Computational discovery and repurposing of chloramphenicol succinate as a potent P2Y₁₄ receptor antagonist for inflammatory bowel disease therapy

Graphical abstract

Highlights

• •Drug repurposing of chloramphenicol succinate (DB07565) as P2Y₁₄R antagonist via integrative in silico screening.
• •DB07565 binds P2Y₁₄R with IC₅₀ = 1.59 nM and low cytotoxicity, promising for IBD therapy.
• •MD simulations and MM/GBSA validate DB07565–P2Y₁₄R binding mode and affinity.
• •Unified computational–experimental pipeline accelerates P2Y₁₄R-targeted drug discovery.
• •Repositioning FDA-approved DB07565 reduces cost, timeline, and risk versus de novo design.

Abstract

Introduction

The P2Y₁₄ receptor (P2Y₁₄R), a Gi-coupled receptor activated by UDP-glucose, plays a critical role in inflammatory responses and immune regulation. Existing P2Y₁₄R antagonists face limitations such as poor bioavailability and structural homogeneity, hindering therapeutic development for inflammatory bowel disease (IBD). Drug repurposing offers a promising strategy to bypass traditional drug discovery challenges by leveraging approved drugs with established safety profiles.

Objectives

This study aimed to computationally identify FDA-approved or experimental drugs as novel P2Y₁₄R antagonists and validate their therapeutic potential for IBD treatment.

Methods

A multi-step computational pipeline integrated structure-based virtual screening (SBVS) of DrugBank drug compounds, molecular docking (Glide XP/AutoDock Vina), molecular dynamics (MD) simulations, and MM/GBSA binding free energy calculations. Top candidates underwent in vitro P2Y₁₄R antagonism assays and cytotoxicity testing. In vivo efficacy was evaluated in a DSS-induced murine colitis model.

Results

Chloramphenicol succinate (DB07565), an antibiotic, emerged as a potent P2Y₁₄R antagonist with nanomolar efficacy (IC₅₀ = 1.585 nM) and minimal cytotoxicity. MD simulations revealed strong interactions with conserved residues (K77, Y102, H184, K277), yielding a binding affinity (ΔG_bind = -54.04 kcal/mol) superior to reference compounds. In vivo, DB07565 alleviated colitis symptoms, reduced colon shortening, and restored gut barrier integrity by enhancing tight junction protein expression (Claudin-1, ZO-1, Occludin).

Conclusion

This study demonstrates that computational repurposing successfully identifies DB07565 as a high-affinity P2Y₁₄R antagonist with therapeutic efficacy in IBD. Its established safety, oral stability, and optimized ADME/T properties position it as a clinically translatable candidate, underscoring the value of integrating SBVS and drug repurposing for accelerating anti-inflammatory drug discovery.

Introduction

Nucleotide extracellular receptors, also known as purinergic receptors, are mainly divided into two types: adenosine (P1) receptors and nucleotide (P2) receptors. In the P2 receptor family, there are two separate subgroups: the ligand-gated ion channel P2X receptors and the metabotropic P2Y receptors (P2YRs) [1,2]. P2Y receptors are broadly distributed in a variety of cell types and tissues, such as dendritic cells, macrophages, neurons, monocytes, platelets, as well as smooth and skeletal muscle cells. These receptors are essential in a variety of biological activities, including the clustering of platelets, the multiplication of smooth muscle cells, the modulation of immune functions, and the protection of neural cells [[2], [3], [4]].

Human P2Y receptors are divided into two subgroups based on their primary G protein interaction profiles. The first subgroup, similar to P2Y₁R, comprises receptors such as P2Y₁, P2Y₂, P2Y₄, P2Y₆, and P2Y₁₁. The second subgroup resembles P2Y₁₂R and includes receptors like P2Y₁₂, P2Y₁₃, and P2Y₁₄ [3]. The P2Y₁₄ receptor (P2Y₁₄R) is present in a range of immune and parenchymal cells. It serves as an inhibitory G-protein (Gi)-linked receptor, playing a role in regulating adenylate cyclase activity (Fig. 1). Activation of P2Y₁₄R has been demonstrated to play a role in pro-inflammatory reactions, neutrophil movement, and mast cell degranulation [[5], [6], [7]]. The functional expression of P2Y₁₄R in immune cells has a causal influence on the behavior of these cells during inflammatory diseases [[7], [8], [9], [10], [11], [12]].

Fig. 1.

Mechanistic diagram of P2Y₁₄R.

The recruitment of neutrophils in the lungs triggered by lipopolysaccharide (LPS) is associated with the upregulation of P2Y₁₄R [13]. Activation of P2Y₁₄R enhances mast cell degranulation [5,6] and couples to Gi proteins, thus suppressing adenylyl cyclase (AC) activity and decreasing the production of cAMP. The resulting reduction in cAMP levels inhibits the activation of the NLRP3 inflammasome, which causes impaired Caspase-1 regulation and enhances IL-1β secretion [[14], [15], [16], [17], [18], [19], [20], [21], [22]]. Previous studies demonstrated that P2Y₁₄R contributes to acute gout flares by modulating NLRP3-mediated pyroptosis and inflammatory cytokine release in macrophages [[23], [24], [25], [26], [27], [28], [29]]. In a previous study conducted by our group [30], we synthesized powerful antagonists of P2Y₁₄R, which demonstrated inhibitory effects on monosodium urate (MSU)-induced pyroptosis in THP-1 cells. This indicates that using small molecules to pharmacologically inhibit P2Y₁₄R might serve as an efficient approach for managing inflammatory conditions. Additionally, the P2Y₁₄R is expressed in a variety of tissues, such as the placenta, fat tissue, brain, spleen, intestines, thymus, heart, and lungs [19]. Recent studies indicate that abnormal expression of P2Y₁₄R in tissue cells plays a role in the progression of ischemic acute kidney injury, inflammatory bowel disease (IBD), and chronic neuropathic pain [12,31,32]. Numerous animal studies highlight the significance of P2Y₁₄R as a possible therapeutic target for addressing insulin resistance in diabetes, macrophage migration to the liver, and local inflammation [33,34]. Therefore, targeting P2Y₁₄R holds therapeutic promise for modulating immune system regulation [35] and treating conditions such as pain [36], diabetes [37], cystic fibrosis, and respiratory disorders [38,39].

