2.1 Rational design of the derivatives
To design novel uPAR inhibitors, we first investigated the binding patterns between uPAR and 22 through a 200 ns-long all-atom MD simulation in aqueous solution. The initial structure pf uPAR:22 complex for the MD simulation was obtained from our previous work [12, 18-20]. We assessed the convergence of simulation via calculating the changes of various structural dynamics-related properties across simulation time, including root-mean-square deviations (RMSD) of backbone or ligand heavy atoms, secondary structure elements, radius of gyration (Rg), solvent-accessible surface area (SASA), and root-mean-square fluctuation (RMSF) of the complex. The results indicated that the interaction with 22 did not apparently alter the conformation of uPAR (Figs. 1A and S1). The RMSD values of 22 heavy atoms converged rapidly at 50 ns with a minor fluctuation (Fig. 1A).
We evaluated the binding contribution of each residue to the binding of 22 to uPAR using MD-based molecular mechanics generalized Bourne surface area (MM-GBSA) [21] to identify the critical interactions between uPAR and 22. The major contributed residues to the binding (≤ −0.5 kcal/mol) were given in Fig. 1B. Notably, R137 was the most substantial contributor to the binding, with an energy contribution of −4.6 kcal/mol, followed by R53 with −2.9 kcal/mol. This was in well agreement with our interaction pattern analysis on the most representative conformation of the complex from the MD simulation (Fig. 1C), which demonstrated that the acetoxy and carbonyl groups of 22 form hydrogen bonds with R137 and R53, respectively. Of note, R53 contacts with uPA Y24 in the crystal structure of the uPAR-uPA-vitronectin complex [22], implying it is a potential binding site to inhibit uPAR-uPA interactions. In addition, P138, L55, and L150 were identified as the most critical hydrophobic residues in uPAR:22 interface, with energy contributions of −2.5, −2.2, and −1.8 kcal/mol, respectively. These residues established a hydrophobic patch interacting with the phenyl rings of 22. This patch is essential for the high affinity and low off-rates in uPAR antagonist peptides and uPA [23-25].
MD simulation demonstrated that 22 made comprehensive interactions with uPAR, highlighting the significant potential of 22 as a lead compound. Meanwhile, there was room for the improvement of the interface patterns between uPAR and 22. For instance, although the acetoxy, carbonyl, and phenyl groups of 22 directly interacted with the binding site residues, the dimethyl ethylamine moiety of 22 (Fig. 1C) did not form any strong and stable interactions with uPAR. Moreover, the binding pocket of 22 consisted of both hydrophilic and hydrophobic residues (Fig. 2A). Based on these findings, the dimethyl ethylamine moiety of 22 was substituted by other groups with different molecular weights, hydrophobicity, acidity, and basicity to optimize the interactions between uPAR and 22 (Fig. 2A). Finally, we designed 68 derivatives of 22 for the following hierarchical binding free energy-based virtual screening (Table. S1).
2.2 Hierarchical binding free energy calculations
The end-point binding free energies of the derivatives bound to uPAR were initially predicted using 200 ns-long MD-based MM-GBSA calculations (13.6 ms-long simulations in total). As indicated in Table S1, 35 derivatives showed more favorable affinities (ΔGMM-GBSA) to uPAR than 22 (−41.9±4.1 kcal/mol). To more accurately predict the binding free energy differences (ΔΔG) between the top 35 derivatives and 22, we further utilized alchemical thermodynamic integration (TI) algorithm. The ΔΔG values for these derivatives ranged from −3.7 to 3.1 kcal/mol (Table S2). Notably, the ΔΔG values for derivatives 221-8, 221-12, 221-17, 221-57, and 221-68 showed a reduction of at least 3 kcal/mol compared to 22 (Fig. 2B). Subsequently, these top five derivatives were selected for further analysis using localized volume-based metadynamics (LV-MetaD).
