From in-silico QSAR modeling to in-vitro MTT assay: experimental validation of novel uPAR leads for triple-negative breast cancer (TNBC) and skin cancer

Abstract

The urokinase plasminogen activator receptor (uPAR) is crucial in processes such as tumor invasion, epithelial–mesenchymal transition, and the metastatic spread of aggressive cancers, including triple-negative breast cancer (MDA-MB-231) and skin cancer (A431). Due to its overexpression and key role in regulating extracellular matrix degradation and cell migration, uPAR stands out as a promising but underutilized therapeutic target. This research combines computational modeling with experimental validation to discover new small-molecule inhibitors of uPAR. A comprehensive QSAR model was developed utilizing 816 structurally diverse uPAR antagonists, resulting in high internal predictivity (R² = 0.84) and external validation accuracy (R²_ext = 0.8014). Key molecular descriptors that influence inhibition, such as com_Nminus_2A, lipo_S_1Ac, fHC3B, and fdonringC7A, have been identified as critical factors in determining electrostatic and steric complementarity. The virtual screening of the ChemDiv database using QSAR methods identified two lead candidates: D685-0061 and C878-1660. Molecular docking analysis demonstrated that C878-1660 interacts with uPAR through advantageous hydrophobic and hydrogen-bond interactions, while D685-0061 displayed relatively weaker binding affinity. The ligand, C878-1660 formed a stable complex (RMSD ~ 1.5 Å over 500 ns), whereas D685-0061 showed higher flexibility (RMSD ~ 2.7 Å). The cytotoxic effects of both ligands were quantified using in vitro MTT assays. D685-0061 demonstrated a higher potency in MDA-MB-231 cells, with an IC₅₀ of 21.34 µM, in contrast to C878-1660, which had an IC₅₀ of 81.82 µM. Conversely, C878-1660 showed greater effectiveness in A431 cells, exhibiting an IC₅₀ of 18.93 µM compared to D685-0061’s IC₅₀ of 28.34 µM. The observed apoptotic morphology was consistent with a dose-dependent relationship, supporting these findings. This study presents a thorough computational-experimental pipeline that identifies two promising lead molecules targeting uPAR and provides mechanistic insights into their binding and cytotoxic profiles. The gathered evidence substantiates the need for further optimization and progression toward preclinical evaluation for uPAR-driven cancers.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-36406-4.

Introduction

Global cancer statistics for 2023 reveal an estimated 20.0 million new cancer cases and 9.7 million cancer deaths worldwide. This data highlights a persistent increase in the disease burden, influenced by factors such as population aging and lifestyle-related risks. Breast cancer is the most frequently diagnosed malignancy worldwide, highlighting the critical necessity for focused therapeutic strategies aimed at aggressive forms like TNBC¹. Cancer progression is a major contributor to morbidity and death globally, with metastasis being one of the most fatal components of the illness^2–5. Despite significant breakthroughs in therapies for primary tumours, progress in the management of metastatic malignancies has been constrained, necessitating the development of novel therapeutic strategies to inhibit the dissemination of cancer cells^6–9. The urokinase-type plasminogen activator receptor (uPAR) is a pivotal component in cancer metastasis, functioning as a membrane-bound receptor that, alongside its ligand urokinase-type plasminogen activator (uPA), is essential in orchestrating the mechanisms that enable cancer cell migration and invasion^10–15. In the last thirty years, uPAR and its related protease uPA have been thoroughly investigated for their contributions to tumor progression and metastasis, as uPA catalyses the transformation of plasminogen into plasmin, an enzyme that aids in the degradation of the extracellular matrix (ECM) and enhances the invasive potential of cancer cells^16–21. The link between uPA and uPAR facilitates many phases of metastasis by permitting tumor cells to destroy extracellular matrix components, hence facilitating invasion of adjacent tissues and dissemination to distant organs^{15,17,20,22–26}. In cancer, uPAR has been associated with many stages of tumor growth, including angiogenesis, immunological regulation, and invasion²⁷. In addition, elevated uPAR expression has been associated with worse prognosis in several malignancies, including breast, colorectal, and prostate cancers^26,28–31. In triple-negative breast cancer (TNBC), a very aggressive and challenging cancer subtype, uPAR expression is significantly heightened, associating with accelerated tumor growth, increased invasion, and metastatic dissemination. uPAR has a role in the control of MMPs, which are essential for ECM breakdown, hence promoting cancer cell migration and the formation of secondary tumours^32–35. Furthermore, uPAR’s participation in epithelial-to-mesenchymal transition (EMT), an essential mechanism in cancer metastasis, has been noted in TNBC, highlighting its significance in the aggressive characteristics of this cancer subtype^36–39. In skin cancer, especially melanoma and non-melanoma types, uPAR is often overexpressed, and its increased levels correlate with tumor aggressiveness. In these malignancies, uPAR promotes ECM breakdown, allowing cancer cells to penetrate and spread^40,41. Moreover, uPAR’s role in immune regulation and inflammatory responses within the tumor microenvironment promotes tumor proliferation and treatment resistance^42–46. This illustrates the significant and essential function of uPAR in facilitating cancer spread and modulating immune responses, hence highlighting its potential as a therapeutic target^46–49. uPAR inhibitors encounter challenges in clinical advancement stemming from inadequate tumor penetration, unintended off-target interactions, and sequestration by soluble uPAR, all of which diminish effective receptor engagement. Furthermore, the presence of compensatory protease networks, along with the lack of validated biomarkers, significantly reduces the effectiveness of therapies in aggressive cancers^50–52. To tackle the difficulty of targeting uPAR in cancer therapy, in-silico approaches, including Quantitative Structure-Activity Relationship (QSAR) modeling^53–55, have emerged as essential tools in the identification of new uPAR inhibitors. QSAR modeling formulates a mathematical correlation between the chemical structure of drugs and their biological activity, allowing researchers to anticipate the inhibitory efficacy of novel compounds against uPAR^56–61. QSAR screening quickly identifies potential uPAR inhibitors, while docking and molecular dynamics elucidate binding relationships and dynamic stability, enabling rational design for aggressive tumors including TNBC and skin cancers^62,63. MD simulations provide a dynamic perspective on protein-ligand interactions, supplying researchers with insights into the flexibility and stability of binding interactions^{60,62,64–68}. PCA identifies conformational changes in uPAR induced by inhibitors, whereas FEL analysis delineates the energetic stability of the complex, highlighting favorable binding states. The use of complementary MTT assays in MDA-MB-231 and A431 cells supports these predictions by measuring cytotoxicity, highlighting the importance of selectively inhibiting uPAR-driven invasion and metastatic progression for therapeutic purposes. The integration of PCA with free-energy landscape analysis provides a systematic approach to delineate conformational dynamics and assess thermodynamic stability in uPAR–inhibitor complexes. PCA identifies the primary movements triggered by ligand interaction, whereas energy-landscape mapping uncovers metastable states and favored binding regions. The integration of QSAR modeling, docking, molecular dynamics, and experimental assays enhances the discovery process of selective uPAR antagonists. This approach establishes a coherent framework for the development of agents aimed at restricting metastatic dissemination in aggressive cancers.

Methodology

QSAR model construction

The reliability of QSAR is contingent upon the availability of high-quality, pertinent data and a comprehensive understanding of the fundamental biological or chemical systems involved. A thorough literature review and meticulous dataset curation are crucial components of the research process. In modern QSAR research, the validity and predictive performance of a model are collectively determined by the use of appropriate descriptors, robust statistical methods, and sound training & test set partitioning, all of which align with OECD principles^69–71. Computational evaluations were accompanied using a Lenovo laptop equipped with an Intel^® Core™ i3-2328 M CPU (2.20 GHz), 12 GB RAM, and a 64-bit operating system. Molecular structures were drawn by using Marvin Sketch, and geometry optimization was conducted using OpenBabel 3.1. Virtual screening was conducted using AutoDock Vina 1.2.0 in conjunction with the NRG Suite. Protein-ligand interactions were examined with Discovery Studio 2021 Client. Molecular descriptors were computed with PyDescriptor, a plugin for PyMOL. The 2D-QSAR model was constructed and verified using QSARINS 2.2.4, while data analysis was conducted using Microsoft Excel 2016.

Dataset collection and optimization

A large, diverse, and high-quality dataset is crucial to the success and use of QSAR and molecular docking investigations in pharmaceutical development. The model’s validity was confirmed by several steps designed to systematically eliminate human-introduced mistakes and biases, guaranteeing its correctness and dependability⁷⁰. A comprehensive search of BindingDB (accessed on February 11, 2023; https://www.bindingdb.org) yielded 1,197 known uPAR antagonists with experimentally validated IC₅₀ values (in nM). Structural data were retrieved in SMILES format. After filtering out duplicates, salts, metal-containing compounds, rule-of-five violators, and entries lacking IC₅₀ values, 516 ligands were selected for further analysis (see Table 1)⁷². The curated collection encompasses a diverse array of chemical variations, including positional and chain isomers, heterocyclic and aromatic frameworks, stereoisomers, and more structural alterations. IC₅₀ measurements demonstrated a wide dynamic range from 0.001 nM to 78,000 nM encompassing four orders of magnitude. IC₅₀ data were converted into pIC₅₀ (− log₁₀IC₅₀) to standardize biological activity for QSAR modeling. Table 1 displays the five most active and five least active compounds according to experimental IC₅₀ values, while comprehensive details including BindingDB IDs, SMILES, IC₅₀. For structural modeling, each SMILES was precisely transformed into its 3D-optimized structures using OpenBabel 3.1 software package, producing SDF files for further analysis⁷¹. Subsequently, the PM3 technique was utilized, which is a semiempirical methodology that incorporates the same structural framework and computational processes as the Austin Model-1 (AM1) approach. The goal was to convert the SDF file format into the MOL2 format to facilitate structural optimization and the assignment of partial charges utilizing MOPAC 2016⁷³.

Table 1.

This dataset’s ten most active and ten least active compounds, along with their experimentally obtained Ki (nM) and pKi (M) values, reflect each of the 10 most and 10 least active compounds, respectively.

Mol Id.	Smiles Notations	Ki in nM	pKi in M
1	NCc1ccc(NC(= O)c2cc(Nc3ncccn3)c3cc(ccc3c2)C(N) = N)cc1	0.62	9.20
2	NC(= N)c1ccc2cc(cc(Nc3ncccn3)c2c1)C(= O)Nc1ccccc1	2	8.69
3	NC(= N)c1ccc2cc(cc(Nc3ncccn3)c2c1)C1CC1c1ccccc1	5	8.30
4	CCC1CNCc2ccc(NC(= O)c3ccc4cc(ccc4c3)C(N) = N)cc12	6.3	8.20
5	CC[C@H](C)[C@@H]1NC(= O)[C@H](Cc2ccc(O)cc2)NC(= O)[C@H](CCCNC(N) = N)NC(= O)[C@H](CO)NC(= O)[C@H](Cc2ccc(O)cc2)NC(= O)[C@H](C)NC(= O)[C@@H]2CCCN2C(= O)[C@@H](N)CSSC[C@H](NC(= O)CNC1 = O)C(O) = O	6.8	8.16
6	C[C@H](NC(= O)[C@@H](CO)NS(= O)(= O)Cc1ccccc1)C(= O)NCc1ccc(cc1)C(N) = N	7.7	8.11
7	NC(= N)c1ccc2[nH]c(cc2c1)-c1cccc(-c2ccccc2)c1O	8	8.09
8	NC(=[NH2+])c1ccc2[nH]c(cc2c1)-c1cccc(-c2ccccc2)c1[O-]	8	8.09
9	NC(=[NH2+])c1cc2cc([nH]c2cc1Cl)-c1cccc(-c2ccccc2)c1O	9	8.04
10	NCc1ccc(NC(= O)c2cc(Nc3ncccn3)c3cc(ccc3c2)C(N) = N)cc1	0.62	9.20
11	Cc1ccc(cc1)S(= O)(= O)NC(Cc1cccc(c1)C(N)N)C(= O)N1CCCCC1	55,000	4.26
12	Nc1ccc2cccc(O)c2n1	56,000	4.25
13	NC(N) = NCCC(= O)N1CCN(CC1)C(= O)[C@H](Cc1cccc(c1)C(N) = N)NS(= O)(= O)CCc1cccc2ccccc12	56,000	4.25
14	CC1CCN(CC1)C(= O)C(Cc1cccc(c1)C(N) = N)NC(= O)OC(C)(C)C	56,000	4.25
15	Cc1c(C)c(c(C)c2CCC(C)(C)Oc12)S(= O)(= O)N[C@H](Cc1cccc(c1)C(N) = N)C(= O)N1CCN(CC1)S(C)(= O) = O	58,000	4.23
16	NC(= N)c1cccc(CC(NS(= O)(= O)c2ccc3ccccc3c2)C(= O)N2CCOCC2)c1	58,000	4.23
17	Cc1cccc(C(= O)Nc2ccc(cc2)N = C(N)N)c1O	60,000	4.22
18	OC[C@@H](NS(= O)(= O)Cc1ccccc1)C(= O)NCC(= O)NCc1ccc(cc1)C(= N)NO	61,000	4.21
19	COc1cccnc1N = C(N)N	62,000	4.20
20	NC(=[NH2+])c1cc2cccnc2s1	63,000	4.20

Normalization of data, computation of molecular descriptors, and objective feature selection (OFS)

PyDescriptor was used to do computations and produced more than 58,000 chemical descriptors for each molecule⁷⁴. The application can generate 58,000 chemical descriptors, including atomic fragments, pharmacophoric patterns, and fingerprints, from 1D to 3D. PyMOL, a popular visualization tool, now includes a Python plugin. Since PyDescriptor is open-source, it works on Windows, Linux, and macOS. The objective feature selection (OFS) technique in QSARINS 2.2.4 eliminated chemical descriptors that were virtually the same in 90% of compounds and extremely closely related (R > 0.90)⁷⁵. The range of molecular descriptors was adequately covered by merging fingerprint, charge-based, 1D to 3D, and atom-pair descriptors, even if the number was reduced to 1081.

Data pretreatment, splitting the data set and subjective feature selection (SFS)

Data pretreatment is essential for removing constant and inter-correlated descriptors. We employed an inter-correlation variance cut-off value of 0.85 and a constant descriptor variance cut-off value of 0.001 to guide our operations. To create a good QSAR model, molecular descriptors must be selected using stepwise regression or genetic algorithms. The dataset was divided into two sets: 80% (328 molecules) for training and 20% (82 molecules) for prediction, ensuring reliable model training and validation⁷⁶.The random split of the dataset was used, with 80% of the dataset being allocated for training and 20% being allocated for testing⁷⁷. The training set was the only source of information utilized to discover the appropriate chemical descriptor combination and number. A new model was evaluated using the external validation set, also known as the prediction set. It is essential to include a sufficient number of chemical descriptors into the model in order to prevent both overfitting and underfitting from occurring when using the model. Thus, a simple graphical method was employed to calculate the appropriate model descriptor number. In MLR models with a variable, or molecular descriptor, the cross-validated coefficient of determination for leave-one-out (Q^2LOO) frequently rises until it reaches its maximum value 36, 55. Q²_LOO then rises somewhat⁷⁸.

