Structural basis for PHD_VC5HCH_NSD1–C2HR_Nizp1 interaction: implications for Sotos syndrome

Abstract

Sotos syndrome is an overgrowth syndrome caused by mutations within the functional domains of NSD1 gene coding for NSD1, a multidomain protein regulating chromatin structure and gene expression. In particular, PHD_VC5HCH_NSD1 tandem domain, composed by a classical (PHD_V) and an atypical (C5HCH) plant homeo-domain (PHD) finger, is target of several pathological missense-mutations. PHD_VC5HCH_NSD1 is also crucial for NSD1-dependent transcriptional regulation and interacts with the C2HR domain of transcriptional repressor Nizp1 (C2HR_Nizp1) in vitro. To get molecular insights into the mechanisms dictating the patho-physiological relevance of the PHD finger tandem domain, we solved its solution structure and provided a structural rationale for the effects of seven Sotos syndrome point-mutations. To investigate PHD_VC5HCH_NSD1 role as structural platform for multiple interactions, we characterized its binding to histone H3 peptides and to C2HR_Nizp1 by ITC and NMR. We observed only very weak electrostatic interactions with histone H3 N-terminal tails, conversely we proved specific binding to C2HR_Nizp1. We solved C2HR_Nizp1 solution structure and generated a 3D model of the complex, corroborated by site-directed mutagenesis. We suggest a mechanistic scenario where NSD1 interactions with cofactors such as Nizp1 are impaired by PHD_VC5HCH_NSD1 pathological mutations, thus impacting on the repression of growth-promoting genes, leading to overgrowth conditions.

INTRODUCTION

The Nuclear receptor-binding SET (Su(var) 3–9, Enhancer of zeste, Trithorax) domain protein 1 (NSD1), is a large multi-domain nuclear protein (2696 amino acids) belonging to a family of three structurally similar mammalian methyltransferases including NSD1, NSD2 and NSD3, that have been all linked to multiple diseases and cancer types (1–4). As NSD prototype, NSD1 is characterized by the presence of different chromatin related domains including two proline–tryptophan–tryptophan–proline domains (PWWP), five plant homeodomains (PHD), a PHD finger-like Cys–His rich domain (C5HCH) (sometimes called PHD_VI), acquired late in the evolution of the NSD family and a catalytic SET domain (5–7). In vitro NSD1-SET domain catalyzes the mono- and di-methylation of H3K36 and tri-methylation of H4K20, even though discrepancies exist as to the level and the target(s) of methylation (8–13). Importantly, mutations in the NSD1 gene lead to different aberrant developmental processes: fusion to Nup98 nucleoporin gene is associated with de novo childhood acute myeloid leukemia (AML) (14–16), and NSD1 mutations and translocations lead to Weaver and Sotos syndromes, inherited congenital malformation overgrowth syndromes leading to delayed motor and cognitive development (17–21). Indeed, mouse NSD1 knock out data show that the protein is essential for correct embryonic development (6). Although the different patho-physiological mechanisms dictated by NSD1 remain elusive, several lines of evidence suggest its direct involvement in context-dependent transcriptional repression or activation. In colon cancer cell lines NSD1 binds near various promoter elements tuning the levels of the various H3K36 methylation forms within the occupied promoter proximal region, regulating multiple genes involved in developmental processes, such as cell growth/cancer and bone morphogenesis (9). Conversely, in neuroblastoma cells NSD1 displays tumour suppressor like properties promoting MEIS1-mediated gene transcriptional repression (8). NSD1 bi-functional transcriptional regulation might be also related not only to the presence of two nuclear-interaction domains, NID^−L and NID^+L (5) but also to the presence of the other chromatin binding domains, that are expected to confer to NSD1 additional chromatin related functions, besides its SET-dependent methyltransferase activity. As a matter of fact, the fifth PHD finger and the adjacent C5HCH domain (PHD_VC5HCH_NSD1) have been shown to have context-dependent activation or repression activities with different patho-physiological outcomes. On the one hand, PHD_VC5HCH_NSD1 is a hot spot for the developmental overgrowth Sotos syndrome, as assessed by the presence of 17 pathological point-mutations within this domain (Supplementary Table S1). On the other hand, in AML context the same tandem domain contributes to inappropriate Hoxa7 and HoxA9 genes activation (22). This is in keeping with the chromatin associated functions generally attributed to these evolutionarily conserved Zn²⁺ binding ‘reader/effector’ modules (∼60 aminoacids long). PHD fingers usually interpret histones post-translational modifications (H3K4 vs. H3K4me3/2; H3R2me0 vs. H3R2me2; H3K36me3; H3K14ac) in a modification and context-specific fashion, thus promoting chromatin changes and/or protein recruitment (23,24). Consistent with its putative role as chromatin reader, murine PHD_VC5HCH_NSD1 (99% identity with human PHD_VC5HCH_NSD1 (Supplementary Figure S1) binds biotinylated H3K4me3_1–21 and H3K9me3_1–21 peptides in vitro (25), even though this interaction has been recently challenged (26). In fact, GST-pulldown assays using unfractionated calf-thymus histones or biotinylated histone peptides did not prove evidence of histone binding to PHD_VC5HCH_NSD1, thus raising a conflict in literature about its actual role as epigenetic reader (26). Notably, PHD fingers are emerging as a robust-conserved structural scaffold working as versatile non-histone binding domains, thereby extending their role to diverse cellular processes, far beyond the well documented histone tail interpretation (27,28). In line with the multifaceted role of PHD fingers, PHD_VC5HCH_NSD1 seems to function as a hub for the interaction with other proteins/domains critical for transcriptional activity, such as the C2HR domain of the transcriptional repressor Nizp1 (NSD1 interacting Zinc-finger protein), one of the few documented NSD1 interactors (7,25,26,29). Nizp1, is a poorly characterized multidomain protein, expressed in several tissues containing an N-terminal SCAN box, a repressor KRAB domain, an atypical C2HR Zinc-finger motif (C2HR_Nizp1) followed by four classical C2H2-type Zinc-fingers (7,29). Intriguingly, according to biochemical in-vitro experiments the interaction with C2HR_Nizp1 seems to be a unique peculiarity of PHD_VC5HCH_NSD1 (7,26), thus implying a functional divergence within the NSD protein family. In order to move a step further in the comprehension of the molecular mechanisms dictating PHD_VC5HCH_NSD1 patho-physiological relevance, we solved its NMR solution structure and provided also a structural rationale for the effects of seven Sotos syndrome point-mutations. To investigate the potential role of PHD_VC5HCH_NSD1 as structural platform for multiple interactions we characterized its binding to histone H3 peptides and to C2HR_Nizp1 by ITC and NMR. We observed only very weak electrostatic interactions with histone H3 N-terminal tail, conversely we proved the existence of a specific interaction (K_d∼ 4 μM) between PHD_VC5HCH_NSD1 and C2HR_Nizp1. We also solved the solution structure of C2HR_Nizp1 and generated a three-dimensional model of its complex with PHD_VC5HCH_NSD1 corroborated by site-directed mutagenesis studies. We suggest a mechanistic scenario where NSD1 interactions with cofactors, such as Nizp1 are impaired by PHD_VC5HCH_NSD1 pathological mutations, thus impacting on the repression of growth-promoting genes, and ultimately leading to overgrowth conditions.

