KDM6B protects T-ALL cells from NOTCH1-induced oncogenic stress

T-cell acute lymphoblastic leukemia (T-ALL) is an aggressive hematopoietic neoplasm resulting from the malignant transformation of T-cell progenitors. While activating NOTCH1 mutations are the dominant genetic drivers of T-ALL, epigenetic dysfunction plays a central role in the pathology of T-ALL and can provide alternative mechanisms to oncogenesis in lieu of or in combination with genetic mutations. The histone demethylase enzyme KDM6A (UTX) is also recurrently mutated in T-ALL patients and functions as a tumor suppressor. However, its gene paralog, KDM6B (JMJD3), is never mutated and can be significantly overexpressed, suggesting it may be necessary for sustaining the disease. Here, we used mouse and human T-ALL models to show that KDM6B is required for T-ALL development and maintenance. Using NOTCH1 gain-of-function retroviral models, mouse cells genetically deficient for Kdm6b were unable to propagate T-ALL. Inactivating KDM6B in human T-ALL patient cells by CRISPR/Cas9 showed KDM6B-targeted cells were significantly outcompeted over time. The dependence of T-ALL cells on KDM6B was proportional to the oncogenic strength of NOTCH1 mutation, with KDM6B required to prevent stress-induced apoptosis from strong NOTCH1 signaling. These studies identify a crucial role for KDM6B in sustaining NOTCH1-driven T-ALL and implicate KDM6B as a novel therapeutic target in these patients.


INTRODUCTION
The genomics era has facilitated remarkable advances in understanding the mechanisms driving the biology and pathogenesis of many blood cancers such as acute lymphoblastic leukemia (ALL). T-cell ALL (T-ALL) accounts for approximately 15% of pediatric and 25% of adult ALL cases respectively. Compared to the more common B-cell ALL, adult T-ALL cases are historically linked with a poor prognosis [1], due to unfavorable clinical features such as high white blood cell count, bulky adenopathy, and central nervous system (CNS) involvement [2]. While the advent of highdose, multi-agent chemotherapy regimens has resulted in longterm event-free survival (EFS) rates of up to 90% in pediatric T-ALL, this clinical benefit has not carried over to the adult T-ALL patient population which have 5-year EFS rates of 40-50%. Clinical outcomes of patients with primary resistant or relapsed leukemia remains poor, and adult T-ALL patients remain at increased risk for induction failure and early relapse [3,4]. The differences in treatment outcomes are likely related to underlying genetic differences between children and adults. More specific therapies are needed for high-risk adult T-ALL patients based on their genetic profile.
T-ALL arises from the accumulation of genomic abnormalities that induce aberrant proliferation, increased cell survival, and impaired differentiation of immature T-cell progenitors. In the past decade, sequencing approaches have identified more than 100 genes mutated in T-ALL patient genomes [5], with NOTCH1 gain-offunction variants being the most common oncogenic drivers [6]. However, efforts to target NOTCH1 have had limited success due to major off-target toxicities [7]. Studies have also revealed a high frequency of mutations in epigenetic regulators in T-ALL patients [8], with approximately 35% of adult T-ALL cases harboring a mutation in some component of the epigenetic machinery [9]. This includes frequent mutations in components (e.g. EZH2, SUZ12, JARID2) of the polycomb repressive 2 (PRC2) complex which establishes the repressive epigenetic mark of H3K27me3 [10,11]. The gene encoding the histone lysine demethylase KDM6A, which catalyzes removal of H3K27me3, is also recurrently mutated in these patients and is thought to function as a cell-type and context-dependent tumor suppressor in this disease [12]. Conversely, its gene paralog KDM6B, also known as JMJD3, is never mutated in T-ALL [13], but rather is often overexpressed in these patients [12] and other hematological disorders [14,15].
This dichotomy presented the hypothesis that while KDM6A may function as a tumor suppressor in T-ALL, KDM6B may actually be necessary for leukemic cell function in this blood cancer.
