Surface Plasmon Resonance (SPR)-Based Workflow for High-Throughput Discovery of CD28-Targeted Small Molecules

Abstract

CD28 is a critical costimulatory receptor involved in T cell activation and immune regulation, making it a compelling target for immunomodulatory therapies. Despite its therapeutic relevance, small molecule CD28 inhibitors remain largely underexplored. To address this gap, we developed a high-throughput screening (HTS) workflow using surface plasmon resonance (SPR) to identify novel CD28-targeted small molecules. To our knowledge, this work represents the first SPR-based HTS platform applied to the discovery of small molecules targeting a stimulatory immune checkpoint receptor. A chemical library composed of diverse 1056 small molecules was screened using a 384-well format. Compounds were evaluated based on level of occupancy (LO), binding response, and dissociation kinetics, resulting in 12 primary hits (1.14% hit rate). Follow-up dose–response SPR screening confirmed micromolar-range affinities for three compounds. Molecular docking and 100 ns molecular dynamics simulations of the top hit, DDS5, revealed a stable complex with CD28, maintained by hydrogen bonding and a persistent interaction with Phe93. Functional validation using a competitive ELISA confirmed that DDS5 inhibited the CD28–CD80 interaction. These results demonstrate that our SPR-based HTS platform is a robust and efficient strategy for discovering CD28-targeted small molecules. The integration of computational evaluation and orthogonal validation further underscores the potential of DDS5 as an early stage immunomodulatory agent.

Introduction

T cell activation is a tightly regulated process that requires both antigen recognition and costimulatory signaling. Among the key costimulatory receptors, Cluster of Differentiation 28 (CD28) plays a central role in initiating and sustaining T cell responses. In the absence of CD28 signaling, engagement of the T cell receptor (TCR) alone can result in T cell anergy, deletion, or regulatory differentiationmechanisms essential for maintaining peripheral tolerance and preventing autoimmunity. CD28 is constitutively expressed on naïve T cells and interacts with its classical ligands, CD80 (B7-1) and CD86 (B7-2), on antigen-presenting cells (APCs). This interaction amplifies TCR-mediated signaling by promoting the activation of downstream pathways such as PI3K-Akt, NF-κB, and MAPK, which collectively enhance interleukin-2 (IL-2) production, cell cycle progression, and antiapoptotic responses.

Dysregulation of CD28 signaling has been implicated in the pathogenesis of several immune-mediated diseases, including inflammatory bowel disease − and rheumatoid arthritis. Consequently, therapeutic strategies targeting CD28 have primarily focused on biologics, such as monoclonal antibodies and Fc fusion proteins, which block ligand binding and downstream signaling. − These approaches aim to dampen excessive T cell activation by preventing CD28 engagement with its ligands, thereby restoring immune balance in chronic inflammatory conditions. While these protein-based therapies have shown efficacy in certain contexts, they are associated with limitations including immunogenicity, limited tissue penetration, and complex manufacturing processes. Moreover, the systemic nature of biologics can lead to off-target effects and prolonged immune suppression, necessitating more selective and controllable therapeutic alternatives.

In contrast, small molecule inhibitors offer several advantages, including oral bioavailability, tunable pharmacokinetics, and reduced risk of antidrug antibody formation. Nevertheless, the development of CD28-targeted small molecule modulators has been hindered by the inherent structural and biophysical challenges of protein–protein interaction (PPI) interfaces, which are often characterized by shallow, flexible, and poorly defined binding topologies. These features limit the ability of small molecules to achieve high-affinity and selective binding. Recent advances in biophysical screening technologies have facilitated the discovery of small molecule modulators targeting PPIs. Among these, surface plasmon resonance (SPR) has emerged as a powerful, label-free technique that enables real-time detection of molecular interactions with high sensitivity and throughput. SPR is particularly well-suited for interrogating challenging targets such as immune checkpoint receptors, as it permits direct measurement of binding kinetics and affinities under near-physiological, solution-phase conditions. Moreover, its compatibility with fragment- and diversity-based libraries makes SPR an ideal platform for the early identification of novel chemotypes against traditionally undruggable targets.