MATERIALS AND METHODS

Sample preparation for NMR and binding assays

Murine NSD1 PHD_vC5HCH (Glu2117–Asp2207), (corresponding to residues Glu2116–Asp2206 in the human sequence NM_022455.4) was amplified from a pSG5 vector containing the cDNA for murine NSD1 (NCBI Reference Sequence: NM_008739.3). Murine C2HR_Nizp1 (residues Glu397–Lys434) was amplified from a pCVM vector containing the cDNA coding for mouse Nizp1 (protein NCBI Reference Sequence: NM_032752.1). Both plasmids were kindly provided by Prof. Chambon (Strasbourg, France). pETM11 and pETM41 (EMBL) plasmids were used to express His-tagged-PHD_vC5HCH_NSD1 and His-MBP-tagged-C2HR_Nizp1 proteins, respectively. The tags were removed by cleavage with TEV-protease. Site-directed mutagenesis was performed by standard overlap extension methods. The DNA constructs were sequenced by Eurofins (Milan, Italy). Both protein domains were expressed in Escherichia coli BL21 (DE3) cells at 28°C overnight after induction with 1mM isopropyl thio-β-d-galactoside (IPTG), in LB medium supplemented with 0.2 mM ZnCl₂. Uniformly ¹⁵N- and ¹³C-¹⁵N-labeled PHD_vC5HCH_NSD1 and C2HR_Nizp1 were expressed by growing E. coli BL21 (DE3) cells in minimal bacterial medium containing ¹⁵NH₄Cl, with or without ¹³C-d-glucose as sole nitrogen and carbon sources. Proteins were purified as described in (30). For binding assays with histone peptide arrays PHD_vC5HCH_NSD1 was cloned into pETM30 expression vector (EMBL) containing an N-terminal His-GST tag. As control only His-GST was used. The molecular masses of the recombinant proteins were checked by mass spectrometry (MALDI). Synthetic histone H3 peptides (H3_1–10, H3_1–21, H3_1–37, H3K4me3_1–21, H3K9me3_1–21) were C-amidated. C2HR_Nizp1 peptides used for ITC and NMR titrations were N-acetylated and C-amidated. They were purchased from Caslo Lyngby, Denmark. Peptide purity (>98%) was confirmed by HPLC and mass spectrometry. The NMR buffer of both PHD_vC5HCH_NSD1 and C2HR_Nizp1 contained 20 mM sodium phosphate pH 6.3, 0.15 M NaCl, 2 mM dithiothreitol (DTT), 50 μM ZnCl₂ (28,31) with 0.15 mM 4,4-dimethyl-4-silapentane-1-sulfonic acid (DSS). D₂O was 10% (v/v) or 100% depending on the experiments.

NMR spectroscopy and resonance assignment

NMR experiments were performed at 295 K on a Bruker Avance 600 MHz equipped with inverse triple-resonance cryoprobe and pulsed field gradients (Bruker, Karlsruhe, Germany). Typical sample concentration was 0.3–0.4 mM. Data were processed using NMRPipe (32) or Topspin 3.2 (Bruker) and analyzed with CCPNmr Analysis 2.1 (33). The ¹H, ¹³C, ¹⁵N chemical shifts of the backbone atoms of both PHD_VC5HCH_NSD1 and C2HR_Nizp1 were obtained from three-dimensional HNCA, CBCA(CO)NH, CBCANH, HNCO experiments. Side chain resonances were assigned through H(CCO)NH, CC(CO)NH, HCCH–TOCSY experiments (34) and classical ¹H–¹H 2D-TOCSY (Total Correlation Spectroscopy). The resonances of the aromatic side chains atoms were assigned using ¹H–¹³C HSQC, (HB)CB(CGCD)HD and the 2D TOCSY and NOESY (Nuclear Overhauser effect spectroscopy) experiments in H₂O and in D₂O. The tautomeric state of the imidazole rings of the histidines were determined performing 2D ¹H–¹⁵N HMQC (Heteronuclear Multiple Quantum Coherence) long-range experiments with transfer delay optimized to transfer the coherence from the Hϵ1 and Hδ2 atoms to the Nϵ2 and Nδ1 atoms (τ = 11 ms) and analyzing the peaks pattern according to (35). Proton–proton distance constraints were obtained from ¹⁵N and ¹³C separated three-dimensional NOESY spectra, 2D ¹H–¹H NOESY in D₂O and H₂O employing 100 ms mixing times. φ/ψ restraints were obtained from backbone chemical shifts using TALOS-N (36). Hydrogen bond restraints were defined from slow-exchanging amide protons identified after exchange in D₂O. ¹H–¹⁵N residual dipolar couplings were measured in isotropic and anisotropic phases generated by an axially compressed acrylamide gel (8%) as described in (37).

Relaxation experiments

Heteronuclear {¹H} ¹⁵N nuclear Overhauser enhancement, longitudinal and transversal ¹⁵N relaxation rates (R1, R2) were measured using standard 2D methods (38). For R1 and R2 experiments a duty-cycle heating compensation was used (39), the two decay curves were sampled at 13 (ranging from 50 to 2000 ms) and 11 (from 12 to 244 ms) different time points, respectively. Both R1 and R2 experiments were collected in random order and using a recovery delay of 2.5 s. R1 and R2 values were fitted to a 2-parameter exponential decay from the intensities using the fitting routine implemented in the analysis program NMRView (40), duplicate measurements are used to determine the uncertainty in the relaxation rate by Monte Carlo methods in NMRView. The tumbling correlation time (τ_c) was estimated from R2/R1. Only residues corresponding to non-overlapping peaks, with heteronuclear NOE > 0.65 were used, residues with anomalous short transverse relaxation time T2 were excluded (41). HSQC spectra measured in an interleaved fashion with and without 4 seconds of proton saturation during recovery delay were recorded for the {¹H}–¹⁵N heteronuclear NOE experiments. The corresponding values were obtained from the ratio between saturated and unsaturated peaks intensities. The uncertainty was calculated as the standard deviation of the noise in the spectrum divided by the intensity of the reference peak.

Structure calculations and validation

PHD_vC5HCH_NSD1 and C2HR_Nizp1 structures were calculated with ARIA 2.3.2 (42) in combination with CNS using experimentally derived restraints (Table 1). All NOEs were assigned manually and calibrated by ARIA, the automated assignment was not used. A total of eight iterations was performed, computing 20 structures in the first seven iterations and 200 in the last iteration. The ARIA default water refinement was performed on the 40 best structures of the final iteration. Initial structures were calculated without Zn²⁺ restraints to verify the position and the geometry of the metal ion ligands, so that the residues involved in Zn²⁺ binding could be identified in an unbiased manner. Several NOEs were observed between metal coordinating residues. After unequivocal identification of the metal coordinating residues, in final refinement calculations the geometry of the Zn²⁺ coordination was fixed through covalent bonds and angles in the CNS parameters (d_Zn-Sγ = 2.3 Å; d_Zn-Nϵ2,Nδ1 = 2.0 Å tetrahedral angle geometry around Zn²⁺ ion, i.e. Sγ–Zn–Sγ = 109.5°, Sγ–Zn–(Nδ1 or Nϵ2) = 109.5°); the proper angles and distances for Zn²⁺ ions coordinating residues were maintained also during water refinement. PROCHECK-NMR (43) and iCING (44) were used to assess the structural quality. The family of the 15 lowest energy structures (no distance or torsional angle restraints violations >0.5 Å or >5°, respectively) of PHD_VC5HCH_NSD1 and C2HR_Nizp1 have been deposited in the PDB (PDB accession codes 2naa and 2nab, respectively). Chemical shift and restraints lists used for structure calculations have been deposited in BioMagResBank (PHD_vC5HCH_NSD1 accession code: 25933; C2HR_Nizp1 accession code 25934). Electrostatic potential maps were calculated using APBS plugin of PyMOL (45). Figures were generated using PyMOL (The PyMOL Molecular Graphics System, Version 1.7.4 Schrödinger, LLC).

Table 1. Summary of conformational constraints and statistics for the 15 best structures of PHD_vC5HCH_NSD1.

Restraints informationa,b	<SA>c
Total number of experimental distance restraints	1344
NOEs (intraresidual/sequential/medium/long)	631/273/137/303
Zn2+ coordination restraints	16
Hydrogen bonds	7
Dihedral angle restraints (Φ/Ψ)	57/57
Residual dipolar couplingsd	22
Deviation from idealized covalent geometry
Bonds (Å)	0.0031 ± 0.0001
Angles (°)	0.498 ± 0.012
Coordinate r.m.s. deviation (Å)e
Ordered backbone atoms (N, Cα, C’)	0.67 ± 0.09
Ordered heavy atoms	1.06 ± 0.09
Ramachandran quality parameters (%)e
Residues in most favored regions	89.40%
Residues in allowed regions	10.60%

^a No distance restraint in any of the structures included in the ensemble was violated by more then 0.5 Å.

