In silico derived small molecules targeting the finger-finger interaction between the histone lysine methyltransferase NSD1 and Nizp1 repressor

Graphical abstract

Abstract

PHD fingers are small chromatin binding domains, that alone or in tandem work as versatile interaction platforms for diversified activities, ranging from the decoding of the modification status of histone tails to the specific recognition of non-histone proteins. They play a crucial role in their host protein as mutations thereof cause several human malignancies. Thus, PHD fingers are starting to be considered as valuable pharmacological targets. While inhibitors or chemical probes of the histone binding activity of PHD fingers are emerging, their druggability as non-histone interaction platform is still unexplored. In the current study, using a computational and experimental pipeline, we provide proof of concept that the tandem PHD finger of Nuclear receptor-binding SET (Su(var)3–9, Enhancer of zeste, Trithorax) domain protein 1 (PHD_VC5HCH_NSD1) is ligandable. Combining virtual screening of a small subset of the ZINC database (Zinc Drug Database, ZDD, 2924 molecules) to NMR binding assays and ITC measurements, we have identified Mitoxantrone dihydrochloride, Quinacrine dihydrochloride and Chloroquine diphosphate as the first molecules able to bind to PHD_VC5HCH_NSD1 and to reduce its documented interaction with the Zinc finger domain (C2HR_Nizp1) of the transcriptional repressor Nizp1 (NSD1-interacting Zn-finger protein). These results pave the way for the design of small molecules with improved effectiveness in inhibiting this finger-finger interaction.

1. Introduction

PHD fingers are small and abundant non-catalytic Zn²⁺ binding domains annotated in more than 170 human chromatin-related proteins [1]. They are often present in single or multiple copies or in tandem with other epigenetic reader domains in so called erasers or writers, i.e. enzymes harboring catalytic domains able to change the chromatin modification status [2], [3], [4]. They are typically involved in the recognition of specific histone marks (H3K4me0, H3K4me3, H3K9me3) to modulate the transcriptional activity of the hosting protein [5], [6]. Mutations of PHD fingers can lead to aberrant protein functions that on turn can trigger different pathological conditions, including neurological and developmental diseases, cancer and immunological disorders [7]. Hence, similarly to other chromatin binding domains such as BRD [8] and PWWP [9], PHD fingers are becoming appealing epigenetic drug targets, thus representing a promising pharmacological alternative to classical enzymatic inhibition strategies of their host protein [10], [11], [12], [13], [14]. Importantly, the versatile structural scaffold of PHD fingers, alone or in tandem with other chromatin binding domains, results in diversified functions that go beyond the perceiving of the epigenetic landscape [15]. In particular, they have emerged as multifaceted interaction platforms, well-suited for bridging their host proteins with other subunits of bigger macromolecular chromatin complexes [12], [16], [17], herewith offering additional pharmacological intervention opportunities. A paradigmatic example of PHD fingers serving as hub for protein–protein interactions is represented by the tandem PHD finger of Nuclear receptor-binding SET (Su(var)3–9, Enhancer of zeste, Trithorax) domain protein 1 (NSD1). It is composed by a canonical (PHD_V) and a degenerate (C5HCH) PHD finger domain that form together an indivisible structure (PHD_VC5HCH_NSD1) [18] (Fig. 1). NSD1 is a histone methyl transferase whose translocation or mutations lead respectively to pediatric acute myeloid leukemia (AML) [19] and to the inherited overgrowth disease Sotos Syndrome [20]. Herein PHD_VC5HCH_NSD1 appears to play a pathophysiological role, as it contributes to oncogenesis and inappropriate Hox genes activation in AML condition [19] and is hotspot for Sotos Syndrome point mutations [18]. Moreover, it mediates the recruitment of its host protein to the transcriptional repressor Nizp1 (NSD1 interaction zinc finger protein) [21], [22] via a direct finger-finger interaction with the Zinc-finger domain of Nizp1 (C2HR_Nizp1), herewith reinforcing its repressive activity [22] (Fig. 1). The pathophysiological role of this repressive complex is still elusive, therefore the identification of chemical probes able to interfere with the PHD_VC5HCH_NSD1/C2HR_Nizp1 interaction would be a desirable tool to dissect its biological outcomes. Modulators of this finger-finger interaction might create new opportunities for the development of inhibitors of aberrant NSD1 function(s) and complement the classical blockade of its histone methyl transferase activity [23]. In this framework we have previously solved the NMR structure of PHD_VC5HCH_NSD1, and shown that it does not recognize Histone H3 tail peptides, but it binds to C2HR_Nizp1. We have generated a three-dimensional model of the complex and shown that the interaction (Kd = 4 μM) is highly specific. In particular, an exposed evolutionary conserved RWR loop of C2HR_Nizp1 accommodates into the PHD_VC5HCH_NSD1 interdomain groove and creates hydrophobic and electrostatic intermolecular interactions [18] (Fig. 2A). The biophysical assessment of this finger-finger interaction has created the premises for studies aiming at the identification of small molecules and/or molecular probes able to interfere with NSD1/Nizp1 complex and modulate NSD1 activity. Identification of protein–protein interactions (PPIs) modulators offers a wide range of opportunities in terms of specific inhibition. However, the targeting of PPIs with small molecules poses formidable challenges, mainly because of the large multifaceted binding surfaces of PPIs that make their druggability quite difficult [24], [25]. Here, coupling the virtual screening of a small subset of the ZINC database (Zinc Drug Database, ZDD) to NMR binding assays and ITC measurements, we have identified 3 compounds able to bind to PHD_VC5HCH_NSD1 and to reduce the interaction with C2HR_Nizp1. These results pave the way for the design of more efficient inhibitors of this finger-finger interaction.