Currently, the number of known antagonistic classes for P2Y₁₄R remains limited [34,[40], [34], [41], [42]]. Notably, Jacobson et al. discovered a highly effective and selective P2Y₁₄R antagonist named PPTN (as shown in Fig. 2), which exhibits an IC₅₀ value of 21.6 nM. However, due to its polar zwitterionic nature, PPTN exhibited reduced oral bioavailability [43]. Despite this limitation, PPTN-based compounds remain the most extensively studied P2Y₁₄R antagonists. In 2022, Jacobson et al. synthesized another potent compound, MRS4738 (compound 2, Fig. 2), which demonstrated the highest specificity for inhibiting P2Y₁₄R (IC₅₀ = 3.11 nM, T_1/2 = 8.98 h) [44]. Challenges related to the poor absorption and solubility of 2-naphthalic acid antagonists have been documented [45]. To address these limitations, Jiang et al. developed compound 3 (IC₅₀ = 1.77 nM, Fig. 2) by introducing an amide linker between two phenyl groups, thereby enhancing the physicochemical properties of 2-naphthenic acid antagonists [46]. In 2020, Jiang et al. disclosed a patent for a 5-amino-1H-pyrazole derivative (compound 4, IC₅₀ = 0.64 μM) as a novel P2Y₁₄R antagonist. Our research group employed two well-validated P2Y₁₄R homology models and implemented a structure-based virtual screening (SBVS) approach by utilizing a stepwise Glide docking protocol. to identify potent P2Y₁₄R antagonists [47,48]. This strategy led to the discovery of two promising hits, compounds 5 and 6 (Fig. 2), which exhibited enhanced antagonistic activity with IC₅₀ values of 2.46 nM and 5.12 nM, respectively. Inspired by these potent virtual screening hits, we rationally designed and synthesized a series of potential P2Y₁₄R antagonists featuring a benzoxazole core. Among these, Antagonist 7 (Fig. 2) emerged as the most effective antagonist for P2Y₁₄R, exhibiting an IC₅₀ value of 2 nM. In 2021, a polyacetylene compound, identified as an essential active component in Dangshen, was successfully isolated. Lobetyolin (Fig. 2) exhibited potent antagonistic activity against P2Y₁₄R with a value of 5.80 μM [49]. In 2023, after optimizing the homology model of P2Y₁₄R, Glide docking was utilized to screen and select 14 natural compounds from TargetMol (USA). Among these, Paederosidic acid derived from Paederia scandens (Fig. 2) demonstrated the highest potency against P2Y₁₄R (IC₅₀ = 8.29 μM) [50,51]. The structural diversity of natural products [52] offers promising opportunities for the identification of new and highly effective P2Y₁₄R antagonists (Fig. 2). However, the current repertoire of P2Y₁₄R antagonists is characterized by limited structural diversity. Therefore, there is a pressing requirement to discover novel P2Y₁₄R antagonists. That exhibit high potency, exceptional specificity, and favorable ADME/T properties [53].

Fig. 2.

Representative chemical structures of known P2Y₁₄R antagonists.

Virtual screening (VS) serves as a fundamental technique in computer-aided drug design (CADD), has substantially advanced the drug design process by virtue of its exceptional efficiency and precision [54,55]. One of the key benefits of VS is its ability to significantly decrease the number of compounds needing synthesis and experimental testing, thereby significantly cutting down on both time and costs [[56], [57], [58]]. Drug repositioning involves discovering novel therapeutic uses for approved or experimental medications beyond their original indications [59]. In light of the high failure rates, substantial expenses, and slow progress associated with traditional drug discovery, repurposing existing drugs for both common and rare diseases is becoming increasingly appealing. This strategy leverages de-risked compounds, potentially reducing developmental costs and timelines [60]. Numerous successful examples now demonstrate the feasibility of this approach. For instance, Dakshanamurthy et al. performed high-throughput computational docking analyses on 3,671 FDA-approved drugs across 2,335 human protein crystal structures, revealing the structural potential of mebendazole, an antiparasitic agent, to inhibit vascular endothelial growth factor receptor 2 (VEGFR2), a key mediator of angiogenesis [61]. This finding was subsequently validated experimentally [60]. Additionally, Sildenafil, originally created for use as a blood pressure medication, was effectively adapted by Pfizer for the treatment of erectile dysfunction [62]. Montelukast, primarily used for asthma management, has also been investigated for its potential effects on neurodegenerative diseases. Thalidomide was initially introduced as a sedative in the late 1950 s, was later identified as an effective treatment for erythema nodosum leprosum (ENL) in 1964 and subsequently emerged as a therapeutic option for multiple myeloma in 1999 [63].

In this research, we utilized a systematic computational methodology to detect potential antagonists of the P2Y₁₄R from both FDA-approved and experimental drug compounds within the DrugBank database. Initially, we assessed five representative P2Y₁₄R-antagonist complexes to differentiate known antagonists from decoys and select the most reliable P2Y₁₄R complex template for structure-based drug design. Subsequently, by employing two distinct docking-based virtual screening strategies including AutoDock Vina [64] integrated into DrugRep (http://cao.labshare.cn/drugrep/) [65] and a combination pipeline utilizing the extra precision (XP) scoring function of Glide docking [66] alongside Prime MM/GBSA binding free energy calculations, we identified top-ranked promising compounds targeting P2Y₁₄R from FDA-approved and experimental drugs for each method, thereby warranting further analysis. Finally, we performed a comprehensive evaluation involving Molecular dynamics (MD) simulations along with the computations of binding free energy/landscape to elucidate interaction patterns between four potential antagonists and P2Y₁₄R. This approach allowed us to scrutinize the dynamic interactions between P2Y₁₄R protein and ligands while examining conformational changes occurring during the dissociation pathway of active compounds. MM-GBSA binding free energy analysis indicated that DB07565 exhibits significantly higher binding affinity (ΔG_bind = -54.04 kcal/mol) compared to the reference Antagonist 7 (ΔG_bind = -44.84 kcal/mol). Consequently, we select DB07565: Chloramphenicol succinate for primary in vitro and in vivo biological testing. The experimental results demonstrated that DB07565 possesses potent P2Y₁₄R antagonistic activity (IC₅₀ = 1.585 nM) and exerts a more pronounced inhibitory effect on DSS-induced experimental colitis. This study presents an efficient computational strategy for the advancement of powerful P2Y₁₄R antagonists featuring innovative chemical structures.

