High-throughput screening with nucleosome substrate identifies small-molecule inhibitors of the human histone lysine methyltransferase NSD2

The histone lysine methyltransferase nuclear receptor-binding SET domain protein 2 (NSD2, also known as WHSC1/MMSET) is an epigenetic modifier and is thought to play a driving role in oncogenesis. Both NSD2 overexpression and point mutations that increase its catalytic activity are associated with several human cancers. Although NSD2 is an attractive therapeutic target, no potent, selective, and bioactive small molecule inhibitors of NSD2 have been reported to date, possibly due to the challenges of developing high-throughput assays for NSD2. Here, to establish a platform for the discovery and development of selective NSD2 inhibitors, we optimized and implemented multiple assays. We performed quantitative high-throughput screening with full-length WT NSD2 and a nucleosome substrate against a diverse collection of bioactive small molecules comprising 16,251 compounds. We further interrogated 174 inhibitory compounds identified in the primary screen with orthogonal and counter assays and with activity assays based on the clinically relevant NSD2 variants E1099K and T1150A. We selected five confirmed inhibitors for follow-up, which included a radiolabeled validation assay, surface plasmon resonance studies, methyltransferase profiling, and histone methylation in cells. We found that all five NSD2 inhibitors bind the catalytic SET domain and one exhibited apparent activity in cells, validating the workflow and providing a template for identifying selective NSD2 inhibitors. In summary, we have established a robust discovery pipeline for identifying potent NSD2 inhibitors from small-molecule libraries.

the nuclear receptor-binding SET domain (NSD) 2 family of histone lysine methyltransferase enzymes, NSD1, NSD2/WHSC1/ MMSET, and NSD3/WHSC1L1, have all been implicated as cancer therapeutic targets (1). The catalytic SET domain (Suppressor of variegation, Enhancer of zeste, Trithorax) of NSD family enzymes catalyzes the mono-and di-methylation of the ⑀-amine of lysine 36 of histone H3 (H3K36), utilizing the methyl donor S-adenosyl-L-methionine (SAM) (Fig. 1A). The nonoverlapping roles of the NSD family of enzymes in normal physiology are attributed to a variety of additional domains, including a PWWP (proline-tryptophan-tryptophan-proline) domain, plant homeodomain, zinc finger domains, and a high-mobility group box domain (1).
NSD2 has been implicated as a therapeutic target for a variety of cancers. Because the gene is located within the Wolf-Hirschhorn syndrome critical region of chromosome 4, NSD2 is also known as Wolf-Hirschhorn syndrome candidate 1 (WHSC1) (2). NSD2 was first described as a gene dysregulated by the t(4;14)(p16.3;q32.3) translocation in ϳ15% of multiple myeloma (MM) cases and is called MMSET (2)(3)(4). The translocation results in a fusion transcript of NSD2 with the immunoglobulin heavy chain and increased NSD2 expression. The t(4;14) translocation can cause overexpression of both NSD2 and fibroblast growth factor receptor 3 (FGFR3) (2,3). However, NSD2 is thought to be the primary oncogenic driver of the t(4;14) ϩ MM subtype because NSD2 is universally overexpressed, whereas FGFR3 is not expressed in ϳ30% of MM cases (4 -6). The role of NSD2 in driving t(4;14) ϩ MM pathogenesis was supported by knockdown of NSD2 in MM t(4;14) ϩ cell lines, which led to reduced growth and tumorigenesis (7)(8)(9)(10)(11).
Notably, NSD2 is among the most frequently mutated genes in pediatric cancer genomes (29). The NSD2 SET domain variant, E1099K, was identified in both acute lymphoblastic leukemia tumors and cell lines with increased H3K36me2 that lack the t(4;14) translocation (21,30). Sequence results of Ͼ1,000 pediatric cancer genomes, representing 21 different cancers, revealed the E1099K variant in 14% of t(12;21) ETV6-RUNX1 containing acute lymphoblastic leukemias (21). NSD2 is also among the most frequently mutated genes found in mantle cell lymphoma tumors, where both E1099K and T1150A variants are observed (31). The E1099K variant has also been reported in chronic lymphocytic leukemia (CLL) and lung and stomach cancers (32)(33)(34)(35). Recombinant NSD2 E1099K showed higher in vitro activity compared with the WT enzyme (21). Ectopic expression of NSD2 E1099K induced H3K36me2 and promoted transformation, whereas knockdown of the enzyme reduced cell line proliferation and tumorigenesis (21).
A major challenge in screening for small molecule inhibitors is that native NSD2 requires nucleosomes as a substrate (17). Interestingly, the apparent specific activity of NSD2 is higher with HeLa-derived nucleosomes compared with recombinant nucleosomes, which has been attributed to unknown modifications of the native substrate (17). Thus, native nucleosomes purified from HeLa are likely a more physiologically relevant substrate than recombinant nucleosomes. Recombinant NSD2 does not act on peptides and is thus not amenable to the commonly adapted histone-derived peptide screening platforms. Here, we report biochemical assay development and pilot-scale screening using full-length recombinant WT, E1099K, and T1150A NSD2 enzymes with native HeLa nucleosomes as a substrate. Chemical libraries were screened in a three-dosepoint quantitative high-throughput screening (qHTS) format (44) with the Methyltransferase-Glo (MTase-Glo TM ) biolumi-nescence assay (45). A number of counter and orthogonal assays were also developed to characterize the hits from the primary screen. Confirmed hits were validated with a radiolabeled SAM substrate-based NSD2 activity assay and further interrogated with binding studies by surface plasmon resonance (SPR), methyltransferase profiling, and activity assessments in U-2 OS cells.

