Computational Cosolvent Mapping Analysis Leads to Identify Salicylic Acid Analogs as Weak Inhibitors of ST2 and IL33 Binding

Abstract

Cytokine signaling initiated by the binding of the cytokine receptors to cytokines plays important roles in immune regulation and diseases. Structurally, cytokine receptors interact with cytokines via an extensive, rugged interface that represents a challenge in inhibitor development. Our computational analysis has previously indicated that butyric acid, mimicking acidic residues, preferentially binds to sites in ST2 (Stimulation-2) that interact with acidic residues of IL33, the endogenous cytokine for ST2. To investigate if a charged group in small molecules facilitates ligand binding to ST2, we developed a biochemical homogeneous time resolved fluorescence assay to determine the inhibition of ST2/IL33 binding by five molecules containing an aromatic ring and a charged group. Three molecules, including niacin, salicylic acid, and benzamidine, exhibit inhibition activities at millimolar concentrations. We further employed the computational cosolvent mapping analysis to identify a shared mode of interaction between niacin, salicylic acid, and ST2. The mode of interaction was further confirmed by four analogous compounds that exhibited similar or improved activities. Our study provided the evidence of inhibition of ST2 and IL33 binding by salicylic acid and analogs. The results suggest that biological activity of salicylic acid may be partly mediated through modulating extracellular cytokine receptors and cytokine interaction.

Graphical Abstract

INTRODUCTION

Membrane-bound cytokine receptors are sensory molecules on immune cells to surveil extracellular cues and instruct immune cells to respond. Action initiated by cytokine receptors is usually triggered by the binding of cytokine receptors with the cognate cytokines generated by various cells challenged by stimuli. Binding between cytokine receptors and cytokines can lead to the activation and differentiation of the immune cells in their response to infection,¹ tissue damage, and repair.^2–4 Stimulation of the immune cells by cytokines is also important for immune cell homeostasis⁵ including the hematopoietic differentiation⁶ and the maintenance of regulator T cells (Treg) by Interleukin 2 (IL2).^7,8 Dysregulated cytokine signaling can lead to delirious outcomes and has been associated with disease manifestation and progression. Examples include the tumor-necrosis factor (TNF) in rheumatoid arthritis⁹ and various cytokines contributing to cytokine storm (cytokine release syndrome) observed in COVID-19 patients.^10–12

Among the cytokine receptors, the IL1 receptor (IL1R) family has five primary members and six auxiliary co-receptors.^13,14 ST2 is a member of the IL1R family and binds to the only known cytokine, IL33. On the cell membrane, ST2 binds with IL33 and IL-1RAcP to signal the ST2/IL33 axis^15,16 that plays important role in Th2 (T-helper 2) cell biology.² An alternatively spliced ST2 isoform, called soluble ST2 (sST2), circulates in the blood stream and functions as a decoy receptor to attenuate IL33 concentration. Excess sST2 was associated with the disease progression in inflammatory bowel disease,^17,18 ulcerative colitis,¹⁹ heart failure,²⁰ acute cardiac allograft rejection,²¹ graft versus host disease,^22,23 and others.^15,16,23,24 Pharmacological intervention to target cytokine receptors has mainly focused on developing antibodies to bind to cytokine receptors or cytokines. Existing antibody therapy includes etanercept, infliximab for TNF, and Canakinumab, Gevokizumab, and LY2189102, which are mainly anti-IL-1β monoclonal antibodies.²⁵ Two ST2 antibody therapies, RG6149/AMG282 and GSK3772847, are currently in phase 2 clinical trials.²⁶

Cytokine receptors bind to cytokines using multiple domains. As an example, three immunoglobulin-like domains (D1, D2, and D3) of ST2 adopt a glove-like shape to bind the globular shape IL33.²⁷ The interaction interface between ST2 and IL33 is 1741.3 Ų.²⁸ This type of interaction was observed in all members of the IL1R family that includes IL1R1/IL1β^29–31 and IL18R/IL18.^32,33 The large contact interface between cytokine receptors and cytokines is devoid of well-defined small-molecule binding pockets³⁴ and is a known challenge for small-molecule inhibitor development. Despite the difficulty, small-molecule inhibitors targeting IL18 to inhibit its binding with IL18R was reported.³⁵ We also reported a set of ST2 inhibitors that inhibited ST2/IL33 binding and demonstrated their in vivo activities in graft versus host disease (GVHD) animal models.³⁶

For efficient inhibitor development, co-crystal structures between the proteins and inhibitors will provide important guidance. However, determination of the crystal structures of our ST2 inhibitors with ST2 remains challenging, although potential binding locations of ST2 inhibitors on ST2 have been proposed using a combination of small angle X-ray scattering experiment and computational modeling.³⁶ To identify small-molecule binding sites in proteins, a number of computational methods can be used instead, including our cosolvent mapping method,³⁷ multiple copies simultaneous search method (MCSS),³⁸ FTmap,³⁹ MDmix,⁴⁰ mixed solvent molecular dynamics (MixMD) method,^41,42 site identification by ligand competitive saturation (SILCS),^43,44 and cosolvent molecular dynamics (MD) simulations based on λ-dynamics.⁴⁵ All these methods employed fragment-like molecules (MW < 300) as probes to identify and assess the binding sites in proteins. In our quest to identify small-molecule binding sites in ST2, we recently employed the butyric acid (BUA), a metabolite, to probe the ectodomain of ST2 using our computational cosolvent mapping analysis.⁴⁶ The rationale was based on the findings that mutation of five acidic residues (E144, E148, D149, D244, and E165) in IL33 led to a decreased binding affinity (K_D) to ST2 by 2 to >167 fold.^47,48 We hypothesized that BUA may mimic the sidechains of glutamic and aspartic acids (Glu, Asp) to interact with ST2 and indeed we found BUA bound to the same locations in ST2 that interacted with E144 and D149 in IL33.