Building regression model

After a thorough evaluation of various combinations using rigorous statistical criteria, six optimised molecular descriptors were identified. In accordance with OECD-compliant modelling principles, QSARINS established a transparent methodology, established applicability, and implemented rigorous performance metrics to guarantee the development of a reliable and mechanistically interpretable predictive model⁷⁹. A genetic algorithm was employed to systematically investigate broad chemical and feature space and pinpoint essential drug-related attributes, thereby reducing the likelihood of converging to local optima. The process initiates with the creation of randomly generated chromosomes, which are then assessed using a fitness function to facilitate iterative optimisation. The following is the definition of the Friedman LOF function, which is often used as the fitness function⁸⁰.

SSE is sum of squared errors. The variable c represents the number of fundamental functions without the constant term. Smoothness factor d. Feature count (p) and data points (n) are different variables in the model. The R² error decreases when the regression model contains more terms, while the LOF measure grows. A model is not over-fit since the length of fit (LOF) measure discourages reckless word addition.

Internal validation

The QSAR model’s accuracy was internally validated using cross-validation. The present research employed leave-one-out cross-validation for internal validation. The equation for cross-validated Q²CV is⁸¹.

Y represents pKi, the observed value. Ypred is the model’s prediction, while Ymean is the training set’s average. A model’s quality is assessed by measuring the training set’s squared correlation coefficient R². Its reliability is restricted by descriptor count. This equation calculates R²adj, a new parameter, to fix the issue⁸².

n is the number of chemicals in the training set, and p is the number of statistical modeling terms that describe those chemicals. The gap between R² and R²adj should be 0.3. All of these things allow the QSAR model to be internally validated. You should have R²tr, Q²LOO, Q²LMO, and R²ex values that are greater than or equal to 0.6. You should also have a CCC value that is greater than or equal to 0.6. Q²-Fn should be more than 0.60⁸³.

External validation

An external validation evaluated the model’s capacity for predicting novel compounds. The process included applying the model equation derived from the training set to a prediction dataset, excluding chemicals not incorporated in the calculation model. The evaluation of the model’s performance was conducted using several measures. The external validation parameters R²ex, Q²F1, Q²F2, and Q²F³ demonstrated elevated values, whereas R²Yscr (the coefficient of determination for Y-randomization), RMSE (Root Mean Square Error), MAE (Mean Absolute Error), RMSEtr (which was inferior to RMSEcv), and the Golbraikh and Tropsha methodologies exhibited diminished values⁸⁴.

Y-randomization test

The random generation of the variable Y is an alternative approach used to assess the model’s robustness. This method involves creating new MLR models by altering the values of the Y variable while keeping the descriptor values same. Through the repeated rearrangement of ‘n’ values, several models are constructed; the R² and Q² metrics of each model are then compared to those of the originally produced models. Lower values of these parameters are favored for the creation of a resilient model⁸⁵.

Applicability domain

To assess the QSAR model, its applicability domain must be identified. The application domain (AD) of a classification or regression model is crucial in chemometrics and QSAR investigations. Despite a solid and tested QSAR model, any chemical compounds may have erroneous predicted characteristic predictions. The models’ extrapolations should be ignored, but AD chemical predictions are reliable. This research employed the Williams plot to evaluate model applicability domains and detect substances. The QSAR model’s applicability domain was determined by plotting standardized residuals versus hat values. QSARINS also uses leverage and model predictions to define the model’s application area in a novel way. If molecules do not respond experimentally to the structural applicability domain, the Insubria graphic is required for placement evaluation. It also lets them compare predictions to experimental molecules⁸⁶.

QSAR based virtual screening

In the virtual screening process employing QSAR methodologies, a protease library was utilized, sourced from the Chemdiv database, which contains 30,000 compounds (https://www.chemdiv.com/catalog/focused-and-targeted-libraries/Protease-Library/).Prior to the calculation of molecular descriptors, the SMILES notations were converted into 3D-optimized structures. Subsequently, all 3D molecular structures were organized into a modeling set. The molecular descriptors were calculated for protease library, encompassing a total of 30,000 protease inhibitors. Furthermore, a total of 30,000 protease inhibitor compounds were subjected to molecular docking via virtual screening utilizing the NRG Suite program. Following validation through ligand-based (QSAR) screening, we selected the compounds exhibiting the highest docking scores (measured in kcal/mol) for subsequent analysis.

Molecular Docking

The hit molecules displaying exceptional results in both virtual screening and QSAR modeling were selected for molecular docking study. To conduct a more thorough comparison and verify that the crystal structure was compatible with the binding pocket of the UPAR protein (PDB:1W11) was selected which showed close resemble to ligand. The structure of UPAR was sourced from the Protein Data Bank, specifically from the PDB file available at https://www.rcsb.org/structure/1W11. Utilizing the UCSF Chimera Tools program, we successfully optimize the proteins and ligands. The UCSF Chimera employed the method with the highest gradient to ascertain 1000 steps. The uPAR protein structure PDB:1W11 was selected because it provides a high-resolution crystallographic model with a co-crystallized native ligand, enabling precise identification of the active binding pocket for docking and MD simulations. The natural ligand phenethylsulfonamidino(P4)-d-seryl(P3)-l-alanyl(P2)-l-argininal(P1) was identified. The initial and final X-rays in the dataset are stored in the pdb:1W11 file. Upon optimization, the protein can be utilized in docking studies and Ramachandran plots were generated prior to and following the optimization process. Molecular docking experiments were conducted using the NRG Suite software. In its capacity as a PyMOL plugin, this open-source application is accessible via the website www.pymol.org. To identify protein surface cavities that can subsequently function as target binding sites, FlexAID may be implemented in docking simulations^87,88. Furthermore, it utilizes a grading system based on surface complementarity that is not heavily dependent on particular geometric factors. The energy settings for the scoring system were developed by studying a wide range of natural and similar shapes, which showed differences of less than two for over 1500 complexes in the PDB binding database that were seen as good examples. The solution to these problems involved many iterations of Monte Carlo optimization on more complex datasets (lower energy decoys) with RMSD values above two^74,89. Genetic algorithms was employed to achieve conformational retrieval, which also simulates covalent binding and ligand and side chain flexibility. The default parameters utilized for the implementation of a docking technique, incorporating both flexibility and rigidity, were aimed at optimizing performance with NRGSuite. The docking outcomes were presented utilizing the Biovia Discovery Studio program. The NRG suite utilizes genetic algorithms to enhance the retrieval of molecular conformations. Additionally, it encompasses models for the adaptability of ligands and side chains, along with covalent docking. To enhance the efficiency of NRGSuite, we utilized a docking methodology that is both adaptable and stringent, with the subsequent default parameters delineated: This analysis utilizes side chain rigidity with spherical binding sites, characterized by a radius of 5 Å. The HET group encapsulates water molecules without restriction. Van der Waals’s magnetic permeability was 0.1, and the kind of solvent is unclear. There exist 1,000 chromosomes, 1,000 generations, a fitness model of sharing, a reproductive model of population expansion, and five TOP complexes. A recognized inhibitor from PDB 1w11 was utilized to confirm the molecular docking methodology.

Molecular dynamics (MD) simulations

Molecular dynamics and simulation (MDS) methods were used to study how stable and consistent the interaction is between C878-1660, D685-0061, and UPAR (PDB:1w11). This work evaluated the stability of a docking complex comprising C878-1660 and D685-0061 with UPAR (PDB: 1w11) and apo-UPAR (PDB: 1w11) during a duration of 500 ns. The system employed the OPLS-2005 force field and incorporated an explicit solvent model using SPC water molecules⁹⁰. In addition, the force field (OPLS4), water model (TIP3P), and convergence criteria (RMSD plateau < 0.5 Å after 500 ns) are used. The fundamental parameters for the orthorhombic box of the explicit SPC water model, measuring 3.0 × 3.0 × 3.0 m, were determined by Desmond. 2018-04⁹¹. The neutralization of complexes was accomplished by the addition of NaCl at a concentration of 0.15 M Na + ions. The molecular dynamics simulation used 18 sodium (Na) ions and 15 chloride (Cl) ions. During the simulation run, we maintained constant temperature and pressure (NPT) and conducted a molecular dynamics simulation (MDS) for 500 ns. The Nose-Hoover chain method facilitated the formation of the NPT ensemble, and the concluding simulation was conducted at 300 K with a relaxation duration of 1 ps over the whole dynamics⁸⁵. The pressure was regulated by a barostat, and was constructed using the Martyna Tuckerman-Klein system for chain coupling, with a relaxation time of only 2 picoseconds⁸⁶. The pressure was maintained at 1 atmosphere and the temperature at 300 K by the isotropic Martyna-Tobias-Klein barostat and the Nose-Hoover thermostat in the Desmond simulation. We employed the NPT ensemble in all cycles, maintaining a temperature of 300 K and a pressure of 1 bar. By employing the RESPA integrator and a time step of 2 femtoseconds, we were able to accurately estimate the bonding interactions. We figured out the long-range electrostatic interactions between the particles by using the particle mesh Ewald method and setting a radius of 9 for Coulomb interactions⁸⁶. The data obtained was used to calculate, RMSD, RMSF, the mean binding energies and standard deviations.

In-Vitro evaluation of anticancer activity by MTT assay (MDA-MD-231)

The rationale for selecting MDA-MB-231 cell lines in MTT assays to study the activity of a ligand targeting the urokinase plasminogen activator receptor (uPAR) stems from the biological and clinical relevance of both the cell line and the uPAR system in aggressive breast cancer³⁵. MDA-MB-231 is a triple-negative breast cancer (TNBC) cell line that lacks estrogen receptor (ER), progesterone receptor (PR), and HER2 expression, rendering it resistant to most conventional targeted therapies⁴⁶. This cell line exhibits high basal levels of uPAR, a cell surface glycoprotein that binds urokinase-type plasminogen activator (uPA) and plays a pivotal role in cancer cell invasion, metastasis, and survival by facilitating pericellular proteolysis and activating downstream signaling cascades such as Ras-MAPK and ERK pathways³⁵. The uPA/uPAR interaction is significantly upregulated in MDA-MB-231 cells and is tightly linked to the cell’s invasive and metastatic capabilities, making these cells an ideal in vitro model for evaluating inhibitors that block this interaction.

Procedure

On Day 1, culture flasks containing 80–90% confluent cells were processed by removing the media, followed by two washes with 1 ml of phosphate-buffered saline (PBS). Cells were then trypsinized with 1 ml of Trypsin-EDTA solution and incubated at 37 °C in a 5% CO₂ environment for 7–8 min until detachment was observed. The flask was gently tapped to facilitate complete cell detachment, and 1 ml of the cell suspension was aliquoted into three 1.5 ml microcentrifuge tubes. The cells were centrifuged at 500 × g for 10 min at 25 °C, the supernatant was carefully removed, and the pellet was resuspended in 1 ml of 1× PBS to disperse cell clumps. A 10 µl aliquot was taken for cell counting, after which the cells were centrifuged again at 200 × g for 10 min at room temperature. The supernatant was discarded, and the pellet was gently resuspended in 1 ml of complete media. Based on the cell count, 5,000 to 10,000 cells were seeded per well in a 96-well plate, ensuring homogeneous suspension by continuous gentle inversion; volumes were adjusted to 100 µl with complete media before incubation at 37 °C for 24 h. On Day 2, the media was removed, and the cells were treated with the test compounds; D685-0061, and C878-1660 (ligands identified in QSAR Based virtual screening and molecular docking based virtual screening) at varying concentrations (5, 10, 20, 40, 80, 160, and 320 µM concentrations were used for both test compounds) in triplicate wells, with a corresponding untreated control set. The total volume in each well was adjusted to 300 µl using complete media, followed by incubation at 37 °C in a CO₂ incubator for 24 h. On Day 3, an MTT assay was conducted by preparing a fresh 5 mg/ml MTT solution in PBS, sterile-filtered through a 0.22 μm filter and stored protected from light. Twenty-five microliters of the MTT solution (final concentration 0.2–0.5 mg/ml) were added to each well and incubated at 37 °C for 2–3 h to allow for formazan crystal formation. Following incubation, media was aspirated, and 100 µl of DMSO was added to dissolve the crystals, mixed gently by pipetting to avoid bubble formation, and incubated overnight at 37 °C. Absorbance was measured at 570 nm to quantify cell viability. The MTT assays were conducted in triplicate wells per condition, and all experiments were independently repeated at least three times. (See supplementary file S4 for detail procedure of MTT Assay)

MTT assay procedure (A431 skin cancer cell lines)

The A431 human epidermoid carcinoma cell line is extensively utilized as an in vitro model for investigating urokinase plasminogen activator receptor (uPAR) function and inhibition in skin cancer due to its inherent overexpression of uPAR, which plays a pivotal role in extracellular matrix degradation, tumor cell invasion, and metastasis. This rationale is supported by multiple studies demonstrating A431’s suitability for dissecting uPAR signaling in the context of skin cancer progression⁹² and UVB-induced uPAR regulation⁹³.

Procedure

On Day 1, cells at approximately 80–90% confluency were prepared by aspirating the culture medium and washing twice with 1 ml phosphate-buffered saline (PBS) to remove residual serum proteins. Cells were then enzymatically detached by adding 1 ml of Trypsin-EDTA solution and incubating the flask at 37 °C in a humidified atmosphere containing 5% CO₂ for 7–8 min, facilitating detachment through proteolytic cleavage of cell adhesion molecules. Complete cell detachment was confirmed visually by observing cell rounding, followed by gentle mechanical agitation of the flask. The cell suspension was transferred into 1.5 ml microcentrifuge tubes and centrifuged at 500 × g for 10 min at room temperature to pellet cells. The supernatant was carefully removed to avoid disturbing the pellet, which was resuspended in 1 ml 1X PBS to dissociate cell aggregates. After gentle mixing, 10 µl of the cell suspension was used for cell counting. Subsequently, cells were centrifuged at 200 × g for 10 min, supernatant discarded, and the pellet resuspended in complete growth medium. Based on the cell count, an aliquot containing 5,000 to 10,000 cells was seeded into each well of a 96-well plate, ensuring homogeneous distribution by continuous gentle mixing during plating, and the volume was adjusted to 100 µl with complete medium. Plates were incubated at 37 °C with 5% CO₂ for 24 h to allow cell attachment and recovery. On Day 2, culture medium was removed and replaced with media containing the test compounds (D685-0061 and C878-1660) at serial concentrations ranging from 5 to 160 µM respectively, applied in triplicate wells, with untreated wells serving as negative controls; the final volume per well was adjusted to 300 µl, and plates were incubated for an additional 24 h under standard culture conditions. On Day 3, MTT reagent was freshly prepared at 5 mg/ml in PBS, sterile-filtered through a 0.22 μm membrane, and protected from light until use, with the final working concentration adjusted to 0.2–0.5 mg/ml. A volume of 25 µl MTT solution was added to each well and incubated for 2–3 h at 37 °C to allow mitochondrial reduction of MTT to insoluble formazan crystals. Subsequently, the medium was aspirated carefully, and 100 µl of dimethyl sulfoxide (DMSO) was added to solubilize the formazan, mixing gently by pipetting to avoid bubble formation. Plates were incubated overnight at 37 °C, and absorbance was measured at 570 nm to quantify cell viability based on metabolic activity. The MTT assays were conducted in triplicate wells per condition, and all experiments were independently repeated at least three times (See supplementary file S5 for detail procedure for MTT Assay, See supplementary file S6 for MTT assay certificate).

