KDM6 demethylases integrate DNA repair gene regulation and loss of KDM6A sensitizes human acute myeloid leukemia to PARP and BCL2 inhibition

Acute myeloid leukemia (AML) is a heterogeneous, aggressive malignancy with dismal prognosis and with limited availability of targeted therapies. Epigenetic deregulation contributes to AML pathogenesis. KDM6 proteins are histone-3-lysine-27-demethylases that play context-dependent roles in AML. We inform that KDM6-demethylase function critically regulates DNA-damage-repair-(DDR) gene expression in AML. Mechanistically, KDM6 expression is regulated by genotoxic stress, with deficiency of KDM6A-(UTX) and KDM6B-(JMJD3) impairing DDR transcriptional activation and compromising repair potential. Acquired KDM6A loss-of-function mutations are implicated in chemoresistance, although a significant percentage of relapsed-AML has upregulated KDM6A. Olaparib treatment reduced engraftment of KDM6A-mutant-AML-patient-derived xenografts, highlighting synthetic lethality using Poly-(ADP-ribose)-polymerase-(PARP)-inhibition. Crucially, a higher KDM6A expression is correlated with venetoclax tolerance. Loss of KDM6A increased mitochondrial activity, BCL2 expression, and sensitized AML cells to venetoclax. Additionally, BCL2A1 associates with venetoclax resistance, and KDM6A loss was accompanied with a downregulated BCL2A1. Corroborating these results, dual targeting of PARP and BCL2 was superior to PARP or BCL2 inhibitor monotherapy in inducing AML apoptosis, and primary AML cells carrying KDM6A-domain mutations were even more sensitive to the combination. Together, our study illustrates a mechanistic rationale in support of a novel combination therapy for AML based on subtype-heterogeneity, and establishes KDM6A as a molecular regulator for determining therapeutic efficacy.

Growing evidence suggests the involvement of KDM6A in acute myeloid leukemia (AML) pathogenesis [4,[8][9][10][11][12]. KDM6A escapes X chromosome inactivation, and Utx-null homozygous female mice spontaneously develop aging-associated myeloid leukemia [9,13]. In addition, KDM6A loss-of-function mutation is implicated in conventional chemotherapy relapse in AML, indicating tumor suppressor function [8,10,14]. KDM6A condensation, which involves a core-intrinsically-disordered-region (cIDR), has been reported to confer tumor-suppressive activity independent of Jumonji C-(JmjC)-demethylase function [14]. Recent studies suggested downregulation of KDM6A expression occurs in about 46% of cytogenetically normal-karyotype and AraC-relapsed AML patients [8]. However, 37% of cases exhibited upregulated KDM6A transcripts. Thus KDM6A must function in a highly contextual fashion since there are subsets of AML cases where expression is on opposite ends of a spectrum. Therefore, the cause and pathophysiological relevance of KDM6A upregulation at chemotherapy relapse, observed in more than a third of the patients, is an open question. Additionally, to what extent KDM6A expression and function are connected with AML-targeted therapy is unknown.
By contrast, KDM6B predominantly plays a context-dependent oncogenic function in hematological malignancies [15,16]. KDM6B regulates transcriptional elongation, and KDM6B expression is upregulated in myelodysplastic syndromes-hematopoietic stem/ progenitors [17,18]. While KDM6A acts as a tumor suppressor and is frequently mutated in T-ALL, KDM6B is essential for the initiation and maintenance of T-ALL [19,20]. However, a subgroup of T-ALL expressing TAL1 is uniquely vulnerable to KDM6A inhibition [21]. Together, KDM6A and KDM6B possess cell type-specific functions in leukemia, with KDM6 proteins and their associated signaling emerging as important focal points for developing targeted therapy. Key cellular processes impacted by KDM6 demethylases include Th-cell development, integrated-stress-response activation, and regulation of DNA double-stranded break repair.
Efficient repair of DNA damage caused by genotoxic stress is important for tissue homeostasis. Tumor cells accumulate considerable levels of DNA damage and require robust DNAdamage-repair (DDR) mechanisms for survival. AML cell survival depends upon an intact DNA repair machinery, with an accumulation of DNA double-stranded-breaks (DSBs) leading to apoptosis [22]. DSBs are among the most lethal DNA aberrations, and are repaired through either homologous-recombination (HR) mediated repair or non-homologous-end-joining (NHEJ) [23]. Targeting DNA repair pathways for cancer therapy has gained momentum over the past few years, with poly(adenosine 5′diphosphate-ribose)-polymerase (PARP) inhibition for HR-deficient tumors have shown promise in clinical settings [24][25][26]. Therefore, identifying molecular regulation of DNA repair pathways important for AML cell survival is essential for developing effective combination targeted therapy.
