Introduction

In eukaryotes, the post-translational modifications (PTMs) of histones play an essential role in the regulation of chromatin structural dynamics and gene expression. Histone methylation, one of the key epigenetic modifications, usually occurs on the lysine (K) or arginine (R) residues at the N-terminal flexible tails of histones, and regulates transcriptional activation or inactivation of specific genes^{[1, 2]}. The nuclear receptor binding SET domain (NSD) family has three SET domain-containing methyltransferases, i.e., NSD1, NSD2, NSD3^[3]. NSD proteins specifically methylate K36 of histone H3 on lysine 36 (H3K36). Importantly, such PTM is associated with multiple human diseases including cancers^[4-7]. Each NSD protein contains a SET domain, a PWWP domain and PHD fingers^[8]. The catalytic SET domain is an evolutionarily conserved motif with 130~150 amino acids, while the PWWP and PHD domains are readers of other proteins or nucleic acids.

NSD1 mutations are closely relative to multiple human diseases. For instance, NSD1 mutations cause Sotos syndrome, which has the symptom of childhood overgrowth, learning disability, and mental retardation^{[4-6, 9, 10]}. NSD1-NUP98 fusion protein, resulting from chromosomal translocations, is associated with childhood acute myeloid leukemia^{[7, 11]}. The inactivating mutation of NSD1 is linked to multiple cancers as well^{[9, 12]}. Therefore, it is of significance to discover small molecule inhibitors to probe the biologic functions of NSD1 and to exploit the potential therapeutic treatment. The SET domains are drug gable targets as have been validated by the inhibitors of DOT1L/KMT4, EZH2/KMT6, and G9a^[13-15]. However, to our best knowledge, no small molecule inhibitors of the NSD1 has yet been identified.

Fragment-based drug discovery (FBDD) has been proven effective and fruitful for the past 20 years, even for challenging protein-protein interaction (PPI) targets with shallow pockets^[16]. FBDD starts from low-molecular-weight compounds (fragments), and usually directly detects the weak interactions between targets and fragments by using biophysical methods, e.g., surface plasmon resonance (SPR), fluorescence spectroscopy, nuclear magnetic resonance (NMR), and X-ray crystallography^{[17, 18]}. The combination of protein-observed and ligand-observed NMR spectroscopy enabled robust detection of ambiguous weak interactions, due to NMR nature of multi-valued structural output^{[19, 20]}.

Here, we applied an automated NMR fragment-based screening (FBS) and identified three weak binders of the NSD1 SET domain. To enhance the screening throughput, we first identified feasible hits from the fragment cocktails by using ligand-observed spectra, e.g., saturation transfer difference (STD) and WaterLOGSY. These experiments were then reproduced for each individual hit. The three specific binders were further validated by the dose-dependent chemical shift perturbations (CSPs) of ¹⁵N-labeled NSD1 SET domain. We finally analyzed the binding mode of the hit using molecular docking.

1 Materials and methods 1.1 Cloning, protein expression and purification

DNA encoding NSD1 SET domain (residues 1 852~2 082) was cloned into the pGEX4T-1 vector (GE Healthcare, Shanghai, China), which was modified with tobacco etch virus (TEV) cleavage sites. The construct was then transformed into Escherichia coli BL21(DE3)-RIL. Cells were grown in minimal medium supplemented with ¹⁵NH₄Cl for NMR samples. The protein expression was generally induced at OD₆₀₀ of 0.8~1.2 using 0.5 mmol/L isopropyl β-D-thiogalactos (IPTG) at 16 ℃ for 24 h. The glutathione S-transferase (GST)-tagged proteins were purified from pretreated bacterial lysates using GSTrap FF (GE Healthcare, Shanghai, China), then treated with TEV to cleave the N-terminal GST tag. Protein was purified further by size exclusion chromatography using a HiLoad 16/60 Superdex 200 column (GE Healthcare, Shanghai, China). The purified protein was stored in Tris buffer [20 mmol/L Tris, 200 mmol/L NaCl, 2 mmol/L tris(2-carboxyethyl)phosphine (TCEP) at pH 6.8].

1.2 NMR fragment-based screening

The whole process of NMR fragment-based screening was performed at 25 ℃ on a 700 MHz spectrometer (Agilent Technologies Santa Clara, CA, USA) equipped with a 96-well autosampler and a cryoprobe^[19]. In brief, the screening samples were prepared containing 0.4 mmol/L compounds in DMSO-d₆ solution (cocktails for primary screening), 50% D₂O and 10 μmol/L unlabeled protein. The binding hits were identified from the primary screening in cocktail and secondary screening for each individual compound based on the signals of STD and WaterLOGSY spectra.

The STD experiment was acquired using the acquisition time of 1 s, 32 dummy scans, and relaxation delay of 0.1 s, followed by a 2 s Gauss pulse train with the irradiation frequency at 20.7 ppm or 250 ppm alternatively. The total acquisition time was 15 min with 256 scans. WaterLOGSY was acquired for 15.1 min with 1 s acquisition time, 1 s relaxation and 1.3 s NOE mixing time and 256 scans.