To address the unmet need for small molecule modulators of CD28, our workflow (Figure ) involved developing a high-throughput screening (HTS) platform based on SPR. Using this platform, we screened a diverse chemical library of small molecules and identified multiple hit compounds (1.14% hit rate) with micromolar binding affinities to CD28. The top candidate demonstrated functional inhibition (Figure ) of the CD28–CD80 interaction using ELISA, supporting its potential as an immunomodulatory agent. To gain mechanistic insight into its mode of action, we performed molecular docking studies and molecular dynamics (MD) simulations, which revealed plausible binding poses at the CD28 interface (Figure ) that may sterically hinder ligand engagement.

1.

Conceptual overview of the SPR-based high-throughput screening workflow for CD28-targeted small molecules.

Results and Discussion

We first defined and developed the affinity-based screening assay using SPR technology. The extracellular domain of the human CD28 protein (residues Asn19-Pro152, UniProt accession number P10747-1) was selected as the target ligand, as it represents the biologically active, glycosylated, disulfide-linked homodimeric form of the protein. This domain is responsible for ligand binding and is structurally well-suited for immobilization in SPR assays. As an essential first step, we optimized the immobilization strategy by selecting the most appropriate sensor chip format to ensure stable and reproducible ligand presentation for high-throughput interaction analysis.

Out of the diverse array of sensor chips compatible with the Biacore instrument, we selected the Sensor Chip CAP based on several key advantages. First, it enables the reversible capture of biotinylated molecules, facilitating chip regeneration and repeated use. Second, the Avitag-labeled target protein exhibits exceptionally high binding affinity to the chip’s modified streptavidin (K _d = 4 × 10^–14 M, biotin–streptavidin), ensuring stable immobilization with minimal dissociation over extended periods of time. Third, the Avitag, a 15-amino acid peptide, is positioned downstream of a polyhistidine (His) tag at the C-terminus of the target protein. This configuration is not expected to interfere with the protein’s native structure or function, as both the His-tag and Avitag are relatively small and terminally located, thereby minimizing potential steric hindrance or conformational disruption.

We conducted a protein concentration scouting experiment ranging from 10 to 50 μg/mL and determined that 50 μg/mL was optimal (Figure S1). This concentration enabled a ligand immobilization level (R _L) of approximately 1750 Response Units (RU), which in turn allowed us to achieve theoretical R _max values (maximum analyte-ligand interaction response) between 14 and 24 RU for our screened compounds (MW_L = 275–475 Da, respectively). Later, we optimized the assay buffer using a positive control, anti-CD28 antibody (reported IC₅₀ ≈ 50 ng/mL in a cell-based assay). We confirmed that 1× PBS-P+ (Cat # 28995084, Cytiva), with or without up to 2% DMSO, did not interfere with the expected protein-antibody binding affinity (data not shown). After verifying that DMSO concentrations up to 2% did not affect protein stability or function, we compared DMSO-supplemented 1× PBS-P+ with DMSO-supplemented 1× HBS-P+ (Cat# BR100671, Cytiva). No difference in RU was observed when a 2 μg/mL (saturating) dose of anti-CD28 antibody was tested (data not shown). Based on these findings, 50 μg/mL of His/Avitag human CD28 protein and a PBS-based buffer supplemented with 2% DMSO were selected for subsequent HTS of small molecules.

A 1056-compound subset from the Discovery Diversity Set (DDS) library (Enamine) was selected to validate the SPR-based screening workflow. This library is particularly well-suited for targeting CD28, a transmembrane costimulatory receptor, due to its enrichment in chemotypes designed to engage G-Protein Coupled Receptors (GPCR)-like and protein–protein interaction (PPI)-like interfaces. Such structural features are advantageous for binding to membrane proteins like CD28, which often present shallow, hydrophobic, or conformationally dynamic binding sites. The library’s emphasis on three-dimensional, sp³-rich scaffolds further enhances the probability of identifying ligands capable of accessing these challenging topologies. In addition, its diverse clustering strategy ensures broad coverage of chemical space, making it a robust starting point for both hit discovery and lead optimization in immunomodulatory drug development.

While a clean screen assay is often used to prefilter compounds that exhibit nonspecific interactions with the sensor chip surface, we elected to omit this step in our workflow. Compounds exhibiting such promiscuous behavior can be effectively identified during the single-concentration screen through elevated signals on the reference flow cell and atypical binding profiles. Moreover, nonspecific binding to the immobilized target may increase baseline responses but does not compromise target protein integrity or obscure the detection of true binders. Given these considerations, and to streamline throughput, we proceeded directly with the primary screen, incorporating analytical flags to account for nonspecific and nondissociating interactions during hit triage.