Here we demonstrate that KDM6 demethylases play an important role in DDR gene regulation in AML opening the potential for improved molecular targeted therapies in AML through epigenetic modulation. Together, our study addresses two important clinical questions: first, PARP inhibition would be effective for KDM6A-deficient AML, and secondly, KDM6A inhibition should potentiate PARP or BCL2 blockade in distinct subtypes of AML where KDM6A expression is upregulated or even maintained above threshold level.

RESULTS
KDM6 demethylases associate with DSB repair gene expression in AML Kdm6a deficient homozygous female mice (Utx -/-) spontaneously develop aging-associated AML [9]. To identify genes regulated by KDM6A in AML development, we re-analyzed the available RNAseq results from Utx -/female mice presenting with AML (ERS1090539, ERS1090541, ERS1090542), compared to Utx +/+ control females (ERS539514, ERS539515) [9]. Utx -/and MLL-AF9 negative AML splenocytes were able to propagate leukemia in secondary recipients. Deficiency of Kdm6a led to 4014 genes being downregulated and 4703 genes being upregulated (FDR: 0.01; Log 2 FC: > 1.5) (Supplementary dataset S1). KDM6A JmjCdemethylase function is predominantly associated with transcriptional activation [27]. Gene ontology (GO) enrichment analysis of the downregulated genes (4014) in Utx -/cells revealed enrichment of several GO terms linked with DNA repair, with the most significant being the double-strand break (DSB) repair (Fig. 1A). The DSB repair term included genes of both HR and NHEJ pathways ( Supplementary Fig. S1A). Re-analysis of ChIP-seq results conducted in Utx +/+ hematopoietic cells showed a total of 1825 Kdm6a ChIP-seq occupied genes [9], which were downregulated upon Kdm6a loss. GO analysis of these 1825 genes (Supplementary dataset S2) further revealed a significant enrichment of DNA repair-associated GO terms, suggesting involvement of Kdm6a demethylase in DNA repair ( Supplementary Fig. S1B).
To dissect the role of KDM6 proteins in regulating DDR gene expression in AML, we generated U937 cells using shRNAexpressing lentivirus vectors against KDM6A or KDM6B or both (hereafter referred as KDM6 deficient cells) ( Supplementary  Fig. S1C). U937 cell, originally isolated from a patient with histiocytic lymphoma, has been defined as a promonocytic myeloid leukemia cell line, capable of monocytic differentiation and has frequently been used as a model for myeloid leukemia. Additionally, U937 cells are relatively resistant to standard chemotherapy including KDM6 small molecule inhibitor GSK-J4, and therefore can serve as a relevant model to characterize targeted therapy. KDM6 knockdown led to an increase in global H3K27me3 and a decrease in H3K27ac levels, with the difference being more prominent in KDM6B knockdown and double knockdown cells (Supplementary Fig. S1D). Deficiency of KDM6A and/or KDM6B did not affect the proliferation of U937 cells ( Supplementary Fig. S1E). KDM6A deficient AML cell lines did not show consistent change in myeloid differentiation (Supplementary Fig. S1F). RNA-seq analysis (FDR: 0.05; Log 2 FC: > 2) suggested that there was significant downregulation (P < 0.05) of >80 DDR genes upon knockdown of KDM6A alone or KDM6B alone or both (Fig. 1B). Gene Set Enrichment Analysis uncovered an enrichment of DDR pathway genes that were significantly downregulated in KDM6A deficient AML cells (Fig. 1C). GO term also indicated enrichment of multiple DNA repair genes, including BRCA and RAD families, in KDM6 deficient AML ( Supplementary Fig. S1G). Together, these findings suggest that KDM6 proteins are associated with DDR gene regulation in AML.

DSB repair activation induces expression of KDM6 in AML
To elucidate the function of KDM6A in mediating DSB repair in AML, we first interrogated irradiation-induced alteration of KDM6A in AML cells. H3K27me3 level influences DSB repair efficiency, and decrease in H3K27me3 associates with radiation dosage, with 10 Gy irradiation causing maximum decrease [28]. Interestingly, a single dose of γ-radiation (10 Gy) induced a time-dependent increase in expression of KDM6A (six out of six AML cell lines tested) and KDM6B (four out of six lines tested) independent of pathological or molecular subtypes ( Fig. 1D; Supplementary  Fig. S1H). Increase in KDM6A expression was observed as early as 30 min in KG1a cells, while OCI-AML-2 and KG1a cells showed maximum induction at 4 h after irradiation (Fig. 1D). Low dose irradiation in AML cells did not sufficiently induce expression of KDM6A or KDM6B ( Supplementary Fig. S2A). In agreement with gene expression alteration, KDM6A protein was upregulated on radiation accompanied with a concomitant decrease in H3K27me3 (Fig. 1E, F; Supplementary Fig. S2B, C). Primary AML cells also showed an increase in KDM6A protein expression and decrease in H3K27me3 level in response to radiation ( Supplementary Fig. S2D, E). There was no significant induction of KDM6A or KDM6B in normal CD34 + CD38 − CD45RA − hematopoietic stem cells upon genotoxic stress ( Supplementary Fig. S2F). Collectively, these results indicate that γ-IR mediated DNA repair induces KDM6 demethylase expression in AML.
