Identification of novel HMs on Cdc42-ITSN signaling
To determine whether this finding from the above reports represented a random encounter of PA or whether biologically significant PAs can be widely captured, we screened the same reported chemical libraries for compounds that display the same or similar PA features employing an Mant-GTP fluorophore-based Cdc42-GEF assay platform. Supporting these studies, ZCL278 promoted GTP loading in the absence of GEF, but inhibited GTP loading in the presence of GEF (Fig. 1A, B). While ZCL367 inhibited Cdc42 GTP loading with or without the presence of an activating GEF (Fig. 1E, F), some compounds display features that go beyond the traditional categories of Cdc42 agonists or antagonists. For example, while ZCL279 did show a trend of increasing GTP loading progressively up to 10 uM concentration, and the experiments have been highly repeatable, the increases were not substantial (Fig. 1C, D). We then calculated the slopes as shown in Fig. S1, which confirmed our observation that ZCL279 promoted Cdc42 GTP loading at lower concentrations and significant inhibition at higher concentrations. The reversal occurred between 10 uM and 100 uM. Experiments with ZCL279 treatment at 50 uM also showed significant inhibition for GTP loading.
This feature of ZCL279 is reminiscent of biphasic, hormetic chemicals14. Using ITSN as reference of endogenous full agonist, ZCL278 and ZCL279 displayed PA features whereas ZCL367 is a bona fide Cdc42 inhibitor or inverse agonist (IA) (Fig. 1G–I). The data from GTPase-Glo™ assay15, which quantitates the amount of GTP remaining after a GTPase reaction, was consistent with HM model compounds as Cdc42-GEF modulators. Therefore, these initial studies provided a foundation with three model classes of Cdc42 modulators, some of which, such as ZCL279, were previously understudied.
Structural simulation of binding mechanisms of different Cdc42-ITSN HM classes
These findings raised the intriguing possibility that there may be more categories of HMs in the existing SMM libraries, and these three model classes of Cdc42 modulators can possibly be defined by their distinct binding mechanisms using computer-aided drug design (CADD) analysis.
We employed SiteMAP of Schrödinger software to identify the top ranked potential protein binding pockets at the interface of Cdc42-ITSN (PDB code: 3QBV). Specific residues and site scores for each binding pocket are summarized (see Supplementary Table S1 and Figs. S2 and S3). For Cdc42 not complexed with ITSN, three different binding pockets were detected, which included the GDP binding site on Cdc42, a region adjacent to the GDP binding site, and the specific residues of the structural motifs “Switch I” (Val36, Phe37) and “Switch II” (Ala59, Tyr64, Leu67, Leu70, Ser71). We named these detected pockets as “GDP”, “Effector”, and “Switch 1&2” pockets, respectively (Fig. 2A). For Cdc42-ITSN complex, besides GDP and Effector pockets, additional three binding pockets were identified including “Cdc42-DH” domain, “DH-PH” domain of ITSN, and “Switch 2 of Cdc42 and DH” domain of ITSN (Fig. 2C).
We then re-docked ZCL278, ZCL279 and ZCL367 to all identified pockets of the Cdc42 and Cdc42-ITSN complex to investigate the binding modes of these model compounds. The first and second preferential binding sites of Cdc42 for these three compounds are shown in Fig. 2B. ZCL278 and ZCL279 bind most preferably to GDP pocket while ZCL367 preferentially binds Effector pocket. The second preferential pockets of Cdc42 for ZCL279, ZCL278 and ZCL367 are Effector, Switch 1&2, and GDP pockets, respectively (Fig. 2B). We designated the binding modes as the preferential binding pocket order (PBPO). In the presence of ITSN, the PBPO for these compounds were different. For ZCL278 and ZCL279, the first preferential binding pocket on Cdc42-ITSN complex is Effector, and the second preferential binding pocket is Cdc42-DH domain of ITSN, while ZCL367 binds to GDP pocket preferably followed by Effector pocket (Fig. 2D). Figure 2E, F showed superimposed images of ZCL278, ZCL279, and ZCL367 on Cdc42 and Cdc42-ITSN complex, respectively. These results revealed that the binding modes of ZCL278 and ZCL279 are alike while ZCL367 interacts with different binding pockets of Cdc42 and Cdc42-ITSN complex from ZCL278 and ZCL279.