Fig. 1.

Schematic representation of NSD1 and Nizp1 domain architecture and of the interaction between PHD_VC5HCH_NSD1 and C2HR_Nizp1. C-terminal residue numbers of NSD1 and Nizp1 and domain boundaries of PHD_VC5HCH_NSD1 and C2HR_Nizp1 are indicated. The representation is not drawn to scale.

Fig. 2.

Ligandability assessment of PHD_VC5HCH_NSD1. A) Data driven docking model of PHD_VC5HCH_NSD1/C2HR_Nizp1 complex [18]. PHD_VC5HCH_NSD1 and C2HR_Nizp1 are represented in surface and cartoon, respectively. B) FTmap consensus sites (CSs) identified on the PHD_VC5HCH_NSD1 structure. The protein and the CSs clusters, ranked from the highest to the lowest populated, are shown in solid and mesh surface, respectively.

2. Materials and methods

2.1. FTMap analysis

Hot spots identification was performed using the FTMap computational map server (www.ftmap.bu.edu). The lowest energy structure of PHD_VC5HCH_NSD1, as determined by NMR, (PDB code PDB 2NAA) was uploaded into the FTMap server and ran according to instructions [26]. The results were visually inspected using PyMol [27].

2.2. Virtual screening

The in silico screening of the ZINC Drug Database (ZDD) against PHD_VC5HCH_NSD1 was performed using the virtual screening workflow of the Schrödinger software suite 2019-3 (Schrödinger, LLC, New York, NY, 2019). The lowest energy structure of PHD_VC5HCH_NSD1 (PDB 2NAA) [18] used for virtual screening was prepared with the Protein Preparation Wizard tool of Maestro [28]. The orientation of the solvent exposed hydroxyl groups of Serine, Threonine and Tyrosine were optimized. A restrained minimization was run using the OPLS3 force field [29] with a root mean square deviation (RMSD) tolerance on heavy atoms of 0.3 Å. The receptor grid was generated defining the centroid of active site residues E2204 and S2123, which are directly involved in PHD_VC5HCH_NSD1/C2HR_Nizp1 interaction [18] with a box dimension of 12 Å × 15 Å × 18 Å. As S2123 plays a fundamental role in complex formation [18], we allowed rotation of its hydroxyl group. LigPrep [30] was used to prepare the 2924 compounds of the ZDD with Epik [31], [32] at 7 ± 2.0 pH units using standard parameters.

Virtual screening was carried out in three steps including high-throughput virtual screening (HTVS), Standard Precision (SP), and extra precision (XP) docking stage, after each step, the best 30, 50 and 20% compounds in terms of docking score were kept, respectively. To obtain a post-processing ligand binding energy a molecular mechanics generalized Born surface area (MM-GBSA) calculation was performed on the final 172 compounds with Prime [33]. The best 50 molecules in terms of ΔG of binding were selected for visual inspection. Among them 19 unique compounds, commercially available and showing good solubility in water were purchased for experimental validation. ADME-related properties of the selected compounds were evaluated using QikProp [30] program running in normal mode. QikProp generates physically relevant descriptors, and uses them to perform ADME predictions, number of violations of Lipinski’s rule of five [34] and of Jorgensen’s rule of three [35].

2.3. Reagents

The following molecules were purchased from Sigma-Merck: ZINC08101116 (Gentamicine sulfate), ZINC03872123 (Clozapine), ZINC00001547 (Hydroxystilbamidine bis(methanesulfonate)), ZINC08214681 (Streptomycin), ZINC08214483 (Amikacin), ZINC08214692 (Tobramycin), ZINC04213094 (Isoemetite), ZINC18098320 (Chlorhexidine), ZINC01530861 (Chloroquine diphosphate), ZINC00601274 (Astemizole), ZINC03830246 (Quinacrine dihydrochloride), ZINC03794794 (Mitoxantrone dihydrochloride), ZINC08214590 (Kanamycin). The following molecules were purchased from Mcule ZINC22443609 (plerixafor), ZINC3830583 (Clindamycin hydrochloride), ZINC52981502 (Netilmicin), ZINC03830924 (Idabucirin HCl), ZINC04097283 (Lyncomicyn hydrochloride) and ZINC14880004 (Rolitetracycline).

