The NSD2 p.E1099K Mutation is Enriched at Relapse and Confers Drug Resistance in a Cell Context Dependent Manner in Pediatric Acute Lymphoblastic Leukemia

Abstract

The NSD2 p.E1099K (EK) mutation is observed in 10% of acute lymphoblastic leukemia (ALL) samples with enrichment at relapse indicating a role in clonal evolution and drug resistance. To discover mechanisms that mediate clonal expansion, we engineered B-ALL cell lines (Reh, 697) to overexpress wildtype (WT) and EK NSD2, but observed no differences in proliferation, clonal growth, or chemosensitivity. To address whether NSD2 EK acts collaboratively with other pathways, we used shRNAs to knockdown expression of NSD2 in B-ALL cell lines heterozygous for NSD2 EK (RS4;11, RCH-ACV, SEM). Knockdown resulted in decreased proliferation in all lines, decreased clonal growth in RCH-ACV, and increased sensitivity to cytotoxic chemotherapeutic agents, although the pattern of drug sensitivity varied among cell lines implying that the oncogenic properties of NSD2 mutations are likely cell context specific and rely on cooperative pathways. Knockdown of both Type II and REIIBP EK isoforms had a greater impact than knockdown of Type II alone, suggesting that both SET containing EK isoforms contribute to phenotypic changes driving relapse. Furthermore, in vivo models using both cell lines and patient samples revealed dramatically enhanced proliferation of NSD2 EK compared to WT and reduced sensitivity to 6-mercaptopurine in the relapse sample relative to diagnosis. Finally, EK-mediated changes in chromatin state and transcriptional output differed dramatically among cell lines further supporting a cell context specific role of NSD2 EK. These results demonstrate a unique role of NSD2 EK in mediating clonal fitness through pleiotropic mechanisms dependent on the genetic and epigenetic landscape.

INTRODUCTION

While five-year survival rates for newly diagnosed pediatric acute lymphoblastic leukemia (ALL) now approach 90%, up to 20% of children will suffer relapse and face a poor prognosis (1). Thus, ALL relapse remains a major cause of death related to cancer in children (2, 3). While new immunological approaches are quite promising, leukemia subclones continue to emerge through the selective pressures of therapy (4). Therefore, targeting the underlying biological pathways of therapy resistance is crucial for preventing and treating relapse.

Comprehensive genomic profiling of diagnosis/relapse pairs from patients with B-precursor ALL (B-ALL) has identified relapse-enriched mutations and copy number changes that confer drug resistance (5–9). Mutations in epigenetic regulators are among the most commonly seen alterations at relapse and have been identified in almost two-thirds of cases (10). We have previously demonstrated the vital role of epigenetic changes in mediating drug sensitivity making targeting such lesions a promising therapeutic strategy (11).

NSD2 (MMSET, WHSC1) encodes a histone 3 lysine 36 (H3K36) methyltransferase that catalyzes the mono- and di-methylation of H3K36. NSD2 alterations have been identified in a variety of cancers, most notably the t(4;14) translocation in multiple myeloma (MM), which confers a poor prognosis (12). The substitution of glutamic acid to lysine at residue 1099 (p.E1099K, EK) within the conserved SET domain occurs in up to 10% of ALL patients (13, 14). NSD2 EK results in increased methyltransferase activity leading to a global increase in H3K36me2 levels and stereotactic inhibition of EZH2-mediated H3K27 trimethylation. This has also been observed in t(4;14) MM and has been shown to lead to increased proliferation, enhanced DNA damage repair, and resistance to DNA damaging agents (13–16).

NSD2 consists of three distinct isoforms associated with oncogenesis: Type I, Type II, and response element II-binding protein (REIIBP) (Supplemental Fig. 1) (17, 18). The canonical isoform, Type II, is the largest, spanning 1365 amino acids with Type I representing the N-terminus and REIIBP the C-terminus of the full-length isoform. While Type I results from alternative splicing, REIIBP is transcribed from a separate internal transcription start site. Type II and REIIBP contain the conserved SET domain responsible for methyltransferase activity and regulation of gene expression through histone methylation (19). Type I can also alter gene expression, however the mechanism(s) remains poorly understood (17). Notably, Type I and II localize to the nucleus, whereas REIIBP localizes to the cytoplasm and nucleolus. Unlike Type II, the impact of EK on REIIBP function has yet to be fully categorized.

Herein, we provide further evidence to support the role of NSD2 EK in pediatric B-ALL disease progression and now demonstrate underlying cellular context has a vital impact on the epigenetic and phenotypic alterations mediated by NSD2 EK. Our data suggest NSD2 EK requires cooperative pathways to exert its pleiotropic effects on cell cycle, proliferation, clonogenicity, and drug response. Our results also suggest that both NSD2 Type II and REIIBP contribute to these effects.

MATERIALS AND METHODS

Cell culture, drug preparation, viral preparation, immunoblotting, phenotypic assays, and mouse xenografts were performed according to methods published previously (7, 8, 20) and additional information can be found in Supplemental Information.