The molecular interactions of these model compounds with Cdc42 in the absence and presence of ITSN are further revealed in Fig. 2G–L (see also Supplementary Figs. S2 and S3). These compounds made interactions with a variety of Cdc42 and Cdc42-ITSN residues via hydrogen bond, halogenic and hydrophobic binding (Supplementary Table S2). The docking result showed that the oxygen atom of hydroxyl group and nitrogen amide group of ZCL367 formed hydrogen bonds with Thr43 and Tyr23 of Cdc42 (Fig. 2K) while in the presence of ITSN, two oxygen atoms of ortho of hydroxyl group made interaction with the Tyr64, Glu62 and Asp63 via hydrogen bonds. Furthermore, Lys1390 of ITSN formed three centered hydrogen bonds with nitrogen atom and oxygen carbonyl of ZCL367 (Fig. 2L). Additionally, carbonyl oxygen of ZCL278 interacted with Gly15 and Lys16 of Cdc42 through hydrogen interactions (Fig. 2G). In the presence of ITSN, Asp38 and Asn39 of Cdc42 as well as Lys 1367 of ITSN formed hydrogen bonds with oxygen atom of carbonyl, amide nitrogen atoms of sulfonamides and thiourea moiety of ZCL278, respectively (Fig. 2H). As shown in Fig. 2I, the nitrogen atom of sulfonamide moiety and oxygen atom hydroxyl group of ZCL279 formed an hydrogen bond with Asp118, Gly15 and Lys16 from Cdc42 (Fig. 2I). In the presence of ITSN, oxygen atom of hydroxyl group of ZCL279 formed hydrogen bond with Glu1244 of ITSN. The other three centered H-bonds were formed between oxygen group of ZCL279 and Asn39 of Cdc42 and Lys1367 of ITSN (Fig. 2J). ZCL278 and ZCL279 showed similar interactions with GDP and Effector pockets of Cdc42-ITSN complex. This analysis suggested that in the presence of ITSN, these compounds could make more interactions with Cdc42 compared to that in the absence of ITSN as a GEF. Docking scores of these compounds also confirmed that, in the presence of ITSN, compounds bind to Cdc42 with a higher score compared to those for Cdc42 without ITSN (Supplementary Table S2). Compared to ZCL278, ZCL279 showed stable interactions with both Cdc42 and Cdc42-ITSN. For ZCL279, the ligand-protein interactions are favorable and strong, confirmed and supported by molecular dynamics (MD) studies (see Supplementary Figs. S4–S6). On the other hand, for ZCL278, the ligand-protein interactions are more moderate. The interactions between Cdc42 and ZCL367 are similarly stable and strong regardless of whether GEF was present or not (see Supplementary Figs. S4–S6).
Homeostatic modulating effects on Cdc42 activation in cultured cells
We then investigated the model compounds that represent the distinct classes (ZCL278, ZCL279, and ZCL367) in a cell-based assay to determine whether their different interaction modes can lead to their differential ability to modulate Cdc42 activation in biological systems. Figure 3 shows that, with a G-LISA platform that measures the GTP bound, active form of Cdc42 from lysates of Swiss 3T3 cells, ZCL278 promoted Cdc42 activation (Fig. 3A). On the other hand, ZCL279 displayed a trend of promoting Cdc42 activation at the concentration up to low micromoles (10 µM), but significantly inhibited the activation at higher concentrations (Fig. 3B). Under the same experimental condition, ZCL367 suppressed Cdc42 activation in a straight concentration-dependent manner (Fig. 3C). These findings are in general agreement with the in vitro GEF assays on the three model compounds.