^b No dihedral angle restraints in any of the structures included in the ensemble was violated by more than 5°.

^c Simulated annealing, statistics refers to the ensemble of 15 structures with the lowest energy.

^d RDC restraints have been used only for residues adopting regular secondary structure.

^e Root mean square deviation between the ensemble of structures <SA> and the lowest energy structure and Ramachandran quality parameters calculated on residues Asp2119–Asp2205.

NMR binding assays

Protein concentrations were determined by UV spectroscopy using the predicted extinction coefficients of 12490 and 6990 M⁻¹cm⁻¹ for PHD_vC5HCH_NSD1 and C2HR_Nizp1, respectively. Peptide concentrations were estimated from their dry-weight. In order to minimize dilution and NMR signal loss, titrations were carried out by adding to the ¹⁵N labeled protein samples (typically 0.2–0.3 mM) small aliquots of concentrated (15 mM) peptides or proteins (1–2 mM) stock solutions. All the solutions were prepared in NMR buffer. For each titration point (stepwise addition of up to 12 and 2 equivalents for histone H3 peptides and for PHD_vC5HCH_NSD1/C2HR_Nizp1 titrations, respectively) a 2D water-flip-back ¹⁵N-edited HSQC spectrum was acquired with 512 (100) complex points, 55 ms (60 ms) acquisition times, apodized by 60° shifted squared (sine) window functions and zero filled to 1024 (512) points for 1H (15N), respectively. Assignment of the PHD_vC5HC_NSD1 in the presence of histone peptides was made by following individual cross-peaks through the titration series. The assignments of PHD_VC5HCH_NSD1 and of C2HR_Nizp1 amides were achieved acquiring HNCA, CBCA(CO)NH experiments in the presence of a 1.5 fold excess of unlabelled CH2R_Nizp1 and unlabelled PHD_VC5HCH_NSD1, respectively. 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 (46).

Isothermal titration calorimetry thermodynamic analysis

ITC titration was performed using a VP-ITC isothermal titration calorimeter (MicroCal LLC, Northampton, MA, USA). Recombinant proteins and synthetic peptides were dialyzed against the same buffer (20 mM NaH₂PO₄/Na₂HPO₄ pH 7.2, 150 mM NaCl, 2 mM β-mercaptoethanol, 10 μM ZnCl₂) at 23°C. Step by step injections of the 0.3–10 mM titrants (H3K9me3_1–21, C2HR_Nizp1 and mutants) solution into a cell containing a 25–200 μM PHD_VC5HCH_NSD1 were performed to finally reach a 2-fold and 7-fold molar excess of CH2R_Nizp1 and H3K9me3_1–21, respectively, with respect to the protein concentration. 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 peptides into buffer and of the buffer into protein samples. The final data were analyzed using the software ORIGIN 7.0 (OriginLab Corp., Northampton, MA, USA).

Quantum mechanics/molecular mechanics calculations

C2HR_Nizp1 structure calculations suggested that the Zn²⁺ ion was coordinated only by the thiols of C407_Nizp1 and C410_Nizp1 and by the Nϵ2 of H423_Nizp1, we therefore verified whether water molecules could complete the Zn²⁺ coordination sphere using hybrid quantum mechanics/molecular mechanics (QM/MM) calculations (Density Functional based Tight Binding) (47). Calculations have been performed on the final structure of a MD simulation (100 ns) performed on the lowest C2HR_Nizp1 NMR energy structure using Amber14 (48), the Amber ff12SB force field and the SPCE water model (49). Adaptive solvent QM/MM calculations (50) were performed including in the active region the side chains of C407_Nizp1, C410_Nizp1, H423_Nizp1, N409_Nizp1 and eleven water molecules. After the equilibration phase of the QM part, a free QM/MM simulation (100 ps) was performed using Sander module (48).

Docking model

The complex model was generated using the information-driven docking software HADDOCK 2.0 (51). 10 NMR input structures for both PHD_VC5HCH_NSD1 and C2HR_Nizp1 (residues S401-E429) were used. Ambiguous interactions restraints (AIRs) were generated based on NMR CSP data. We defined as active those residues showing a significant amide chemical shift displacement (defined as CSP > <CSP> + 1σ, whereby σ is the standard deviation) upon ligand binding and as passive those residues within a radius of 4 Å from the active ones. Among them, only amino acids with high solvent accessibility (relative all-atoms accessibilities > 40%) were considered (Supplementary Table S2). Naccess V2.1.1 (Hubbard, S.J. & Thornton, J.M. (1993), ‘NACCESS’, Computer Program, Department of Biochemistry and Molecular Biology, University College London) was used for the solvent accessible area calculations. Unambiguous restraints were applied to maintain the Zn²⁺ ion coordination sphere. Calculations generated 1000, 1000, 500 structures for the rigid body docking (it0), the semi-flexible refinement (it1) and the explicit solvent refinement (water), respectively. In it0, each model was the result of five internal docking trials (with randomization of the starting orientations), whereby for each model the pose rotated by 180° around the normal to the interface was also sampled. In the it1 step, the semi-flexible regions were automatically defined. OPLS force field (52) and TIP3P water model were used. Tetra-coordinated Zn²⁺ ions in PHD_VC5HCH_NSD1 were treated according to bonded ZAFF parameters (53), whereas the Zn²⁺ ion in C2HR_Nizp1 was treated according to the aforementioned QM/MM calculations. The final 500 structures obtained after water refinement were scored according to their Haddock Score. The latter (defined as HADDOCK_score = 1.0 EvdW+0.2 Eelec+1.0 Edesolv+0.1 EAIR) is a weighted combination of van der Waals and electrostatic energy terms, represented by Lennard–Jones and Coulomb potentials, empirical desolvation term developed by Recio et al. (54) and ambiguous interaction restraint energy term, which reflects the accordance of the model to the input restraints. HADDOCK models were clustered (55) based on their interface root mean square deviation (r.m.s.d.), setting the cutoff and the minimum number of models in a cluster to 4 Å and to 10, respectively. To remove any bias of the cluster size on the cluster statistics, the final overall score of each cluster was calculated on the five lowest HADDOCK scores models in that cluster (Supplementary Table S3).