Cells and reagents

The B-lineage leukemia cell lines RS4;11, Reh (ATCC, Manassas, VA), KOPN8, 697, RCH-ACV (DSMZ, Braunschweig, Germany), and SEM (kindly gifted by Jun Yang, St. Jude Children’s Hospital) were grown in RPMI1640 medium. HEK293T (ATCC) cells were grown in DMEM medium. All media were supplemented with 10% FBS, 1% penicillin/streptomycin under 5% CO2 at 37°C. No cell lines were used beyond passage 20. Each leukemia line was validated by short tandem repeat analysis through ATCC except for RCH-ACV which was purchased from DSMZ directly. DSMZ routinely verifies cell lines and provides authentication information with each order. Cell lines were routinely monitored for mycoplasma contamination by PCR using ATCC Universal Mycoplasma Detection Kit (20–1012K).

Primary patient sample gene expression analysis

Primary patient samples were obtained from the Children’s Oncology Group (COG) Biobank. All subjects provided consent for banking and future research use of these specimens in accordance with the regulations of the institutional review boards of all participating institutions. Whole genome sequencing and/or whole exome sequencing was performed as previously described (9). Data were aggregated for somatic SNVs/indels detected in diagnosis/relapse tumors of each case to generate two-dimensional scatter plots to visualize variant allele frequency (VAF) between diagnosis and relapse tumors. A small number of somatic variants, including NSD2 EK, were subject to ultra-deep targeted re-sequencing (>500,000X) to define their pre-existence status in diagnosis tumors. CleanDeepSeq was used to obtain allele counts of corresponding markers (21). To detect somatic mutations with very low frequency, WT genomic sites flanking the designed amplicons with same genotype as the marker of interest were used to derive a background error-rate histogram of corresponding mutation type and the VAF of the marker of interest was compared against the background error-rate histogram by using Z-test (two-sided). Markers with P < 0.001 were called somatic in the corresponding sample.

In vivo xenograft experiments

Patient-derived xenograft (PDX) models using non-obese diabetic/severe combined immunodeficiency (NOD/SCID/Il2rgtm1wjl/SzJ) mice were generated from a diagnostic (NSD2 WT) and relapse (NSD2 EK) pair as previously described (22). Mice (4–5 per treatment arm, per sample) were randomized to receive 6-mercaptopurine [6-MP] (25 mg/kg/day by gavage) or vehicle control following the detection of at least 1% peripheral blood (PB) blasts (defined as CD19+/CD45+ by flow cytometry) for 5 days per week until sacrifice. Average disease burden at the time of treatment initiation was not different between treatment and control arms. Disease burden was assessed weekly by flow cytometric measurement of PB and splenic and bone marrow blasts were compared at sacrifice using published techniques (8).

RNA-seq

Gene expression was assessed by RNA-seq according to standard protocols (20). RNA was extracted using the QIAGEN RNeasy Mini Kit and quality was verified by an Agilent Bioanalyzer 2100 (PICO chip). RNA-seq libraries were sequenced using 54 base pair reads on the Illumina Genome Analyzer GAIIx. Image collection and analysis was completed using the Illumina CASAVA pipeline. Adapters from RNA-seq paired-end reads were trimmed and low-quality bases (<30) were removed using Trimmomatic (v0.33) (23). Alignment was performed using STAR (v2.5.3a) (24) to human genome (hg19) and bases with mapping quality <30 were removed. Raw counts of sequencing reads were obtained from HTseq2. Normalized genome browser tracks were generated using BEDTools (v2.26.0) (25). Differential gene expression analysis was performed using DESeq2 (26). For each gene, a two-sample t-test was applied to obtain the p-value for significance of differential expression between test and control. Genes detected as differentially expressed (p<0.002) were defined as up or down modulated according to the sign of t-statistics. All other genes were classified as “not changed” in expression. Genes with absolute fold change ≥1.5 and p-value ≤0.05 were selected for pathway analysis with Enrichr and KEGG 2016 to determine pathways significantly altered as indicated by combined scores (p-value and z-score) (27, 28).

ChIP-seq

Cells were cross-linked in 1% formaldehyde, lysed and then sheared chromatin was immunoprecipitated with antibodies targeting histone marks [H3K36me2 (Abcam, ab9049), H3K27me3 (Abcam, ab6002), H3K27ac (Abcam, ab4729) and H3K9ac (Millipore, 06–942)]. A small aliquot was set aside prior to immunoprecipitation as input DNA. Quality control was performed using an Agilent Technologies 2100 Bioanalyzer prior to library preparation. ChIP-seq libraries were generated using KAPA HyperPrep kit, quantified using KAPA qPCR kit and sequenced using the HiSeq 2500 for paired-end 50-bp reads. ChIP-seq paired-end reads were trimmed and low-quality bases (<30) were removed using Trimmomatic (v0.33) (23). Reads were mapped to the human genome (hg19) using bowtie2 and bases with mapping quality <30 were removed. Due to the broad/diffuse peaks created by H3K36me2 and H3K27me3, peaks for these marks were called by SICER that uses a cluster enrichment based analysis (29). H3K27ac and H3K9ac peaks were called using MACS2 (--broad) (30). Differential binding analysis of peak data was performed using DiffBind (31) and nearest genes were annotated using ChIPseeker (32). H3K27ac-promoter regions, H3K9ac-promoter regions, H3K36me2-promoter and gene body regions (exclude intergenic regions) were selected for further downstream analysis. RNAseq heatmaps were aligned with the histone mark called for each gene from DiffBind output files to assess effect of the mark on gene expression. Bigwig files and profile plots were generated using deepTools2 (v2.3.3) using RPGC normalization (--smoothLength 2500 --binSize 500) (33). Profile plots were generated across significantly up- and downregulated genes and immediate upstream and downstream 10kb regions were divided into 50 bins of 200 bp each to compare knockdown (NSD2-low) lines against the NT control (NSD2-high). If the closest upstream gene was at least 30 kb away from the TSS of the current gene, the intergenic region was visualized to detect profile patterns in those regions using the first and last 10 kb of each intergenic region. For distribution analysis regions were defined in mutually exclusive manor as promoter = plus/minus 5kb from TSS, gene body = exon + intron + utr, and intergenic = intergenic + downstream regions.

