Optimization of galectin-3 binding agents by in situ multiple compound synthesis and native mass spectrometry

Abstract

Developing high-affinity drug lead critically depends on the efficient structural optimization of a lead compound. However, the dynamic nature of the process by which drugs bind to target molecules often complicates the rational design of effective substituents, even if their three-dimensional structure is available. Herein, we report a streamlined strategy for the optimization of a galectin-3 binder through the integration of in situ chemistry and native mass spectrometry. This method involves the simultaneous synthesis of multiple analogues in a single reaction vessel, followed by the direct identification of high-affinity derivatives using native mass spectrometry. This approach enables the detection of ligand–protein complexes under native conditions. Through the further structural modification of the identified hit compound, we finally identified high-affinity galectin-3 binders. The results demonstrated that our strategy would reduce the time and effort required for the structural optimization stage of a lead compound.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-38570-z.

Introduction

Structural optimization of a lead compound is a key step in obtaining highly active drug candidates. During this process, substituents that form preferred contacts with the target molecule should be introduced to the lead structure to improve their properties, including affinity and selectivity. However, this optimization process often requires a time-consuming trial-and-error approach. One reason for this inefficiency is the dynamic and complex nature of the binding process of a drug molecule to its target molecule, which involves factors such as conformational flexibility, effects of the solvation, and induced fit of the target molecules. Due to these fundamental challenges, even when the three-dimensional structure of the target molecule is known, ligand design by medicinal chemists does not always succeed. Furthermore, computer-assisted, structure-based approaches often fail to capture such dynamic behavior, which limits the accuracy of drug design.

To address this issue, target-guided synthesis (TGS), including dynamic combinatorial chemistry (DCC)^1–4 and kinetic target-guided synthesis (KTGS)^5–7, has been developed as a promising strategy, allowing the target molecule to guide selection of its ideal binding partners. (Fig. 1a). However, these TGS strategies rely on sophisticated analytical techniques to identify high-affinity ligands from a mixture of candidate compounds, which imposes limitations on their broader applicability. In addition to these general limitations, both DCC and KTGS have specific requirements related to the inherent characteristics of each method. DCC amplifies high-affinity compounds by shifting the equilibrium. However, in the case of the most potent product having low thermodynamic stability, which results in a low concentration of the potent compound at equilibrium, detecting the potent compound becomes more difficult^8,9. KTGS utilizes the acceleration of chemical reactions between multiple fragments within the binding pocket of the target molecule. However, direct affinity evaluations are necessary to confirm the activity of the resulting products, as reaction acceleration alone indicates the stabilization of the transition state of the reaction between the fragments. In 2012, Sindelar and Wanner reported a pioneering strategy to overcome these problems. In their report, mass spectrometry (MS) binding assay of compound mixture prepared by DCC strategy, which allowed the seamless synthesis of compounds and hit identification, was used to explore a GABA transporter 1 (GAT-1) ligand (Fig. 1b).¹⁰ By using MS binding assay, one of the compound groups which contains the most potent binder could be efficiently identified. However, the identification of a mixture containing highly active compounds requires a “marker” compound with known affinity, and the information can only be obtained indirectly from the decrease in the binding ratio of the “marker” compounds. Moreover, it is not possible to identify a highly active compound in the mixtures based on that information alone. Therefore, a deconvolution step was still necessary to identify the potent compound within that group, indicating room for improvement. This background prompted us to develop a method that enables the simultaneous synthesis of compounds and the subsequent detection of high-affinity compounds within a mixture of the synthesized compounds. Namely, direct detection of ligand–protein complexes would enable the identification of the highly active compounds even in the absence of reference marker compounds. Furthermore, we anticipated that integrating such detection techniques with the synthesis of compound mixtures would allow for a seamless workflow that unifies synthesis and screening, facilitating the efficient discovery of highly active compounds.

Fig. 1.

Research concept presented in this study.