2.4. Sample preparation for NMR and binding assays

Murine PHDvC5HCH_NSD1 (Glu2117–Asp2207, NCBI Reference Sequence: NM 008739.3) and murine C2HR_Nizp1 (residues Glu397–Lys434, corresponding to Glu398–Lys435 in the human sequence, NCBI Reference Sequence: NM 032752.3) were expressed in E. coli and purified as described in [18]. As the mouse PHD_VC5HCH_NSD1 sequence shares 99% identity with the human one (residues Glu2116–Asp2206 in the human sequence NM 022455.4), we adopted the human sequence numbering scheme to map ligand interactions. Uniformly ¹⁵N-labeled PHDvC5HCH_NSD1 was expressed by growing E. coli BL21 (DE3) cells in minimal bacterial medium containing ¹⁵NH₄Cl, as sole nitrogen source. The NMR buffer of both PHDvC5HCH_NSD1 and C2HR_Nizp1 contained D₂O 10% (v/v) 20 mM sodium phosphate pH 6.3, 0.15 M NaCl, 2 mM dithiothreitol (DTT), 50 μM ZnCl₂ with 0.15 mM 4,4-dimethyl-4-silapentane-1-sulfonic acid (DSS). Stock solution of ligands (10 mM) contained 20 mM phosphate buffer, pH 6.3, 0.15 M NaCl.

2.5. NMR measurements

NMR spectra were recorded at 295 K on a Bruker Avance 600 MHz spectrometer (Karlsruhe, Germany) equipped with a triple-resonance TCI cryoprobe with an x, y, z-shielded pulsed-field gradient coil. Spectra were processed with Topspin™ 3.2 (Bruker) and analyzed with CcpNmr Analysis 2.3 [36]. ¹H-¹⁵N-HSQC assignments of PHDvC5HCH_NSD1 were taken from the BMRB databank (accession number: 25933). ¹H assignment of the 1–3 has been performed acquiring and analyzing classical 2D ¹H–¹H TOCSY (TOtal Correlation SpectroscopY, t_mix = 60 ms), 2D ¹H¹H NOESY (Nuclear Overhauser Effect SpectroscopY, t_mix = 200 ms, T = 295 K) in 20 mM phosphate buffer, pH 6.3, 0.15 M NaCl.

Titrations. For NMR titrations, at each titration point a 2D water-flip-back ¹H-¹⁵N-edited HSQC spectrum was acquired with 2048 (160) complex points for ¹H (¹⁵N), respectively, apodized by 90° shifted squared (sine) window functions and zero filled to 256 points for indirect dimension. Assignment of the labelled proteins in the presence of the ligands was obtained following individual cross-peaks through the titration series. For each residue the weighted average of the ¹H and ¹⁵N chemical shift perturbation (CSP) was calculated as CSP = [(Δδ²HN + Δδ²N/25)/2]^1/2 [37].

Ligand titrations have been performed on ¹⁵N-PHDvC5HCH_NSD1 adding 0.5, 1, 2, 3, 6, 10, 20 equivalents of ligands to the labelled protein. In order to minimize dilution and NMR signal loss, titrations were carried out by adding small aliquots of concentrated ligands (typically 10 mM, dissolved in 20 mM phosphate buffer, pH 6.3, 0.15 M NaCl) to the ¹⁵N labelled protein samples (0.13 mM, dissolved in NMR buffer).

Dissociation constant estimation. The apparent dissociation constants of the ligands-¹⁵N- PHDvC5HCH_NSD1 interactions were estimated from least-squares fitting of CSPs as a function of total ligand concentration according to the following equation:

$${\delta }_i=\frac{b-\sqrt{b^2-4ac}}{2a}$$

with a = 8(K_a/δ_b)∙[P_t], b = 1 + K_a∙([L_ti] + [P_t]), and c = δ_b∙K_a∙[L_ti], [L_ti] is the total ligand concentration at each titration point, [P_t] is the total protein concentration, K_a = 1/K_d is the association constant, and δ_b is the chemical shift of the resonance in the complex, δ_i is the absolute change in chemical shift for each titration point. The K_d were obtained from the average plus the standard deviation of the fitting values of 7 residues, whose resonances were well resolved in the HSQC spectra and had CSP > avg + sd. K_d and δ_b were used as fitting parameters using the Xmgrace program (http://plasma-gate.weizmann.ac.il/Grace/).