ChIPseq and RNAseq data were then integrated to determine whether the observed epigenetic changes were associated with commensurate changes in transcriptional output. The log2 fold changes in histone peaks at promoters (for activation marks) and promoter/gene body (repressive marks) were then compared to the log2 fold changes in gene expression. Box plots were generated to display changes in marks relative to gene expression. A similar approach was performed for enhancers and superenhancers which were associated with the nearest neighboring gene.

ATAC-seq

ATAC libraries were generated based on the protocol by Buenrostro et al (34) with one modification. Cells were lysed via centrifugation for 1 minute at 500xg. Nuclei were tagmented using Nextera (Illumina) Tagmentation DNA buffer and enzyme. PCR amplification was performed as described in protocol. Reads were mapped to the human genome (hg19) using Bowtie2(v2.3.4.1) and those with a mapping quality <30 were removed. Duplicated reads were removed using Sambamba (v0.6.8). Peaks were called using MACS2(v2.1.1). Differential binding analysis of peaks was performed using DiffBind. Peaks were assigned to nearest neighboring gene.

Data Deposition

All raw data files generated through high-throughput sequencing discussed in this publication have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo) and are accessible through GEO Series accession number GSE149159.

Statistical analysis

Statistical significance was calculated using unpaired t-test or one-way ANOVA with post-hoc Tukey test for IC50 significance.

RESULTS

NSD2 EK is Detected at Diagnosis as a Minor Subclone in a Majority of Patients who Harbor Dominant EK Mutations at Relapse

In a recently collected cohort of matched diagnosis/remission/relapse samples from patients enrolled on COG Relapse B-ALL trial (AALL1331), NSD2 EK was present at relapse in 7/82 trios. A previously published analysis of 20 trios also harbored NSD2 EK at relapse (9) resulting in an overall frequency at relapse of 12%. Among all 10 NSD2 EK relapse cases (Fig. 1), three cases had similar VAFs at diagnosis and relapse while four cases revealed NSD2 EK as part of a minor subclone at diagnosis (VAF <0.1) that evolved into the major clone at relapse (VAF 0.49–0.72). Similarly, three patients harbored NSD2 EK solely at relapse suggesting that these mutations were acquired or that a minor clone may have been present at diagnosis but below sensitivity of detection. The enrichment of NSD2 EK at relapse in 7/10 patients implies a role in shaping clonal evolution and drug resistance.

Figure 1:

Patients Harboring NSD2 EK at Relapse in COG diagnosis/relapse Paired Cohorts. (A & B) Enrichment of NSD2 EK mutation at relapse is observed in 7 out of 10 Patients. Variant allele frequency (VAF) at diagnosis and relapse for 10 B-ALL patient samples.

Overexpression of NSD2 WT and EK in Leukemia Cell Lines Has No Impact on Tumor Phenotype

To determine the role of NSD2 EK in clonal evolution, B-ALL cell lines (Reh and 697) were engineered to overexpress c-myc tagged Type II WT or EK NSD2. As a control, both cell lines were stably infected with the empty destination vector (EV). Overexpression of both WT and EK resulted in increased H3K36me2 and decreased H3K27me3 consistent with enzymatic hyperactivity (Supplemental Fig. 2a). However, overexpression of either WT or EK did not affect clonogenicity, cell proliferation, or drug sensitivity (Supplemental Fig. 2b–d). Furthermore, we overexpressed either the WT or EK REIIBP isoform in Reh cells. Like Type II overexpression, REIIBP overexpression was insufficient to impart oncogenic properties in a NSD2 WT background B-ALL cell line (Supplemental Fig. 3a–c).

NSD2 EK Requires Cooperating Oncogenic Pathways to Impact Tumorigenicity and Drug Resistance

Due to the lack of an oncogenic phenotype in our overexpression experiments, we hypothesized NSD2 EK might require other cooperating pathways to endow leukemic cells with a clonal advantage. Therefore, we generated NSD2 knockdown (KD) cell lines using short-hairpin (sh) RNAs targeting different NSD2 isoforms in B-ALL cell lines that naturally harbor a heterozygous NSD2 EK mutation (RS4;11, RCH-ACV, and SEM) (Fig. 2a and Supplemental Fig. 1). RS4;11 and SEM both contain KMT2A-AFF1 translocations but SEM also contains a CDKN2a deletion and a TP53 mutation. RCH-ACV harbors a TCF3-PBX1 translocation and mutations in EGFR, HRAS and NRAS. For a full list of all mutations in each cell line, please refer to the Broad Institute’s Cancer Cell Line Encyclopedia (CCLE, https://portals.broadinstitute.org/ccle).