Previously, we developed a structural optimization process of complex natural products utilizing in situ chemistry, where several analogues could be synthesized in nano- or micro-mole scales, and following biological evaluation of these analogues without purification processes^11,12. In this methodology, all analogues were synthesized in the individual reaction mixture composed of a pair of molecules and reagents. We envisioned that if the same in situ chemistry could be applied for the synthesis of multiple compounds in the same reaction vessel, it would be beneficial to identify a suitable combination of the introduced substituent and possible binding sites, as well as to suppress the reaction solvent and biological material used in the screening stages. Herein, we report an efficient strategy for the structural optimization of a hit compound to develop a galectin-3 binder. For this first compound screening, we utilized native mass spectrometry (nMS), which is a promising technology to investigate intact protein complexes with high sensitivity^13–15. The nMS method has been used to qualitatively and quantitatively analyze the ligand–protein binding by noncovalent interactions^16–21 and applied to the compound screening for the inhibitor development of therapeutic-target proteins^22–27. In this study, after the synthesis of multiple analogues in the same reaction vessel, nMS with affinity selection that can directly detect ligand–protein complexes successfully identified a high-affinity derivative. After careful investigation of the binding mode of the hit compound and derivatization, we obtained a potent β pocket binder, whose affinity was comparable to that of well-studied α pocket binders. These results demonstrate that this approach efficiently identifies suitable fragments without the need to determine the target pocket in advance (Fig. 1b).

Result and discussion

Initial investigation of in situ synthesis and nMS screening

Galectin-3 is a glycoprotein that binds to a β-galactoside²⁸, and regulates cell growth²⁹, angiogenesis³⁰, cell adhesion^31,32 and differentiation^33,34. Recent studies also revealed that galectin-3 plays an important role in the growth and invasion of certain cancer cells, such as hepatocellular carcinoma, lung cancer, and thyroid carcinoma^35,36. Moreover, Cai’s group reported that a galectin-3 inhibitor GB1107 (1) increases the lysis of lung cancer cells caused by anti-PD-L1 antibody via the upregulation of STAT3 phosphorylation³⁷. These studies indicate the potential of galectin-3 binders as novel anti-cancer drugs. Various galectin-3 binders have been reported to date, and most of them are derived from galactose^38–41, which is an intrinsic binder of galectin-3. Hence, our ligand design was based on the known galectin-3 inhibitor GB1107 (1) reported by Nilsson et al. (Fig. 2a).⁴² The 4-(3,4,5-trifluorophenyl)-triazole-substituted galactose moiety of GB1107 (1) is recognized by multiple hydrogen bonds formed by R144, H158, R162, and W181, and the thiophenyl substituent interacts with W181 and G182. Their structure–activity relationship study (SAR) also suggested that the substituent on the C3 position of the galactose moiety contributes to the high affinity to the sugar-binding site of galectin-3. Therefore, we planned to synthesize various analogues by attaching substituents to the anomeric carbon of the 4-(3,4,5-trifluorophenyl)-triazole-substituted galactose moiety. Since both α and β-isomer was reported as galectin-3 binders, the β-isomer was selected as the core unit. The most important challenge in this project is maintaining the reliability of the screening results while synthesizing multiple compounds as mixtures. Since the affinity for the target molecule is defined as the concentration of the compound when it binds (K_d), it is ideal that the concentration of each molecule in the mixture is the same. However, it is difficult to synthesize a mixture containing the same amount of molecules because even in the same chemical reaction, the reactivity varies depending on the chemical structure of the compounds or reagents. In the example of HTS with synthetic peptide mixtures^9,43, the equivalent for each reagent was precisely optimized to generate almost the same amount of products, but this process is laborious and severely compromises the throughput. In view of the above, we used a dynamic covalent bond for in situ derivatization. In the whole system, the composition of each product is under thermodynamic control, and by setting conditions that bias the equilibrium toward the product, the product can be obtained independently of the reactivity of the reagents. As a result, analogues were designed that can be synthesized from aldehyde core 2 (GalAld) containing a 4-(3,4,5-trifluorophenyl)-triazole moiety and hydrazides using the reversible hydrazone formation reaction.

Fig. 2.

(a) Analogue design based on GB-1107. (b) Time course of each hydrazone formation reaction in the mixture of 2, BZ20, BZ14 and BZ21.

