The Vero African green monkey kidney cells in this study were obtained from American Type Culture Collection (ATCC), Cat# CCL-81. Vero cells were cultured in Dulbecco’s modified Eagle medium (DMEM, Corning, NY) containing 10% fetal bovine serum (FBS, Phoenix Scientific, St. Joseph, MO), 1% penicillin-streptomycin (Corning, NY), and GlutaMax (Thermo Fisher Scientific, Waltham, MA). Cells were maintained at 37°C in 5% CO₂. Generation and propagation of recombinant vesicular stomatitis virus (rVSV) encoding enhanced green fluorescent protein (eGFP) in the first position and replacing VSV G with the EBOV-GP (EBOV/H.sap-tc/COD/76/Yambuku-Mayinga) or Lassa virus (LASV)-GP (Josiah) were previously described (26, 27).

Docking studies were carried out using Glide-SP (Schrödinger modeling suite, versions 2024-1 to 2024-3) to obtain the binding mode for tamoxifen into the proposed binding site. Standard option joint with the SP algorithm was applied for pose generation and evaluation. The X-ray structure for the Zaire EBOV-GP in complex with toremifene (PDB: 5JQ7) was used as a template for the docking.

The details of chemical synthesis and characterization are described in Supplementary Material . The SuFEx-based library synthesis was adapted from a method described previously (17) in 384-well plate format. Briefly, to a DMSO solution of the iminosulfur oxydifluoride derivative 1 was added amine library in DMSO and sodium phosphate buffer (pH 9.0, 0.2 M) subsequently. The compounds were synthesized with difluoride concentration at 1 mM and a solvent mixture DMSO:buffer 3:1. The reaction mixtures were shaken at room temperature overnight and then used directly for activity measurement with 5000-fold dilution.

The virus was titrated to achieve an infection rate of approximately 50%. Vero cells were seeded at a density of 2.0 × 10⁴ cells/well in a 384-well plate and incubated for 24 hours at 37°C in 5% CO₂. The cell culture medium was then replaced with 20 μL/well of virus culture medium (DMEM containing 2% FBS, 1% penicillin-streptomycin, and GlutaMax), and 0.2 μL/well of the compound library in DMSO was added using the Bravo Automated Liquid Handling Platform (Agilent Technologies, Santa Clara, CA). Subsequently, 20 μL/well of virus solution was added, and the cells were incubated for 16 hours at 37°C in 5% CO₂. Infected cells were stained with Hoechst (Thermo Fisher Scientific, Waltham, MA) to visualize nuclei and fixed with 4% paraformaldehyde. Infectivity of VSV pseudotype was measured by automated enumeration of eGFP⁺ cells using a Cytation 5 reader (BioTek, Winooski, VT), as previously described (26). Quantification was done using Gen5 data analysis software (BioTek, Winooski, VT). For each plate, DMSO-treated infected and non-infected control wells were included to calculate the inhibition rate as: Inhibition rate (%) = (Infection control - Sample)/(Infection control - Non-infection control) × 100.

Dose-response assays were performed under the same experimental conditions as the library screening. Compounds were prepared in two-fold serial dilutions in DMSO and 0.2 μL/well was added to the cell culture media using the Bravo Automated Liquid Handling Platform. The EC₅₀ values were calculated using GraphPad Prism 10.

Cells were seeded under the same conditions as the antiviral assay. The cell culture medium was replaced with 40 μL/well of virus infection medium without virus, and compound solutions prepared as two-fold serial dilutions were added at 0.2 μL/well using the Bravo platform. After 16 hours, Promega^® CellTiter-Glo^® 2.0 (Promega, Madison, WI) was added, and cell viability was assessed following the manufacturer’s instruction. The CC₅₀ values were calculated using GraphPad Prism 10.

The gene fragment encoding the extracellular domain of the Zaire EBOV (strain Mayinga-76) glycoprotein (UniProt ID: KB-Q05320) was synthesized as described previously (28). This gene was inserted into the mammalian expression vector pHCMV3, with an Ig Cκ leader sequence at the N-terminus to enable secretion. A foldon trimerization domain from bacteriophage T4 fibritin was added to the C-terminus to promote trimerization, along with a 6×His tag for affinity purification.