Saturation Transfer Difference (STD) experiments. STD experiments have been performed on 2 mM ligands in the presence of 0.1 mM PHDvC5HCH_NSD1 in NMR buffer. STD experiments were acquired using a pulse scheme (Bruker pulse sequence: stddiffesgp.3) with excitation sculpting with gradients for water suppression and spin-lock field to suppress protein signals. The spectra were acquired using 128 scans, a spectral width of 9600 Hz, 64 K data points for acquisition. For protein saturation, a train of 60 Gaussian shaped pulses of 50 ms was applied, for a total saturation times of 3 s, with relaxation delays of 3 s. On- and off-resonance irradiations were set at 0 ppm and at 107 ppm, respectively. STD spectra were obtained by internal subtraction of the on-resonance spectrum from the off-resonance spectrum. To analyze the STD effect we used the amplification factor (AF_STD) [38]. The AF_STD at a given ligand concentration ([L]_T) was obtained by multiplying the relative STD effect of a given ligand hydrogen (I_STD/I₀) (where I_STD and I₀ correspond to the peak integral in the STD spectrum and in the off-resonance spectrum, respectively, and I_STD = I₀-I_SAT, where I_SAT is the intensity of the saturated peak) with the molar ratio of ligand in excess relative to the protein ([L]_T/[P]):

$${\mathit{AF}}_{\mathit{STD}}=\frac{I_{0-}{I}_{\mathit{SAT}}}{I_0}\times \frac{{[L]}_T}{[P]}=\frac{I_{\mathit{STD}}}{I_0}\times \frac{{[L]}_T}{[P]}$$

To obtain an epitope mapping we quantitatively expressed the differences in A_STD for the different hydrogens calculating the relative STD effects (STD%), whereby for each ligand the hydrogen with highest A_STD was set to 100%, and used as a reference to calculate the relative STD effects for the other protons.

2.6. Isothermal titration calorimetry competition experiments

ITC titrations were performed at 23 °C using a VP-ITC isothermal titration calorimeter (MicroCal LLC, Northampton, MA, USA). Recombinant proteins and ligands were dissolved in the same buffer (20 mM NaH₂PO₄ /Na₂HPO₄ pH 7.2, 0.15 M NaCl, 2 mM beta-mercaptoethanol, 10 μM ZnCl₂). The ΔG of binding between C2HR_Nizp1 and PHD_VC5HCH_NSD1 was previously determined in [18]. To monitor the inhibition activity of 1–3, we stepwise titrated 1.5–3 mM C2HR_Nizp1 into a cell containing a pre-formed protein–ligand complex (formed by 0.15 mM PHD_VC5HCH_NSD1 with a 20-fold excess of either molecule 1, 2 or 3) up to a three or five-fold molar excess. The quantity of heat absorbed or released in the process was measured. Control experiments were performed under identical conditions to determine the dilution heat of the titrant C2HR_Nizp1 into buffer and of the buffer into protein–ligand samples. Data were analyzed with the software ORIGIN 7.0®.

3. Results

3.1. Assessment of PHD_VC5HCH_NSD1 ligandability

PPIs are traditionally classified as difficult pharmacological targets because of their ‘poor druggability’ reputation, mainly due to the structural and dynamic complexity of the interfacial features [24], [39]. However, recent successful studies have challenged this assumption demonstrating that the targeting of protein–protein interfaces is feasible, thus increasing the possibilities of pharmacological intervenes [40]. In particular, PPIs can be mediated by so-called “hot spots”, where specific hydrophobic interactions and, to a lesser extent, polar interactions drive most of the affinity [41]. This is the case for PHD_VC5HCH_NSD1/C2HR_Nizp1 interaction where hydrophobic and polar contacts formed by the R₄₁₅W₄₁₆R₄₁₇ (RWR) loop of C2HR_Nizp1 and the PHD_VC5HCH_NSD1 interdomain groove represent the major driving force for complex formation, as assessed by mutagenesis studies [18]. Prompted by these results, we applied FTMap [26] to assess in silico the ligandability of the surface of PHD_VC5HCH_NSD1. Herein, the druggability of a target protein was defined based on two criteria: i) the number of probe clusters within a consensus site (CS) higher than 16 and ii) the existence of at least one additional weaker hot spot within 8 Å from the CS itself. FTMap identified 7 CSs (Fig. 2B), with the first two being the largest one that fulfilled the druggability requirements. In particular, the highest populated site (CS1) was composed by 20 probe clusters and was close to one weaker hot spot (CS7, consisting of 4 probes). The second one (CS2) contained 19 probes and was within 8 Å from two weaker hot spots (CS3, and CS4 consisting of 16 and 13 probes, respectively). Interestingly, CS2 was located at the interface between PHDv_NSD1 and C5HCH_NSD1, in the C2HR_Nizp1 interaction surface (Fig. 2), supporting the notion that this region could be druggable and appropriate for a virtual screening campaign.