Primary assay development and optimization
To identify small molecule inhibitors of the WT NSD2, the MTase-Glo assay was optimized for use as a primary assay, with whole nucleosomes as the substrate, in 1,536-well format with 4-l reaction volumes (Fig. 1B). The MTase-Glo assay reagent measures methyltransferase activity through the coupling of The assays are robust with S/B values Ͼ3 and ZЈ-factor values ϳ0.9. E, titrations of DMSO demonstrated that the assay performance is not diminished by the introduction of vehicle up to 1.7% (mean Ϯ S.D.; n ϭ 3). F, linear correlation is observed between the WT NSD2 enzyme concentration and methyltransferase activity (mean Ϯ S.D.; n ϭ 8).
High-throughput NSD2 screening with nucleosome substrate the NSD2 reaction product S-adenosyl-L-homocysteine (SAH) to a bioluminescent signal (45). The assay was further optimized for the NSD2 variants E1099K (Fig. 1C) and T1150A (Fig.  1D), to enable insight into cross-inhibition of clinically-relevant NSD2 enzymes. The enzyme concentrations were optimized to allow a robust signal-to-background ratio (Ͼ3.0) in a 15-min reaction at room temperature while consuming Յ20% substrate. The three optimized assays were robust with ZЈ-factor values near 0.9. There was no reduction in assay performance in the presence of up to 1.7% DMSO, which represents a triple aliquot of fixed volume (23 nl) transferred via a Kalypsys pintool equipped with a 1,536-pin array (Fig. 1E) (46). A titration of WT NSD2 demonstrated a linear correlation with methyltransferase activity as expected (Fig. 1F).

qHTS for NSD2 inhibitors
The primary screen was carried out by performing qHTS against 16,251 compounds at three concentrations (115, 57.5, and 11.5 M) in 1,536-well plates (Table 1). Commerciallyavailable libraries screened included the LOPAC (1,280 compounds), Prestwick (1,360 compounds), MicroSource (2,000 compounds), and Tocris (1,304 compounds) collections. Addi-tionally, four NCATS libraries were screened, including an epigenetics-focused collection (284 compounds), a natural products library (2,108 compounds), the NPACT library (5,099 compounds), and the NCATS Pharmaceutical Collection (2,816 compounds) (47). These libraries, enriched with pharmacologically active compounds, were selected to evaluate the suitability of the primary and secondary assays to identify NSD2 inhibitors. The overall quality of the primary screen data from 43 plates was high, with average values for the ZЈ-factor of 0.92 Ϯ 0.02, signal-to-background (S/B) of 3.28 Ϯ 0.31, and coefficient of variation (CV) for the DMSO control of 1.68 Ϯ 0.49% (Fig. 2, A-C). The screen was conducted over a period of 4 months, so these data demonstrate excellent day-to-day reproducibility.