Sindelar and Warner reported that an excess amount of hydrazine is applied to multiple aldehyde mixtures to make the equilibrium favor the product system. However, it is unclear whether this device can cancel the differences in the reactivity of the individual aldehydes. Thus, we conducted the hydrazone formation reaction of 2 and three hydrazides (BZ09, BZ14, and BZ21), which were expected to show different nucleophilicity because of the steric and electronic effect of the substituent, in DMSO as a model system for the in situ synthesis of multiple compounds. All four components were mixed in DMSO, and the ratios of the resulting products GalAldBZ09, GalAldBZ14 and GalAldBZ21 were monitored using LC–MS (Fig. 2b).^32–41 In this experiment, a slight excess amount of 2 was used to consume all hydrazides with the expectation that the consumption of all hydrazides would form the same amount of hydrazone as the initial concentration of each hydrazide. As expected, while the initial kinetics, which attribute the relative nucleophilicity of each hydrazide, varied with the substituents on the aromatic ring, after 192 h, the reaction almost reached a steady state, and the deviation of the concentration of each hydrazide decreased. Additionally, it was possible to obtain a mixture with a composition equivalent to that obtained by a long reaction by mixing each compound in DMSO and immediately concentrating the mixture under reduced pressure. This result was beneficial in terms of synthesizing the compound in a short period.

The above results confirmed that the concentration of multiple compounds could be controlled by the initial concentration of hydrazides, and almost the same amount of analogues could be synthesized in the same reaction mixture. Next, we focused on a suitable analytical method to detect the highly affinity compound in the mixture. In recent years, affinity-selection mass spectrometry (ASMS), which can separate compounds based on their m/z values, has been developed as a method for identifying compounds that show affinity for target molecules from a mixture of a large number of compounds^{16,20,21,24–27,44–46}. The current limitation of ASMS is that its analyte is only a ligand, not a protein–ligand complex, because the mass analysis was conducted under protein-denatured conditions containing a high concentration of organic solvents such as methanol. To overcome this limitation, we used nMS, which allows us to observe ligand–protein complexes directly and would be suitable for identifying high-affinity derivatives. Since nMS is conducted under native conditions, such as ammonium acetate solution, the protein enables the selection of a high-affinity compound from a compound mixture, and the resulting complex can be directly detected by nMS. Because the analytical method is based on the difference in the m/z value of each compound, the molecular weight of each compound in the mixture should be different. Therefore, hydrazides with different molecular weights (Figure S1, Tables S1–S5, BZ01, BZ02, BZ12, BZ19, BZ20, BZ23 and BZ42) were selected for the initial investigation, and a mixture of hydrazones from these hydrazides and aldehyde 2 (1.2 eq. for total hydrazides) was prepared in a single microtube (Fig. 3a). In this situation, a mixture of compounds composed of the remaining GalAld (4.2 μM) and the generated hydrazones GalAldBZXX (3.0 μM, XX = 01, 02, 12, 19, 20, 23 and 42) were obtained, assuming that all hydrazone formation reactions proceeded quantitatively. The nMS method was applied to the compound mixture after adding galectin-3 (3.0 μM). Nano ESI–MS spectra afforded several signals corresponding to the protein–ligand complexes, and the intensity of galectin-3 in the complex with some hydrazones, such as GalAldBZ01 and GalAldBZ20, was dominant in the spectra, suggesting that these hydrazones showed higher binding affinity to galectin-3.

Fig. 3.

(a) Initial experiment of nMS. NanoESI mass spectrum of galectin-3 and its complex with compounds was expanded for 7 + charged ion. (b) Flow diagram of the screening to identify hit compounds for galectin-3. Three compounds (GalAldBZ01, GalAldBZ20 and GalAldBZ41) were finally selected, and their binding affinities to galectin-3 were estimated by the ITC measurements. The numbers on the horizontal axis for MS screening and DSF screening represent the number XX in GalAldBZXX. For nMS, DSF and ITC experiments, at least three independent measurements were performed.