3.2. Virtual screening: targeting the PHD_VC5HCH_NSD1/C2HR_Nizp1 interaction surface

We used the Schrödinger Suite 2019-3 (Schrödinger Inc., LLC) to dock into PHD_VC5HCH_NSD1 (pdb code: 2NAA) the ZINC Drug Database, composed by 2924 commercially available compounds, including worldwide commercially available approved drugs as pure compounds, Drugbank approved molecules and nutraceuticals, and Food and Drug Administration approved drugs. The screening workflow consisted in three sequential docking steps comprising a high-throughput virtual screening (HTVS), a Standard Precision (SP), and an extra precision (XP) docking stage. 172 compounds emerging from the XP step were subsequently re-ranked according to molecular mechanics generalized Born surface area (MM-GBSA) calculations (Fig. 3). The best 50 molecules in terms of binding energy were selected for visual inspection. Finally, 19 unique compounds, commercially available and showing good water solubility were selected for further experimental validation (Supplementary Table S1). The “drug-likeness”, the bioavailability and the pharmacokinetic profiles of the candidates were assessed on the basis of Lipinski’s “Rule of Five” (Ro5) [34], Jorgensen’s “Rule of three” (Ro3) [35] and ADME properties (absorption, distribution, metabolism and excretion), respectively (Supplementary Table S1). As proposed by Jorgensen and Duffy [35] to assess the pharmacokinetic profiles of the selected compounds we used an overall ADME-compliance score (#stars), indicating the number of property descriptors computed by QikProp that fall outside the 95% range of similar values for known drugs. Herein, we observed that 21.1% and 26.3% of the compounds showed no Lipinski and Jorgensen violations, respectively, while the 85% and 100% showed ≤2 Ro5 and Ro3 violations, respectively (Supplementary Fig. S1). Moreover, the 79% of the selected candidates showed suitable ADME properties with up to a maximum of 5 molecular descriptors and predicted properties falling outside the normal range of known drugs, thus indicating that the selected molecules are expected to have good pharmacokinetic properties (Supplementary Fig. S2).

Fig. 3.

Virtual screening workflow.

3.3. Hits validation

The water-soluble and commercially available hits identified in the virtual screening (19 molecules) (Table 1) were subsequently tested for their binding to PHD_VC5HCH_NSD1 by NMR spectroscopy using protein-based methods [42] to assess possible interactions and mapping thereof on PHD_VC5HCH_NSD1 structure. We stepwise titrated purified ¹⁵N-labeled PHD_VC5HCH_NSD1 (0.13 mM) with increasing concentrations of ligand (up to 2.5 mM), whereby for each titration point we recorded heteronuclear single-quantum coherence (HSQC) spectra. 3 molecules turned out to be negative for binding, as judged by their ¹⁵N HSQC spectra that superimposed perfectly on a reference experiment (Supplementary Fig. S3A), and 7 molecules induced protein aggregation. 6 molecules caused very weak and scattered chemical shift perturbations (CSPs) on the protein surface, suggestive of aspecific interactions (Supplementary Fig. S4). Only 3 compounds (1, ZINC03794794 – Mitoxantrone dihydrochloride; 2, ZINC03830246 – Quinacrine dihydrochloride; 3, ZINC01530861 – Chloroquine diphosphate) (Fig. 4A) proved to be positive, and induced several small but significant CSPs, indicative of weak interactions occurring in the fast exchange regime on the NMR time scale (Fig. 4B). By fitting the shifts of selected NMR peaks upon addition of increasing ligands concentrations we estimated a K_d of 1.2 ± 0.4 mM for 1, 1.4 ± 0.3 mM for 2 and 4.7 ± 1.4 mM for 3 (Fig. 4C, Table 1). Overall, the pattern of significant CSPs (CSP > avg + sd) was similar for the titrated molecules, with resonances of C2124_NSD1, G2125_NSD1, A2144_NSD1, C2146_NSD1 S2180_NSD1, H2205_NSD1 shifting upon addition of 1–3 (Fig. 5A, Supplementary Table S2). Here, projection onto the docking poses of the residues significantly shifting during ligands titrations allowed to generate a mapping of the interactions, which indicates that the three hits targeted one face of the inter-domain groove between PHDv and C5HCH (Fig. 5B). Interestingly, the resonances of residues mostly affected by ligands binding coincided or were nearby those significantly affected by C2HR_Nzip1 interaction (Fig. 5C, Supplementary Table S2). We next performed saturation transfer difference (STD) experiments as orthogonal NMR methods to further validate the binding of molecules 1–3. STD is a well-recognized ligand-based NMR method, commonly used to probe ligand–protein interactions of medium-weak affinity (high nM to mM) [43]. The experiment is based on the intermolecular saturation transfer from the protein to the binding part of the ligand. Thus, the part of the ligand in direct proximity to the protein shows the most intense NMR signals, herewith allowing for the identification of those ligand hydrogens which are closest to the receptor [43]. Reference and STD spectra performed on a solution containing 2 mM ligand and 0.1 mM PHD_VC5HCH_NSD1 exhibited narrow and well resolved lines. Significant STD effects were observed for ligands 1–3 unequivocally demonstrating their binding to PHD_VC5HCH_NSD1 (Fig. 6A,B). To obtain group ligand epitopes [38], i.e. to identify those hydrogens in close contact with PHD_VC5HCH_NSD1, for each molecule we calculated the relative STD effect (STD%) at 2 s of saturation time. We observed that for the three molecules the aromatic protons displayed the highest relative STD effect (with 70–100% saturation), implying their close proximity to the protein. Conversely, the aliphatic protons contributed to a lesser extent suggesting only a modest involvement in the interaction (Fig. 6A,B). Overall, MM-GBSA refined docking poses of the three ligands were consistent with NMR chemical shift perturbations and in part with STD epitope mapping (Fig. 5, Fig. 7). Ligand Interaction Diagrams of the binding site showed that the aromatic moieties of all the compounds bound to the hydrophobic groove at the interface of PHDv and C5HCH establishing critical apolar interactions with residues L2147_NSD1, C2178_NSD1, P2179_NSD1, F2182_NSD1, E2204_NSD1 and H2205_NSD1 (Fig. 7A,B). In the docking poses all the compounds created also polar interactions with residues around the hydrophobic interface groove, whereby each compound established at least two hydrogen bonds with S2180_NSD1 and the carbonyl of S2123_NSD1. Molecule 1 and molecule 3 established additional polar contacts with E2120_NSD1, C2121_NSD1, C2178_NSD1, H2205_NSD1, and with the carbonyl of C2124_NSD1, respectively (Fig. 7). However, the relatively low percentage of saturation in STD experiments observed for the methylene nearby the hydroxyl groups of molecule 1 and the N-ethyl groups of molecules 2 and 3 suggests that these aliphatic chains do not stably interact with PHDvC5HCH_NSD1. Collectively, validation of the virtual screening hits through ligand based and protein based NMR experiments allowed to identify three aromatic ligands that specifically target the inter-domain region of PHDvC5HCH_NSD1 that is recognized by the RWR motif of C2HR_Nizp1.