Intrigued by the observation, we reasoned that using aromatic molecules with a charged group as probes can offer an opportunity for inhibitor development if their binding interaction with ST2 were identified because the aromatic ring can be subject to diverse modifications. Because binding affinities of fragment molecules (MW < 300) to proteins are typically at high micro- or milli-molar concentration,⁴⁹ high solubility of the probes may be required for experimental validation. Based on these considerations, we initially selected five bioactive metabolites to probe binding sites in ST2 in this work. They included nicotinic acid (or niacin, NA), salicylic acid (SA), nicotinamide (NAM), benzamidine (BZA), and 3-(aminomethyl)pyridine (AMP). Both NA and SA have a carboxylic acid group to mimic the acidic group of Glu and Asp. NAM is an analog of NA with the carboxylic acid replaced by an amide group. In contrast, both BZA and AMP are structurally similar to NA except with opposite charges. Both NA and SA are expected to bind to basic sites on ST2, whereas BZA and AMP will preferentially bind to acidic sites on ST2. The amide group of NAM will be used to examine the influences of the hydrogen bond strength between NAM and NA on their recognition of the basic or acidic sites on ST2.

To validate our computational analysis, we developed the biochemical binding assay between ST2 and IL33 using the homogeneous time resolved fluorescence (HTRF) technology⁵⁰ to determine the inhibition activity (or IC₅₀) of probes. Our experimental activity evaluation and computational cosolvent mapping analysis led us to identify NA and SA as weak inhibitors of ST2/IL33 binding. Aided by the shared binding mode of SA and NA with ST2 obtained from the cosolvent molecular dynamics (MD) simulations, we selected seven compounds analogous to SA for evaluation. We found that two additional fragment-like molecules exhibited 2-fold higher inhibition activities than SA. Because of the prominent use of SA in immune-mediated diseases⁵¹ and its use as a scaffold for inhibitor development,^52,53 our study have implications for investigating the role of SA in cytokine signaling and using SA as a scaffold molecule for inhibitor development.

METHODS

ST2/IL33 HTRF Assay.

The buffer used in the assay contains 9 mL phosphate-buffered saline (PBS), 18.7 μL Tween 20 (Sigma-Aldrich), and 0.375 mL of 5% bovine serum albumin (BSA, Gemini) in PBS. In the assay condition optimization, 5 μL of recombinant human ST2-Fc-Histag (R&D Systems, #523-ST) and 5 μL of biotinylated recombinant human IL33⁵⁴ at varying concentrations were added into each well first. After 1 h of incubation at room temperature, 5 μL of anti-human IgG-Tb cryptate (or anti-6His Tb cryptate, Cisbio) and 5 μL of streptavidin-d2 (Cisbio) prepared by the manufacturer’s instruction were added into each well and incubated for another 2 h at room temperature. In each well with a total volume of 20 μL, we varied final concentrations of hST2-Fc-Histag from 0.666 to 486 nM in 3-fold dilution and IL33 from 0.75 to 60.75 nM in 3-fold dilution. The well with hST2-Fc-Histag alone was used as a negative control, whereas the ST2/IL33 mixture a positive control. A multi-mode reader (BioTek Synergy Neo2) was used to measure time-resolved fluorescence values excited at λ_ex = 330 nm and emitted at λ_em = 665 and 620 nm. The HTRF signal ratio (665 nm/620 nm) was used to determine the energy transfer efficiency, and the maximum signal ratio (665 nm/620 nm), providing the highest dynamic range of the signal, was obtained when 4.5 nM of hST2-Fc-Histag and 6.75 nM of IL33 were used. These concentrations of hST2-Fc-Histag and IL33 were adopted in the competitive inhibition assay. In the competitive inhibition assay, 4 μL of hST2-Fc-Histag at 22.5 nM and 4 μL of biotinylated IL-33 at 33.75 nM were added into each well followed by the addition of 2 μL of probe molecules or un-biotinylated IL33 in a serial dilution and incubated at room temperature for 1 h. Five μL of anti-human IgG-Tb cryptate (or anti-6His Tb cryptate, Cisbio) and 5 μL of streptavidin-d2 (Cisbio) were added into each well and incubated for 2 h at room temperature. The well with hST2-Fc-Histag alone was used as a negative control, whereas the ST2/IL33 mixture was used as the positive control. The HTRF signal ratio (665 nm/620 nm) determined by the multi-mode reader was used to calculate the IC₅₀ values using the sigmoidal 4-parameter logistic model in Prism 8.0 (GraphPad Software, La Jolla, California USA). The pH measurement was performed using 1 mL of solution and the accumet AB15 plus instrument.

Reagents.

Human ST2-Fc-Histag was purchased from the R&D system (catalog#523-ST), and human IL33 and biotinylated IL33 were produced inhouse as reported previously.⁵⁴ Donor beads, acceptor beads, and the detection buffer for the HTRF assay were obtained from CicBio (cat al o g # 61 HI2TLF, 6 1H FCTAA, 610 S ADLF, 61DB10RDF). Nicotinic acid (CAS#59-67-6, catalog#AA00E-ZEC), nicotinamide (CAS#98-92-0, catalog#AA00IJXK), salicylic acid (CAS#69-72-7, catalog#AA00FTEJ), 3-(aminomethyl)pyridine (CAS#3731-52-0, catalog#AA003TZB), and compounds 1 to 7 (CAS#86-48-6, catalog#AA0032PI, CAS#345-29-9, catalog#AA0033Y8, CAS#345-16-4, catalog#AA00I6QO, CAS#1666-28-0, catalog#AA001WRS, CAS#89-55-4, catalog#AA00345T, CAS#434-45-7, catalog#AA0032TD, CAS#26944-43-4, catalog#AA00C25H) were procured from AA Blocks, and benzamidine (CAS#618-39-3, catalog#30306) was from AstaTech. The purity of these compounds are >95%.

Force Field Parameters for the Cosolvent Molecule.

We used the Antechamber module in the AMBER18 program suite⁵⁵ to derive force field parameters of the cosolvent molecules. Structures of cosolvent molecules were first optimized via the quantum chemistry (QM) calculation at Restricted Hartree Fock (RHF) level using the 6-31G**, or 6-31G(d,p), basis set in the Gaussian program.⁵⁶ Point charges of the cosolvent molecules were obtained by fitting to the QM generated electrostatic field based on the RESP fitting method.⁵⁷ Non-charge force field parameters of the cosolvent molecules were provided by the parmchk2 program. Negative charged NA and SA and positive charged BZA and AMP were used to prepare the cosolvent molecules unless stated otherwise.