RESULTS AND DISCUSSION

PHD_VC5HCH_NSD1 tandem domain forms a unique structural entity

We determined the three-dimensional solution structure of PHD_VC5HCH_NSD1 using multidimensional heteronuclear NMR spectroscopy (Figure 1A, B, Table 1). As the mouse PHD_VC5HCH_NSD1 sequence shares 99% identity with the human one (Supplementary Figure S1), we adopted the human sequence numbering scheme to map Sotos syndrome mutants (see further). Residues E2116-Q2163_NSD1 adopt the classical PHD finger fold consisting of an initial loop, relatively well defined from residues D2126_NSD1 to L2130_NSD1, a two-stranded antiparallel β-sheet formed by L2130-S2132_NSD1 (β1) and V2141-H2143_NSD1 (β2), followed by a short α-helix (A2144-L2147_NSD1), and the so-called variable L3 loop (N2148-E2158_NSD1) (56). A C-terminal 3–10 α-helix (P2160-H2162_NSD1) links the domain to the adjacent C5HCH domain (residues C2164-D2206_NSD1), whose sole elements of secondary structures consist of two short antiparallel β-sheets, β3-β4 (S2173-C2175_NSD1; S2180-F2182_NSD1), β5-β6 (L2191-S2194_NSD1; G2198-C2202_NSD1) arranged orthogonally one to each other (Figure 1A and B). The tandem domain is stabilized by four Zn²⁺ ions, coordinated respectively by C2121_NSD1, C2124_NSD1, C2146_NSD1 and H2143_NSD1 (Zn²⁺_I), C2133_NSD1, C2138_NSD1, C2159_NSD1 and H2162_NSD1 (Zn²⁺_II), C2164_NSD1, C2167_NSD1, C2183_NSD1 and H2186_NSD1 (Zn²⁺_III), C2175_NSD1, C2178_NSD1, C2202_NSD1 and H2205_NSD1 (Zn²⁺_IV) (Figure 1A). Metal ions binding via the imidazole rings occurs through Nδ of H2143_NSD1 and H2186_NSD1 and through Nϵ atoms for His2162_NSD1 and H2205_NSD1, as assessed from HMQC long range experiments (data not shown). Several conserved hydrophobic residues within the NSD family (F2122_NSD1,Y2142_NSD1, W2161_NSD1, F2174_NSD1, P2179_NSD1) (Supplementary Figure S1A) establish direct inter-domain interactions (Figure 1B) creating a hydrophobic groove surrounded by negatively charged residues (Figure 1C). Of note, when expressed as single domains both the PHD_V and the C5HCH domains are unstable in solution, as assessed respectively, by their low solubility and NMR peaks duplication (data not shown), thus supporting the notion that the two Zn²⁺ binding domains are intimately linked, forming an indivisible structural entity. Global and local dynamics of the tandem domain, including motions ranging from picoseconds to milliseconds time scales, were obtained analyzing ¹⁵N relaxation parameters from 79 residues that gave sufficiently resolved backbone peaks in the ¹H-¹⁵N HSQC spectrum. The R2/R1 ratio for residues located in both PHD_V and C5HCH is relatively uniform indicating that in solution the molecule behaves as a monomer and tumbles as a single body with a calculated correlation time of 7.96±0.41 ns (Supplementary Figure S2A). The steady-state heteronuclear {¹H}–¹⁵N NOE, a sensitive indicator of internal motions on a sub-nanosecond timescale, revealed that most residues displayed values around 0.8. The N- and C-termini, the L3-loop within PHD_V and the loop linking β5 and β6 showed lower values, indicating a significant mobility in the ps-ns time scale, in agreement with the paucity of inter-residue NOEs in this region (Figure 1D). The single domains are arranged in a ‘face to side’ manner, similarly to what observed in the homologue PHD_VC5HCH_NSD3 structure (65% sequence identity) (Supplementary Figures S1 and S3A). The fold similarity between the two NSD tandem domains is high, with a r.m.s.d. of 1.6 Å over 91 Cα atoms; major differences between the two structures are observable in the L3 loop, due to the presence of two consecutive Prolines (P1355-P1356_NSD3) in PHD_VC5HCH_NSD3 occupied by an Arginine and a Proline in the equivalent positions in PHD_VC5HCH_NSD1 (R2152-P2153_NSD1) (Supplementary Figure S3A). The ‘face to side’ orientation appears to be a hallmark of the NSD family, and differs from the ‘face to back’ arrangement adopted by the tandem PHD fingers of the histone acetyltransferase MOZ (monocytic leukemia Zinc-finger protein) (57,58) or of the adaptor component of BAF chromatin remodeling complex Dpf3b (59) (Supplementary Fig. S3B and C), thus defining a different structural class of PHD tandem domains.

Figure 1.

Solution structure of PHD_VC5HCH_NSD1. (A) Superposition of the best 15 NMR structures PHD_V; the L3 loop and C5HCH are coloured in orange, gold and green, respectively. Zn²⁺ binding residues and Zn²⁺ ions are represented in sticks and spheres, respectively. (B) Cartoon representation of PHD_VC5HCH_NSD1 highlighting elements of secondary structure. Residues forming the domain hydrophobic core at the domains interface are shown in sticks and dotted space-filling representation. Zn²⁺ ions are represented in spheres. Amino acids in this and in the following Figures are numbered according to the human NSD1 sequence. (C) Electrostatic surface potential of PHD_VC5HCH_NSD1. The structures on the left and on the right are represented in the same orientation as in (A) and (B), respectively. (D) Backbone dynamics of PHD_VC5HCH_NSD1. Dotted line indicates the {¹H}–¹⁵N heteronuclear NOE value threshold of 0.65. Elements of secondary structure are indicated on the top.

Structural analysis of Sotos syndrome point mutations

Of the 17 Sotos syndrome missense mutations targeting PHD_VC5HCH_NSD1, 14 involve Zn²⁺-coordinating residues and 3 affect amino-acids outside the metal binding site (Supplementary Table S1). As already observed in other PHD finger domains (30,31), the pathological effect of mutations involving Zn²⁺ coordinating Cysteines is ascribable to loss of function effects, due to domain unfolding caused by Zn²⁺ depletion. Conversely, the structural effect of Histidine mutations is less predictable, as they are often dispensable to preserve Zn²⁺ coordination and fold integrity (60) (see also further). In order to have a complete overview of the structural impact of the non-Cysteine pathological mutations we expressed and purified seven non-Cysteine Sotos mutants and compared their NMR spectra to the wild-type one (Figure 2A and B, Supplementary Figure S3E). The 1D-¹H spectra of H2143E_NSD1, H2143Y_NSD1, H2162R_NSD1 were characterized by distinctive line-broadening effects, indicative of aggregation and/or unfolding, thus revealing their fundamental role in Zn²⁺ coordination and domain stability (Figure 2A). Unexpectedly, H2205R_NSD1 mutation did not affect the domain fold as shown by the good peak dispersion of its 2D ¹H–¹⁵N HSQC spectrum (Figure 2C). Conceivably, the three Zn²⁺ binding Cysteines are sufficient to maintain metal ion coordination, and the methylene groups of the Arginine-mutant can partially replace the hydrophobic interactions established by H2205_NSD1 with M2177_NSD1 (Figure 2B). Finally, among the mutants outside the Zn²⁺ binding sites, Y2142N_NSD1 and F2182I_NSD1 involved two conserved aromatic residues establishing stabilizing hydrophobic interactions with F2122_NSD1, K2140_NSD1, W2157_NSD1, P2160_NSD1, W2161_NSD1, V2166_NSD1, M2177_NSD1, P2179_NSD1, M2190_NSD1 and L2191_NSD1) (Figure 2B). Mutations of both Y2142 _NSD1 and F2182 _NSD1 resulted in domain unfolding and/or aggregation, as assessed by the broad lines and reduced peak dispersion of their ¹H-1D spectra (Figure 2A). Conversely, mutation of the solvent exposed R2152_NSD1 into Glutamine (Figure 2B, C), did not affect the domain fold, as assessed by the remarkable similarity between the wild-type and mutant ¹H–¹⁵N-HSQC spectra (Figure 2C). In the context of Sotos syndrome, fold-destroying mutants most likely result in functional inactivation of NSD1, well in keeping with NSD1 haploinsufficiency being the primary pathogenic mechanism in the syndrome (61). However, a more complex scenario at molecular level could arise from fold-preserving mutations, whose effects might not be simply ascribed to a generic loss of function event. It is conceivable that the change in electrostatic potential induced by both the H2205R_NSD1 and R2152Q_NSD1 fold-preserving mutations results in downstream effects on the molecular complexes recruited by NSD1 and consequently on the transcriptional events regulated by these macromolecular assemblies. The functional impact of fold-preserving mutations has been already observed in other PHD finger containing proteins, such as Autoimmune regulator protein (AIRE), where PHD fingers pathological mutations acted on AIRE chromatin-associated interactome, thus reducing the activation of its target genes (31).

Figure 2.

Structural effects of Sotos syndrome mutations. (A) 1D ¹H spectra (amide region) of wild-type and pathological mutants of PHD_vC5HCH_NSD1. (B) Cartoon representation of PHD_VC5HCH_NSD1 (gray). The side chains of pathological point mutations are shown in blue sticks, residues involved in hydrophobic contacts are shown in grey sticks with dotted space-filled representation. (C) 2D ¹H–¹⁵N HSQC spectra (amide region) of wild-type and pathological mutants of PHD_vC5HCH_NSD1.