Hit confirmation and secondary screening
Of the 16,251 compounds evaluated in the primary screen, 536 compounds were active with a maximum response Ն50%, corresponding to a hit rate of 3.3% (Fig. 2D). Promiscuity scores were calculated for each hit, and any compounds with a promiscuity score higher than 0.2 were eliminated from the cherrypicking. Next, the remaining hits were evaluated by a structural Table 1 Optimized protocol for the NSD2 MTase-Glo UltraGlo luciferase qHTS assay High-throughput NSD2 screening with nucleosome substrate filter to eliminate the pan-assay interference compounds (PAINS) (48). Finally, 289 hit molecules were selected as cherry picks and prepared in 11-point concentration series from library stock solutions for further studies (Fig. 2E).
First, 174 of the cherry picks were confirmed as active in the primary assay (a 60% confirmation rate). Actives were defined as having concentration-response curves (CRCs) in the classes of 1.1, 1.2, 2.1, 2.2, and 3. In brief, classes 1.1 and 1.2 are the highest-confidence complete CRCs containing both upper and lower asymptotes with efficacies Ͼ80% and Յ80%, respectively. Classes 2.1 and 2.2 are incomplete CRCs having only one asymptote with efficacies Ͼ80% and Յ80%, respectively. Class 3 CRCs show activity at only the highest concentration or are poorly fit. Class 4 CRCs are inactive, having a curve-fit of insufficient efficacy or lacking a fit altogether (44). Second, an MTase-Glo counter screen was implemented without NSD2 but containing 200 nM SAH (mimicking 20% substrate conversion), which identified 37 assay interference compounds that might act by inhibiting the coupling enzymes, luciferase or the luminescent signal. Potential redox cycling by compounds was assessed with an Amplex Red assay performed in the presence of reducing agents (49), and 63 compounds were found to be active with a threshold of 3. To provide additional evidence for on-target activity against NSD2, the orthogonal EPIgeneous homogeneous time-resolved fluorescence (HTRF) methyltransferase assay was utilized with reaction conditions identical to the primary assay (Fig. 3). The assay measures the NSD2 reaction product SAH, which competitively displaces d2-labeled SAH that is pre-bound to anti-SAH labeled with Lumi4-Tb, resulting in a loss of FRET signal (50).
Among the 37 compounds found to be active with the MTase-Glo counter assay was a known luciferase inhibitor NCGC00183809 (Fig. 4A). The compound shares structural commonality with the firefly luciferase inhibitor PTC-124 (51,52). As such, it was likely a false-positive hit that inhibits the UltraGlo luciferase enzyme utilized in the MTase-Glo assay. This notion was supported by the results of the orthogonal assay, which did not indicate inhibition of NSD2 activity. UltraGlo luciferase is a genetically evolved firefly luciferase containing 70 mutations to improve its robustness, thermal stability, and resistance to interference compounds (45,53,54), which is consistent with the 158-fold reduced potency of NCGC00183809 to UltraGlo compared with firefly luciferase (Fig. 4A). The nonspecific methyltransferase inhibitor 3-deazaneplanocin A (DZNep) showed similar activities in both the MTase-Glo primary and counter assays; however, no inhibition of NSD2 activity was observed with the orthogonal assay (Fig. 4B).
After the orthogonal assay, 48 confirmed hits remained that were not active in the two counter screens (Fig. 2E). Of the 48 confirmed inhibitors, 45 were also active against the E1099K Figure 2. Quantitative high-throughput screening and secondary screening for inhibitors of NSD2 WT and variant enzymes. A-D, statistics from the qHTS primary screen against full-length WT NSD2 with the MTase-Glo methyltransferase assay for 43 1,536-well plates. A, overall assay performance of the primary screen was robust with an average ZЈ-factor value of 0.92. B, average signal-to-background ratio was 3.28. C, average % CV for the DMSO control was 1.68. D, of the 16,251 compounds screened, 536 were identified as active (3.3% hit rate). E, primary and secondary assay triage process. Eight compound libraries (16,251 compounds) were screened in qHTS format at three concentrations, resulting in 536 hits. From the initial hit list, 289 compounds were selected as cherry picks and further evaluated in 11-point dose response. Of the 289 cherry picks, the activities of 48 were confirmed against NSD2 with the primary and orthogonal assays with no activity observed by the counter assays. The majority of these compounds was also active against the NSD2 variants E1099K (45 compounds) and T1150A (44 compounds). High-throughput NSD2 screening with nucleosome substrate enzyme, and 44 showed activity against the T1150A variant. One of these hits is the nonspecific histone lysine methyltransferase inhibitor chaetocin, which was reported to inhibit NSD1, NSD2, and NSD3 (39). Chaetocin inhibited both WT and variant NSD2 enzymes without showing activity by the MTase-Glo counter assay or Amplex Red assay (Fig. 4C). The hit DA3003-1 is known to be redox-active (49,55), and this was corroborated by the Amplex Red counter screen (Fig. 4D). Redox activity is undesirable, because it can result in nonspecific modulation of proteins, activation of cell pathways with redox-switches, and cytotoxicity (56). DA3003-1 was nevertheless selected for follow-up studies due to its submicromolar potency with the primary assay (IC 50 ϭ 0.9 M). In addition to chaetocin and DA3003-1, three other hits were selected for further studies that have not been previously linked to NSD2 inhibition: PF-03882845, TC LPA5 4, and ABT-199 (Table 2). In anticipation of a full-fledged HTS, these five compounds were selected to validate our post-HTS workflow, which is intended to further evaluate compound activities.

Potency evaluation with the HotSpot radiolabel assay
Biochemical activities of the five selected hit compounds against WT and variant NSD2 enzymes were further validated by the radioisotope-based HotSpot assay (Fig. 5). The HotSpot assay incorporates [ 3 H]SAM to assess total histone methylation by direct measurement of the filter-bound tritiated substrate, without the need for coupling enzymes or anti-bodies (38). Both DA3003-1 and chaetocin inhibited WT and variant NSD2 enzymes with submicromolar potencies, whereas PF-03882845, TC LPA5 4, and ABT-199 inhibited the enzymes at low-micromolar concentrations ( Table 2). No activity was observed with the orthogonal assay. C, confirmation of chaetocin as an NSD2 inhibitor. Chaetocin was identified as a hit from the primary screen, and the activity was confirmed by primary (IC 50 ϭ 8.5 M; Hill slope ϭ Ϫ1.09) and orthogonal assays (IC 50 ϭ 67 M; Hill slope ϭ Ϫ1.37). No activity was observed with the MTase-Glo counter assay or Amplex Red assay. Chaetocin also inhibits the NSD2 variants E1099K (IC 50 ϭ 19 M; Hill slope ϭ Ϫ0.93) and T1150A (IC 50 ϭ 9.6 M; Hill slope ϭ Ϫ1.15). Data from the MTase-Glo Counter and Amplex Red assays resulted from single experiments. All other data are shown as mean values Ϯ S.D. of n technical replicates (n ϭ 3 for WT NSD2 MTase-Glo and n ϭ 2 for HTRF orthogonal, NSD2 E1099K, and NSD2 T1150A). D, primary assay showed inhibition of WT (IC 50 ϭ 0.9 M; Hill slope ϭ Ϫ1.4), E1099K (IC 50 ϭ 2.8 M; Hill slope ϭ Ϫ1.51), and T1150A NSD2 (IC 50 ϭ 2.9 M; Hill slope ϭ Ϫ1.44) enzymes by DA3003-1, whereas no activity was observed with the counter assay. Inhibition of WT NSD2 was further supported by the orthogonal assay. However, DA3003-1 showed activity with the Amplex Red assay (IC 50 ϭ 0.9 M; Hill slope ϭ 1.25) indicating redox activity.