Subsequently, sourced from the 10 mM stock plates, we prepared the assay 384-well plates containing 100 μM of the small molecules in assay buffer supplemented with 2% DMSO. Additionally, negative control samples, containing only assay buffer with the same DMSO concentration but no small molecules, and positive control samples, containing anti-CD28 antibody at 2 μg/mL, were included in the 96-well plate where the remaining SPR reagents were also present. All plates were sealed with the appropriate foil to prevent sample evaporation throughout the duration of the experiment. Subsequently, the plates were placed into the Biacore hotel trays, and the HTS was carried out over a 19-h period (full details on the method in the Supporting Information). After raw data collection, the following data analysis workflow was designed and applied to identify primary hit candidates for subsequent binding affinity experiments using the Biacore Insight Evaluation Software (Cytiva).

After applying solvent correction to all samples, we verified that the immobilization levels across the channels were consistent (ranging from 1764.8 to 1820.2 RU) and recorded the individual values for subsequent analysis. Afterward, R _max [eq ] and Level of Occupancy [LO, eq ] parameters were calculated for each screened compound. Briefly, R _max is the maximum possible SPR signal resulting from the interaction between a ligand and an analyte and is represented in RU. This value is calculated based on the molecular weight of the analyte (small molecule), the molecular weight of the immobilized ligand (CD28 protein), the amount of ligand immobilized onto the chip surface, and the valency. Since our target protein is a homodimer, the expected number of analyte binding sites per ligand equals two. In the LO equation, the level of occupancy refers to the extent to which the available binding sites in the CD28 homodimer are occupied by a specific analyte. As shown in Figure , the analyte binding late corresponds to the RU value obtained for a specific analyte at the end of the association time (last 5 s).

$$R_{\max }(\mathrm{RU})=\frac{\mathrm{Analyte}\mathrm{molecular}\mathrm{weight}(\mathrm{Da})}{\mathrm{Immobilized}\mathrm{ligand}\mathrm{molecular}\mathrm{weight}(\mathrm{Da})}\times \mathrm{Immobilized}\mathrm{ligand}\mathrm{level}(\mathrm{RU})\times 2(\mathrm{ho}\mathrm{mod}\mathrm{imer})$$ 1

$$\mathrm{Level}\mathrm{of}\mathrm{Occupancy}(\%)=\frac{\mathrm{Analyte}\mathrm{binding}\mathrm{late}(\mathrm{RU})}{R_{\max }(\mathrm{RU})}\times 100$$ 2

2.

Sensorgrams of primary small molecule hits identified through HTS and selected for follow-up binding affinity studies. Sensorgrams represent the analysis step within the designed HTS method used to evaluate 1056 compounds. All compounds were injected over the sensor chip at 100 μM with a 60-s association phase and a 100-s dissociation phase at a flow rate of 30 μL/min in running buffer (1× PBS-P+ supplemented with 2% DMSO). Wash step with 50% DMSO in Milli-Q water and a carry-over control using running buffer (20 s at 30 μL/min) followed, both bypassing the sensor chip flow cells. Compounds were selected for follow-up binding affinity studies if they exhibited a Level of Occupancy (%) >50% and <200% and were not flagged as nonspecific binders. Sensorgrams are color-coded to distinguish each screened compound. Specifically, DDS1 showed LO between 100 and 200% with slow-dissociating behavior; DDS7 and DDS10 showed LO between 50 and 100% with slow-dissociating behavior; and the remaining compounds showed LO between 50 and 100% without additional flags. Abbreviations: C–O: carry over control.

Additionally, we added markers to label each compound with one or several of the following categories: nonspecific binder (NSB), non- or low-dissociating binder (NDB), high level of occupancy compound (High LO), and potential hit. Compounds that exhibited a relative response equal to or greater than 5 RU on the reference flow cell at the beginning of the dissociation phase (Analyte stability early, Figure ) were classified as NSB (123 out of 1056 total screened compounds, 11.65% rate).The typical behavior of small molecules is characterized by rapid association and dissociation with the target protein. Full association with the target ligand generally occurs within a few seconds, followed by a stable plateau phase as long as the small molecule continues to flow over the sensor chip. Once the flow stops, a rapid dissociation is expected. Therefore, compounds that exhibited minimal or no dissociation during the dissociation phase were flagged as NDB. For High LO classification, small molecules were divided into two groups: those with occupancy levels exceeding 200% were excluded from further consideration (1 out of 1056 compounds, 0.09% rate), while those with occupancy between 100 and 200% were flagged. Although an LO higher than 100% is theoretically impossible, given that the homodimer has only two binding sites, we allowed some margin for experimental variability. However, if subsequent binding affinity experiments showed a linear dose–response behavior, the analyte would be discarded as a true binder. Finally, compounds with occupancy levels between 50 and 100% were flagged as potential hits.