Figure 2:

Knockdown of NSD2 in B-ALL Cell Lines Harboring EK Mutations Leads to Decreased Proliferation and Colony Formation in a Cell Context Dependent Manner. (A) Western blot analysis of whole cell lysates from RS4;11, RCH-ACV and SEM NSD2 knockdown cell lines. (B) Proliferation curves of RS4;11, RCH-ACV and SEM NDS2 knockdown cell lines as counted by trypan blue over 7–10 days. Each cell line was plated in triplicate with bars representing the mean +/− Standard Deviation. A statistical significance of <0.001 for all lines was determined by nonlinear regression exponential growth. (C) MethoCult™ colony forming assay in RCH-ACV NSD2 knockdown cell lines plated at 1000 cells/400 uL in duplicate per cell line. Colonies were stained with MTT and counted after 14 days in culture. Statistical significance determined by unpaired t test *p<.05, **p<.01.

NSD2 KD in cell lines harboring the heterozygous NSD2 EK mutation resulted in decreased proliferation compared to non-targeting (NT) controls (Fig. 2b). NSD2 KD in RCH-ACV demonstrated decreased colony formation (Fig. 2c) similar to what was reported previously in RS4;11(35). To determine if the observed phenotypes was due to reduced expression of mutant NSD2 specifically, expression was also knocked down in WT NSD2 cell lines (Reh, 697, and KOPN8) (Supplemental Fig. 4a). No differences in growth or clonogenic survival were found upon NSD2 KD in any of the WT cell lines, suggesting the phenotypic changes were due to reduction of NSD2 EK specifically (Supplemental Fig. 4b–c).

Notably, NSD2 KD in heterozygous EK lines also resulted in increased chemosensitivity (Fig. 3 and Supplemental Table 1). The RS4;11 NSD2 KD lines were significantly more sensitive to mercaptopurine (6-MP) (sh2: 3.4-fold lower IC50; p<.0001, sh5: 1.3-fold; p<.003), but not to other agents, including thioguanine (6-TG) (Fig. 3a). The impact of decreased NSD2 expression in RCH-ACV cells created a more diverse chemotherapy response profile with increased sensitivity to 6-MP (sh2: 4.0-fold; p<.0001, sh5: 2.4-fold; p<.0001), 6-TG (sh2: 1.4-fold; p<.01, sh5 1.5-fold; p<.005), prednisolone (Pred) (sh2: 2.6-fold, p<.001; sh5: 2.3-fold p<.003) and doxorubicin (Dox) (sh2: 1.9-fold; p<.007, sh5: 1.2-fold, n.s.) relative to NT (Fig. 3b). No difference was observed with methotrexate (MTX) (Supplemental Table 1). Activated caspase 3 levels were measured using flow cytometry after 120 hours of exposure to 6-MP, 6-TG, and Dox in RS4;11 and RCH-ACV NT and KD cells (Supplemental Fig. 5a–b). A significantly greater percentage of cells had activated caspase 3 in 6-MP treated RS4;11 sh2 and 6-MP, 6-TG, and Dox treated RCH-ACV sh2 lines versus their NT controls, which verified the cytotoxicity results across cell types. Additionally, like Swaroop et. al, we observed increased apoptosis in RCH-ACV NSD2 KD cells at baseline, however caspase activation following treatments were significantly higher than baseline levels (Supplemental Fig. 5b) (36). SEM KD and NT cells had similar responses to 6-MP, 6-TG, MTX, and Dox treatment (Fig. 3c). Paradoxically, in SEM, only sh5 showed increased sensitivity to Pred (7.8-fold; p<.009) (Fig. 3c), suggesting a role for Type I in glucocorticoid resistance, consistent with a recent report demonstrating Type I can modulate gene expression (17).

Figure 3:

Knockdown of NSD2 in B-ALL Cell Lines Harboring EK Mutations Leads to Changes in Drug Response in a Cell Context Dependent Manner. (A-C) Representative cytotoxicity curves assessed by CellTiter-Glo® of RS4;11(A), RCH-ACV(B) and SEM (C) NSD2 knockdown cell lines exposed to various chemotherapy agents, each plated in triplicate and bars represent mean +/− Standard Deviation. All experiments were repeated at least three times. IC50 dot plots are shown as insets. Dots represent mean from individual experiments. Bar represents median. Statistical significance determined by One-way ANOVA *p<.05, **p<.01, ***p<.001.

Overall, sh2, which targets both EK containing isoforms, had a more dramatic impact on the cellular phenotypes than sh5. To address whether these findings were due to the greater degree of knockdown observed with sh2 (Fig. 2a), a third shRNA, sh1 (which also targets Type II and REIIBP) was tested (Supplemental Fig. 1). This resulted in less of a reduction in NSD2 expression compared to sh2, yet also showed analogous 6-MP sensitivity. (Supplemental Fig. 6). These findings suggest that both isoforms may be involved in mediating the impact of NSD2 EK. . Furthermore, in contrast to mutant lines, NSD2 KD of SET containing isoforms in WT lines (Reh, 697, and KOPN8) failed to show any impact on response to 6-MP, 6-TG, Dox, or MTX compared to NT controls (Supplemental Fig. 7 and Supplemental Table 1). Collectively, these results again indicate NSD2 EK has a pleiotropic impact on drug resistance that is dependent upon cell context. The one exception was KD in 697 and KOPN8 (NSD2 WT) cells which led to a paradoxical increase in resistance to Pred (1.9-fold; p<.0003, 58.6-fold; p <.005, respectively), suggesting expression of NSD2 WT actually sensitizes these cells to glucocorticoids.