Table 2 Biochemical activities of five compounds that inhibit NSD2 WT and variant enzymes
All values are derived from one experiment unless followed by a symbol, in which case the value represents a global fit to n technical replicate experiments (*, n ϭ 2; ⍀ , n ϭ 3; ⌽ , n ϭ 4; ॖ , n ϭ 5).

High-throughput NSD2 screening with nucleosome substrate
Although chaetocin is known to inhibit NSD2, inhibition by DA3003-1, PF-03882845, TC LPA5 4, and ABT-199 has not been reported. Notably, the HotSpot assay was consistently more sensitive than the MTase-Glo and HTRF assays for both WT and variant enzymes (Table 2). Overall, the reaction conditions were very similar. The MTase-Glo and HTRF assays utilized 8 nM NSD2, 500 nM nucleosomes, 50 mM Tris-HCl, pH 8.8, 5 mM MgCl 2 , 50 mM NaCl, 1 mM TCEP, 0.01% Tween; and the HotSpot assay used 10 nM NSD2, 400 nM nucleosomes, 50 mM Tris-HCl, pH 8.5, 5 mM MgCl 2 , 50 mM NaCl, 1 mM DTT, 0.01% Brij35. A comparison of IC 50 values determined by the HotSpot assay for all five compounds from reactions containing either 1 mM TCEP or DTT did not suggest that potency differences were due to the reducing agent (data not shown). Each of the five compounds inhibited WT, E1099K, and T1150A NSD2 enzymes with similar potencies ( Table 2).

Direct binding of inhibitors to the NSD2 SET domain
To support an on-target mechanism of action for the five inhibitors, SPR was used to determine whether each inhibitor interacts with the catalytic SET domain of NSD2. As expected, the two positive controls, cofactor SAM and product SAH, both bound the NSD2 SET domain with low micromolar affinities (Table 3). With the exception of chaetocin, the inhibitors bound the NSD2 SET domain stoichiometrically, with dissoci-ation constants (K d ) comparable with the in vitro IC 50 values determined by the HotSpot assay ( Table 2). Chaetocin bound the SET domain with an apparent dissociation constant of 20 nM. Additionally, the data indicated super-stoichiometric binding, which might be due to a binding ratio higher than 1:1 that is consistent with chaetocin's two disulfide moieties forming adducts with the NSD2 protein or compound aggregation (data not shown) (57). These data indicate that all five compounds might mediate inhibition of NSD2 by directly binding to the catalytic SET domain.

Selectivity assessment by methyltransferase profiling
Methyltransferase profiling of the five NSD2 inhibitors was performed to examine the selectivity toward NSD2 compared with other methyltransferases. Activities of the inhibitors were evaluated against 35 other methyltransferases, including NSD1 and NSD3, with the HotSpot methyltransferase assay technology (Table 4). DA3003-1 was the least selective methyltransferase inhibitor and showed potent inhibition of nearly every enzyme except DOT1L (no activity) and GLP (weak inhibition). Furthermore, DA3003-1 inhibited 27 enzymes potently with submicromolar IC 50 values. Both PF-03882845 and TC LPA5 4 inhibited most enzymes, albeit with weak potencies, but showed some selectivity to the PRMT5-MEP50 complex and the MLL4 complex. Chaetocin inhibited 18 methyltransferases, 12 with submicromolar potencies (including all NSD enzymes) and two with IC 50 values above 75 M. ABT-199 inhibited 23 methyltransferases, two of which have IC 50 values above 65 M. The NSDs were among the most potently inhibited enzymes, with IC 50 values in the low-micromolar range.

