PRC2 Inhibitors Overcome Glucocorticoid Resistance Driven by NSD2 Mutation in Pediatric Acute Lymphoblastic Leukemia

With the progressive improvements in the efficacy of multi-agent regimens, five-year survival rates for children with acute lymphoblastic leukemia (ALL) have increased up to 90%. Nevertheless, ALL remains one of the most frequent causes of cancer-related death in children because relapsed disease is highly resistant to chemotherapy (1–3). Genome-wide analyses provided insight into the pathogenesis of ALL, revealing frequent mutation of transcriptional factors and epigenetic regulators (4). More recent studies have highlighted that mutations found in relapsed ALL such as CREBBP, FPGS, MSH2, NR3C1, NT5C2, PRPS1, TP53, and NSD2 (5, 6). NR3C1, encoding the glucocorticoid receptor (GR), as well as CREBBP mutations drive glucocorticoid (GC) resistance due to lack of GC transcriptional response (5, 7, 8). Mutations of nucleotide metabolism enzymes NT5C2 and PRPS1 and DNA mismatch-repair gene MSH2 cause resistance to 6-mercaptopurine (6-MP; refs. 5, 9–12). FPGS, a folate metabolism gene, is linked to methotrexate resistance (5). Mutations of TP53 are associated with treatment failure and poor outcomes (5, 13). Minor ALL subclones with NSD2 mutations at diagnosis persisted to become the dominant clone at early relapse, suggesting that NSD2 mutations also drive therapy resistance in ALL.

NSD2 encodes a histone methyltransferase specific for di-methylation of histone H3 at lysine 36 (H3K36me2; ref. 14). NSD2 is overexpressed in multiple myeloma with t(4;14) and causes a global increase in H3K36me2, a decrease in H3K27me3, aberrant gene expression, and resistance to DNA damage (15, 16). Although the frequency of NSD2 mutations is approximately 3% in pediatric ALL, NSD2 mutations are enriched in patients with B-cell ALL (B-ALL) with t(12;21) ETV6–RUNX1 (20%) and t(1;19) TCF3–PBX1 (15%; refs.17, 18) and are recurrently found in relapsed ALL (6.8%), with the most common site being p.E1099K, within the enzymatic SET domain (5). This mutation increases NSD2 activity, but not protein levels, and results in a global increase in H3K36me2 and decrease in H3K27me3 (17, 19, 20). Comparison of NSD2 p.E1099K cell lines and isogenic NSD2 wild-type (WT) cells with the mutation eliminated by gene editing showed that NSD2 p.E1099K stimulated cell growth, migration, and an aberrant gene signature enriched in mesenchymal and neuronal genes. Xenografted NSD2 p.E1099K cell lines showed tropism to the central nervous system, suggesting a possible mechanism for ALL relapse, correlating with increased expression of neural adhesion molecules and chemokines (20).

The enhanced occurrence of NSD2 mutations in early relapse (5, 6) suggests that the aberrant protein might lead to resistance to ALL therapy. Therefore, we performed high-throughput drug screening in isogenic ALL cell lines and found that NSD2 p.E1099K cells were resistant to GCs but not to other chemotherapeutic agents. The NSD2-mutant protein blocked GR expression, autoactivation, and the ability of GR to bind apoptosis-related GC-responsive genes. Pretreatment with polycomb repressive complex 2 (PRC2) inhibitors significantly reversed this form of resistance, suggesting a novel therapeutic approach for relapsed ALL with NSD2 mutation.

To explore which component of ALL therapy may be affected by NSD2 mutation, we performed high-throughput screening of isogenic NSD2 p.E1099K and NSD2 WT RCH-ACV cell lines with nearly 10,000 compounds, including the NCAT Pharmaceutical Collection of all approved drugs, and many experimental therapeutics with defined mechanisms of action (21). Although the screen did not identify compounds strongly selective for NSD2-mutant cells, we found that mutant cells were specifically resistant to GCs (Fig. 1A; Supplementary Fig. S1A; Supplementary Table S1) but not to other agents used in pediatric ALL regimens such as vincristine, daunorubicin, cyclophosphamide, l-asparaginase, and 6-MP (Fig. 1B; Supplementary Fig. S1B). In a validation study, NSD2 p.E1099K RCH-ACV cells were resistant to dexamethasone, but as sensitive to daunorubicin, vincristine, and 6-MP as isogenic WT cells (Fig. 1C).

We next investigated the GC sensitivity of samples from six pediatric ALL patient-derived xenograft (PDX) models (22, 23): four with NSD2 p.E1099K, one with NSD2 p.T1150A, and one with NSD2 p.G1246S (Supplementary Fig. S1C; Supplementary Table S2). NSD2 p.E1099K and p.T1150A are located in the enzymatic SET domain, whereas NSD2 p.G1246S is located in a PHD domain near the C-terminus of the protein. Sample ALL-58 was derived from a patient with B-ALL with NSD2 p.T1150A at diagnosis but NSD2 p.E1099K at relapse from chemotherapy (ALL-123). All of these PDX samples were treated ex vivo with dexamethasone, revealing that PDX cells with NSD2 p.E1099K and NSD2 p.G1246S were resistant to dexamethasone, whereas one with NSD2 p.T1150A was as sensitive as cells with WT NSD2 (Fig. 1D; Supplementary Table S3). NSD2 p.E1099K was associated with increased H3K36me2 and a minor decrease of H3K27me3 (Fig. 1E; Supplementary Fig. S1D). In accordance with in vitro results, NOD/SCID recipient mice injected with NSD2 WT cells were sensitive to dexamethasone, whereas those with NSD2 p.E1099K were completely resistant with no change in tumor burden or survival with dexamethasone treatment (Fig. 1F). Dexamethasone induced apoptosis in NSD2 WT cells but not in NSD2 p.E1099K cells ex vivo (Fig. 1G and H). Administration of dexamethasone to a series of ALL cell lines with mutant NSD2 p.E1099K [B-ALL: RCH-ACV, SEM; T-cell ALL (T-ALL): RPMI-8402, HPB-ALL, MOLT13] or WT NSD2 (B-ALL: 697, SUP-B15; T-ALL: CEM, MOLT4; refs. 17, 19) showed that T-ALL cell lines were generally not as sensitive as B-ALL cell lines to dexamethasone (Supplementary Fig. S1E). Compared with cell lines with NSD2 WT, B-ALL cell lines with NSD2 p.E1099K were more resistant to dexamethasone but showed some decrease in proliferation with treatment. By contrast, the T-ALL cell lines with NSD2 p.E1099K mutation were completely resistant to dexamethasone.