In a previous study, we demonstrated with mouse models that Kdm6b was necessary for leukemia development in certain contexts [16]. Here, we show KDM6B is required for T-ALL maintenance using mouse and human genetic models. Utilizing a knockin mouse model harbouring point mutations in Kdm6b that abolish its catalytic activity confirmed that the pro-tumorigenic role of Kdm6b in T-ALL was largely related to its molecular function as a histone demethylase enzyme. Analysis of human T-ALL patient specimens determined that the genetic dependence of these cells on KDM6B was directly related to the signal strength of different NOTCH1 mutations, with the patients with most robust NOTCH1 signaling being most dependent on KDM6B. KDM6B was required to protect these cells from stress-induced apoptosis, a finding corroborated in mouse models using a "weak" Notch1 mutation which indicated no reliance on Kdm6b for T-ALL cell function in this mutational background. Cumulatively, these studies reveal a novel role for KDM6B as a genetic dependency in T-ALL driven by strong NOTCH1 gain-of-function mutations and highlight a new therapeutic target for this patient population.

Western blot
Cells were washed twice with PBS, sonicated for 5 min, and the cell lysates were loaded on SDS-PAGE gels, transferred to polyvinylidene fluoride membranes (Fisher Scientific, IPVH00010), and blocked with 5% non-fat milk in Tris-buffered saline with 0.1% Tween 20 detergent (TBST). The blots were incubated with the indicated primary antibodies at 4°C overnight. Western blots were probed with antibodies to detect NICD (Cell signaling, #4147 S) and β-ACTIN (Santa Cruz, SC-47778). The next day, the blots were washed with TBST and incubated with the suitable mouse or rabbit secondary antibody (Fisher Scientific, 55965-84-9, 10794347) at room temperature for one hour. Detection was performed using horseradish peroxidase-conjugated secondary mouse or rabbit antibody and chemiluminescence HRP substrate (Millipore #WBKLS0100). The blots were then visualized on a Bio-Rad ChemiDOC Touch Imaging System (Bio-Rad Laboratories).

RNA-seq
RNA was isolated using the NucleoSpin RNA XS kit (Macherey-Nagel #740902.250). The SMARTer Ultra Low RNA kit (Clontech) was used to prepare libraries for RNA-seq as per manufacturer instructions. Sequencing was performed with an Illumina NovaSeq S4 2 × 150. Analysis was performed using Partek flow software. Raw RNA-seq reads were aligned to the Ensembl release 99 (Mouse), and to the Ensembl release 105 (Human) using STAR version 2.7.3a. Quality control was performed to confirm the efficacy of the alignment and the quality of the reads. Gene counts were normalized and Gene Specific Analysis (GSA) was performed on the normalized counts to generate the differential analysis. Gene set enrichment analysis (GSEA) was generated to identify dysregulated gene sets. Primary data is available under GEO accession number GSE214576.