Inhibitor activities in a cell-based assay
To evaluate the activities of the five NSD2 inhibitors in cells, U-2 OS human osteosarcoma cells were chosen due to a relatively high expression of endogenous NSD2 protein (18). To establish a baseline for reduced NSD2 activity, siRNA was used to knock down NSD2 in U-2 OS cells (Fig. 6A). Variations in NSD2 transcript knockdown were observed among the three unique siRNA molecules, and greater reductions in NSD2 corresponded to lower levels of H3K36me2 (Fig. 6, B and C). Knockdown of NSD2 did not appear to reduce the proliferation of U-2 OS cells (Fig. 6D). Next, the cells were treated with each of the five compounds in 10-point dose response from 0.0025 to 50 M for 96 h. NSD2 inhibition should result in reduced H3K36me2. Total histone H3 and H3K36me2 levels were measured by Western blot analysis. The positive control DZNep was tested at a concentration of 10 M in parallel with each  High-throughput NSD2 screening with nucleosome substrate compound and was also evaluated in dose response (Fig. 6E). Densitometry was used to quantify both H3K36me2 and total H3, and the densities of H3K36me2 were normalized to those of H3 (Fig. 6F). The growth of U-2 OS cells over 96 h in the presence of test compounds at the same concentrations was also evaluated (Fig. 6G). The IC 50 value of DZNep for reducing H3K36me2 levels in U-2 OS cells was 390 nM with a modest dose-dependent reduction in U-2 OS confluency (IC 50 ϭ 180 nM). Similar to DZNep, DA3003-1 treatment also resulted in a dose-dependent reduction in H3K36me2 (IC 50 ϭ 545 nM); however, higher drug concentrations resulted in cytotoxicity (CC 50 ϭ 270 nM) similar to chaetocin and the control bortezomib, both of which were cytotoxic at all concentrations tested. PF-03882845 reduced H3K36me2 (IC 50 ϭ 3.2 M) over a range of concentrations that had negligible influence on growth; however, substantial cell death at concentrations above 16.7 M was observed during the Western blot analysis experiment (Fig. 6E), so data from the two highest concentrations were excluded from the potency calculation. TC LPA5 4 did not reduce H3K36me2 over the concentrations tested, and minimal reduction in growth was observed. ABT-199 modestly reduced H3K36me2 levels at 16.7 M; however, higher concentrations were cytotoxic.

Discussion
A number of histone lysine methyltransferases have been implicated as attractive therapeutic targets in the field of oncology, and small molecule inhibitors are in different stages of  50 values for all test compounds are colored as a heat map with green indicating potent inhibition and red indicating weak inhibition. No color indicates no measurable activity. All values are derived from a single experiment except for those of WT NSD2, which result from a global fit of multiple data sets as indicated in Table 2.