Materials and methods

Preparation of P2Y₁₄R-antagonist complexes and validation dataset

Five P2Y₁₄R-antagonist complexes (Fig. 3), comprising two P2Y₁₄R homology models (MMC3 and MMC4) reported by Trujillo et al. [67], two virtual screening hits from Glide docking, and one rational design/modification antagonist interacting with P2Y₁₄R developed in our group, were selected, optimized, and incorporated into the drug repurposing pipeline. The Protein Preparation Wizard module in Schrödinger 9.0 [67] was used to eliminate all water molecules, repair broken side chains, and add missing hydrogen atoms. Partial charges and protonation states for each homology model were assigned using the OPLS_2005 force field [30]. The binding pocket for each P2Y₁₄R complex was determined by utilizing the Receptor Grid Generation component of Schrödinger 9.0, with the active site center set as the centroid of the co-ligand. All other parameters were kept at their default settings.

Fig. 3.

Structural superposition of the five P2Y₁₄R complexes.

The known human P2Y₁₄R antagonists were retrieved from the BindingDB database [68]. After removing duplicates and selecting compounds with documented antagonistic activity (IC₅₀ or K_i value), a total of 68 P2Y₁₄R antagonists were retained for further analysis. Non-antagonists (decoy compounds) were generated using the DUD-E [69]. DUD-E is designed to benchmark molecular docking programs by providing challenging decoys, allowing standardized evaluation and comparison of different docking algorithms and scoring functions. For each antagonist, 50 decoys, possessing similar physicochemical characteristics yet distinct two-dimensional structures were generated. to ensure diversity in the dataset. The LigPrep module from Schrödinger was utilized to create ionized states and tautomers at a pH of 7.0 by employing the Epik module, while preserving the original chiralities for known antagonists with available 3D structures. For decoy compounds lacking pre-existing 3D structures, all possible chiral centers were enumerated (up to a maximum of 32 stereoisomers per molecule). Ultimately, the validation dataset consisted of 94 antagonist structures and 6,581 non-antagonist structures.

The screened library, comprising of 2,588 FDA-approved drug compounds and 6,251 FDA-experimental drug compounds (accessed on January 4th, 2023) [70] was generated utilizing the LigPrep module within the Schrödinger software suite. For each drug molecule, tautomers were created at a pH of 7.0 ± 2.0 utilizing the Epik tool while preserving specified chiralities (varying other chiral centers). Stereoisomers were generated by limiting the maximum number of stereoisomers to 32. The prepared library consisted of 6,546 protomers for FDA-approved drugs and 12,120 protomers for experimental drugs.

Validation of molecular docking procedure

Molecular docking was carried out using the Glide tool within the Schrödinger software package, employing both Standard Precision (SP) and Extra Precision (XP) modes. Selecting a suitable docking mode and complex structure plays a crucial role in enhancing the effectiveness of structure-based virtual screening. To identify the most effective virtual screening approach, we conducted a systematic assessment of the “discriminatory capacity” of molecular docking in distinguishing known P2Y₁₄R antagonists from decoy compounds. This evaluation employed two scoring functions within Glide docking: SP and XP based on five P2Y₁₄R-antagonist complex structures.

Structure-based virtual screening pipeline

Considering the intrinsic constraints of individual docking-based virtual screening (VS) approaches, it is unrealistic to expect any single approach to consistently perform optimally across all scenarios [71]. As highlighted by Sheridan and Kearsley, “various approaches might produce different sets of active compounds for a particular biological function, and the effectiveness of a given method can differ when applied to various activities” [72]. To make use of the diverse advantages of various VS approaches, integrated strategies have been introduced, either sequentially or in parallel. Based on our previous studies [71,73], AutoDock Vina strikes a good balance between speed and accuracy, making it an excellent choice for large-scale virtual screening of compound libraries. It allows for a quick yet effective initial assessment of potential binders, helping us rapidly identify compounds that may interact favorably with the target receptor. Meanwhile, Glide XP, especially when combined with MM/GBSA for binding free energy calculations, is known for its high precision in predicting binding poses. This approach provides a more detailed and accurate evaluation of molecular interactions, which is essential for gaining deeper insights into how potential antagonists interact with P2Y₁₄R. In this study, we employed two consecutive state-of-the-art SBVS methods: AutoDock Vina from DrugRep and a hybrid approach combining Glide XP docking and Prime MM-GBSA binding free energy calculations, to identify potential antagonists of P2Y₁₄R from the DrugBank database. Each docking method possesses distinct strengths and limitations. By integrating multiple approaches, we aimed to conduct a more comprehensive evaluation-ensuring that no potential hits are overlooked while also enhancing the likelihood of identifying a diverse array of potential antagonists with varying chemical scaffolds. First, the CurPocket [74,75] embedded in the DrugRep was utilized to determine the binding site for the P2Y₁₄R complex with optimal “discrimination power”. The protomers of both FDA-approved and experimental drugs from DrugBank were then docked into this binding pocket and scored using AutoDock Vina with default settings in the DrugRep platform. Similarly, a binding pocket was constructed by employing the Receptor Grid Generation mode in Glide by defining the active site center as the centroid of co-ligand. These well-prepared protomers were also subjected to docking utilizing the XP scoring function from Glide. To enhance the precision of binding affinity predictions, Prime MM/GBSA minimization in Schrödinger software was employed to re-evaluate the docking poses. Finally, the top 10 compounds ranked for each method, from both approved and experimental drugs in DrugBank were selected for further analysis.

System setup and molecular dynamic simulations

The apo form of P2Y₁₄R, together with its complexes formed with the top-ranked compounds (DB00320, DB04764, DB02706, and DB07565), as well as the validated antagonist 7 from our earlier research, were analyzed using the “Orientations of Proteins in Membranes (OPM)” database. Based on the orientation of P2Y₁₂R bound to 2MeSADP (PDB: 4PXZ [76]), a proposed membrane orientation was determined for each structure.

Each protein–ligand complex was incorporated into a pre-equilibrated 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) lipid bilayer (80 Å × 80 Å). This bilayer was created using the VMD Membrane Plugin tool. During the process of protein insertion, any overlapping lipids within a distance of 0.6 Å were methodically eliminated.