Cosolvent System Preparation.

We used the concentration of the water molecule in pure liquid (55.5 M) as a reference to calculate the ratio between cosolvent and water molecules corresponding to the assigned molar concentration in each cosolvent system (see Table 1 and the notation to each system). Deprotonated SA and NA and protonated BZA and AMP forms were used throughout the study. The initial configuration of the cosolvent system was obtained by randomly placing the cosolvent molecules and/or counter ions in a box of water molecules using the software Packmol.⁵⁸ Counter ions were included if needed to neutralize the charges of the cosolvent system. The GPU implemented PMEMD⁵⁹ in AMBER18 was used to perform the MD simulations. Preparation of the equilibrated cosolvent system was done in the following steps. First, the cosolvent system was subject to a 3000-step minimization using 1000 steps of conjugated gradient followed by 2000 steps of the steepest decent. A second 100 ps of constant pressure, constant temperature (NPT) simulation was done to progressively raise the temperature from 0 to 150 K (0−2 ps) and from 150 to 298 K (2−6 ps) and set at 298 K (6−100 ps). The system was then allowed for 6 ns equilibration under the NPT condition. The final snapshot of the cosolvent system at 6 ns was used to incubate ST2 for cosolvent MD simulations. To calculate the radial distribution function (RDF) and self-diffusion coefficients (D_i), production runs at windows of 7−12, 13−18, and 19−26 ns were used. Trajectories in each segment of the production runs were used to calculate RDF and D_i listed in Table 1.

Table 1.

Self-Diffusion Coefficients (D_i) of the Probes and Water in the Cosolvent Systems^a

	7–12 ns		13–18 ns		19–26 ns
	probe	water	probe	water	probe	water	probe#	water#	mM	<Di>	diffusion length (48 ns)
NA1	2.98	5.77	3.98	5.93	1.33	5.65	1	5600	9.91	2.76	115.14
NA3	1.65	5.76	2.21	5.84	3.15	5.78	3	5700	29.21	2.34	105.95
NA5	1.36	5.66	2.26	5.57	2.29	5.52	5	2800	99.11	1.97	97.30
SA1	1.13	5.75	0.64	5.79	0.43	5.80	1	5600	9.91	0.73	59.32
SA3	0.42	5.75	2.36	5.70	0.73	5.76	3	5600	29.73	1.17	75.01
SA5	1.09	5.71	1.42	5.70	1.06	5.64	5	2800	99.11	1.19	75.56
BZA1	1.06	5.64	0.37	5.83	0.63	5.91	1	5600	9.91	0.69	57.55
BZA3	4.62	5.78	1.42	5.67	1.13	5.79	3	5700	29.21	2.39	107.16
BZA5	1.02	5.67	1.21	5.85	1.47	5.49	5	2800	99.11	1.23	76.94
NAM3	1.33	5.85	3.21	5.86	1.45	5.70	3	5700	29.21	2.00	97.93
NAM5	2.23	5.68	1.07	5.79	2.69	5.70	5	2800	99.11	2.00	97.91
AMP3	0.92	5.84	2.02	5.74	2.77	5.84	3	5600	29.73	1.90	95.60
AMP5	2.10	5.55	2.11	5.59	2.00	5.68	5	2800	99.11	2.07	99.63

^aThe numbers (1,3,5) following the probe name refer to different probe concentrations from 10 and 30 to 100 mM. D_i values of the probes at different cosolvent systems are calculated from 8 ns of MD simulations. Numbers of probes and water in each cosolvent systems are provided. The free diffusion length of the probes expanding over 48 ns is calculated based on the average D_i (<D_i>). The units for D_i and the diffusion length are 10⁻⁹ m²/s and Å, respectively.

Cosolvent MD Simulations and Mapping Analysis.

After the equilibrated cosolvent systems were prepared, we followed the same procedures of conducting cosolvent MD simulations and the cosolvent mapping analysis reported previously.^37,60,61 Briefly, ST2 (PDB ID: 4KC3⁴⁷) and LilrB2 (PDB ID: 6BCS⁶²) were extracted from the crystal structures. Three N-acetylglucosamine modifications on N95, N140, and N151 in ST2 were not in the ST2/IL33 interaction interface and were removed.⁵⁴ No post-translational modification on LilrB2 was found. Both ST2 and LilrB2 were neutralized with counter ions and embedded in each cosolvent system for MD simulations. The numbers of probe and water molecules in each system were listed in Table 2. Cosolvent MD simulations were performed to determine the persistent retention of the probes on ST2 and LilrB2 throughout the entire 48 ns. A grid-based counting method was used to calculate the observed occupancy (N_p) of the probes at the grid point in space and compared with the expected occupancy (N₀) in the cosolvent system alone. A higher N_p/N₀ value indicated preference of the probes at the location and was attributed to the binding interaction between the probes and proteins. In the cosolvent mapping analysis, N₀ can be used as an adjustable parameter to progressively reveal preferential locations of the probes at the protein binding site and these locations were displayed as envelop functions.⁶³ Here, we used N₀ = 0.00432 for all carbon atoms of the probes in the cosolvent mapping analysis.³⁷

Table 2.

Number of Probe and Water Molecules Used in the Cosolvent MD Simulations (BZZ and BZY Refer to Two Different Neutral Forms of BZA)

	probe	water
NA1	4	32,257
NA3	14	31,639
NA5	46	28,885
SA1	4	31,674
SA3	15	31,517
SA5	61	43,650
BZA1	2	31,612
BZA3	14	32,278
BZA5	48	31,369
NAM3	15	29,707
NAM5	43	28,685
AMP3	17	31,638
AMP5	49	31,247
BZZ3	13	31,300
BZY3	13	31,189

RESULTS

Development of the ST2/IL33 Binding Assay Using the HTRF Technology and the Inhibition of ST2/IL33 Binding by the Probes.