In order to assess potential oncogenic properties of NSD2 EK in vivo, xenograft models were generated from isogenic cell lines and a diagnostic and relapse primary B-ALL patient sample (PAWWLL) that harbored NSD2 EK only at relapse (patient-derived xenograft, PDX). Importantly, the effects on growth were supported by significantly decreased tumor burden in mice injected with RS4;11 cells expressing sh2 compared to NT (Fig. 4a). However, we were not able to recapitulate the response to 6-MP treatment likely due to the much slower growth of EK knock down cells in vivo (sh2 2-fold lower in vitro vs >10-fold lower in vivo). Likewise, the relapse patient sample harboring EK resulted in more aggressive disease compared to the diagnosis sample as evidenced by comparing absolute splenic blast count at sacrifice in both relapse and diagnostic samples treated with vehicle controls (Fig. 4b). While both the relapse and diagnostic samples retained sensitivity to 6-MP, the relative reduction in absolute blast count was far greater at diagnosis (25-fold) compared to relapse (11-fold decrease) (Fig. 4b). Furthermore, bone marrow blast percentage was dramatically reduced with 6-MP treatment in the diagnosis sample relative to vehicle control whereas the relapse sample showed no difference post-treatment (Fig. 4c). Western blot analysis demonstrates increased H3K36me2 in a sample from the relapse-derived mice, suggesting hyperactivation of NSD2 (Fig. 4d). While this in vivo model supports the role of NSD2 EK in proliferation and resistance to 6-MP, it is difficult to ascertain if these findings are solely due to the gain of NSD2 EK since additional mutations are acquired at relapse. (Supplemental Table 2).

Figure 4:

NSD2 EK Leads to Significant Impact on Tumor Growth in vivo. (A) Quantification of average flux determined by analyzing BLI of mice injected with RS4;11 NT or sh2 expressing cells treated with Vehicle or 6-mercaptopurine (6-MP). Bars represent mean +/− Standard Deviation. N=4 per condition. (B&C) Quantification of absolute splenic blast count as calculated by spleen cell number (x10^6) x percentage of blasts in spleen (B) and bone marrow blast percent (C) at sacrifice from Patient Derived Xenograft (PDX) generated from diagnosis (NSD2 WT)/relapse (NSD2 EK) patient pair. Mice were treated with vehicle (n=5 for both diagnosis and relapse) or 6-MP (25 mg/kg/day) (n=4 for diagnosis, n=5 for relapse) 5 days a week until sacrifice. Bars represent mean +/− Standard Deviation. (D) Western blot analysis of whole cell lysate from cells taken from spleen of mice showing increased H3K36me2 in mouse from relapse sample (“R”) compared to mouse from diagnosis (“D”) sample.

A major mechanism of cell death mediated by both 6-MP and 6-TG relates to incorporation of thioguanine nucleotides (TGN) into DNA leading to double stranded breaks, cell cycle arrest, and apoptosis (7). Thus, the impact of NSD2 KD on 6-MP, but not 6-TG sensitivity seen in RS4;11, was unexpected. To investigate this selectivity, cell cycle progression and DNA-TG incorporation were measured following thiopurine treatment. Over the course of 6-MP treatment, the RS4;11 KD cells demonstrated a greater degree of cell cycle arrest, as demonstrated by a decrease in cells actively cycling (p<.01), compared to the NT cells. However, following 6-TG treatment, NT and sh2 expressing cells displayed an equivalent drop in the percentage of cells cycling (Fig. 5a). These data support our cytotoxicity results that revealed increased sensitivity to 6-MP but not 6-TG. In contrast, KD in RCH-ACV cells led to a significantly greater decrease in cycling cells over time compared to NT upon treatment with both 6-MP (p<.01) and 6-TG (p<.01) (Fig. 5b). Furthermore, RCH-ACV KD cells showed increased DNA-TG levels over time after 6-MP treatment compared to NT (p<.001), consistent with increased sensitivity through the DNA-TG-induced nucleotide mismatching and futile mismatch repair attempts. In contrast, while RS4;11 KD showed slower accumulation of DNA-TGs, presumably due to slower proliferation, by 120 hours of exposure to 6-MP, NT and KD cells exhibited no significant difference of DNA-TGs (Fig. 5c). SEM NT and KD cell lines did not reveal significant differences in cell cycle progression or TGN incorporation upon treatment with 6-MP, as expected based on the cell viability data (data not shown). However, this short-term exposure may not reliably mimic the long-term maintenance therapy of ALL. Still, these data indicate NSD2 EK can impart thiopurine resistance through alternative mechanisms in addition to circumventing the canonical thiopurine induced mismatch repair dependent process.