Using CRISPR/Cas9, we reverted NSD2 p.E1099K to NSD2 WT in the RCH-ACV cell line (20). We created additional isogenic SEM and RPMI-8402 cells where NSD2 p.E1099K was also corrected (Supplementary Fig. S2A and S2B). Disruption of the allele encoding NSD2 p.E1099K significantly decreased H3K36me2 and increased H3K27me3 (Fig. 2A; Supplementary Fig. S2C). Although NSD2-mutant cells were partially (RCH-ACV and SEM) or completely (RPMI-8402) resistant to dexamethasone, reversion of NSD2 p.E1099K to WT conferred GC sensitivity (Fig. 2B). Dexamethasone failed to induce apoptosis in NSD2-mutant cells, especially in RPMI-8402, whereas administration of dexamethasone triggered apoptosis and cell-cycle arrest of isogenic WT cells (Fig. 2C and D; Supplementary Fig. S2D and S2E). In vivo, female and male mice xenografted with NSD2 p.E1099K cells had a higher tumor burden including central nervous system infiltration and shorter survival time than those harboring NSD2 WT cells (Fig. 2E, left and middle). Mice xenografted with NSD2 p.E1099K cells were resistant to dexamethasone, whereas treatment of mice harboring isogenic NSD2 WT cells led to significant tumor reduction and extension of survival (Fig. 2E, right).

To complement reversion of NSD2 p.E1099K, we inserted this mutation into the CEM cell line (Supplementary Fig. S2B). Knock-in of NSD2 p.E1099K increased global levels of H3K36me2 and decreased H3K27me3 (Fig. 2A; Supplementary Fig. S2C), and the CEM cell line became completely resistant to dexamethasone (Fig. 2C and D; Supplementary Fig. S2D and S2E). Collectively, these gain-of-function and loss-of-function studies in isogenic gene-edited ALL cells established a causal link between NSD2 p.E1099K and GC resistance.

To characterize the pathways altered by NSD2 p.E1099K mutation in response to GC, we performed RNA sequencing (RNA-seq) of these isogenic cell lines with dexamethasone treatment. The transcriptional response to GC was modest in NSD2-mutant RCH-ACV (157 up and 36 down), SEM (363 up and 234 down), and CEM cells (129 up and 28 down) and virtually absent in RPMI-8402 cells (24 up and 7 down; Fig. 3A), which was consistent with the partial sensitivity of RCH-ACV, SEM, and CEM cells to dexamethasone and the lack of GC response in RPMI-8402 cells. With elimination of the NSD2 p.E1099K mutation, the transcriptional response to GC [Fold change (FC) > 1.5, P < 0.05] was dramatically increased in all four cell lines as RCH-ACV (1,208 up and 778 down), SEM (791 up and 569 down), RPMI-8402 (2,043 up and 1,621 down), and CEM (1,816 up and 1,290 down), correlating with the GC sensitivity in Fig. 2B. We further analyzed the genes upregulated by dexamethasone only in NSD2 WT cells to ascertain which genes might play a role in inhibition of cell growth (1,064 genes in RCH-ACV, 527 genes in SEM, 2,024 genes in RPMI-8402, and 1,695 genes in CEM; Fig. 3B). We found 39 genes activated in common among NSD2 WT cell lines (Supplementary Fig. S3A; Supplementary Table S4). Some genes were upregulated in both NSD2 WT and NSD2 p.E1099K cells but were more highly induced in NSD2 WT cells in response to dexamethasone. Therefore, we combined the genes upregulated only in NSD2 WT cells and the genes upregulated in both WT and mutant cells that showed ≥1.5-fold more induction in WT cells, yielding a total of 75 activated genes activated by GC, which were enriched in apoptosis-related processes (Fig. 3C and D; Supplementary Table S4). Among the genes significantly activated by GC after removal of NSD2 p.E1099K were the proapoptotic BH3-only genes, BCL2L11 and BMF (encoding BIM and BMF proteins), together with NFKBIA (Fig. 3E and F). Additionally, there were 21 genes commonly downregulated in response to GC in NSD2 WT cells ascertained by the overlap from RCH-ACV (760 genes), SEM (459 genes), RPMI-8402 (1,621 genes), and CEM (1,268 genes; Supplementary Fig. S3B and S3C; Supplementary Table S4). The overlap of genes downregulated in NSD2 WT cells and the genes downregulated in both WT and mutant cells decreased >1.5-fold more in WT cells and yielded a limited set of 22 genes (Supplementary Fig. S3C; Supplementary Table S4) that function in metabolic pathways (Supplementary Fig. S3D). These data indicate that GC-mediated transcription is blocked by NSD2 p.E1099K, potentially contributing to ALL relapse.

To determine how NSD2 p.E1099K blocks GC function and GR response, we assayed the state of chromatin in isogenic NSD2-mutant and WT cells using the assay for transposase accessible chromatin using sequencing (ATAC-seq). In some of the NSD2-mutant cell lines, there were more sites of open chromatin at baseline prior to GC treatment (Supplementary Fig. S4A), consistent with our previous work showing that overexpression of NSD2 and increased levels of H3K36me2 lead to more accessible chromatin (24). After addition of dexamethasone, there were very modest increases in chromatin accessibility in the ALL cell lines harboring NSD2 p.E1099K. By contrast, with the reversion of NSD2 p.E1099K mutation to WT, GC induced significant chromatin accessibility in RCH-ACV, SEM, and RPMI-8420 cell lines. In GC-sensitive cells, GCs execute their functions via interacting with GR in the cytoplasm. The GC–GR complex translocates into the nucleus and binds either directly through a cognate sequence or tethered to other transcription factors to activate gene transcription through recruitment of cofactors such as the histone acetyl transferases (P300, CREBBP, and PCAF), histone methyl transferases (CARM1), mediator components, and the BRG1 and BRM SWI/SNF ATPases that remodel chromatin and increase accessibility (23, 25). With the insertion of NSD2 p.E1099K into CEM cells harboring WT NSD2, GC induced much less chromatin accessibility (Fig. 4A; Supplementary Fig. S4A). By contrast, in NSD2 WT cells, dexamethasone led to an increase in accessibility peaks upstream and downstream of gene bodies (Supplementary Fig. S4B). Among the genes displaying differential chromatin accessibility was that of BCL2L11 encoding BIM. GC-mediated chromatin accessibility at the BCL2L11 promoter and enhancer was significantly decreased in all NSD2 p.E1099K cells (Fig. 4B).