CRISPR/Cas9 and targeted deep sequencing
Synthetic guide RNAs (gRNAs) were designed using the UCSC Genome browser software. gRNAs were selected to target the catalytic domain of KDM6B to maximize specificity (lowest off-target effects) and efficiency (on-target cleavage efficiency) [21]. The four highest-scoring gRNAs were tested by nucleofecting HEL cells. 48-hours post-nucleofection, DNA was extracted from cells in each group and sent for next-generation sequencing (NGS). The two gRNAs with the highest targeting efficiency were used for experimentation. Primary human T-ALL cells were nucleofected with Cas9/ribonucleoprotein (IDT, #1074181) complexed with gRNAs as previously described [22]. Synthetic gRNAs (Synthego) sequences are as follows: KDM6B_18a: TGCTCCGTCAACATCAACAT, KDM6B_18d: CGCGGTGCACGAGCACTACT. A gRNA targeting the AAVS1 locus was used as a negative control: GGGGCCACUAGGGACAGGAU. Postnucleofection, cells were left to recover in StemSpan SFEM II Media (StemCell Technologies #09655) supplemented with Pen-Strep (50 Units/ mL), human stem cell factor (SCF; 50 ng/mL), human IL-2 (50 U/mL), and human IL-7 (10 ng/mL). 48 hours later, approximately 100,000 cells were set aside to measure targeted variant allele fraction (VAF) using PCR amplicon-based deep sequencing. The remaining cells were transplanted into NSG mice via tail vein injections, where each mouse received 250,000 nucleofected cells. Serial monthly bleeds were performed where CD45 + CD7+ human T-ALL cells were purified, DNA was extracted using the PureLink Genomic DNA Mini Kit (Thermo Fisher Scientific, # K1820-02), and VAF was monitored over time. The following primer pairs were used to generate amplicons: AAVS1_Forward-ACAGGAGGTGGGGGTTAGAC plus AAVS1_Reverse-CCCCTATGTCCACTTCA; KDM6B_18_Forward-CAGCCAAT GAGGGCAGAG plus KDM6B_18_Reverse-CACAGGTCAGGTGGGAACTG. Amplicon-based deep-sequencing libraries were sequenced using the Illumina MiSeq platform and data analyzed using CRISPResso2 [23].

AnnexinV staining
Cells were stained with antibodies to identify T-ALL cells (mouse: CD45.2+ GFP+; human: hCD45 + CD7+), then washed twice with cold PBS and incubated at room temperature for 15 min in 1x AnnexinV binding buffer N. Issa et al.

Ki67 cell cycle assay
Cells were stained with antibodies to identify T-ALL cells (mouse: CD45.2+ GFP+; human: hCD45 + CD7+), then washed twice with cold PBS. The cell pellet was loosened by vortexing and three mL of cold 70% ethanol was added dropwise to the cell pellet while vortexing. Samples were incubated with 70% ethanol at −20°C for one hour. After fixation, samples were washed three times with the Cell Staining Buffer (BioLegend, # 420201) then stained with fluorochrome-conjugated Ki67 antibody (BioLegend, # 652413) at 1:100 dilution at the concentration of 1.0 × 10 6 /mL. Samples were incubated in the dark for 30 min, then washed twice with Cell Staining Buffer, and resuspended in 0.5 mL cell staining buffer for flow cytometric analysis.

Study approval
All animal procedures were approved by the Institutional Animal Care and Use Committee at Washington University. Human T-ALL patient samples were obtained with written consent in accordance with the Declaration of Helsinki protocol. Because all patient samples were de-identified and the study team had no access to individual patient health information (PHI), the Washington University Institutional Review Board (IRB) and Human Research Protection Office (HRPO) determined this to be a nonhuman study.

RESULTS
Kdm6b is essential for maintenance of NOTCH1-mutant T-ALL cells A retroviral model of NOTCH1 intracellular domain (NICD) expression that recapitulates many of the human pathologies of NOTCH1mutant T-ALL [24] was used to examine the role of Kdm6b in T-ALL development. A conditional knockout model was generated by crossing the Vav-Cre driver [18]  To circumvent any potential complications of initial T-ALL cell engraftment, an analogous experiment was performed using cells from inducible Mx1-Cre:Kdm6b fl/fl mice. NICD-GFP transduced Sca-1+ BM cells were transplanted and then deletion of Kdm6b was induced 5-weeks post-transplant when engraftment levels were comparable (Supplementary Fig. 2A). Following injection of pIpC, there was no difference in overall survival between recipients of Mx1-Cre:Kdm6b fl/fl and Mx1-Cre:Kdm6b +/+ (control) T-ALL cells ( Supplementary Fig. 2B). However, while injection of pIpC was relatively efficient at inducing Mx1-Cre and recombination of floxed alleles in Mx1-Cre:Kdm6b fl/fl T-ALL cells one-week post-treatment, the cells that underwent recombination were rapidly outcompeted by three-weeks post-treatment ( Supplementary Fig. 2C). Thus, the observed survival outcome of Mx1-Cre:Kdm6b fl/fl recipients was likely due to T-ALL cells that escaped recombination rapidly outcompeting cells successfully induced for Kdm6b deletion. This is an indirect indication that Kdm6b is necessary for the maintenance of NOTCH1-mutant T-ALL cells.