It is interesting to note that with the 15-minute treatment durations as shown above (Fig. 3A), 50 and 100 µM ZCL278 appeared to display a setback in activation pattern in that they displayed antagonistic effects although not quite as potent as the inhibitory ZCL367 (compare Fig. 3A and 3C). Therefore, we set to determine whether the differential effects of ZCL278, ZCL279, and ZCL367 on Cdc42 activation are time dependent. Figure 3D shows that Cdc42 activation by ZCL278 at 50 µM formed an initial peak at 10 minutes. ZCL278 then displayed an antagonistic effect at 30 minutes followed by resuming a strong agonistic effect up to 6 hours without showing the plateau. As a comparison, ZCL279 at 50 µM, although showing moderate activating effects at 10 minutes, suppressed Cdc42 activation thereafter in a time-dependent manner (Fig. 3E). Therefore, in general, ZCL278 promoted Cdc42 activation whereas ZCL279 manifested a biphasic and hormetic effect on Cdc42 activation in cell-based assay, consistent with the in vitro findings.
Bradykinin (BK) is widely known to activate Cdc42 signaling in a variety of cells16,17,18 including Swiss 3T3 cells. Using G-LISA assays, GTP bound, active Cdc42 levels in lysates of serum-starved Swiss 3T3 cells treated with ZCL278, ZCL279 and ZCL367 at 1 and 50 µM in different time points were quantified via PAK binding. Figure 3G, H showed that when Cdc42 is prior activated by BK, ZCL278 at low concentrations sustained a higher GTP bound Cdc42 level after treatment for 15 minutes and 6 hours when compared to control (vehicle) but antagonized the BK stimulation. On the other hand, ZCL367 treatments for 15 minutes and 6 hours at 1 and 50 µM all led to a decreased level of GTP bound Cdc42 compared to control. However, at high concentrations ZCL278 and ZCL279 acted like ZCL367 that inhibited Cdc42 activation to below control level (Fig. 3F–H). This is surprising as GEF assays in vitro and G-LISA in cells all demonstrated that ZCL279 promoted Cdc42 activation at low micromolar concentrations.
HM effects of model compounds on actin remodeling in cultured cells
Next, we explored the phenotypic effects of these three model compounds on Swiss 3T3 cells because BK induced a robust actin remodeling due to Cdc42 activation8,10,16. Serum-starved Swiss 3T3 cells showed distinct focal adhesions and ridges around cell edges (Fig. 4A: Arrows and Arrowhead). BK stimulation led to a loss of focal adhesion while the smooth cell edges gave way to microspikes or filopodia (Fig. 4B: Asterisks). ZCL278 at 50 µM promoted microspikes or filopodia much like BK (Fig. 4C: Asterisks). When the cells were treated with BK, ZCL278 antagonized the BK effects on the induction of microspikes or filopodia (Fig. 4D). ZCL279 at 1-10 µM did stimulate the formation of microspikes and filopodia, although at 50 µM, it showed moderate formation of microspikes or filopodia with additional morphological alterations in the overall shape of the cell, such as elongation or spreading (Fig. 4E: Asterisks). This morphological anomaly was exacerbated when the cells were treated with BK and ZCL279 (Fig. 4F). Therefore, consistent with the findings from both GEF and G-LISA examinations, the effects of ZCL279 at higher concentrations can be hormetic. ZCL367 suppressed microspikes and filopodia and recovered ridges and focal adhesions in both BK stimulated or unstimulated cells (Fig. 4G, H: Arrows and Arrowheads).
HMs modified the social interaction behavior in AD mice in vivo
Cdc42-ITSN dysregulation has long been suspected to contribute to neurogenerative diseases such as Alzheimer’s diseases (AD) and related dementias (ADRD)19,20,21. We recently discovered that small GTPase signaling such as RhoA was spatially dysregulated in the brain of triple transgenic mouse model (3xTg-AD) bearing AD-like mutations in presenilin-1 (PS1M146V), amyloid precursor protein (APPsw), and microtubule-associated protein tau (TauP301L)22,23. When these mice were examined by unbiased RNA transcriptomic analyses using GeoMx, changes in Cdc42 signaling modulatory elements including downregulation of Arhgef9 mRNA and upregulation of Arhgap21 mRNA were seen in the 3xTg-AD mouse brain (Fig. 4I–K). The dysregulation of the major Cdc42 effector protein in layer 3 and 4 cortical neurons in brain, cdc42ep3, was reported in psychiatric disorders such as schizophrenia24. We found that cdc42ep3 was upregulated in the 3xTg-AD mouse brain (Fig. 4L). We then examined the social contact behavior of 3xTg-AD mice. Under the condition when they were left unapproached, 3xTg-AD mice showed reduced spontaneous activity25. However, when they were approached and lifted by an experimenter, they quickly ran away (Fig. 4M). All three model compounds, to varying degrees, attenuated this contact avoidance behavior of 3xTg-AD mice (Fig. 4M). AD associated protein phospho-tau (S396) levels were reduced in 3xTg-AD mice upon treatment with ZCL279 (Fig. 4N). These effects were consistent with the ADME (Absorption, Distribution, Metabolism, and Excretion) and pharmacokinetics (PK) studies that these compounds passed blood-brain barrier26, demonstrating their potential functionality in modulating AD-like neuropathogenesis.