In lymphocytes or ALL cells, GCs execute their biological functions by cooperation between GR and CCCTC-binding factor (CTCF) binding at promoter and enhancer region to modulate chromatin modification, recruiting transcriptional machinery to open chromatin and induce gene transcription (22, 23). CTCF protein levels were unchanged in NSD2-mutant versus WT cells (Supplementary Fig. S4C). We previously showed that the BCL2L11 enhancer can become hypermethylated and inaccessible in GC-resistant cells (23). In T-ALL cell lines, but not B-ALL cell lines in which NSD2 was converted to WT, there was a modest loss of DNA methylation of the BCL2L11 enhancer compared with levels in isogenic NSD2 p.E1099K-mutant cells (Supplementary Fig. S4D and S4E). Nevertheless, the enhancer became highly responsive to GC, suggesting that hypermethylation and silencing of the BCL2L11 enhancer is not the mechanism of GC resistance in NSD2-mutant cells. Chromatin immunoprecipitation sequencing (ChIP-seq) revealed that GC-induced GR and CTCF binding was largely blocked in NSD2-mutant cells, while in isogenic NSD2 WT cells, there was robust stimulation of binding (Fig. 4C), predominantly to upstream and exon regions (Supplementary Fig. S5A). This was particularly evident for BCL2L11 whose GC-mediated expression is blocked in NSD2-mutant cells. GC stimulated GR and CTCF binding to a previously identified enhancer [intronic GR-binding region (IGR); ref. 23] in isogenic NSD2 WT cells, but induced binding was partially blocked in RCH-ACV and completely absent in RPMI-8402 cells (Fig. 4D). Similarly, GC-induced GR and CTCF binding to the BMF and NFKBIA genes was blunted (RHC-ACV) or abrogated (RPMI-8402) in NSD2 p.E1099K cells (Supplementary Fig. S5B). Induced GR and CTCF binding in NSD2 WT cells was accompanied by increased H3K27 acetylation (H3K27ac) at the BCL2L11 IGR as well as enhancers and promoters of the BMF and NFKBIA genes (Fig. 4D; Supplementary Fig. S5B).

We integrated the RNA-seq, ATAC-seq, and ChIP-seq data sets of isogenic RCH-ACV and RPMI-8402 cell lines in response to GC to identify genes directly targeted by GR. Specifically, we sought induced GR-binding peaks ±2 kb from the transcriptional start site of genes and identified 200 genes bound and activated by GR in response to GC in common between RCH-ACV and RPMI-8420 cells, many of which were related to GC-mediated apoptosis (Supplementary Table S5; Supplementary Fig. S5C). Most of these genes were activated by GC in all four cell lines and included BCL2L11, BMF, NFKBIA, and NR3C1. In NSD2 WT RCH-ACV and RPMI-8402 cells, GR bound to 115 downregulated genes in common, largely encoding cell cycle regulatory proteins. Collectively, our data indicate that the global decrease in GR and CTCF binding across the genome, including at BCL2L11, and resulting defect in GC-mediated gene activation, underlie GC resistance in NSD2-mutant cells.

GC-mediated cell death is a threshold phenomenon, and induction of GC-responsive genes is sensitive to the level of the GR (26) with knockdown of the GR and loss of positive autoregulation of the NR3C1 gene associated with GC resistance of ALL (27). Interestingly, before the addition of dexamethasone to the cultures, NR3C1 was one of only three genes (NR3C1, ARNTL2, and FAM78A) downregulated in common in the RCH-ACV, SEM, RPMI-8402, and CEM NSD2 p.E1099K-mutant cells compared with NSD2 WT cell lines (Supplementary Fig. S6A). In NSD2-mutant cell lines and PDX cells, treatment with GC failed to induce transcription of NR3C1. Knock-in of NSD2 p.E1099K decreased basal NR3C1 expression in CEM cells and blocked GC-mediated induction of NR3C1. After the reversion of NSD2 p.E1099K mutation to WT, dexamethasone treatment led to increased NR3C1 expression (Fig. 5A and B; Supplementary Fig. S6B and S6C). Although basal and induced levels of GR were lower in NSD2-mutant ALL cells, GR translocation was not affected by the NSD2 p.E1099K mutation (Supplementary Fig. S6D). Increased NR3C1 expression was associated with enhanced GR and CTCF binding and H3K27ac enrichment at the NR3C1 promoter to induce feed-forward transcriptional induction of GR (Supplementary Fig. S6E). Increased GR induced expression of proapoptotic BCL2L11/BIM in NSD2 WT cell lines and PDX cells (Fig. 5B; Supplementary Fig. S6B and S6C).

To confirm that the failure of GC treatment was due to low GR expression, we overexpressed the GR in GC-resistant NSD2 p.E1099K RCH-ACV and RPMI-8402 cells (Fig. 5C; Supplementary Fig. S6F). High-dose (10 μmol/L) dexamethasone led to only a 20% loss of viability in RCH-ACV cells and minimal cell death in RPMI-8402 cells. With overexpressed GR, the NSD2-mutant cells displayed an IC₅₀ of 10 nmol/L to dexamethasone (Fig. 5D), approaching the profile of NSD2 mutation reverted WT cells (IC₅₀ of ∼100 nmol/L; Fig. 2B). Overexpression of GR allowed for GC-mediated apoptosis (Fig. 5D; Supplementary Fig. S6G) and restoration of GC-mediated induction of BCL2L11, BMF, and NFKBIA (Fig. 5E; Supplementary Fig. S6H). ChIP-qPCR showed that overexpression of GR significantly increased H3K27ac modification at the enhancers of BCL2L11, BMF, and NFKBIA in the presence of dexamethasone but did not affect H3K27me3 levels at the BCL2L11 promoter, which does not bind GR (Supplementary Fig. S7A). This indicates that the chromatin of NSD2-mutant cells can respond to activated, GC-bound GR and supports the idea that the major defect in NSD2-mutant ALL cells is an insufficiently high level of GR.

The failure of GC to activate NR3C1 in NSD2-mutant cells was associated with significant accumulation of H3K27me3 at the promoter of NR3C1, which was completely absent in NSD2 WT RCH-ACV cells (Fig. 5F). GC treatment did not decrease H3K27me3 levels at the NR3C1 promoter in NSD2-mutant cells. Similarly, NSD2 mutation was associated with increased H3K27me3 upstream of the BCL2L11 transcriptional start site, also unresponsive to GC treatment (Fig. 5F). In NSD2 WT cells, GR and CTCF binding near the start site of transcription NR3C1 was induced by GC in association with increased chromatin accessibility and accumulation of H3K27ac (Supplementary Fig. S6E). In NSD2-mutant cells, there was decreased induction of GR binding after GC treatment and evidence of decreased GR activity with reduced induction of open chromatin and less accumulation of H3K27ac in response to GC.

