioxidant, anti-inflammatory, and other activities. These compounds serve as effective lead compounds for developing PBUT binding competitors, and their structures could be further optimized or their analogues screened for improved dialysis efficiency in the future.

Previous studies have demonstrated that PβCD, when used as an adsorbent, effectively enhances the clearance rate of pCS and lowers the concentration of free PBUTs through adsorption²⁶. This facilitates the continuous dissociation of undissociated PBUTs from HSA. If PβCD is utilized in conjunction with a competing displacing agent, it has the potential to maximize the dialysis clearance rate of PBUTs while reducing the necessary dosage of competitive replacement agents. This subsequently lessens the impact of introducing competitive replacement agents, as most drugs require liver metabolism and subsequent kidney excretion, which can be burdensome for ESKD patients. Consequently, combining our candidate molecules with the adsorbent PβCD might prove to be an effective strategy for the removal of PBUTs via dialysis. Nonetheless, further experimental verification is required to confirm this effect.

Conclusion

In summary, we utilized a structure-based virtual screening method to identify a class of compounds with a new scaffold (Fig. 2H) that can increase the removal efficiency of PBUTs. Among them, ZINC000008791789, ZINC000012297018, and ZINC000012296493 emerged as highly effective competitive substitutes for PBUTs. Their displacement efficiency surpasses that of LA, the best competitive displacement agent reported by predecessors. Although their clinical effects need further study in the future, this study, in general, demonstrates the feasibility of virtual screening in discovering competitive substitutes for PBUTs. It provides a theoretical basis and reference for the future development of new competitive substitutes for PBUTs.

Materials and methods

Protein structure preparation and redocking calculation

The protein structure of HSA (PDB ID: 2BXH), obtained from RCSB PDB (https://www.rcsb.org/), underwent a three-step preparation process: repairing missing atoms using WHAT IF (https://swift.cmbi.umcn.nl/), and adding hydrogens and Gasteiger electric charges using Reduce²⁷ and AutoDock Tools 1.5.6 (https://ccsb.scripps.edu/mgltools/downloads/)²⁸.

The search space was defined on the basis of the coordinates of the IS bound to site II and site I of HSA, the center coordinates of site II were x: 6.236, y: 2.681, and z: -15.064, with a search extent of x: 23.25, y: 23.25, and z: 20.25. Site I had center coordinates of x: 3.597, y: -5.893, and z: 7.275, with search extents of x: 31.5, y: 24.75, and z: 24.75. utilizing a grid box with a 0.375 Å spacing between points in each dimension, covering the entire binding region.

Redocking calculations were performed on the native ligand using AutoDock Vina 1.1.2 (https://vina.scripps.edu/downloads/)²⁹. The best-ranked pose was then compared with the experimental pose of this ligand in the crystallographic structure 2BXH, evaluating the RMSD value between these two molecular positions using in-house scripts in PyMol.

Ligand library preparation

Compounds similar to LA were obtained from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/)³⁰, and compounds from the “world” and “natural-product” subsets of the ZINC database (https://zinc15.docking.org/)³¹were separately downloaded. Compounds with at least 85% similarity to LA were retrieved from PubChem using a search method based on the Tanimoto index³². The 3D structures of these compounds were obtained, and the protonated/deprotonated species of each molecule at pH 7.4 were generated using Open Babel version 3.1.1 (https://github.com/openbabel/openbabel)³³.

The other part of the dataset consisted of compounds approved by major regulatory institutions, including the Food and Drug Administration (FDA), as well as marketable natural compounds recognized as secondary metabolites or their analogues. These compounds were sourced from the “world” and “natural-product” subsets of the ZINC database, respectively. All of the compound information was organized into individual files on the basis of unique 3D structures through a shell script, and docking files were prepared using AutoDock Tools 1.5.6 (https://ccsb.scripps.edu/mgltools/downloads/)²⁸. The preparation process involved computing the correct partial charges and detecting the torsion bonds. In the end, the number of compound structures successfully transformed into pdbqt files in each subset was 4178 in PubChem-similarity to LA, 6405 in ZINC-world, and 205,868 in ZINC-natural, respectively.