Profiling of HMs by biochemical activity with Cdc42-GEF approach
Since these three model classes of Cdc42 modulators demonstrated significant functionality in BK-Cdc42 activation of actin remodeling and modified AD-like social interaction behavior in a mouse model, we questioned whether there are more compound classes in our top ranked SMMs which may display the same or similar properties. The top ranked compounds were initially obtained by high throughput virtual screening based on structural fitness and binding scores8. Because higher binding scores do not necessarily translate to higher functionality, we performed exhaustive analyses with over 2,500 GEF assays to profile the GTP loading activities on top ranked compounds derived from molecular docking-based high throughput virtual screening.
This GEF-based activity screening revealed a spectrum of Cdc42 HMs that can be categorized into five classes of different functionalities (Fig. 5A–J and Supplementary Table S3 and Fig. S7). They displayed different modes of pharmacological activity in the absence and presence of GEF, respectively. Among the five classes, three classes were found to possess activities similar to those observed for the model compounds ZCL278, ZCL279, and ZCL367. The compounds of ZCL278 class, or Class I, promoted GTP loading to Cdc42 in the absence of GEF but reduced this activity in the presence of GEF (Fig. 5A, B). The compounds of ZCL279 class, or Class II, elevated GTP loading to Cdc42 at the concentration up to low micromoles (from 0.2 nM to 10 µM) but suppressed GTP loading at higher concentrations with and without GEF (Fig. 5C, D). The compounds of ZCL367 class, or Class III, suppressed GTP loading to Cdc42 with and without GEF (Fig. 5E, F).
The GEF-based activity screening identified two additional classes of compounds that we had no prior knowledge. They displayed different modes of action from the ZCL278 (Class I), ZCL279 (Class II), and ZCL367 (Class III) classes. The compounds of ZCL1139 class, or Class IV, promoted GTP loading to Cdc42 regardless of GEF presence (Fig. 5G, H), thus presenting the features of bona fide activators or agonists. The compounds of ZCL1107 class, or Class V, promoted GTP loading to Cdc42 at low concentrations up to 10 µM, but reduced GTP loading to Cdc42 at high concentrations in the absence of GEF, much like ZCL279 class or Class II (compare Fig. 5I with Fig. 5C). However, when GEF is present, the Class V compounds promoted GTP loading to Cdc42 even at 100 µM concentration (Fig. 5J).
Independent profiling of HMs by CADD
To explore whether there are more efficient and economic ways to profile HMs other than GEF-based activity screening, we applied several Schrödinger-enabled CADD methodologies to predict HM classification.
We first used classic docking score to instill the PBPO in Cdc42-ITSN for our model compounds. Then, this PBPO strategy was applied to profile the top ranked compounds derived from the high throughput screening of SMM libraries, which was based on the five binding pockets in interface of Cdc42-ITSN (Fig. 2C). Most compounds showed preferential binding to Effector and GDP binding pockets (Fig. 5K). We compared the PBPO of all identified HMs to that of each model compound. For example, PBPO of Cdc42-ITSN for ZCL279 is: Effector > Cdc42-DH domain of ITSN > GDP > DH-PH domain of ITSN > Switch 2-DH domain of ITSN. Among the compounds from the HM libraries, ZCL993 and ZCL1124 were found with having the same PBPO as ZCL279. This method was also employed to assign compounds to ZCL278, ZCL367, ZCL1139 and ZCL1107 classes based on their PBPO (Fig. 5K).