Deficiency of KDM6 impairs DDR gene expression and DSB repair in AML Efficient DSB repair has been shown to promote survival of AML cells [22]. To identify target genes that sensitize AML cells to genotoxic stress, we leveraged a previously reported genome-wide pooled lentiviral shRNA screening performed utilizing TEX cells in response to one and three rounds of 1 Gy γ-IR ( Fig. 2A) [29,30]. "Leukemia stem cell (LSC)-like" human hematopoietic cell line TEX was generated via TLS-ERG leukemia fusion oncogene expression in cord-blood derived hematopoietic stem and progenitor cells (HSPCs), which maintains functional heterogeneity, cytokine dependency, and a functional p53 pathway [31,32]. Interestingly, KDM6A knockdown, similar to loss of other crucial DNA repair genes, significantly impaired proliferation in two out of three clones, suggesting a radioprotective function ( Fig. 2A, B). Treatment of U937 cells using γ-IR induced HR gene expression ( Fig. 2C; Supplementary Fig. S2G). On contrary, induction of DDR gene and protein expression was significantly impaired in KDM6A or KDM6B deficient AML cell lines, which was accompanied with an altered cell survival and proliferation ( Fig. 2C; Supplementary Fig. S2H-K). Similar results were obtained using a different shRNA expression construct targeting KDM6A. Ectopic expression of full-length KDM6A, but not JmjC mutant, restored DDR gene expression ( Supplementary Fig. S2K). KDM6A knockdown in normal CD34 + HSPCs had a modest effect in radiation-induced DDR gene expression ( Supplementary Fig. S2L). . qRT-PCR values were normalized to GAPDH. Data are representative of at least three independent experiments. Statistics were calculated with Student's t test; error bars represent means ± SD. *P < 0.05 or **P < 0.01 were considered to be statistically significant.
Earlier we demonstrated that treatment of AML cells using a KDM6 small molecule inhibitor GSK-J4 causes a selective increase in H3K27me3 [4]. KDM6A primarily plays tumor suppressor role in demethylase-independent mechanisms [9,14]. However, we and others reported that KDM6 inhibition in AML cells, with intact KDM6 expression, using GSK-J4 attenuates leukemia cell survival and leukemia development [4,12]. Apart from KG1a, all cells displayed IC 50 greater than 2 µM, a dose we used as a sub-lethal concentration for subsequent experiments ( Supplementary  Fig. S3A Fig. S4F). Consistent with these findings, KDM6 knockdown in human AML also caused a reduced expression of MRN ( Supplementary Fig. S4G). Collectively, these results underscore that KDM6 proteins play a critical role in maintaining an elevated expression of DSB recognition genes in AML cells, and KDM6 deficiency or inhibition causes an impaired DSB repair response.

KDM6A regulates chromatin accessibility and transcriptional activation at DDR loci
To understand the mechanism of KDM6A-mediated DDR gene regulation, we conducted qChIP experiments. Treatment with γ-IR resulted in a significant increase in chromatin occupancy of KDM6A at the transcription start sites (TSS) and promoter-proximal elements of BRCA and RAD family genes in AML cells ( Fig. 3A; Supplementary Fig. S4H). In addition, qChIP analysis suggested that deficiency of KDM6A was associated with a concomitant increase in occupancy of H3K27me3 at these loci in both untreated (0 h) and radiation treated (2 h) AML cells, further indicating demethylase-dependent transcriptional regulation ( Fig. 3B; Supplementary Fig. S4I). γ-IR caused a reduction in H3K27me3 at the HR promoter-proximal loci in control cells, however, a similar decrease in locus-specific H3K27me3 occupancy was either absent or negligible in KDM6A deficient cells ( Fig. 3B; Supplementary Fig. S4I). KDM6A has been shown to functionally interact with SWI/SNF ATP-dependent chromatin remodeling complex to regulate chromatin accessibility, and influence gene expression [1,9,33]. We had previously conducted ChIP-seq with the SMARCC1 (BAF155) core subunit of SWI/SNF in primary AML samples (GSE108976) [11,34]. Reanalysis of genes which showed enrichment for both SMARCC1 and KDM6A revealed a substantial overlap between SMARCC1 and KDM6A targets (2785 genes; dataset S3, P < 0.05), with the majority of these co-occupied genes also showing enrichment of H3K27ac, while being devoid of H3K27me3 (1676 genes) ( Fig. 3C; Supplementary dataset S4). These KDM6A targets that include DNA repair genes may represent potential candidates for co-regulation by KDM6A demethylase and SWI/SNF (Fig. 3D). qChIP further demonstrated that concomitant with an increase in KDM6A occupancy there was a significant enrichment of SMARCC1, CBP, and H3K27ac at the TSS and promoter regions of BRCA and RAD genes after 2 h of radiation exposure (Fig. 3E). Radiation-induced increase in chromatin occupancy of SMARCC1, CBP and H3K27ac were also observed at the KDM6A promoter itself (Fig. 3F), suggesting KDM6A and SWI/SNF cooperation in DDR gene regulation.