Figure 5:

NSD2 EK Mediates Resistance to 6-MP through Multiple Mechanisms. (A&B) Cell cycle analysis measured by EdU incorporation for RS4;11 (A) and RCH-ACV (B) cell lines. Percentage of cells cycling were normalized to untreated controls at each time point. The IC50 of the more sensitive line (sh2) was chosen for this analysis. Representative flow cytometry images at 120 hours shown on right. Rectangle represents EdU positive (cycling) cells. C) DNA-TGN incorporation was measured using liquid chromatography-tandem mass spectrometry following 6-MP treatment of RS4;11 (left) and RCH-ACV (right) lines. Bars represent mean +/− Standard Deviation. Statistical significance determined by unpaired t test *p<.05, **p<.01, ***p<.001.

NSD2 EK Activates Unique Transcriptional Programs That Are Cell Context Dependent

To determine the downstream transcriptional program mediated by NSD2 EK, RNAseq was performed on Reh and 697 cells overexpressing either WT or EK NSD2, and EV controls (Supplemental Table 3). Interestingly, in both Reh and 697, the WT and EK lines (relative to EV controls) shared approximately 50% of upregulated and 35% of downregulated genes (Supplemental Fig. 8a) suggesting that while WT and EK NSD2 do share some transcriptional targets, EK may modulate a unique set of genes. Furthermore, when comparing EK overexpression in Reh cells to 697 cells, only 5 upregulated genes (ADAP2, PCNXL2, PLXNA2, SCBA and NSD2) and one suppressed gene (FLRT3) were shared with similar results noted between Reh and 697 WT cell lines (Supplemental Fig. 8b). The limited overlap in differentially regulated genes and pathways modulated by overexpressed EK and WT NSD2 between cell lines implies the transcriptional programs modulated by NSD2 are quite dependent upon cell context (Supplemental Fig. 8c).

To determine the extent to which NSD2 mediated epigenetic changes play a role in gene expression, we performed ChIPseq on Reh WT and EK NSD2 overexpression lines. As expected, we saw a global increase in the H3K36me2 mark across the genome as well as a global decrease in the H3K27me3 mark. When aligning the significantly differentially expressed up and downregulated genes from RNAseq alongside the respective ChIPseq data, we observed a strong correlation between gain of H3K36me2 and loss of H3K27me3 with increased gene expression, indicating NSD2 mediated epigenetic changes drive gene expression in cells with hyperactivated NSD2 (Supplemental Fig. 9).

We next examined gene expression in our NSD2 KD cell lines naturally harboring the EK mutation by RNAseq (Supplemental Table 3). We first compared transcriptional programs between NT (NSD2^high) and sh2 (NSD2^low) lines. We observed only 4.4% (143/3257) of upregulated genes and 1.7% (39/2319) of downregulated genes were shared between all three cell lines (Fig. 6a) indicating significant diversity in transcriptional reprogramming by NSD2 between cell types. Pathway analysis on genes modulated by NSD2 reduction in each cell line revealed few common pathways shared among cell lines (focal adhesion, and cell adhesion molecules), some of which are in agreement with previously published data (Supplemental Fig. 10) (37). As our functional assays implied a possible role for REIIBP, we also examined differences in gene expression modulated by sh2 (Type II and REIIBP) and sh5 (Type I and II). Interestingly, reduction of Type II/REIIBP resulted in a greater impact on gene expression compared to reduction of Type I/II further supporting a possible role of REIIBP in NSD2-mediated oncogenesis (Fig. 6b). Notably, the majority of genes modulated by sh5 are a subset of those modulated by sh2, again suggesting REIIBP may modulate gene expression and impact the phenotype of NSD2 EK cancer cells.

Figure 6:

Knockdown of NSD2 in NSD2 EK B-ALL Cell Lines Reveals Distinct Genetic Signatures and Chromatin Accessibility. (A) Overlap of differentially expressed genes, assessed by RNAseq, in RS4;11, RCH-ACV and SEM NT (NSD2^high) relative to each cell line’s sh2 knockdown using an absolute fold change of 1.5, p<0.05. (B) Overlap of shared up- (top) and downregulated (bottom) genes in NT vs. sh2 against NT vs. sh5 in RS4;11, RCH-ACV and SEM. (C) Quantification of peaks gained or lost, as assessed by ATACseq, in RS4;11, RCH-ACV and SEM NT relative to each cell line’s sh2 knockdown using an absolute fold change of 1.5, p<0.05 (D) Overlap of nearest genes to peaks gained or lost among the three lines. Promoters defined as +/− 5kb from TSS, Non-promoters defined as all other regions excluding promoters.