T-ALL cells undergo apoptosis in the absence of Kdm6b
To begin to understand the molecular mechanism of functional dependence for NOTCH1-mutant T-ALL cells on Kdm6b, gene expression analysis was performed by RNA-seq on T-ALL cells (CD45.2+ NICD-GFP+) isolated from the blood of recipient mice four-weeks post-transplant. Comparison of gene expression profiles of control and Kdm6b-KO (Supplemental Table 1) identified 210 differentially expressed genes (>2-fold change, p < 0.05, FDR < 0.05). The majority of these genes showed decreased expression in Kdm6b-KO T-ALL cells ( Fig. 2A). Gene set enrichment analysis (GSEA) identified three significantly differential pathways-"E2F targets", "G2M checkpoint", and "mitotic spindle", all of which were downregulated in Kdm6b-KO T-ALL cells (Fig. 2B). Given this, we hypothesize that Kdm6b may be necessary for cell cycle progression of T-ALL cells. To examine this, Ki67 cell cycle assay was performed on T-ALL cells from peripheral blood of recipient mice at four-weeks post-transplant (Fig. 2C). Kdm6b-KO T-ALL cells showed a slight yet significant decrease in the proportion of cells in the G 2 /M phase of cell cycle (Fig. 2D). While this was suggestive of a slight cell cycle arrest of Kdm6b-KO T-ALL cells, it appeared unlikely that this relatively minor difference could account for the large discrepancy in T-ALL cell maintenance.
To determine if the cell cycle differences had any impact on cell survival, apoptosis analysis was performed on T-ALL cells by AnnexinV staining (Fig. 2E). There was a striking increase in T-ALL cell apoptosis in the absence of Kdm6b (Fig. 2F). This suggests that the primary function of Kdm6b in NOTCH1-mutant T-ALL cells may be to prevent leukemia cell apoptosis.
Pro-tumorigenic function of Kdmb6 in T-ALL is demethylasedependent KDM6B is described to function primarily as a H3K27me3 histone demethylase enzyme [25,26], yet recent studies have suggested that KDM6B may have other demethylase-independent activities [13,27,28]. To determine if the function of Kdm6b in T-ALL was dependent on histone demethylase function, a mouse model was utilized with two engineered point mutations in the same germline allele of Kdm6b (Kdm6b H1388A;H1390A ) that abolish the catalytic activity of the protein [25,29]. As homozygous Kdm6b H1388A;H1390A mice are embryonic lethal (Dr. Kai Ge, unpublished data), we crossed Kdm6b H1388A;H1390A/+ mice to Vav-Cre:Kdm6b fl/fl mice to generate Vav-Cre:Kdm6b fl/H1388A;H1390A such that the only copy of Kdm6b expressed in hematopoietic cells is catalytically dead (CD). The NICD retroviral transduction and transplantation model was repeated to include this genotype. The significant difference in survival of recipient mice transplanted with NICD-GFP expressing Kdm6b-HET and Kdm6b-KO cells allowed evaluation of the catalytic function of Kdm6b in this activity. If the survival of mice receiving NICD-GFP expressing Kdm6b-CD cells resembled that of Kdm6b-KO cells, then the function of Kdm6b in T-ALL is demethylase-dependent. However, if the survival of these mice mirrored that of NICD-GFP expressing Kdm6b-HET cells, this suggests the function of Kdm6b in T-ALL is histone demethylase-independent. Despite comparable levels of initial engraftment, analysis of recipient mice eight-weeks posttransplant revealed a dearth of Kdm6b-CD T-ALL cells in the peripheral blood (Fig. 3A). T-ALL cells were unable to propagate in a Kdm6b-CD genetic background, replicating the overall survival of mice receiving Kdm6b-KO cells (Fig. 3B). The few mice that developed Kdm6b-KO and Kdm6b-CD T-ALL in these experiments were found to have retained one copy of the wild-type Kdm6b allele (functionally a Kdm6b-HET). These results strongly suggest that the function of Kdm6b in sustaining NOTCH1-mutant T-ALL cells is demethylase-dependent.