To interrogate changes in chromatin accessibility on KDM6 loss, we performed bulk ATAC-seq in KDM6A and KDM6B deficient AML cells and compared them with unaltered control cells. In concordance with transcriptional activation function of KDM6, the number of transcription factor (TF) motifs enriched in control cells that lost accessibility in KDM6 deficient cells was much higher than the number of motifs, which gained accessibility in KDM6A/B deficient AML cells (Fig. 3G, H). Motif comparison revealed greater than 90% overlap in KDM6A or KDM6B deficient cells. There was a significant loss in the binding potential of TCF, CEBP, FOXO, and HOXA family, which usually promote HR gene expression (Fig. 3G). Alternatively, there was an increase in binding potential of IRF, PU, and PRDM (Fig. 3H), which have been shown to suppress DDR and induce genomic instability. Collectively, these results indicate that changes in chromatin accessibility correlate with lower abundance of TF binding sites required for optimal DNA gene regulation. Together this may account for the observed repression of DDR gene expression in KDM6 deficient AML, thus compromising DNA repair.

KDM6A loss renders AML cells sensitive to PARP inhibition
We next assessed whether reduced KDM6 levels would sensitize AML cells to inhibition of PARP-1 signaling. Analysis of OHSU AML (n = 672), containing de novo and relapsed AML cases with varying molecular subtypes, showed a significant inverse correlation between KDM6A and PARP-1 expression (Fig. 4A, B; Supplementary Fig. S4J). There was no major change in PARP1 expression in KDM6 deficient U937 cells ( Supplementary Fig. S4K). PARP inhibition using olaparib for 72 h decreased intracellular PAR level and induced apoptosis in AML cells ( Supplementary Fig. S5A, B). In general treatment with olaparib at concentrations below IC 50 caused a cytostatic rather than cytotoxic effect; significant cell apoptosis was observed only at higher concentrations (Supplementary Fig. S5B). Drug dose-response analysis indicated that AML cells co-treated with GSK-J4 were significantly more sensitive to olaparib compared to the controls ( Fig. 4C; Supplementary  Fig. S5C). Except KG1a cells, both TP53 wild type (OCI-AML-2, OCI-AML-5, MOLM-13) and TP53 mutated (U937, NB4) AMLs were susceptible to olaparib in response to KDM6 inhibition (Fig. 4C). Similarly, deficiency of KDM6A sensitized AML cells to PARP inhibition (Fig. 4D). Additionally, KDM6 loss caused a differential sensitivity of select AML subtypes to a conventional chemotherapeutic agent like AraC, although daunorubicin treatment did not appreciably alter AML sensitivity ( Supplementary Fig. S5D-F). Fig. 2 Loss of KDM6 in AML cells impairs DDR gene expression and double-stranded break (DSB) repair. A Schema representing screening assay for radiosensitive genes using pooled targeted lentiviral shRNA library in TEX leukemia cells. B Scatter plots showing distribution of KDM6A along with a few other known DNA repair-associated genes with respect to the overall gene sets analyzed in the shRNA screening in response to 1 round (upper panel) or 3 rounds (lower panel) of radiation-recovery cycles. The values on the Y-axes denote the ratio [IR (treatment)/NT (control)] of individual shRNA corresponding to each gene. Analysis of clone abundance (average of four replicates) of KDM6A targeting shRNA clones after 1 round or 3 rounds of γ-IR (1 Gy) and recovery cycles (right two panels). C qRT-PCR analysis (normalized to 0 h) of HR genes in control and KDM6 deficient U937 cells treated with 10 Gy of γ-IR (n = 2). D qRT-PCR analysis (normalized to 0 h) of HR genes in DMSO (control) and GSK-J4 treated OCI-AML-2 cells treated with 10 Gy γ-IR (n = 2). E qRT-PCR analysis (normalized to 0 h) of HR genes in DMSO and GSK-J4 treated OCI-AML-5 cells irradiated with 10 Gy (n = 2). F Neutral comet assay showing the distance of comet tail measured in control or KDM6A deficient AML cells after 2 h of treatment with 10 Gy of γ-IR (n = 2). G Immunofluorescence analysis (left) and quantitation (right) of γH2A.X foci per nucleus (n = 40-50) in control or KDM6A deficient AML cells at different time points after treatment with 10 Gy of γ-IR (n = 2). qRT-PCR values were normalized to GAPDH. Data are representative of two to three independent experiments. Statistics were calculated with Student's t test; error bars represent means ± SD. *P < 0.05 or ***P < 0.001 were considered to be statistically significant.