Synthesis and evaluation of hydrazone library

The initial investigation of sequential synthesis and nMS screening worked well as expected, and we focused on obtaining a high-affinity ligand and determining its binding site of the ligand used. Thirty-five hydrazides, BZ01-14, BZ19-21, BZ23-28, BZ30-32, BZ34, BZ35, BZ38, BZ40-44, and BZ47, were used in this stage (Fig. 3b, S1, S2). These hydrazides were divided into five groups to distinguish each compound in the crude mixture of hydrazone formation by the m/z value (to avoid overlapping spectra). That is, five individual mixtures were prepared, which were obtained by the hydrazone formation of each hydrazone mixture and GalAld. In this experiment, the concentrations of all hydrazone products and the remaining GalAld were 3 μM and 4.2 μM, respectively, assuming that all reactions proceeded quantitatively. The nMS method was applied to each crude mixture by adding galectin-3 (3.0 μM). The cut-off value for identifying the hit compounds was set to 1.26 (an average value of all compounds). As a result, 15 hydrazones were identified as the primary hit compounds in nMS, with GalAldBZ01 showing the highest intensity in the MS spectra. To further screen the hit compounds, a differential scanning fluorimetry (DSF) assay was conducted on the 15 hit compounds. GalAldBZ20 and 41 resulted in relatively large protein stabilization (ΔT_m > 1 K), suggesting that these compounds exhibit higher binding affinities among all 35 hydrazones. As observed for compounds 24 and 35, discrepancies were observed between the nMS and DSF assays, suggesting the possibility of the false positive in the nMS screening. This fact may arise from the differences in the ionization efficiency or gas-phase stability of the complexes depending on the compounds during MS analysis. In contrast, such false positive was effectively eliminated by the subsequent DSF assay. As a result of the initial filtering by nMS screening, the number of purified compounds required for synthesis for the DSF assay was reduced to 15. These results demonstrate that the present approach, in which compounds are synthesized as a mixture, worked effectively to reduce the cost associated with compound synthesis. Based on the results of the MS screening and DSF assay, we selected GalAldBZ01, 20, and 41 as hit compounds to quantitatively evaluate their binding affinities. The dissociation constants (K_d) of GalAldBZ01, 20, and 41, estimated by isothermal titration calorimetry (ITC), were 1.8, 0.4 and 1.2 μM, respectively. Importantly, GalAldBZ20, which contains a nitro group, exhibited the highest binding affinity. Considering the structural difference between the phenyl group (GalAldBZ01) and the 3-nitrophenyl group (GalAldBZ20), there should be some attractive interactions, such as a hydrogen bond between the nitro group and galectin-3.

Interaction analysis of the hit compound

The next challenge is to reveal how GalAldBZ20 interacts with galectin-3 to show a high affinity. Focusing on the interactions between galectin-3 and known galectin-3 binders, there are two potential binding pockets, α or β pockets, around the sugar-binding site. (Fig. 4a).^42,47 Notably, most of the galectin-3 binders reported to date are α-pocket binders, and we anticipated that GalAldBZ20 was also act as a α-pocket binder. To elucidate the detailed binding mode of GalAldBZ20 to galectin-3, the X-ray structure of the galectin-3 complex with GalAldBZ20 was solved at a resolution of 1.30 Å. Regardless of the high-resolution structure, an electron density map corresponding only to the 3-nitrophenyl group was not observed, probably because of the structural flexibility of the moiety or multiple binding modes (Fig. 4b). Thus, we observed interactions between GalAldBZ20 and galectin-3 using solution nuclear magnetic resonance (NMR). The ¹H-¹⁵N heteronuclear single quantum coherence (HSQC) spectrum of [¹⁵N]-galectin-3 with GalAldBZ20, the signal assignments of which were shown in Figure S3, revealed two or more signals derived from a single amide moiety for amino acid residues around the β pocket of galectin-3 (Fig. 4c, S4). These results suggest multiple binding modes of the 3-nitrophenyl group to galectin-3 and validate the unobserved electron density map in X-ray crystallography. To further investigate the interactions of GalAldBZ20 with galectin-3, molecular dynamics (MD) simulation of the complex was performed, supporting the hypothesis that the 3-nitrophenyl ring contacts the β pocket of galectin-3 (Fig. 4b). X-ray crystallography, NMR spectroscopy, and MD simulations suggest that GalAldBZ20 interacts with the β pocket, an unexpected result considering that the majority of existing galectin-3 binders are α pocket binders. To best of our knowledge, the binding affinity of the most potent β pocket (K_d = 540 nM)^42,47 was 14-fold weaker than that of GB1107 (K_d = 37 nM), an α pocket binder. Hence, we hypothesized that further structural optimization from GalAldBZ20 would achieve a binding affinity comparable to existing α-pocket binders. Namely, it was expected that the conversion of the hydrazone linkage of GalAldBZ20 would generate additional interactions with the amino acid residues of the β pocket, strengthening the binding affinity and structural rigidity, which would improve the electron density map of the refined compounds in the galectin-3 complex.