All protein-membrane systems were hydrated using TIP3P water molecules and balanced with 0.154 M Na+/Cl- counterions. The force field parameters for the ligands were determined by calculating RESP (restraint electrostatic potential) atomic partial charges at the HF/6-31G* theoretical level. This was performed on geometry-optimized structures using the Gaussian 16 quantum chemistry software. The AMBER ff14SB and lipid14 force fields were applied for the receptor and POPC lipids, respectively. A four-stage energy minimization protocol was executed to eliminate unfavorable contacts. Initially, a 20,000-step energy minimization was conducted on water molecules with all other atoms fixed. Then, POPC lipids underwent energy minimization for 20,000 steps, followed by a 20,000-step minimization for the ligand and receptor. Ultimately, all atoms in the system were released for a 20,000-step energy minimization.

All the studied systems were progressively heated from 0 to 310 K over a period of 250 ps under NVT conditions, followed by a 5 ns pre-equilibration phase. The production molecular dynamics (MD) simulations were conducted at 310 K and 1 atm, employing a Langevin thermostat for temperature control and a Berendsen barostat for pressure regulation. Each system underwent production MD simulations lasting 1 μs, maintained at 310 K and 1 atm through the use of Langevin thermostat and Berendsen barostat controls. Long-range electrostatic interactions were handled using the particle mesh Ewald (PME) method with 10 Å cutoff, [77], while SHAKE algorithm [78] was utilized to restrict all covalent bonds that involve hydrogen.

All molecular dynamics (MD) simulations were carried out using the Amber 20 software, with an integration time step set to 2 fs. The production run trajectories, after equilibration, were analyzed utilizing the CCPTRAJ module [79], yielding root-mean square deviations (RMSDs), root-mean square fluctuations (RMSFs), radius of gyration (Rg), dynamical cross-correlation matrices (DCCMs), and free energy landscape (FEL) analyses.

Computational protocols for protein–ligand interaction energy analysis

The molecular mechanics/Generalized born surface area (MM/GBSA) method was utilized to evaluate the binding free energies (ΔG_bind) for each protein–ligand complex. This assessment was based on snapshots from the final 500 ns of the equilibrated trajectory. The binding free energy (ΔG_bind) was determined using the following equation:

$$\begin{array}{l}\mathrm{\Delta }{G}_{\text{bind}}=G_{\mathit{com}}-(G_{\mathit{rec}}+G_{\mathit{lig}})\\ =\mathrm{\Delta }E{}_{\mathit{MM}}+\mathrm{\Delta }{G}_{\mathit{solvation}}-T\mathrm{\Delta }S\\ =\mathrm{\Delta }{E}_{int}+\mathrm{\Delta }{E}_{\mathit{ele}}+\mathrm{\Delta }{E}_{\mathit{vdW}}+\mathrm{\Delta }{G}_{\mathit{GB}}+\mathrm{\Delta }{G}_{\mathit{SA}}-T\mathrm{\Delta }S\end{array}$$

where. ΔE_MM can be determined using molecular mechanics methodology, which indicates the gas-phase energy interaction between the protein and the ligand, encompassing van der Waals interaction (ΔE_vdW), electrostatic interaction (ΔE_ele), polar (ΔG_GB) and nonpolar (ΔG_SA) components to the solvation free energy (ΔG_solvation). TΔS denotes the conformational entropy upon ligand binding. For membrane-bound protein systems, MM/GBSA calculations differ from those in solvated protein systems due to the incorporation of the implicit membrane model. The membrane, acting as a partially polar medium, significantly influences the ligand-binding process. Parameters for MM/GBSA calculations in membrane-bound systems were configured according to Amber tutorial manual. Specifically, MD trajectories were processed using CPPTRAJ to eliminate solvent molecules, membrane components, and counterions from the receptor-ligand complex. Free energies of all systems were calculated using a heterogeneous dielectric membrane model within SANDER/GBSA calculations. The polar contribution to the solvation energy (ΔG_GB) was calculated by employing the modified GB model (GB^OBC1, igb = 8) [80], which was developed by Onufriev. The dielectric constants for the solute (ε_in) and solvent (ε_out) were set to 2 and 80, respectively. The nonpolar contribution to the solvation energy (ΔG_SA) was determined by evaluating the alterations in solvent-accessible surface areas (ΔSASA) using the LCPO algorithm [81]. This can be expressed as ΔG_SA = γ × ΔSASA + β, with γ and β being assigned values of 0.0072 kcal/(mol·Å²) and 0 kcal/(mol·Å²), respectively. TΔS was neglected due to excessive computational demands and limited accuracy. Other parameters were default values in MMPBSA.py. Residue-specific ligand-P2Y₁₄R interaction energies were computed according to the equation: ΔG_{ligand‑residue} = ΔE_vdW + ΔE_ele + ΔG_GB + ΔG_SA. Apart from ΔG_SA, which was computed using the ICOSA algorithm, the other terms were determined with the same parameters used in the aforementioned MM/GBSA energy calculations.

In vitro P2Y₁₄R antagonistic activities screening

Human embryonic kidney 293 (HEK293) cells, which stably express the human P2Y₁₄ receptor (hP2Y₁₄-HEK293 cells), were obtained from Keygen Biotech Co., Ltd. These cells were maintained in Dulbecco's Modified Eagle Medium (DMEM) containing 1 % penicillin/streptomycin and 10 % fetal bovine serum (FBS) under conditions of 37 °C with 5 % CO₂. Prior to conducting activity assays, the cells were seeded at a concentration of 10,000 cells per well into 384-well plates that were white with clear bottoms. They were then incubated for 30 min at 37 °C in the presence of either DMSO or antagonists. Then examine the P2Y₁₄R inhibition activity through testing the intracellular cAMP using a cAMP-Glo^TM Assay kit from Promega (v1502, Madison, Wisconsin, USA). Data were collected using a multi-mode microplate reader.

Cell viability assays

HT-29 cells were purchased from the BeNa Culture Collection Co, ltd, and cells were cultured in DMEM containing 1 % penicillin/streptomycin and 10 % FBS at 37 °C in a 5 % CO₂ environment. Prior to assessing cell viability, cells were seeded at a density of 10,000 cells per well in 96-well plates and cultured with either DMSO or antagonists for 24 h at 37 °C. Subsequently, cell viability was evaluated using the Cell Counting Kit-8 (BS350A, Biosharp, Shanghai, China) following the manufacturer's guidelines.

