Natural killer (NK) cells are key players in cancer immunosurveillance, targeting and lysing tumour and virus-infected cells through non-specific mechanisms1,2. In recent years, there has been growing interest in harnessing NK cells for cancer immunotherapy, with strategies ranging from bi-and tri-specific killer engagers that redirect NK cells to tumours, to the development of chimeric antigen receptor NK (CAR-NK) cells3,6,7. While these approaches show promise, antibody-based immunotherapies remain extremely expensive, and their efficacy across cancer types is limited. In particular, breast and prostate cancers often show low response rates, and there is significant heterogeneity in treatment outcomes across different tumour lesions within the same patient4,5.
The effectiveness of NK cells’ cytotoxic response depends on the engagement of specific receptors, such as NKp30 (NCR3), which are critical for initiating cytotoxic responses. However, cancer cells have multiple mechanisms of immune evasion, downregulating natural cytotoxicity receptors (NCRs) expression or saturating them with activating ligands6,8. Moreover, the tumour microenvironment is also highly immunosuppressive, due to the secretion of cytokines such as IL-10 and TGF-β9,10. Together with the high costs, these factors underscore the significant limitations of current NK cell-based therapies3.
Addressing these challenges, the structure of the NCRs provides a starting point to design specific ligands to trigger NK cells’ cytolytic activity, as demonstrated by the synthesis of small peptides that bind NKp30 and trigger cytokine release in the NK92-MI cell line11,12 . Building on this, small-molecule drugs such as the toll-like receptor 7 (TLR7) agonist imiquimod have shown success in immune activation, though their applications remain limited by non-specificity and a focus on topical use13-15.
Beyond imiquimod, the development of small-molecule agents that target pattern recognition receptors (PRRs), immune checkpoints, and oncogenic pathways has expanded the therapeutic landscape. In parallel, adoptive NK cell transfer therapies, supported by agents like IL-15, have demonstrated efficacy in promoting NK cell survival and inducing remission in certain cancers16. However, the need for more accessible and scalable therapies remains pressing.
Designing NKp30-Targeting Ligands
Within this paradigm, identifying a small organic hit compound able to engage NKp30 could pave the way for a novel drug design strategy. To this end, a computational approach was applied to design a selective NKp30-targeting small organic molecule (SOM).
The target region of NKp30 was defined based on its interaction with the natural peptide ligand B7-H6. Structural and mutagenesis studies have identified a small binding interface on NKp30, predominantly composed of hydrophobic amino acids (including Trp55, Val66, and Leu93), which forms a ridge extending inward to create a cavity11,17,18. Importantly, alterations in key residues, such as Ser52 and Phe85, have been shown to significantly reduce binding affinity, highlighting the sensitivity of NKp30 and B7-H6 interactions to structural changes19.
To identify scaffolds capable of interacting with the NKp30 target region, a diversity-focused virtual library of approximately 0.5 million compounds was retrieved from ChemBank20. This library consisted of molecules under 600 Da, each containing no more than four aromatic rings, five nitrogen, and five oxygen atoms. The unbound conformation of NKp30 (PDB ID 3NOI)17 was used as the receptor model, allowing broader ligand interactions to identify novel scaffolds. A flexible ligand–rigid receptor docking approach was applied to screen all compounds against the NKp30 binding site 11.
The 10 ligands with more favourable binding energies (SI, Table S1) were visually inspected to confirm proper orientation within the binding site. Top candidates were filtered for the presence of pan-assay interfering groups (PAINs)21, aggregators22-24, and reactive Michael acceptor groups25. Ligands with a biphenyl core (1-3, Fig. 1a) demonstrated better pocket occupancy compared to other compounds, likely due to the biphenyl moiety ability to adopt dihedral planes other than 0 and 180º, resulting in a larger overall volume. These ligands also follow the tendency to bury their most hydrophobic domain in the deep region of the binding pocket (Fig. 1b, hit 1) when compared with hits missing this feature (Fig. 1b, hit 2). Hit-compounds 1 to 3 were identified as promising leads, with binding energies below -8.4 kcal/mol. Compounds 1 and 2 also feature a bis(4H-3,1-benzoxazin-4-one) core that was deemed synthetic accessible upon retrosynthesis analysis. Compound 4 (Fig. 1c), retaining this core, was designed to have one terminal substituted phenyl ring replaced with an alkyl chain, introduced for future derivatization. This scaffold was further diversified by introducing various substituents at the remaining terminal phenyl group (SI, Table S2). covering a broad range of physicochemical properties, such as hydrogen bond donors and acceptors, polar surface area, and carbon atom count. These compounds were docked against the unbound conformation of NKp30 as before, revealing consistent hydrophobic interactions between the aromatic moieties and key amino acid residues located in the deep region of the pocket. Binding energies averaged -8.37 kcal/mol, ranging from -9.9 to -7.6 kcal/mol. Moreover, the adjacent six-membered ring fused with the biphenyl moiety interacted with Leu80 and Arg67 through potential hydrogen bonds (Fig. 1d). Additional hydrophobic contacts were observed between compound 4 and residues Leu48, Ile50, Gly51, Val53 and Leu86 (Fig. 1e).