To determine the contribution of enhanced NSD2 activity to the repression of NR3C1, we performed ChIP-seq for the H3K36me2 mark in isogenic RCH-ACV NSD2-mutant and WT cells and integrated this with H3K27me3 modification and RNA-seq gene-expression data. This revealed, in a manner similar to our findings and those of others in cells expressing high levels of NSD2 (28, 29) or harboring p.E1099K-mutant NSD2 (30), a widespread increase in H3K36me2, particularly in intergenic regions (Fig. 6A and B), accompanied at many genes by a decrease in H3K27me3. H3K36me2 levels did not change with GC treatment, and the increase in H3K36me2 was not uniform across the genome. In RCH-ACV cells, we identified 478 genes whose expression was activated in the presence of the NSD2 mutation when compared with isogenic NSD2 WT cells in association with loss of H3K27me3 at their promoters, concomitant with a broad gain of H3K36me2, an example being the neural adhesion molecule NCAM1 (Fig. 6A; Supplementary Table S6). NR3C1 sat within a dense cluster of H3K36me2 modification in NSD2 WT cells, but in NSD2-mutant cells, H3K36me2 was present more broadly across the local genomic region. In the vicinity of the NR3C1 promoter, H3K36me2 was sparser, and increased H3K27me3 was found near the transcriptional start site (Fig. 6B) in association with repressed GR expression. BCL2L11, whose induction is blocked in NSD2-mutant cells, displays no change in H3K36me2 patterns in NSD2-mutant versus WT cells yet an increase of H3K27me3 (Fig. 6C), potentially due to depressed binding of GR and replacement by a repressive factor. BMF did not show changes in H3K27me3 in NSD2-mutant versus WT cells, whereas GILZ showed a modest increase of H3K27me3 in NSD2-mutant cells and only a 20% decrease in baseline expression (Supplementary Fig. S7B), suggesting that the global shifts of H3K36me2 and H3K27me3 were less important for the regulation of these GC-responsive genes than the loss of GR expression.

There were 245 genes that gained H3K27me3 at their promoters and showed decreased expression in NSD2 p.E1099K cells (Supplementary Table S6). Analysis of the promoters of genes activated and repressed in association with mutant NSD2 showed highly divergent patterns of DNA-binding motifs (Fig. 6D). Promoters of genes activated upon removal of H3K27me3 in NSD2-mutant cells were enriched for binding sites of highly expressed lymphoid transcriptional activators, including EBF1, IRF2, STAT5A, and ZNF143, the latter a mediator of enhancer–promoter looping (31). By contrast, genes repressed and demonstrating increased H3K27me3 in their promoters in the presence of mutant NSD2 had motifs for transcription factors poorly expressed in lymphoid cells, GATA5 and SOX1, did not have the types of binding sites found in the activated promoters, and were enriched for binding sites for the BCL6 transcriptional repressor. Collectively, these data indicate that NSD2 mutation causes global changes in chromatin in a heterogeneous manner, inappropriately activating and repressing genes, notably for its phenotypic effects, repressing the NR3C1 encoding the GR.

Given the accumulation of H3K27me3 at the NR3C1 promoter in NSD2 p.E1099K-mutant cells, we hypothesized that inhibition of EZH2 or other PRC2 responsible for creation of H3K27me3 might then reactivate the NR3C1 locus and restore autoactivation of GR expression and activity. NSD2 p.E1099K cells were treated with three different EZH2 inhibitors (EPZ-6438, GSK126, and UNC1999) as well as an inhibitor of the EZH2-interacting protein EED (EED226). Each of the inhibitors caused a global decrease in the level of H3K27me3 in NSD2-mutant cells with no impact on H3K36me2 (Supplementary Fig. S8A). The inhibitors had a modest effect on cell viability at the highest doses tested. However, pretreatment of NSD2-mutant B-ALL and T-ALL cells with any of the PRC2 inhibitors for three to seven days followed by treatment with dexamethasone potently inhibited cell growth and induced apoptosis (Fig. 7A–C; Supplementary Fig. S8B and S8C). Similarly, partial knockdown of EZH2 in RCH-ACV cells increases sensitivity to GC treatment (Supplementary Fig. S8D–S8G). GSK126, a highly selective inhibitor of EZH2 that we found to inhibit the proliferation of EZH2-mutant lymphoma (32, 33) and KDM6A-mutant multiple myeloma (34), was most effective. T-ALL and B-ALL PDX cells harboring NSD2 p.E1099K were pretreated with GSK126 followed by dexamethasone treatment (Fig. 7D). The EZH2 inhibitor on its own had a mild effect, extending median survival by six to nine days. Dexamethasone had no effect on the B-ALL xenograft and extended survival of mice harboring the T-ALL xenograft by a few days. By contrast, the combination therapy extended the survival for the B-ALL and T-ALL PDX mice by four weeks or more (Fig. 7D).

After pretreatment of NSD2-mutant cell lines with PRC2 inhibitors, dexamethasone robustly induced BCL2L11, BMF, and NFKBIA expression (Fig. 7E). PRC2 inhibitor treatment increased basal levels of NR3C1 and allowed dexamethasone to further upregulate NR3C1 expression (Fig. 7E; Supplementary Fig. S9A). ChIP analysis of the NR3C1 promoter showed that EZH2 inhibitor pretreatment led to a decrease of H3K27me3 at the start site of transcription (Fig. 7F). Furthermore, pretreatment with EZH2 inhibitor restored the GR autoactivation loop allowing GR levels to rise in NSD2-mutant cells, increasing binding of GR at the NR3C1 promoter and across the genome, including the BCL2L11 enhancer (Fig. 7F and G; Supplementary Fig. S9B). Although longer-period pretreatment of NSD2-mutant cells with PRC2 inhibitors (14 days) enhanced the activity of dexamethasone, NSD2 WT cells were already highly sensitive to dexamethasone, and no additive effect of EZH2 inhibitor could be detected, although we did note that the combination treatment enhanced activation of BCL2L11 but not NR3C1, BMF, or NFKBIA (Supplementary Fig. S9C), offering a theoretical benefit in non–NSD2-mutant ALL.

Genetic studies offer a caveat to the use of EZH2 inhibitors as disruption of the PRC2 complex may occur in T-ALL (35), and EZH2 knockdown or knockout in T-ALL cell lines led to resistance to conventional chemotherapy (36). Pretreatment of NSD2-mutant RCH-ACV (B-ALL) and RPMI-8402 (T-ALL) cells with EZH2 inhibitor did not prevent cytotoxic effects of common ALL agents, including daunorubicin, vincristine, and 6-MP (Supplementary Fig. S9D). Taken together, these data demonstrate that pharmacologic inhibition of PRC2 activity may provide a promising treatment for patients with ALL with NSD2 p.E1099K mutation.

The elucidation of the molecular biology of ALL has uncovered mechanisms of disease pathogenesis, often involving altered function of transcription factors necessary for normal lymphocyte differentiation. However, no matter the underlying molecular lesion, most T-ALL and B-ALL leukemic cells, like normal lymphocytes, undergo apoptosis in response to GC, and ALL treatment regimens contain this agent. Accordingly, a growing literature indicates that anomalies of GR function are enriched in relapsed ALL, suggesting that failure of the GR to activate its downstream targets, including the critical proapoptotic regulator BCL2L11/BIM (23, 37), is a common cause of relapse. Mechanisms of resistance to GC in ALL include mutation or deletion of the GR (5, 38). Anomalies of the mineralocorticoid receptor, which can bind in concert with the GR, are also found in relapsed ALL (5). GC resistance can also result from increased AKT activity in ALL, leading to phosphorylation of the GR, preventing its nuclear import and binding to targets (39) or enhanced proteolytic cleavage of the GR (40). GR activates genes in part by recruitment of CREBBP, and inactivating mutations of this cofactor also confer GC resistance (8). Furthermore, decreased expression of SWI/SNF chromatin remodeling components is associated with primary GC resistance (41).