The plasmid was transfected into human Expi293S cells to produce the EBOV-GP trimers. After six days of incubation, the culture medium containing the secreted trimers was collected. The protein was purified using Ni-NTA affinity chromatography with Ni Sepharose Excel resin and dialyzed overnight into 1× PBS. The sample was concentrated and further purified via size-exclusion chromatography on a Superdex 200 HiLoad 16/600 column pre-equilibrated with 1× TBS. The protein peak corresponding to the EBOV-GP trimers was identified, collected, and concentrated to a final concentration of 2 mg/mL.

Recombinant EBOV-GP protein was labeled using the Monolith Protein Labeling Kit RED-tris-NTA 2nd Generation dye (Cat #MO-L018, NanoTemper Technologies, Germany) following the manufacturer’s instructions. Specifically, 125 nM EBOV-GP was incubated with 25 nM dye in 25 mM HEPES pH 7.5, 100 mM NaCl, 0.005% tween-20 in the dark at room temperature for 30 min. The sample was centrifuged for 10 min at 4°C and 14,000 g and the supernatant was transferred to a fresh tube. To determine the K _D of EBOV-GP to tamoxifen and its analogs, 125 nM labeled EBOV-GP was incubated with increasing concentrations of small molecules in the same buffer with 1% DMSO. Samples were loaded into standard glass capillaries (Monolith NT.155 Capillaries) and analyzed by MST using a Monolith NT.115 Blue/Red, LED power and IR laser power of 60%. Samples showed no aggregation according to post-run analysis using the Monolith data collection software (NanoTemper). Fraction bound and error were generated by NanoTemper software (MO.Affinity Analysis) and K _D values were determined using GraphPad Prism 10 and nonlinear fit of one-site specific binding.

For crystallization of EBOV-GP trimers, the protein was concentrated to 8 mg/mL in a buffer containing 20 mM Tris (pH 8.0) and 150 mM NaCl. Crystals were grown at 20°C in a solution of 9% PEG 6000 and 100 mM sodium citrate (pH 5.2). Complex structures were obtained by soaking EBOV-GP crystals in a 5 mM solution of the target compound for a few minutes, followed by cryoprotection with 20% glycerol and rapid plunging into liquid nitrogen for storage before data collection. Diffraction data were collected at the Stanford Synchrotron Radiation Lightsource (SSRL) on beamline BL12-1. EBOV-GP crystals soaked with compound 4R diffracted to a resolution of 2.59 Å. Data indexing, integration, and scaling were performed using HKL2000 (29). Structures were solved by molecular replacement (MR) with Phaser in Phenix (30) with PDB 6F6I as the MR model, and subsequent model building and refinement were conducted using Coot and Phenix.refine (31, 32). Structural quality was assessed with MolProbity (33), and further validation was performed using the PDB validation server. Data collection and refinement statistics are summarized in Supplementary Table 1 .

The first step of the SuFEx-enabled HTMC workflow requires installation of a SuFExable difluoride moiety on the lead molecule ( Figure 1A ). This difluoride analog can be readily synthesized by reacting thionyl tetrafluoride (O=SF₄) gas with the lead compound functionalized with a primary amine. A structural analysis of EBOV-GP protein + ligand complex was performed to determine the modification site on the lead molecule. Due to the absence of an experimentally determined crystal structure for the complex of EBOV-GP with tamoxifen, molecular docking was used to predict the binding mode of this ligand. For this purpose, we employed a crystal structure of toremifene (i.e. an analog of tamoxifen) bound to EBOV-GP (PDB entry: 5JQ7) as a template (34). Given the chemical structural similarity between tamoxifen and toremifene, we expected that both molecules would share similar binding conformations.

The docking suggests that tamoxifen aligns closely with the binding conformation of toremifene, preserving key interactions with crucial residues within the active site ( Supplementary Figure 1 ). The aromatic rings B and C of tamoxifen adopt a similar orientation to the corresponding rings in toremifene, forming a π-stacking between these rings and Tyr517, as well as an ionic interaction between the protonated tertiary amine and the negatively charged side chain of Glu100. Additionally, the docking analysis highlights a solvent-exposed area at the meta position on the aromatic ring A, making it a prime candidate site for further functionalization. By targeting these solvent-exposed positions, we hypothesized that expanding the molecule into unoccupied regions of the binding pocket could form additional interactions with adjacent residues, thus enhancing both binding affinity and cellular anti-infective efficacy.