High-throughput NSD2 screening with nucleosome substrate
preclinical and clinical development (59,60). Selective inhibitors of NSD2 are of major interest both to advance basic research and for therapeutic development. However, NSD2 is regarded as a challenging target (61), and no selective small molecule inhibitors of NSD2 have been reported to date. Challenges in studying NSD2 in vitro include that the target of NSD2 methylation depends on the nature of the substrate (17), and full-length NSD2 is only active against a nucleosome substrate (38), which may be cost prohibitive in many cases.
A wide variety of methyltransferase assays have been described in the literature (38,45,(62)(63)(64)(65)(66)(67); however, few optimized assays have been reported for NSD2 in plate formats to enable HTS. Recently, a radiolabeled [ 3 H]SAM assay was optimized for the NSD2 SET domain with both histone octamer (ZЈ ϭ 0.69) and nucleosome (ZЈ ϭ 0.75) substrates in a 384-well format (39). The assay utilizing a nucleosome substrate was used to screen 1,040 compounds from the Prestwick Chemical Library at a single concentration of 25 M. Although the reported assays show robust performance in the 384-well format, the use of radiolabeled reagents for HTS is a challenge for many laboratories due to safety regulations and disposal costs.
Herein, we report the implementation and validation of two optimized homogeneous NSD2 activity assays in the highly miniaturized 1,536-well format for the identification of small molecule inhibitors from chemical libraries. The assays were utilized to screen the full-length WT NSD2 enzyme against a nucleosome substrate in qHTS format with three concentrations of test compounds. The use of qHTS reduces both false-positive and false-negative hits common to single-point HTS and facilitates selection of actives (44). For the primary screen, we used the recently reported Methyltransferase-Glo assay reagent with a sensitive biolumines- High-throughput NSD2 screening with nucleosome substrate cent readout (45). A similar approach has been applied for the discovery of NSD1 inhibitors by HTS (68). After screening eight libraries, including numerous pharmacologically active collections, containing over 16,000 compounds, many hits were identified, including chaetocin, which is known to inhibit NSD2 (39). By incorporating orthogonal and counter screens, hits were prioritized for subsequent follow-up studies.
Among the confirmed active inhibitors, DA3003-1, PF-03882845, chaetocin, TC LPA5 4, and ABT-199 were selected for further characterization. In vitro potencies were determined by the HotSpot assay, which is a direct readout of the NSD2 reaction product (38). The HotSpot assay is very similar in format to the traditional gold standard radioisotope detection used in conjunction with gel electrophoresis or MS. The five compounds inhibited WT NSD2 as well as the E1099K and T1150A variant enzymes.
DA3003-1 is a cell-permeable Cdc25 phosphatase inhibitor that potently and irreversibly inhibits all Cdc25 isoforms, including Cdc25A (IC 50 ϭ 29 nM), Cdc25B2 (IC 50 ϭ 95 nM), and Cdc25C (IC 50 ϭ 89 nM) (69). It is known that DA3003-1 is capable of redox cycling (49,55), which was verified here by the Amplex Red counter screen. In addition to potent inhibition of NSD2 activity, our data demonstrate that DA3003-1 bound the SET domain with a strong affinity (K d ϭ 370 nM, Table 3), which is comparable with its potency (IC 50 ϭ 170 nM, Table 2). Together, this suggests that DA3003-1 inhibits NSD2 through a direct interaction with the catalytic SET domain, although it is most likely nonspecific. Notably, DA3003-1 inhibited 27 of the 38 methyltransferase enzymes tested with submicromolar potencies and another eight with IC 50 (Table 4). Both compounds inhibited WT NSD2 with an IC 50 value near 8 M. Also, both compounds were most potent against the PRMT5-MEP50 complex with nearly identical IC 50 values. Furthermore, both compounds had ϳ3-fold weaker potency against the MLL4 complex. Compared with the PRMT5-MEP50 complex, the potency values of PF-03882845 were at least 8-fold weaker against 31 other methyltransferases. Similarly, compared with the PRMT5-MEP50 complex, the potency value of TC LPA5 4 is at least 7-fold weaker against 28 other methyltransferases. No activity or toxicity was observed for TC LPA5 4 in U-2 OS cells.
The fungal mycotoxin chaetocin is known to inhibit NSD2 (39), so identifying it as a hit further validated our screening approach. Of the 38 methyltransferases profiled, chaetocin inhibited 12 with submicromolar potencies. Notably, the methyltransferase profiling indicated chaetocin potencies of 740 nM against SUV39H1 and 570 nM against SUV39H2, which is consistent with a previously reported value of 600 nM (72). Chaetocin was initially reported to be a specific inhibitor of the histone lysine methyltransferase SU(VAR)3-9 both in vitro and in vivo (72). However, the two disulfide bonds of chaetocin can complicate bioassay interpretation because of the potential for redox activity and covalent modification of proteins (36). Indeed, reports have indicated that the activity against histone lysine methyltransferases is due to chemical modification of the enzyme by the disulfide groups (73,74). Superstoichiometric binding of chaetocin to the NSD2 SET domain was observed by surface plasmon resonance, which might be due to the formation of direct compound-thiol adducts. The affinity of chaetocin to the NSD2 SET domain was strong (K d ϭ 20 nM) with a biochemical potency about 7-fold weaker. Chaetocin was cytotoxic to U-2 OS cells at all concentrations tested.
ABT-199, also known as GDC-0199 or venetoclax, binds BCL-2 with a subnanomolar affinity (K i Ͻ0.010 nM) and is approved by the Food and Drug Administration for the treatment of CLL (75,76). The biochemical potency of ABT-199 against WT NSD2 (IC 50 ϭ 1.7 M) was similar against the NSD2 variants, NSD1 and NSD3 ( Table 4). The compound was most potent against the MLL1 and MLL4 complexes. The affinity of ABT-199 to the NSD2 SET domain (K d ϭ 8.3 M) is nearly 5-fold weaker than the biochemical potency. At a concentration of 50 M, ABT-199 was cytotoxic, consistent with its use in oncology, but any influences on cellular H3K36me2 levels were negligible.
DZNep is a carbocyclic analog of adenosine and a derivative of the antibiotic neplanocin-A. It was initially reported as a competitive inhibitor of SAH hydrolase at picomolar concentrations (77). DZNep has since been reported as a global histone methylation inhibitor when used at substantially higher concentrations (36,78,79). Although DZNep was identified as a hit from the primary screen, similar activities were observed with the MTase-Glo primary and counter assays, and no inhibition of NSD2 activity was observed with the orthogonal assay. The potency of DZNep in reducing H3K36me2 in U-2 OS cells (IC 50 ϭ 390 nM; Fig. 6F) is similar to its potency in reducing H3K27me3 in SU-DHL-6 cells (IC 50 ϭ 160 nM) when assessed by the same method of Western blot analysis (80). DZNep has been also shown to reduce H3K36me2 in SW480 cells at a concentration of 5 M (81).
The purpose of this study was to establish an HTS discovery pipeline for NSD2 and to evaluate the workflow for identifying High-throughput NSD2 screening with nucleosome substrate high-quality tool inhibitors of NSD2. The majority of the molecules screened were from pharmacologically active libraries that served to validate the primary and secondary assays. The identification of known methyltransferase inhibitors, including chaetocin and DZNep, further validated the workflow. During this pilot, many known interference compounds were identified by the secondary assays, thereby demonstrating how such bad actors behave among the various assays. Five actives selected from the primary screen were shown to bind the catalytic SET domain and inhibit NSD2 activity in vitro. Although these studies confirm inhibition of NSD2, they do not rule out inhibition by intractable mechanisms of action, such as nonspecific reactivity, redox, or aggregation. Two of the five compounds reduced H3K36me2 in U-2 OS cells, but the mechanisms are likely to be complicated and involve multiple targets. These studies provide a basis for the future discovery and development of novel selective NSD2 inhibitors by establishing a robust workflow for identifying and triaging hits from highthroughput screens.