To examine the impact of NSD2 on chromatin accessibility, we also performed ATACseq in our NSD2 KD cell lines naturally harboring NSD2 EK (Supplemental Fig. 11b). Similar to RNAseq analysis, we compared open and closed chromatin regions between NT (NSD2^high) and sh2 (NSD2^low) lines. The most variable region of peak alterations was within the intergenic space (Figure 6c). Interestingly, RCH-ACV and RS4;11 NSD2^high lines each had significant gains of open chromatin within their intergenic regions, whereas SEM NSD2^high displayed marked peak loss within their intergenic regions indicating a decrease in accessible chromatin. Within promoter regions, RCH-ACV and RS4;11 showed modest increases in peak gains, whereas SEM displayed the most peak losses. Gains and losses of ATAC promoter peaks correlated with increases and decreases in transcript expression as assessed by RNAseq in RS4:11 and RCH-ACV but no correlation was observed in the SEM cell line (Supplemental Fig 11a). Gene bodies were the least variable region as chromatin accessibility was unchanged across all three NSD2^high cell types. Interestingly, only 0.11% of promoter region gains and 1.3% of intergenic regions with peak gains were shared among all three cell lines (Fig. 6d), consistent with our RNAseq data showing very little overlap of changes in transcriptional output across cell lines. Overall, the data indicate NSD2 EK effects chromatin accessibility in a cell context dependent manner with the most profound impact on accessibility within intergenic regions.

Lastly, to examine changes in the epigenetic landscape driven by NSD2 EK, we performed ChIPseq on our RS4;11 NT and KD lines. We saw a dramatic increase in the H3K36me2 mark across intergenic regions (Fig. 7a–b) with a pronounced decrease in H3K27me3 evenly across the genome (data not shown) in NT (NSD2^high) compared to KD (NSD2^low), which coincides with dramatic changes in chromatin accessibility within the intergenic regions as well as previous data reported in MM (37). We noted a strong correlation of increased gene expression with an increase in the H3K36me2 at promoters and gene bodies as well as an increase in promoter distribution of the activating H3K9ac and H3K27ac marks (Fig. 7c). Despite the genome-wide heightened H3K36me2, a subset of genes displayed a decrease in expression along with loss of H3K36me2 in RS4;11 NT relative to the KD (Fig. 7d). Likewise, in spite of the genome wide loss of H3K27me3, a small subset of genes maintained this repressive mark and lost expression compared to the KD line (Fig. 7d). These findings suggest that NDS2 EK mediated epigenetic changes that contribute to gene expression are due to direct histone methyltransferase activity of NSD2 as well as downstream effects on the epigenetic landscape.

Figure 7.

Knockdown of NSD2 in NSD2 EK B-ALL Cell Lines Leads to a Unique Epigenetic Signature that Modulates Gene Expression. (A) Pie chart of distribution of histone marks as assessed by ChIPseq. (B) Profile plots of distribution of H3K36me2 in RS4;11 NT and sh2 in intergenic (top) and intragenic (bottom) regions of the genome. Underneath each are IGV genome browser views of H3K36me2 distribution within a representative locus on chromosome 8. (C) Box plot of changes in histone marks as assessed by ChIPseq analysis in RS4;11 NT (relative to NSD2 sh2) aligned with gene expression data (**p<.01, ***p<.001, ****p<.0001). (D) Heatmaps of top differentially expressed genes aligned with histone marks between RS4;11 NT and sh2.

DISCUSSION

Relapsed B-ALL remains a leading cause of cancer mortality in children. The outgrowth of pre-existing intrinsic or secondary drug resistant subclones under the selective pressures of chemotherapy accounts for the majority of disease recurrences (38, 39). Mutations in epigenetic modifiers are found in a majority of patients at relapse, suggesting a role for the epigenome in mediating clonal evolution. NSD2 EK occurs in up to 10% of B-ALL and appears to be particularly enriched in ETV6/RUNX1 and TCF3/PBX1 biological subtypes (9, 14, 35). Furthermore, a recent publication identified NSD2 mutations were enriched at relapse in B-ALL, all of which were early relapses (<36 months from diagnosis) further supporting a role in drug resistance (40). Our analysis of relapse samples with NSD2 EK mutations shows that in the majority of cases (7/10), either the mutation was not detected at diagnosis or was present in a very minor subclone indicating a role in clonal evolution.

In our aim to discover the underlying mechanism(s) that drive NSD2 EK mediated clonal evolution, we utilized two preclinical models. Unlike previous studies in MM and ALL that modulated endogenous expression in cell lines harboring a NSD2 EK allele (16, 37), we demonstrate that overexpression of WT NSD2 (mimicking MM) or EK NSD2 (mimicking B-ALL) in WT B-ALL lines do not lead to enhanced oncogenic properties, despite increased H3K36me2 and decreased H3K27me3. Conversely, knockdown of NSD2 in cell lines that harbor heterozygous NSD2 EK resulted in diminished oncogenic properties, which was not observed upon NSD2 KD in WT lines, implying that NSD2 EK requires collaborative pathways to exert oncogenic properties. These results agree with previous work where knockdown of NSD2 in individual cell lines harboring the mutation resulted in decreased proliferation and clonal growth (14, 35, 37), but herein we demonstrate significant differences among cell lines in these properties and, importantly, a highly variable impact on chemosensitivity to individual agents used in ALL treatment. Moreover, the impact of NSD2 EK on proliferation and drug resistance was confirmed using in vivo models.

While we and others have shown relapse enriched somatic alterations result in either pan-resistance (6, 8) or resistance to a single class of agents such as glucocorticoids or thiopurines (20, 41–43), remarkably, our data shows that NSD2 EK imparts resistance to a variety of agents depending on cell type. Knockdown of NSD2 EK in RS4;11 resulted in increased sensitivity to 6-MP and in RCH-ACV, widespread sensitivity to 6-MP, 6-TG, Dox, and Pred. Li et al. also observed increased glucocorticoid sensitivity in RCH-ACV (44). No alteration in drug sensitivity was seen in SEM mediated by the knockdown of both SET containing isoforms. This varied pattern of resistance further supports a model where NSD2 EK collaborates with other oncogenic programs to impart a clonal advantage.