In vivo anti-inflammatory evaluation of DB07565

Male, 7–8-week-old, mice with C57BL/6 J background were used to establish experimental colitis model. These mice were randomly separated into five groups based on their weight, with six mice in each group: control group, model group, 5-ASA, DB07565 (100 μM) group and DB07565 (50 μM) group. Except the control group, the animals received 3 % (w/v) DSS (MP Biomedicals, Solon, Ohio) in their drinking water to trigger experimental colitis for 7 days. For DB07565 (100 μM) group and DB07565 (50 μM) group, DB07565 was first dissolved in DMSO and then diluted with PBS to achieve the desired concentration. It was administered rectally at a volume of 100 µl per mouse on a daily basis. Additionally, mice in the 5-ASA group received oral administration of 5-ASA at a dose of 50 mg/kg each day. The entire cohort of mice was observed daily to track changes in body weight and Disease Activity Index (DAI). 7 days after DSS treatment, All the experimental mice were euthanized, and the lengths of their colons were recorded. The remaining colon tissues were then harvested for subsequent analysis.

Histopathological analysis

For the histological evaluation of colitis, colon samples were preserved in 4 % paraformaldehyde (PFA) and subsequently embedded in paraffin wax. Tissue sections, each measuring five micrometers in thickness, were stained using hematoxylin and eosin (H&E) and analyzed for signs of colitis according to previously established methods. Microscopic images were obtained using a BX53 microscope (Olympus).

Immunohistochemistry and immunofluorescence

Immunofluorescence staining was performed using paraffin-embedded tissues as mentioned before. The paraffin slides were deparaffinized, rehydrated, blocked and treated according to a standard protocol. The levels of tight junction proteins related to the intestinal barrier were assessed by incubating tissue samples with primary antibodies against Claudin-1 (bioss, bs-10011R), Occludin (bioss, DF6919), and ZO-1 (Affinity, AF5145) at 4 ℃ overnight. This was followed by a 1-hour incubation with the appropriate secondary antibody. To visualize nuclear positioning, all slides were stained with DAPI for 10 min. Microscopic imaging was performed using a BX53 microscope (Olympus).

Results and discussion

Performance of docking-based validation and reliable P2Y₁₄R complex selection

Validating homology models is essential for reliable docking studies. Ramachandran plots was applied to detect structural outliers and the results indicated that 96.3 % of P2Y₁₄R residues are in favored regions, 3.7 % in allowed regions, and none in disallowed regions (Fig. S1 in Supporting Materials). Recently, the crystal structure of human P2Y₁₄R was published on May 13, 2025. We therefore aligned our homology model with the new crystal structure (UDP-Glucose bound purinergic receptor P2Y₁₄R, PDB ID: 9J0B) [82] to evaluate template accuracy and identify structural differences (Fig. S2). The comparison showed a high degree of conformational similarity between our homology model (orange) and the crystal structure (purple), with a main-chain RMSD of only 1.20 Å. This confirms that our P2Y₁₄R homology model closely matches the actual structure, supporting the reliability of our template. These validation results collectively show that our homology models are of high quality and suitable for reliable virtual screening. Our prior investigations have revealed substantial variations in the predictive capacity of molecular docking when different docking templates are applied to a specific target [83,84]. We evaluated the discriminatory ability of five P2Y₁₄R complexes in distinguishing known antagonists from decoy compounds using two scoring modes (SP and XP) in Glide docking. Each compound within the meticulously curated validation set was docked into the five P2Y₁₄R complexes using both SP and XP scoring modes in Glide docking, with their discriminatory powers assessed via the Student's t-test method. Although it is commonly believed that the XP scoring mode is more accurate than the SP scoring mode [85], our findings indicate that SP scoring function does not consistently underperform relative to XP scoring (Fig. 4 and Table S1 in Supporting Materials). Notably, for two P2Y₁₄R complexes (MMC3 and MMC4) constructed by Trujillo et al., Glide SP docking demonstrated superior discriminatory power compared to Glide XP docking, as evidenced by lower p-values. The Glide docking method showed the highest discriminatory power (p-value = 2.95 × 10^-7) for the P2Y₁₄R-compound 6 complex using the XP scoring function. ROC analysis (Fig. S3 in Supporting Materials) of five P2Y₁₄R complexes, based on active and decoy compounds from the DUD-E dataset, confirmed that Glide XP docking achieved the best performance in distinguishing actives from decoys, with an AUC of 0.756 for P2Y₁₄R-compound 6.

Fig. 4.

Distribution of docking scores between known antagonists and decoy compounds using SP and XP scoring modes of Glide docking for five P2Y₁₄R complexes.

Drug repurposing based on two classic docking-based virtual screening strategies

The FDA-approved/experimental medications are widely utilized due to their reasonable cost, favorable ADME/T properties, and minimal likelihood of inducing side effects. These characteristics underscore their significant clinical applicability. To enhance the accuracy of identifying candidate compounds, two molecular docking techniques were employed for virtual screening. Initially, AutoDock Vina from DrugRep was used to screen two FDA compound libraries (approved drugs database and experimental drugs database), resulting in the choice of the top 10 drug compounds that exhibit the most promising docking scores (Fig. 5, Table S2 in Supporting Materials). Subsequently, Glide docking using the XP scoring mode was applied to screen the Drugbank libraries. The binding poses of the drug compounds were then rescored via Prime MM/GBSA minimization, retaining only the top 10 drug compounds with optimal MM/GBSA scores for each database (Fig. 6, Table S3 in Supporting Materials). Detailed interaction patterns of the top 10 approved and experimental drugs, identified as potential antagonists of P2Y₁₄R through DrugRep or Glide XP docking combining with MM/GBSA minimization virtual screening strategies, are presented in Figs. S4–S7 in the Supporting Materials. Despite sharing a similar binding pocket (Fig. S8 in Supporting Materials), the potential P2Y₁₄R antagonists identified through AutoDock Vina of DrugRep and Glide XP docking combined with MM/GBSA minimizations (Fig. 5, Fig. 6) exhibited diverse chemical structures and significant variations. These findings indicate that employing two distinct structure-based virtual screening methods can improve the chemical diversity of virtual screening hits. Finally, leveraging predictions from AutoDock Vina (DB00320: Dihydroergotamine; DB04764: [4-(3-Aminomethyl-phenyl)-piperidin-1-YL]-(5-phenethyl-pyridin-3-YL)-methanone) and Glide XP docking combined with MM/GBSA minimizations (DB07565: Chloramphenicol succinate; DB02706: Mercaptocarboxylate inhibitor), we identified the top1-ranked FDA-approved/experimental drugs for further detailed analysis of their interaction mechanisms with P2Y₁₄R.