Taking all these criteria into account, those small molecules that had no NSB flag and their LO was between 50 and 200% were selected for subsequent binding affinity experiments (12 out of 1056 compounds, 1.14% hit rate) (Table and Figure ). Among these, nine compounds presented a LO between 50 and 100% and no additional flags: DDS2, DDS3, DDS4, DDS5, DDS6, DDS8, DDS9, DDS11, and DDS122. Compound DDS1 exhibited a higher-than-expected LO (189.70%) and a slow-dissociating behavior. Compounds DDS7 and DDS10 fell within the acceptable LO but displayed a slow-dissociating behavior2. This HTS and subsequent data analysis workflow enabled the rapid classification, ranking, and prioritization of primary hits for follow-up binding affinity experiments. Selection was based on each compound’s binding response and behavior toward CD28, while excluding small molecules exhibiting aberrant or undesirable binding characteristics.

1. List of Primary Small Molecule Hits Identified through High-Throughput Screening .

analyte	analyte MW (Da)	analyte binding level (RU)	ligand	ligand MW (Da)	immobilized ligand level (RU)	R max (RU)	level of occupancy (%)
DDS1	341.33	33.2	Biotinylated Human CD28 Protein	70,000	1792.5	17.48	189.70
DDS2	298.26	14.4			1776.8	15.14	95.08
DDS3	288.30	13.8			1786.5	14.72	93.47
DDS4	321.39	12.2			1792.5	16.46	74.27
DDS5	305.42	11.3			1798.1	15.69	72.12
DDS6	305.37	10.7			1798.1	15.69	67.96
DDS7	301.30	9.6			1786.5	15.38	62.35
DDS8	444.52	12.3			1780.8	22.62	54.38
DDS9	335.33	9.0			1764.8	16.91	53.51
DDS10	300.42	8.0			1776.8	15.25	52.46
DDS11	309.38	8.3			1792.8	15.85	52.16
DDS12	364.44	9.5			1792.5	18.66	50.77

^aCompounds were selected for follow-up binding affinity studies if they exhibited a Level of Occupancy (%) >50% and <200% and were not flagged as non-specific binders. For each compound, the table includes molecular weight (MW), analyte binding level (corresponding to the Analyte Binding Rate_1 Relative [RU] at the end of the association phase), ligand name and MW, immobilized ligand level, the calculated R _max value, and the corresponding Level of Occupancy.

To further characterize the binding interactions of the identified CD28 binders, we investigated whether each compound exhibited a dose-dependent response. Using SPR, we successfully measured K _d values in the micromolar range for three compounds (Table and Figure ). Additionally, two compounds demonstrated weaker binding affinities, with K _d values in the low millimolar range (data not shown). The remaining compounds did not display a clear dose–response relationship (Table ).

2. Binding Affinities and Structural Information of Selected Primary Small Molecule Hits .

^aThis table summarizes the binding affinities (K_d) and chemical structures of all compounds tested for interaction with the CD28 protein. Binding affinities are reported in micromolar (μM) or millimolar (mM) concentrations, with standard deviations where available. ND indicates that the binding affinity was not determined due to lack of dose–response effect. Non-specific binder denotes compounds that exhibited a dose–response effect but did not show selective binding to the active flow cell.

3.

Binding affinity of selected small molecules to immobilized human CD28 protein. (A) Serial dilutions of compound DDS2 (ranging from 400 to 35.12 μM, including a 0 μM reference; 1.5-fold dilution series) were injected onto a Sensor Chip CAP with immobilized human CD28 protein using a multicycle kinetics method. (B) Serial dilutions of compound DDS5 (ranging from 400 to 35.12 μM, including a 0 μM reference; 1.5-fold dilution series) were similarly injected using the same method. (C) Serial dilutions of compound DDS9 (ranging from 500 to 66.41 μM, including a 0 μM reference; 1.4-fold dilution series) were also injected under the same conditions. Binding curves were analyzed using nonlinear curve fitting based on steady-state affinity analysis. The sensorgrams show relative Response Units (RU) over time during a 90-s association phase and a 240-s dissociation phase, representing one of three independent experiments.