KDM6B is required to sustain disease in a subset of T-ALL patients To investigate the translational potential of our findings, KDM6B was inactivated in primary human patient T-ALL cells by CRISPR/ Cas9 gene targeting. Four guide RNAs (gRNAs) were designed to target the exons encoding the catalytic domain of KDM6B [30]. The two gRNAs that showed the highest targeting efficiency in HEL cells were used for experimentation. A gRNA that targets the inert AAVS1 locus was used as a negative control. For genetic consistency, only NOTCH1-mutant T-ALL patient samples were used in these experiments (Table 1). gRNA/Cas9 ribonucleoprotein (RNP) complexes [31] were nucleofected into primary human T-ALL cells and 48-hours post-nucleofection, 2.5 × 10 5 nucleofected cells were xenografted into three NSG mice per gRNA per patient, with the remaining T-ALL cells processed for genomic DNA extraction to quantify initial CRISPR targeting efficiency by amplicon-based deep sequencing. In general, no significant differences were noted between the AAVS1 and KDM6B gRNA groups across different patient samples as assessed by monthly analysis of the peripheral blood, time to morbidity, or leukemic burden in the BM and spleen of moribund mice. However, analysis of the variant allele fraction (VAF) of CRISPR-targeted cells over time revealed a striking trend and segregated the response of patient cells into two distinct entities. For 3/8 adult T-ALLs, the VAF of the edited KDM6B alleles remained relatively constant over time, suggesting that these specific patient samples were insensitive to genetic loss of KDM6B (Fig. 4A, Supplementary  Fig. 3A; "non-responders"). In contrast, 5/8 T-ALL samples showed a marked reduction in KDM6B VAF over time, showing the edited cells were rapidly outcompeted in vivo (Fig. 4B, Supplementary  Fig. 3B). Patient samples that showed >50% decrease in VAF of KDM6B at the conclusion of the transplant compared to the input were termed "responders". Analysis of VAF over the experimental time course showed that in most specimens, the VAF of control AAVS1 edited alleles remained relatively consistent (Fig. 4C). However, a similar comparison of the VAF of KDM6B edited alleles showed that while the targeted cells were efficiently retained in the non-responder patient specimens, the targeted cells were rapidly outcompeted over time in the responder T-ALL patient cells (Fig. 4C). While the focus of these translational studies was adult T-ALL samples due to the inferior clinical outcomes of this demographic, we also performed analogous CRISPR/Cas9 experiments with a limited set of NOTCH1-mutant pediatric T-ALL primary specimens. Of the three samples tested, one classified as a non-responder ( Supplementary Fig. 4A) whereas two acted as responders with rapid depletion of KDM6B targeted cells (Supplementary Fig. 4B). This was reinforced by dynamic comparisons of AAVS1 versus KDM6B VAFs over the experimental time course ( Supplementary  Fig. 4C). Cumulatively, these experiments with primary T-ALL patient samples suggests that as in mouse models, KDM6B was required for the efficient propagation of T-ALL cells, but only for a specific subset of patients.
We sought to determine if genetic dependency on KDM6B in T-ALL patients correlated with clinical outcomes. For the three  pediatric specimens subject to functional analysis, all of the patients survived their T-ALL. The two KDM6B "responder" pediatric patients achieved durable complete remission (CR) after induction chemotherapy, whereas the KDM6B "non-responder" suffered CNS relapse after the initial response, but underwent successful allogeneic stem cell transplantation. Only one of the eight adult T-ALL patients surveyed in this study survived their disease (a KDM6B "responder"). Due to the complex treatment histories and transplant complications, it was not possible to correlate clinical outcomes with KDM6B reliance in this relatively small patient cohort.