Analysis of the Beat AML dataset indicated that cells with lower KDM6A expression may harbor FLT3-ITD mutation ( Supplementary  Fig. S6A). In agreement, we observed that FLT3-ITD expressing KDM6A deficient AML cells were relatively more sensitive to olaparib compared to the controls (Supplementary Fig. S6B). To investigate olaparib sensitivity in vivo, we transplanted control and KDM6A deficient U937 into NOD.Cg-Prkdc scid /J (NOD.SCID) mice ( Supplementary Fig. S6C). KDM6A loss alone did not affect the overall engraftment potential. In support of our prediction, compared to vehicle-treated cells, olaparib administration resulted in a significant decrease in the engraftment of KDM6A deficient, but not control, AML (Fig. 4E). To further confirm, we established AML patient-derived xenograft models carrying KDM6A nonsense mutation implicated in relapse (Fig. 4F). There was a significant reduction of human CD45 + CD33 + cells in the bone marrow in mice treated with olaparib compared to vehicle-treated group (Fig. 4G). Together these results suggest that KDM6A loss increases the sensitivity of AML cells to PARP inhibition.
Deficiency of KDM6A increases the susceptibility of AML to BCL2 blockade BCL2 inhibitor venetoclax has shown promise in the clinical setting, although a majority of the initial responders relapse [35,36]. Beat AML analysis indicated that monocytic (Mono) AML cases associate with venetoclax resistance [37], as well KDM6A hi expressing male AMLs are relatively more tolerant to venetoclax (Fig. 5A, B). We re-analyzed the available RNA-seq dataset from venetoclax-resistant Mono-AML ROS low LSCs, and compared it with venetoclax-sensitive Prim-AML Ros low LSCs [38]. In agreement with earlier findings, Mono-AML showed a relatively lower BCL2, and there was a significant increase in BCL2A1 expression in Mono-AML compared to Prim-AML ( Fig. 5C; Supplementary  Fig. S6D). BCL2A1 is a predictive biomarker of venetoclax resistance in AML and induces resistance to BCL2 inhibitor ABT-737 in CLL [39,40]. Consistent with these findings, the OHSU AML dataset further suggested a positive correlation between KDM6A and BCL2A1 expression, while expression of BCL2A1 and BCL2 was inversely correlated ( Fig. 5D; Supplementary Fig. S6E). KDM6A downregulation induced BCL2, which was accompanied with a concomitant decrease in BCL2A1 gene expression (Fig. 5E Fig. S6I). Gain in fulllength KDM6A, but not TPR or JmjC deletion mutants to some extent, induced BCL2A1 expression in THP1 cells ( Supplementary  Fig. S6J). Although cIDR-deleted KDM6A mutant did not restraint BCL2A1 induction, chimeric IDRs partly restored BCL2A1 expression ( Supplementary Fig. S6J). In addition, qChIP analysis performed in our AML cell lines panel identified KDM6A occupancy at BCL2 or BCL2A1 TSS and promoter regions ( Supplementary Fig. S6K, L). While KDM6A deficiency resulted in increased occupancy of p300 and H3K27ac at BCL2 promoter, there was a significantly reduced p300 binding at BCL2A1 promoter region in KDM6A deficient AML cells ( Fig. 5G; Supplementary Fig. S6M). Mechanistically, qChIP studies further revealed that deficiency of KDM6A was associated with a significant reduction of H3K27ac occupancy, accompanied with an increase in H3K27me3 binding, at BCL2A1 loci ( Supplementary Fig. S6N). Although H3K27ac level was increased at BCL2 promoter in KDM6A deficient AML cell lines, we did not observe a consistent change in H3K27me3 occupancy, suggesting indirect regulation of BCL2 gene expression (Fig. 5G, Supplementary Fig. S6N).