Fig. 5.

Docking poses and chemical structures of (a) the top10 FDA approved drugs (b) the top10 FDA experimental drugs predicted as potential P2Y₁₄R antagonists using DrugRep.

Fig. 6.

Docking poses and chemical structures of (a) the top10 FDA approved drugs (b) the top10 FDA experimental drugs predicted as potential P2Y₁₄R antagonists using Glide XP docking and MM/GBSA minimizations.

Molecular dynamics simulation and P2Y₁₄R-ligand interaction energy

The interaction patterns of the top1-ranked potential P2Y₁₄R antagonists, identified through dual computational pipelines including AutoDock Vina and Glide XP/MM-GBSA binding free energy calculations, were depicted in Fig. 7. Although these compounds share a common orthosteric binding domain within P2Y₁₄R, the 2D protein–ligand interaction diagrams reveal distinct residue interaction profiles across the screened compounds.

Fig. 7.

The potential P2Y₁₄R antagonists with the highest ranking from DrugBank, which include both approved and experimental drugs, were predicted using (a) AutoDock Vina of DrugRep and (b) Glide XP docking followed by MM/GBSA minimizations.

However, considering the intrinsic conformational flexibility of P2Y₁₄R as a GPCR, different ligands are likely to interact with P2Y₁₄R in diverse manners, modulating its activity by inducing distinct conformational changes. Therefore, all-atom molecular dynamics (MD) simulations were carried out in conjunction with MM/GBSA free energy calculations to investigate ligand-induced dynamic perturbations and identify high-efficacy antagonists.

System stability study

To provide a thorough understanding of the system dynamics and stability, we included the calculation and analyses of root-mean-square deviation (RMSD), radius of gyration (Rg) and root-mean-square fluctuation (RMSF). Root-mean-square deviation (RMSD) based on MD trajectories (Figs. S9–S10) indicates that ligand-bound systems (P2Y₁₄R·7, P2Y₁₄R·DB02706, P2Y₁₄R·DB07565, P2Y₁₄R·DB00320 and P2Y₁₄R·DB04764) exhibit lower RMSD values compared to the apo-P2Y₁₄R system (Average RMSD: 4.01 vs 3.91, 3.63, 3.56, 3.91 and 3.62 Å, Figs. S9a–b and e). This phenomenon suggests that the screened compound can enhance the structural stabilization of P2Y₁₄R and achieve equilibrium after ∼500 ns MD simulations. The ligand RMSD values provide a quantitative measure of the dynamic behavior of the ligands within the binding pocket with value of 2.81, 2.62. 2.57, 3.10, 2.45 Å for antagonist 7, DB00320, DB02706, DB04764 and DB07565, respectively (Figs. S9c–d and f). Among them, DB07565 exhibit lower dynamic fluctuation, suggesting stronger interaction affinity of DB07565 with P2Y₁₄R during MD simulations.

The radius of gyration (Rg) has been calculated for the protein–ligand complexes to assess the overall compactness of P2Y₁₄R system. It can be observed from Fig. S10a that the Rg value of antagonist 7 and DB07565 bound system are larger than that of DB00320, DB02706, DB04764 systems (21.88 and 22.06 vs 21.62, 21.29 and 21.46 Å). Accordingly, antagonist 7 and DB07565 binding can induce lower RMSD value and higher Rg value, suggesting a complex interplay of local stabilization and global conformational changes. Locally, the binding site might become more rigid and stable (decreased RMSD), while globally, the receptor might adopt a more extended conformation (increased Rg). The antagonist binding might trigger allosteric effects, leading to changes in distant regions of the receptor, which would increase the overall size of the receptor (Rg) while stabilizing the local binding site (RMSD). Moreover, receptor might exist in a dynamic equilibrium between different conformational states, that antagonist binding could shift this equilibrium towards a state with a more extended conformation (higher Rg) while stabilizing the binding site (lower RMSD).

Residue fluctuation profiles (Fig. S10b) demonstrate that both antagonist 7 and DB07565 stabilize the transmembrane regions, TM5 (170–200) and TM6 (240–260), likely due to their strong interactions with residues in these domains. Furthermore, these compounds interacted with residues in extracellular loop 2 (ECL2), which contributes to reduced fluctuations in both ECL2 and TM4. In the P2Y₁₄R·7 complex, fluctuations were significantly increased in TM3 (110–124), ICL2 (125–132) and TM4 (133–140). Additionally, DB07565 can also amplify the flexibility of ICL2 (130–132) and intra region of TM4 (133–139), consistent with the above speculations.

P2Y₁₄R-antagonist contact energy analysis

The MM/GBSA free energy calculation results summarized in Table 1 indicated that all the screened compounds exhibit enhanced binding free energy (ΔG_bind), exceeding that of the reference antagonist compound 7. DB07565 demonstrated the highest binding affinity, with a ΔG_bind value of −54.04 kcal/mol, establishing it as a promising P2Y₁₄R antagonist candidate. This improved binding was primarily driven by strong electrostatic interactions (ΔE_ele = -164.68 kcal/mol) and optimal hydrophobic complementarity (ΔE_vdW = −44.84 kcal/mol).

Table 1.

The binding free energy and the respective energy components of the screened compounds interacting with P2Y₁₄R (kcal/mol).

ID	ΔEvdW	ΔEele	ΔGGB	ΔGSA	ΔGbind
Antagonist 7	−46.26	−10.58	17.36	−5.28	−44.84
DB00320	−46.42	−11.34	13.8	−6.9	−50.86
DB04764	−43.58	−47.94	50.82	−4.84	−45.54
DB02706	−46.92	−130.66	135.22	−4.72	−47.26
DB07565	−44.84	−164.68	160.62	−4.82	−54.04

Furthermore, the per-residue interaction analyses presented in Fig. 8 and Table S4 identified the essential residues participating in the binding process. Antagonist 7 formed energetically favorable contacts with F76, K77 and G80 in TM2, A98, Y102 in TM3, C172 in ECL2, H184 in TM5 (Fig. 8a–b). We have performed hydrogen bond (H-bond) analysis, including H-bond counts (Fraction) and detailed H-bond interactions between the ligands and key residues of the P2Y₁₄R. From Table 2, we can observe that two strong H-bond contacts are formed between Antagonist 7 and Tyr102 and Lys77 in P2Y₁₄R with fraction of 0.83 and 0.66, respectively.