Analysis of the binding energies in relation to molecular structure revealed a negative correlation between the topological polar surface area (TPSA) of the R1 substituent (Fig. 1f) and the binding affinity. In contrast, the volume of R1 had a positive contribution to the binding energy. Positional analysis indicates that ortho-substituents enhance binding, whereas meta-substituents, while positioned away from the deep region of the binding pocket (compounds 6 and 8–10, Fig. 1c), are well tolerated. Among the different types of substituents examined, the addition of nitro groups leads to an improved binding energy. Bulkier ortho-substituents such as chlorine atoms and nitro groups, fit well within the binding pocket, assuming the expected orientation (5 to 7, Fig. 1c). The binding affinities of these compounds correlated closely with the volume of the substituents, reinforcing the relevance of size in optimizing ligand-receptor interaction (Fig. 1e).
Notably, the similar binding affinities observed for both Cl-- and NO₂-substituted compounds suggest that volume plays a limited role in these cases, indicating that meta substitutions are well-tolerated. Additionally, the introduction of a second meta group, as seen in compounds 9 and 10, produced minimal changes in binding affinity, implying that the binding pocket accommodates bulky groups at the 3 and 5 positions of the terminal phenyl ring.
Based on the docking results, molecules of types F and H (SI, Table S2) were excluded from further consideration due to their less favourable binding affinities. Although type B compounds demonstrated relatively good affinities, they were not selected for additional studies as they closely resembled types C and D. Consequently, molecules from types A, C, D, E, and G were prioritized for further investigation, and seven ligands from these groups were subsequently synthesized.
A robust synthetic entry for the bis(4H-3,1-benzoxazin-4-one) scaffold was achieved using bis-anthranilic acid (11) as starting material, which was reacted with activated pentanoic acid to yield the amide derivative 12 (Fig. 2a). Subsequent reaction of 12 with activated carboxylic acids, followed by cyclodehydration, yielded compound family 13 (13a–g). Although these compounds are stable in organic solution, they undergo prompt hydrolysis in aqueous conditions, leading to the opening of the 6H-1,3-oxazin-6-one ring on the alkyl chain side, producing family 14 derivatives (Fig. 2a). Hydrolysis on the opposite benzoxazinone moiety proceeds significantly slower, likely due to steric hindrance at that site.
Evaluating NKp30 binding
The ability of compounds from families 13 and 14 to bind to NKp30 was assayed by a mass spectrometry-based approach. Commercial rhNKp30 was found, by SDS-PAGE, to be extensively and aberrantly glycosylated (SI, Fig. S1), spanning a molecular weight range of almost 6 kDa 27. The necessary deglycosylation was successfully achieved enzymatically with Peptide:N-glycosidase F (PNGaseF). While this process is known to convert glycosylated asparagines into aspartic acid residues27, rhNKp30 only contains three Asn residues (Asn58, Asn88, and Asn97), which are not part of the B7-H6 binding site. Thus, no concerns arose from PNGaseF conversion of glycosyl-Asn to Asp in rhNKp30 as it does not interfere with the pocket structure of ligand affinity. A ligand-observed assay (MS binding assay), relying on the detection and quantification of the ligands released from protein-ligand complexes28 was employed to assess relative binding affinities.
An equimolar mixture of compounds 13a to 13g was incubated with rhNKp30deglyc at a 1:1 molar ratio overnight at 37 °C. After ultrafiltration to remove unbound ligands, the protein-bound ligands were released and analysed by HPLC-ESI-HRMS/MS. This allowed identification of the bound ligands by comparison with a blank control (without protein). Lysozyme, with a similar molecular weight and globular shape as rhNKp30deglyc, was used as a control to identify non-specific binding, given its distinct binding site composition of both hydrophobic and polar residues29,30. The NKp30 binding site, in contrast, is predominantly hydrophobic11.