Guided by the predicted structure, we designed a synthetic route ( Supplementary Figure 2 ) to introduce a difluoride group at the meta position of ring A of tamoxifen based on previous reports for the synthesis of analogs with the triphenylethylene core (25). The difluoride-containing analog 1 was successfully synthesized, and its chemical structure was verified by NMR and LC-MS analyses, as described in the Supplementary Material .

With analog 1 in hand, we assessed its antiviral activity and cytotoxicity, along with that of tamoxifen, using a virus-like particle (VLP) assay (26). The assay employs vesicular stomatitis virus (VSV) particles displaying EBOV-GP (VSV-EBOV-GP), in place of the native glycoprotein G, to infect Vero cells in culture. This approach obviated the need for high-security BSL4 facilities required for working with authentic filoviruses. The VSV-EBOV-GP also encodes an enhanced GFP, which allows for direct quantification of infected cells by fluorescence imaging. The use of a cell-based assay provides advantages for drug discovery, as it more accurately reflects the compounds’ activity in a biologically relevant environment and accelerates the profiling of cell-active compounds. This infection model enabled us to evaluate the inhibitor’s effectiveness in blocking the processing of viral GP, a crucial step for EBOV infection. Additionally, the VSV assay provided insight into whether the proposed structural modifications were tolerated by the lead compound without significantly compromising its activity.

The results showed that the lead compound tamoxifen has an EC₅₀ value of 0.19 μM, while our difluoride-functionalized analog 1 exhibited an EC₅₀ of 0.38 μM, with both values being within the same magnitude ( Figure 1B ). Importantly, compound 1 did not exhibit cytotoxicity up to 10 µM ( Figure 1C ). These findings indicate that the structural modifications to introduce the difluoride moiety do not critically reduce the biological activity of the lead compound, consistent with the docking predictions. The results support the effectiveness of the docking-guided design and validate the choice of the modification site on the tamoxifen aromatic ring.

Once an appropriate SuFExable derivative of the lead compound has been identified, the next step in the HTMC workflow ( Figure 1A ) is the diversification reaction with an amine-fragment library. This strategy allows the preparation of a focused library of lead compound analogs with expanded diversity containing sulfamide or sulfuramidimidoyl fluoride linkages, from primary or secondary amines respectively ( Figure 2A ).

A library of 2,496 primary and secondary amine fragments were individually reacted with the difluoride-containing analog in 384-well plates overnight at room temperature using a 1:3 phosphate buffer:DMSO solvent mixture (see Materials and Methods for further details). The reaction conditions employed for diversification were determined in a preliminary assessment of the difluoride reactivity with a small set of representative amines. Reagents concentration, temperature, pH, and solvent composition were evaluated to identify the optimal conditions for generating a high-quality library, allowing us to screen the compound set without additional purification. The synthesis of the library was performed using an automated liquid handling robot Agilent Bravo BenchCel, enabling efficient generation of 2,496 analogs in a single batch. An LC-MS analysis of a randomly selected wells was performed to evaluate the quality of the library. The results indicated a generally high conversion rate, validating the suitability of the library for the D2B screen in the subsequent step.

The synthesized library was screened at 200 nM using the VSV-based EBOV entry assay. The library solutions were dispensed over Vero cells in 384-well plates. Cells treated with the compound were infected with VSV-EBOV-GP, and 16 hours post-infection, their nuclei were stained and then fixed. The infection rate was assessed by measuring eGFP expression levels, normalized to the number of nuclei. This approach allowed us to screen the compound set in singlicate with an assay Z′-factor of 0.77, indicating robust assay performance. The screening results are illustrated in Figure 2B as a scatter plot of % inhibition. After validating the hits by triplicate, the most potent molecules were manually synthesized in milligram quantities for further dose-dependent biochemical and cell-based characterization. This approach allowed us to confirm the efficacy and refine the profile of the most promising candidates. The structure of the representative hits synthesized for validation is shown in Figure 2A (compounds 2-4), and the dose-response inhibition profile in Figure 2C . Compound 4 was identified as one of the top hits in the primary screening, with an EC₅₀ value of 241 nM. Notably, compound 4 demonstrated an almost 2-fold increase in cellular antiviral potency compared to 1.