Results and discussion

QSAR model

The regression analysis presented in the QSAR model provides an in-depth evaluation of the relationship between various predictor variables and the dependent variable through their coefficients, standardized coefficients, standard errors, confidence intervals, and p-values (See supplementary material file S1 for detail QSAR results).

pIC₅₀ = 4.211 (± 0.157) + -0.659 (± 0.072) * com_Nminus_2A + -4.505 (± 0.803) * lipo_S_1Ac + 0.413 (± 0.049) * fHC3B + 0.131 (± 0.01) * fdonringC7A + -0.498 (± 0.045) * fsp3Nsp2C6B + 1.162 (± 0.11) * fnotringOsp2N9B + 1.195 (± 0.105) * sp2N_S_5B.

The regression model demonstrates a high level of significance for all predictors, with a p-value of 0. The intercept, quantified at 4.2107, serves as the baseline value in this analysis. Negative correlations were identified for com_Nminus_2A with a value of -0.4375, lipo_S_1Ac at -0.4633, and fsp3Nsp2C6B showing − 0.5399. Conversely, fHC3B presents a positive contribution of 0.422, fdonringC7A at 0.6783, and fnotringOsp2N9B with a value of 0.5369. The predictor sp2N_S_5B exhibits the highest influence, with a value of 0.9711, underscoring its significant effect. The presence of tight confidence intervals, coupled with small standard errors, serves to validate the precision of the measurements obtained. The model exhibits robustness, characterized by significant predictors that indicate both positive and negative effects, while also demonstrating high statistical reliability (See supplementary file S2 for calculated molecular descriptors used in the developed QSAR model).

QSAR validation parameters

The regression analysis yielded an R² of 0.8459 and an adjusted R² of 0.8425, with a minor difference of 0.0034. The lack-of-fit (LOF) was 0.1410, while Kxx was 0.3316, and Delta K was − 0.0043. The root mean square error (RMSE) for training was 0.3596, with a mean absolute error (MAE) of 0.3045. Residual sum of squares (RSS) for training was 42.5333, the concordance correlation coefficient (CCC) for training was 0.9165, standard deviation (s) was 0.3640, and the F-statistic was 251.6706. For internal validation, Q²loo was 0.8386, with an R²-Q²loo difference of 0.0073, RMSE cross-validation (cv) of 0.3680, and MAE cv of 0.3117. The predictive residual sum of squares (PRESS cv) was 44.5466, while CCC cv reached 0.9125. The Q²LMO value was 0.8371, with R²Yscr at 0.0212, Q²Yscr at -0.0288, and RMSE AV Yscr at 0.9061. External validation revealed an RMSE of 0.3766, MAE of 0.3311, and PRESS of 11.6315, with an external R² (R²ext) of 0.8014. The Q² values were Q²-F1: 0.8012, Q²-F2: 0.8008, and Q²-F3: 0.8309, while CCC ext stood at 0.8919. The average r²m was 0.7152, with an r²m delta of 0.1364. The calculated regression angle for external data from the diagonal was − 5.5891°. Leave-one-out (LOO) predictions showed that Exp(x) vs. Pred(y) had an R² of 0.8386, R’²o of 0.8097, k’ of 0.9965, closeness (Clos’) of 0.0345, and r’²m of 0.6960, while Pred(x) vs. Exp(y) had an R² of 0.8386, R²o of 0.8386, k of 0.9999, closeness of 0.0000, and r²m of 0.8356. External predictions using the model equation showed that Exp(x) vs. Pred(y) had an R² of 0.8014, R’²o of 0.7643, k’ of 0.9969, closeness (Clos’) of 0.0463, and r’²m of 0.6470, whereas Pred(x) vs. Exp(y) had an R² of 0.8014, R²o of 0.8009, k of 0.9994, closeness of 0.0006, and r²m of 0.7834 (See Table 2).

Table 2.

Presentation of QSAR results in terms of correlation coefficient.

Variable	Coeff.	Std. Coeff.	Std. Err.	(+/-) co. Int. 95%	P-value
Intercept	4.2107	4.2107	0.0797	0.1568	0
com_Nminus_2A	-0.6594	-0.6594	0.0364	0.0717	0
lipo_S_1Ac	-4.5047	-4.5047	0.4084	0.8034	0
fHC3B	0.4129	0.4129	0.025	0.0493	0
fdonringC7A	0.1306	0.1306	0.0053	0.0104	0
fsp3Nsp2C6B	-0.4982	-0.4982	0.0229	0.0451	0
fnotringOsp2N9B	1.1615	1.1615	0.0558	0.1099	0
sp2N_S_5B	1.1951	1.1951	0.0533	0.1048	0

Correlation matrix

The table presents the correlation matrix among seven variables: com_Nminus_2A, lipo_S_1Ac, fHC3B, fdonringC7A, fsp3Nsp2C6B, fnotringOsp2N9B, and sp2N_S_5B, indicating the strength and direction of their linear relationships. The diagonal elements are all equal to 1, as they represent the self-correlation of each variable. com_Nminus_2A exhibits weak positive correlations with lipo_S_1Ac (0.0908), fHC3B (0.0945), fdonringC7A (0.1144), fnotringOsp2N9B (0.3396), and sp2N_S_5B (0.0907), while it has a slight negative correlation with fsp3Nsp2C6B (-0.0504). Lipo_S_1Ac shows a relatively strong positive correlation with sp2N_S_5B (0.8558), indicating a strong association between these two variables, while it has weaker correlations with fHC3B (0.2423) and com_Nminus_2A (0.0908) (See Table 3).

Table 3.

Presentation of the correlation matrix for the molecular descriptor obtained in the developed QSAR model.

	com_Nminus_2A	lipo_S_1Ac	fHC3B	fdonringC7A	fsp3Nsp2C6B	fnotringOsp2N9B	sp2N_S_5B
com_Nminus_2A	1
lipo_S_1Ac	0.0908	1
fHC3B	0.0945	0.2423	1
fdonringC7A	0.1144	-0.2449	-0.4023	1
fsp3Nsp2C6B	-0.0504	-0.1693	-0.0517	0.3789	1
fnotringOsp2N9B	0.3396	-0.2492	-0.3296	0.2466	-0.1128	1
sp2N_S_5B	0.0907	0.8558	0.3165	-0.3321	-0.1387	-0.295	1

However, it is negatively correlated with fdonringC7A (-0.2449), fsp3Nsp2C6B (-0.1693), and fnotringOsp2N9B (-0.2492). fHC3B has a moderate positive correlation with sp2N_S_5B (0.3165) and weaker positive correlations with com_Nminus_2A (0.0945) and lipo_S_1Ac (0.2423), while it exhibits negative correlations with fdonringC7A (-0.4023), fsp3Nsp2C6B (-0.0517), and fnotringOsp2N9B (-0.3296). fdonringC7A is positively correlated with fsp3Nsp2C6B (0.3789) and fnotringOsp2N9B (0.2466) but negatively correlated with sp2N_S_5B (-0.3321), fHC3B (-0.4023), and lipo_S_1Ac (-0.2449). fsp3Nsp2C6B has a moderate positive correlation with fdonringC7A (0.3789) but weak negative correlations with lipo_S_1Ac (-0.1693), fHC3B (-0.0517), fnotringOsp2N9B (-0.1128), and sp2N_S_5B (-0.1387). fnotringOsp2N9B is positively correlated with com_Nminus_2A (0.3396) and fdonringC7A (0.2466) but negatively correlated with lipo_S_1Ac (-0.2492), fHC3B (-0.3296), fsp3Nsp2C6B (-0.1128), and sp2N_S_5B (-0.295). sp2N_S_5B, which has the strongest correlation with lipo_S_1Ac (0.8558), is weakly positively correlated with fHC3B (0.3165) and com_Nminus_2A (0.0907) but negatively correlated with fdonringC7A (-0.3321), fsp3Nsp2C6B (-0.1387), and fnotringOsp2N9B (-0.295). The correlations suggest that lipo_S_1Ac and sp2N_S_5B are strongly linked, while fdonringC7A and fHC3B share a notable negative correlation. These relationships indicate potential dependencies among the variables, which could be useful for predictive modelling or understanding underlying patterns in the dataset (See Table 4).

Table 4.

Portrayal of QSAR results comprising experimentally determined Ki, predicted Ki, and residual values for first 20 molecules only.

ID	Name	Status	Exp. endpoint	Pred. by model eq.	Pred.Mod.Eq.Res.	Pred. LOO	Pred. LOO Res.	HAT i/i (h*=0.0729)	Std.Pred.Mod.Eq. Res.	Std.Pred.LOO Res.
1	25,097	Prediction	5.95	5.75	-0.20	-	-	0.02	-0.57	-0.57
2	25,099	Training	6.05	6.54	0.48	6.54	0.49	0.01	1.34	1.35
3	25,100	Training	5	5.75	0.75	5.76	0.76	0.02	2.08	2.14
4	25,104	Training	5.30	4.99	-0.30	4.98	-0.31	0.02	-0.85	-0.87
5	25,106	Training	6.12	5.75	-0.37	5.74	-0.38	0.02	-1.04	-1.06
6	25,512	Prediction	6.69	6.54	-0.15	-	-	0.01	-0.43	-0.43
7	25,514	Prediction	6.95	6.82	-0.13	-	-	0.01	-0.37	-0.37
8	25,515	Training	7.95	7.32	-0.63	7.31	-0.64	0.01	-1.76	-1.79
9	25,526	Training	6.53	7.19	0.65	7.20	0.66	0.01	1.81	1.84
10	25,527	Training	6.74	6.82	0.07	6.82	0.07	0.01	0.21	0.21
11	25,528	Training	6.60	6.53	-0.06	6.53	-0.06	0.02	-0.18	-0.19
12	25,532	Training	6.31	6.82	0.51	6.83	0.52	0.01	1.42	1.44
13	25,717	Prediction	5.63	5.66	0.02	-	-	0.02	0.07	0.07
14	26,057	Prediction	5.22	5.75	0.52	-	-	0.02	1.47	1.47
15	26,060	Training	5.29	5.75	0.45	5.76	0.47	0.02	1.27	1.30
16	26,064	Training	7.19	6.95	-0.24	6.94	-0.24	0.01	-0.66	-0.67
17	26,065	Training	6.55	6.82	0.27	6.82	0.27	0.01	0.74	0.76
18	26,066	Training	7.02	6.95	-0.06	6.95	-0.06	0.01	-0.18	-0.19
19	26,067	Training	7.30	6.95	-0.34	6.94	-0.35	0.01	-0.96	-0.97
20	26,068	Training	5.02	5.75	0.72	5.76	0.74	0.02	2.02	2.07

Graphical analysis

The scatter plot compares experimental and model-predicted values of a physicochemical variable (See Fig. 1). Training and prediction datasets show strong alignment along the y = x line, indicating robust accuracy. Minor deviations and outliers reveal residual variance and instances of reduced predictive performance across the evaluated range, highlighting both fit and limitations.

Fig. 1.

Presentation of Scatter plot displaying the experimental end point(activity) and predicted end point(activity) by QSAR model.

Molecular descriptors such as com_Nminus_2A, lipo_S_1Ac, and fsp3Nsp2C6B serve to quantify structural features that provide insights into model performance. Residual analysis presents the predicted values plotted against errors ranging from − 1 to 1, utilizing dataset-coded points (see Fig. 2). The random dispersion observed around the zero line suggests a satisfactory distribution of errors, which in turn reinforces the validity of the model while reducing the likelihood of overfitting or biassed predictions.

Fig. 2.

Portrayal of residual plot for the Pred. endpoint vs. Residuals.

The residuals are concentrated around zero, suggesting that the model performs consistently, although there is slight heteroscedasticity observed at higher prediction levels. The absence of directional patterns indicates a low level of bias. Descriptor labels indicate the variables that are utilised. The Williams plot displays standardised residuals that indicate the presence of several distant outliers, which are influential points that could potentially impact the robustness of the predictive model (See Fig. 3).

Fig. 3.

Presentation of Williams plot for the comprehensive view of the model’s residuals against the predicted values, allowing us to identify potential outliers or influential observations.

Outliers or influential data points are typically those residuals that deviate substantially from the baseline (the horizontal dashed line at zero) as shown in In-subria plot (See Fig. 4). Notably, several points have high standardized residuals, such as points labelled 203, 32, 33, and 34, with labels positioned far above or below the dashed line. These outlier data points could have a great influence on the model’s overall fit, and further investigation into their nature is warranted. Points such as 409, which lies significantly away from the main cluster on the x-axis, might be indicating an extreme prediction error / outliers or a unique characteristic of that particular data point that does not fit well with the model’s general pattern. On the left side of the plot, the data points around 37 and 63 exhibit considerable deviations from the zero residual line, which are also potential candidates for being influential outliers. These data points should be examined more closely, as they may suggest issues such as data entry errors, measurement anomalies, or cases where the model may not be capturing the underlying data relationships effectively. In conclusion, the Williams plot highlights several outlier compounds, such as those numbered 203, 32, 33, 34, 409, 37, and 63, whose extreme residuals deviate from the general trend and could potentially have a disproportionate effect on the model.

Fig. 4.

Display of Insubria plot for the developed QAR model.

PC1 and PC2 plot analysis

The PCA plot demonstrates the variance structure along PC1 (35.84%) and PC2 (19.55%). Observations from training and prediction exhibit clustered distributions, alongside some isolated outliers. The biplot presented illustrates variable loadings, with the length and direction of the arrows (e.g., fnotringOsp2N9B, com_Nminus_2A, lipo_S_1Ac) representing the contributions and correlations among the variables (see Figs. 5 and 6). Variables that are strongly projected significantly influence major data patterns, allowing for the interpretation of key descriptors and the relationships among multiple variables.

Fig. 5.

Display of first principal component (PC1) Vs second principal component (PC2) scatter plot.

Fig. 6.

The biplot visualizes how different variables contribute to the principal components PC1 and PC2. Arrows indicate the loadings of each variable, showing their influence on these components.