KDM6B restrains expression of pro-apoptotic genes in KDM6B-dependent T-ALLs
To explore the potential clinical application of KDM6B inhibition, the small molecule GSK-J4 has been proposed as a H3K27 histone demethylase family inhibitor that has evidence of effectiveness in T-ALL [12]. We exposed two "responder" and two "non-responder" patient samples to increasing concentrations of GSK-J4 but did not note any differences in sensitivity (Fig. 5A). Moreover, the sensitivity was not affected by CRISPR-mediated KDM6B gene deletion (Fig. 5B). Thus, GSK-J4 does not appear to be specific enough for KDM6B inhibition in T-ALL, and any benefit may be due to inhibition of KDM6A or other off-target effects on αketoglutarate-dependent enzymes [32].
To begin to elucidate the mechanisms of KDM6B dependence in human patient samples, adult "responder" and "non-responder" T-ALL specimens were targeted with AAVS1 and KDM6B gRNAs and cultured in vitro for seven days. Due to the rapid depletion of KDM6B-targeted cells in vivo, this approach was taken to study molecular profiles of KDM6B-inhibited T-ALL while the tumor still contained a high fraction of edited cells. Global transcriptomic analysis was performed after confirming the high targeting efficiency of the edited cells (Fig. 5C) RNA-seq data showed that samples grouped more strongly by individual patient heterogeneity rather than dependence on KDM6B (Fig. 5D). Due to this, further analysis was performed to aggregate responder versus non-responder samples under control conditions (AAVS1 gRNA) compared KDM6B inhibition (KDM6B gRNA). Under control conditions, only 216 genes were differentially expressed between responders and non-responders (>2-fold expression difference, p-value <0.05). However, following KDM6B targeting, 1229 genes became significantly differentially expressed (Supplemental Table 2). GSEA of the two specimen     groups under control conditions revealed many of the same pathways identified in mouse RNA-seq such as "E2F targets" and "G2M checkpoint" (Fig. 5E). However, GSEA of the KDM6B targeted cells (Fig. 5F) revealed specific enrichment of "apoptosis" pathway genes (Fig. 5G) that was not observed in the AAVS1-targeted cell comparison. Thus, KDM6B appears necessary for preventing apoptosis in a subset of NOTCH1-mutant human T-ALL patients, corroborating previous findings in mouse models. To analyze this further, AnnexinV staining was performed on T-ALL patient specimens cultured in vitro for seven days following CRISPR with either AAVS1 or KDM6B gRNAs. While some responder patients showed a subtle yet significant increase in the proportion of apoptotic cells following targeting with KDM6B (Fig. 5H), in general, this genetic manipulation was insufficient to induce robust T-ALL cell death (Fig. 5I). This is likely due to complicating factors such as variable CRISPR targeting efficiency, cell culture conditions optimized to maintain viability of T-ALL cells, and the lack of cell competition/selective pressures of an in vivo experiment.
Genetic signature of KDM6B dependence in T-ALL We attempted to integrate transcriptome and mutation profile to identify why only a subset of T-ALL patient specimens were sensitive to KDM6B inhibition. Expression levels of KDM6B itself or the paralog KDM6A were not different between adult responders and non-responders (Fig. 6A). Exome sequencing data of human T-ALL patient samples was examined to determine if the dependence on KDM6B for certain specimens correlated with the type of NOTCH1 mutation or the nature of co-operating mutations. The VAF of NOTCH1 mutations in the primary tumor samples was not different between responders and nonresponders (Fig. 6B). Analysis of the mutational distribution mutations showed that the NOTCH1 mutations found in nonresponders were all localized between amino acids 1585-1593 in the N-terminus of the heterodimerization (HD) domain (Fig. 6C).