Molecular docking

In the screening of competitive binders, AutoDock Vina 1.1.2 (https://vina.scripps.edu/downloads/)²⁹ was employed to dock all of the structures from the ligand library to the binding site of HSA. A two-step approach was implemented for the docking screen: initially, a low exhaustiveness value was set to rapidly identify the top-scoring molecules, followed by the use of a high exhaustiveness value to determine the optimal binding conformation for each compound with HSA. The top-scoring structures (binding energy ≤ -9.0 kcal/mol) obtained from the initial low-precision calculations underwent further high-exhaustiveness docking for site II and preliminary docking for site I. The docking calculations were executed on a 320-core CPU cluster computer using a shell script.

ADMET prediction and visual inspection

The SMILES of all of the structures were submitted to ADMETLab 2.0 (https://admetmesh.scbdd.com) for the calculation and prediction of their physicochemical properties, medicinal chemistry measures, ADME endpoints, toxicity endpoints, toxicophore rules, and PAINS-substructure. Molecules displaying significant carcinogenic toxicity, genotoxicity, hepatotoxicity, and the potential for blood-brain barrier penetration (indicating a potential neurotoxicity risk) were excluded. The remaining candidates underwent additional visual inspection, including the analysis of HSA residues that form hydrogen bonds or salt bridges, and the evaluation of the compatibility of ligand hydrophobic groups with the protein binding pocket. Molecules with high flexibility, long aliphatic chains, and those predominantly exposed outside the pocket were excluded, even if they demonstrated good binding affinity. Visual analysis of these interactions is done using the Protein-Ligand Interaction Profiler (PLIP) online tool (https://plip-tool.biotec.tu-dresden.de/plip-web/plip/index) calculations and PyMOL version 2.5.2.

Molecular dynamics simulation

To explore the dynamic interactions between HSA and the final candidates in a solvent environment, molecular dynamics (MD) simulations were conducted using GROMACS version 2019.6 (https://manual.gromacs.org/2019.6/download.html)³⁴. The initial coordinates for the protein-ligand complex and ligand conformations were obtained from the docking study. Topology and coordinate files were prepared with the Amber99SB ILDN force field³⁵ for the HSA protein and converted using the AnteChamber PYthon Parser interfacE (ACPYPE) version 2022.6.6 (https://alanwilter.github.io/acpype/)³⁶ for ligands.

A cubic box with a minimum distance of 1.0 nm between the solute and box edge was subsequently constructed for each protein-ligand complex system, and it was filled with TIP3 water molecules³⁷. Counterion treatment was applied for system neutralization. A 50,000-step energy minimization with a step size of 1 fs was performed using the steepest descent algorithm, with a maximum force set to 150.0 kJ mol⁻¹ nm⁻¹as the convergence criterion. Following this, the minimized structures underwent equilibration for 250 ps under an isothermal-isochoric (NVT) ensemble and 1 ns under an isothermal-isobaric (NPT) ensemble, utilizing the Bussi-Donadio-Parrinello thermostat and Parrinello–Rahman methods^38,39. The target equilibration temperature was set to 300 K, and the target pressure was 1 bar.

Finally, a 200 ns MD production simulation was conducted at 300 K and 1 bar, employing a cut-off of 1.2 nm for short-range electrostatic and van der Waals interactions, the Particle Mesh Ewald (PME) method for long-range electrostatic interactions, and a storage frequency of 100 ps for complex coordinates. The MD trajectories generated were analysed using virtual molecular dynamics version 1.9.4 alpha 51 (https://www.ks.uiuc.edu/Research/vmd/)⁴⁰.

MM-PBSA calculation

To calculate the binding free energy of each protein-ligand complex, the MD trajectories were analysed using gmx_MMPBSA 1.6.1 (https://github.com/Valdes-Tresanco-MS/gmx_MMPBSA)⁴¹. The free energy of the complex was calculated according to Eq. (1):

1

The , , and refer to the energy of protein-ligand complexes, proteins, and small molecule compounds, respectively. All three methods consider the vacuum and the dissolved energy. EvdW and Ecoul represent the energy of the molecular mechanical term (energy under vacuum). The Poisson–Boltzmann equation is used to estimate the dissolved free energy term, whereas the nonpolar dissolved energy term is calculated from the solvent accessible surface area.