PHD_VC5HCH_NSD1 binds to C2HR_Nizp1 with low-micromolar affinity

Over the last decade PHD fingers as single modules or in combination with other chromatin binding domains have been intensively studied as histone binding recognition domains (62,63). Conversely, the possibility that PHD fingers can directly interact with non-histone proteins has been limited to only few examples, including the PHD fingers of MLL1 (28), of Pygo2 (27) and of Sp140 (64), interacting respectively with Cyp33, Bcl9 and Pin1. In line with the emerging multifaceted role of PHD fingers as structural hubs for multiple interactions, early yeast-two-hybrid screening and GST-pulldown assays, corroborated by site directed mutagenesis, have suggested a direct interaction between murine NSD1 and the nuclear repressor Nizp1 (also known as Zfp496) via the PHD_VC5HCH_NSD1 and C2HR_Nizp1 domains (7,25). Prompted by this observation, we wondered whether we could confirm and thermodynamically characterize this PHD-finger/non-histone protein interaction. Indeed, evaluation of the binding through isothermal titration calorimetry indicated that a Nizp1 fragment (residues E397-K434_Nizp1 as defined in (7)) containing the putative Zn²⁺-binding C2HR_Nizp1 domain binds to PHD_VC5HCH_NSD1 with a dissociation constant of 3.8 ± 0.7 μM in a 1:1 ratio (Figure 3A). At room temperature (T = 23°C) the reaction is endothermic (ΔH = 4.1 ± 0.1 kcal/mol) and entropy driven (TΔS = 11.4 kcal/mol) suggesting a prominent role of hydrophobic interactions in complex formation.

Figure 3.

Interaction between PHD_VC5HCH_NSD1 and C2HR_Nizp1. (A) ITC-binding curves of C2HR_Nizp1 to PHD_VC5HCH_NSD1. 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). (B) Superposition of ¹H–¹⁵N HSQC spectra of PHD_VC5HCH_NSD1 without (black) and with (red) C2HR_Nizp1. (C) Superposition of ¹H-¹⁵N HSQC spectra of C2HR_Nizp1 without (black) and with (red) PHD_VC5HCH_NSD1. The peak with the down shift to 11.8 ppm upon PHD_VC5HCH_NSD1 addition corresponds to the backbone amide of R415_Nizp1, suggesting the presence of an H-bond upon complex formation. Peaks shifting more than <CSP>+1σ are explicitly labeled.

Solution structure of C2HR_Nizp1

To get further molecular insights into the interaction between PHD_VC5HCH_NSD1 and C2HR_Nizp1 we first solved C2HR_Nizp1 solution structure by classical homonuclear and heteronuclear experiments (Table 2). Despite the presence of an Arginine (R427_Nizp1) replacing the fourth core Histidine in this atypical Zinc-finger sequence (Supplementary Figure S1B), the domain has a Zn²⁺ dependent fold, as assessed by the increased chemical shift dispersion of its 1D-¹H spectrum upon addition of one Zn²⁺ equivalent (Figure 4A). The domain adopts the typical Zinc-finger topology composed by a small two-stranded antiparallel β-sheet, encompassing residues Y405-C407_Nizp1 (β1) and G411-F414_Nizp (β2), followed by a short positively charged α-helix (R417-H423_Nizp1) (Figure 4B). The N-terminus (E397-K403_Nizp1) and the C-terminus (R425-K434_Nizp1) are extremely flexible, in line with their low heteronuclear NOE values (Figure 4C). A network of hydrophobic interactions involving F414_Nizp1, F420_Nizp1, L424_Nizp1 also contributes to domain stability (Figure 4B). Of note, W416_Nizp1 is located on a solvent exposed loop (RWR-loop) and protrudes out of the domain suggesting a possible functional role for this residue (Figure 4B). Finally, C2HR_Nizp1 is highly positively charged with a positive patch comprising the surface of the α-helix and the RWR-loop, well suited to interact with the negatively charged surface of PHD_VC5HCH_NSD1 (Figure 4D). Of note, the RWR-loop appears to be well conserved along evolution, in line with its possible functional role (Supplementary Figure S1B). Preliminary structure calculations performed without including the Zn²⁺ ion revealed that three residues out of the C2HR_Nizp1 signature (C407_Nizp1, C410_Nizp1, H423_Nizp1) are near in space, at the appropriate distance to coordinate the metal ion, whereas R427_Nizp1 (Figure 4B), located on the flexible C-terminal tail is distant from these residues and is not involved in Zn²⁺ binding. We also excluded that N409_Nizp1 side chain (Figure 4B), which was in proximity to the metal binding site, could bind Zn²⁺ ion completing its coordination sphere, as its mutation into Alanine did not have any impact on the domain fold, as assessed by comparison between the mutant and wild-type 1D ¹H NMR spectra (Supplementary Figure S4A). Final structure calculations including distance restraints for the Zn-S⁻ and Zn-Nϵ, did not highlight any backbone carbonyl or suitable side chain in the proximity of the metal ion that could complete its coordination. Since the three-coordinated Zn²⁺ ion is partially solvent exposed, water molecules can easily fill its coordination sphere, as often observed in the catalytic Zn²⁺-binding sites of enzymes or in other Zn²⁺ binding modules (60,65). Indeed, QM/MM (Quantum Mechanics/Molecular Mechanics) calculations confirmed the presence of one water molecule completing the tetrahedral Zn²⁺ ion coordination sphere (Figure 4B and Supplementary Figure S4B).

Table 2. Summary of conformational constraints and statistics for the 15 lowest energy structures of C2HR_Nizp1.

Restraints Information a,b	<SA>c
Total number of experimental distance restraints	569
C2HRNizp1 (intraresidual/sequential/medium/long)	191/156/122/100
Zn2+ coordination restraints	3
Hydrogen bonds	1
Dihedral angle restraints (Φ/Ψ)	20/20
Deviation from idealized covalent geometry
Bonds (Å)	0.0031 ± 0.0001
Angles (°)	0.481 ± 0.031
Coordinate R.m.s. Deviation (Å)d
Ordered backbone atoms (N, Cα, C’)	0.462 ± 0.146
Ordered heavy atoms	1.034 ± 0.34
Ramachandran Quality Parameters (%)e
Residues in most favoured regions	93.00%
Residues in allowed regions	7.00%

^a No dihedral angle restraints in any of the structures included in the ensemble was violated by more than 5°.

^b Simulated annealing, statistics refers to the ensemble of 15 structures with the lowest energy.

^c No distance restraint in any of the structures included in the ensemble was violated by more then 0.5 Å.

^d Root mean squared deviation between the ensemble of structures <SA> and the lowest energy structure and Ramachandran quality parameters calculated on residues Ser404–Ser426.

Figure 4.

Solution structure of C2HR_Nizp1. (A) 1D ¹H spectrum of a synthetic peptide corresponding to C2HR_Nizp1 (0.2 mM) with (bottom) and without (top) one equivalent of ZnCl₂. (B) Superposition of the best 15 NMR structures of C2HR_Nizp1, with the two β-strands, the α helix and the loops coloured in yellow, red and green, respectively. Zn²⁺ binding residues and the Zn²⁺ ions are represented in sticks and spheres, respectively. The hydrophobic core, the RWR_Nizp1 loop and R427_Nizp1 are represented in lines. A zoom into the Zn²⁺ binding site of a representative structure from QM/MM calculations is shown in the inset, highlighting the presence of one water molecule completing the tetrahedral Zn²⁺ ion coordination sphere (highlighted by red dotted lines). H423_Nizp1, C407_Nizp1, N409_Nizp1, C410_Nizp1 and the water molecule are represented in sticks. (C) Backbone dynamics of C2HR_Nizp1. The dotted line indicates the {¹H}–¹⁵N heteronuclear NOE value threshold of 0.65. Elements of secondary structure are indicated on the top of the figure. (D) Electrostatic surface potential of C2HR_Nizp1.