While the MM-GBSA and TI methods can evaluate binding affinity, they have limitations in capturing induced-fit effects. To overcome this, we have developed an enhanced sampling-based approach, called LV-MetaD [17]. This approach has been successfully applied in several flexible protein-ligand complexes with a discrepancy of around 0.5 kcal/mol between experimental and calculated values. We first calculated the binding free energy (ΔGLV-MetaD) of the binding of 22 to using this method to evaluate its performance in this case. The multiple binding/unbinding recrossing events were observed on uPAR:22 complex during a 4 μs-long LV-MetaD simulation, highlighting the satisfactory binding/unbinding sampling (Fig. S2). This was further corroborated by the convergence of one-dimensional free energy profiles, measured as a function of the center-of-mass distance between uPAR DII backbone atoms and the 22 heavy atoms (ρ), across different simulation timescales (Fig. S2). Additionally, we reweighted the potential bias to two collective variables (CVs): ρ and the number of intermolecular hydrogen bonds to construct a two-dimensional free energy landscape of 22 binding/unbinding process. As shown in Fig. 3A, the landscape featured a prominent minimum energy basin (depicted in dark blue). The energy difference between bound and unbound states determined the binding free energy between 22 and uPAR. The calculated ΔGLV-MetaD for 22 was −6.7±0.2 kcal/mol, aligning well with our previous experimental data measured by surface plasmon resonance (−6.1±0.1 kcal/mol) [12], thereby underscoring the precision of LV-MetaD in this work.
Subsequently, we applied the same methodology on uPAR in complexes with the top five derivatives (Figs. S2 and Fig. 3). The derivatives (221-8, 221-12, 221-17, 221-57, and 221-68) demonstrated more favorable binding affinities to uPAR than 22, with ΔGLV-MetaD values of −9.5±0.3, −9.3±0.4, −9.1±0.4, −7.6±0.2, and −8.5±0.3 kcal/mol, respectively (Fig. 3G), suggesting that the five derivatives have significant potential as potent uPAR inhibitors.
2.3 Predictions of ADMET properties
Predicting ADMET (absorption, distribution, metabolism, excretion, and toxicity) properties provides valuable information about the safety, efficacy, and overall pharmacokinetic behavior of a drug candidate in drug development [26-28]. We used the SwissADME webserver to predict the ADMET properties of the five derivatives [29]. As shown in Table 1, 221-68 showed more favorable aqueous solubility (LogS) than 22, with values of -3.73 and -3.96, respectively. Both 221-68 and 221-12 exhibited octanol/water partition coefficient (LogP) values indicative of orally administered drugs, recorded at 2.80 and 2.81, respectively. Furthermore, all compounds, except 221-12, demonstrated high gastrointestinal (GI) absorption potential. None of the compounds showed a significant probability of crossing the blood-brain barrier (BBB). Adherence to Lipinski's rules of drug-likeness [30], was observed in all compounds, except for 221-17. the Pan-Assay Interference Compounds (PAINS) alert system identified no false positives in the remaining compounds, with 221-17 being the exception. Consequently, due to their favorable ADMET profiles, 221-8, 221-57, and 221-68 were selected for following synthesis and biological evaluation.
Table 1. The drug physicochemical and pharmacokinetic properties for 22 and top five derivatives predicted by the SwissADME webserver.
|
MWa (Da)
|
LogSb
(g/L)
|
ALogPc
|
GId absorption
|
BBBe permeant
|
Lipinski
|
PAINf
|
22
|
415.53
|
-3.96
|
2.81
|
High
|
No
|
Yes
|
0 alert
|
221-68
|
415.53
|
-3.73
|
2.80
|
High
|
No
|
Yes
|
0 alert
|
221-17
|
503.57
|
-4.94
|
3.44
|
High
|
No
|
1 violation
|
1 alert
|
221-57
|
457.54
|
-4.31
|
3.41
|
High
|
No
|
Yes
|
0 alert
|
221-12
|
494.52
|
-5.11
|
2.97
|
Low
|
No
|
Yes
|
0 alert
|
221-8
|
491.56
|
-5.27
|
3.91
|
High
|
No
|
Yes
|
0 alert
|
a Molecular Weight.
b Solubility in aqueous solution.
c Concentration of a ligand in octanol/concentration of drug in aqueous solution.
d Gastrointestinal adsorption.
e Blood-brain-barrier permeant.
f Pan-assay interference compounds
2.4 In vitro evaluation
The detail information of the synthesis for 221-8, 221-57, and 221-68 could be found in the Supporting Information. To evaluate the anti-metastatic and the uPAR-targeting abilities of these derivatives, we utilized uPAR high-expressing human prostate cancer cells (PC-3) and uPAR low-expressing human breast cancer cells (MDA-MB-231) in cytotoxicity assays and Transwell invasion experiments.