Together, these two plots provide a comprehensive view of both the structure of the data (from the first plot) and the contribution of individual variables to this structure (from the second plot). By analysing the patterns in the plot includes yellow and blue points, where yellow represents training data and blue represents predicted data. Several points are labelled with numbers that highlight certain compounds with extreme values. Notably, points like 409, 200, 209, 347, 391, and 392 are situated far away from the main cluster of data. These data points are outliers, suggesting that the model does not predict these particular values accurately. Points such as 409 and 392 exhibit significant discrepancies between predicted and observed values, possibly indicating that they are influential outliers or cases where the model’s prediction do not cover.

Mechanistic interpretation

The QSAR model delineates the critical molecular descriptors that influence the activity of uPAR inhibitors (See Table 5). The variable com_Nminus_2A, with an average value of 0.75, exhibits a negative correlation with activity. This suggests that the presence of nearby negatively charged nitrogen atoms may impede binding interactions, likely due to the effects of electrostatic repulsion. The compound lipo_S_1Ac, with an average value of -0.43, exhibits a negative correlation. This indicates that increased lipophilicity in proximity to sulphur atoms may lead to a reduction in affinity. In contrast, fHC3B (average 0.68) and fdonringC7A (average 13.5) exhibit a positive correlation with activity, which contributes to increased molecular flexibility and optimal geometry for receptor binding. The notation fsp3Nsp2C6B, with an average value of 0.43, characterizes sp2 carbons that are six bonds away from sp3 nitrogens, thereby playing a crucial role in maintaining structural balance within binding interactions. The observed negative correlation suggests that this bonding configuration leads to a lower level of activity. This reduction is likely attributable to factors such as rigidity or unfavourable electronic effects that constrain receptor flexibility. fnotringOsp2N9B (avg 0.24) exhibits a positive correlation with activity, indicating the presence of advantageous nitrogen–oxygen interactions, which may include hydrogen bonding or electrostatic effects. In a similar manner, sp2N_S_5B (average 0.37) exhibits a positive correlation, indicating that sulphur atoms located near sp2-hybridized nitrogens improve binding affinity and positively influence receptor interaction.

Table 5.

Illustration of molecular description obtained in the developed QSAR model.

Molecular descriptor	Description	Average	Range	Software used	Correlation
Com_Nminus_2A	Occurrence of negatively charge nitrogen atoms within 2 A From the center of the molecule	0.75	0–2	PyDescriptor	Negative
lipo_S_1Ac	Sum of partial charges of lipophilic atoms after 1 Angstrom from sulphur atoms	0.04	-0.46 - 0.43	PyDescriptor	Negative
fHC3B	Frequency of occurrence of carbon atom exactly 3 bonds from the hydrogen atom	0.68	0–5	PyDescriptor	Positive
fdonringC7A	Frequency of occurrence of ring carbon atom exactly at 7 A from the donor atom	13.5	0 − 26	PyDescriptor	Positive
fsp3Nsp2C6B	Frequency of occurrence of sp2 hybridized carbon atom exactly at 6 bonds from the sp3 hybridized nitrogen atom	0.43	0–4	PyDescriptor	Negative
fnotringOsp2N9B	Frequency of occurrence of sp2 hybridized nitrogen atom exactly at 9 bonds from non-ring oxygen atom	0.24	0–1	PyDescriptor	Positive
sp2N_S_5B	Occurrence of Sulphur atom within five bonds from the sp2 hybridized nitrogen atom	0.37	0–2	PyDescriptor	Positive

This highlights the importance of sulfur-nitrogen interactions, which could enhance binding through van der Waals or electrostatic forces, making this feature advantageous for uPAR inhibition. In conclusion, these molecular descriptors emphasize the significance of optimizing lipophilicity, charge distribution, and atom-specific interactions, such as nitrogen-sulfur and carbon-hydrogen bonding, to enhance the potency of uPAR inhibitors. These findings provide a valuable framework for the rational design of more effective inhibitors targeting the uPAR receptor.

com_Nminus_2A The presented activity cliff analysis based on the descriptor com_Nminus_2A, which denotes the occurrence of negatively charged nitrogen atoms within a 2 Å radius from the molecular centroid, underscores the importance of electronic and spatial characteristics in optimizing urokinase-type plasminogen activator receptor (uPAR) inhibitory activity. This descriptor exhibits a negative correlation with biological activity in the developed QSAR model, indicating that a lower number of centrally located negatively charged nitrogen atoms is beneficial for achieving stronger receptor binding and increased Ki potency. A pronounced activity cliff is observed between compound 27 (Ki = 5100 nM, pKi = 5.29, com_Nminus_2A = 2) and compound 31 (Ki = 280 nM, pKi = 6.55, com_Nminus_2A = 1), where the reduction of one nitrogen atom near the molecular center of mass significantly improves bioactivity (see Fig. 7). This suggests that excess localized electron density, likely due to two centrally located electron-rich nitrogen atoms in compound 27, leads to unfavourable electrostatic repulsion or misorientation in the binding pocket, disrupting optimal receptor interactions. Conversely, compound 31 achieves a more favorable spatial orientation with its reduced nitrogen count, resulting in a tighter and more stable drug–receptor complex.

Fig. 7.

Demonstration of molecular descriptor com_Nminus_2A.

This trend is consistent across other molecular pairs, such as 802 (Ki:53nM, pKi:7.27 M, com_Nminus_2A:1) with 793 (Ki:520 nM, pKi: 5.29 M, com_Nminus_2A:2), 173(Ki:3570nM, pKi:5.44 M, com_Nminus_2A:2) with 845 (Ki:2498 nM, pKi:5.60 M, com_Nminus_2A:1), and, 837 (Ki:1208 nM, pKi:5.91 M, com_Nminus_2A:1) with 173 (Ki: 3570 nM, pKi: 5.44 M, com_Nminus_2A:2), also support this observation These observations highlight a generalizable structure–activity relationship wherein minimizing the number of central negatively charged nitrogen atoms enhances uPAR binding affinity. For future structural optimization, it is recommended to consider the relocation or removal of nitrogen functionalities such as amine, imidazole, or pyridine for the purpose of structural optimisation. The transformation of amines into neutral amides, the substitution of imine nitrogen with carbon, or the relocation of nitrogen to the peripheries of the molecule can lead to a decrease in the com_Nminus_2A count. Replacing nitrogen-rich heterocycles with carbon and oxygen groups enhances hydrophobic interactions and maintains electronic neutrality. Overall, this descriptor serves as a critical parameter for fine-tuning the electrostatic complementarity between drug and receptor, enabling medicinal chemists to engineer molecules with enhanced affinity and selectivity through functional group deletion or substitution strategies aimed at reducing com_Nminus_2A.

lipo_S_1Ac (Sum of partial charges of lipophilic atoms after 1 Angstrom from Sulphur atoms): This molecular descriptor highlights the electronic characteristic of the molecule which plays crucial role in drug receptor interactions. The impact of molecular descriptor can by observed by comparing the match molecular pairs 313(Ki:99, pKi:7.004 M, lipo_S_1Ac:0.21), 316(Ki:327, pKi: 6.48 M, lipo_S_1Ac:0.215), 318(Ki:277, pKi: 6.55 M, lipo_S_1Ac:0.21), and 433 (Ki:531, pKi:531 M, lipo_S_1Ac:0.371) in which increase in the partial positive charges decreases the UPAR inhibition, thus, substitution of electronegative functional group such as chlorine is desirable to enhance the biological activity. The activity cliff analysis of UPAR inhibitors, based on the molecular descriptor lipo_S_1Ac, provides valuable insights into optimizing drug design strategies. Lipo_S_1Ac represents the sum of partial charges on lipophilic atoms located within 1 Å from sulfur atoms and is helpful in understanding the electronic characteristics that affect the interaction of inhibitors with the UPAR receptor. In this analysis, four compounds;313, 316, 318, and 433, are compared in terms of their Ki values, which measure the inhibitory potency. Compound 433, with a lipo_S_1Ac value of 0.371, demonstrates the weakest inhibitory activity (Ki: 531 nM), whereas compounds 313, 316, and 318, with lipo_S_1Ac values around 0.215, exhibit significantly better activity, with Ki values of 99 nM, 327 nM, and 277 nM, respectively (See Fig. 8).

Fig. 8.

Demonstration of molecular descriptor lipo_S_1Ac.

This suggests that the partial positive charge around the Sulphur atoms in the molecule plays a crucial role in determining inhibitory potency. Specifically, an increase in the partial positive charges on the Sulphur atoms correlates with a decrease in UPAR inhibition. This finding indicates that lowering the partial positive charge density around the Sulphur atoms could enhance the inhibitory activity of these molecules. A potential strategy to achieve this is through the incorporation of electronegative substituents, such as chlorine or fluorine, which could help reduce the partial positive charge density near the sulfur atoms and thereby improve receptor binding.

For example, compound 313, which contains a chlorine atom, demonstrates the strongest inhibitory activity (Ki: 99 nM), supporting the idea that electronegative substituents contribute positively to the biological activity of these inhibitors. Furthermore, the comparison between the compounds also reveals that lipophilicity plays an important role in their biological activity. Lipo_S_1Ac serves as a measure of lipophilicity, and it is evident that the compounds with a lipo_S_1Ac value close to 0.215 (compounds: 313, 316, and 318) tend to show better inhibition than those with a higher value (compound:433). This suggests that an optimal balance of lipophilicity is essential for effective receptor binding and activity. Future drug design efforts should focus on fine-tuning lipophilicity, ensuring that compounds have an appropriate lipo_S_1Ac value, which could improve both potency and pharmacokinetic properties. Moreover, the analysis implies that fine-tuning the electronic distribution around the sulfur atoms in these molecules is critical. The substitution of electronegative functional groups, particularly near the sulfur atoms, could reduce the electron density around these atoms, leading to better receptor interactions and higher inhibitory potency. Another approach to optimize UPAR inhibitors could involve altering the molecular scaffold to maintain an optimal lipo_S_1Ac value while ensuring the electronic environment around sulfur is suitably adjusted for stronger receptor binding. The inclusion of other electronegative groups like fluorine or bromine could be explored in the future to further optimize the electronic and lipophilic characteristics of these molecules. In summary, by strategically incorporating electronegative substituents and fine-tuning lipophilicity, drug design can be directed toward enhancing UPAR inhibition. These strategies will help optimize the electronic characteristics of the molecules, improving their receptor-binding ability and ultimately resulting in more effective therapeutic agents for diseases related to UPAR. This detailed analysis not only highlights the importance of molecular descriptors such as lipo_S_1Ac in drug design but also underscores the need for a holistic approach that balances electronic, lipophilic, and structural modifications to achieve optimal biological activity.

fHC3B (frequency of occurrence of carbon atom exactly 3 bonds from the hydrogen atom). This molecular descriptor captures key steric and lipophilic characteristics of the molecule, which are fundamental determinants in ligand–receptor interactions and overall pharmacological activity and, positively correlated with the urokinase plasminogen receptor inhibitory activity (Ki). Its relevance becomes apparent when comparing compound 830 (Ki = 904 nM, pKi = 6.04, fHC3B = 2) with compound 842 (Ki = 287 nM, pKi = 6.54, fHC3B = 3) (see Fig. 9). Notably, compound 842 exhibits a higher number of hydrogen atoms located precisely three bonds away from a carbon atom, a structural feature that appears to significantly enhance its binding affinity, as evidenced by its lower Ki value. This suggests that the additional hydrogens reduce steric hindrance between the ligand and the urokinase-type plasminogen activator receptor (uPAR), thereby promoting a more favorable ligand-receptor complementarity. As a result, compound 842 demonstrates a superior bioactivity profile compared to compound 830. This analysis highlights how subtle topological variations—such as hydrogen positioning can modulate molecular interactions at the receptor interface.

Fig. 9.

Illustration of molecular descriptor fHC3B.

Therefore, this descriptor proves valuable in rational drug design, where the optimization of steric and hydrophobic features can critically influence receptor engagement and therapeutic potential. The comparative activity cliff analysis of the presented molecular dataset reveals that the descriptor fHC3B defined as the frequency of carbon atoms located exactly three bonds from a hydrogen atom acts as a critical determinant of urokinase-type plasminogen activator receptor (uPAR) inhibition. This descriptor encodes both steric and lipophilic properties, which are essential for optimal ligand–receptor binding. An upward trend in bioactivity with increasing fHC3B is consistently observed across compound pairs, as seen with compound 830 (Ki = 904 nM, fHC3B = 2) versus compound 842 (Ki = 287 nM, fHC3B = 3), and compound 795 (Ki = 934 nM, fHC3B = 2) versus compound 802 (Ki = 53 nM, fHC3B = 3). In both comparisons, the compounds with higher fHC3B values demonstrate significantly enhanced potency. This implies that the presence of hydrogens three bonds from carbon atoms alleviates steric congestion and enhances receptor complementarity, thereby improving binding affinity. From a structural optimization perspective, future analogs should prioritize the addition of small, hydrogen-rich alkyl groups, such as methyl, ethyl, or methylene bridges in regions where fHC3B can be increased without introducing electronic interference or polarity that may disrupt receptor interactions (See Table 6).

Table 6.

Display of optimization strategies for the compounds; 830, 842, 795, and 802.

Compound	Activity (Ki / pKi)	fHC3B	Key Structural Features	Optimization Strategy	Expected Impact
830	904 nM / 6.04	2	Nitrile group, planar aromatic system	Replace CN with –CH₃ or –CH₂CH₃, Addition of methylene near aromatic ring	↑ fHC3B, ↓ steric hindrance, ↑ flexibility
842	287 nM / 6.54	3	Methoxy group, fused ring system	Retain current scaffold, Addition of small alkyl substitutions on ring	Maintain/increase fHC3B and hydrophobicity
795	934 nM / 6.03	2	Polyheterocycle with amide & oxygen	Remove/reduce polar amide , Replace rigid linker with CH₂ or CH₃	↑ fHC3B, ↓ polarity, ↑ receptor fit
802	53 nM / 7.27	3	Piperazine ring, tricyclic core	Retain core structure, Addition of isosteric oxazole replacements	↑ stability, maintain fHC3B, enhance binding

In compound 830, for example, the introduction of a methylene spacer adjacent to the aromatic system or replacement of electron-withdrawing nitrile groups with non-polar alkyl chains could increase hydrogen density at favorable topological that contribute to a higher fHC3B while avoiding excessive substitution may further preserve or enhance activity. Similarly, in compound 802, the piperazine-linked tricyclic core provides both structural rigidity and topological hydrogen richness, contributing to its potent activity (Ki = 53 nM). Here, optimization could involve isosteric replacement of the oxazole ring or ring expansion/contraction to fine-tune hydrophobic interactions and further increase fHC3B. In low-potency analogs like compound 795, the presence of bulky heteroatoms such as oxygen and nitrogen in rigid arrangements likely reduces fHC3B and increases steric hindrance. Removing non-essential polar functional groups, e.g., an amide or carbamate and replacing them with hydrogen-dense moieties may improve receptor binding by restoring spatial freedom. Thus, increasing fHC3B through strategic functional group additions (alkylation, methylene insertion) or removals (desubstitution of polar or bulky groups) can be systematically used to drive higher uPAR inhibitory activity, making fHC3B a valuable descriptor in medicinal chemistry efforts aimed at optimizing steric and lipophilic complementarity with the receptor binding domain.

fdonringC7A (Frequency of occurrence of ring carbon atom exactly at 7 A from the donor atom) The molecular descriptor fdonringC7A plays a crucial role in identifying key pharmacophoric elements, specifically the donor group and the ring carbon atom, both of which significantly contribute to stabilizing drug-receptor interactions. These contributions occur through electrostatic and hydrophobic forces, which are essential for optimizing binding affinity. In the context of the quantitative structure-activity relationship (QSAR) model, the descriptor fdonringC7A exhibits a positive coefficient, suggesting that an increase in its value correlates with enhanced electrostatic and hydrophobic interactions between the ligand and its receptor. This enhancement leads to a stronger drug-receptor binding affinity, which is reflected in a reduction of the inhibition constant (Ki). Consequently, a higher fdonringC7A descriptor value results in a more favorable interaction profile, potentially improving the efficacy of the drug. Thus, fdonringC7A is not only a critical feature for understanding the molecular interactions but also serves as a predictor for optimizing drug design in QSAR models, where the precise manipulation of this descriptor can fine-tune drug-receptor binding properties and contribute to better pharmacological outcomes. The influence of the fdonringC7A molecular descriptor on uPAR inhibitory activity is evident when examining the matched molecular pair 522 (Ki: 1900 nM, pKi: 5.72, fdonringC7A: 16) and 778 (Ki: 10000 nM, pKi: 5.0, fdonringC7A: 15). Notably, the alteration of a single carbon atom markedly shifts the binding affinity (Ki), indicating a substantial impact on the compound’s bioactivity. Specifically, the inclusion of one additional carbon atom in compound 522 corresponds to a significantly enhanced uPAR inhibitory effect relative to 778 (See Fig. 10).