Mapping of PHD_VC5HCH_NSD1/C2HR_Nizp1 interaction surface

To map the interaction surface between PHD_VC5HCH_NSD1 and C2HR_Nizp1 we performed CSP experiments titrating ¹⁵N-PHD_VC5HCH_NSD1 with unlabeled C2HR_Nizp1. The CSP method is extremely sensitive and versatile, well suited to detect both strong and weak interactions, to map binding sites and detect residues both directly interacting with the ligand and/or indirectly affected by the association. Considerable changes in the slow- to intermediate-exchange regime occurred in PHD_VC5HCH_NSD1 ¹H–¹⁵N HSQC spectrum upon addition of unlabeled C2HR_Nizp1 (Figure 3B). At equimolar ratio PHD_VC5HCH_NSD1 resonances with significant amide CSP or disappearing upon binding, clustered on the negatively charged surface of the tandem domain, mainly involving residues at the interface between PHD_V and C5HCH (D2119_NSD1, F2122_NSD1, S2123_NSD1, C2124_NSD1, H2143_NSD1, L2147_NSD1, W2157_NSD1, E2158_NSD1, M2177_NSD1, E2204_NSD1) (Figure 5A and B). The reverse NMR titration, in which unlabelled PHD_VC5HCH_NSD1 was stepwise added into ¹⁵N-C2HR_Nizp1 (Figure 3C) indicated that the C2HR_Nizp1 α-helix and the solvent exposed RWR-loop were affected by the binding, as assessed by the remarkable amide chemical shifts of residues R415_Nizp1, W416_Nizp1, R417_Nizp1, V418_Nizp1, F420_Nizp1, R422_Nizp1 (Figure 5C and D). Relaxation measurements acquired on ¹⁵N-PHD_VC5HCH_NSD1 or on ¹⁵N-C2HR_Nizp1 in the presence of unlabelled C2HR_Nizp1 or PHD_VC5HCH_NSD1, respectively, yielded a correlation time of ∼12 ns, well in agreement with a 1:1 complex (MW ∼15 kDa) tumbling in solution as a unique structural entity (Supplementary Figure S2 A and B).

Figure 5.

Mapping of the interaction between PHD_VC5HCH_NSD1 and C2HR_Nizp1. (A) Histograms showing the averaged backbone chemical shift differences observed in PHD_VC5HCH_NSD1¹H–¹⁵N HSQC (0.2 mM) upon addition of a 2-fold excess of C2HCR_Nizp1. Missing bars correspond to either Prolines or amides that are not visible because of exchange with the solvent. The asterisk indicates peak disappearance (Glu2204_NSD1) upon binding. The last two bars in the plot correspond to the indole amides of W2157_NSD1 and W2168_NSD1, respectively. (B) Cartoon and surface representation of PHD_VC5HCH_NSD1. Residues showing significant chemical shift perturbation are represented in sticks and coloured in blue. (C) Histograms showing the averaged backbone chemical shift differences observed in C2HCR_Nizp1¹H–¹⁵N HSQC (0.2mM) upon addition of a 2-fold excess of PHD_VC5HCH_NSD1 In all the histograms, the dotted line represents <CSP>, the dashed line represents <CSP> + 1σ value, where σ corresponds to standard deviation. (D) Cartoon and surface representation of C2HR_Nizp1 (yellow). Residues showing significant chemical shift perturbation are represented in sticks and are coloured in blue.

Three-dimensional model of PHD_VC5HCH_NSD1/C2HR_Nizp1 complex

Although most of the backbone resonances assignment of bound PHD_VC5HCH_NSD1 and C2HR_Nizp1 was accomplished, due to unfavourable dynamic regime and to sample precipitation and degradation at concentrations beyond 0.3 mM, it was not possible to detect intermolecular NOEs to determine the complex structure using the standard NOE-based approach. Thus, we took advantage of the well-established HADDOCK 2.0 software (51) to obtain a structural model of the complex, exploiting experimental CSP data, as described in Materials and Methods. Docking models were clustered based on the interface r.m.s.d. resulting in 11 clusters (Figure 6A, Supplementary Table S2). Importantly, in all the clusters C2HR_Nizp1 docks inside PHD_VC5HCH_NSD1 placing its α helix in the interdomain groove formed by PHD_V and C5HCH (Figure 6B). In particular in Cluster 1, showing the best HADDOCK score (Figure 6A, Supplementary Table S2), C2HR_Nizp1 interacts with PHD_VC5HCH_NSD1 inserting its W416_Nizp1 side chain into the hydrophobic pocket generated by H2205_NSD1, F2122_NSD1, M2177_NSD1, P2179_NSD1, R415_Nizp1 and V418_Nizp1 (Figure 6C-left). A set of polar contacts involving the Arginines of the RWR signature seems to contribute to the binding: the guanidinium group of R417_Nizp1 forms a salt-bridge with the carboxylate of E2204_NSD1; moreover R415_Nizp1 establishes electrostatic interactions with the carboxylate of E2120_NSD1 through its guanidinium group and forms through its amide proton a hydrogen bond with the carbonyl of S2123_NSD1 (Figure 6C, middle). Finally, R422_Nizp1 points toward the interdomain groove, interacting with L2147_NSD1 and W2157_NSD1 (Figure 6C right)_. All the residues establishing favourable intermolecular interactions were consistent with the observed chemical shift perturbations. In an effort to assess the validity of our HADDOCK model we made a series of single point mutations around the binding interface of both PHD_VC5HCH_NSD1 and C2HR_Nizp1 to investigate their effect in NMR and/or calorimetric titrations (Table 3). In particular, mutation of the exposed W416_Nizp1 into Alanine had a very strong impact, with little or virtually no detectable binding in NMR titrations, implying almost complete abrogation of the interaction (Supplementary Figure S5B, Table 3). This result strongly supports the role of hydrophobic interaction in this binding process, in agreement with the entropically driven reaction, most likely deriving from the desolvation of the intermolecular binding interface. Both R417_Nizp1 and E2204_NSD1 mutations into Alanine resulted in one order of magnitude reduction in binding affinity and a reduction of the average chemical shifts displacements (Table 3) corroborating the presence of a salt-bridge between these two residues (Figure 6C, left). Mutation of R415_Nizp1 did not have a relevant impact in binding, suggesting that the major interaction with PHD_VC5HCH_NSD1 occurs via a hydrogen bond between its amide and the carbonyl of S2123_NSD1, in line with the extreme down-shift of S2123_NSD1 amide proton resonance upon binding (Figures 3C and 6C, middle). R422A_Nizp1, with only a 2-fold increase in the Kd, had a minor impact in binding (Table 3), implying a secondary role for this residue in substrate recognition. Unfortunately, we could not assess the role of E2120_NSD1, F2122_NSD1 and S2123_NSD1 in binding as their mutations into Alanine caused low expression yields and partial PHD_VC5HCH_NSD1 unfolding. Mutation of the R427_Nizp1 within the C2HR signature did not influence binding, in agreement with both the model and the chemical shift mapping which did not highlight any relevant role of this residue in the binding. We also verified the impact of fold-preserving Sotos syndrome point mutations in the binding to C2HR_Nizp1. As expected, R2152Q_NSD1 mutation, located on the opposite surface with respect to C2HR_Nizp1 binding site, did not affect complex formation (Table 3). Conversely, H2205R_NSD1 mutation, located near the binding site identified by CSP, reduced the binding of almost one order of magnitude, probably because of repulsive interactions with R417_Nizp1 and of the loss of stabilizing hydrophobic interactions between H2205_NSD1 and W416_Nizp1 (Figure 6C, left). Based on these findings and on the repressive role attributed to Nizp1 and NSD1 interaction (7), it is conceivable that Sotos syndrome mutations targeting PHD_VC5HCH_NSD1 and hampering interaction with Nizp1 might have an impact on the repression of growth-promoting genes, eventually resulting in overgrowth disease.