Interestingly, none of the three flagged compounds from the initial set of 12 hits exhibited a dose–response effect. Instead, we observed either nonspecific binding to both the active and inactive flow cells of the sensor chip or a lack of dose-dependent behavior. This highlights the robustness of our screening approach: from the original 1056-compound library, eight unflagged hits were identified in a single day (excluding one that was not commercially available), and five of these ultimately confirmed specific binding to CD28.

The use of SPR in a solution-phase format enabled direct, real-time measurement of binding kinetics, a key advantage over traditional HTS methods, which often rely on end point or indirect readouts. This kinetic resolution was instrumental in distinguishing true binders from nonspecific interactions, thereby reducing the likelihood of false positives. We attribute this success to the high specificity and sensitivity of the SPR-based approach, which contrasts with other HTS platforms such as microscale thermophoresis, known for higher false-positive rates due to indirect detection and limited kinetic insight. Our methodology thus provides a more reliable alternative for screening structurally complex PPI targets like CD28.

Following the identification of three CD28 binders with defined binding affinities, we sought to characterize their binding mode to CD28 and evaluate their potential to inhibit the CD28–CD80 interaction. As a proof of concept to validate the robustness of our HTS platform, the top-ranking compound, DDS5, was selected for further mechanistic and functional validation experiments.

The costimulatory receptor CD28 and its inhibitory counterpart CTLA-4 share a common ligand binding mechanism through their interaction with B7 family proteins (CD80/CD86). Previous crystallographic studies have established that the primary ligand binding site for both CD28 and CTLA-4 is located within a conserved region comprising residues 99–104. Mutagenesis experiments have demonstrated that alterations within this sequence result in a substantial reduction (>90%) in ligand binding affinity, confirming its critical role in receptor–ligand recognition. However, the primary ligand binding region of CD28 presents significant challenges for small molecule drug development. Analysis of the three-dimensional structure (PDB ID: 1YJD) revealed that the CD28–CD80 binding interface is characterized by a relatively shallow, extended surface lacking the well-defined binding clefts typically required for high-affinity small molecule interactions. Such flat protein–protein interaction surfaces are notoriously difficult to target with drug-like compounds due to their limited capacity to provide the multiple contact points necessary for selective binding.

Although the primary ligand-binding site of CD28 presents significant challenges for small-molecule intervention due to its shallow and extended topology, structural analysis has identified a secondary binding pocket with more favorable features for small molecule targeting. This CD28-specific cavity is positioned adjacent to the canonical ligand-binding region and is formed by a discontinuity in β-strand G. The pocket is delineated by key residues including His38, Phe93, Lys95, Asp106, and Lys109 forming the surrounding walls, with Val5, Asn107, and Ser110 comprising the base of the binding site (Figure A, highlighted region). These structural attributes suggest that the pocket may serve as a viable target for structure-based design of small molecule CD28 modulators.

4.

Structural analysis and molecular docking of CD28-DDS5 binding site. (A) Crystal structure of CD28 (PDB ID: 1YJD) showing the primary ligand binding site, predicted by the PrankWeb server to be the most likely binding site (secondary pocket, blue color). This secondary pocket is formed by a discontinuity in β-strand G and is surrounded by key residues His38, Phe93, Lys95, Asp106, and Lys109. The adjacent (primary) binding site is highlighted in red. (B) Docked conformation of compound DDS5 within the identified binding pocket, showing optimal fit and induced conformational changes. (C) 3D bound complex of DDS5 with CD28. (D) Detailed view of DDS5 binding interactions, highlighting the critical hydrogen bond formation with Cys94 and multiple hydrophobic contacts with surrounding residues.

The therapeutic relevance of targeting regions adjacent to primary binding sites is supported by precedent from CTLA-4 research. Structural studies of the anti-CTLA-4 antibody tremelimumab have demonstrated that nine of the ten residues within the heavy chain complementarity-determining region 3 (HCDR3) loop (residues 101–110) are directly involved in CTLA-4 binding. These findings led to the hypothesis that cyclic peptides derived from the HCDR3 sequence could effectively block CD80/CD86 binding to CTLA-4. Given the high degree of sequence and structural homology between CTLA-4 and CD28, it is plausible that analogous binding regions on CD28 may similarly be exploited for therapeutic modulation. In particular, the identified secondary pocket on CD28composed of residues His38, Phe93, Lys95, Asp106, Asn107, Lys109, and Ser110may represent a viable target for small-molecule intervention.