Additionally, corroborating these results GSK-J4 treatment at respective IC 50 doses induced BCL2 expression in select AML subtypes (except MOLM-13 and Kasumi1 cells) ( Fig. 5H; Supplementary Fig. S6O, P). Collectively, these findings indicate that KDM6A differentially regulates BCL2 family gene expression, and KDM6A loss correlates with BCL2 induction. BCL2 induction commonly associates with venetoclax function [41]. GSK-J4 mediated BCL2 induction in AML subtypes further prompted us to interrogate venetoclax sensitivity. Indeed, dose response analysis revealed that TP53 wild type (OCI-AML-2, OCI-AML-5) as well as TP53 mutant (NB4, KG1a) AML cells co-treated with either varying doses of GSK-J4 or constant doses of GSK-J4, set at half of the IC 50 concentrations of respective cell types, were significantly more sensitive to venetoclax compared to the monotherapies alone (Fig. 6A). Although MOLM-13 partially responded to this combination, Kasumi1 cells did not show any effect (Fig. 6A). Similarly, deficiency of KDM6A also sensitized AML cells to BCL2 inhibition (Fig. 6B). Kasumi1 and U937 cells appeared to be intrinsically resistant lines to venetoclax, which correlated with a significant increase in BCL2A1 expression ( Supplementary Fig. S6Q). Previous study suggested that sensitization of neuroblastoma cells to venetoclax upon KDM6 inhibition was accompanied with upregulation of BBC3 (PUMA), IRE1α, and ATF4 [42]. Unlike KG1a and OCI-AML-5 cells, KDM6 inhibition did not induce expression of BBC3, IRE1α, and ATF4 in MOLM-13, Kasumi1, and U937 cell lines (Supplementary Fig. S6R). In addition, olaparib treatment resulted in an increase in mitochondrial membrane potential (MMP) in KDM6A deficient AML cells compared to control cells (Fig. 6C). Inhibition of KDM6 and PARP also resulted in an increase in MMP in AML cells (Fig. 6D). Furthermore, MV4-11 venetoclax resistant (Ven-res) cells showed a decrease in MMP and ROS level compared to venetoclax sensitive (Ven-sen) group ( Supplementary Fig. S7A-D). KDM6 inhibition restored ROS in MV4-11 Ven-res cells, which was further increased in presence of olaparib ( Supplementary Fig. S7C, D). Although this combination had synergistic effects in inducing significant cell death in both primary de novo AML cells as well as in the ven-sen group, we did not observe similar effects in the ven-res AML cells (Supplementary Fig. S7E-G). Intriguingly, we provide evidence that changes in BCL2 expression and mitochondrial activity associated with KDM6A loss, may account for venetoclax tolerance in AML.
Dual inhibition of PARP and BCL2 synergizes in AML Next, we investigated whether inhibition of PARP and BCL2 would have a combination effect in controlling AML cell survival. Cotreatment of olaparib and venetoclax were superior in inhibiting AML cell viability compared to the monotherapies alone ( Supplementary Fig. S7H). Combination of olaparib and venetoclax showed synergistic effects in reducing cell survival in select AML subtypes including OCI-AML-2, OCI-AML-5, KG1a, NB4, and U937 ( Supplementary Fig. S7H). We did not observe drug synergism in MOLM-13, Kasumi1, and HL60 cells (Supplementary Fig. S7H). Similarly, dual inhibition of PARP and BCL2 signaling induced apoptosis in AML ( Supplementary Fig. S8A-C). Interestingly, Motif analysis shows a gain in chromatin accessibility in KDM6A deficient AML. Statistics were calculated with Student's t test; error bars represent means ± SD. *P < 0.05 was considered to be statistically significant.
KDM6A deficient AML cell lines were even more sensitive to the combination therapy ( Supplementary Fig. S8D, E).
To further confirm, we argued that KDM6A silencing may not necessarily mimic pathologically occurring KDM6A mutations. Therefore, we compared drug sensitivities in primary AML samples carrying either wild-type KDM6A or different acquired domain mutants of KDM6A ( Fig. 7A; Supplementary Fig. S8F). Corroborating our findings, olaparib and venetoclax treatment showed a stronger synergistic effect in inhibiting viability of KDM6A-domain mutant primary AML samples compared to KDM6A-wild type cases ( Fig. 7B, C). Although we could not test drug efficacy in cIDRmutant, both TPR and JmjC mutants had dramatic loss of cell viability in response to olaparib and venetoclax (Fig. 7C). NPM1 mut AML848978 only showed a marginal response to the combination (Fig. 7C). Similarly, combination of PARP and BCL2 inhibition led to an increase in apoptosis in KDM6A mutant primary AML cells compared to the wild type control cells ( Fig. 7D; Supplementary  Fig. S8G). Additionally, combination of olaparib and venetoclax significantly impaired clonogenic potential of AML0646 HSPCs during serial replating analysis ( Supplementary Fig. S8H). Normal HSPCs were relatively more tolerant to olaparib (average IC 50 : 13.56 µM compared to 0.42 µM in KDM6A mut primary AML) and venetoclax (average IC 50 :11.46 µM compared to 0.18 µM in KDM6A mut primary AML) ( Supplementary Fig. S8I, J). Overall, KDM6A loss had the most profound effect by compromising DNA damage response and inducing BCL2, thus rendering AML cells sensitive to PARP and BCL2 blockade (Fig. 7E). In sum, we provide evidence and rationale supporting pre/clinical testing of the novel combination targeted therapy for human AML, and posit KDM6A as an important regulator in determining therapeutic efficacy in AML subtypes.