Mutations in the NOTCH1 HD domain act as gain-of-function by inducing ligand-independent activation of the receptor [33,34], but the mutation pattern was not significantly different between responders and non-responders (Fig. 6C). With regard to cooperating mutations, the sample size precluded generation of definitive conclusions, however, some patterns of mutational cooperation emerged (Fig. 6D). Patients in the responder group were more likely to have multiple NOTCH1 mutations with 2/7 patients having >1 NOTCH1 mutation (Fig. 6E) compared to 0/4 in the non-responder group. Additionally, FBXW7 mutations were identified in 2/7 responder patients (mutually exclusive of responder patients with multiple NOTCH1 mutations) compared to 0/4 non-responder patients with FBXW7 mutations (Fig. 6D). FBXW7 is a ubiquitin ligase that negatively regulates stability of NICD. Mutations in FBXW7 occur in~15% of T-ALL patients [35,36] which abrogate binding to NICD, thereby prolonging its half-life and potentiating NOTCH1 signaling. Both mutational patterns observed in KDM6B responders (multiple NOTCH1 mutations, NOTCH1 plus FBXW7 mutations) serve to increase the strength of NOTCH1 signaling in T-ALL cells.
KDM6B protects T-ALL cells from strong NOTCH1 signalinginduced apoptosis Cancers must develop mechanisms of tolerating various infidelities of oncogene-induced cellular stress. This includes T-ALL where NOTCH1 mutations hijack the cellular stress response machinery to facilitate transformation by a number of mechanisms including upregulation of the heat shock response pathway [37] and increased nucleotide biosynthesis to overcome metabolic stress [22]. As KDM6B responders were enriched for genetic mutations that further amplified NOTCH1 signaling in T-ALL patients, we hypothesized KDM6B may be required to protect T-ALL cells from strong NOTCH1-induced oncogenic stress.
To examine NOTCH1 signal strength in T-ALL patient cells, a western blot was performed to observe stability of the cleaved NICD protein. NICD levels were markedly higher in responder patient samples compared to KDM6B non-responders (Fig. 7A). However, protein stability of NICD does not appear to be influenced directly by KDM6B as NICD levels were comparable in patient samples following CRISPR/Cas9-mediated deletion of KDM6B (Fig. 7B). Consistent with increased NICD stability, there was increased expression of NOTCH1 target genes in T-ALL [38] such as CCND3, CDK4 and CDK6 in human responder samples versus nonresponder patients, and increased expression of classical NOTCH1 signaling genes such as Hes1, Hey1 and Dtx1 in mouse Kdm6b-KO T-ALL cells (Fig. 7C). To investigate the hypothesis that KDM6B is only required to protect T-ALL cells faced with strong NOTCH1 signaling, we utilized a retroviral model expressing a T-ALL mutant form of the NOTCH1 receptor (NOTCH1 L1601P -ΔPEST [39]) that induces weaker NOTCH1 signaling than the NICD-expressing retrovirus used in earlier studies (Fig. 7D). We performed the same retroviral transduction experiment on Sca-1-enriched WBM from adult control, Kdm6b-HET and Kdm6b-KO mice. Four-weeks posttransplant, AnnexinV staining was performed on peripheral blood T-ALL cells to examine apoptosis. In contrast to NICD-driven T-ALL which induced significant apoptosis in Kdm6b-KO cells (Fig. 2F), there was no difference in apoptosis between the T-ALL genotypes expressing NOTCH1 L1601P -ΔPEST (Fig. 7E). Consistent with the apoptosis data, long-term monitoring of these mice revealed that NOTCH1 L1601P -ΔPEST-driven T-ALL developed at the same incidence amongst all genotypes of donor cells (Fig. 7F). Cumulatively, these data show that Kdm6b protects T-ALL cells from oncogene stressinduced apoptosis where potent NOTCH1 activating mutations induce strong NOTCH1 signaling.

DISCUSSION
Technological advances in genome sequencing technology have allowed unparalleled resolution into the spectrum of mutations present in cancer genomes, and virtually all of the important genetic drivers of T-ALL have now been identified [10,[40][41][42]. Resulting investments into generating genetic tools to model how these mutations contribute to leukemogenesis has generated mechanistic insight into transformation of malignant T-cells. Unfortunately, this has not necessarily translated into new therapeutic approaches. An alternative strategy to identifying new drug targets in T-ALL is to define leukemia-essential genes, that is genes which are never mutated in a particular disease, indicating they may be essential for the survival and propagation of the malignant cells. Therapeutic inhibition of these gene products could thus potentially render the cells incompetent for malignant transformation. This study identifies KDM6B as an essential gene for NOTCH1-mutant T-ALL cells. Mouse hematopoietic progenitor cells deficient for Kdm6b were unable to initiate T-ALL driven by strong NOTCH1 activating mutations, and primary T-ALL cells from human patients targeted for KDM6B loss-offunction by CRISPR/Cas9 gene editing were rapidly outcompeted by non-edited cells. These outcomes suggest KDM6B mutations are never observed in T-ALL patients as this would result in loss of these clones due to competitive disadvantage.
For consideration of therapeutic targeting, we have previously shown that while genetic inhibition of Kdm6b leads to a competitive disadvantage of mouse hematopoietic stem and progenitor cells (HSPCs), this defect was only evident under stress conditions such as serial bone marrow transplantation or inflammation [16]. Mice with complete absence of Kdm6b in the hematopoietic system are viable and show no overt phenotypes, suggesting inhibition of KDM6B (at least in the short term) can be tolerated by HSPCs and should no cause overt hematopoietic toxicity as a drug candidate. The small molecule GSK-J4 has been identified to inhibit KDM6B [32] and been shown to have efficacy against certain T-ALL cells in vitro [12]. However, we show here that the activity of GSK-J4 against T-ALL cells was not dependent on KDM6B (Fig. 5B). GSK-J4 is a catalytic site inhibitor of the JmjC domain conserved between KDM6B and KDM6A. Despite the high degree of sequence similarity between these proteins, recent studies have revealed surprisingly contrasting roles for these proteins in cellular reprogramming and cancer, including T-ALL where KDM6A functions as a gender-specific tumor suppressor [43]. Thus, a more selective inhibitor that specifically targets KDM6B is necessary for translational applications. A more detailed understanding of the specific molecular functions by which KDM6B protects NOTCH1-mutant cells from apoptosis is required to achieve this. Our previous studies showed that the function of Kdm6b in normal hematopoiesis appears largely unrelated to chromatin regulation, as H3K27me3 patterns were not significantly different wild-type and Kdm6b-null HSPCs [16]. In contrast, the protective function of Kdm6b in NOTCH1-mutant T-ALL was dependent on histone demethylase activity as catalytically-dead HSPCs were also unable to propagate T-ALL (Fig. 3B). This justifies a screen for more specific enzymatic inhibitors of KDM6B as a therapy for NOTCH1-mutant T-ALL.
Translation of our findings from mouse models using primary patient samples identified a subgroup of NOTCH1-mutant T-ALL cells that were particularly sensitive to genetic inhibition of KDM6B. Although all primary samples used contained NOTCH1 mutations for genetic consistency, there is considerable heterogeneity in the spectrum of NOTCH1 variants within T-ALL patients [44]. We found the "responder" patients had the highest level of NICD expression, conveying strong NOTCH1 signaling. Moreover, the "responder" group also contained the only patients with multiple NOTCH1 mutations and patients which also carried FBXW7 mutations which further amplify NOTCH1 signaling [36]. We hypothesize that KDM6B is required in this subgroup of patients to protect the leukemia cells from NOTCH1 oncogene stress-induced apoptosis, although the precise mechanism of this protection conveyed by KDM6B remains to be determined in future studies. Cumulatively, this study identifies a novel role for KDM6B in maintenance of T-ALL driven by strong NOTCH1 signaling, which opens the possibility for a new class of therapeutic discovery for this subset of T-ALL patients.

DATA AVAILABILITY
Primary RNA-sequencing data are available through NCBI Gene Expression Omnibus under GEO accession number GSE214576.