To further evaluate this hypothesis, we employed the SiteMap module within the Maestro software suite (Schrödinger). SiteMap utilizes physics-based scoring criteria to identify and rank potential binding pockets while filtering out regions unlikely to exhibit druggable properties. Consistent with our manual structural analysis, SiteMap ranked the secondary pocket adjacent to the canonical ligand-binding site as the second most favorable site for small-molecule binding, reinforcing its potential as a novel druggable interface on CD28 (data not shown). To provide additional validation, we employed the PrankWeb server for complementary binding site prediction. PrankWeb utilizes machine learning algorithms trained on protein–ligand complexes to predict ligandable sites and assess their drugability. Remarkably, this independent analysis identified our proposed secondary binding site as the most druggable pocket with a score of 6.91 and an average conservation score of 0.78.

The top hit compound, DDS5, which exhibited the strongest binding affinity to CD28, was subjected to molecular docking analysis using the Induced Fit Docking (IFD) protocol wizard of Maestro Schrodinger. This approach accounts for the flexibility of both the ligand and the receptor, allowing for realistic conformational adjustments during the binding process. DDS5 demonstrated favorable binding characteristics within the previously identified secondary pocket (Figure B). The docking results suggested that the compound induces local conformational rearrangements within the binding site, optimizing the geometry to accommodate the ligand. Visual inspection of the predicted binding pose revealed the formation of a key hydrogen bond with Cys94, along with multiple hydrophobic interactions involving adjacent residues (Figure D). The combination of polar and nonpolar contacts contributes to a well-anchored and specific binding mode, supporting the potential of DDS5 as a structurally viable modulator of CD28 function.

To evaluate the stability of the predicted protein–ligand complex, we performed two independent 100 ns MD simulations: one for unbound CD28 and the other for CD28 in complex with DDS5. Throughout the simulation period, DDS5 remained stably bound to CD28, as evidenced by the root-mean-square deviation (RMSD) trajectory (Figure A, blue line). RMSD analysis revealed an average deviation of approximately 1.3 Å, indicating minimal structural perturbation and a highly stable binding conformation.

5.

MD simulation analysis of compound DDS5 stability in CD28 binding site. (A) Root mean square deviation (RMSD) trajectory of DDS5–CD28 complex over 100 ns simulation (blue line), demonstrating stable binding with average RMSD of ∼1.3 Å. (B) Protein–ligand contact histogram for DDS5 showing sustained interaction with Phe93 throughout the simulation period, confirming binding stability and the critical role of this residue.

The closeness of the RMSD values of the bound and unbound CD28 complexes are derived from the subsequent MD simulations and reflect the overall structural stability of the protein–ligand complex once formed. These values indicate that following the initial induced-fit docking and system equilibration, the complex remains stable throughout the simulation trajectory, which supports the validity of the binding pose generated by the IFD protocol. To further support the stability of the interaction, protein–ligand contact analysis was conducted, which demonstrated that DDS5 maintained persistent interactions with Phe93 across the full 100 ns simulation (Figure B). This sustained contact highlights both the structural integrity of the complex and the critical role of Phe93 as a potential anchor residue for ligand recognition and stabilization.

Finally, we performed a quantitative assessment of the CD28–CD80 disruption by our DDS5 candidate using an ELISA-based approach. As previously discussed, disrupting the CD28–CD80 interaction offers a promising strategy to modulate T cell costimulation, with therapeutic implications in not only in autoimmunity, but also in transplant rejection, and cancer. − In pathological conditions involving hyperactive immune responses, blocking this pathway can help suppress excessive inflammation and restore immune balance. To evaluate the ability of small molecules to interfere with this interaction, we conducted a competitive ELISA using immobilized CD28 and biotinylated CD80 as the ligand. Compound DDS5 was tested across a range of concentrations and exhibited a dose-dependent inhibition of CD80 binding with an IC_{5 0} value of 332 μM (Figure ). This finding is consistent with its moderate binding affinity (K _d = 175.57 ± 66.84 μM). Together, these results provide confirmation that our top compound directly engages CD28 and can interfere its binding with CD80, supporting its potential as an early stage inhibitor for further optimization.