Fig. 4.

(a) Two binding modes of known galectin-3 binders. (b) Plausible binding model of GalAldBZ20 to galectin-3 based on the X-ray structural and NMR analyses (left) and MD simulation (right). 2|Fo|-|Fc| map countered at 1.6 σ for GalAldBZ20 (blue mesh) was overlayed on the cartoon model. Magenta-colored residues showing multiple signals in ¹H-¹⁵N HSQC spectrum for [¹⁵N]-galectin-3 complex with GalAldBZ20. (c) Overlay of ¹H-¹⁵N HSQC spectra for R144 and N164 of [¹⁵N]-galectin-3 in the absence (black) and presence of GalAldBZ20 (red) or compound 5 (blue).

Optimization of the linkage and evaluation of the analogues

We designed three analogues of GalAldBZ20, all of which contain 4-(3,4,5-trifluorophenyl)-triazole-substituted galactose and 3-nitrophenyl moieties, and the different linkers, which contain multiple hetero atoms that could form some polar contacts with the protein, connecting them (Fig. 5). The thermodynamic parameters of 3–5 binding to galectin-3 estimated by ITC are summarized in Table 1. Compounds 3 and 4, which possess an amide bond and an oxadiazole, respectively, showed binding affinities one order of magnitude weaker than that of the parent compound GalAldBZ20. Compound 3 was enthalpically disfavored upon binding to galectin-3, presumably because of the less polar contact between the amide linkage of 3 and galectin-3. On the other hand, 4 showed enthalpy-driven binding to galectin-3, suggesting that the oxadiazole moiety can form a hydrogen bond to galectin-3. However, the entropic disfavor of 4 could be due to the 3-nitrophenyl ring being placed in a different position from GalAldBZ20, which may result in a reduced binding affinity. This phenomenon is probably caused by the loss of hydrophilic interactions formed between the aromatic ring and the β-pocket. Fortunately, N-galactoside 5 exhibited almost tenfold higher binding affinity (K_d = 39 nM), suggesting that the alkoxyamine moiety formed some attractive interactions with galectin-3, preserving the mode of interaction of the 3-nitrophenyl ring. This binding affinity is comparable to that of GB1107, a monosaccharide compound reported by Nilsson et al.⁴² We performed X-ray crystallography to determine the binding pose of 5 to galectin-3, and solved the complex structure at 1.05 Å resolution, in which the obvious electron density of the 3-nitropheyl group was observed (Fig. 6a, Table S6). The 3-nitropheyl ring was bound to the β pocket as expected, and multiple hydrogen bonds identified by a Protein–Ligand Interaction Profiler (PLIP) tool⁴⁸ strengthened the binding affinity between the alkoxylamine moiety and galectin-3 (Fig. 6b). The distinctive interactions observed with 5 were the formation of two hydrogen bonds between the alkoxyamino group of 5 and the R162 side chain. Additionally, the galactose moiety of 5 forms three hydrogen bonds with the H158, R162 and N174 side chains, and the 3-nitrophenyl group engages in a cation-π interaction with the R162 side chain (Fig. 6b). The 3-nitrophenyl group can further form two hydrogen bonds with the N160 side chain, although the PLIP tool did not detect (Figure S5).