Signal-to-noise (S/N) ratios were calculated by comparing peak areas in the protein incubations against the blank, with S/N values >10 considered indicative of positive binding31. Compounds 13b, 13c, and 13e showed significant binding to rhNKp30deglyc, with no relevant interaction with lysozyme (Fig. 3a). The remaining compounds did not interact significantly with either protein. Further MS/MS analysis revealed hydrolysis of compound 13b, reversing cyclodehydration on the alkyl chain side. This phenomenon (water attack on the carbonyl C atom of the alkyl-substituted benzoxazinone group), observed across all tested compounds, yields products 14a-g. Since ligand release was performed in anhydrous methanol, the hydrolysis likely occurred during the protein incubation step.In this competitive binding assay, both hydrolysed and non-hydrolysed forms were evaluated for receptor interaction. After normalizing peak areas from the extracted ion chromatograms, compound 14c, the hydrolysed form of 13c, was confirmed as a strong NKp30 binder, with no significant interaction observed with lysozyme (Fig.3a).
Navigating Stability and Solubility
At this stage, compound 14c was identified as the lead candidate based on the protein-ligand binding studies. Although the docking results were not definitive, they suggested that 13c and 14c share a similar binding mode, revealed by the superposition of all binding poses (Figs. 3b,c). This indicates that the pharmacophore is likely centred around a dinitrobenzene ring linked to a benzoxazinone core, further connected to a phenyl ring with a carbonyl group in the meta position (Fig. 3d).
The alkyl chain, which significantly contributed to the hydrophobicity of these compounds, was deemed non-essential for affinity to the NKp30 binding site and was replaced by a polar group to improve aqueous solubility. A shorter, chlorine-substituted 1-carbon chain (15, Fig. 2a) was introduced to facilitate subsequent derivatization with 1,4-diazabicyclo[2.2.2]octane (DABCO) (16 and 17, Fig. 2a). DABCO was selected over other tertiary amines due to its superior nucleophilicity and the reduced steric hindrance around its amine groups. Additionally, the introduction of a non-labile charged group was expected to enhance solubility and orient the hydrophobic portion of the molecule more effectively within the binding pocket.
Compound 17 was synthesized as a more soluble surrogate of compound 13c. However, during biological assays (see next section), compound 17 exhibited significant inter-day and inter-assay variability, beyond what could be attributed to donor variability alone.
NMR studies conducted under buffered conditions (pH 7.4) revealed that 17 was rapidly hydrolysed to 18 in aqueous medium, with an half-life of ca. 4 min (SI, Fig. S2).
The increase in solubility appeared to render the benzoxazinone rings more susceptible to hydrolysis, leading to concerns that the resulting compound 18 might be too flexible to function as an effective NKp30 ligand. Furthermore, the free carboxylic acid in compound 18 could disrupt critical hydrophobic interactions necessary for NKp30 activation. Despite these limitations, compound 18 was synthesized and tested for biological activity.
To overcome hydrolysis of compound 17, a more water-stable analogue was envisioned. Thus, the benzoxazinone moiety was replaced with a 4-quinazolinone moiety, which is less prone to nucleophilic attack by water, yielding compound 21 (NK815A, Fig 2b). Among all synthesized compounds, 17, 18, and 21 were selected for biological testing. Compound 17 was chosen as a precursor for the presumably active compound 22, while compound 18, anticipated to exhibit inactivity due to its high flexibility, served as a control.
Impact on NK cell cytotoxic activity
NK cells mediate cytotoxicity through both direct and indirect mechanisms, involving perforin- and granzyme-mediated target cell lysis, as well as the release of cytokines and chemokines11. Thus, assessing NK cell activation in response to putative ligands requires the evaluation of both cytolytic activity and cytokine production.
In this study, prior to the assessment of the effect of the synthesized compounds in NK cells, validation of each model was essential, given the limited research in this specific area. Model validation was conducted by assessing the target cell killing capacity and quantifying cytokine production following stimulation with B7-H6, the natural ligand for NKp30 NK cells were isolated from peripheral blood mononuclear cells (PBMCs), from healthy donors, using negative selection via magnetic-activated cell sorting (MACS). This approach yields highly pure, untouched NK cell populations with minimal granulocyte contamination, as previously described33,34.