Fig. 10.

Portrayal of molecular descriptor fdonringC7A.

Conversely, the removal of a carbon atom in compound 778 results in diminished bioactivity, suggesting a crucial role for hydrophobic and electrostatic interactions modulated by this structural feature. The marked difference in uPAR inhibitory activity between compound 522 (Ki = 1900 nM, fdonringC7A = 16) and compound 778 (Ki = 10000 nM, fdonringC7A = 15) highlights an activity cliff driven by a single carbon atom variation affecting the fdonringC7A descriptor. This molecular descriptor, quantifying the frequency of ring carbon atoms located precisely 7 Å from donor atoms, is directly associated with enhanced hydrophobic and electrostatic drug-receptor interactions. Compound 522, with a higher fdonringC7A, benefits from optimal spatial positioning of a donor group near a hydrophobic ring system, resulting in a stronger affinity. To optimize future compounds, functional group modifications should aim to increase the fdonringC7A value by either adding donor groups (e.g., –OH, –NH₂, –NH–) or extending hydrophobic aromatic systems (e.g., adding phenyl, benzyl, or substituted aryl groups) in spatial proximity (~ 7 Å) to these donors. For instance, incorporating a methoxyphenyl, piperidine, or morpholine moiety could improve hydrophobicity while contributing donor capability or favorable geometry. Alternatively, introducing a sulfonamide or urea linker could serve dual roles as hydrogen bond donors/acceptors while extending ring systems. Conversely, the removal of such moieties, as seen in 778, disrupts the optimal pharmacophore arrangement, leading to reduced affinity. Thus, rational design should maintain or enhance both the donor presence and the adjacent ring system to leverage the fdonringC7A descriptor for optimized uPAR inhibition. Subsequently, some additional match molecular pair also support the similar trend, such as 829 with 843, 661 with 678, 158 with 640 and few more to mentioned. These observations highlight the sensitivity of the uPAR binding pocket to subtle structural modifications and underscore the importance of hydrophobic contribution in ligand optimization. Consequently, rational drug design strategies targeting uPAR should consider incorporating such structural motifs, particularly carbon-rich functionalities, to enhance binding affinity and improve pharmacological efficacy.

fsp3Nsp2C6B (Frequency of occurrence of sp2 hybridized carbon atom exactly at 6 bonds from the sp3 hybridized nitrogen atom). The molecular descriptor encapsulates a distinct integration of lipophilic and electronic attributes characterized by the presence of an sp²-hybridized carbon atom and an sp³-hybridized nitrogen atom. Within the quantitative structure–activity relationship (QSAR) model developed for urokinase-type plasminogen activator receptor (uPAR) inhibition, this descriptor exhibits a negative regression coefficient, indicating an inverse relationship with bioactivity, that an increase in its magnitude correlates with a reduction in inhibitory potency against uPAR. This suggests that molecular structures exhibiting a greater prevalence or spatial emphasis on these hybridization features are less efficacious in modulating uPAR function. Furthermore, the descriptor highlights the relevance of the topological (graph-theoretical) distance between the sp²-hybridized carbon and sp³-hybridized nitrogen atoms as a critical structural parameter. This spatial separation likely influences molecular conformation, electronic distribution, or interaction potential with the uPAR binding site, thereby modulating the bioactivity profile. The inclusion and statistical significance of this descriptor underscore the nuanced interplay between electronic configuration and three-dimensional molecular topology in determining receptor binding efficiency, thereby offering mechanistic insights valuable for rational drug design aimed at enhancing uPAR inhibition. The influence of the molecular descriptor can be observed by comparing the molecule 548 with the molecule 99. Molecule 548 exhibits a higher fsp³Nsp²C6B value, indicating a greater topological distance between sp²-hybridized carbon atoms and sp³-hybridized nitrogen atoms, which correlates with a reduction in its inhibitory potency against uPAR (See Fig. 11).

Fig. 11.

Display of molecular descriptor fsp³Nsp²C6B.

This increased separation is likely due to the presence of electron-withdrawing functional groups, particularly the chlorine (Cl) atom and the sulfonamide group (-SO₂NH) attached to the aromatic ring. These groups contribute to the electronic redistribution within the molecule, increasing the spatial separation between the sp² carbon and sp³ nitrogen. To optimize the bioactivity of molecule 548, removal of these electron-withdrawing groups could be considered, as they contribute to an unfavourable spatial configuration for receptor binding. Specifically, eliminating the chlorine and sulfonamide groups would decrease the electronic repulsion between the hybridized atoms, thereby reducing their spatial separation. Furthermore, the introduction of electron-donating groups, such as hydroxyl (-OH) or amine (-NH₂) functionalities, could be strategically incorporated onto the aromatic ring to enhance the electron density around the sp² carbons. This would decrease the distance between the sp² carbon and sp³ nitrogen, facilitating better molecular alignment and improving the binding affinity to the uPAR receptor. In contrast, molecule 99 demonstrates a lower fsp³Nsp²C6B value, suggesting a more favorable spatial configuration with a reduced distance between the sp²-hybridized carbon and sp³-hybridized nitrogen atoms, which is indicative of enhanced bioactivity. The presence of hydroxyethyl (-CH₂CH₂OH) and carboxamide (-NHCO) groups facilitates the closer spatial proximity of these hybridized atoms, optimizing the molecular conformation for uPAR receptor interaction. To further optimize molecule 99, fine-tuning of the functional groups could involve the addition of further alkyl or hydroxyl substituents, which would reduce the spatial separation between the hybridized atoms, enhancing their interaction potential with the receptor. The strategic incorporation of electron-donating groups would likely reinforce the favorable conformation and further augment receptor binding, thus improving the overall inhibitory potency. Therefore, structural optimization for both molecules should focus on the modulation of electron-donating and electron-withdrawing groups to reduce the distance between sp² carbon and sp³ nitrogen atoms, which directly influences the molecular interaction with uPAR and consequently enhances inhibitory potency. Some more pair of molecules also support this observation and follow the similar trend, such as 148 with 631, 146 with 611, 147 with 612, 228 with 611, 612 with 631, 228 with 615, 446 with 449, 128 with 227, and few more to mention.

fnotringOsp2N9B (Frequency of occurrence of sp2 hybridized nitrogen atom exactly at 9 bonds from non-ring oxygen atom) The molecular descriptor in focus highlights the combined effect of two electronic pharmacophoric features: the sp²-hybridized nitrogen atom (sp²N) and an oxygen atom located on the ring at a topological distance of nine bonds. This descriptor reveals a positive correlation with UPAR inhibition, significantly enhancing the binding affinity of molecules through electrostatic interactions, thereby elevating the value of this particular descriptor. The impact of this molecular feature is clearly demonstrated by comparing a set of three molecular pairs: molecule 146 (Ki: 2100 nM, pKi: 5.67 M, fnotringOsp2N9B: 1), molecule 611 (Ki: 6600 nM, pKi: 5.18 M, fnotringOsp2N9B: 0), and molecule 612 (Ki: 8700 nM, pKi: 5.06 M, fnotringOsp2N9B: 0). The analysis reveals that molecule 146, which contains a single sp²-hybridized nitrogen atom, shows significantly improved UPAR inhibitory potency (Ki), with a Ki value of 2100 nM. This suggests that the presence of sp²N plays a crucial role in augmenting the molecule’s bioactivity. In contrast, molecules 611 and 612, which lack the sp²-hybridized nitrogen atom, exhibit diminished bioactivity, as evidenced by their higher Ki values of 6600 nM and 8700 nM, respectively (See Fig. 12). The absence of this sp²N feature in these molecules correlates with a significant decrease in their UPAR inhibitory potency, thus underlining the importance of this pharmacophoric element.

Fig. 12.

Illustration of molecular descriptor fnotringOsp2N9B.

The interaction between these two electronic features, sp²N and the on-ring oxygen, facilitates a stronger and more stable binding interaction, ultimately leading to a more potent UPAR inhibition. These findings suggest that in future drug design strategies, maintaining the presence of the sp²-hybridized nitrogen atom in the molecular structure is crucial for optimizing the inhibitory potency against UPAR. Moreover, the combination of these electronic features could be leveraged to further refine drug candidates, enhancing their binding affinity and therapeutic efficacy. Some more pairs of molecules follow the same trends, such as 50 with 511, 56 with 508, 50 with 497, 56 with 491, 56 with 490, 45 with 394, 50 with 484, 446 with 450, and few more to mention. Thus, the incorporation of sp²-hybridized nitrogen, coupled with a strategically placed oxygen atom, could provide a promising approach for developing more effective UPAR inhibitors and improving drug design in general. The molecular descriptor fnotringOsp2N9B emphasizes the synergistic effect of two key pharmacophoric features: the sp²-hybridized nitrogen atom (sp²N) and an oxygen atom located at a topological distance of nine bonds.

The descriptor is positively correlated with UPAR inhibition, significantly enhancing binding affinity through electrostatic interactions, which in turn increases the value of this molecular descriptor. The relevance of these features is demonstrated by comparing the activity of three molecular pairs: molecule 146 (Ki: 2100 nM, pKi: 5.67 M, fnotringOsp2N9B: 1), molecule 611 (Ki: 6600 nM, pKi: 5.18 M, fnotringOsp2N9B: 0), and molecule 612 (Ki: 8700 nM, pKi: 5.06 M, fnotringOsp2N9B: 0). Molecular 146, which contains an sp²-hybridized nitrogen atom, exhibits significantly enhanced UPAR inhibitory potency (Ki), suggesting that the presence of this sp²N feature is critical for augmenting molecular bioactivity. In contrast, molecules 611 and 612, which lack the sp²-hybridized nitrogen, display diminished UPAR inhibitory potency, as indicated by their elevated Ki values. This underscores the pivotal role of the sp²N in determining the bioactivity profile. The sp²-hybridized nitrogen and an oxygen located nine bonds apart contribute to electrostatic complementarity, thereby stabilising the UPAR complex and enhancing inhibitory potency. Maintaining the sp²N feature is crucial for future designs. Adjusting electron density through the incorporation of electron-withdrawing groups and fine-tuning the steric characteristics surrounding the oxygen can enhance binding interactions, minimise steric hindrance, and produce more effective UPAR inhibitors.

sp2N_S_5B

The molecular descriptor represents the occurrence of a sulfur atom within five bonds from a sp2 hybridized nitrogen atom, also shows a positive correlation with activity (average value of 0.37, range 0 to 2). This molecular descriptor displays positive correlation with UPAR inhibition (Ki), therefore increase in the value of descriptor enhances binding with UPAR through van der Waals or electrostatic forces, making this feature advantageous for uPAR inhibition as, this descriptor highlights the importance of sulfur-nitrogen combination. The molecular descriptor captures the frequency of occurrence of a sulfur atom positioned within a five-bond topological distance from an sp²-hybridized nitrogen atom. Statistically, this descriptor exhibits an average value of 0.37, with an observed range between 0 and 2 across analysed compounds. This descriptor quantifies a sulfur atom within five bonds of an sp²-hybridized nitrogen and shows a positive correlation with uPAR inhibition (K_i). Increased S sp²N proximity enhances binding through van der Waals and electrostatic interactions. Its polarizable sulfur and electron-accepting nitrogen jointly stabilize ligand receptor interactions, making this motif a favorable pharmacophoric element in uPAR-targeted design. The activity cliff between compounds 821 and 13 can be attributed to the molecular descriptor sp²N_S_5B, which quantifies the presence of a sulfur atom located within a five-bond topological distance from an sp²-hybridized nitrogen atom. Compound 821, with a sp²N_S_5B value of 2, demonstrates significantly higher potency in uPAR inhibition (Ki = 10 nM, pKi = 8) compared to compound 13, which has a sp²N_S_5B value of 0 and a reduced activity (Ki = 11 nM, pKi = 7.95) (See Fig. 13).

Fig. 13.

Portrayal of molecular descriptor sp²N_S_5B for the molecules 821 and 13 only.

The sulfur atom, with its polarizability and moderate lipophilicity, contributes to hydrophobic interactions, while the sp² nitrogen, often acting as a hydrogen bond acceptor or involved in π-stacking, stabilizes the ligand–receptor complex. In contrast, compound 13, which lacks this critical feature, exhibits diminished receptor binding, resulting in reduced inhibitory potency. To optimize uPAR inhibition and address the observed activity cliff, future structural modifications should focus on maintaining or increasing the sp²N_S_5B value. Incorporating sulfur-containing heterocycles (e.g., thiazoles, thiophenes) near sp² nitrogens, positioned to maintain the five-bond separation, could enhance binding interactions. Additionally, amide or sulfamide groups could be incorporated to retain this spatial configuration while introducing further hydrogen-bonding capabilities. The inclusion of sulfur-rich substituents would enhance van der Waals and electrostatic interactions, optimizing the receptor binding affinity. From an ADMET perspective, compounds with higher sp²N_S_5B values, such as compound 821, are likely to demonstrate improved absorption due to enhanced lipophilicity and increased membrane permeability driven by the sulfur-containing group. The increased potency observed in compound 821 suggests that the spatial configuration of sulfur and nitrogen also contributes to optimal tissue distribution, particularly in tissues expressing uPAR. Furthermore, the incorporation of sulfur may offer resistance to metabolic degradation, increasing metabolic stability. The improved binding affinity of compound 821 implies that it may have a higher therapeutic index and selectivity, reducing the likelihood of off-target toxicity. Therefore, targeting the sp²N_S_5B descriptor in the rational design of uPAR inhibitors can optimize both biological activity and pharmacokinetic properties, improving their therapeutic potential.