Methyltransferase-Glo assay
MTase-Glo assays were performed by a multistep format in white solid bottom 1,536-well plates (Greiner,  Plates were then incubated for 30 min at room temperature prior to reaction initiation with 1 l of 4 M (1 M final) SAM in reaction buffer and incubated at room temperature for 15 min. Upon completion, methyltransferase conversion of SAM to SAH was then detected using a two-step detection system where 1 l of MTase-Glo reagent was first added to each well to convert SAH to ADP for 30 min at room temperature. Secondly, 5 l of MTase-Glo Detection Solution was added to each well and allowed to incubate for 30 min at room temperature to convert ADP to ATP, which was then measured by luminescence detection using a ViewLux uHTS Microplate Imager (PerkinElmer Life Sciences) and compared with control samples to determine relative activity.

Methyltransferase-Glo counter assay
MTase-Glo counter assay was performed with identical procedures as with the MTase-Glo primary assay but without NSD2 enzyme or nucleosomes. Instead, 200 nM SAH was added to mimic the reaction with 20% substrate conversion.

EPIgeneous HTRF methyltransferase assay
The EPIgeneous HTRF methyltransferase assay was performed using white solid bottom 1,536-well plates (Greiner, catalog no. 789175-F). First, 23 nl of compounds (or DMSO control) were pin-transferred into 3 l of reaction buffer (50 mM Tris-HCl, pH 8.8, 5 mM MgCl 2 , 50 mM NaCl, 1 mM TCEP and 0.01% Tween 20) containing 666.7 nM (500 nM final) nucleosomes and either 10.7 nM (8 nM final) WT NSD2 enzyme or no enzyme (low activity control, columns 2-3). Plates were then incubated for 30 min at room temperature prior to reaction initiation with 1 l of 4 M (1 M final) SAM in reaction buffer and incubated at room temperature for 15 min. After the incubation period, 0.8 l of EPIgeneous Detection Buffer One was added to each well, followed by a 10-min incubation at room temperature. Next, anti-SAH/Lumi4-Tb solution was prepared according to the manufacturer's instructions, and 1.6 l of the solution were added to each well. Finally, the SAH-d2 conjugate was prepared as a 32-fold dilution according to the manufacturer's instructions, and 1.6 l of the solution was added to each well. The assay plate was allowed to incubate for 1 h at room temperature before detection of the HTRF signal using an Envision plate reader (PerkinElmer Life Sciences).

Amplex red (10-acetyl-3,7-dihydroxyphenoxazine) assay
The assay was adapted from a previously described protocol to assess redox cycling of compounds in the presence of reducing agents (49). 23 nl of compounds (or DMSO control) were pin-transferred into 2.5 l of HBSS (ThermoFisher Scientific; containing 1.26 mM CaCl 2 , 0.49 mM MgCl 2 , 1 g/liter D-glucose) in black 1,536-well plates. Compound fluorescence was measured immediately (READ 0) using a ViewLux uHTS microplate imager (PerkinElmer Life Sciences) equipped with excitation 525/20 and emission 598/25 filters. 2.5 l of a 2ϫ Amplex Red solution (100 M Amplex Red (Cayman Chemical, Ann Arbor, MI), 200 M DTT (ThermoFisher Scientific), and 2 units/ml horseradish peroxidase (Sigma); diluted in HBSS and protected from light) were added to each well. Fluorescence was measured after a 15-min incubation at room temperature High-throughput NSD2 screening with nucleosome substrate (READ 1), using ViewLux settings identical to read 0. Activity was calculated using corrected fluorescence values (READ 1 minus READ 0), which were compared with control samples (negative ϭ vehicle; positive ϭ 46 M walrycin B).

HotSpot methyltransferase assay
The HotSpot radioisotope-based methyltransferase assays were performed as described previously (58,84) with the following modifications. Standard substrate concentrations were 5 M peptide or protein substrate, or 0.05 mg/ml for nucleosomes and core histones, and 1 M SAM, unless otherwise mentioned. For control compound IC 50 determinations, the test compounds were diluted in DMSO and then added to the enzyme/substrate mixtures in nanoliter aliquots by using an acoustic technology (Echo550; Labcyte) with a 20-min preincubation. The reaction was initiated by the addition of [ 3 H]SAM (tritiated SAM, PerkinElmer Life Sciences) and incubated at 30°C for 1 h. The reaction was detected by a filter-binding method. Data analysis was performed using GraphPad Prism software.

Surface plasmon resonance
Surface plasmon resonance measurements were performed using a Biacore 8K (GE Healthcare) at 25°C. Human recombinant NSD2-SET domain was immobilized to a Serial-S CM5 Sensorchip (GE Healthcare) using the classic amine-coupling method in immobilization buffer containing 10 mM HEPES, 150 mM NaCl, 0.5 mM TCEP, and 0.05% v/v surfactant P20. Single cycle kinetic measurements were performed in running buffer containing 50 mM Tris, pH 8.8, 50 mM NaCl, 0.5 mM TCEP, 5 mM MgCl 2 , 0.05% surfactant P20, and 2% DMSO. Compounds were diluted in a 3-fold series with running buffer, and the DMSO concentration was carefully matched to 2%. Compound solutions were then injected over the prepared sensorchip at a flow rate of 80 l/min for 80 s and allowed to dissociate over a period of 200 -600 s. Data analysis was performed with Biacore 8K evaluation software using the 1:1 kinetic binding model.