We additionally applied a structure-based pharmacophore hypothesis generation for the model compounds. The docked ZCL278 and ZCL279 on Effector pocket, ZCL367 and ZCL1139 on GDP pocket of Cdc42-ITSN as well as ZCL1107 on Cdc42-DH domain of Cdc42-ITSN complex (their preferential binding pockets) were used to build pharmacophore hypothesis using Phase module of Schrödinger suite27. Pharmacophore modeling provides six built-in types of pharmacophore features: hydrogen bond acceptor (A), hydrogen bond donor (D), hydrophobic (H), negative ionizable (N), positive ionizable (P), and aromatic ring (R). The identified pharmacophoric hypotheses features for ZCL278, ZCL279, ZCL367, ZCL1139 and ZCL1107 were RNRAR, RNDRRR, DDRDDR, RAARD and RAA containing five, six, four and three features, respectively (see Supplementary Fig. S8A–E). We examined all top ranked HMs and identified for matches with pharmacophore hypotheses of ZCL278, ZCL279, ZCL367, ZCL1139 and ZCL1107 (see Supplementary Table S4). For example, ZCL993 was again found with having matched pharmacophoric feature to ZCL279.
Shape-based methods to align and score ligands were also employed for flexible ligand superposition and virtual screening as ligand-based classification28,29. Using Schrödinger’s Phase Shape, 3D ligand alignments for model compounds and the rest of the 44 HMs with the matching scores to assign HMs to ZCL278, ZCL279, ZCL367, ZCL1139 and ZCL1107 classes are shown in Supplementary Data (Table S4 and Fig. S7F–J). For instance, ZCL993 and some other compounds were identified with similar shapes to ZCL279.
The CADD analysis with PBPO revealed similar HM profiles in a color-coded spectrum to these five distinct classes of Cdc42 HMs identified by GEF-based activity profiling (Fig. 5A–K). Although more data points may help their better match-up, this comparison provided some insights how to best find functionally similar compounds to the five distinct classes that were identified in our primary high throughput screening. In a schematic overview, the number of identified HMs based on the screening methods in the intersection of ITSN and Cdc42 is shown (Fig. 5L–P). Some compounds that show similarities to the model compounds were identified as the same compounds using different screening methods (see Supplementary Table S4). As shown in Fig. 5L–P, the red square boxes on the connecting lines between these methods indicate the overlap, showing the count of compounds identified by both methods. For instance, NSC136024, NSC158437, NSC94024 were identified as ZCL278-like Class I compounds based on both GEF activity assay and Shape screening methods. Also, in ZCL279-like Class II compounds, ZCL993 was found by all employed screening methods. For ZCL367-like Class III compounds, ZCL1111 was identified according to GEF activity assay and Molecular modeling screening methods. Molecular modeling, Shape screening and GEF assay revealed that ZCL1105 could be in ZCL1139-like Class IV and ZCL1142 was similar to ZCL1107-like Class V. These compounds identified with different screening methods (shown in red square boxes in Fig. 5L–P) help to understand the extent of consistency and reliability of the screening methods in identifying potential modulators of Cdc42 activation.
These studies showed the potential of each screening method to find a compound which can mimic the GEF-validated pharmacological functionality of interests (Fig. 5). The average prediction rate with each screening method for all five GEF-validated HM classes are as follow: Shape screening (54.2%) > Molecular docking (34.2%) > Molecular modeling (30.8%). However, considering that the number of compounds identified by different screening methods are not identical, we re-calculated the average prediction rate for each screening method based on the total number of compounds identified by the given screening method {Prediction rate (%) =(X/Y)*100, where X=the number of identified compounds which are the same between the GEF assay and the given screening method, and Y=total number of identified compounds for the given screening method}. The normalized average prediction rate for all five HM classes obtained with Molecular docking screening method was slightly better than that obtained with Shape screening method (21.4 vs. 20.6). Overall, the results indicated that with PBPO instilled, the redesigned Molecular docking was likely an effective CADD method for prediction of an HM category that can mimic the pharmacological functionality based on GEF-activity screening.