Unexpectedly, NSD2 EK confers resistance to 6-MP exclusively in RS4;11 and significantly more resistance to 6-MP versus 6-TG in RCH-ACV. Generally, both prodrugs manifest cell death through the generation and incorporation of TGNs into DNA, but we validated a novel mechanism of resistance in RS4;11 by demonstrating no difference in DNA-TG levels between NT and NSD2 KD cells. Unlike 6-TG, 6-MP may impact cellular metabolism by inhibition of de novo purine synthesis and GTP homeostasis (45, 46). Elucidating the exact mechanism(s) by which NSD2 EK mediates 6-MP resistance is critical, as 6-MP is a backbone of ALL maintenance therapy.

The phenotypic diversity among cell lines was also seen when examining changes in chromatin state and transcriptional output across cell lines. Similar to previous work in t(4;14) MM by Popovic et al, our ChIPseq analysis of RS4;11 showed EK mediated H3K36me2 enrichment was distributed widely across the genome and not confined to promoters and gene bodies (15). Upon further analysis of changes in chromatin structure mediated by modulation of NSD2 in the three cell lines containing NSD2 EK alleles, we demonstrate diversity in the amount and location of differential peaks yet, within all lines, most changes occurred in intergenic regions, an observation also seen in MM (15). Interestingly, RS4;11 NT displays a majority of peak gains in the intergenic region that complements our ChIPseq findings, which demonstrated gains of H3K36me2 in intergenic regions. Presence of NSD2 EK impacted chromatin accessibility in RCH-ACV and SEM lines mostly in the intergenic regions as well. However, cell context clearly has a role since the activating mutation conveys a variable signature in each cell line as overlap of peaks between cell types is limited.

The importance of cell context was reinforced by RNAseq analysis of knockdown of NSD2 in heterozygous EK cell lines where little overlap was seen in transcriptome changes. Minimal overlap was also observed between the NSD2 EK up- and downregulated pathways seen in each individual cell type. Interestingly changes in chromatin accessibility determined by ATACseq did not always correlate with the magnitude of changes measured by RNAseq. For example, SEM NSD2^high possessed the largest set of upregulated genes, but displayed the greatest loss of chromatin accessibility at the promoter and the intergenic regions. Conversely, RCH-ACV NSD2^high possessed the largest number of downregulated genes, but displayed the greatest gains of accessibility at the promoter and the intergenic regions. Taken together, the greatest influence of NSD2 EK may not be related to proximal transcription elements, but rather on distal regulatory regions that impact transcriptional reprogramming (47).

Our study provided additional novel insights related to the biological properties of the NSD2 EK mutation in ALL. First, by comparing the impact of overexpressed WT NSD2 (mimicking MM) vs. NSD2 EK (mimicking B ALL), we are able to show that despite some overlap in changes in the transcriptional output within the same cell line, a remarkable degree of diversity exists. The EK substitution occurs within the SET domain in a loop proximal to the substrate binding pocket which may contribute to alterations in NSD2 substrate binding. This hypothesis is supported by a recent report that showed NSD2 EK has greater affinity for chromatin and enhanced nucleosome complex stability (37), which may explain the transcriptional diversity observed between our NSD2 WT and EK overexpression lines. A second unique aspect of our study is our ability to assess the possible contribution of both SET containing isoforms in transformation and clonal evolution. In RS4;11 and RCH-ACV, combined knockdown of Type II and REIIBP (sh2) demonstrated a greater reduction in cell proliferation and more pronounced effect on chemosensitivity compared to cells with simultaneous knockdown of Type I and Type II (sh5). In a t(4;14) MM cell model, the concurrent reduction of both SET domain containing isoforms also imparted the most drastic impact on cell phenotypes (48). While both SET isoforms have been implicated in the methylation of many histone substrates and play roles in transcriptional regulation, DNA repair, and RNA processing, the major contribution of Type II to oncogenic programming appears to be related to generation of H3K36me2 (49). Studies have demonstrated REIIBP has H3K27 methyltransferase activity resulting in transcriptional repression (50), H3K79 methyltransferase activity resulting in transcriptional activation (51), and a role in RNA processing (18). Therefore, distinct aspects of both the EK activating mutation and differences in the function and transcriptional reprogramming of Type II and REIIBP may account for some of the cell context specific differences noted.

Overall, our data indicate NSD2 EK can endow B-ALL cells with a clonal advantage through divergent mechanisms depending on the cell context, mediated by unique genetic or epigenetic signatures. In this regard, the mechanism of action of NSD2 EK is significantly different from other relapse-enriched alterations that are associated with specific resistance to individual agents used in therapy such as thiopurines or glucocorticoids. The pleotropic effect mediated by NSD2 EK may provide tumor cells with even greater plasticity to respond to the selective pressures of therapy.

IMPLICATIONS.

NSD2 p.E1099K mutation leads to drug resistance and a clonal advantage in childhood B-ALL.