Figure 6.

HADDOCK model of PHD_VC5HCH_NSD1 and C2HR_Nizp1. (A) HADDOCK score versus r.m.s.d. from the lowest energy complex structure in terms of HADDOCK score (a.u.). Circles correspond to the best five structures of each cluster, black squares correspond to the cluster averages with the standard deviation indicated by bars. (B) Representative pose of the best docking model in terms of HADDOCK score. PHD_VC5HCH_NSD1 and C2HR_Nizp1 are shown as surface and cartoon, respectively. Active residues are shown in hot pink and yellow sticks in PHD_VC5HCH_NSD1 and C2HR_Nizp1, respectively. (C) Details of the relevant interactions between PHD_VC5HCH_NSD1 and C2HR_Nizp1. Hydrophobic and electrostatic interactions between RWR_Nizp1 loop and PHD_vC5HCH_NSD1 are shown on the left and central panels, respectively. Interactions of R422_Nizp1 are shown on the right panel. Hydrophobic interactions are highlighted with dotted space-filling representation and polar interactions with dotted lines. Zn²⁺ ions are shown as grey (PHD_VC5HCH_NSD1) and gold spheres (C2HR_Nizp1)

Table 3. Thermodynamic parameters, stoichiometry (n) and dissociation constants (Kd) between PHD_VC5HCH_NSD1 and C2HR_Nizp1 wild type (WT) and mutants measured by ITC (T = 22°C).

PHDvC5HCHNSD1	C2HRNizp1	ΔH (kcal/mol)	-TΔS (kcal/mol)	ΔG (kcal/mol)	n	Kd (μM)
WT	WT	4.1 ± 0.1	11.4± 0.1	-7.3 ± 0.1	1.1 ± 0.1	3.8 ± 0.7
R2152Q	WT	3.5 ± 0.2	10.7 ± 0.1	-7.2 ± 0.1	1.0 ± 0.1	4.9 ± 0.6
E2204A	WT	0.4 ± 0.1	6.3 ± 0.2	-5.9 ± 0.2	0.9 ± 0.1	45.2 ± 6.9
H2205R	WT	1.0 ± 0.1	7.2 ± 0.1	6.2 ± 0.1	1.0 ± 0.1	27.2 ± 5.3
WT	R415A	2.2 ± 0.1	9.3 ± 0.1	-7.1 ± 0.1	1.1 ± 0.1	6.1 ± 0.4
WT	W416A	N.D.	N.D.	N.D.	N.D.	N.D.
WT	R417A	1.5 ± 0.1	7.5 ± 0.1	6.0 ± 0.1	1.1 ± 0.1	36.0 ± 6.0
WT	R422A	0.7 ± 0.1	7.7 ± 0.1	7.0 ± 0.1	0.9 ± 0.1	7.0 ± 0.2
WT	R427A	1.7 ± 0.1	9.0 ± 0.1	-7.3 ± 0.1	1.0 ± 0.1	4.0 ± 0.3

Biophysical assessment of PHD_VC5HCH_NSD1 interaction with histone H3 N-terminal tail peptides

Until now biochemical assays aiming at the in vitro assessment of the effective histone binding ability of PHD_VC5HCH_NSD1 were not conclusive. On one hand recent pulldown assays of GST-PHD_VC5HCH_NSD1 with unfractionated calf thymus histones or with biotinylated histone H3 peptides did not provide evidence of histone binding (26). On the other hand histone binding assays using MODified Histone Peptide Array TM (Activ Motif) (Supplementary Figure S6A,B) were in line with previous pull-down assays of GST-PHD_VC5HCH_NSD1 with modified biotinylated peptides (21 aminoacids long) (25), suggesting an interaction preference for H3K4me3 and H3K9me3. Because of the qualitative nature of biochemical binding assays, that are often prone both to false positive and negative results (66–68), we biophysically verified this putative interaction using NMR CSP. At variance to what observed in titrations with C2HR_Nizp1, addition of up to 12 molar excess of unmodified or modified histone peptides (H3_1–21, H3K4me3_1–21 and H3K9me3_1–21) induced in PHD_VC5HCH_NSD1 ¹H–¹⁵N HSQC spectrum very small changes in the fast exchange regime, suggestive of extremely weak interactions (with an average CSP value <CSP> of ∼0.02 ppm) (Figure 7A, Supplementary Figure S6C). As the binding curves did not reach saturation, it was not possible to reliably fit the binding curves obtained during the individual titrations, implying affinities in the high mM range, i.e. orders of magnitude higher with respect to the micromolar affinity usually observed in PHD-histone tail interactions (69,70). Accordingly, isothermal titration calorimetry experiments did not develop sufficient heat to allow Kd determination (data not shown). Notably, in all the titrations, irrespectively of the peptides modification status, residues exhibiting significant amide chemical shift (CSP > <CSP> + 1σ) did not involve the classical histone binding surface, composed by the N-terminus of the PHD finger, the first β-strand, and the conserved PXGXW binding pocket. Peaks significantly shifting upon H3 peptides addition belonged to residues located on both PHD_V and C5HCH (D2119_NSD1, E2120_NSD1, S2123_NSD1, C2124_NSD1, T2150_NSD1, K2151_NSD1, W2157_NSD1, E2158_NSD1, H2162_NSD1, C2164_NSD1, D2165_NSD1, S2181_NSD1, H2205_NSD1), mainly, at the interface between PHD_V and C5HCH, on the negatively charged surface of the tandem domain (Figures 1C and 7B). Of note, these residues did not form a continuous interaction surface when mapped on the surface of the tandem domain (Figure 7B). Intriguingly, NMR titrations performed reducing or increasing the number of positively charged residues using shorter (H3₁–₁₀) or longer peptides (H3₁–₃₆), did not change the chemical shift differences profiles, but rather resulted in smaller and higher chemical shifts displacements, respectively (Supplementary Figure S6D). Taken together, these observations strongly support the non-specific electrostatic nature of PHD_VC5HCH_NSD1 histone H3 tail interaction. Indeed, titration of the domain with up to 300 mM of Lysine induced amide chemical shifts in the same region (Supplementary Figure S6D). Overall, in contrast to what observed in previous binding assays with biotinylated histone peptides (25), but in agreement with recent GST-pulldown experiments with unfractionated calf thymus histones and biotinylated histone peptides (26), our biophysical data exclude a role of PHD_VC5HCH_NSD1 in decoding methylated/unmethylated H3K4 and H3K9 epigenetic marks. We cannot rule out that, despite the extensive modification coverage guaranteed by the peptides arrays, PHD_VC5HCH_NSD1 has an as-of-yet unidentified histone modification(s) specificity. It is conceivable that the specific recruitment of full-length NSD1 to chromatin might be mediated by the combinatorial interactions of the other chromatin readers (four PHD fingers and two PWWP) present in the full-length protein. Herein, the acidic surface of PHD_VC5HCH_NSD1 might work as an electrostatic platform reinforcing the interaction with the basic histone tails. Indeed, it is not unusual that isolated readers displaying in vitro weak or non-specific interactions with their histone peptide target, in the context of the full length protein reach physiologically relevant binding affinities range, thanks to multivalent interactions and/or contacts using scaffolding domains that bridge their host proteins with other subunits of the chromatin bound complex (62). As the C2HR_Nizp1 binding site involves PHD_VC5HCH_NSD1 interdomain space, we wondered whether the weak interaction with the histone H3 N-terminal tail could compete with complex formation. The ¹H–¹⁵N HSQC spectra of both ¹⁵N-PHD_VC5HCH_NSD1/C2HR_Nizp1, and of ¹⁵N-C2HR_Nizp1/PHD_VC5HCH_NSD1 (1:1) did not show any substantial change upon addition of an excess of H3_1–21 peptide (1:12), excluding a competing influence of the histone H3 tail on PHD_VC5HCH_NSD1/C2HR_Nizp1 interaction (Supplementary Figure S5A).

Figure 7.

Mapping of the interaction with histone H3 amino-terminal peptides. (A) Histograms showing the averaged backbone chemical shift perturbations (CSP) observed in ¹H–¹⁵N HSQC of PHD_VC5HCH_NSD1 (0.2 mM) upon addition of a twelve-fold excess of H3_1–21, H3K4me3_1–21, H3K9me3_1–21. The last two bars in the plot correspond to the indole amides of W2157_NSD1 and W2168_NSD1, respectively. The dotted line represents <CSP>, the dashed line represents <CSP> + 1σ, where σ corresponds to standard deviation. Missing bars correspond to either Prolines or amides that are not visible because of exchange with the solvent. (B) Surface and cartoon representation of PHD_VC5HCH_NSD1. Residues showing significant chemical shift perturbation (CSP > <CSP>+1σ) are represented in cyan.

Functional divergence of the PHD_VC5HCH domain within the NSD family

Despite the high sequence similarity shared by the PHD_VC5HCH domain within the NSD family (PHD_VC5HCH_NSD1 shares 61% and 64% identity with PHD_VC5HCH_NSD2 and PHD_VC5HCH_NSD3, respectively) (Supplementary Figure S1A), they appear to have evolved to exert different functions. As a matter of fact, PHD_VC5HCH_NSD1 does not function as classical epigenetic reader, whereas PHD_VC5HCH_NSD2 and PHD_VC5HCH_NSD3 preferentially recognize H3K4me0 (demonstrated only biochemically) and H3K9me3 marks, respectively (26). Based on sequence comparison and on the available structural knowledge (this study and the 3D structure of the PHD_VC5HCH_NSD3/H3K9me3 complex (26)), we hypothesized that PHD_VC5HCH_NSD1 inability to efficiently recognize the positively charged histone H3 tail could be ascribable to electrostatic repulsions, generated by two basic residues (R2117_NSD1, K2134_NSD1) on the canonical histone interaction surface (Supplementary Figures S1A, S3D). Notably, the equivalent positions in NSD2 and NSD3 are occupied by a neutral aminoacid and by an acidic residue forming a salt-bridge with H3R2, as highlighted by the PHD_VC5HCH_NSD3/H3K9me3 crystallographic structure (Supplementary Figure S3D). However, simultaneous mutations of these basic residues (R2117A_NSD1, K2134D_NSD1) aiming at converting PHD_VC5HCH_NSD1 into a classical histone H3 reader, failed to induce the expected functional twist (Supplementary Figure S6E). Collectively, these results suggest that a more complex combination of amino-acids substitutions, possibly related to yet unidentified conformational and dynamic effects might dictate the different histone recognition propensities within the NSD family. Following the same line, PHD_VC5HCH domain functional divergence seems also to reflect into different C2HR_Nizp1 binding abilities attributed to the three NSD proteins. In fact yeast two hybrid screenings on both C5HCH_NSD2 and C5HCH_NSD3 (71) along with GST-pulldown assays on PHD_VC5HCH_NSD3 (26) did not support their direct interaction with C2HR_Nizp1. Conceivably, also in this case a combination of dynamical effects and of minor sequence variations targeting the interaction surface might account for possible differences in C2HR_Nizp1 binding specificity within the NSD family. Future structural, dynamical and biophysical studies i. investigating the effective binding ability of the other NSD members to C2HR_Nizp1 and ii. verifying the putative histone H3 interaction with PHD_VC5HCH_NSD2 are warranted to complete our picture on the structure-function relationship within the PHD_VC5HCH-NSD family.

Concluding remarks

In the last years, PHD fingers, alone or in tandem with other chromatin binding domains, have emerged as multifaceted interaction platforms, whose functions go far beyond the perceiving of the epigenetic landscape (62,63). As a matter of fact PHD fingers have been shown to work often as scaffolding domains bridging their host proteins with other subunits of bigger macromolecular complexes (62,63,72). Herein, PHD_VC5HCH_NSD1 seems to represent a paradigmatic example for the PHD domain functional versatility: on one hand it has been proposed to decode the methylation status of histone H3K4 and H3K9 (25), on the other hand it has been shown to mediate the recruitment of NSD1 to reinforce Nizp1 transcriptional repressor activity via its direct binding to C2HR_Nizp1 Zinc-finger domain (25,29,73). Notably, structural and biophysical studies proving all these interactions and providing molecular details on them were lacking. Here, we have shown that the tandem domain PHD_VC5HCH_NSD1 forms an indivisible structural entity that interacts only very weakly with histone H3 N-terminal tail peptides. The binding occurs via non-specific electrostatic forces, irrespectively of histone H3 methylation and involves a distinct surface with respect to the classical histone H3 tail binding groove. While the role of PHD_VC5HCH_NSD1 as classical epigenetic reader is highly questionable, ITC and NMR CSP experiments clearly demonstrate a finger-finger micromolar interaction between C2HR_Nizp1 and PHD_VC5HCH_NSD1. Via a data driven docking model, supported by site-directed mutagenesis, we have provided first molecular insights into the recruitment of the co-repressor NSD1 to the repressor Nizp1 via PHD_VC5HCH_NSD1 and a solvent exposed RWR motif of C2HR_Nizp1. In particular, the RWR loop of Nizp1 accommodate into the PHD_VC5HCH_NSD1 interdomain groove establishing hydrophobic and electrostatic interactions. Interestingly, both C5HCH_NSD1 and C2HR_Nizp1 domains have been acquired late in evolution (7) and their high conservation in vertebrates support an evolutionary preservation of this protein-protein interaction (Supplementary Figure S1). Recruitment of co-repressors through short sequence motifs is a recurrent theme in transcriptional repression, as observed for example in the engagement of Groucho corepressor via the WPRW motif of Drosophila transcriptional repressors Dorsal and Hairy (74). The prominent functional role exerted by the RWR_Nizp1 signature within the C2HR_Nizp1 domain strengthens also the notion that this atypical Zinc-finger motif is more than just a degenerate evolutionary vestige of the classical Zinc-finger C2H2 motif (7). This is in keeping with previous studies showing that Nizp1 transcriptional repressor activity depends on C2HR domain integrity, as fold destroying mutations abolished its binding to NSD1, thus reducing transcriptional repression (7). Importantly, in the Sotos syndrome context we have shown that the fold preserving mutation H2205R_NSD1 also diminishes the interaction with C2HR_Nizp1 supporting a mechanistic scenario in which impairment of the repressive NSD1/Nizp1 complex might alter the physiological down-regulation of growth-promoting genes, thus resulting in the overgrowth phenotype, typical of Sotos syndrome. In this context, it is conceivable that other nuclear proteins besides Nizp1 bind to NSD1, whereby the PHD finger domains, in line with their functional versatility, might mediate additional protein interactions. Future studies dedicated to NSD1/Nizp1 dependent transcriptome and interactome in normal and pathological conditions will be instrumental in understanding the patho-physiological role of NSD1/Nizp1 interaction and in devising new strategies for successful treatments in overgrowth diseases.

ACCESSION NUMBERS. DB accession codes: 2naa; 2nab BioMagResBank 25933; 25934.

SUPPLEMENTARY DATA

Supplementary Data are available at NAR Online.

FUNDING

Italian Ministry for Health [RF-2013–02354880 to G.M and G.T.]; Italian Association Cancer Research [AIRC-13149 to G.M.]; Fondazione Telethon [ TCP99035 to G.M.]; Fondazione Veronesi [to F.M.G.]; Marie Curie Co-funding of Regional, National and International Programmes [PCOFUND-GA-2010-267264 INVEST to M.A.C.R.]; Fondazione Cariplo [2009-3577 to M.G.]. Funding for open access charge: Italian Ministry for Health [RF-2013-02354880] or Italian Association Cancer Research [AIRC-13149].

Conflict of interest statement. None declared.