siRNA knockdown of NSD2
All siRNA transfections were performed using 6-well plates. Within five wells, 128 l of siRNA (either 40 nM Silencer Select Negative control, 5 nM siWHSC1 C (Assay ID 42650), 5 nM siWHSC1 B (Assay ID 42732), 5 nM siWHSC1 A (Assay ID 42800), or 20 nM AllStars Cell Death Control) were complexed with 3.2 l of RNAiMax transfection reagent in 1 ml of McCoy's 5A for 15 min at ambient temperature. The sixth well contained cells only without siRNA and lipid complex. One hundred thousand cells in 1 ml of McCoy's, 20% FBS were added to each well. The plates were maintained at room temperature for 45 min before incubation at 37°C, 5% CO 2 . After 72 h, the cells were utilized for an NSD2 transcript analysis, a cell growth assay, and Western blot analysis of H3K36me2 and total H3 levels. For the transcript analysis, RNA was extracted using an RNeasy mini kit. Following RNA extraction, cDNA was generated with the High-Capacity RNA-to-cDNA kit. RT-PCR was conducted using a TaqMan Gene Expression Master Mix, Taq-Man WHSC1 human (Hs00983720_m1), and human GAPDH endogenous control (4352934E). For the cell growth assay, 1,750 cells were plated in 40-l volumes (McCoy's 5A ϩ 10% FBS ϩ 0.5ϫ penicillin/streptomycin) in a 384-well cell carrier plate (PerkinElmer Life Sciences), and the plates were sealed with Breathe-Easy sealing membrane (Sigma). Images were captured every 4 h up to 72 h with an IncuCyte ZOOM System (Essen BioScience). For the Western blot analysis, the cells were lysed with 1ϫ SDS sample buffer and blotted as described below.

Western blotting analyses
Test compounds were dissolved with DMSO to a 10 mM stock. U-2 OS cells were seeded in 12-well plates at a density of 0.5 ϫ 10 6 /well in complete culture medium and placed into the incubator at 37°C, 5% CO 2 . After overnight incubation, the cells were treated with test compounds (10-point concentration series with 3-fold dilution, 0.0025-50 M) or with reference compound DZnep (10 M single dose) and allowed to incubate for an additional 96 h. Following the incubation with compound, culture medium was removed, and the cells were washed once with ice-cold PBS. The cells were lysed with 1ϫ SDS sample buffer (62.5 mM Tris-HCl, pH 6.8, 2% SDS w/v, 10% glycerol, 0.01% w/v bromphenol blue, 50 mM DTT), and the lysates were sonicated three times in 3-s increments at 15 A. Cell lysate samples (14 l) were subjected to SDS-PAGE and transferred onto nitrocellulose membranes by the iBlot dry blotting system. The membranes were blocked with 2% nonfat milk blocking buffer for 1 h and then probed with anti-histone H3 (dimethyl Lys-36) primary antibody overnight. Anti-rabbit IgG IRDye 680RD secondary antibody was used to detect the primary antibody. Then the blots were washed three times with 1ϫ TBS buffer plus 0.01% Tween 20 and re-probed with antihistone H3 primary antibody and anti-mouse IgG IRDye 800CW secondary antibody. The membranes were scanned with a LI-COR Odyssey Fc Imaging System. The specific bands of interest were quantified by LI-COR Image Studio Lite software.

U-2 OS cell growth assay
U-2 OS (ATCC HTB-96) cells were obtained directly from the ATCC and cultured according to the recommended culturing conditions. At cell passage 2, 1,750 cells were plated in 40-l volumes (McCoy's 5A ϩ 10% FBS ϩ 0.5ϫ penicillin/streptomycin) in a 384-well cell carrier plate (PerkinElmer Life Sci-High-throughput NSD2 screening with nucleosome substrate ences) and incubated overnight. At 16 h, test compounds were delivered in 195-nl aliquots by pin transfer. The plates were sealed with Breathe-Easy sealing membrane (Sigma), and images were captured every 4 h up to 96 h with an IncuCyte ZOOM System (Essen BioScience).

Data analysis
Data normalization and curve fitting were performed using in-house informatics tools. Briefly, raw plate reads for each titration point were first normalized relative to the DMSO-only wells (100% activity) and no enzyme control wells (0% activity) and were then corrected by applying a plate-wise block pattern correction algorithm to remove any plate edge effects and systematic background noise. Active compounds from the primary HTS were defined as having a maximum response Ն50%.
To determine compound activities from the 11-point qHTS, the concentration-response data for each sample were plotted and modeled by a four-parameter logistic fit yielding IC 50 and efficacy (maximal response) values as described previously (44). The activities were designated as classes 1-4 according to the type of concentration-response curve observed. Active compounds were defined as having concentration-response curves in the classes of 1-3. The promiscuity score for each compound was defined as (number of assays in which the compound is active)/(total number of assays in which the compound was tested). A compound with a promiscuity score higher than 0.2 was considered as a "frequent hitter" to be eliminated from the follow-up studies.