NSD2 p.E1099K-mutant ALL cell lines and xenografts were highly resistant to GC, and resistance was reversed by removal of the mutation, whereas installation of the mutation into sensitive cells created resistance. Additionally, NSD2-mutant ALL cells tended to migrate to the brain, this correlating with elevated expression of genes such as NCAM1, GABRB2, SATB2, PECAM1, and FGF13. Aggressive invasive behavior and GC resistance together may explain the association of NSD2 mutation with early relapse of ALL. Unbiased high-throughput screening as well as validation studies showed that NSD2-mutant and isogenic WT cells were similarly sensitive to all other conventional chemotherapy agents, including vincristine, 6-MP, and daunorubicin. This contrasts with a recent report in which knockdown of NSD2 protein in ALL cells was associated with increased sensitivity to chemotherapy (30). Given the important role of NSD2 in DNA damage response, the reduction of NSD2 below baseline levels would be expected to inhibit DNA repair and augment the action of chemotherapy (15, 42). By contrast, conversion of mutant NSD2 to WT would not be expected to disrupt the ability of NSD2 to affect PTEN activity (16) or interplay with PARP1 (43), potential mechanisms of action of NSD2 in DNA repair. Our system, which more closely mimics the clinical situation, suggests that GC resistance is the major cause of relapse in NSD2-mutant ALL.

Genome-wide studies with isogenic NSD2-mutant and WT cells showed that NSD2 p.E1099K was associated with lower basal binding of the GR across the genome and poor induction of GR binding in response to GC. Furthermore, NSD2 p.E1099K mutation was associated with decreased chromatin accessibility and H3K27ac change in response to GC and decreased (RCH-ACV) or the complete absence (RPMI-8402) of GC-mediated gene induction. NSD2 p.E1099K mutation led to decreased basal GR expression and blocked the ability of GR to autoactivate the expression of NR3C1. Prior work showed that GR levels need to reach a threshold to activate critical downstream targets such as proapoptotic BCL2L11/BIM (27, 44). Our integrated RNA-seq and ChIP-seq data identified genes bound and activated by GR in response to GC in NSD2 WT cells that enriched for proapoptotic genes and genes bound and repressed by dexamethasone treatment including positive regulators of the cell cycle. These genes and pathways represent additional potential targets to augment the effect of GC in ALL. Accordingly, enforced expression of exogenous GR in the NSD2-mutant cells restored GC-mediated cell killing.

Although NSD2 p.E1099K causes a genome-wide increase in H3K36me2 and decrease in H3K27me3, we found that a subset of genes such as NR3C1 and BCL2L11 were repressed in NSD2-mutant cells, in association with the accumulation of H3K27me3 at their promoters. When the NSD2 p.E1099K mutation was removed from ALL cells, H3K27me3 was absent from the NR3C1 promoter, and basal expression of NR3C1 increased. Although GC-mediated GR binding to NR3C1 and BCl2L11 genes was inhibited in NSD2 p.E1099K cells, GR more readily bound to these genes in NSD2 WT cells, correlating with their activation and induction of apoptosis. The NSD2 p.E1099K mutation led to low levels of GR that failed to bind the BCL2L11 IGR and induce H3K27ac and gene activation. However, there was no increase in H3K27me3 at the IGR in NSD2-mutant cells, suggesting that the locus remained accessible and poised for transcriptional activation. GC resistance can be associated with DNA hypermethylation and loss of accessibility of the BCL2L11 IGR, reversed by pretreatment with a DNA methyltransferase inhibitor (23). However, there was no overall increase in methylation of the BCL2L11 IGR in NSD2-mutant RCH-ACV and SEM B-ALL lines, and only a modest increase in RPMI-8402 and CEM T-ALL cells, suggesting that silencing of the BCL2L11 enhancer is not the mechanism of GC resistance in NSD2-mutant cells. This is further backed by the finding that enforced expression of GR in NSD2-mutant ALL cells led to GC-mediated histone acetylation of IGR enhancer and activation of BC2L11 expression.

Pretreatment of NSD2-mutant cells with EZH2 inhibitors removed the repressive H3K27me3 mark from the NR3C1 promoter, allowed GR binding to NR3C1 to be induced in the presence of GC treatment and further increase GR expression, resulting in genome-wide binding and activation of BCL2L11 and other proapoptotic regulators. Prior work suggested that EZH2 inhibitors could augment the antitumor effects of GCs. The EZH2 inhibitor tazemetostat (EPZ-6438) increased the ability of GC to inhibit the growth of lymphoma cell lines and increased expression of GR target genes such as SESTRIN1 and GILZ but did not affect NR3C1 gene expression (45), suggesting a different mode of action from that described here.

Mutations of EZH2 and other PRC2 subunit genes are often found in AML and ALL, promoting transformation by cooperating with other oncogenic mutations (35). Loss of EZH2 in T-ALL is also associated with resistance to chemotherapeutic agents (36). Although NSD2 mutation causes a global loss of H3K27me3, in some ways resembling loss of EZH2, it differs in other ways in that H3K27me3 is redistributed and leads to aberrant gene repression, making inhibition of EZH2 activity a useful therapeutic strategy in both multiple myeloma (29) and ALL. Treatment of NSD2-mutant cells with an EZH2 inhibitor for seven days before the addition of conventional chemotherapy agents, such as vincristine, daunorubicin, or 6-MP, did not block the cytotoxic activity of chemotherapeutic agents, but instead modestly accentuated their ability to inhibit leukemic cell growth. This suggests a functional difference between short-term therapeutic inhibition of the PRC2 complex and the epigenetic rewiring that occurs after permanent loss of PRC2 components that contributes to tumor progression and resistance. Our data indicate that combination therapy of an EZH2 inhibitor with GCs alone and even in combination with chemotherapy may be a highly effective strategy in the treatment of relapsed ALL with NSD2 mutation.