To determine the ST2/IL33 binding, we first developed a biochemical binding assay using the HTRF technology. In this assay, donor and acceptor beads are used to bind to tagged ST2 and biotinylated IL33 respectively (Figure 1A). Association between ST2 and IL33 will bring the donor and the acceptor beads in close proximity and energy transfer from the excited donor beads to the acceptor beads will lead to emission of the acceptors beads signaling the binding event.⁵⁰ We first used anti-Histag (histidine-tag) donor beads for ST2-Fc-Histag and the streptavidin coupled acceptor beads for IL33 in our HTRF assay development. We varied the ST2-Fc-Histag and IL33 concentrations to determine the optimal energy transfer signal in the HTRF assay. Using the optimized condition, we determined that the IC₅₀ value of unlabeled IL33 to ST2-Fc-Histag/IL33 binding was 2.25 ± 0.98 nM (2.94 and 1.5 nM, see Figure 1B). The IC₅₀ value was comparable to the K_D value (2.23 nM) we obtained previously based on the biolayer interferometry (BLI) experiment⁵⁴ and other reports.^27,48 We then determined the inhibition of ST2/IL33 binding by five probes (see Figure 2A) and found that salicylic acid had the lowest IC₅₀ value at 2.47 mM followed by nicotinic acid (IC₅₀ = 5.89 mM, see Figure 2B, and the percentage of inhibition is shown in Figure S1). In comparison, benzamidine containing a positive charge group (pKa = 11.6⁶⁴) had an IC₅₀ value of 8.75 mM, whereas 3-(aminomethyl)pyridine, also carrying a positive charge (pKa = 8.34), gave no inhibition activity. The neutral analog of nicotinic acid, i.e., nicotinamide, showed some inhibitory activity only at 30 mM (the highest concentration used in the assay).

Figure 1.

IC₅₀ values of IL33 against ST2/IL33 binding based on the HTRF assay. (A) Principle of the HTRF assay. Titration of unlabeled IL33 to the ST2/IL33 binding using (B) anti-Histag and (C) anti-Fc donor beads and the calculated IC₅₀ values are displayed. The 665/620 ratios at 0.01 and 100 mM corresponding to the negative and positive controls were assigned to aid the curve fitting. Results were based on two independent measurements colored in black and red. Each experiment is performed in duplicate.

Figure 2.

Inhibition of ST2/IL33 binding interaction by the cosolvent probe molecules. (A) Chemical structures of five probe molecules. (B) Similar to Figure 1 except that titration of five probes were used. The 665/620 ratios at 0.01 and 100 mM corresponding to the negative and positive controls were assigned to aid the curve fitting. Each experiment is performed in duplicate.

Because binding of the anti-Histag to the positively charged polyhistidine-tag (i.e., Histag) may be interfered by positively charged probe molecules at high concentrations, we decided to develop a second complementary assay using donor beads that bind to Fc in ST2-Fc-Histag. When the Fc donor beads were used, we found that the IC₅₀ value of ST2-Fc-Histag/IL33 was 3.70 ± 0.29 nM (3.49 and 3.90 nM, Figure 1C), similar to the values determined using the anti-Histag donor beads. Using the Fc donor beads, we obtained similar IC₅₀ values for SA and BZA and the IC₅₀ value of NA decreased by 2.8-fold (Figure 2B). Both NAM and AMP were found inactive up to 30 mM. After we determined the activities of the probe molecules, we conducted cosolvent mapping analysis to corroborate the experimental data.

Hydration Structures and Mobility of Probes in the Cosolvent Systems.

To examine the force field parameters of the probes, we studied the probes in the equilibrium cosolvent systems. In the cosolvent systems, we used probe concentrations much higher than their IC₅₀ values to increase the number of available probes in the simulation box to interact with ST2 in the subsequent cosolvent mapping analysis. A higher probe concentration can also allow the detection of the binding site location where the probe interacts with ST2 at a lower affinity. The probe concentrations in the cosolvent MD simulations, however, do not necessarily recapitulate the experimental equilibrium binding kinetics to ST2 that requires a longer simulation time. For NA, SA, and BZA, 10, 30, and 100 mM were prepared, whereas only 30 and 100 mM of NAM and AMP were included because of their inactivity, respectively.

First, we analyzed the hydrogen bond strengths of the probes by studying the radial distribution functions (RDFs) between the polar atoms in each probe and the oxygen atom of water (Ow). We found that the RDFs between the polar atoms in each probe and Ow were not affected by the concentrations of the probes (see Figure S2), suggesting that the systems were dilute solutions. As an example, we focused on discussing RDFs of each probe at 30 mM in Figure 3. We started by analyzing the hydration of oxygen atoms in the probes. Generally, the acidic carboxylic oxygen atoms in NA (O8, O9) and SA (O8, O9) have shorter distances with Ow (measured by the first peak in Figure 3) than bulk water, indicating expected stronger hydrogen bonds compared to those between two water molecules. Despite a stronger hydrogen bond, the number of neighboring water molecules surrounding the acidic oxygen atoms was similar to that in the bulk water (Figure 3). When compared to the carboxylic oxygen, the carbonyl oxygen of NAM (O7) formed a hydrogen bond with a fewer number of water molecules in the first solvation shell. The structureless plateau level of g(R) starting at 3 Å between SA (O10) and Ow indicated no specific hydrogen bonds were formed (Figure 3). As for the nitrogen atoms, a higher number of water molecules surrounding the charged nitrogen atoms in BZA (N8, N9) and AMP (N6) in the first solvation shell were identified; however, the peaks were at longer distances than the Ow-Ow distance or the acidic oxygen atoms with Ow (Figure 3). For neutral nitrogen atoms, the probability of forming structural hydrogen bonds with water molecules decreased for NA (N1) and NAM (N1, N6). The height of the first peak in N1(AMP)-Ow was lower than the height of bulk Ow-Ow, which suggested N1(AMP) is a poorer hydrogen bond acceptor than water. The above analyses characterized the capability of the polar atoms in the probes to form hydrogen bonds with the amino acids in ST2 in the following study.

Figure 3.