Fig. 8.

(a) Per-residue interaction decomposition of the antagonist-P2Y₁₄R binding free energy for P2Y₁₄R·7 and P2Y₁₄R·DB07565. (b) The 3D-cocrystal structure of P2Y₁₄R with DB07565. (c) Schematic representation of the GPCR-membrane complex.

Table 2.

The hydrogen bond analysis of Antagonist 7 and DB07565 bound P2Y₁₄R system.

#Acceptor	Donor	Frames	Fraction	AvgDist	AvgAng
Antagonist 7@O2	Tyr102@OH	2075	0.83	2.74	160.01
Antagonist 7@O2	Lys77@NH3	1655	0.66	2.85	151.55
DB07565@O4	Lys77@NH3	2373	0.95	2.76	156.85
DB07565@O7	Tyr102@OH	1656	0.66	2.71	160.94
DB07565@O6	Lys277@NH3	1585	0.63	2.81	153.25
DB07565@O4	Tyr256@OH	1167	0.47	2.69	164.11
DB07565@O3	Cys172@NH	511	0.20	2.91	152.22
DB07565@O4	His184@NH2	320	0.14	2.84	158.23

In the DB07565-bound system, the phenyl ring of DB07565 tightly positioned within the narrow pocket that links the orthosteric binding site to the EL environment (Fig. 7), anchoring it in a slightly deeper position through strong interactions with multiple residues. This necessitates greater conformational rearrangement of the protein, thus improving the binding affinity between the protein and ligand in the DB07565-P2Y₁₄R complex. In details, DB07565 can tightly interacts with residues in TM2 (T73, F76, K77 and G80), TM3(V93, C94, S97, A98, F101, and Y102), ECL2 (T168, I170, K171, C172 and I173), TM5(H184), TM6 (Y256) and TM7(K277), as shown in Fig. 8 and Table S4. Notably, DB007565 can also form favorable H-Bond interactions with Lys77 and Tyr102 with the fraction of 0.95 and 0.66, which is consistent with the strong binding affinity between P2Y₁₄R and DB007565 with the value of −5.7 and −2.6 kcal/mol, respectively (Table 2). Moreover, additional H-bonds exist between DB007565 and Lys277, Tyr256, Cys172 (Fraction: 0.63, 0.47, 0.20), suggesting DB007565 as more promising antagonist of P2Y₁₄R. Moreover, K77 formed a strong pi-cation interaction with the phenyl ring of DB07565 (−5.7 kcal/mol), while DB07565 established hydrogen bonds with Y102, His184, Cys172, Y256 and K277, with binding affinity of −2.6, −1.42, −3.58, −1.32 and −3.36 kcal/mol, respectively. Highly conserved positively charged residues in P2Y₁₂-like receptors, such as K277, engage in ionic interactions with negatively charged phosphates. DB07565 interacted with the functionally conserved residues K277 and Y256, and demonstrated pi–pi stacking with Y102. The highly conserved motif K-E-X-X-L in TM7 has been identified in P2Y₁₂-like receptors through sequence alignments, corresponds to residues K277, E278, F279, T280 and L281 in the P2Y₁₄R. Additionally, DB07565 engaged in favorable hydrophobic interactions with residues such as A98 in TM8 and C172 in ECL2. Notably, many stable/strong ligand-residues interactions observed in DB07565 (with Lys77, Tyr102, His184 and Lys277) are also present in various validated P2Y₁₄R antagonist systems. This suggests that these interactions are crucial for antagonist binding and further confirms DB07565 as a potential potent P2Y₁₄R antagonist.

Conformational landscape remodeling in antagonist-bound system

As a class A GPCR, P2Y₁₄R inherently exhibits dynamic behavior, undergoing ligand-dependent conformational transitions that regulate its allosteric coupling to G_i-mediated signaling cascades. Accumulating evidence demonstrates that pharmacologically distinct ligands (e.g., antagonist) modulate P2Y₁₄R function differently by interacting with the receptor in divergent manners and inducing ligand-specific conformational perturbations, thereby highlighting the complexity of its dynamic characteristics. Therefore, in our study, elucidating the conformational changes in the DB07565-bound P2Y₁₄R complex is crucial for identifying the key factors governing activation modulation and further unraveling the antagonistic mechanism of DB07565.

Dynamical cross-correlation matrix (DCCM) analysis

To assess the impact of antagonist binding on P2Y₁₄R’s internal dynamics, Cα-based dynamical cross-correlation matrixes (DCCMs) were performed to quantify the inter-residue motion coupling based on MD trajectories.

In the apo system (Fig. 9a), pronounced anti-correlated motions were observed of the extracellular part of TM6 with TM5 (R₁), TM4 (R₂), and TM3 (R₃). Conversely, the R₄-R₇ region exhibited positive motion correlations between the extracellular part of TM4 and TM5 (R₄), TM3 and TM5 (R₅), ECL2 and TM5, as well as TM3 and TM4 (R₇). In contrast, antagonist binding significantly induced a global rearrangement of dynamic network in P2Y₁₄R, and Antagonist 7 and DB07565 exhibit similar DCCM profile (Fig. 9–c). Specifically, in the presence of Antagonist 7 and DB07565, TM5 and TM6 (R₁), TM4 and TM6 (R₂), and TM3 and TM6 (R₃) displayed similar motion direction. Moreover, the positive correlations of TM4, TM3 with TM5 were notably reduced, and ECL2 and the extracellular part of TM5 exhibited negative motion movement, indicating the close of ligand binding site because of the favorable interactions between DB07565 and specific residues in ECL2.

Fig. 9.

The DCCM analysis of apo and antagonistic P2Y₁₄R system.