NSD2 aberrantly activates a set of neural and stromal genes not native to the hematopoietic lineages (20), which may play a role in the biology of relapsed disease. Increased gene expression results from the ability of the H3K36me2 histone modification created by NSD2 to antagonize the activity of EZH2 (46). When NSD2 is overexpressed in t(4;14) multiple myeloma, domains of H3K36me2 expand and regions of H3K27me3 shrink, derepressing many genes. Global levels of H3K27me3 decline in NSD2-mutant and overexpressing cells due to the ability of H3K36me2 to antagonize EZH2 activity as increased H3K27 demethylation (14, 20), although the demethylases responsible have not yet been identified. Meanwhile, EZH2 relocalizes and represses other specific gene sets (29, 47). Therapy resistance in NSD2-mutant ALL is linked to the ability of NSD2 p.E1099K to repress basal levels of NR3C1 due to aberrant accumulation of the PRC2, preventing autoactivation of the locus. Our preliminary investigation of the nature of the promoters of genes activated or repressed in the presence of mutant NSD2 indicated that genes activated by NSD2 mutation tended to be enriched for many lymphoid transcriptional activators, whereas the promoters of repressed genes were enriched for a very different set of DNA-binding motifs, including that of the BCL6 transcriptional repressor. We hypothesize that newly replicating chromatin exposed to overactive NSD2 loses H3K27me3 and becomes more accessible. As a result, promoters rich in activator binding sites are more likely to be reactivated by binding highly expressed transcription factors. Promoters lacking such binding sites and having sites for repressors are then targeted by displaced EZH2. Further studies will be required to determine which factors guide aberrant repression of genes such as NR3C1. Nevertheless, our studies suggest a way to overcome this aberrant epigenetic regulation and suggest a specific approach to the treatment of children with relapsed/refractory ALL associated with an activating NSD2 mutation.

In addition to the most prevalent NSD2 p.E1099K mutation, other NSD2 mutations (p.T1150A and p.G1246S) were also identified in relapsed ALL. Interestingly, one PDX sample with the NSD2 p.G1246S mutation was resistant to GC, whereas one with NSD2 p.T1150A was sensitive to GC. Therefore, whether these other NSD2 mutations affect GR expression or function through mechanisms similar to that of NSD2 p.E1099K will require further investigation.

RCH-ACV (DSMZ, RRID:CVCL_1851), SEM (DSMZ, RRID:CVCL_0095), RPMI-8402 (ATCC, RRID:CVCL_1667), CCRF-CEM (ATCC, RRID:CVCL_0207), HPB-ALL (DSMZ, RRID:CVCL_1820), MOLT13 (DSMZ, RRID:CVCL_1422), 697 (DSMZ, RRID: CVCL_0079), SUP-B15 (ATCC, RRID:CVCL_0103), MOLT4 (ATCC, RRID: CVCL_0013), and HEK 293T (ATCC, RRID:CVCL_0063) cell lines were purchased in 2016 and 2017, authenticated using short tandem repeat DNA profiling in 2019, and tested regularly for Mycoplasma contamination. All cell lines were used for the experiments within five passages.

CRISPR/Cas9 gene editing was performed to revert the NSD2 p.E1099K mutation to NSD2 WT in ALL cell line SEM and RPMI-8402 or knock-in this mutation into the CCRF-CEM cell line as described (34). Annealed crRNA:tracrRNA ribonucleotides were loaded onto Cas9 protein (IDT), mixed with a single-strand donor DNA (ssODN) template, and electroporated into ALL cell lines using the Neon Transfection System (Life Technologies). The crRNA recognition sites are NSD2 WT: CAGCATTATCTGCATACCAG; NSD2 p.E1099K mutation: TGGGGAGCTGATCGACAAGG. The ssODNs are NSD2 p.E1099K mutation: GGGTGATGT CGTTCTCGTGTGCGTGCTTGATTCTCGCCATGCACTCTTCCTTGTCGATCAGCTCCCCAACGTACTCGTTAACAAATTCTCCCTGGAAATC; NSD2 WT: GGGTGATGTCGTTCTCGTGTGCG TGCTTGATTCTCGCCATGCACTCTTCCTCGTCGATCAGCTCCCCAACGTACTCGTTAACAAATTCTCCTGGAAATC. Extracted DNA from single colonies was sequenced to identify the genotype.

High-throughput drug screening of isogenic RCH-ACV cell lines was performed at the National Center for Advancing Translational Science (NCATS), NIH, as previously described (48). Briefly, isogenic RCH-ACV NSD2 p.E1099K-mutant and WT cells were plated at 500 cells per well in a 1,536-well plate with a 72-hour incubation with compounds prior to addition of CellTiter-Glo to assess cell viability. NCATS libraries screened include NPC, MIPE 5.0, Kinase, and NPACT. To determine compound activity in the qHTS assay, the concentration-response data for each sample were plotted and modeled by a four-parameter logistic fit yielding IC₅₀ and efficacy (maximal response) values. The area under the curve (AUC) of the dose–response curve ensures that both efficacy (magnitude of cell killing) and potency (concentration that elicits cell killing) are accounted for in the analysis of activity.

HEK 293T cells were transfected with luciferase vector tdTomato-pFU-L2T (gift of Dr. Marcus Peter, Northwestern University, Evanston, IL) and lentiviral packaging plasmids psPAX2 (RRID:Addgene_12260) and VSV-G (RRID:Addgene_12259; both gifts of Didier Trono, School of Life Sciences, École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland), with FuGENE 6 Kit (Promega, #E2312). Viral supernatant was collected and passed through 0.45-μm filter at 48 hours and 72 hours posttransfection. Cells were infected with luciferase virus and sorted by BD FACSAria II (BD Biosciences) using the fluorescent Tomato protein. Expanded isogenic ALL cell lines were used for xenograft experiments.

HEK 293T cells were transfected with FCIV1 and FCIV1-rat GR vector (RRID:Addgene_72701, EGFP; gift of Freddy D. Jeanneteauc, University of Montpellier, Montpellier, France) and lentiviral packaging plasmids (psPAX2 and VSVG) using FuGENE 6 Kit (49). ALL cell lines were infected by packaged lentivirus, and the positive cells were sorted with EGFP using BD FACSAria II (BD Biosciences).

ALL cell lines were cultured in RPMI1640 or Iscove's modified Dulbecco's medium and treated with dexamethasone (Sigma-Aldrich, #D9184; 0–10 μmol/L) for 24 to 72 hours. PDX cells were cultured in QBSF medium and treated with dexamethasone (0–10 μmol/L) for 48 hours. The cells were used for cellular viability, apoptosis, cell cycle, immunoblotting, RNA-seq, ATAC-seq, and ChIP-PCR assays. In vivo, luciferase-expressing ALL cell lines or established ALL PDX cells (3 × 10⁶ cells) were injected by the tail vein into NOD/SCID mice (IMSR catalog no. JAX: 001303, RRID:IMSR_JAX:001303). Tumor growth was monitored by bioluminescence imaging (IVIS Spectrum, Xenogen) or detecting human CD45⁺ (hCD45⁺) cells in peripheral blood. When the luciferase was visible in imaging or hCD45⁺ cells reached 1% in peripheral blood, dexamethasone (Sigma-Aldrich, #D2915; isogenic cell line: 5 mg/kg for 14 days; PDX: 15 mg/kg, 5 days/week) and vehicle (PBS) were administered intraperitoneally. Finally, the leukemic cells were harvested for further experiments. Experimental protocol (#201509176) was approved by the University of Florida Institutional Animal Care and Use Committee.