Microdialysis in vitro

The microdialysis system used consists of a microdialysis syringe pump (CMA402, Stockholm, Sweden) and dialysis probes (CMA Microdialysis AB, Torshamnsgatan 30 A, 16440 Kista, Sweden). The membrane fitted on the probes is 4 mm in length and 0.5 mm in outside diameter, with a molecular weight cut-off of 20,000 Da. Indoxyl sulfate (IS) and p-cresol sulfate (pCS) were purchased from MedChemExpress Co., Ltd. (St. NJ, USA). LA was provided by Dalian Meilun Biotechnology Co., Ltd. (Dalian, China) (Lot: F0501AS). The compounds ZINC000008791789, ZINC000012297018, ZINC000012296493, and ZINC000004090361 were purchased from InterBioScreen Co., Ltd. (Russia). ZINC000031161007 (Chrysin-7-O-glucuronide) was obtained from Shanghai Yuanye Bio-Technology Co., Ltd. (Shanghai, China)⁴². The perfusate (compound sodium chloride solution) was provided by Double-Crane Pharmaceutical Co., Ltd. (Anhui, China).

The IS and pCS standards were accurately weighed and dissolved in a mixture at a concentration of 50 µg/mL. IS and pCS were each prepared as 1 mg/mL stocks using ultrapure water. The positive control LA was prepared as an 8 mM stock, and five samples were prepared as 16 mM stocks, stored at -80 °C, and serially diluted before the experiment for later use. The total volume of the entire dialysis system was 400 µL. First, 20 µL of IS and 20 µL of pCS solution were added to 320 µL of rat mixed plasma and coincubated at 37 °C for 30 min. Subsequently, 40 µL of the sample solution was added, 40 µL of LA solution was added to the positive control group, and 40 µL of the same solvent was added to the blank group. The mixture was coincubated at 37 °C for 1 h (n = 3), resulting in a final concentration of 50 µg/mL for both the IS and pCS. The microdialysis system was pre-perfused with a compound sodium chloride solution and equilibrated in the blank control system for 1 h. The perfusion flow rate was set at 2 µL/min, the sample temperature was maintained at 37 °C using a thermostatic metal bath (ALB-H4, Thermo BATH), and the dialysate was collected for 1 h.

Determination of IS and pCS in dialysate by HPLC-fluorescence

The concentrations of IS and pCS in the dialysate were determined via a high-performance liquid chromatography system (HPLC, Agilent 1260 Infinity II-FLD, Agilent Co., Ltd., USA) with a Zorbax XBD-C18 column (4.6 mm×50 mm, 1.8 μm, Agilent, USA). The column and autosampler temperatures were maintained at 35 °C and 4 °C, respectively. The mobile phase consisted of (A) ammonium formate (0.1%) and (B) acetonitrile. The conditions for isocratic elution were optimized as follows: 25% B (0–6 min). The IS was eluted over approximately 2.5 min, and pCS was eluted over approximately 3.2 min. IS and pCS were detected by fluorescence at specific excitation/emission wavelengths (IS: 280/375 nm; pCS: 260/300 nm). The injection volume was set to 5 µL⁴³.

Statistical analysis

Two independent-sample t-test were used to analyse differences between groups. A significant difference was indicated by P < 0.05, whereas a highly significant difference was denoted by P < 0.01. The dialysis efficacy of the IS and pCS (%) was calculated as follows: response of the IS or pCS in the dialysate / response of the IS or pCS in the standard sample × 100%. The improvement rates of dialysis efficacy (%) were calculated as follows: (Dialysis efficacy of samples treated with LA or 5 compounds – Dialysis efficacy of blank control samples) / Dialysis efficacy of blank control samples × 100%.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Yan Wang and Shengtian Zhao conceived and designed the project, writing – review & editing. Shasha Liu and Ping Wang analysed the data, validation, made the figures, writing – original draft.

Data availability

The data, including input and parameter files for molecular docking and molecular dynamics simulations, as well as the scripts for the molecular dynamics simulations results analysis, can be found at the Github repository: https://github.com/Labw405/lab405.

Declarations

Competing interests

The authors declare no competing interests.