Fig. 5.

Chemical structures of GalAldBZ20 analogues (compounds 3–5) and compound five analogues (compounds 6 and 7.

Table 1.

Thermodynamic parameters of GalAldBz20 and compounds 3–7.

Comp.	Kd (µM)	ΔH (kcal/mol)	− TΔS (kcal/mol)	ΔG (kcal/mol)
GalAldBZ20	0.41 ± 0.06	− 14.7 ± 0.3	5.92	− 8.72
3	3.14 ± 0.98	− 9.2 ± 1.0	1.66	− 7.51
4	4.45 ± 1.03	− 17.0 ± 2.3	9.67	− 7.30
5	0.039 ± 0.008	− 14.7 ± 0.2	4.60	− 10.10
6	0.64 ± 0.14	− 15.8 ± 0.6	7.37	− 8.45
7	1.37 ± 0.17	− 17.5 ± 0.5	9.45	− 8.00

Fig. 6.

(a) X-ray co-crystal structure of compound 5 bound to galectin-3. 2|Fo|-|Fc| map countered at 1.6 σ for compound 5 (blue mesh) was overlayed on the cartoon model. (b) Interactions between compound 5 and galectin-3 were identified by the PLIP tool. Blue solid lines and orange broken lines represent the hydrogen bonds and cation-π interactions, respectively. (c) Weighted average chemical shift differences in HSQC spectra of galectin-3 in the presence of GalAldBZ20 and compound 5. The dashed line represents the mean plus a standard deviation. (d) RMSF (Å) of alpha carbon atoms as a function of residue number. The MD analysis simulations of compound 5 bound to galectin-3 were performed for 500 nsec. Error bars represent SD (n = 3).

To verify the contribution of the 3-nitrophenyl group of 5 to the interaction with galectin-3, we performed the ITC measurements for 6 (R = H) and 7 (R = Me), analogues of 5 (Fig. 5, Table 1). The binding affinity of 6 to galectin-3 was approximately ten-fold lower than that of 5. Compound 7, containing the hydrophobic 3-methylphenyl group, showed an even worse binding affinity. These results are consistent with the prediction from the crystal structure that the 3-nitrophenyl group of 5 forms hydrogen bonds with the N160 side chain (Figure S5).

It is noteworthy that the binding affinity of 5 interacting with the β pocket is comparable with that of GB1107 (K_d = 37 nM)⁴⁹, an α pocket binder (Fig. 4a). This result indicated that the improvement in the binding affinity could be achieved by selecting suitable substituents for each pocket, regardless of the pocket type in galectin-3. Considering that another galectin family galectin-1, which does not have a β pocket, the interaction with the β pocket would be beneficial from the viewpoint of selectivity against galectin-1^41,47.

To further examine the binding modes of GalAldBZ20 and 5 to galectine-3, we compared the ¹H-¹⁵N HSQC spectra of galectin-3 in the presence of GalAldBZ20 or 5, showing relatively large chemical shift difference for some amino acid residues in the bonding sites such as R144, N164, E165, and N167 (Fig. 6c). Of great interest is the observed difference that there were no significant structural changes around the binding pocket of galectin-3 in the X-ray crystal structures. To rationalize the NMR results, we performed long MD simulations (500 ns, three times) for galection-3 with 5, GalAldBZ20 or without a ligand. The root mean square fluctuation (RMSF) values were calculated for each α-carbon atom of the amino acid residues. The results suggested that the variety and presence or absence of a ligand had little effect on the dynamic fluctuations of the protein (Fig. 6d). When comparing GalAldBZ20 with 5, it is noteworthy that the density map of the substituent derived from the hydrazide was observed in the X-ray structural analysis of galectin-3 only in the case of 5. This fact, along with the differences observed in the NMR analysis, suggests that the difference in how long the 3-nitrophenyl group is in contact with the β pocket of galectin-3 and the interaction with the introduced alkoxyamino group may be responsible for the longer contact time.