1.3 NMR CSP calculation

The ¹⁵N-labeled NSD1 proteins were concentrated to 0.1 mmol/L in Tris buffer. The compounds were then titrated into ¹⁵N-labeled NSD1. The HSQC spectra were collected for each titration points on Agilent 700 MHz spectrometer at the ligand/protein molar ratios of 0, 0.25, 0.5, 1, 2, 5, respectively. The binding constant was best-fitted using the following equation assuming a 1:1 binding mode^[24]

$ \begin{array}{l} \mathit{\Delta }\delta = \mathit{\Delta }{\delta _{\max }}\\ \left\{ {{K_D} + {{[P]}_t} + {{[L]}_t} - {{\left[ {{{\left( {{K_D} + {{[P]}_t} + {{[L]}_t}} \right)}^2} - \left( {4*{{[P]}_t}{{[L]}_t}} \right)} \right]}^{1/2}}} \right\}/\left( {2{{[P]}_t}} \right) \end{array} $

(1)

where [P]_t and [L]_t are total concentrations of ligand and protein, respectively, Δδ is the chemical shift change with respect to the free state; maximum chemical shift change (Δδ_max) and binding affinity (K_D) were best-fits from the dose-dependent chemical shift changes.

2 Results and discussion

NSD1 is a 2 969 amino-acid protein containing four PHD fingers, two PWWP domains, and one catalytic SET domain [Fig. 1(a)]. Although the crystal structure of NSD1 containing AWS, SET and post-SET sequence motifs (amino acids 1 852~2 082) has been resolved (PDB code: 3OOI), little is known about its solution structure and dynamics. The ¹H-¹⁵N heteronuclear single quantum coherence (HSQC) spectrum is powerful to assess the protein folding and aggregation status. We hence expressed and purified the ¹⁵N-labeled protein fused to a GST tag with TEV protease cleavage site [Fig. 1(b) and 1(c)]. The transverse relaxation optimized HSQC (TROSY-HSQC) spectrum suggests that the NSD1 SET domain is mono-dispersed [Fig. 1(d)], suitable for following FBS.

Our automated NMR FBS^[19] was utilized to uncover hits of the NSD1 SET domain. NMR FBS integrates the ligand-observed and protein-observed experiments^[17]. The protein-observed NMR FBS method provides valuable information of binding sites and affinity. However, protein-observed NMR FBS requires a significant amount of ¹⁵N-labeled proteins and a high quality of the 2D ¹H-¹⁵N HSQC spectrum. Conversely, the ligand-observed NMR FBS experiments record the changes of ligand signals in the presence of a small portion of the unlabeled protein. STD and WaterLOGSY are the most widely used spectra in NMR FBS. To enhance the throughput of primary screening, we pooled the fragment library of 890 compounds into 89 cocktails with 10 compounds each. The compound and protein concentrations were set to 0.4 and 0.01 mmol/L, respectively. A total of 16 hits were identified as they presented signals in STD and inverted sign of signals in WaterLOGSY [Fig. 2(a)]. The 16 hits were further filtered by a secondary screening using the same set of aforementioned experiments for each individual compound [Fig. 2(b)]. Such a cross validation procedure produced 7 hits.

The hits identified from the above ligand-observed FBS were then validated by CSPs of the ¹⁵N labeled NSD1 SET domain. The hits were titrated to 0.12 mmol/L ¹⁵N-labeled NSD1 SET domain at a final ligand/protein molar ratio varying from 0 to 5 [Fig. 3(a)]. Three hits perturbed a common set of residues, indicating that these hits bind specifically to the NSD1 SET domain. The binding affinities determined from the dose-dependent CSPs [Fig. 3(b)] were 0.12, 0.54 and 1.22 mmol/L for the three hits, respectively (Table 1). The hit rate and the reasonable ligand efficiencies (binding energy per non-proton atom) indicate the NSD1 SET domain is a drug gable target.

The target-ligand interaction modes are essential for rational drug discovery. However, we failed to crystallize the NSD1 SET domain complexes, probably due to the low binding affinities of the hits. We hence simulated the complex structural models using the molecular docking software AutoDock, in a way as has been published previously^[21-23]. Notably, the lowest-energy docking pose demonstrated compound 1 bound to the AdoMet binding pocket [Fig. 4(a) and 4(b)] and formed hydrogen bonds with residues H2021 and K2071, respectively [Fig. 4(c)]. This interaction pattern is consistent with that of AdoMet as revealed in the complex crystal structure^[8] [Fig. 4(d)]. We recently demonstrated that paramagnetic NMR spectroscopy facilitated the filtration of the best-fit docking pose^[21-23]. We are now working on the backbone chemical shift assignment of NSD1 SET domain and subsequent acquisition of NMR structural restraints.

3 Conclusion

NSD1 is a potential therapeutic target for multiple human cancers and Sotos syndrome. We discovered new hits by using NMR FBS against the NSD1 SET domain. Three specific binders were confirmed by NMR CSPs. The molecular docking poses exhibited a similar pattern to the natural substrate AdoMet. Our study shall enable the following structure-guided hit-to-lead evolution targeting the NSD1 SET domain.