To further understand the structure-activity relationships, we synthesized a small library of analogs of the identified molecules and subjected to dose-response antiviral and cytotoxicity studies. Specifically, we focused on modifications to the oxolane ring ( Table 1 , Supplementary Figure 3 ). These modifications included altering the ring size (5), investigating different substitution patterns on the oxolane ring (6, 7), and replacing the oxolane with an N-methyl pyrrolidine ring (8) as well as other N-substituted analogs with varying polarity and chain lengths (9 – 15). Despite exploration of these broad structural modifications, no significant improvement in activity was observed with the ring changes. Since compound 4 was identified as a racemic mixture, we synthesized and evaluated both R and S enantiomers ( Table 1 ). Interestingly, the S-configured compound (4S) exhibited a superior EC₅₀ value of 92 nM, compared to its R-counterpart (4R) of 351 nM, indicating the importance of the chirality on the improved potency and specific interaction with GP-protein. The cytotoxicity studies showed variable CC₅₀ values among the synthesized analogs. Compounds 4S and 4R did not show significant cytotoxicity below 10 µM with an excellent selectivity index (SI) of 150 for 4S ( Table 1 ).

To confirm the direct interaction of 4R and 4S with EBOV-GP, the equilibrium dissociation constants (K _D) for the interaction inhibitor-protein were measured. The recombinant protein was expressed in Expi293S cells and purified using Ni-NTA affinity and size-exclusion chromatography. Then, we evaluated the interaction of both enantiomers as well as the parent tamoxifen with EBOV-GP by microscale thermophoresis (MST). As shown in Figure 3A , 4S displayed a stronger binding affinity against EBOV-GP, with a K _D value of 0.63 µM, compared to 4R with a K _D value of 6.4 µM. In comparison, the corresponding value for tamoxifen was 32 µM. This trend mirrors the observed differences in cellular potency, where 4S demonstrated enhanced efficacy at low submicromolar concentrations in blocking EBOV entry, while 4R required higher concentrations to achieve a comparable effect ( Table 1 ). It is worth noting the 50-fold affinity enhancement in the biochemical assay between 4S and tamoxifen, providing further validation of improved potency of the inhibitor developed through our HTMC platform.

We then determined the X-ray structure of 4R in complex with EBOV-GP to 2.59 Å resolution ( Figure 4 , Supplementary Table 1 ). The electron density for 4R is well defined ( Figure 4B ) and the compound binds within the extensive hydrophobic cavity of EBOV-GP, a known target site for other inhibitors. It adopts a similar pose to its analog, toremifene, with conserved interactions involving the three aromatic rings and the dimethylethanamine group of the parent scaffold (34). Compared to toremifene, 4R occupies a larger portion of the available cavity, with its additional moiety engaging a hydrophobic subpocket, potentially contributing to its improved binding. Unfortunately, we were unable to obtain a structure for 4S, as it renders the crystal unstable, possibly due to conformational changes when the compound is soaked into the crystal, and leads to poor diffraction. Nevertheless, the complex structure of EBOV-GP and 4R validates the docking model of our lead molecule and provides structural insights into the improved potency of the developed molecules.

Finally, we evaluated the viral selectivity and specificity of compound 4S by comparing the inhibitory activity against VSV-EBOV-GP to VSV-G and VSV particles displaying LASV-GP (VSV-LASV-GP, Figure 3B ). Tamoxifen and our compound 4S showed selective inhibition of VSV-EBOV-GP but not VSV-G or VSV-LASV-GP, indicating that its antiviral activity is specific to EBOV-GP. This result supports the hypothesis that compound 4S does not induce non-specific effects on viral entry mechanisms shared by multiple viruses or inhibit proteins involved in later stages of the EBOV infection, such as cathepsins, which are known to mediate viral entry through endosomal processing. Furthermore, the lack of activity in VSV-G and VSV-LASV-GP models reduces the likelihood that the compound’s mechanism involves general endosomal disruption or interference with acidic environments. This selectivity, in concert with the chiral preference of 4S over its R-isomer in binding EBOV-GP, suggests that the compound engages specific molecular interactions with EBOV-GP that are crucial for its activity.