Conclusions

In conclusion, the combination of nMS with in situ chemical synthesis utilizing hydrazone formation, where equilibrium is governed by the concentration of each fragment, has proven to be a powerful strategy for the structural optimization of a lead compound. In this study, we serendipitously obtained a β pocket binder that exhibits high affinity comparable to that of the α pocket binder, demonstrating that this approach is particularly effective for identifying an optimal fragment for proteins with multiple binding sites. Once the optimal combination of a fragment and a protein pocket was identified, the linker moiety was subjected to structural optimization. Although it is unresolved how the biological processes following binding to galectin-3 change between α- and β-pocket binders, compound 5, identified in a series of studies, is expected to be a lead compound for new anticancer drugs owing to its high affinity for galectin-3. The developed strategy, which focused on the strengths of chemical equilibrium and native mass spectrometry, would be applicable for the drug discovery of challenging target molecules.

Methods

Preparation of mixed hydrazone library

Core aldehyde 2 and hydrazides were weighed into microtubes and diluted with DMSO to prepare 10 mM stock solutions. Thirty-five hydrazides were divided into five groups [each group has seven different hydrazides (see Table S1–S5)]. For each group, all seven hydrazide solutions (10 mM in DMSO, 5 µL, respectively) were mixed in a microtube. The stock solution containing the core aldehyde (10 mM in DMSO, 42 µL) was pipetted to the mixture of hydrazides. The mixture was concentrated by centrifugal concentrator to afford a mixture of hydrazones. The hydrazones in the microtube were dissolved with DMSO (50 µL) to obtain the mixture of the hydrazone (1 mM in DMSO).

Preparation of galectin-3 protein

DNA sequences of galectin-3 (residues 118–250) were inserted into the pET-47b vector. Galectin-3 proteins were expressed in E. coli BL21(DE3) as His-tagged fused proteins. The cells were grown in LB medium at 37 °C. Overexpression of proteins were induced by 0.25 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) after reducing the incubation temperature to 15 °C. After overnight cultivation, E. coli cells were harvested by centrifugation. The cell pellets were resuspended in 30 mM HEPES buffer (pH 7.4) containing 400 mM NaCl, 40 mM imidazole, 0.1 mM 4-(2-aminoethyl)benzenesulfonylfluoride hydrochloride (AEBSF), and 0.1% NP-40. The cells were lysed by sonication and the lysates were clarified by centrifugation. The supernatant was applied to HisTrap FastFlow column (Cytiva), and the eluent in 30 mM HEPES buffer (pH 7.4) containing 400 mM NaCl and 500 mM imidazole was then subjected to SEC (Superdex 200 26/60, Cytiva) in 30 mM HEPES buffer (pH 7.4) containing 400 mM NaCl. After removal of the His-tag by digestion with HRV3C protease, the protein solution was passed through the HisTrap FastFlow column to remove the uncleaved protein. Galectin-3_118–250 were further purified by SEC (Superdex 200 26/60) in PBS (pH 7.4) containing 1 mM dithiothreitol (DTT).

For preparation of ¹³C, ¹⁵N-double labeled galectin-3 (residues 112–250), M9 minimal media containing 1.0 g/L ¹⁵NH₄Cl and 2.5 g/L ¹³C-glucose was used instead of LB media. The expression and purification for the labeled protein was performed in the same protocol as the non-labeled protein.

NanoESI-MS

The solvent of galectin-3_118–250 was exchanged to 200 mM ammonium acetate using a Bio-Rad Micro Bio-Spin™ 6 column (Hercules, CA). The concentration of galectin-3_118–250 was determined by UV absorbance at 280 nm using a DeNovix DS-11 spectrometer (Wilmington, DE, USA). For the native mass spectrometry, galectin-3_118–250, GalAld and 7 different kinds of compounds (GalAldBZ01, GalAldBZ02, GalAldBZ12, GalAldBZ19, GalAldBZ20, GalAldBZ23 and GalAldBZ42 for group 1, GalAldBZ04, GalAldBZ06, GalAldBZ10, GalAldBZ13, GalAldBZ30, GalAldBZ31 and GalAldBZ35 for group 2, GalAldBZ05, GalAldBZ11, GalAldBZ14, GalAldBZ21, GalAldBZ27, GalAldBZ32 and GalAldBZ38 for group 3, GalAldBZ08, GalAldBZ09, GalAldBZ26, GalAldBZ28, GalAldBZ34, GalAldBZ41 and GalAldBZ47 for group 4, GalAldBZ03, GalAldBZ07, GalAldBZ24, GalAldBZ25, GalAldBZ40, GalAldBZ43 and GalAldBZ44 for group 5,) were mixed at concentrations of 3 μM for galectin-3_118–250, 4.2 μM for GalAld, and 3 μM for each compound, respectively, in 200 mM ammonium acetate and 2% DMSO, and incubated for 1 h in 100 µL reaction volume at 25 °C.