Radial distribution functions between the probes and water in the 30 mM cosolvent systems. Atoms of the cosolvent probes are labeled according to Figure 2A. Radial distribution functions between the polar heavy atoms on each probe and oxygen (Ow) of the water are calculated from 8 ns of MD simulations. Each cosolvent system contains three cosolvent molecules and are annotated as −1, −2, and −3. As an example in NA, the probes are labeled as NA3−1, NA3−2, and NA3−3. The radial distribution function between oxygen atoms in water (Ow-Ow) is included as a reference.

To ensure that probes will reach the protein binding sites during the timescale of the cosolvent MD simulations, we calculated the self-diffusion coefficients (D_i) of probes in the cosolvent systems. Using the water molecules as a reference, we found that the D_i of the water molecule (D_H2O) varies between 5.55 and 5.91 × 10⁻⁹ m²/s in all five cosolvent systems with probes at 10, 30, and 100 mM. The diffusion rates of water in our simulations were comparable to previously reported data using the TIP3P water model.^65,66 Among the probes, D_NA decreased by 28% when the concentration increased from 10 to 100 mM (Table 1). D_BZA increased by ∼2-fold when the concentration increased. For other probes, D_SA, D_NAM, and D_AMP remained similar irrespective of the probe concentrations. Overall, the probe molecules had similar diffusion rates (0.69−2.76 × 10⁻⁹ m²/s) at three concentrations. Based on the averaged self-diffusion coefficients, each probe molecule was expected to have free diffusion paths between 58 and 115 Å in 48 ns of simulations. Because the radius of gyration of ST2 typically varied from 33 to 21 Å in the 48 ns of simulations, most probes were expected to make contact with ST2 during the simulations.

Cosolvent Mapping Analysis of ST2.

We have previously reported that ST2^D1D2 and IL33 interaction accounted for 83% of the total binding free energy between ST2 and IL33 using biolayer interferogram (BLI).⁵⁴ Although the fragment-like probes may bind to multiple locations in ST2, we focused our cosolvent mapping analysis on the binding interface between ST2^D1D2 and IL33 to study the primary mode of inhibition. Binding of probes at the interface will directly impact on the ST2/IL33 binding. Using NA as the probe, we recurringly identified binding sites S1 and S2 on ST2^D1D2 (see Figure 4) at three different concentrations (10−100 mM). A larger shape at S1 would suggest that more than one NA may interact with ST2 at S1. Similarly, preferential bindings of SA at 10, 30, and 100 mM and NAM at 100 mM to S1 were also found (envelop shapes shown in Figure 4 and Figure S3E). At 100 mM of NAM, another S3 site emerged (Figure 4). The consensus binding site locations detected by NA and SA indicated that their inhibitory activities may be attributed to their binding to sites S1 and S2.

Figure 4.

Cosolvent mapping analyses of ST2 using different cosolvent concentrations. Envelop shapes indicating the long resident times of probes in ST2 are shown by mapping to the crystal structure of ST2 as a reference (deepteal). ST2^D1D2 conformation remains close to the crystal structure throughout the simulations; thus, the crystal structure of ST2^D1D2 is an ideal reference for comparison. The purple, green, and oranges shapes correspond to maps derived from 10, 30, and 100 mM of cosolvent molecules in the simulations. IL33 (gray) is included for comparison, and the positive and negative charged residues are colored in blue and red, respectively. Major binding sites in ST2 (S1−S3) are circled. The yellow arc denotes the dominant interaction interface between ST2 and IL33 that accounted for 83% of total binding free energy.

No prolonged NAM binding at ST2^D1D2 was found at 30 mM (Figure 4), but bindings of NAM at S1, S2, and S3 at the interface were found at 100 mM that partly explained the observed residual inhibition activity at 30 mM in Figure 2B. Our cosolvent mapping analysis, however, did not detect retention of BZA and AMP at the ST2^D1D2/IL33 interaction interface. To examine if ST2 was trapped in distinctive conformations to preclude the binding of BZA and AMP to ST2^D1D2 in the simulations, we found no indication after analyzing ST2 conformations in the cosolvent simulations (data not shown and the final snapshots of ST2 conformations are provided in Figure S4). The IL33 binding interface at ST2^D1D2 appeared to preferentially attract negatively charged NA and SA instead of positively charged probes based on our cosolvent mapping analysis. Further, the inhibitory activity of BZA to ST2/IL33 binding cannot be explained by the cosolvent mapping analysis.

BZA Binding Sites in Li1rb2 Were Identified by the Cosolvent Mapping Analysis.

One of the reasons that we did not identify preferential BZA binding sites in ST2 can be caused by the deficiency of the cosolvent mapping method using the positively charged probes such as BZA. Another possible reason may be attributed to the interference of the HTRF signals by the buffer condition because a high BZA concentration may change the pH value (to be discussed later). To address the first issue, we found a co-crystal structure of leukocyte immunoglobulin-like receptor subfamily B member 2 (LilrB2) with BZA (PDBID: 6BCS).⁶² LilrB2 is a receptor of class I MHC antigens and contains four Ig-like domains homologous to the D1-D3 domains in ST2. In the crystal structure, four BZA molecules in the D1-D2 of LilrB2 (LilrB2^D1D2) were resolved. We removed BZA from the crystal structure and embedded LilrB2^D1D2 in 100 mM of BZA to perform the cosolvent MD simulation. Based on the cosolvent mapping analysis of 48 ns simulations, we identified two BZA binding sites in LilrB2^D1D2 (meshed shapes in Figure 5A). In the larger binding site between LilrB2^D1 and LilrB2^D2, the meshed shape from mapping (represented by BZA709 or BZA and molecule number 709 in the simulation) colocalized with a BZA molecule in the crystal structure. A second smaller binding site (a meshed shape nearby BZA218) was found at the concave interface between LilrB2^D1 and LilrB2^D2 that was close to another BZA in the crystal structure. After examining the final snapshots, we found that BZA218 bound to the same location as a BZA molecule in the crystal structure and interacted with D13 of LilrB2^D1D2. By tracking the trajectory of BZA218, we found that BZA218 reached the binding site to interact with D13 after 34 ns of simulations (Figure 5B). In comparison, BZA709 bound to the first binding site and interacted with D178 after 5 ns of simulations (Figure 5B). The initial distance between BZA709 and D178 was 9.18 Å; thus, recognition of BZA709 to the binding site was driven by the unbiased cosolvent MD simulation. Results of our cosolvent mapping analysis of LilrB2^D1D2 suggested that the method should be applicable to identify BZA binding sites in ST2 if they existed.