Free energy landscape analysis

In the ligand-free P2Y₁₄R system, several protein states at local energy minima were identified, including State1, State2 ∼ 2′ and State3 ∼ 3′’. Among these, State1 exhibited the lowest energy (Fig. 10). Compared with State2 ∼ 2′ and State3 ∼ 3′’, State1 undergone significant conformational changes in transmembrane helices TM3, TM5, TM6 and TM7, which are characteristic of the active receptor conformation. Specifically, in State1, the intracellular portion of TM6 shifted outward, away from TM3, thereby enabling TM7 to move inward and approach TM3 (Fig. 11a and b). This eventually created a cavity on the intracellular side of the receptor, which aids in the binding of downstream effector proteins such as Gi proteins. Conversely, the binding of antagonist 7 and DB07565 both prevented the inward movement of TM6 and TM5, thereby increasing the distance and weakening the energetic interactions between TM3 and TM6. This is a conserved feature of the inactive conformation, further supporting the potential of DB07565 as a promising P2Y₁₄R antagonist (Fig. 11b).

Fig. 10.

The free energy landscape analysis of apo-P2Y₁₄R system and corresponding conformation in local energy minima.

Fig. 11.

The free energy landscape analysis of (a) P2Y₁₄R·7 and (b) P2Y₁₄R·DB07565 system and corresponding conformations in local energy minima.

Biological testing

The biological activity of DB07565 purchased from TargetMol, USA was evaluated using both in vitro and in vivo models. First, the in vitro P2Y₁₄R inhibition activity of DB07565 was assessed in hP2Y₁₄-HEK293 cells by examining its influence on UDPG-induced responses. The results demonstrated that DB07565 acts as a potent P2Y₁₄R antagonist with satisfactory antagonistic activity (IC₅₀ = 1.585 nM) (Fig. 12a). Additionally, the cytotoxicity of DB07565 was evaluated in HT-29 cells, revealing low cytotoxicity at the concentration of 100 μM (Fig. 12b).

Fig. 12.

The in vitro biological testing of DB07565. (a) In vitro P2Y₁₄R antagonistic activity of DB07565 (N = 3). (b) The effect of DB07565 on cell viability of HT-29 cells (N = 5).

Based on the in vitro findings, we further investigated the biological activity of DB07565 in DSS-induced experimental colitis, a well-established model of inflammatory bowel disease (IBD), which has been reported to benefit from pharmacological antagonism of P2Y₁₄R (Fig. 13a). The results demonstrated that treatment with DB07565 significantly attenuated DSS-induced colitis, as evidenced by reduced weight loss, alleviated diarrhea, decreased rectal bleeding, and prevented colon shortening caused by DSS administration (Fig. 13b). Consistent with body weight and disease activity index (DAI) data, histopathological analysis revealed that DB07565 mitigated DSS-induced tissue damage and inflammatory cell infiltration in the colon (Fig. 13c and d).

Fig. 13.

The biological testing of DB07565 in DSS-induced experimental colitis model in mice. (a) Experimental flow chart. (b) Body weight and disease activity index evaluation of changes in mice during the disease process (N = 6). (c) The H&E staining in the colon tissues of DSS-treated mice (Scale bar = 100 μm). (d) The length of colons of mice 7 days after DSS treatment (N = 6). ^###P < 0.001 vs Control group, *P < 0.05, ^**P < 0.01, ^***P < 0.001 vs Model group.

Furthermore, immunofluorescence staining of tight junction proteins (Claudin-1, Occludin, and ZO-1) confirmed that DB07565 treatment markedly enhanced gut barrier function (Fig. 14). Collectively, these findings suggest that DB07565 ameliorates DSS-induced colitis by targeting P2Y₁₄R and protecting mucosal barrier integrity.

Fig. 14.

The influence of DB07565 on intestinal barrier function. Immunofluorescent images of colon tissues stained with Claudin-1, ZO-1 and Occludin, the principal components of tight junction (Scale bar = 50 μm).

Conclusion

The P2Y₁₄ receptor (P2Y₁₄R) has been identified as a key therapeutic target in purinergic signaling due to its significant pathophysiological roles across various disorders. To overcome challenges in drug development, we employed an integrative computational strategy that combined structure-based virtual screening with the repurposing of FDA-approved and experimental compounds. This approach identified four candidate P2Y₁₄R antagonists, among which Chloramphenicol succinate (DB07565) exhibited superior binding characteristics validated through multi-parametric analyses involving microsecond-scale molecular dynamics simulations and MM/GBSA binding free energy analyses. Mechanistic investigations uncovered a conserved K77-Y102-H184-K277 pharmacophore essential for receptor antagonism. Biological validation confirmed DB07565′s potent inhibition of P2Y₁₄R with negligible cytotoxicity at 100 µM in vitro. In murine colitis models, DB07565 demonstrated dose-dependent therapeutic efficacy comparable to first-line inflammatory bowel disease (IBD) treatments, achieving substantial symptom alleviation even at low doses. As a prodrug of chloramphenicol, DB07565 exhibits myelosuppressive toxicity after being metabolized into chloramphenicol in the body, thereby posing potential risks in clinical applications. In this study, we found that the prodrug form of DB07565 exerts an inhibitory effect on P2Y₁₄R, with improved ADME/T properties and a distinct chemotype. To circumvent systemic conversion, we proposed direct rectal administration of DB07565. Notably, previous studies have demonstrated its excellent stability in intestinal fluid, with less than 10 % degradation observed within 60 min, thus providing a pharmacokinetic rationale for its rectal delivery [86]. This study not only identifies a clinically translatable P2Y₁₄R antagonist for IBD treatment but also illustrates how computational repurposing can accelerate therapeutic discovery by circumventing early-stage development obstacles. Our findings contribute to GPCR-targeted drug design and establish a framework for developing next-generation anti-inflammatory therapies via structure-guided drug discovery.

Data availability

The relevant contents supporting the findings of this study are available in the supplementary material of this paper.

Compliance with Ethics Requirements

All animal experiments were conducted according to the animal ethics requirements of China Pharmaceutical University (Ethics Approval Number: 2022-01-037).

Author contributions

ST., KW., and CXL contributed equally to this work; Conceptualization: ST., QHH., XTK., and HQL.; Methodology: KW., and CXL.; Analysis: KW., CXL., XYC., ZDZ., QS., and SFR.; Writing: KW., and CXL.; Supervision: ST., QHH., HQL., and XTK. All authors have read and agreed to the published version of the manuscript.

Funding

This study was supported by the National Natural Science Foundation of China (82373725), the Priority Academic Program Development of the Jiangsu Higher Education Institutes (PAPD), the Science and Technology Program of Suzhou (Grant Nos. SKY2023127 and SKY2022103), Shandong Laboratory Program (SYS202205) and China Postdoctoral Science Foundation, No. 76 General Fund (2024M763660) .

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: