Pharmacological targeting of a PWWP domain demonstrates cooperative control of NSD2 localization

NSD2 is the primary enzyme responsible for the dimethylation of lysine 36 of histone 3 (H3K36me2), a mark associated with active gene transcription and intergenic DNA methylation. In addition to a methyltransferase domain, NSD2 harbors two PWWP and five PHD domains believed to serve as chromatin reading modules, but their exact function in the regulation of NSD2 activity remains underexplored. Here we report a first-in-class chemical probe targeting the N-terminal PWWP (PWWP1) domain of NSD2. UNC6934 binds potently (Kd of 91 ± 8 nM) to PWWP1, antagonizes its interaction with nucleosomal H3K36me2, and selectively engages endogenous NSD2 in cells. Crystal structures show that UNC6934 occupies the canonical H3K36me2-binding pocket of PWWP1 which is juxtaposed to the DNA-binding surface. In cells, UNC6934 induces accumulation of endogenous NSD2 in the nucleolus, phenocopying the localization defects of NSD2 protein isoforms lacking PWWP1 as a result of translocations prevalent in multiple myeloma. Mutation of other NSD2 chromatin reader domains also increases NSD2 nucleolar localization, and enhances the effect of UNC6934. Finally we identified two C-terminal nucleolar localization sequences in NSD2 that appear to drive nucleolar accumulation when one or more chromatin reader domains are disabled. These data support a model in which NSD2 chromatin engagement is achieved in a cooperative manner and subcellular localization is controlled by multiple competitive structural determinants. This chemical probe and the accompanying negative control, UNC7145, will be useful tools in defining NSD2 biology.


Introduction
Nuclear receptor-binding SET domain-containing 2 (NSD2, also known as MMSET and WHSC1) is a protein lysine methyltransferase that belongs to the NSD family, which also includes NSD1 and NSD3.
Functionally, NSD2 is responsible for the bulk of H3K36me2 in diverse cell types. Dimethylation of H3K36 by both NSD1 and NSD2 recruits DNMT3a at intergenic regions to control DNA methylation and regulate development and homeostasis 8,9 . NSD2 is also required for efficient non-homologous end-joining and homologous recombination, two canonical DNA repair pathways 10,11 .
In addition to its catalytic domain, NSD2 has multiple protein-protein interaction (PPI) modules with known or potential chromatin reading functions, including five PHD (plant homeodomain) and two PWWP (proline-tryptophan-tryptophan-proline) domains 2 , as well as a putative DNA-binding HMG-box (high mobility group box) domain (Fig. 1a). Mounting evidence suggests that these domains play important roles in NSD2 function, but the individual and/or collective roles of the NSD2 chromatin reader domains are still being elucidated 2 . Many PWWP domains are known H3K36me2,3 reading modules that engage methyl-lysine while simultaneously interacting with nucleosomal DNA adjacent to H3K36 12,13 .
The isolated N-terminal PWWP domain of NSD2 (NSD2-PWWP1) binds H3K36 di-and trimethylated nucleosomes; this interaction presumably is mediated by a conserved aromatic cage and stabilizes NSD2 at chromatin 8 . Mutation of the aromatic cage residues abrogates NSD2-PWWP1 binding to nucleosomal H3K36me2, but has only modest effect on global H3K36 methylation level in cells 8 . However, H3K36 methylation has been shown to be abolished upon mutation of the second PHD domain (PHD2) 14 .
Chromatin association is preserved with the PHD2 mutant but lost upon combined truncation of PWWP1, PHD1, PHD2 and part of PHD3 (i.e. the RE-IIBP isoform) due to cytoplasmic retention 14 . Conversely, nucleolar accumulation was observed in the case of NSD2-PWWP1 truncated variants 15,16 . Overall, these data suggest a complex regulatory system for NSD2 in which distinct combinations of structural modules contribute to subcellular localization, substrate engagement, and lysine methylation.
Due to their role in disease and biology, there has been much recent interest in targeting the NSD family of methyltransferases with small molecules. High-quality, cell-active, and selective inhibitors of NSD2 and NSD3 catalytic activity remain elusive; however, irreversible small molecule inhibitors of the NSD1 SET domain that demonstrate on target activity in NUP98-NSD1 leukemia cells have recently been reported 17 . We have had recent success targeting the PWWP domains of NSD3 and NSD2 suggesting that PWWP domains are druggable, while the PHD domains have so far not been targetable. Specifically, we reported a potent chemical probe targeting the N-terminal PWWP domain of NSD3 that repressed MYC mRNA levels and reduced the proliferation of leukemia cell lines 18 . Additionally, we recently described the development of the first antagonist of NSD2-PWWP1 which binds with modest potency and abrogates H3K36me2 binding 19 . Therefore, we hypothesized that targeting the PWWP domain(s) of NSD2 with highly potent and selective chemical probes may be a strategy to modulate NSD2 engagement with chromatin, subcellular localization, and/or catalytic function.
In this study, we report a first-in-class chemical probe, UNC6934, that selectively binds in the aromatic cage of NSD2-PWWP1, thereby disrupting its interaction with H3K36me2 nucleosomes. UNC6934 potently and selectively binds full-length NSD2 in cells and induces partial disengagement from chromatin, consistent with a cooperative chromatin binding mechanism relying on multiple protein interfaces. UNC6934 promotes nucleolar localization of NSD2, phenocopying previously characterized PWWP1-disrupting mutations prevalent in t(4;14) multiple myelomas 15,16 . Furthermore, we identified two active nucleolar localization sequences in NSD2 and demonstrated cooperativity between multiple chromatin reader modules to prevent nucleolar sequestration. Our data demonstrate that UNC6934 is a potent and selective drug-like molecule suitable as a high-quality chemical probe to interrogate the function of NSD2-PWWP1.  19 led to MRT866 and finally the chemical probe UNC6934. UNC7145 is a structurally similar negative control compound.

Discovery of a potent ligand targeting NSD2-PWWP1
We recently described the use of virtual screening, target class screening, and ligand-based scaffold hopping approaches to identify ligands of the NSD2 PWWP domains as starting points for further development 19 . This initial effort led to compound 3f which binds NSD2-PWWP1 with a K d of 3.4 ± 0.4 µM as determined by surface plasmon resonance (SPR). Based on the crystal structure of 3f in complex with NSD2 (PDB 6UE6) molecular docking simulations predicted that a benzoxazinone bicyclic ring would favorably replace the cyanophenyl group of 3f. We confirmed that compound MRT866 binds NSD2-PWWP1 with a K d of 349 ± 19 nM (Fig. 1b, Supplementary Fig. 1) and occupies the aromatic cage of PWWP1 similarly to compound 3f as determined by x-ray crystallography (PDB 7LMT, Supplementary Fig.1 and Supplementary Table 1). The benzoxazinone ring of MRT866 makes more extensive Van der Waals interactions with NSD2-PWWP1 than 3f and engages in an additional hydrogen bond with the side-chain of Q321. Further structure-based optimization focused on the replacement of the thiophene ring and resulted in UNC6934 which binds NSD2-PWWP1 with a K d of 91 ± 8 nM by SPR ( Fig. 1b and 2a). Interestingly, conversion of the cyclopropyl group of UNC6934 to an isopropyl moiety  (Fig. 1b) resulted in no appreciable binding up to 20 µM and therefore UNC7145 is an ideal negative control compound. To confirm that UNC6934 is binding in the methyl-lysine binding pocket of NSD2-PWWP1, we generated NSD2-PWWP1 with a key aromatic cage mutant (F266A). In contrast to wild type protein, UNC6934 did not result in a significant thermal stabilization of the NSD2-PWWP1 F266A mutant (Supplementary Fig. 2). Furthermore, UNC6934 is selective for NSD2-PWWP1 over 14 other human PWWP domains as assessed by differential scanning fluorimetry (DSF) (Fig. 2b) and did not inhibit any of a panel of 33 methyltransferase domains including the H3K36 methyltransferases, NSD1, NSD2, NSD3, and SETD2 (Supplementary Fig. 3). UNC7145 was similarly inactive against all PWWP domains and methyltransferases tested (Fig. 2a, Supplementary Fig. 3).  NSD2-PWWP1 is postulated to stabilize the binding of NSD2 on chromatin, primarily through the recognition of H3K36me2 8 . Therefore, we next used an AlphaScreen-based proximity assay to investigate the effect of UNC6934 on the interaction of NSD2 with recombinant semi-synthetic designer nucleosomes (dNucs) 9 . We first evaluated His-tagged NSD2-PWWP1 binding to a range of lysinemethylated semi-synthetic dNucs (me0, me1, me2, and me3 at H3K4, H3K9, H3K27, H3K36, and H4K20), and confirmed that NSD2-PWWP1 binds di-and tri-methyl H3K36 with a preference for the former (Supplementary Fig. 4), as previously reported 8 . UNC6934 disrupted the interaction between NSD2-PWWP1 and nucleosomal H3K36me2 in a dose-dependent manner with an IC 50 of 104 ± 13 nM, while UNC7145 had no measurable effect (Fig. 2c).
To determine whether the PWWP1 domain is necessary for mediating the interaction between full-length NSD2 (fl-NSD2) and nucleosomes, we similarly tested the effect of UNC6934 on fl-NSD2 binding to nucleosomal H3K36me2 in the AlphaScreen assay. Unlike with NSD2-PWWP1, we found that UNC6934 was unable to disrupt the interaction between fl-NSD2 and nucleosomal H3K36me2 (Fig. 2d). With H3K36me2 being proximal to nucleosomal DNA, we reasoned that electrostatic interactions between fl-NSD2 and DNA may abrogate disengagement of the protein by UNC6934. To test this hypothesis, we repeated the experiment in the presence of an excess of salmon sperm DNA (SSD), which is a commonly used blocker of non-specific DNA interactions. Under these conditions, UNC6934 could effectively disengage fl-NSD2 from nucleosomal H3K36me2 (IC 50 = 78 ± 29 nM) (Fig. 2e). Together, these results indicate that NSD2 binding to nucleosomes is multivalent, and therefore UNC6934 can disengage fl-NSD2 from H3K36me2-modified nucleosomes only in the presence of excess competitive DNA.

UNC6934 occupies the H3K36me2 binding pocket of NSD2-PWWP1
PWWP domains have both a methyl-lysine-binding pocket and a DNA-binding surface 12 . To better understand these interactions, we solved the crystal structures of NSD2-PWWP1 in complex with Table 1). While DNA binds a basic surface area where the side-chains of K304, K309 and K312 are engaged in direct electrostatic interactions with the DNA phosphate backbone ( Fig. 3a, PDB 5VC8), UNC6934 occupies the canonical methyl-lysine-binding pocket adjacent to the DNA binding surface in an arrangement where the cyclopropyl ring is deeply inserted in the aromatic cage composed of Y233, W236 and F266 (Fig. 3b, c; PDB 6XCG). The extremely tight fit at the aromatic cage rationalizes the lack of binding of the corresponding negative control, UNC7145, where a bulkier isopropyl replaces the cyclopropyl group (Fig. 1b). The non-bonded carbons of an isopropyl group are separated by ~2.5 Å (ex: PDB 1NA3), while the corresponding atoms are 1.5Å apart in the cyclopropyl ring of UNC6934. The structure further explains the observed loss of binding of UNC6934 to the F266A NSD2-PWWP1 mutant (Supplementary Fig. 2). Unlike the DNA binding surface, the UNC6934 binding pocket is mildly electronegative, as would be expected for a site accommodating a positively charged methyl-lysine side-chain. Interestingly, the pocket is partly occluded in the apo structure and undergoes a conformational rearrangement of three side-chains (Y233, F266, E272) upon UNC6934 binding (Fig.   3d). Overall, our structural data confirms that UNC6934 competes directly with H3K36me2 for binding to NSD2-PWWP1 to disrupt high-affinity binding to H3K36me2 nucleosomes. Our structural data also helps to explain the exquisite selectivity of UNC6934 for NSD2-PWWP1 over other PWWP domains. Mapping the side-chains positioned within 5 Å of the bound ligand in our cocrystal structure onto a multiple sequence alignment of all human PWWP domains identified the degree ed s a ig.

UNC6934 (Supplementary
ng A tic nd e: ate er ee of conservation among binding pocket residues exploited by UNC6934. We find that the binding pocket of NSD3-PWWP1 is by far the closest to that of NSD2-PWWP1, as only three of the side-chains lining the binding pocket are not conserved between the two proteins ( Supplementary Fig. 5). In NSD2, G268 is at the bottom of a cavity accommodating the benzoxazinone of UNC6934 and replacing this residue with the serine of NSD3 would be expected to occlude ligand binding. This is consistent with the ability of UNC6934 to stabilize NSD2-PWWP1, but not NSD3-PWWP1 or any of the other PWWP domains in a thermal shift assay (Fig. 2b). These data strongly suggest that UNC6934 is likely selective for NSD2-PWWP1 versus any other human PWWP domain. Further supporting the overall selectivity of the compound, UNC6934 and UNC7145 were profiled against a set of 90 central nervous system receptors, channels, and transporters (Supplementary Table 2). Of those that were inhibited by UNC6934 greater than 50% at 10 µM (2 in total), the human sodium-dependent serotonin transporter receptor was the only protein inhibited by UNC6934 with a measurable inhibitory constant (Ki = 1.4 ± 0.8 µM).

UNC6934 selectively engages NSD2-PWWP1 in cells
To profile ligand selectivity and target engagement in a cellular context, we synthesized UNC7096, a biotin-labeled affinity reagent containing a close analog of UNC6934 for chemical pulldown experiments ( Fig. 4a). We first verified that UNC7096 is a high-affinity NSD2-PWWP ligand by SPR, measuring a K d of 46 nM, comparable to UNC6934 (Supplementary Fig. 6). UNC7096 efficiently enriched both NSD2-Short (MMSETI) and NSD2-Long (MMSETII) isoforms from KMS11 whole cell lysates as determined by western blotting (Fig. 4b). Chemiprecipitation of NSD2 by UNC7096 could also be blocked by pre-incubation of KMS11 lysates with 20 µM of UNC6934, but not the negative control compound UNC7145 (Fig. 4b). Label-free proteomic analysis of the UNC7096 pulldown experiments identified NSD2 as the only protein significantly depleted by competition with UNC6934 (Fig. 4c), whereas pre-incubation of lysates with UNC7145 did not considerably alter the enrichment profile.

Perturbation of NSD2's Chromatin Binding Domains Promotes Localization to the Nucleolus
Reader domains are critical for the recruitment and positioning of epigenetic proteins at defined loci across the genome and small molecule antagonism of reader domains is known to alter the localization of chromatin-associated target proteins 21,22 . We therefore reasoned that PWWP1 antagonism by UNC6934 may affect NSD2 localization within the nucleus. To test this hypothesis, we used confocal microscopy to evaluate the localization of endogenous NSD2 in U2OS cells treated for four hours with 5 µM UNC6934 or negative control UNC7145 (Supplementary Fig. 8). Upon treatment with UNC6934, we observed an increase in NSD2 signal within nucleoli, the sub-nuclear membrane-less organelles that house ribosomal DNA for pre-ribosomal transcription and processing. Interestingly, t(4;14) chromosome translocations in multiple myeloma, which juxtapose the IgH enhancer and NSD2 promoting overexpression of NSD2, can result in truncation/inactivation of PWWP1 and nucleolar enrichment, suggesting that the PWWP1 domain contributes to the exclusion of NSD2 from the nucleolus 15,16 . Therefore, UNC6934 appears to phenocopy NSD2 N-terminal PWWP1 truncations while UNC7145 has no effect.
To validate this result, we repeated the imaging experiments while co-staining for the nucleolar marker fibrillarin and measured the extent of NSD2 and fibrillarin co-localization using Pearson correlation ( Fig.   5a, b). The Pearson correlation coefficient (PCC) is a common statistic used to describe co-localization; it measures the correlation in signal intensity between two fluorescent molecules (values range between -1 and 1 representing no correlation and absolute correlation, respectively). We observed a significant increase in the correlation between NSD2 and fibrillarin signal in response to UNC6934, confirming an increase in the nucleolar localization of NSD2. These data support the engagement of endogenous NSD2 by UNC6934 in cells and suggest that loss of H3K36me2 binding by NSD2-PWWP1 promotes nucleolar accumulation of NSD2. While we found no significant effect on ribosomal RNA transcription in response to UNC6934 (Supplementary Fig. 9), we do observe a steady-state pool of nucleolar NSD2 that is sensitive to RNA polymerase I inhibition (actinomycin D; 50 nM) or genotoxic agents (doxorubicin; 1 µM), conditions known to significantly alter the protein composition of nucleoli 23,24 (Supplementary Fig.   10). Overall, these results indicate that UNC6934 mediated antagonism of PWWP1 leads to accumulation of endogenous NSD2 in the nucleolus. To test if subnuclear localization is exclusively mediated by the PWWP1 domain, we mutated GFPtagged NSD2 at several critical sites within distinct chromatin-recruitment modules. This included an Nterminal short linear motif that engages BET proteins (K125A) 21,25 , two key aromatic cage residues in the PWWP1 domain (W236A and F266A) 12 , two PHD2 mutants that disrupt recruitment to target loci and disable H3K36me2 methyltransferase activity in cells (H762R and H762Y) 14 , a presumptive inactivating aromatic cage mutation in PWWP2 (W894A), and a catalytic-dead mutant of the methyltransferase domain (Y1092A) 1 . We found that disruption of any one of the canonical reader domains (PWWP1, PHD2, and PWWP2) promoted enrichment of NSD2 in nucleoli ( Fig. 6a, b). These observations suggest that it is the loss of chromatin binding that leads to the nucleolar retention of NSD2, and that multiple NSD2 reader domains cooperate to maintain appropriate nuclear sublocalization. (c) Assessing domain cooperativity by treating NSD2-GFP point mutants with DMSO control, 5 µM UNC7145 or 5 µM UNC6934 and measuring co-localization by PCC (n=5, significant p-values derived from a Welch's unpaired t-test compared to the DMSO control for each panel are indicated in order as ** = 0.0059, **** = 8.7 x 10 -5 , ** = 0.0062). (d) Computational prediction of Nucleolar Localization Sequences in NSD2 using the Nucleolar localization sequence Detector algorithm (NOD) 26 . (e) Representative fluorescent images of cells expressing GFP tagged with putative nucleolar localization sequences from NOD.
We next used UNC6934 to test the cooperativity of the NSD2 reader domains towards nucleolar localization. To do so, we treated cells transfected with RFP-fibrillarin and GFP-tagged NSD2 (wild-type, W236A, H762R, or W894A) for four hours with 5 µM UNC6934 or UNC7145. In cells expressing wildtype NSD2-GFP we observed an increase in nucleolar signal upon treatment with UNC6934 .
Importantly, while the PWWP1 aromatic cage mutant (W326A) had a higher baseline nucleolar localization compared to WT, no further increase was observed upon UNC6934 treatment, again supporting PWWP1-dependent activity of the probe (Fig. 6c). However, in cells expressing the PHD2 and PWWP2 mutants, we observed an additive effect with UNC6934 treatment. In addition to higher baseline nucleolar localization due to their respective mutations, there was a further increase in nucleolar colocalization upon UNC6934 treatment compared to both the DMSO control and the negative control UNC7145. These observations support a model in which NSD2 reader domains act cooperatively in recruiting NSD2 to chromatin and preventing nucleolar sequestration. Antagonizing the interaction between NSD2-PWWP1 and H3K36me2 may therefore not be sufficient to fully disengage full-length NSD2 from chromatin. Indeed, we only observed a modest release of full-length NSD2 from chromatin upon treatment with UNC6934 in cell fractionation experiments, as well as no changes in global H3K36me2 levels or the proliferation of KMS11 t(4;14) multiple myeloma cells grown on bone marrow stroma in response to UNC6934 (Supplementary Fig. 11).
To further define the features of NSD2 that drive its nucleolar targeting, we used NoD (Nucleolar localization sequence Detector) 27 to computationally predict putative nucleolar localization sequences (NoLS) within NSD2 (Fig. 6d). Of the three NoLS predicted with high-confidence, we found that two were arginine-rich sequences within the C-terminus that could robustly target GFP to nucleoli (Fig. 6e).
These results suggest a competitive situation between chromatin reader-domains and NoLS sequences, and support a model in which perturbation of NSD2 chromatin-binding modules enables the activity of Cterminal NoLS's to dominate, leading to nucleolar accumulation.

Discussion
Here we describe the discovery of UNC6934, the first chemical probe to target NSD2, a potent and wellcharacterized driver of both hematological malignancies and solid tumours. UNC6934 follows the recent discovery of BI-9321, which targets the closely-related PWWP1 domain of NSD3 18 . UNC6934 and BI-9321 are chemically distinct and selective, establishing PWWP domains as tractable targets for future chemical biology and drug discovery efforts.
Additionally, we found no significant off-targets by chemical proteomics and in vitro screening of functionally relevant proteins, including a large number of human PWWP domains, methyltransferases, and membrane proteins. Further demonstration of cellular activity and specific target engagement is evident from the robust changes in endogenous NSD2 localization in response to PWWP1 antagonism by UNC6934. We also show limited cytotoxicity by UNC6934 and its negative control counterpart UNC7145, signifying their suitability for cell biology experiments exploring the function of the NSD2 PWWP1 reader domain.
Nucleoli are dynamic membrane-less nuclear structures that not only act as the site of ribosome transcription and pre-assembly, but also as integral organizational hubs in the regulation of many noncanonical functions 31 . It is now widely appreciated that the shuttling of proteins between the nucleolus and nucleoplasm is a critical feature of nuclear biology, regulating many processes, including stress response, DNA repair, recombination, and transcription [31][32][33] . Here we define active nucleolar localization sequences in the NSD2 C-terminus, which likely drive nucleolar sequestration of NSD2 in response to modulation of its chromatin-binding domains. Altering the balance of nucleoplasmic versus nucleolar NSD2 through PWWP1 antagonism by UNC6934 did not have a significant effect on ribosome transcription or global levels of H3K36me2, suggesting instead that these features may provide a mechanism to rapidly tune the sub-nuclear localization of NSD2 in response to stimuli. Supporting this idea, a recent report showed that epigenetic proteins, including NSD2, are sequestered within the nucleolus in response to heat shock stress as a mechanism for subsequent rapid recovery and epigenome maintenance (data from Azkanaz et al. highlighting NSD2 shown as Supplementary Fig. 12) 34 . Given our observations of a steady-state pool of nucleolar NSD2, the question remains how balance and control of nucleolar-nucleoplasmic NSD2 levels may influence NSD2 function in normal and disease biology.
Cytoplasmic localization of an NSD2 variant lacking PWWP1 and the first 3 PHD domains was also reported 14 , suggesting that the sub-cellular compartmentalization of NSD2 is fine-tuned by its reader domains, which could be achieved by masking sub-cellular localization sequences or by engagement of subcellular-specific substrates.
Our data highlight the multivalent nature of NSD2 recruitment to chromatin, whereby NSD2 chromatin reader domains and DNA binding interfaces act cooperatively to coordinate its activity on chromatin.
These findings highlight the utility of UNC6934 as a tool to interrogate the contributions of NSD2-PWWP1 in the interplay between reader domains. Finally, because of its role in multiple myeloma and other cancers, NSD2 has long been a drug target of interest, but despite much community effort, there is no selective, cell-active inhibitor of its catalytic activity. UNC6934 provides a clear starting point for the development of bifunctional molecules, like PROTACs, able to induce proteasomal degradation of NSD2 to antagonize its function in disease.

Acknowledgements
The Structural Genomics Consortium is a registered charity (no: 1097737) that receives funds from;

Competing interests
EpiCypher is a commercial developer and supplier of reagents and platforms used in this study: recombinant semi-synthetic modified nucleosomes (dNucs) and the dCypher® binding assay.

Expression and Purification of biotinylated NSD2-PWWP1
Construct and Expression: DNA fragment encoding human NSD2 (residues 208-368) was amplified by PCR and sub-cloned into p28BIOH-LIC vector, downstream of an AviTag and the upstream of a poly-histidine coding region.

Molecular Docking
The X-ray structure of the PWWP domain of NSD2 in complex with 3f (PDB ID: 6UE6) was prepared with PrepWizard (Schrodinger, New York) using the standard protocol, including the addition of hydrogens, the assignment of bond order, assessment of the correct protonation states, and a restrained minimization using the OPLS3 force field. Receptor grids were calculated at the centroid of the ligand with the option to dock ligands of similar size and a hydrogen bonding constraint with the backbone of A270 was defined.
Over 6,000 commercially available chemical analogs of 3f were prepared with LigPrep (Schrodinger, New York). The resulting library was then docked using Glide SP (Schrodinger, New York) with default settings. Also, the core docking option was turned on to allow only ligand poses that have their core aligned within 1.0 Å of the reference core (the cyclopropyl and the amide group of 3f). Only 448 compounds fitted and were ranked by Glide. Finally, after a visual inspection, 20 compounds were ordered. Optimization and SAR leading from MRT866 to UNC6934 was guided by free energy perturbation and will be presented elsewhere.

dCypher binding assays
Recombinant semi-synthetic designer nucleosomes (dNucs) were from EpiCypher. Two phases of dCypher ® testing on the PerkinElmer AlphaScreen ® platform were performed as previously described 9 .

Selectivity assays
Selectivity of UNC6934 for NSD2-PWWP1 over 14 other PWWP domains was tested using differential scanning fluorimetry (DSF) as previously described 35  The diffraction data for NSD2-PWWP1+UNC6934 was collected at 100K on the home source Rigaku FR-E superbright and data set was processed using the HKL-3000 suite 38 46 .

Cell Culture
Cell lines were cultured according to standard aseptic mammalian tissue culture protocols in 5% CO 2 at by variance stabilizing normalization and tested for differential enrichment relative to pulldowns competed with DMSO vehicle control.

Fluorescence microscopy
For immunofluorescence, cells were fixed with 2% formaldehyde in 1x phosphate buffered saline (PBS) for 10 minutes at room temperature, followed by 3 washes in 1x PBS and permeabilization with 0.25% For confocal microscopy, images were acquired with Quorum Spinning Disk Confocal Microscope equipped with 405, 491, 561, and 642 nm lasers (Zeiss) and processed with Volocity software (Perkin Elmer) and ImageJ. For localization measurements of fluorescent fusion proteins, images were acquired with a EVOS™ FL Auto 2 Imaging System (Thermo Scientific™ Invitrogen™). Co-localization measurements were quantified using a custom CellProfiler (v3.1.9) 47 analysis pipeline.

Western blotting for global H3K36me2
Cells treated with compound for 72 hours before harvesting by centrifugation at 300 x g for 5 min. Cells Proliferation was monitored by counting GFP-expressing cells over time using an IncuCyte live-cell imaging and analysis platform (Sartorius).

5-ethyl uridine (5-EU) incorporation assay
5-EU incorporation assays to measure changes in nucleolar transcription were performed as previously

General Chemistry Procedures
Reactions were carried out using conventional glassware. All reagents and solvents were used as received unless otherwise stated. Reagents were of 95% purity or greater, and solvents were reagent grade unless otherwise stated. Any anhydrous solvents used were purchased as "anhydrous" grade and used without further drying. "Room" or ambient temperature varied between 20-25˚C. Analytical thin layer chromatography (TLC) was carried out using glass plates pre-coated with silica gel (Merck) impregnated with fluorescent indicator (254 nm). TLC plates were visualized by illumination with a 254 nm UV lamp.
Analytical LCMS data for all compounds were acquired using an Agilent 1260 Infinity II system with the UV detector set to 254 nm. Samples were injected (<10 µL) onto an Agilent ZORBAX Eclipse Plus C18, To a 50 mL flask equipped with a stir bar was added methyl 4-formylbenzoate (1.0 g, 1 Eq, 6.1 mmol) and methanol (10 mL), followed by cyclopropylamine (0.35 g, 0.43 mL, 1 Eq, 6.1 mmol). The flask was capped and stirred at room temperature overnight. The next day, the flask was cooled in an ice water bath and sodium borohydride (0.46 g, 2 Eq, 12 mmol) was added portionwise. Borohydride addition was accompanied by effervescence and heating of the solution. After 4 hours, at which time the reaction had come to room temperature, the reaction was quenched by addition of saturated sodium bicarbonate and extracted three times with ethyl acetate. The combined organic layers were washed once more with saturated sodium bicarbonate, once with brine, then dried over sodium sulfate and concentrated to an oil.
Normal phase chromatography over silica (0-100% ethyl acetate in hexanes) provided the free base as a colorless free-flowing oil. The oil was dissolved in 25 mL of diethyl ether and cooled in an ice water bath, and trifluoroacetic acid (1.2 g, 0.80 mL, 1.7 Eq, 10 mmol) was added dropwise. A voluminous white solid formed, which was filtered and washed rigorously with diethyl ether to provide S1-P ( To a scintillation vial was added S3-P (18 mg, 1 Eq, 50 µmol), tert-butyl 4-aminobenzoate (19 mg, 2 Eq, 0.10 mmol), EDC (19 mg, 2 Eq, 0.10 mmol), DMAP (12 mg, 2 Eq, 0.10 mmol), and DMF (0.2 mL). The reaction was heated to 50 ˚C and stirred overnight. The next day, the reaction was partitioned between water and ethyl acetate. The layers were separated, and the aqueous layer was extracted twice more with ethyl acetate. The combined organic layers were washed twice with water, once with saturated sodium bicarbonate, and once with brine, then dried over sodium sulfate and concentrated to an off-white residue.
The next day, the reaction was diluted with distilled water and purified directly by reverse phase chromatography (10-100% methanol in water + 0.1% TFA) and lyophilized to provide UNC7096 (6.51 mg, 5.26 µmol, 81%) as a white hygroscopic solid. 1

Data availability
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE 50 partner repository with the dataset identifier PXD017641.
The structure of NSD2-PWWP1 in complex with MRT866 and UNC06934 were deposited to the Protein Data Bank with accession number 7LMT and 6XCG respectively.