Figure 5.

Mapping the BZA binding sites in LilrB2 using the cosolvent mapping method. (A) Alignment of the final snapshot of LilrB2 in 100 mM BZA simulations (green color) with the crystal structure of LilrB2 with BZA (PDB ID: 6BCS, gray color). BZA in the crystal structure and the simulation is colored in yellow and green, respectively. Residues interacting with BZA are labeled. The mesh shapes in cyan color indicate the locations that BZA resides for an extended period of time during the 48 ns of simulations. Oxygen and nitrogen atoms are colored in red and blue, and the hydrogen bonds are depicted in red dashed lines. (B) Distances between Cδ of D13 and C7 of BZA218 (black), Cδ of D178, and C7 of BZA709 (red) during the 48 ns of simulations are shown.

pH-Dependent Signal Changes of the ST2/IL33 Binding in the HTRF Assay.

We remained puzzled why no persistent BZA binding to the ST2 and IL33 interaction interface was identified from our mapping analysis despite the observed inhibition activity of BZA (IC₅₀ ≈ 8 mM). Because the inhibitory activities of the probes started at relatively high concentration from 1 to 30 mM, we asked if the probes at high concentrations may change the pH values of the buffer and influence the IC₅₀ value measurement. Based on reported data, the pKa of the five probes varies from 2.9 (salicylic acid)^67,68 to 11.6 (benzamidine)⁶⁴ and the anti-Histag donor beads bind to the poly-histidine group in ST2-Fc-Histag. We first tested the pH values of 30 mM salicylic acid and benzamidine in the buffer, and found that the pH values were 4.1 and 10.5 respectively (pH = 4.1/SA, 5.1/NA, 7.2/NAM, 9.4/AMP, 10.5/BZA). We then determined the signals of the fluorescent energy transfer between ST2-Fc-Histag and three biotinylated IL33 concentrations (2.25, 6.75, and 20.25 nM) in two experiments at pH = 6−9 and 4−11 (Figure 6A,B). Using 6.75 nM of biotinylated IL33 (used in our optimized HTRF assay), we found that the signals were unaltered at pH = 6−9 (Figure 6A). When using biotinylated IL33 at 2.25 and 20.25 nM, up to 28 and 18% decreases of signals were found at pH = 6 and 9, respectively, but not a complete loss of signals. Decrease of the pH value further to 4 did not affect the signals using 6.75 nM of biotinylated IL33 (Figure 6B). When the pH value was raised to 10 and 11, the signal decreased markedly by 50% at 6.75 nM of biotinylated IL33 (Figure 6B). In the buffer with BZA, the pH value started at 7.5 (0.12 mM) and increased to 10.1 (3.3 mM) and reached 10.8 (30 mM, see Figure 6C). Decreased HTRF signals, caused by high pH values that affected the protonation states of the Histag in ST2-Fc-Histag, can be expected at >3.3 mM of BZA. Based on these data, inhibition of BZA to ST2/IL33 binding can be partly attributed to the decreased signals using the anti-Histag donors at high BZA concentrations that gave high pH values.

Figure 6.

Dependence of the energy transfer signals on the pH values in the ST2/IL33 HTRF assay. (A, B) Two independent experiments and each in duplicate are performed at 2.25, 6.75, and 20.25 nM of biotinylated IL33 concentrations and varying pH values. The anti-Histag donor beads are used in both experiments. For the data shown in Figures 1 and 2, 6.75 nM of biotinylated IL33 is used in the assay. (C) Dependence of pH values on the concentrations of BZA. (D) Cosolvent mapping analysis on ST2 using the neutral form of BZA at 30 mM. Two proton positions on N9 of BZA are used in two separate 48 ns cosolvent MD simulations.

The effect of pH on the HTRF signals, however, cannot explain the inhibitory activity of BZA using the anti-Fc donor beads that bind Fc in ST2-Fc-Histag. At pH = 10, the neutral form of BZA is predicted to be at 99.37% in solution calculated by the Chemicalize program (version 19.9, 2019, https://chemicalize.com developed by ChemAxon, http://www.chemaxon.com). We used two neutral forms (different proton orientation on N9) of BZA (Figure 6D) at 30 mM to perform the cosolvent mapping analysis. Binding sites of BZA at S1 and S2 on ST2^D1D2 were subsequently identified in both forms of neutral BZA (Figure 6D). In comparison, the neutral form of SA accounts for 5.8% at pH = 4. Therefore, the acidic form of SA is the primary species to inhibit ST2/IL33 binding giving the IC₅₀ value of 1.77 mM. Taken together, a high concentration of BZA may generate neutral BZA to bind to ST2^D1D2 and inhibit the ST2/IL33 binding.

Shared Binding Modes of NA and SA to ST2^D1D2.

Identification of similar binding sites by NA and SA in our cosolvent mapping analysis led us to ask if both probes bind to ST2^D1D2 similarly. We first analyzed the final snapshots of ST2 incubated with NA and SA at different concentrations and found one NA or SA bound to S1 in ST2^D1D2 in each case (Figure 7B−F, Figure S3D). Both NA and SA used a carboxylic acid group to interact with K22 of ST2^D1D2 and an aromatic ring to interact with R35 via a cation−pi interaction. This mode of interaction is shared by NA and SA (Figure 7G, Figure S3D). At 100 mM of SA, we further found a second SA molecule (SA5-9123 in Figure 7F) near the S1 site and formed a salt−bridge interaction with R198. A second NA (NA5-18020 in Figure 7D) was also found to interact with R198 similarly to SA5-9123 at 100 mM of NA concentration. Although the crystal structure of ST2 with IL33 showed that three acidic residues (E148, D149, and D244) in IL33 interact with ST2 at a similar location (cf, Figure 7A,H), the binding modes of NA and SA to ST2 clearly indicated that a small binding site was induced by the rearrangement of K22 and R35 in ST2 (Figure 7I).