Isogenic NSD2-mutant cell lines were pretreated with different doses of PRC2 inhibitors, including GSK126 (Active Biochem, #A-1275), EPZ-6438 (Active Biochem, #A-1623-2), UNC1999 (Tocris, #4904), and EED226 (Selleckchem, #S8496) for 7 or 14 days. The cells were divided into two groups. One group was continuously treated with PRC2 inhibitors, the other was treated with PRC2 inhibitors and dexamethasone. Twenty-four to 72 hours later, the cells were used for cellular viability, apoptosis, PCR, immunoblotting, and ChIP-qPCR assays. In vivo, ALL PDX cells (3 × 10⁶ cells) were injected by the tail vein into female NOD/SCID mice. Recipient mice were randomized into four groups (vehicle, GSK126, dexamethasone, GSK126 and dexamethasone). The pretreatment of GSK126 started when hCD45⁺ cells reached 1% in the peripheral blood in PDX recipients. GSK126 (50 mg/kg) or vehicle was administered intraperitoneally in 20% captisol (Selleckchem, #S4592) adjusted to pH 4–4.5 with 1 N acetic acid (33, 34). One week later, dexamethasone (15 mg/kg, 5 days/week) was administered intraperitoneally over four weeks. Tumor growth was monitored by detecting hCD45⁺ cells using a BD Fortessa flow cytometer (BD Biosciences). Finally, the leukemic cells were harvested for further experiments.

Single-cell suspensions from the recipient mice were stained with fluorochrome-conjugated human CD45 antibody (BD, RRID: AB_398600) and detected using a BD Fortessa flow cytometer (BD Biosciences). Apoptosis was detected with the Annexin V FITC Apoptosis Detection Kit (eBioscience, #BMS500FI-300). The harvested cells were washed with PBS, resuspended in 1× Binding Buffer, and incubated with 5 μL of PE Annexin V for 10 minutes and 10 μL of propidium iodide (PI). Stained cells were required by flow cytometry BD Fortessa within one hour. Cell cycle was detected with PI/RNase staining assay (BD Pharmingen, #550825). The harvested cells were washed with PBS and 5 mL of cold 70% ethanol was added drop by drop. Cells were incubated at 4°C for 30 minutes and washed twice to remove the ethanol. To stain, 1 × 10⁶ cells were resuspended in 0.5 mL of PI/RNase staining buffer and incubated for 15 minutes at room temperature prior to flow cytometry on a BD Fortessa flow cytometer (BD Biosciences).

Cell viability was determined by the CellTiter-Glo assay (Promega, #G7572). The opaque-walled multiwell plates with cell lines in culture medium were equilibrated at room temperature for approximately 30 minutes. Equal volume of CellTiter-Glo Reagent to the volume of cell culture medium was added into each well and mixed for two minutes on a shaker to induce cell lysis. The plates were incubated at room temperature for 10 minutes to stabilize luminescent signal. The cellular viability was measured using CLARIOstar (BMG LABTECH).

Cytoplasmic and nuclear fragments were prepared using the Nuclear Complex Co-IP Kit (Active Motif). Proteins were separated using SDS-PAGE and transferred to nitrocellulose membranes using iBlot 2 (Invitrogen). Histone modifications and protein levels were determined using antibodies including anti-H3K36me2 (Abcam, RRID:AB_1280939), anti-H3K27me3 (Cell Signaling Technology, RRID:AB_2616029), anti-H3 (Cell Signaling Technology, RRID:AB_2756816), anti-GR (Cell Signaling Technology, RRID:AB_11179215), anti-Max (Santa Cruz Biotechnology, RRID:AB_627913), anti-Tubulin (Cell Signaling Technology, RRID:AB_2288042), anti-BIM (Abcam, RRID:AB_725697), and anti-Vimentin (Cell Signaling Technology, RRID:AB_10695459) followed by staining with Alexa Fluor 680 donkey anti-rabbit IgG (Invitrogen, RRID:AB_2633283) or Alexa Fluor 790 goat anti-mouse IgG (Invitrogen, RRID:AB_2633279). The membranes were scanned using an Li-Cor Odyssey CLx Imaging System (Li-Cor).

Total RNA was isolated with the Direct-zol RNA MiniPrep Plus Kit (Zymo Research, #R2070), and cDNA was synthesized with the iScript cDNA Synthesis Kit (Bio-Rad, #1708841). Primers were designed and purchased from IDT (Supplementary Table S7). Quantitative real-time PCR was performed with SYBR Green Supermix (Bio-Rad, #1725271) using a CFX96 Touch Real-Time PCR Detection System (Bio-Rad) and normalized to PPP1R15B, a gene whose expression was consistently unchanged in response to GC treatment in all NSD2-mutant and WT cell lines.

Total RNA was extracted with the Direct-zol RNA MiniPrep Plus Kit. Library preparation and sequencing were performed by Novogene using the TruSeq RNA Library Preparation Kit and NovaSeq 6000 (Illumina) with PE150. Short reads were filtered and trimmed using trimmomatic (v 0.36; RRID:SCR_011848). Quality control (QC) on the original and trimmed reads was performed using FastQC (v 0.11.4; RRID:SCR_014583) and MultiQC (v 1.1; RRID:SCR_014982). The reads were aligned to the transcriptome using STAR (v 2.7.3a; RRID:SCR_004463). Transcript abundance was quantified using RSEM (v 1.2.31; RRID:SCR_013027). Differential expression analysis was performed using DESeq2 (RRID:SCR_015687), with an FDR-corrected P value threshold of 0.05 and fold change of 1.5. Gene Ontology analysis of biological processes was performed using Enrichr (ref. 50; RRID:SCR_001575), WebGestalt (ref. 51; RRID:SCR_006786), and Venny 2.1.0 (RRID:SCR_016561).