Table 1.

Molecules emerging from VS selected for experimental validation. For positive hits the dissociation constant (K_d) are reported. For negative hits the type of interaction (aspecific, aggregation (*) or no binding (–) is reported.

#	ZINC code	Commercial Name	Interaction/Kd
1	ZINC03794794	Mitoxantrone dihydrochloride	1.2 ± 0.4 mM
2	ZINC03830246	Quinacrine dihydrochloride	1.4 ± 0.3 mM
3	ZINC01530861	Chloroquine diphosphate	4.7 ± 1.4 mM
4	ZINC18098320	Chlorhexidine	*
5	ZINC08101116	Gentamicin sulfate	Aspecific
6	ZINC22443609	Plerixafor	*
7	ZINC08214590	Kanamycin	–
8	ZINC14880004	Rolitetracycline	*
9	ZINC03872123	Clozapine	Aspecific
10	ZINC03830583	Clindamicin hydrochloride	–
11	ZINC00001547	Hydroxystilbamidine bismethanesulfonate	*
12	ZINC52981502	Netilmicin	Aspecific
13	ZINC08214681	Streptomycin	Aspecific
14	ZINC08214483	Amikacin	*
15	ZINC08214692	Tobramycin	Aspecific
16	ZINC00601274	Astemizole	*
17	ZINC03830924	Idabucirin hydrochloride	*
18	ZINC04097283	Lyncomicyn hydrochloride	–
19	ZINC04213094	Isoemetite	Aspecific

Fig. 4.

Molecules 1–3 bind to ¹⁵N labelled PHD_VC5HCH_NSD1. A) The chemical structures of the 3 hits. B) Superposition of ¹⁵N HSQC spectra of PHD_VC5HCH_NSD1 (0.13 mM, 20 mM phosphate buffer, pH 6.3, 0.15 M NaCl, T = 295 K) in the absence (in black) and in the presence (red) of twenty-fold excess of 1 (top), 2 (middle), and 3 (bottom). C) Plot of normalized chemical shift change versus concentration for selected residues, when 1 (top), 2 (middle) and 3 (bottom) are titrated into ¹⁵N-PHD_VC5HCH _NSD1. Dissociation constants were measured by fitting average CSP at each concentration of titrant. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 5.

Mapping of the ligands binding sites. A) Histograms showing chemical shift perturbations (CSPs) of ¹⁵N‐labeled PHD_VC5HCH_NSD1 amides (0.1 mM) upon addition of 20‐fold excess of 1 (top), 2 (middle), 3 (bottom). Missing residues are prolines. Green circles indicate residues whose amides resonances are not present in the free protein because of exchange with the solvent. Red triangles indicate residues whose amide resonances disappear upon ligand addition (C2146_NSD1, L2147_NSD1, N2148_NSD1, W2157_NSD1, E2158_NSD1 for 1 and L2147_NSD1 for 2). Residues with CSPs > avg + sd elicited by 1 (top), 2 (middle), 3 (bottom) (or disappearing upon ligand binding) are projected in cyan on B) their corresponding binding pose; on C) PHD_VC5HCH_NSD1 in complex with C2HR_Nizp1 (cartoon representation). Residues whose resonances were perturbed upon addition of C2HR_Nizp1 are shown in yellow. Residues affected by both the ligands and C2HR_Nzip1 are highlighted in magenta. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 6.