Figure 7.

Binding modes of NA and SA to ST2 obtained from the final snapshot of ST2 from the cosolvent MD simulations. (A) Interaction between ST2 (gray) and IL33 (orange) at the shared binding location of NA and SA to ST2 is shown. (B−F) Final snapshots of NA and SA bound to the same location of ST2 in different cosolvent concentrations are displayed. The numbers in parentheses are the molecule indexes in each simulation. (G) Alignment of all probes (colored in yellow) shown in (B−F). (H, I) Comparison between the ST2/IL33 and ST2/SA interactions. A surface representation of ST2 is used to highlight the induced binding site in ST2/SA interaction absent from that in ST2/IL33. Oxygen and nitrogen atoms are colored in red and blue, and the hydrogen bonds are depicted in cyan dashed lines. In (I), blue and red colored surfaces correspond to positive and negative charged regions. All snapshots were from the final conformation except NA5-18020, which was taken at 36 ns.

To study the time course of NA and SA recognition to ST2, we monitored the trajectories of the NA and SA molecules that interacted with ST2 identified in the final snapshots (Figure 7, Figure S3C). Figure 8 showed that the binding interaction between NA and SA with K22 preceded their interactions with R35. Figure 8 also revealed that the initial encounter of K22 by the acidic group in NA and SA can lead to strong and persistent interaction (distance <3.5 Å) for a period of time, such as those in SA5 and NA5. At 30 mM of SA (or SA3), we found that SA3-13796 was not initially placed near R35 (Figure S3A) but was in the proximity of K22 and R35 after equilibration (Figure S3B). SA3-13796 drifted away at around 18 ns and returned to interact with K22 and R35 after around 22 ns of simulation and remained at the location throughout the simulation (Figure S3F). Recognition of K22 and R35 in ST2 by NA and SA, however, was not biased because the probes were not placed close to the binding site at the beginning of the simulations. In the simulation of ST2 with SA at 100 mM, we found that the second SA molecule (SA5-9123, Figure 7F) bound to R198 after SA5-438 docked at S1 (Figure 8). Although SA5-438 and SA5−9123 were close to each other, we did not find intermolecular interaction between them. Thus, SA5-9123 was not recruited to interact with R198 through an interaction with SA5-438. Unlike SA5-9123, the second NA5-18020 formed a salt−bridge interaction with R198 for a shorter period of time between 27 and 37 ns of simulations (Figure 8).

Figure 8.

Trajectories of NA and SA binding to site 1 on ST2 in the cosolvent MD simulations. Migration of NA and SA to site 1 on ST2 is monitored by the distances between C7 of NA or SA and the guanidinium carbon atom (Cζ) in R35 and R198 or the terminal amine atom (Nζ) in K22. The numbers in parentheses correspond to the molecule indexes in each simulation. SA1, SA5, NA1, NA3, and NA5 denote 10 and 100 mM of SA and 10, 30, and 100 mM of NA in the cosolvent systems, respectively.

SA Analogs Inhibited the ST2/IL33 Binding in the HTRF Assay.

The structural insights of the shared binding modes of NA and SA to S1 indicated that the aromatic ring opposite to the carboxylic acid is exposed to the solvent (Figure 7G). We then asked if substituted SA analogs may retain similar or improved activities to ST2/IL33 binding. We purchased seven SA analogs and performed the initial screening using two concentrations at 1 and 10 mM (Figure 9A). Four (2, 3, 4, and 5) of the seven compounds (Figure 9B) exhibited inhibition at two concentrations (Figure 9A). We further determined the IC₅₀ values of the four compounds. Based on the data (Figure 9C), bromination or fluorination to SA at the 3-(3, 5) or 4-position (2, 4) gave comparable or 2-fold higher inhibition activities than SA (IC₅₀ = 1.77 ± 0.51 mM), suggesting that the halogen atoms may not be in an enclosed pocket to exert strong interaction. In comparison, replacing the carboxylic group by a neutral isostere, the triflouroacetyl group, (6 and 7) abolished the activities (Figure 9B). Compound 1 at 10 mM exhibited poor solubility, and the IC₅₀ value of 1 cannot be determined accurately. The structure−activity relationship of the SA analogs is consistent with the binding mode of SA from our analysis and can be useful for inhibitor development in the future.

Figure 9.

Screening and IC₅₀ values of salicylic acid analogs determined by the ST2/IL33 HTRF assay. (A) Screening of seven SA analogs at 1 and 10 mM. The data were from duplicate experiments. (B) Chemical structures of the SA analogs. IC₅₀ values in mM were provided in parentheses. Representative titration curves of four active compounds. The IC₅₀ values in mM were calculated using GraphPad Prism software and from duplicate experiments.

DISCUSSION AND CONCLUSIONS

Cytokine signaling has played important roles in lymphocyte homeostasis, immune response to infection, and tissue repairs. Dysregulated acute or chronic production of cytokines can lead to pathological conditions and diseases. Although antibodies have been shown effective to target either cytokine receptors or cytokines and have been approved for clinical uses, known small-molecule inhibitors targeting cytokine receptors were rare. The progress is hampered by the challenge to identify binding sites in the extensive and shallow interface suitable for small-molecule binding to interrupt the cytokine receptors and cytokine interaction.

In this work, we performed a computational cosolvent mapping analysis³⁷ by using bioactive aromatic molecules to probe the binding interface between IL33 and ST2, a member of the IL1R family. To support our computational analysis, we developed a biochemical HTRF assay that measured the binding of ST2 with IL33 to determine the activity of probe molecules. Five probes, including, NA, SA, NAM, BZA, and AMP, were first selected in this study. In the HTRF assay measurement, we found that SA, BZA, and NA inhibited the ST2/IL33 binding at IC₅₀ values of ∼1.77, 8, and 11 mM, whereas NAM and AMP had no activity up to 30 mM. The millimolar IC₅₀ values are consistent with the expectation that low complexity fragment molecules are weak inhibitors but possess greater potential for development.^49,69 Our assay data confirmed our hypothesis that fragment molecules mimicking acidic residues may interrupt ST2/IL33 binding and were concordant with the binding data between ST2 and mutant IL33 reported previously.^47,48

Previous studies indicated that several acidic residues, when mutated individually, contributed importantly to ST2/IL33 binding.^47,48 The implication is that the probe molecules may bind to multiple locations in ST2 that interacted with different acidic residues in IL33, though with different affinities. Our cosolvent mapping analysis elucidated that acidic probes (NA and SA) bound to locations in ST2^D1D2 that interacted with E148, D149, and D244 in IL33 at three different cosolvent concentrations (10, 30, and 100 mM). In contrast, the basic probes (BZA and AMP) did not bind persistently to the ST2/IL33 interaction interface and NAM (an analog of NA) was clustered to some sites in ST2^D1D2 only at 100 mM.

Because BZA had an IC₅₀ value of 8 mM to ST2/IL33 binding in our HTRF assay, we asked if high millimolar probe concentrations in the assay may change the pH values in the buffer to influence the signals associated with the ST2/IL33 binding events in the HTRF assay. We found that the fluorescent energy transfer (FRET) signals between ST2 and IL33 remained robust without substantial decreases at low to moderate pH values. However, the signals in the HTRF assay reduced up to 50% at pH = 10−11. We attributed the reduced FRET signals at high pH values to partly contributed to the sigmoidal curves of BZA (pKa = 11.6) used in the IC₅₀ value calculations. To compare with BZA, the pKa of AMP is 8.34 and no inhibition activity of AMP was found up to 30 mM. Other undetermined factors may still contribute to the inhibitory activity of BZA in our assay including the potential protein aggregation at high pH mediated by the Fc domain⁷⁰ of ST2-Fc-Histag. When the neutral form of BZA was used in the cosolvent mapping analysis, binding events of BZA at S1/S2 and other regions emerged. Whether neutral BZA inhibits ST2/IL33 binding at the IC₅₀ = 8 mM remains to be determined.

To provide structural insight of binding of NA and SA to ST2, we examined the data from MD simulations and identified a shared binding mode of NA and SA at site S1 on ST2. Our analysis elucidated that the formation of a salt−bridge interaction between the carboxylic group on NA and SA and K22 in ST2 preceded an induced cation−pi interaction between R35 in ST2 and aromatic rings on NA and SA. The small binding pocket formed via the binding of NA and SA to ST2 differed markedly from the mode of interaction between ST2 and IL33 at the same location. Although NA and SA may bind to multiple sites on ST2 including S1 and S2 to exert inhibition at high concentrations, we cannot rule out that SA or NA may possess allosteric inhibition activities.^61,71 The structural insight further led us to select seven compounds analogous to SA to confirm our findings. Four of the seven showed either comparable or higher inhibitory activities than SA that validated our analysis. Our study provided a starting point for ST2 inhibitor development using SA as a template.

The in silico virtual screening (VS) method can be another powerful way to identify hit compounds^73,74 from billions of compounds in a chemical library including a recent study that targeted cannabinoid receptors (CB) and Rho associated coiled-coil containing protein kinase 1 (ROCK1).⁷⁴ However, VS methods use a static protein structure for hit identification and subsequent MD simulations are frequently needed to validate and refine the un-optimized protein−ligand interaction such as the protocol used by Mai et al.’s study.⁷² The cosolvent-based simulations^{37,38,40–42,45} have the advantage of accounting for the dynamic interaction between the protein and the ligand and allowing for reshaping the binding site induced by the ligands. Although the cosolvent-based simulations are computationally more expensive than the VS methods for screening the same number of compounds, both approaches can be effective in different applications. While the VS methods are efficient to screen complex compounds in a large library against a known binding site, the cosolvent-based methods aimed to use a smaller set of probe molecules with different physical properties to identify novel unknown binding sites and the probe molecules have greater potential for extensive chemical modifications than the complex molecules identified from the VS methods. Therefore, both methods can complement each other and build on the strengths of each other to achieve the goal of discovering hit compounds against important protein targets.

As an example in the VS approach, Mai et al.⁷² recently constructed a pharmacophore model based on several key amino acids on IL33 that interact with ST2. Using the pharmacophore model and the virtual screening approach, they identified ZINC59514725 and suggested that ZINC59514725 may be more potent than iST2-1³⁶ through computational calculations. Interestingly, ZINC5914725 contained two benzoic acid groups and was predicted to interact with basic residues including K22 in ST2.We have explored the possibility of acquiring ZINC59514725 in order to confirm the activity of the compound in our assay but was unsuccessful in acquiring it. Therefore, the inhibitory activity of ZINC59514725 to ST2/IL33 binding remains to be determined.

Despite the challenges of targeting PPI between cytokine and cytokine receptors using small molecules, increasing reports suggested the feasibility of the approach including the discovery of small-molecules binding to IL1β⁷⁵ and IL18³⁵ and our ST2 inhibitors.³⁶ In the most recent example, binding of NSC80734 to IL18 was shown to block the binding of IL18 with IL18BP and inhibited the IL-18 induced secretion of IFN-γ in KG-1 cells.³⁵ Our finding in this work suggested an extrinsic inhibitory activity of ST2 by SA at a high concentration (∼2 mM). SA and aspirin is known to inhibit COX-1, modulate COX2 activity, and exhibit anti-inflammatory action.^51,76 Cancer stem cells (CSC) treated with aspirin reduced their chemotherapy-resistance acquisition that was attributed to the decreased translocation of NF-κB in CSC.⁷⁷ Although SA inhibits ST2/IL33 binding at high concentrations, the significance of SA action on ST2 to impact ST2/IL33 signaling in physiological conditions remains to be studied.

In conclusion, we have demonstrated the use of fragment-like bioactive probe molecules to identify potential small-molecule binding sites on ST2 using computational simulations, analyses, and a biochemical assay. The inhibitory activities of the probes to ST2/IL33 binding determined from the HTRF assay support our computational cosolvent mapping analysis. Characterization of the binding modes of the probes to ST2 further provides the opportunity to conduct follow-up chemical modifications based on the bioactive probe molecules. The potential of developing probes into more potent ST2 inhibitors and the applicability of the approach to other targets engaged in protein−protein interaction will be reported in the future.