QSAR based virtual screening

QSAR-based virtual screening (VS) represents a computational methodology in drug discovery that establishes a relationship between molecular descriptors and biological activity. This is achieved through the application of machine learning techniques, GA-MLR, which are utilised to develop QSAR models. Prior to the development of QSAR model, molecular descriptors were calculated for the external dataset obtained from the chemdiv (See Table 7).

Table 7.

Presentation of hit molecules identified by QSAR-Based virtual screening along with their molecular id, calculated descriptor values, and predicted IC₅₀ values (_PIC_{50 in M}).

mol_name	com_Nminus_2A	lipo_S_1Ac	fHC3B	fdonringC7A	fsp3Nsp2C6B	fnotringOsp2N9B	sp2N_S_5B	pki by QSAR VS
D685-0061	0	0	0	19	0	2	0	9.02
G610-0057	0	0	0	19	0	2	1	9.02
F871-0668	0	0	0	27	0	1	0	8.91
F523-0244	0	0	0	18	0	2	0	8.89
F523-0400	0	0	0	18	0	2	0	8.89
8018 − 7244	0	0	0	17	0	2	0	8.76
C301-8561	0	0	0	17	0	2	0	8.76
C301-9184	0	0	0	17	0	2	3	8.76
D374-0045	0	0	0	17	0	2	0	8.76
D524-2413	0	0	0	17	0	2	0	8.76
E640-0223	0	0	0	17	0	2	2	8.76
F518-0612	0	0	0	17	0	2	0	8.76
D426-0031	0	0	0	25	0	1	0	8.64
D488-0331	0	0	0	24	0	1	2	8.51
C301-7568	0	0	0	15	0	2	0	8.5
D374-0091	0	0	0	15	0	2	0	8.5
D560-0168	0	0	0	15	0	2	0	8.5
G802-1124	0	0	0	15	0	2	0	8.5
C073-4494	0	0	0	32	0	0	0	8.40
C073-4530	0	0	0	32	0	0	0	8.40

The activity of novel compounds is predicted by these models, which leads to a reduction in both experimental testing time and resource expenditure. Effective QSAR applications encompass the identification of active compounds for malaria, schistosomiasis, tuberculosis, and viral infections. In present work, QSAR-based virtual screening and molecular docking analyses have identified D685-0061 as a potent inhibitor of the urokinase plasminogen activator receptor (UPAR), exhibiting a pIC₅₀ value of 9.024 M. This compound demonstrates superior performance compared to other analogues and is associated with advantageous electronic and topological characteristics, such as electron-rich, hydrogen bond-donating groups and sp²-hybridized heteroatoms. The presence of these attributes facilitates binding with UPAR, and the lack of steric or lipophilic hindrances allows for better receptor accommodation. This method illustrates the efficacy of QSAR-based virtual screening in systematic drug design and the identification of new lead compounds. In parallel, ligand C878-1660, selected through molecular docking, demonstrated the most favorable binding free energy of -8.4897 kcal/mol and an RMSD of 3.3844 Å, indicative of a thermodynamically favorable yet conformationally flexible interaction within the UPAR active site. While the RMSD marginally exceeds the typical threshold for optimal pose convergence (< 2 Å), it remains acceptable in light of the strong binding energy and may suggest a dynamic binding mode that enables interaction with multiple receptor conformers (See Table 8). Comparative assessment against other docked ligands such as C073-4530 (-8.3269 kcal/mol, RMSD = 2.6259 Å), C636-2326 (-8.1544 kcal/mol, RMSD = 0.9611 Å), and D576-0310 (-7.9171 kcal/mol, RMSD = 3.4891 Å) affirms the relative superiority of C878-1660 in terms of ligand-receptor affinity. The presence of C878-1660 in the docking dataset and not in the QSAR list, and the converse for D685-0061, underscores the orthogonal validation achieved through dual-platform screening (See Table 8).

Table 8.

Presentation of structures of top five hit molecules identified by molecular Docking-Based virtual screening along with their id, Docking scores and RMSD values.

Name of the Molecule	Structures	Docking Score (kcal/mol)	RMSD
C878-1660		-8.48	3.38
C073-4530		-8.32	2.62
C636-2326		-8.15	0.96
C301-7568		-8.15	3.36
D430-0515		-8.02	1.01

Collectively, D685-0061 is prioritized for its exceptional ligand-based potency prediction, while 0356 − 0096 is advanced based on its superior structure-based interaction energy and plausible binding adaptability. This dual-strategy approach enhances the confidence in their selection, offering a rational foundation for their progression into experimental pipelines for UPAR-targeted therapeutic development, including biochemical inhibition assays, receptor-ligand crystallographic studies, and molecular dynamics simulations to further characterize binding stability, specificity, and induced-fit behaviour at the atomic level.

Molecular Docking

In evaluating the molecular docking interactions of ligands C878_1660 and D685_0061 with the urokinase-type plasminogen activator (uPA) against the co-crystallized reference ligand SK1 (CHEMBL158405) within the PDB:1W11 structure, a detailed comparative interaction profile reveals key differences and similarities that highlight the binding efficiency and potential therapeutic viability of these compounds. The redocking validation of a known uPAR inhibitor (SK1) was carried out using AutoDock Vina, employing its empirical free energy scoring function that estimates binding affinity based on a combination of steric, hydrophobic, hydrogen-bonding, and torsional energy terms. The overlay of the experimental (cyan) and redocked (green) conformations is shown in the image, and the resulting RMSD value of 1.901 Å confirms a highly reliable docking accuracy (RMSD < 2 Å) (see Fig. 14). The native ligand SK1 exhibits robust hydrophobic and polar interactions, including significant hydrophobic contacts with GLN192 and VAL213 at distances of 3.55 Å and 3.63 Å, respectively, stabilizing the ligand within the active site cleft (See Fig. 14). Furthermore, SK1 forms a comprehensive hydrogen bonding network with residues LEU0, ASP189, SER190, SER214, GLY216, and GLY219, with bond lengths ranging from 1.85 Å to 2.23 Å and donor angles between 129.18° and 169.76°, supporting high-affinity binding through seven hydrogen bonds. These interactions are reinforced by water bridges involving ARG217 and LEU222 that further tether the ligand, enhancing the hydration shell’s contribution to overall stability. In contrast, ligand C878_1660 maintains only two hydrophobic contacts, notably with GLN192 (3.70 Å) and TRP215 (3.60 Å), and engages in two relatively weaker hydrogen bonds with GLN192 and GLY216, both exhibiting longer distances (3.44–3.97 Å) and lower donor angles (115.39° and 131.17°), which suggests a potentially less stable binding conformation.

Fig. 14.

Presentation of 3D interactions and surface fit of ligand SK1 (CHEMBL158405) with Co-crystal structure of Urokinase Type Plasminogen Activator (PDB:1w11).

Notably, C878_1660 forms water bridges via SER146 and GLN192 at comparable distances (~ 3.36 Å) but with lower water angles (72.51°–83.58°), indicating a less geometrically favorable orientation for optimal solvation support. On the other hand, ligand D685_0061 displays a more complex hydrogen bonding pattern than C878_1660, forming three hydrogen bonds with GLN192 and GLY219 with bond lengths from 2.04 Å to 2.68 Å and donor angles around 130°, resembling the compact and more favorable interactions seen with SK1(See Fig. 15).

Fig. 15.

Presentation of 3D interactions and surface fit of ligand C878_1660 with Co-crystal structure of Urokinase Type Plasminogen Activator (PDB:1w11).

However, unlike SK1, which engages with a broader range of residues including several serines and aspartates critical for catalytic function, D685_0061’s interactions are predominantly limited to GLN192 and GLY219, suggesting a more localized binding profile. Additionally, its dual hydrophobic contacts with GLN192 at 3.51 Å and 3.71 Å and absence of water bridges indicate limited engagement with the hydration network essential for stability in dynamic biological environments (see Fig. 16). Overall, while SK1 presents a comprehensive interaction profile encompassing multiple contact types and structural stabilization mechanisms indicative of high binding affinity and specificity, C878_1660 and D685_0061, though engaging crucial residues like GLN192 and GLY219, lack the breadth and structural depth of interaction exhibited by the native ligand. C878_1660’s reliance on fewer and weaker polar contacts, although supported by modest water bridges, contrasts with D685_0061’s stronger direct hydrogen bonds but absence of hydration support, illustrating complementary but incomplete mimicking of the SK1 interaction landscape.

Fig. 16.

Presentation of 3D interactions and surface fit of ligand D685_0061 with Co-crystal structure of Urokinase Type Plasminogen Activator (PDB:1w11).

These distinctions suggest that while both candidate ligands show promise, especially D685_0061 for its tighter hydrogen bonds, further structural optimization would be necessary to fully replicate the high-affinity, multisite binding characteristic of SK1 within the uPA active site. The binding affinity of inhibitors to urokinase-type plasminogen activator (uPA) is governed by a complex interplay of structural features within both the protein target and the inhibitor. Several key elements within the active site of uPA and the molecular architecture of the inhibitors contribute to the overall stability and efficacy of the interaction. The uPA structure is characterized by a catalytic domain responsible for cleaving plasminogen to plasmin, a critical enzyme in fibrinolysis. Inhibitors targeting uPA typically interact with residues in the active site that are involved in substrate binding and catalytic processing. A primary structural feature that enhances the binding affinity of inhibitors is the formation of hydrophobic interactions within the enzyme’s substrate-binding pocket. These interactions are critical in stabilizing the inhibitor within the active site and preventing its dissociation. Residues such as GLN192, VAL213, and TRP215, present in the uPA structure, play a pivotal role in forming these hydrophobic interactions with the ligand. GLN192, in particular, is a crucial residue for binding as it forms strong van der Waals contacts with the ligand, especially in ligands such as SK1, C878_1660, and D685_0061. Additionally, the uPA active site is rich in polar residues that facilitate hydrogen bonding with the inhibitor, further enhancing binding specificity and strength.

For instance, residues like SER190, SER214, GLY216, and GLY219 are known to form hydrogen bonds with the ligand, with optimal bond angles and distances that promote tight binding. Hydrogen bonds contribute significantly to the inhibitor’s affinity by creating a stable, energy-favorable interaction between the ligand and the protein. The geometry of these hydrogen bonds is particularly important; donor-acceptor angles, bond lengths, and the orientation of the side chains in uPA’s active site must align to enable maximal interaction with the inhibitor. For example, SK1 forms multiple hydrogen bonds with key residues like LEU0, ASP189, and SER190, reinforcing the ligand’s position within the binding site and providing a tight fit that precludes competitor molecules from displacing the inhibitor. Beyond hydrogen bonds, the interaction of water molecules within the active site can also be critical to enhancing binding affinity. Water bridges, which are formed by water molecules that interact with both the inhibitor and the protein, contribute to stabilizing the ligand-protein complex. In uPA, water bridges formed by residues like ARG217 and LEU222 have been shown to enhance the solvation of the ligand and maintain its structural integrity within the active site. These interactions help preserve the flexibility of the protein, allowing it to accommodate various inhibitors while maintaining optimal catalytic function. The water-mediated interactions are often observed in inhibitors like SK1, which benefit from the additional stabilization provided by the water network. Another critical feature influencing uPA inhibitor binding affinity is the flexibility and shape of the inhibitor itself. Inhibitors with rigid structures, such as those that tightly fit into the enzyme’s binding pocket, tend to have higher binding affinity.

Conversely, inhibitors with flexible or bulky structures may experience steric clashes with the active site, reducing their affinity. Ligands such as C878_1660 and D685_0061 demonstrate the importance of optimal molecular geometry for effective binding, as deviations from the ideal geometry give rise to a weaker interaction, as seen in the longer hydrogen bond lengths and less favorable donor-acceptor angles in these inhibitors. The affinity of uPA inhibitors is influenced by a combination of structural factors that work together, including shape complementarity, an optimized distribution of charges, and stabilising electrostatic interactions, particularly with residues ASP189 and GLY216. The presence of functional groups that facilitate hydrogen bonding and polar interactions increases specificity, whereas the size of these groups affects their interaction with active or allosteric sites. Conformational adjustments induced by ligand binding enhance the strength of the interaction. High-affinity inhibitors like SK1 are characterized by distinct thermodynamic parameters, specifically low dissociation constants and a favourable balance between enthalpy and entropy, which set them apart from weaker ligands such as C878_1660 and D685_0061. The potency of uPA inhibitors for cancer, thrombosis, and fibrosis is determined by a combination of hydrophobic, electrostatic, and geometric features, which inform the rational design of next-generation inhibitors (See supplementary file S3 for detailed docking interactions of pdb: 1w11 ligand, C878_1660 and D685_0061 ligands).

Molecular dynamics (MD) simulation

The 500 ns molecular dynamics simulation, which compares apo PDB:1w11 with its complexes involving ligands C878-1660 and D685-0061, elucidates the effects of ligand binding on the stability of the protein. The apo form exhibits a progressive increase in RMSD from approximately 1.0 to 2.2 Å, indicating flexible yet stable dynamics and inherent structural variation in the absence of ligand constraints. The C878-1660 complex exhibits a stable RMSD of approximately 1.5–1.6 Å following an initial deviation of 1.3 Å ( see Fig. 17). This stability suggests a reduction in flexibility and an increase in conformational stability, attributed to the presence of stabilising ligand–protein interactions, including hydrogen bonding and hydrophobic contacts.

Fig. 17.

RMSD analysis of the apo protein PDB:1w11 and its complexes with ligands C878-1660 and D685-0061 during 500 ns molecular dynamics simulations. The top-left panel shows the protein backbone RMSD trajectory of the unbound apo protein PDB:1w11, illustrating a gradual increase in structural deviation from approximately 1.0 Å to 2.2 Å over time. The top-right panel presents the RMSD of the protein (blue) and ligand (red) in the C878-1660_dock + PDB:1w11 complex, where the protein remains relatively stable (~ 1.5 Å) until ~ 250 ns, after which ligand RMSD increases, indicating ligand repositioning. The bottom panel depicts RMSD trajectories for the D685-0061_Dock + PDB:1w11 complex, showing an increase in protein RMSD from ~ 1.2 Å to 2.7 Å and a late surge in ligand RMSD beyond 14 Å around 400 ns, signifying ligand dissociation or major conformational rearrangement.