22 and 221-8 exhibited slight dose-dependent cytotoxicity in PC-3 cells (Fig. 4A and Table S3). In contrast, 221-57 and 221-68 did not show cytotoxicity in PC-3 cells up to a concentration of 100 μM. These results indicated that none of the compounds significantly affected the viability of PC-3 cells. Transwell invasion assays on PC-3 cells (Figs. 4B-C and Table S4) revealed that all four compounds significantly inhibited PC-3 cell invasion in a dose-dependent manner. At a concentration of 100 μM, 221-68 and 221-8 reduced PC-3 cell invasion by approximately 65% and 63%, respectively, outperforming 22, which achieved around 55% inhibition. While 221-57 also inhibited PC-3 cell invasion in a dose-dependent manner, its effectiveness was the lowest, reducing cell invasion by about 45% at 100 μM. Analysis of dose-dependent fitting curves for PC-3 cell invasion inhibition (Fig. 4D) enabled the determination of the half-maximal inhibitory concentration (IC50) values. Fig. 4D and Table S4 revealed that the inhibitory activity of 221-68 was more than double that of 22, with the IC50 values of 25.02±8.46 μM and 54.13±0.02 μM, respectively. 221-8, with the IC50 value of 42.02±0.13 μM, showed about a 30% enhancement in efficacy compared to 22. These findings suggested that each of the three derivatives possesses anti-invasive properties against PC-3 cells, and derivatives 221-68 and 221-8 outperformed 22 in terms of their anti-tumor invasion effectiveness, which underscores the validity of our drug design approach.
In the case of MDA-MB-231 cells, 22 was the only one to demonstrate modest growth inhibitory activity (Fig. 5A and Table S5). At a concentration of 100 μM, 22 inhibited MDA-MB-231 cell growth by approximately 33%. The other three compounds did not show cytotoxic effects on MDA-MB-231 cells. Transwell invasion assays for MDA-MB-231 cells (Figs. 5B-C and Table S6) indicated a mild dose-dependent inhibitory impact of the four compounds on cell invasion. Fig. 5D illustrated the dose-dependent fitting curves for the invasion inhibition of MDA-MB-231 cells by 22, 221-8, 221-57, and 221-68, with the corresponding IC50 values of 233.70±7.00 μM, 329.20±87.40 μM, 457.30±219.60 μM, and 160.45±24.40 μM, respectively (detailed in Table S6). The anti-metastatic effects of 22 and its derivatives on MDA-MB-231 cells, characterized by low uPAR expression, were markedly less potent than those on PC-3 cells featured with high uPAR expression (Fig. S8), identifying the uPAR-targeting ability of these compounds.
Taken together, in vitro experiments demonstrated that the three derivatives (221-68, 221-8, and 221-57) exerted uPAR-dependent inhibitory effects on cancer cell invasion at the micromolar without cytotoxicity. Additionally, 221-68 and 221-8 displayed superior inhibitory activities compared to 22.
2.5 Binding mechanisms of the derivatives to uPAR revealed by MD simulations
To investigate the molecular mechanisms underlying the enhanced inhibitory activity of 221-68 and 221-8, we extended the all-atom MD simulations on the uPAR:221-68 and uPAR:221-8 complexes from 200 ns to 1000 ns. The sampling and convergence of the MD simulations were evaluated by calculating the RMSD of uPAR in complexes with 221-68 and 221-8 (Figs. 6A-B). The analyses of secondary structural elements, Rg, SASA, and RMSF of uPAR when bound to 221-68 and 221-8 revealed that their binding did not result in significant structural changes in uPAR (Fig. S1). Cluster analysis identified the most representative conformations of 221-68 and 221-8 for simulations (Figs. 6C-D). The interface of these two inhibitors with uPAR was similar to that of 22. Both derivatives formed hydrogen bonds with residues R53 and R137. Moreover, 221-68 established an additional hydrogen bond with the side chain of D254 through its pentane-1-amine group. Hydrogen bonding analysis showed a 0.67 existence probability of this hydrogen bond during the simulation (Fig. 6E). Moreover, the hydrogen bonding stability between R53 and 221-68 was enhanced compared to 22, with its existence probability increasing from 0.43 to 0.93. As a result, 221-68 formed an average of 2.4±0.4 hydrogen bonds with uPAR, which is twice the average number formed by 22 (Fig. 6F), in well agreement with previous findings of Transwell invasion assays. Similarly, 221-8 established a new hydrogen bond with T51 for approximately 43% of the simulation duration, and its hydrogen bonding interactions with R53 were also more stable than those between 22 and R53, with probabilities of 0.63 and 0.43, respectively. Furthermore, the average number of hydrogen bonds between 221-8 and uPAR (1.9±0.5) was significantly higher than that of 22 (1.2±0.3). Consequently, the enhanced binding affinities of 221-68 and 221-8, relative to 22, could be due to their altered functional groups forming additional hydrogen bonds with uPAR. These outcomes once again identified the reasonability of our drug design strategy.