The mass spectra were obtained using SYNAPT G2-HDMS mass spectrometer (Waters, Milford, MA) equipped with a nanoelectrospray ionization (nanoESI) source^19,50. 4 µL sample solution was deposited in a purchased (HUMANIX, Hiroshima, Japan) or self-made nanoESI emitter with 5 μm outer diameter and placed in the nanoESI source. For observation of the intact complex ions under native conditions, the following parameters were used: ionization voltage 0.6–0.7 V, ion source temperature 70 °C, 20 V of sampling cone voltage, trap collision energy 4 V, trap Ar gas flow 1.5 L/min. Data were processed using MassLynx v.4.2 software.

Differential scanning fluorimetry (DSF)—thermal shift assay

DSF measurements were performed using sample solutions containing 6.5 μM galectin-3_118–250 and 13 μM compounds in PBS (pH 7.4) with 1 mM DTT and 2% DMSO, and incubated with SYPRO Orange protein gel stain (Thermo Fisher Scientific, Waltham, MA, USA) diluted 1000-fold (w/w). 20 μL of the protein/dye solution were aliquoted into a 96-well polymerase chain reaction (PCR) plate, and an emission of 610 nm was measured using a Bio-Rad CFX98 Real-Time System (Hercules, CA, USA), with the temperature being raised from 25 to 90 °C at a rate of 1 °C/min.

ITC measurement

MicroCal PEAQ-ITC (Malvern Panalytical) was used for the ITC measurements. galectin-3_118–250 were dissolved in PBS (pH 7.4) with 1 mM DTT and 2% DMSO. A compound solution (20 μM) in PBS (pH 7.4) with 1 mM DTT and 2% DMSO was placed in the cell, whereas the micro-syringe was loaded with a protein solution (300 μM). The titrations were conducted at 298 K using 13 successive injections (the first at 0.4 μL and the remaining 12 at 3.0 μL) with a duration of 6 s per injection and 150 s between injections. The data were analyzed using a one-site model on Microcal PEAQ-ITC analysis software. For each interaction, at least three independent titration experiments were performed.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

S.I. designed and directed the research and A.K. and T.K. wrote the main manuscript text. Y.K. and K.H. synthesized compounds. K.H. and A.K. performed the MD simulation. R.T., S.A. and T.K. conducted the protein expression and the interaction analyses. All authors reviewed the manuscript.

Funding

This research was supported in part by JSPS Grant-in-Aid for Scientific Research (A) (Grant Number JP25H01014 to S.I.), Grant-in-Aid for Scientific Research (B) (Grant Number JP22H03143, JP22H02738 and JP19H03345 to S.I.), Grant-in-Aid for Scientific Research (C) (Grant Number JP23K05720 to T.K. and JP24K09343 to S.A.), Grant-in-Aid for Challenging Research (Exploratory) (grant number JP23K18171 to S.I), JST SPRING (grant number JPMJSP2119 to K.H.), and Life Science Research (Basis for Supporting Innovative Drug Discovery and Life Science Research (BINDS)) from AMED under Grant Number 25ama121039.

Data availability

All data supporting the findings of this manuscript are included in this published article and its supplementary information file. The coordinates of protein–ligand complex have been deposited in the Protein Data Bank (PDB code: 9UG4). As the manuscript is currently under peer review, the deposited data are not yet publicly released.

Declarations

Competing interests

The authors declare no competing interests.