DISCUSSION
In this study we illustrate a mechanistic connection between KDM6 function to impaired DNA repair and BCL2 dependence in AML cell survival. Although venetoclax tolerance is primarily determined by BCL2 expression, and BCLA1 associates with resistance, molecular epi/genetic regulation of these two key proteins is unknown. We provide the first evidence in support of a central regulation integrated by KDM6A demethylase towards BCL2 and BCL2A1 expression important for AML pathogenesis. Our findings that KDM6A was an important regulator for determining efficacy of both PARP and BCL2 blockade; provide support for molecular subtype guided combination targeted therapy for AML. Venetoclax in combination with other small molecule inhibitors has shown better efficacy than venetoclax alone [35,43]. Combination therapy using venetoclax with Complex I inhibitor, MAPK pathway inhibitor or cytarabine has shown promise in preclinical AML models [44][45][46]. In addition, combining KDM6 pharmacological inhibition with venetoclax has been shown to be effective in MYCN-amplified neuroblastoma [42]. Although BCL2 inhibition has been used in combination with hypomethylating agents, their effectiveness in synergy with PARP blockade in AML remains unexplored. In agreement with our findings, an ongoing study indicates that PARP Inhibition using talazoparib can enhance antileukemic activity of venetoclax in preclinical human AML models [Blood (2021)  While HR-mediated DSB repair is indispensable for survival of MLL-AF9 transformed AML, most therapy-related AML has an abnormal DSB response [22,47]. KDM6 inhibition was shown to induce DNA damage in differentiating ES cells [48]. Inhibition of KDM6 catalytic activity impairs HR-mediated DSB repair and augments radiosensitivity in solid tumors [28,49]. Therefore, unlike the demethylase-independent, tumor suppressor function of Utx in AML development, DDR gene regulation is dependent on KDM6A demethylase function [9,11]. In addition, we provide evidence for KDM6A and SWI/SNF cooperation in regulating DDR gene expression. Different subunits of the SWI/SNF complex have been implicated to have non-transcriptional roles in DSB repair. For example, the BRG bromodomain was shown to directly interact with γ-H2A.X and promote chromatin remodeling around DSBs [50]. Also ARID2 facilitates RAD51 recruitment and HR-mediated repair [51]. We did not observe induction of KDM6A or KDM6B transcript in normal HSPCs upon DNA damage, which is in agreement with the distinct DNA damage response of human HSPCs. For example, human HSPCs, in contrast to their lineage-committed progeny, have attenuated irradiation-induced or endonuclease-induced DNA double-strand break repair activity, which is correlated with diminished expression of multiple DSB repair transcripts and more persistent 53BP1and γH2A.X foci [52,53].
Tumors deficient in BRCA genes have suppressed repair system and respond to PARP inhibition [24]. However, AML patients have a low mutational burden for BRCA, and only select subtypes have been shown to have defective DDR that respond well to PARP inhibition. AML1-ETO and PML-RARα driven AML have suppressed expression of key HR-associated genes, and are sensitive to olaparib, whereas MLL-AF9 harboring AML is HR proficient and insensitive to PARP inhibition [25]. Only when used in combination with cytotoxic drugs like cytarabine or daunorubicin does MLL-AF9 AML respond to PARP inhibition [54,55]. Therefore, inducing a "BRCAness" phenotype, through epigenetic modulation expands the range of AML patients, previously unresponsive to treatment, that might respond to PARP inhibitors. In accordance with this, we illustrate that KDM6 attenuation in general sensitizes AML to PARP inhibition.