Throughout the simulation, the C878-1660 complex keeps the protein RMSD low and stable, indicating that ligand binding encourages a structurally restricted conformation that is comparable to the crystal structure and limits large-scale backbone motions. A noticeable increase in ligand RMSD at around 250 ns suggests partial dissociation or structural change within the binding pocket. The protein backbone is nevertheless stable in spite of this, suggesting structural robustness and continued overall fold integrity even in the presence of ligand mobility. As opposed to C878-1660, the D685-0061 complex exhibits a consistent rise in protein RMSD from ~ 1.2 Å to 2.7 Å by 500 ns, indicating a diminished stabilising impact. Up until 400 ns, the contemporaneous ligand RMSD is constant; however, it then spikes over 14 Å, indicating ligand egress or a significant change to an unfavourable binding mode. Elevated protein RMSD, which indicates linked destabilisation of the ligand–protein system, occurs concurrently with this event. These divergent behaviours, which represent variations in binding affinity, interaction strength, and conformational control, show that C878-1660 imposes higher protein stiffness whereas D685-0061 causes or does not prevent structural drift. The overall RMSD trajectories analysis demonstrates the specific effects of ligand binding on the dynamics of the protein. The apo protein demonstrates flexible conformational sampling; however, the binding with C878-1660 stabilises the protein and decreases backbone fluctuations. D685-0061, on the other hand, promotes increased flexibility and destabilisation, suggesting a weaker or less persistent binding interaction. Interestingly, the repetitive molecular dynamics (MD) simulations of both ligand complexes with PDB:1w11 exhibited similar protein stability and predictable dynamic behavior across runs. Protein RMSD was consistent at 1.5–2.5 Å across all systems, showing structural stability throughout simulation. The C878-1660 complex exhibited consistent RMSD values, with ligand variations limited to 2–4 Å, indicating robust and stable binding throughout the trajectory. In contrast, the D685-0061 complex showed higher RMSD fluctuations, particularly beyond 400 ns, when the ligand showed temporary detachment or conformational rearrangements. In comparison, numerous simulations validated the reproducibility of both systems, however C878-1660 revealed better conformational stability and binding affinity with PDB:1w11 than D685-0061, which showed more flexible and less stable interactions (see supplemetary material for repated md simulation of ligands; C878-1660 and D685-0061) .

The flexibility of apo PDB:1w11 and its ligand-bound forms differs significantly, according to the 500 ns RMSF study. With RMSF peaks exceeding 2.5 Å in solvent-exposed loops and termini, the apo protein exhibits heterogeneous mobility, demonstrating the inherent flexibility necessary for functional adaptation. α-helices and β-sheets are stable areas (RMSF < 1.0 Å), which preserve structural integrity (see Fig. 18).

Fig. 18.

Residue-wise Root Mean Square Fluctuation (RMSF) profiles of apo protein PDB:1w11 and its ligand-bound complexes with C878-1660 and D685-0061 over 500 ns molecular dynamics simulations. The top-left panel depicts the RMSF profile of the unbound apo protein, showing intrinsic residue flexibility with peaks corresponding to solvent-exposed loops and termini. The top-right panel illustrates the RMSF for the C878-1660_dock complex, demonstrating reduced fluctuations in regions proximal to the binding site, indicative of ligand-induced stabilization without major global rigidity changes. The bottom panel shows the RMSF profile of the D685-0061_Dock + PDB:1w11 complex, where highlighted regions (green bars) reveal increased residue-level fluctuations, suggesting enhanced local flexibility or destabilization associated with ligand binding.

RMSF is decreased close to the binding pocket upon ligand C878-1660 engagement, suggesting localised stabilisation via van der Waals, hydrophobic, and hydrogen bonding interactions. In remote areas, this ligand-induced stiffness restricts backbone motion without compromising flexibility, indicating localised stabilisation as opposed to global stabilisation.This selective modulation is consistent with the maintenance of the protein’s functional dynamics, facilitating conformational states favorable for biological activity or allosteric communication. The dynamic restraint exerted by C878-1660 binding thus reflects an optimized interaction mode that stabilizes the native fold without compromising the overall conformational landscape.In stark contrast, binding of the D685-0061 ligand profoundly alters the RMSF profile by amplifying residue-level fluctuations across multiple protein regions, suggesting increased conformational plasticity or potential destabilization upon ligand engagement. The D685-0061 complex has much more flexibility than the apo and C878-1660-bound forms, according to the RMSF study, with residues varying over 3.0 Å in areas related to allosteric regulation and ligand binding. These increased movements are consistent with ligand displacement and decreased stability and suggest weaker ligand–protein interactions, local destabilisation, or incomplete unfolding. By stabilising important residues via advantageous hydrogen bonding and hydrophobic interactions, C878-1660 binding, on the other hand, reduces fluctuations close to the binding pocket. According to comparative profiles, D685-0061 encourages disorder and structural heterogeneity, while C878-1660 increases stiffness and conformational stability. Mechanistically, D685-0061’s inadequate fit promotes solvent exposure and destabilises secondary structures, whereas C878-1660’s high complementarity restricts entropy and maintains a well-defined pocket. In general, ligand chemistry clearly influences protein dynamics stabilization by C878-1660 against destabilisation by D685-0061 and offers important information for medication design that aims to dynamically modulate the flexibility of uPAR proteins.

PDB:1w11 is stabilised by ligand C878-1660 binding, which supports conformational stability by decreasing flexibility in important residues while preserving overall dynamics. On the other hand, D685-0061 causes destabilisation and decreased complex integrity via increasing residue fluctuations. The C878-1660 complex’s high affinity is driven by a dense network of non-covalent contacts, as seen in the 2D interaction diagram. Important residues like HIS57, CYS58, and HIS99 improve ligand orientation and selectivity by forming strong hydrogen bonds (< 3.5 Å). Through van der Waals interactions, hydrophobic residues (LEU97, VAL41, TYR40) stabilise the ligand’s aromatic and cyclohexyl groups, while water-mediated bridges involving GLY193, GLY216, and ARG217 aid in dynamic stabilisation. C878-1660’s heterocyclic moiety and core aromatic scaffold allow for multipoint binding across the active site, enhancing selectivity and affinity. These results demonstrate how structure-based drug design for targeted uPAR inhibition may be guided by optimised ligand complementarity, which can also improve stability(See Fig. 19).

Fig. 19.

Detailed 2D interaction map and interaction fraction bar plot of the ligand C878-1660 docked to the protein PDB:1w11. The 2D interaction diagram illustrates key hydrogen bonds, hydrophobic contacts, and water-mediated interactions between the ligand and surrounding residues, highlighting critical binding site amino acids including HIS 57, CYS 58, VAL 41, and GLY 193. Water molecules facilitate indirect interactions, enhancing ligand stabilization within the active site. The accompanying bar plot quantifies the interaction fractions observed during molecular dynamics simulations, with HIS 57 exhibiting the highest interaction frequency, followed by notable contributions from CYS 58, VAL 41, TRP 216, and other residues.

By displaying residue-level contact frequency throughout the simulation, the bar plot enhances the 2D interaction maps and emphasises the endurance of the interaction between C878-1660 or D685-0061 and PDB:1w11. HIS57 has the strongest contact (fraction > 1.0) for C878-1660, which is in line with its crucial catalytic function. Further stability is supplied by CYS58, VAL41, and TRP216 (0.2–0.3), and ligand placement is facilitated by the temporary connections formed by LEU97 and ARG217. This suggests a well-balanced binding mechanism that combines hydrophobic packing and hydrogen bonding, further stabilised by solvent-mediated interactions that improve affinity and residence duration. The 2D map and interaction bar plot for D685-0061 show a dynamic, intricate network of hydrophobic interactions and hydrogen bonds (see Fig. 20). The ligand is anchored by key residues ASN145, SER146, GLN192, ARG217, and GLY219 via water-mediated interactions and long-lasting hydrogen bonds. GLY219 has the greatest interaction percentage (~ 0.9), indicating flexibility and almost constant contact in solvent-exposed loops. SER146 and GLN192 provide adaptive support, whereas ARG217 and ASN145 (0.4–0.5) act as secondary stabilisers. The binding interface as a whole exhibits ligand-specific adaptation, with D685-0061 forming larger, weaker, and more dynamic connections and C878-1660 achieving stronger, more localised stabilisation. This explains their varying impacts on ligand effectiveness and protein stability (See Fig. 20).

Fig. 20.

2D interaction diagram and interaction fraction bar plot of the ligand D685-0061 docked to the protein PDB:1w11. The 2D map illustrates key hydrogen bonds and water-mediated contacts between the ligand and critical residues such as ASN 145, SER 146, ARG 217, GLY 219, and GLN 192, highlighting their roles in stabilizing the ligand within the binding pocket. The presence of water molecules facilitates indirect interactions, enhancing the ligand-protein interface. The bar plot quantifies the frequency of these interactions throughout the molecular dynamics simulation, with GLY 219 and ARG 217 showing the highest interaction fractions, indicating persistent ligand engagement.

Instead of a few prominent interactions, a network of moderate-strength hydrogen bonds, hydrophobic contacts, and water-mediated bridges sustains the binding of D685-0061 to PDB:1w11. ASN145 and SER146 help with flexibility and adaptability, whereas key residues like GLY219 and ARG217 attach the ligand and have the largest interaction percentages. The ligand’s nonpolar portions are stabilised by hydrophobic residues such as LEU222 and VAL213, which enhance affinity and complement polar interactions. Water molecules act as a bridge, enhancing solvent compatibility and dynamic flexibility. Consistent ligand coordination in D685-0061 is highlighted by ligand interaction timelines spanning 500 ns, which show prolonged hydrogen bonding with ASN145, SER146, GLN192, ARG217, and GLY219 as well. SER146, LEU222, and VAL213 are involved in hydrophobic and water-mediated interactions that provide extra stability and flexibility. The ligand stays firmly fixed even if hydrogen bonds sometimes change. Strong and stable binding inside the active site is ensured by C878-1660’s more robust interaction network, which is dominated by long-lasting hydrogen bonds and water bridges with HIS57 and CYS58. With C878-1660 generating greater anchoring and D685-0061 displaying more dynamic flexibility, our results together demonstrate how ligand-specific interaction patterns govern stability and binding kinetics information that is useful for improving uPAR-targeted drug design (See Fig. 21).

Fig. 21.

Comparative timeline plots of key molecular dynamics parameters for the C878-1660 and D685-0061 ligand-bound complexes with PDB:1w11 over 500 ns simulation. Each panel displays time series of protein RMSD, radius of gyration (Rg), intramolecular hydrogen bonds, molecular surface area (MolSA), solvent-accessible surface area (SASA), and polar surface area (PSA).

In keeping with its catalytic function in ligand recognition and inhibition, the C878-1660 complex exhibits a sustained association with HIS57. Water-mediated bridges via GLY193, GLY216, and ARG217 improve interaction stability, whereas hydrophobic interactions with VAL41 and TRP216 sporadically stabilise nonpolar areas. In contrast to D685-0061, C878-1660 shows a variety of dynamic interactions, often alternating between different forms of contact, indicating a stable but adaptable binding mechanism with possible entropic benefits. Key residues are engaged by both ligands, but C878-1660 exhibits stronger, more extensive, and longer-lasting contacts, suggesting a greater affinity. Conversely, D685-0061 exhibits sporadic contacts, which is consistent with its weaker stability and higher RMSF/RMSD ratios. Analysis of the torsion profiles further separates the two ligands. Through advantageous interconversion between low-energy states, C878-1660 exhibits a number of rotatable bonds with wide angular distributions, demonstrating structural flexibility that improves binding adaptability and affinity (See Fig. 22). Stable hydrogen bonding patterns and optimal fitting inside the binding pocket are supported by this flexibility. On the other hand, D685-0061 has a more stiff conformation and limited flexibility due to its narrow torsional ranges and fewer rotatable bonds. These results together demonstrate how the dynamic interaction network and conformational diversity of C878-1660 support sustained suppression of uPAR, while the stiffness of D685-0061 correlates to weaker, temporary binding, providing important information for structure-based medication optimisation (See Fig. 22).

Fig. 22.

Torsion angle distribution and conformational analysis of ligand C878-1660 docked to UPAR protein (PDB:1w11). The plots display histograms and polar representations of the key rotatable bonds within the ligand structure, illustrating the range of torsion angles sampled during molecular dynamics simulations.

Limited torsional flexibility is shown by ligand D685-0061, which binds in a stable orientation to reduce entropic costs but limit flexibility in response to protein motions. Constrained rotation is confirmed by polar plots, which show concentrated torsion angles. The binding effectiveness may be impacted by this stiffness, which preserves important interactions but restricts conformational shift accommodation. This behaviour may be explained by the ligand’s stiff aromatic–amide connections and few rotatable bonds. The torsional flexibility of C878-1660, on the other hand, allows for dynamic adaptability and numerous contact networks, which enhance stability and broaden conformational compatibility with uPAR ( see Fig. 23).

Fig. 23.

Torsion angle distribution and conformational dynamics of ligand D685-0061 docked with UPAR protein (PDB:1w11). The torsion plots display histograms and polar projections representing key rotatable bonds within the ligand, illustrating restricted conformational sampling characterized by predominant torsion angle populations.

Torsional study of C878-1660 and D685-0061 demonstrates how ligand effectiveness is influenced by flexibility as opposed to stiffness. The wide torsional range of C878-1660 increases affinity by encouraging adaptive binding and entropic stabilisation. The stiffness of D685-0061 restricts adaptation while favouring enthalpic accuracy. Optimising uPAR-targeted ligand affinity, specificity, and pharmacokinetic performance requires striking a balance between torsional flexibility.

Principal component analysis (PCA) and free energy landscape study

Based on all-atom molecular dynamics simulations, Principal Component Analysis (PCA) and Free Energy Landscape (FEL) techniques were used to examine the conformational and energetic impacts of ligands C878-1660 and D685-0061 on uPAR (see Figs. 24 and 25). By capturing dominating functional movements, PCA simplified the complex configurational space into a crucial subspace. The conformational ensemble of the uPAR–C878-1660 complex was strongly clustered, with the top three main components explaining more than 60% of the variation. While low cosine content (< 0.3) revealed steady, convergent essential dynamics typical of a well-defined conformational state, its steep eigenvalue decay suggested a low-dimensional landscape dominated by a single deep energy basin. With eigenvalues diffused over many dimensions, the uPAR-D685-0061 complex displays dispersed main component projections, suggesting more conformational flexibility and variability. Dynamic diversity within its conformational ensemble is further confirmed by increased cosine content and decreased rmsip overlap across simulated replicates.

Fig. 24.

Principal Component Analysis (PCA) scatter plots depicting the conformational sampling of the urokinase plasminogen activator receptor (uPAR) bound to ligand C878-1660. The left panel illustrates the projection of molecular dynamics simulation frames onto the first and second principal components (PC2 vs. PC1), while the right panel shows the projection onto the third and second principal components (PC3 vs. PC2). The colour gradient represents simulation progression over time (frame index), indicating the trajectory of conformational states sampled during the MD simulation. The distinct clustering and trajectory continuity suggest a restricted conformational landscape stabilized by ligand binding.

Fig. 25.