ChIPmentation was performed as previously described (52). ALL cell lines were cross-linked with 0.8% formaldehyde and quenched with 0.125 mol/L glycine. Extracted chromatin was sonicated using Covaris E220 Focused-ultrasonicator (Covaris). After preclearing with Prot.A/G Dynabeads (Invitrogen, #10001D and 10003D), DNA was immunoprecipitated using antibodies including anti-GR (Cell Signaling Technology, RRID:AB_11179215), anti-CTCF (Cell Signaling Technology, RRID:AB_ 2086791), anti-H3K27ac (Active Motif, RRID:AB_2561016), anti-H3K27me3 (Cell Signaling Technology, RRID:AB_2616029), anti-H3K36me2 (Cell Signaling Technology, RRID: AB_1030983), and IgG (Cell Signaling Technology, RRID:AB_1031062) overnight at −4°C. Precipitated chromatin was washed and incubated in Tagmentation buffer containing Tagment enzyme (Illumina, #20034210). DNA was purified after reversal cross-linking. ChIP-qPCR was performed with SYBR Green Supermix using a CFX96 Touch Real-Time PCR Detection System (Bio-Rad) and normalized to IgG control (Primers: Supplementary Table S7). For ChIP-seq, the purified DNA samples were amplified for the library with index primers using the KAPA HiFi HotStart Kit (Roche, #7959028001). The libraries were cleaned, size-selected using Ampure XP Beads (Beckman Coulter, #A63880), and pooled for Novaseq 6000 S4 2 × 150 flow cell (ICBR Next-Gen Sequencing Core, University of Florida). The input sequences were trimmed using trimmomatic. QC was performed before and after trimming using FastQC. The input sequences were then aligned to the GRCh38 genome using Bowtie (v 2.3.5.1; RRID:SCR_016368). Peak detection was performed using MACS (v 2.1.2; RRID:SCR_013291), and motif finding on peak regions was performed with the HOMER (RRID:SCR_010881) FindMotifs function.

Nuclei were isolated from 200,000 cells, and ATAC-seq was performed on aliquots of 50,000 nuclei as described (53). ATAC-seq libraries were purified by size selection using Ampure XP Beads, and library quality was assessed by fragment analyzer and qPCR for GAPDH and a genomic desert region (Primers: Supplementary Table S7). Sequencing of pooled libraries was performed on a Novaseq 6000 S4 2 × 150 flow cell (ICBR Next-Gen Sequencing Core, University of Florida). Reads were trimmed using trimmomatic (v 0.36), and QC on the original and trimmed reads was performed using FastQC (v 0.11.4) and MultiQC (v 1.1). The reads were aligned to the human genome version GRCh38 using Bowtie (v. 2.3.3), and ATAC peak calling was performed using the MACS (v 2.1.2). Differential peak analysis was performed using DASA (https://github.com/uf-icbr-bioinformatics/dasa). Custom scripts were used to produce heat maps and enrichment plots.

Statistical analyses were performed using the GraphPad Prism 8 software (RRID:SCR_002798). The statistical significance was evaluated using two-way ANOVA (Tukey multiple comparisons test), one-way ANOVA (Dunnett multiple comparisons test), and unpaired t test. Survival of NOD/SCID mice was estimated with the log-rank test and represented with Kaplan–Meier curves. Statistical significance was considered as *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

All next-generation sequencing data have been deposited to the NCBI Sequence Read Archive (PRJNA669144). Drug screen data have been deposited to PubChem (AID 1645876 and 1645877). Source data for all main and supplementary figures are available in the Supplementary Data.

A.A. Ferrando reports grants from NIH–NCI and Columbia University during the conduct of the study; grants from Leukemia and Lymphoma Society, SEAS-HICCC, Alex Lemonade Stand Foundation, The Pershing Square Sohn Cancer Research Alliance, The Chemotherapy Foundation, and NIH–NCI, personal fees from Bristol-Myers, Ayala Pharmaceuticals, VantAI, and SpringWorks Therapeutics outside the submitted work; in addition, A.A. Ferrando has a patent 20070077245 issued, a patent 20100087358 issued, a patent 20110118192 issued, a patent 20130064810 issued, a patent 8633179 issued, a patent 20150299801 issued, a patent 20160102367 issued, a patent 9574241 issued, a patent 20100093684 issued, a patent 8716233 issued, and a patent 9624491 issued; and additional financial interests: royalties from the commercial distribution of the CUTLL1 cell line. R.B. Lock reports grants from Cancer Australia, The Kid's Cancer Project, the Anthony Rothe Memorial Trust, The National Health and Medical Research Council Australia, and the Cancer Institute NSW during the conduct of the study. J.D. Licht reports grants from Celgene and grants and personal fees from Samuel Waxman Cancer Research Foundation during the conduct of the study. No disclosures were reported by the other authors.

J. Li: Conceptualization, resources, data curation, software, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. J. Hlavka-Zhang: Resources, data curation, formal analysis, and methodology. J.H. Shrimp: Resources, data curation, software, formal analysis, and methodology. C. Piper: Resources, data curation, investigation, and methodology. D. Dupéré-Richér: Investigation, methodology, writing–original draft. J.S. Roth: Data curation, formal analysis, investigation, and methodology. D. Jing: Data curation, formal analysis, funding acquisition, investigation, and methodology. H.L. Casellas Román: Data curation and methodology. C. Troche: Methodology. A. Swaroop: Methodology. M. Kulis: Methodology. J.A. Oyer: Methodology. C.M. Will: Methodology. M. Shen: Data curation, software, formal analysis, and methodology. A. Riva: Data curation, software, and formal analysis. R.L. Bennett: Conceptualization, resources, data curation, investigation, methodology, writing–original draft, project administration, writing–review and editing. A.A. Ferrando: Conceptualization, writing–review and editing. M.D. Hall: Conceptualization, resources, supervision, funding acquisition, writing–review and editing. R.B. Lock: Conceptualization, resources, supervision, funding acquisition, writing–review and editing. J.D. Licht: Conceptualization, resources, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing.

This research was supported by R01 CA195732, a Leukemia and Lymphoma Society Specialized Center of Excellence Grant, the Samuel Waxman Cancer Research Foundation, Celgene, Florida Department of Health 8LA01, the Lauri Strauss Leukemia Foundation, the Harry T. Mangurian, Jr. Foundation (all to J.D. Licht), the Leukemia and Lymphoma Society Special Fellow Grants (3399-20 to J. Li and 3392–19 to D. Dupéré-Richér), the Rally Foundation for Childhood Cancer Research and Bear Necessities Pediatric Cancer Foundation 19FN10 and 20IC26 (J. Li), the Priority-driven Collaborative Cancer Research Scheme and cofunded by Cancer Australia and The Kids' Cancer Project Grant APP1129129 (to R.B. Lock), the Anthony Rothe Memorial Trust (D. Jing and R.B. Lock), the National Health and Medical Research Council of Australia NHMRC Fellowships APP1059804 and APP1157871 (R.B. Lock), and the Cancer Institute NSW Early Career Fellowship 15/ECF/1-02 (D. Jing). J.H. Shrimp, J.S. Roth, M. Shen, and M.D. Hall acknowledge the funding from the Intramural Research Program, NCATS, and NIH. Data analysis was performed at the Bioinformatics and Flow Cytometry Core at the Interdisciplinary Center for Biotechnology Research (UF ICBR), a shared research facility under UF Research and the UF Health Cancer Center at the University of Florida. We thank UF Research for supporting use of the HiPerGator High Performance Computing Infrastructure.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.