6.

Inhibition of CD28–CD80 interaction by compound DDS5. Dose-dependent inhibition of CD28–CD80 binding by compound DDS5 was measured using a competitive ELISA assay. The compound demonstrated an IC_{5 0} value of 332 μM. The maximum luminescence signal with no compound added was considered as 100% binding, while the minimum luminescence observed at the highest compound concentration was considered as 0% binding. All intermediate values were normalized accordingly between these two reference points. IC₅₀ calculations were performed using GraphPad Prism, applying the log­(inhibitor) vs response – Variable slope (four-parameter) nonlinear regression model. The graph shows the raw data from a single experiment with two technical replicates. The data are presented as the mean ± standard deviation.

Conclusions

In this study, we report the first successful implementation of a SPR-based HTS platform for the identification of small-molecule modulators targeting CD28a historically challenging immune checkpoint receptor. By leveraging recent technological advancements in SPR, including enhanced sensitivity, throughput, and data resolution, we established a robust, label-free, and solution-phase method capable of directly quantifying binding kinetics and affinities. This makes SPR particularly advantageous for interrogating challenging targets such as CD28, which present flat, dynamic, and poorly defined PPI interfaces.

Our platform enabled the rapid and reproducible screening of a structurally diverse library of 1056 compounds, culminating in a hit rate of 1.14%, with 5 of 12 primary hits (41.7%) exhibiting dose-dependent binding and quantifiable affinities. Notably, the single-concentration HTS campaign was completed within a single day, underscoring the operational efficiency and reliability of the workflow. The integration of biophysical screening with computational modeling and functional validation represents a comprehensive strategy for triaging hits early in the discovery process.

Our top hit compound, DDS5, demonstrated micromolar binding affinity, formed a structurally stable complex with CD28 as shown by MD simulations, and inhibited CD28–CD80 binding in a competitive ELISA. These results provide the first functional evidence of CD28–CD80 disruption by a nonpeptidic small molecule. Several of the identified hits share a 1,4-substituted 1,2,3-triazole scaffold, a motif frequently found in inhibitors of diverse proteins and PPIs. While such scaffolds are broadly used in drug discovery, the specific compounds identified here have not been previously reported as CD28 modulators, highlighting their novelty and potential for further optimization. Docking studies further suggested a mechanism involving steric hindrance within a previously unexploited secondary pocket on CD28, providing a rational foundation for future structure–activity relationship (SAR) and hit-to-lead optimization campaigns. While additional biophysical validation assays were not feasible at this early stage, future orthogonal validation will be incorporated as more potent analogs are developed.

Beyond the specific case of CD28, this study underscores the broader utility of SPR as a central platform in modern drug discovery. SPR not only accelerates early stage screening and target validation but also enables detailed mechanistic interrogation, lead prioritization, and risk reduction in downstream development. Compared to other biophysical techniques such as thermal shift assays, X-ray crystallography, and NMR, SPR offers unique advantagesreal-time kinetic resolution, low sample consumption, and broad molecular compatibilitymaking it indispensable for fragment-based design, mode-of-action studies, and selectivity profiling.

Taken together, our work establishes a blueprint for SPR-enabled discovery of small-molecule immune modulators and highlights its potential to expand the therapeutic landscape for T cell–targeted immunotherapies. Furthermore, the approach may be extended to other costimulatory pathways, such as ICOS, and integrated into multitarget discovery pipelines aimed at addressing complex immune-driven diseases. By demonstrating the druggability of CD28 with small molecules, this work contributes to redefining what is possible in the chemical modulation of immune checkpoint biology.

Glossary

Abbreviations

CAP

capture sensor chip

CD28

cluster of differentiation 28

CD80

cluster of differentiation 80 (B7-1)

CD86

cluster of differentiation 86 (B7-2)

DDS

Discovery Diversity Set

ELISA

enzyme-linked immunosorbent assay

GPCR

G protein-coupled receptor

HTS

high-throughput screening

IC_{5 0}

half-maximal inhibitory concentration

K _d

equilibrium dissociation constant

LO

level of occupancy

MD

molecular dynamics

PPI

protein–protein interaction

R _max

maximum theoretical response

RU

response units

SPR

surface plasmon resonance

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.5c10222.

• Experimental procedures and chromatographic and mass spectra data for the identified hit compounds (PDF)

The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.

The authors declare no competing financial interest.