Ligands epitope mapping. A) Aromatic and aliphatic regions of the STD experiment performed on 1 (top), 2 (middle), 3 (bottom). Resonance assignments are indicated according to (B). On- and off-resonance spectra are shown in the lower and upper panels, respectively. B) Low (0% < STD% < 30%), medium (30% ≤ 8STD% < 70%), and high (STD% ≥ 870%) relative STD percentage are mapped with green, yellow, and red circles, respectively on the 2D chemical structure. ¹H assignments of 1–3 are summarized in Supplementary Table S3). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 7.

Binding pose of 1–3. A) Ligand Interaction diagram and B) binding pose of 1 (left), 2 (center) and 3 (right). Molecules and protein interacting residues are shown in orange and blue sticks, respectively. Hydrogen bond, salt bridge, π-stacking and hydrophobic interactions are shown with blue line, yellow, green and grey dot lines, respectively. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

4. Molecules 1–3 reduce the interaction between C2HR_Nizp1 and PHD_VC5HCH_NSD1

As both the CSPs mapping of the interaction on PHD_VC5HCH_NSD1 and the docking poses suggested that 1–3 obstructed to a certain extent the C2HR_Nizp1 binding site (Fig. 1B, Fig. 7), we asked whether these molecules, despite their low binding affinity, were able to interfere at least in part in the interaction between PHD_VC5HCH_NSD1 and C2HR_Nizp1. To verify this hypothesis, we performed ITC experiments titrating C2HR_Nizp1 into PHD_VC5HCH_NSD1 solutions that were previously saturated with a twenty-fold excess of 1–3 and compared the results with ITC titrations in the absence of ligands. Herein, we observed that all the three molecules reduced the interaction between C2HR_Nizp1 and PHD_VC5HCH_NSD1 by one order of magnitude, with the dissociation constant increasing from 4 μM to 40–50 μM in the absence and presence of ligands, respectively (Fig. 8A–D, Table 2). Saturation of PHD_VC5HC_NSD1 with 1–3 reduced the contributions of both the enthalpic and the entropic terms to the ΔG of binding. These results, in accordance with the observed CSPs in NMR titrations, support the notion that 1–3 target at least in part the C2HR_Nizp1 binding site herewith weakening the interaction between PHD_VC5HCH_NSD1 and_. C2HR_Nizp1.

Fig. 8.

Molecules 1–3 reduce the interaction between C2HR_Nizp1 and PHD_VC5HCH_NSD1. ITC-binding curves of C2HR_Nizp1 to PHD_VC5HCH_NSD1, in the presence of twenty-fold excess of A) 1, B) 2, C) 3 and D) in the absence [18] of 1–3. The upper panel shows the sequential heat pulses for domain-domain binding, and the lower panel shows the integrated data, corrected for heat of dilution and fit to a single-site-binding model using a nonlinear least-squares method (line). The dissociation constants are indicated.

Table 2.

Thermodynamic parameters of the interaction between C2HR_Nizp1 and PHDvC5HCH_NSD1 in the absence and in the presence of a twenty-fold excess of molecules 1–3. ΔG, ΔH, ΔS, stoichiometry (n) and dissociation constants (K_d) as measured by ITC experiments at T = 293.15 K are indicated.

	ΔG (kcal/mol)	ΔH (kcal/mol)	ΔS (cal/mol)	N	Kd (μM)
No ligand [18]	−7.3 ± 0.1	4.1 ± 0.1	38.7 ± 0.1	1.1 ± 0.1	3.8 ± 0.7
1	−5.9 ± 0.1	3.6 ± 0.1	32.2 ± 0.2	0.9 ± 0.1	40.5 ± 2.2
2	−5.8 ± 0.1	2.5 ± 0.2	27.8 ± 0.1	1.0 ± 0.1	54.1 ± 1.1
3	−5.9 ± 0.1	2.7 ± 0.1	29.0 ± 0.2	1.0 ± 0.1	± 1.0