We also demonstrate altered chromatin accessibility in KDM6deficient AML. The majority of these changes entailed loss in accessibility to TFs, like TCF and FOXM, supporting KDM6A's function as a transcriptional activator. Loss of TCF was reported to attenuate DSB repair and sensitizes colorectal cancer cells to radiotherapy [56]. TCF target NEIL1, a base excision repair gene, is downregulated in KDM6A deficient cells. In addition, FOXM regulates transcription of BRIP1, which cooperates with BRCA1 to promote HR repair [57]. BRIP1 expression is also downregulated in KDM6A-deficient AML. HOXA9 is the mediator of resistance of MLL-AF9 leukemia to olaparib [25]. It promotes the transcription of key HR genes involved in DSB repair, like MCM9, NABP, BLM, ATM, RAD51C, RPA1, BRCA1, and BRCA2. Importantly, among genes downregulated in KDM6A deficient cells are NABP, ATM, BRCA1, and BRCA2. Additionally, our findings indicate a putative association Fig. 4 KDM6A deficiency sensitizes AML to PARP inhibition. A KDM6A and PARP1 mRNA expression z-scores (RNASeq v2 RSEM) heatmap cluster from OHSU AML dataset. B Gene expression correlation analysis of PARP1 with KDM6A in OHSU AML cohort (n = 672). C Percent viability of AML cells treated with varying doses (from 1 nM to 1 mM) of GSK-J4 alone (red) or olaparib alone (green) or in combination (blue) for 72 h. Data represent average of two to three independent experiments with similar results. IC 50 values are tabulated and combination index (Ci) at ED 50 was calculated using CompuSyn v 1.0. Ci < 1 was considered as drug synergism. D IC 50 of olaparib of control or KDM6A deficient AML cells cultured for 48 h. Data represent average of two to three independent experiments with similar results. E Bone marrow engraftment analysis of human CD45 + cells in NOD.SCIDs after treatment with vehicle or olaparib (n = 5 for each treatment group). F Schema representing bone marrow engraftment analysis performed in KDM6A mutant AML patient-derived xenografts (PDX) in response to PARP inhibition. G Flow cytometry contour plots (left) and quantitative analysis (right) showing engraftment of human CD33 + CD45 + cells in NSG mice after being treated with vehicle or olaparib (n = 10 for each treatment group). Statistics were calculated with Student's t test; error bars represent means ± SD. *P < 0.05 was considered to be statistically significant.
of olaparib sensitivity with KDM6A expression and FLT3-ITD mutation. FLT3-ITD AML occurs in about 30% of all AML patients, has a high leukemic burden, poor prognosis, and routinely relapse [58]. FLT3-ITD has been shown to drive increased ROS production, resulting in extensive DNA damage accumulation [59]. Therefore, together with low levels of KDM6A and impaired HR, it represents a suitable target for PARP inhibition. Indeed it has been demonstrated that FLT3-ITD AML is highly sensitive to olaparib [60].
Loss of KDM6A expression and acquired resistance for conventional chemotherapy [8] led to the impetus to further interrogate potential synthetic lethal vulnerabilities in AML In sum, we present a molecular framework highlighting that absence of KDM6A is an important mediator of compromised DDR in different AML subtypes and determining response to PARP inhibition. Collectively, our results are in agreement with previous findings showing KDM6A tumor suppressor properties. Importantly, our findings greatly extend this field both mechanistically but also in terms of clinical relevance as it not only illustrates the efficacy of PARP blockade in KDM6A deficient AML, but also highlights proof of concept for epigenetic modulation guided combination targeted therapy (PARP and BCL2 blockade) in a different subtype of AML where KDM6A expression is upregulated or intact. Although bi-allelic Utx deficiency causes evolution to myeloid neoplasms, perhaps minimal KDM6 activity is important for survival of human AML cells similar to what observed in TET2 deficient AML [61]. Transcriptional adaptation in response to genetic, epigenetic or metabolic perturbations remains a cardinal phenomenon of AML evolution. Adaptive chromatin remodeling mediated by KDM6 proteins was found to be important for the persistence and drug tolerance of glioblastoma stem cells [62]. Future studies should investigate to what extent qRT-PCR and qChIP values were normalized to GAPDH and IgG, respectively. Data represent two to three independent experiments. Statistics were calculated with Student's t test; error bars represent means ± SD if not specified otherwise. *P < 0.05 or ***P < 0.001 were considered to be statistically significant. KDM6 proteins cooperate with clonal hematopoiesis-associated mutational burden and impinge on chromatin topology and epigenomic landscape in AML pathophysiology. KDM6 demethylases have been implicated in solid tumors, and both PARP and BCL2 inhibitors are already being tested in cancer patients, suggesting a broader scope of application. To conclude, KDM6A emerges to be a common regulator for susceptibility of AML to both PARP and BCL2 inhibition, expanding the possibility to characterize effective combination targeted therapy for AML subtypes in pre/ clinical settings.