NK-induced cytotoxicity was assessed using HCT116 cells as target via a calcein-based method. Adherent HCT116 cells were loaded with non-fluorescent cell-permeant calcein-acetoxymethyl (calcein-AM), which is converted intracellularly to membrane-impermeable fluorescent calcein. Upon NK cell addition, calcein fluorescence in the culture supernatant correlates with target-cell lysis35,36. Despite donor variability, all NK cell populations displayed the ability to kill target cells without prior sensitization, a hallmark of NK cells (Fig. 4a). NK cell activity was further analysed by measuring B7-H6-induced granzyme B release. After 24 h incubation with this natural NKp30 agonist, granzyme B levels in the culture medium significantly increased upon stimulation with 0.1 µg/mL (3.6 nM) rhB7-H6, but this effect diminished at concentrations above 0.2 µg/mL37 (Fig. 4B). These results demonstrate that MACS-isolated untouched NK cells from PBMCs are functionally active and responsive to external stimuli, validating their reliability as a model system for investigating NK cell activation mechanisms.
To evaluate the activity of the synthesized lead compound, NK cells were incubated for 24 hours with NK815A in serum-free culture medium, and cytokine levels in the culture supernatant were measured by ELISA. IFN-γ levels showed a 40% increase compared to the control at a 40 nM of NK815A, while TNF-α displayed a 20% increase at the same concentration (Fig. 4c). Granzyme B release from NK815A-activated NK cells followed a similar trend, with a 1.2-fold increase relative to control (Fig. 4c). The limited cytokine elevation can be attributed to the absence of crosstalk with other cell types, which play a crucial role in modulating cytokine production by NK cells38. In fact, when PBMCs were incubated with NK815A, TNF-α and IFN-γ expression increased by approximately 1.8-fold and 5.0-fold, respectively, relative to control (Fig. S3).
Granzyme B and perforins are secreted by NK cells during degranulation, a process that was monitored in real time using a fluorescent assay that tracks the shift of acridine orange (AO) fluorescence from orange (AOH⁺ in intracellular acidic granules) to green (AO at neutral pH) upon release39. When isolated NK cells were incubated with the natural NKp30 agonist B7-H6 (Fig. 4d), the green/orange fluorescence ratio increased (indicating degranulation) starting at 0.2 µg/mL of B7-H6, following a well-defined bell-shaped dose-response curve. Similarly, NK815A induced an increase in that ratio, beginning at 1.25 µM (Fig. 4e). Kinetically, B7-H6 showed a faster response, peaking at 0.4 µg/mL, whereas NK815A displayed a slower but sustained degranulation response up to 5 µM (Figs. 4d,e).
Given that NK815A triggered both cytokine production and degranulation in NK cells, its potential to enhance NK-mediated cytolysis was tested using calcein-labeled HCT116 cells. Incubation of labelled HCT116 cultures with NK815A alone did not induce calcein release (Fig. 4f), confirming the absence of both cytotoxicity and direct activity on target cells. However, co-incubation of NK cells with NK815A and labelled HCT116 cells resulted in a 2-fold increase in specific cell lysis compared to control (Fig. 4g). This response followed a bell-shaped curve, with an estimated IC50 of 4.6 µM and an EC50 of 176 nM (Fig. 4h), demonstrating that NK815A promotes NK-mediated lysis (Fig. 4h).
The effect of NK815A on NK cell cytolytic activity was further evaluated in the presence of IL-2 (250 U/mL). The normalized effect of NK815A on the cytotoxicity of NK cells against HCT116 cells (Fig. 4i) was found to be independent of IL-2.
The non-cytotoxic nature of NK815A was confirmed in HCT116 (Fig. 4j) and MCF-7 (Fig. 4k) cells, with no cell viability reduction observed in MTT assays. Furthermore, NK815A also did not induce lactate dehydrogenase release from PBMCs (Fig. 4l), underscoring its safety profile.
Taken together, these results demonstrate that NK815A effectively enhances NK cell cytotoxicity through NKp30 activation, promoting both cytokine production and degranulation. NK815A-induced NK cell-mediated lysis was robust, with an estimated IC50 of 4.6 µM. Importantly, NK815A exhibited no direct cytotoxicity on target cells or PBMCs, confirming its specificity and safety, making it a promising candidate for NK activation towards cell-based cancer immunotherapy.