Two-dimensional and three-dimensional representations of the Free Energy Landscape (FEL) for the urokinase plasminogen activator receptor (uPAR) complexed with ligand C878-1660. The left panel shows a 2D FEL contour map plotted as a function of root mean square deviation (RMSD) and radius of gyration (Rg), with Gibbs free energy values (kJ/mol) indicated by the colour gradient. The right panel presents a 3D FEL surface plot with the same reaction coordinates and energy scale, illustrating the energetic basins and barriers governing conformational stability. The prominent global minimum region (blue) indicates the most thermodynamically favorable conformational state, signifying ligand-induced stabilization of uPAR’s structural ensemble.

On the other hand, the uPAR-C878-1660 complex exhibits a single deep global minimum and a small number of shallow local minima, indicating a kinetically constrained, thermodynamically stable state. According to its free energy landscape, C878-1660 binding stabilises the receptor’s binding pocket and secondary structures while lowering conformational entropy to favour a dominant ligand-bound configuration by confining uPAR’s conformational space through strong hydrogen bonds and hydrophobic interactions (See Fig. 26).

Fig. 26.

Principal Component Analysis (PCA) projections illustrating the conformational sampling of the urokinase plasminogen activator receptor (uPAR) in complex with ligand D685-0061. The left panel shows the distribution of molecular dynamics simulation frames along the first and second principal components (PC2 vs. PC1), while the right panel displays the projection along the third and second principal components (PC3 vs. PC2). The color gradient corresponds to simulation time (frame index), representing the progression of sampled conformational states. The broader distribution and multiple clusters suggest increased conformational heterogeneity and dynamic flexibility induced by ligand binding.

The FEL of D685-0061 exhibits several shallow basins that are divided by low barriers, facilitating swift transitions between metastable uPAR conformations. This indicates an improvement in structural plasticity and provides support for mechanisms such as induced-fit or conformational selection. The analysis of RMSD-based clustering supports this observation: C878-1660 confines uPAR to a limited number of compact, stable clusters characterised by low variability in the domains critical for uPA recognition. In contrast, D685-0061 generates numerous smaller, heterogeneous clusters that are predominantly located in flexible loops and inter domain regions. Temporal PCA–FEL mapping reveals that C878-1660 restricts uPAR to a single predominant basin, whereas D685-0061 facilitates ongoing transitions between basins. This suggests a contrast between enthalpy-driven stabilisation and entropy-enhanced dynamics (See Fig. 27).

Fig. 27.

Two-dimensional and three-dimensional Free Energy Landscape (FEL) representations for the urokinase plasminogen activator receptor (uPAR) bound to ligand D685-0061. The 2D FEL contour map (left) displays Gibbs free energy as a function of root mean square deviation (RMSD) and radius of gyration (Rg), with lower free energy regions highlighted in blue indicating thermodynamically favoured conformations. The 3D FEL surface plot (right) illustrates the energetic landscape topology, revealing multiple minima and a rugged energy surface consistent with increased conformational heterogeneity and dynamic flexibility induced by ligand binding.

The uPAR–D685-0061 complex exhibits a balance of moderate enthalpic interactions alongside increased conformational entropy, facilitating a flexible binding interface that could support allosteric regulation or transient interactions. Comprehensive multi-replica molecular dynamics sampling, coupled with validated convergence, offers a reliable perspective on the dynamics of uPAR. The analyses conducted through PCA and FEL indicate that C878-1660 functions as a structural clamp, thereby stabilising a singular conformation that is conducive to prolonged inhibition. In contrast, D685-0061 promotes a wider range of conformational diversity, which is associated with partial agonism or allosteric effects. The findings emphasise the influence of ligand chemistry on the energy landscape of uPAR and illustrate the importance of dynamic modelling in the design of selective, functionally optimised uPAR-targeting molecules.

In-Vitro cell line (MDA-MB-231) MTT assay

The ligands D685-0061 and C878-1660 demonstrate unique cytotoxic effects in MDA-MB-231 cells. D685-0061 exhibits significantly increased potency, evidenced by an IC₅₀ of approximately 21 µM, aligning with QSAR-based predictions, and leads to notable morphological degeneration at low micromolar concentrations (see Fig. 28). In contrast, C878-1660 exhibits reduced cytotoxicity (IC₅₀ ~82 µM) despite demonstrating a favourable docking affinity, which likely indicates limited efficacy within the intracellular environment. Morphological disruption induced by C878-1660 occurs predominantly around its IC₅₀ (see Fig. 29), while D685-0061 initiates early cell rounding, detachment, and nuclear condensation, leading to significant monolayer loss at elevated concentrations. The results demonstrate that D685-0061 exhibits a greater capacity to inhibit uPAR-mediated oncogenic signaling.

Fig. 28.

Morphological changes in MDA-MB-231 breast cancer cells treated with increasing concentrations of ligand D685-0061 (5–320 µM). Progressive cell rounding, detachment, and fragmentation indicate dose-dependent cytotoxicity. High doses result in extensive cell death and loss of cell viability.

Fig. 29.

Dose-response curve showing normalized cell inhibition by compound D685-0061, with an IC₅₀ value of 21.34. The plot illustrates decreasing cell viability as concentration increases.

The activity of the ligand is likely a result of engagement at the active site or through allosteric mechanisms, leading to effective inhibition of uPAR and a decrease in cell viability. D685-0061 induces significant cytotoxic effects and morphological alterations, thereby reinforcing its potential as a primary inhibitor that necessitates further kinetic, signalling, apoptotic, and in vivo validation. The integration of transcriptomic and proteomic data has the potential to elucidate response pathways and mechanisms of resistance. The increased IC₅₀ of C878-1660 indicates inadequate binding or insufficient receptor blockade, highlighting the necessity for structural optimisation.

In line with QSAR predictions that incorporate multidimensional chemical-biological aspects beyond static docking estimations, D685-0061 exhibits better pharmacological performance than C878-1660. According to morphological data, D685-0061 activates caspase mediated and mitochondrial cell-death processes by causing cytoskeletal collapse, membrane blebbing, and the production of apoptotic bodies (see Fig. 30). Strong uPAR inhibition, suppression of uPA-driven proteolysis, and disruption of adhesion and cytoskeletal networks, which makes cells more susceptible to apoptosis, are probably the causes of its increased cytotoxicity. Its improved anti-proliferative effect on MDA-MB-231 cells and lower IC₅₀ are confirmed by MTT results and morphological alterations, indicating its therapeutic potential (see Fig. 31).

Fig. 30.

Morphological changes in MDA-MB-231 breast cancer cells treated with increasing concentrations of ligand C878-1660 (5–320 µM). Progressive cell rounding, detachment, and fragmentation indicate dose-dependent cytotoxicity. High doses result in extensive cell death and loss of cell viability.

Fig. 31.

Dose-response curve showing normalized cell inhibition by compound D685-0061, with an IC₅₀ value of 21.34. The plot illustrates decreasing cell viability as concentration increases.

Conversely, while C878-1660 demonstrates measurable biological activity, its comparatively higher IC₅₀ and subdued morphological effects necessitate further refinement to achieve clinically relevant efficacy. These insights reinforce the critical importance of integrated computational-experimental workflows in identifying and validating potent uPAR inhibitors with therapeutic potential against metastatic breast cancer phenotypes.

MTT assay using A431 skin cancer cell lines

The evaluation of cytotoxic efficacy of ligands D685-0061 and C878-1660 on the A431 epidermoid carcinoma cell line was performed using the MTT assay, a colorimetric assay that quantifies cell viability by measuring the enzymatic reduction of 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) to insoluble formazan crystals by mitochondrial succinate dehydrogenase. The reduction in formazan formation correlates directly with loss of mitochondrial metabolic activity, thereby serving as a surrogate marker for cell viability and proliferation. C878-1660 exhibited a potent inhibitory effect with an IC₅₀ of 18.93 µM, indicating a relatively high efficacy in impairing A431 cell viability (See Fig. 32).

Fig. 32.

Dose-dependent inhibition of A431 skin cancer cell viability by ligand C878-1660. The graph illustrates the normalized percentage of cell inhibition relative to control across increasing concentrations of C878-1660. The calculated IC₅₀ value of 18.93 µM indicates the concentration required to reduce cell viability by 50%, demonstrating potent cytotoxic activity in the MTT assay. Data are presented as mean ± SEM from triplicate experiments.

A431 control cells exhibited a robust epithelial morphology characterised by strong adhesion and a well-preserved cytoskeletal structure. The administration of C878-1660 at concentrations of 5–10 µM resulted in early indicators of cytotoxicity, such as decreased cell spreading and mild blebbing. These observations suggest a disruption of focal adhesions and the integrin-FAK survival signalling pathway. At a concentration of 20 µM, the cells displayed distinct apoptotic characteristics, including rounding and detachment, which align with the observed actin collapse and the process of anoikis. Increased concentrations (40–80 µM) resulted in enhanced apoptosis characterised by cellular condensation, pyknosis, and fragmentation, which can be attributed to mitochondrial dysfunction and the activation of caspases. At a concentration of 160 µM, significant rupture and debris indicated the occurrence of late-stage apoptosis or secondary necrosis (See Fig. 33).

Fig. 33.

Morphological effects of C878-1660 on A431 skin cancer cells at increasing concentrations (5 µM to 160 µM). Phase-contrast microscopy images reveal progressive cytotoxicity characterized by reduced cell density, cell rounding, membrane blebbing, and loss of adherence with increasing ligand concentration.

In contrast, D685-0061 demonstrated a somewhat diminished cytotoxic potency, with an IC₅₀ of 28.34 µM(see Fig. 34), requiring higher concentrations to elicit similar anti-proliferative effects. Morphological examination revealed a delay in the onset of apoptosis-associated changes. At 5 µM and 10 µM, A431 cells retained the majority of their epithelial characteristics, though mild cellular stress manifested as decreased spreading and early membrane ruffling was apparent.

Fig. 34.

Dose-dependent inhibition of A431 skin cancer cell viability by ligand D685-0061. The graph displays normalized cell inhibition percentages as a function of increasing ligand concentration on a logarithmic scale. The IC₅₀ value of 28.34 µM indicates the concentration at which 50% of the cell population is inhibited, demonstrating moderate cytotoxic activity in the MTT assay. Data represent mean ± SEM from triplicate experiments.

The observed cytomorphological changes for both compounds align with well-established apoptotic processes involving the intrinsic (mitochondrial) pathway, characterized by loss of mitochondrial membrane potential, release of apoptogenic factors (cytochrome c, Smac/DIABLO), and subsequent activation of caspase cascades (See Fig. 35). The associated disruption of cytoskeletal architecture and focal adhesion complexes underscores the role of integrin signaling in maintaining cell survival and anchorage-dependent growth, whose inhibition by these ligands contributes to anoikic and cell death.

Fig. 35.

Morphological changes in A431 and MDA-MB-231 cancer cell lines treated with increasing concentrations of ligand D685-0061 (5 µM to 160 µM). Phase-contrast microscopy reveals dose-dependent cytotoxicity characterized by decreased cell confluence, cell rounding, membrane blebbing, and detachment from the culture substrate.

Strong adhesion, intact cytoskeletal integrity, and good epithelial shape were all shown by A431 control cells. Early cytotoxic indications, such as decreased spreading and moderate blebbing, were brought on by treatment with C878-1660 at 5–10 µM. These signs suggested that focal adhesions and integrin-FAK survival signalling were disrupted. Cells showed distinct apoptotic characteristics, including rounding and detachment, at 20 µM, which are consistent with anoikis and actin collapse. Because of caspase activation and mitochondrial malfunction, higher dosages (40–80 µM) caused advanced apoptosis characterised by condensation, pyknosis, and fragmentation. Debris and widespread rupture at 160 µM indicated secondary necrosis or late-stage apoptosis. In addition, in molecular dynamics simulations, C878-1660 forms a more stable uPAR complex with lower RMSD values, constrained conformational drift, and enthalpically stabilized binding pockets than D685-0061, but its lower potency in TNBC (MDA-MB-231) cells may be due to downstream biological factors not captured by MD. The study reveals that D685-0061 has higher cytotoxicity, causing apoptotic morphology at lower doses (IC₅₀ = 21.34 µM vs. 81.82 µM for C878-1660). D685-0061 may better interfere with TNBC cells’ uPAR-dependent signaling networks, such as integrin/FAK pathways, cytoskeletal adhesion complexes, and mitochondrial apoptotic machinery, than C878-1660, which may have reduced cellular uptake, less efficient receptor engagement in TNBC, or limited intracellular distribution. Thus, cell permeability, efflux susceptibility, metabolic stability, or off-target regulation of TNBC-specific pathways may explain the computational-experimental discrepancy.

Conclusion

The integrated computational-experimental framework facilitated the identification of two promising uPAR inhibitors, C878-1660 and D685-0061, via a workflow that combined robust QSAR modeling, virtual screening, molecular docking, long-timescale MD simulations, and in-vitro cytotoxicity assays. The QSAR model identified crucial structural factors influencing uPAR inhibition, facilitating the selection process from a library of 30,000 compounds. Additionally, docking studies and 500-ns OPLS4/TIP3P simulations validated the stability of binding dynamics, especially for C878-1660. The differential cytotoxicity observed in MDA-MB-231 and A431 cells provides additional validation of their biological significance. Future research should concentrate on optimizing structure–activity relationships based on the identified descriptors, expanding to include analogue libraries, and conducting free-energy perturbation analyses to enhance potency predictions. It is essential to conduct validation in 3D spheroid and co-culture models, followed by in-vivo pharmacokinetic and anti-metastatic studies. Furthermore, evaluating the impacts on uPAR-regulated pathways, markers of epithelial-mesenchymal transition (EMT), and proteolytic signaling will enhance our understanding of the mechanisms involved and facilitate advancement toward preclinical development.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Harshal Badukle : Conceived, drafting, review, and supervised this study.Rahul D. Jawarkar : Conceived, drafting, review, and supervised this study.Umang Shah : created images, writing and drafted manuscript.Somdutta Chaudhary : Collected and curated the data.Abdullah Yahya Abdullah Alzahrani : created images, writing and drafted manuscript.Abdul Samad : created images, writing and drafted manuscript.Sami A. Al-Hussain : collected and curated the data.Aamal A. Al-Mutairi : collected and curated the data.Magdi E.A. Zaki : Conceived, drafting, review, and supervised this study.

Funding

This work was supported and funded by the Deanship of Scientific Research at Imam Mohammad Ibn Saud Islamic University (IMSIU) (grant number IMSIU-DDRSP2601).

Data availability

The data of molecules used in the present work are available as supplementary material files; S1 to S6.

Declarations

Competing interests

The authors declare no competing interests.

Cell line studies & source of cell line

The MMT assay was carried out at Department of Pharmaceutical Chemistry, Ramanbhai Patel College of Pharmacy, CHARUSAT Campus, Changa, Gujarat 388421. The source of cell line used in the study were procured from the National Center for Cell Science, Pune.