5. Discussion and conclusions

PHD fingers are small epigenetic readers able to decode the modification status of histones [5], [6]. They are involved in several diseases, especially in cancer, thus their pharmacological targeting is starting to be considered as a complementary therapeutic strategy to the classical enzymatic inhibition of their host proteins [44]. These small Zn²⁺ binding domains are versatile structural platforms well suited for diversified protein–protein interactions. This is the case for the PHD fingers of MLL1 [45] of Pygo2 [46] and of Sp140 [47], interacting respectively with Cyp33, Bcl9 and Pin1, or for the second PHD finger of AIRE, that functions as a hub for multiple protein–protein interactions [16]. While inhibitors or chemical probes of the histone binding activity of PHD fingers are starting to emerge [11], [12], [13], the druggability of PHD fingers (alone or in tandem) as non-histone interaction platform remains completely unexplored. This is possibly due to the limited information on both the biological significance and the molecular details at the basis of these interactions. Hence, identification of chemical probes targeting these complexes would be extremely enlightening in terms of their structural and functional understanding, even though tackling large PPI targets with synthetic molecules still remains a considerable endeavor [24]. In the current study we have applied a computational and experimental pipeline to investigate the ligandability of PHD_VC5HCH_NSD1 as finger-finger interaction platform. To this aim we took advantage of a 3D model of PHD_VC5HCH_NSD1/C2HR_Nizp1 interaction [18] and inspected PHD_VC5HCH_NSD1 structure for possible hotspots suitable for small molecules targeting. These so-called “sticky zones” have emerged as collecting sites of most of the binding energy of inhibitors, thus opening new possibilities for the rational selection of binding molecules [48], [49]. Indeed FTmap analysis [26] predicted the presence of two putative contiguous ligand binding hotspots, that partially coincided with PHD_VC5HCH_NSD1 interaction surface with C2HR_Nizp1. To further explore the ligandability of PHD_VC5HCH_NSD1 we performed a structure-based in silico screening followed by NMR validation of the best hits emerging from our computational protocol. We found three structurally related molecules: the type II topoisomerase inhibitor mitoxantrone (1) used as salvage therapy for AML [50], the antimalaric drugs chloroquine [51] (3) and its acridine analog quinacrine (2), repositioned for treatment of cancer [52]. They all were able to bind with millimolar affinity to PHD_VC5HCH_NSD1. This value is well in the range of the Kds measured for first hits derived from VS campaigns aiming at the identification of PPI inhibitors. For example, a millimolar affinity has been measured by NMR for Benzimidazoles able to compete with the histone-binding pocket of the Pygo PHD Finger [12]. Similarly, an NMR fragment based screening of the PWWP1 domain of NSD3 yielded hits with mM affinities [9]. Importantly, mapping of the interaction by NMR CSPs methods confirmed that the three molecules target the C2HR_Nizp1 binding pocket located at the interface between PHD_V and C5HCH. This surface is in accordance with the one defined by the Glide docking poses (Supplementary Fig. S5) and with the predicted hotspot region, with 1 accommodating in the CS4 pocket, and 2 and 3 each occupying two adjacent sites, i.e. CS1, CS2 and CS3, CS4, respectively. Orthogonal ligand-based NMR experiments confirmed binding. In particular, epitope mapping via STD experiments suggested that the fused aromatic rings, that are in common to the three molecules, mostly contribute to the interaction. Most likely they create apolar interactions with the small hydrophobic patch at the interdomain interface. Intriguingly, the quinolinyl, acridyl and antraquinone moities of chloroqine (3), quinacrine (2) and mitoxantrone (1), respectively, are reminiscent of the indole moiety of the RWR motif of C2HR_Nizp1 and might mimic its crucial interaction with PHD_VC5HCH_NSD1. As a matter of fact, saturation of PHD_VC5HCH_NSD1 with an excess of the single ligands resulted in a reduction of the finger-finger interaction strength by one order of magnitude, as assessed by ITC measurements. We are aware of the fact that the affinities of the identified compounds are still not appropriate for in vivo applications. PPI inhibitors with in vivo efficacy typically have high nanomolar-low micromolar affinities to their targets. This applies for example for the molecules JQ1 and nutlin, inhibitors of the BET-H4_acetylated [53] and of MDM-p53 interactions [54], respectively. Thus, future work should be dedicated to optimize the chemical scaffolds identified in this work in order to generate more potent inhibitors. Our results suggest that the molecules that we have identified mimic only in part the interactions that are crucial for PHD_VC5HCH_NSD1/C2HR_Nip1 complex formation. We hypothesize that fragment-based approaches in which the identification and linkage of fragments targeting contiguous hotspots could be a successful strategy. Alternatively, peptidomimetic approaches could be a viable strategy to target non-contiguous hotspots [48], [53]. Design of peptidomimetic of C2HR_Nizp1 coupled to click-chemistry stabilization are currently ongoing. Nevertheless, the results obtained in this study, show for the first time that a finger-finger interaction, that involves a relatively large interaction surface composed on one side by a RWR signature and by an α-helix, and on the other side by a shallow hydrophobic groove, has proven tractable by small-molecule inhibition. The identified compounds may be thus considered as interesting leads for the development of new and more efficient NSD1/Nizp1 inhibitors. The determination of a high resolution three-dimensional structure of the PHD_VC5HCH_NSD1/C2HR_Nizp1 complex will be fundamental for the tailored design of ligands (peptidomimetics, small molecules) targeting the non-histone binding properties of these tandem PHD fingers.

6. Author statement

AB performed the experiments, analyzed the results and wrote the manuscript. MG performed the virtual screening analyzed the results and wrote the manuscript. GQ performed NMR experiments. GM, AB, MG conceived and designed throughout the studies, analyzed and interpreted all the achieved results. GM wrote the manuscript. All authors reviewed and approved the final manuscript.

CRediT authorship contribution statement

Andrea Berardi: Investigation, Formal analysis, Methodology, Validation, Visualization Data curation Writing original draft. Michela Ghitti: Investigation, Formal analysis, Methodology, Validation, Software, Visualization Data curation. Giacomo Quilici: Investigation, Methodology. Giovanna Musco: Conceptualization, Supervision, Writing original drfat, review and editing, funding acquisition.

Appendix A. Supplementary data

The following are the Supplementary data to this article: