KDM6A promotes imatinib resistance through YY1-mediated transcriptional upregulation of TRKA independently of its demethylase activity in chronic myelogenous leukemia

Rationale: Despite landmark therapy of chronic myelogenous leukemia (CML) with tyrosine kinase inhibitors (TKIs), drug resistance remains problematic. Cancer pathogenesis involves epigenetic dysregulation and in particular, histone lysine demethylases (KDMs) have been implicated in TKI resistance. We sought to identify KDMs with altered expression in CML and define their contribution to imatinib resistance. Methods: Bioinformatics screening compared KDM expression in CML versus normal bone marrow with shRNA knockdown and flow cytometry used to measure effects on imatinib-induced apoptosis in K562 cells. Transcriptomic analyses were performed against KDM6A CRISPR knockout/shRNA knockdown K562 cells along with gene rescue experiments using wildtype and mutant demethylase-dead KDM6A constructs. Co-immunoprecipitation, luciferase reporter and ChIP were employed to elucidate mechanisms of KDM6A-dependent resistance. Results: Amongst five KDMs upregulated in CML, only KDM6A depletion sensitized CML cells to imatinib-induced apoptosis. Re-introduction of demethylase-dead KDM6A as well as wild-type KDM6A restored imatinib resistance. RNA-seq identified NTRK1 gene downregulation after depletion of KDM6A. Moreover, NTRK1 expression positively correlated with KDM6A in a subset of clinical CML samples and KDM6A knockdown in fresh CML isolates decreased NTRK1 encoded protein (TRKA) expression. Mechanistically, KDM6A was recruited to the NTRK1 promoter by the transcription factor YY1 with subsequent TRKA upregulation activating down-stream survival pathways to invoke imatinib resistance. Conclusion: Contrary to its reported role as a tumor suppressor and independent of its demethylase function, KDM6A promotes imatinib-resistance in CML cells. The identification of the KDM6A/YY1/TRKA axis as a novel imatinib-resistance mechanism represents an unexplored avenue to overcome TKI resistance in CML.


Introduction
Chronic myelogenous leukemia (CML) is characterized by the Philadelphia (Ph) chromosome that results in the expression of the constitutively active tyrosine kinase BCR-ABL [1]. Targeting BCR-ABL using tyrosine kinase inhibitors (TKIs) has become the standard of care for CML patients [2]. Treatment with imatinib, the first of this class of drugs, can achieve a complete hematologic and cytogenetic response, but a multitude of patients eventually develop resistance or intolerance [3]. A major mechanism responsible for the resistance of CML to imatinib is point mutations in the BCR-ABL kinase domain [4]. Although second-generation TKIs including dasatinib and nilotinib can overcome the resistance caused by many imatinib-resistant mutants, they remain ineffective against the T315I "gatekeeper" mutation [5]. The third-generation TKI ponatinib can potently overcome the resistance caused by T315I and other BCR-ABL mutations, but its application is frequently complicated by unacceptable cardiovascular toxicity associated with its broad-spectrum inhibition profile [6]. It seems therefore that an alternative strategy to overcome resistance of CML to TKIs is to target downstream molecular modules essential for CML cell survival.
There is increasing evidence that indicates that epigenetic dysregulation is involved in the pathogenesis of cancer [7,8]. This has led to the development of drugs targeting DNA methyltransferases and histone methyltransferases [9]. Another class of epigenetic modifiers that are emerging as molecular targets for cancer treatment are histone lysine demethylases (KDMs), which constitute two broad family divisions: the lysine-specific demethylases represented by KDM1A and KDM1B that act on mono-and dimethylated lysines at lysine 4 or lysine 9 of H3 and the Jumanji (JmjC) domain-containing KDMs that contains five subfamilies (KDM2-7) that catalyse the demethylation of mono-, di-and trimethylated lysines in both histones and non-histone substrates [10]. Of the latter, KDM6A (also known as UTX) has been shown to be important for embryogenesis as female mice homozygous for catalytically inactive KDM6A have severe developmental defects [11]. Mechanistically, KDM6A functions as a component of the MLL3/4-COMPASS (complex of proteins associated with Set1)-like complex to co-activate gene transcription likely through removing repressive lysine 27 histone 3 methylation marks [12,13]. Nevertheless, KDM6A may also act independently of its demethylase activity [14]. KDM6A has also been broadly implicated as a tumor suppressor gene where its mutational loss commonly occurs in cancer cell lines of diverse tissue origins [15]. KDM6A mutations have been confirmed in corresponding patient samples [16,17], for example, over 30% of bladder cancers contain KDM6A mutations and these mutations also occur to a lesser extent in hematological malignancies including acute lymphoblastic leukemia (subtypes of T-cell and B-cell ALL) along with chronic myelomonocytic leukemia (CMML) [18,19]. Conversely, some KDMs including KDM6A have been shown to be upregulated in human primary acute myelogenous leukemia (AML) cells and inhibiting histone demethylase activity in these cells reduces their survival [20]. In this report, we have explored the potential roles of KDMs in resistance of CML to imatinib. We show here that KDM6A is commonly upregulated in CML cells and its expression is important for CML cell survival upon treatment with imatinib. Notably, KDM6A functions independently of its demethylase activity to promote YY1-mediated transcriptional upregulation of TRKA. Moreover, we demonstrate that KDM6A-mediated activation of TRKA is required for protection of CML cells against imatinib afforded by the neurotrophin nerve growth factor (NGF). These results suggest that targeting KDM6A represents a useful strategy for overcoming resistance of CML to TKIs.

KDM6A is upregulated and confers resistance to imatinib in CML cells
Through interrogating datasets acquired from Oncomine, we derived a list of KDMs that were upregulated in CML in comparison with either normal bone marrow or peripheral blood mononuclear cells (PBMCs), including KDM1B, KDM4B, KDM5B, KDM6A and KDM6B ( Figure  S1A-B). Strikingly, although shRNA knockdown of the individual KDMs did not impinge on the viability of K562 CML cells that harbor wild-type BCR-ABL ( Figure 1A), knockdown of KDM6A but not the other KDMs rendered K562 cells more sensitive to apoptosis induced by imatinib ( Figure 1B and Figure  S1C). This effect of KDM6A knockdown was confirmed using two independent shRNAs ( Figure  1C-E). Moreover, K562 cells with KDM6A knocked out using the lentiCRISPR v2/Cas9 system ( Figure  S1D) appeared markedly more sensitive to imatinib than control counterparts ( Figure 1F-G). This significantly reduced the IC50 for imatinib from 1.15 to 0.24 µM ( Figure 1H) indicating that KDM6A plays a role in protecting CML cells against imatinib. Depletion of KDM6A by either knockout or shRNAs caused no changes in the expression of BCR-ABL or its phosphorylation at Tyr177 ( Figure 1F), indicating that protection of CML cells against imatinib by KDM6A is not due to alterations in the expression or activity of BCR-ABL. The role of KDM6A in protection of CML cells against imatinib is not cell line dependent as knockdown of KDM6A also sensitized MEG-01 CML cells that carry wild-type BCR-ABL to imatinib-induced apoptosis ( Figure 1E and Figure  S2A). Notably in the absence of imatinib treatment, depletion of KDM6A in both K562 and MEG-01 cells had no appreciable effect on cell growth ( Figure  S2B-C).
Interrogation of bone marrow samples from newly diagnosed CML patients against control patients with idiopathic thrombocytopenic purpura showed an increase in KDM6A mRNA expression associated with CML ( Figure 1I and Table S1). Moreover, a group of bone marrow samples from patients who were refractory to imatinib treatment displayed even higher levels of KDM6A mRNA than those who responded to the treatment ( Figure 1J and Table S1). The general increase in KDM6A in CML along with the high levels observed in imatinib refractory CML cases proposes that the KDM6A-dependent mechanism of imatinib resistance is also active in patients.

Figure 1. KDM6A is upregulated in CML and depletion of KDM6A sensitizes CML cells to imatinib-induced apoptosis. (A)
Knockdown efficiency of shRNA-bearing lentiviruses against individual KDMs in K562 cells compared to a scrambled shRNA control. Cells were transduced for 72 h and relative expression determined by qPCR. shKDM6A: shKDM6A#3. Mean ± s.d. are given for three independent experiments. Unpaired, two-tailed Student's t-test; ***p < 0.001. (B) The rates of apoptosis in shCtrl versus individual shKDM K562 cells after 24 h treatment with DMSO vehicle control or 0.5 μM imatinib. The percentage of early apoptotic cells was determined by FACS using annexin V/PI double staining. shKDM6A: shKDM6A#3. Mean ± s.d. are given for three independent experiments. Unpaired, two-tailed Student's t-test; ***p < 0.001. (C, D) Knockdown efficiency of two independent shRNAs against KDM6A in K562 and MEG-01 cells stably transduced with lentivirus was determined by qPCR (C) and Western blotting (D). Mean ± s.d. are given for three independent experiments. One-way ANOVA; ***p < 0.001. (E) The rates of apoptosis in shCtrl versus shKDM6A K562 cells and MEG-01 cells were determined by FACS after treatment with 0.5 μM imatinib for 24 h. Mean ± s.d. are given for three independent experiments. One-way ANOVA; ***p < 0.001.

KDM6A protects CML cells against imatinib independently of its demethylase activity
Since the biological roles of KDMs are predominantly but not exclusively dependent on their demethylase activities, we investigated whether the demethylase activity of KDM6A is required for its protection of CML cells against imatinib. We employed GSK-J4, an inhibitor affecting the demethylase activity of KDM6A and related demethylases [21,22]. Dose titration of GSK-J4 against K562 cells identified 100 nM could optimally increase global H3K27me3 levels without adversely affecting cell growth (Figure 2A and Figure S2D). However, GSK-J4 treatment did not affect the sensitivity of K562 cells to imatinib ( Figure 2B), implying that the demethylase activity of KDM6A is dispensable for protection of CML cells against imatinib. To verify this notion, we introduced an enzymatically-dead KDM6A mutant (KDM6A-ED) carrying H1146A and E1148A mutations into K562 cells with endogenous KDM6A knocked out ( Figure 2C) [14]. While re-expression of wild-type KDM6A restored resistance to imatinib, the expression of the KDM6A-ED similarly rescued the imatinib-sensitive phenotype ( Figure 2D-F). Thus, KDM6A protects CML cells from imatinib independently of its demethylase activity.

KDM6A protects CML cells against imatinib through TRKA
To identify genes and signaling pathways associated with the KDM6A-dependent protection of CML cells against imatinib, we carried out comparative transcriptomic analyses of K562 cells and its imatinib-resistant subline K562/G01 both with and without KDM6A depletion. These analyses identified 144 differentially regulated genes (DEGs) in K562 cells before and after KDM6A depletion with 2601 DEGs in K562/G01 cells ( Figure S3A-B, respectively). Application of the 33 genes in common to hierarchical clustering analysis of gene expression profiles showed ten genes were specifically enriched in KEGG pathways with KDM6A associated with 'transcriptional misregulation in cancer' (Figure 3A-B). Strikingly, downregulation of NTRK1 encoding tropomyosin receptor kinase A (TRKA), a high affinity receptor for nerve growth factor (NGF), was most prominently associated with three KEGG pathways, namely 'transcriptional misregulation in cancer', 'Pathways in cancer' and the 'MAPK signaling pathway' ( Figure 3C). Alternative analysis of the DEGs using GSEA obtained a top ranked list of six enriched pathways including two annotated pathways that include NTRK1, namely 'Neuroactive_Ligand_Receptor_Interaction' and "Thyroid Cancer' ( Figure S3C). The association of TRKA with these pathways proposed its potential involvement in resistance to imatinib. We confirmed that NTRK1 mRNA and protein (TRKA) were downregulated after KDM6A knockout/knockdown in K562 cells using qPCR ( Figure 3D) and Western blotting ( Figure 3E). Similarly, knockdown of KDM6A resulted in downregulation of TRKA at both the mRNA and protein levels in MEG-01 cells ( Figure S3D-E). Consistently, re-expression of wildtype KDM6A in K562 cells with KDM6A knocked out rescues TRKA expression and overexpression of KDM6A caused, albeit moderately, upregulation of TRKA in K562 cells ( Figure 3F). Together, these results indicate that KDM6A drives the expression of TRKA in CML cells. Substantiating this, the expression levels of TRKA were upregulated and positively correlated with the levels of KDM6A in clinical CML samples ( Figure  3G-J). As anticipated, treating K562 cells with GSK-J4 to inhibit the demethylase activity of KDM6A did not impinge on TRKA expression ( Figure 3K). Moreover, ectopic expression of the KDM6A-ED mutant, similar to re-expression of wild-type KDM6A, restored TRKA expression ( Figure 4F).
In agreement with previous reports that TRKA signaling protects CML cells from imatinib [23]. knockdown of TRKA by shRNAs enhanced sensitivity of K562 cells to imatinib-induced apoptosis ( Figure 4A-B and Figure S4A). To test whether TRKA is required for protection of CML cells by KDM6A, we introduced exogenous TRKA into K562 cells with KDM6A knocked out ( Figure 4C). Indeed, ectopic expression of TRKA partially rescued the imatinib-senstitive phenotype caused by knockout of KDM6A ( Figure 4D and Figure S4B). Therefore, TRKA is the functional effector of KDM6A in protection of CML cells against imatinib.

KDM6A functions as a transcriptional activator of NTRK1
Having established that KDM6A promotes the expression of TRKA independently of its demethylase activity, we tested whether other functional domains of KDM6A are necessary for its effect on TRKA expression. Introduction of the KDM6A truncation mutant with the tetratricopeptide repeat (TPR), Jumanji (JMJC) domains or MF (middle fragment) into K562 cells with KDM6A knocked out ( Figure 4E and Figure S5A-B) failed to restore the expression of TRKA ( Figure 4F), suggesting that intact KDM6A is necessary for upregulation of TRKA.
To examine how KDM6A promotes TRKA expression, the ~5 kb upstream region of the NTRK1 proximal promoter was divided into two over-lapping segments designated p1 and p2 and these were cloned into luciferase reporter plasmids ( Figure 4G). Co-introduction of p1 or p2 construct with KDM6A demonstrated that KDM6A strongly enhanced the transcriptional activity of the p1 reporter which encompasses the -5080/-2838 region ( Figure 4H), whereas it had a minimal effect on the p2 reporter that spanned -2837 bp upstream of the transcription start site (TSS). Instructively, luciferase reporter results obtained with the KDM6A-ED mutant phenocopied those of wildtype KDM6A ( Figure 4H). Further subdividing p1 into two further fragments demonstrated that both constructs (p3 and p4) were responsive to NTRK1 expression, albeit with approximately half the magnitude of response compared to p1 ( Figure 4H). These data suggest that a KDM6A functional site exists within the -5080/-2838 region of NTRK1. In support, chromatin immunoprecipitation (ChIP) assays conducted against target sequences across the NTRK1 promoter showed that KDM6A was most strongly enriched with the C-primers which amplifies sequences within the -4038 to -2837 region common to the p1, p3 and p4 fragments ( Figure 4I). To further confirm the ability of the demethylase dead KDM6A mutant to bind to the NTRK1 promoter and influence transcription, we repeated the ChIP-qPCR assays in K562-KO cells reconstituted with either wildtype or KDM6A-ED. Ectopically expressed KDM6A and KDM6A-ED both significantly bound to -4038 to -2837 region but not a control region ( Figure 4J and Figure S6A-B). Moreover, there were no significant changes in histone tri-methylation observed in ChIP assays against H3K27me3 ( Figure 4K and Figure S6C-D) and nor were there global changes in H3K27me3 after KDM6A knockout ( Figure S6G). Similarly, the global levels of of histone acetylation were also unchanged ( Figure S6G). Interestingly the H3K27ac ChIP-qPCR levels associated with the NTRK1 promoter were reduced following KDM6A knockout but could be restored by re-expression of either KDM6A wildtype or ED mutant proteins ( Figure 4L and Figure S6E-F). Based on a previous report showing the histone acetylase CBP (p300) associated with KDM6A to affect changes in H3K27ac levels [24], we further checked if CBP associated with KDM6A in K562 cells. However, no associated CBP was observed in KDM6A immunoprecipitates although as expected, enrichment of RNA Pol II was observed ( Figure S6H). Collectively, these results establish that KDM6A acts as a transcriptional activator of NTRK1 independently of its demethylase activity.   (A-D, F, H-L) represent three independent experiments. Mean ± s.d. are given for three independent experiments. Unpaired, two-tailed Student's t-test; ***p < 0.001; n.s., not significant.

KDM6A is necessary for YY1-mediated transcription of NTRK1
To clarify the mechanism through which KDM6A promotes transcriptional activation of NTRK1, bioinformatics analyses conducted on the NTRK1 promoter determined that the -5080/-2838 fragment was enriched in binding sites for the transcription factors YY1 and C-FOS. Instructively, knockdown of YY1 but not c-FOS led to downregulation of TRKA expression ( Figure 5A), suggesting that YY1 plays an important role in transcriptional activation of NTRK1. ChIP assays revealed that YY1 indeed bound to the -4038/-2838 fragment ( Figure 5B and Figure S6I-J). We next examined the relationship between YY1, KDM6A and the transcriptional regulation of NTRK1. Knockdown of YY1 resulted in a dramatic decrease in KDM6A binding to the NTRK1 promoter ( Figure 5C and Figure S6K). Moreover, re-expression of wild-type KDM6A could not restore the expression of TRKA in K562 cells with deletion of KDM6A and YY1 knockdown ( Figure  5D), indicating that KDM6A-dependent transcriptional upregulation of TRKA requires YY1. Intriguingly, co-immunoprecipitation assays demonstrated binding between KDM6A and YY1 ( Figure 5E-F), suggesting they are components of a complex necessary for transcriptional regulation of NTRK1.
We also tested the importance of YY1 in KDM6A-mediated protection of CML cells against imatinib. Knockdown of YY1 sensitized CML cells to imatinib-induced apoptosis akin to the effect of KDM6A knockdown ( Figure 5G). Thus, YY1 is necessary for protection of CML cells against imatinib by KDM6A.

KDM6A-mediated activation of NTRK1 is required for NGF protection of CML cells against imatinib
We examined the role of KDM6A-mediated activation of NTRK1 in resistance of CML cells to imatinib caused by NGF [23]. Knockout of KDM6A diminished the protection of K562 cells against imatinib afforded by treatment with NGF ( Figure 6A). However, re-expression of wild-type KDM6A or introduction of the KDM6A-ED mutant rescued the protective effect ( Figure 6A), recapitulating the effect of ectopic expression of TRKA in KDM6A KO cells ( Figure 6B and Figure S4B). These results indicate that KDM6A-driven expression of TRKA is needed for resistance of CML cells to imatinib triggered by NGF. Mechanistically, KDM6A knockout abolished NGF-induced activation (phosphorylation) of Akt and ERK1/2 ( Figure 6C), which are known to be involved in NGF-mediated protection of CML cells against imatinib [23], whereas re-expression of wild-type KDM6A or introduction of the KDM6A-ED mutant re-enabled NGF to trigger activation of Akt and ERK1/2 in K562 cells with KDM6A knocked out ( Figure 6D). Moreover, the effects of KDM6A on NGF-triggered activation of Akt and ERK1/2 were also observed in MEG-01 cells with KDM6A knocked down by shRNA ( Figure 6E). Similar to these findings in K562 and MEG-01 cells, knockdown of KDM6A in KU812 CML cells as well as the imatinib-resistant K562/G01 subline was able to abrogate the protective effects of NGF after treatment with imatinib ( Figure  S7A-C and Figure S7D-F, respectively). Lastly, to confirm that the KDM6A-TRKA mechanism revealed in the cell line experiments was relevant to CML in vivo, we undertook knockdown of KDM6A in fresh bone marrow isolates from three newly diagnosed CML patients (Table S1). Importantly, depletion of KDM6A decreased TRKA expression in all three samples ( Figure S8A), providing proof of concept that the regulatory axis exists in bona fide CML cells. NGF did not rescue the cells from increased apoptotic cell death when KDM6A was knocked down ( Figure S8B-C), indicating KDM6A intrinsically confers resistance to imatinib through TRKA in CML patients.

Discussion
Despite the recent advances in the development of novel targeted therapies, drug resistance along with intolerance remains a major obstacle for curative treatment of CML using TKIs. The above results demonstrate that the increased expression of KDM6A plays a role in resistance of CML cells to the TKI imatinib. Noticeably, this effect of KDM6A is not due to its demethylase activity. Instead, KDM6A acts as a partner of the transcription factor YY1 to promote transcriptional activation of TRKA in CML cells ( Figure 6F). Regardless, these results point to the potential of targeting KDM6A in overcoming resistance of CML to TKIs, in particular, when the resistance is conferred by NGF signaling.
Since inactivating mutations in the KDM6A gene are found in a number of human cancers and its inactivation invokes oncogenic phenotypes in various experimental systems, it is thought that KDM6A primarily functions as a tumor suppressor [15,18,25,26]. Moreover, KDM6A is highly expressed in hematopoietic stem and progenitor lineages [27], suggestive of its importance in the hematopoietic compartment. Indeed, mice deficient in Kdm6a exhibit disturbances in normal hematopoiesis and develop acute myelogenous leukemia, albeit with a long latency [28]. Furthermore, loss of KDM6A has recently been reported to confer drug resistance in human AML cells [29]. On the other hand, KDM6A also displays an oncogenic role in varying settings [30,31]. For example, it not only acts as tumor suppressor in certain T-ALL cases [32], but KDM6A overexpression in TAL1-expressing T-ALL increasesd their growth [33]. Similarly, KDM6B which is the most closely homologous KDM to KDM6A [34] is known to be essential for the initiation and maintenance of T-ALL by modulating H3K27 methylation [32]. Furthermore, high KDM6A expression was associated with worse survival of breast cancer patients and overexpression of KDM6A promoted breast cancer cell growth in vitro [35,36]. It seems therefore that KDM6A impinges on cancer pathogenesis in a highly tissue type-and context-dependent fashion.
Here we found that five KDMs were upregulated in CML cells derived from patients compared with normal bone marrow cells including KDM6A and KDM6B. However, only knockdown of KDM6A rendered cultured CML cells more sensitive to killing by imatinib. Moreover, this mechanism was shown to be relevant to imatinib-selected CML cells expressing elevated levels of BCR-ABL, collectively indicating that KDM6A fulfils an intrinsic survival role for CML cells upon inhibition of BCR-ABL signaling. Nevertheless, KDM6A expression was comparatively upregulated in imatinib-resistant versus imatinib-sensitive CML cells from patients including some patients displaying supraphysiological levels. How this occurs remains to be determined but to our knowledge there are only 0.15% KDM6A gene amplifications reported for the most common alterations in KDM6A [37].
An important finding of this study was that KDM6A protected CML cells against imatinib independently of its demethylase activity. This was firstly suggested by the inability of GSK-J4, which inhibits the demethylase activity of KDM6A, to reduce the sensitivity of CML cells to imatinib, and was further substantiated by the observation that an enzyme-dead KDM6A mutant, similar to wild-type KDM6A, recovered the resistance to imatinib in CML cells with endogenous KDM6A knocked out. Although the biological functions of KDM6A are primarily attributable to its nature as a demethylase [38], demethylase activity-independent roles of KDM6A are emerging [14,39,40]. For example, demethylase-independent mechanisms were responsible for KDM6A-mediated regulation of gene expression during mouse embryonic development [11], whereas KDM6A suppressed squamous-like pancreatic cancer growth in male mice similarly independently of its demethylase activity [41]. Our results have expanded the list of demethylase-independent functions of KDM6A by showing that its demethylase activity is dispensable for protection of CML cells against imatinib.
How does KDM6A execute its protective role against imatinib in CML cells? Our results revealed that this was achieved by transcriptional activation of the NGF receptor TRKA. Activation of pro-survival signaling by the NGF/TRKA axis has previously been associated with treatment resistance in hematological malignancies [42][43][44] lending support to the idea of employing TRKA inhibitors.
Notably, pan-pharmacological inhibition of TRKs enhanced sensitivity of diffuse large B-cell lymphoma (DLBCL) to the monoclonal antibody against CD20 rituximab [45], whereas treatment with NGF protected TRKA-transformed myeloid 32D cells against irradiation [46]. Of particular interest, NGF/TRKA signaling has been shown to be constitutively active and play an important role against imatinib through activation of the Akt pathway in BCR-ABL positive CML cells [23]. It would be interesting to further identify the levels of TRKA and downstream signaling pathways of NGF before and after imatinib treatment in CML-sensitive and CML-resistant patients. Recently it has been suggested that TRKA inhibitors could be useful in targeting CML cells bearing oncogenic NTRK-fusion proteins [44]. TRKA inhibitors are not presently indicated for CML treatment nor other hematological malignancies but our findings implore further consideration of this concept. Indeed the correlation between KDM6A and TRKA mRNA expression in patient-derived samples and demonstration of their functional dependency in ex vivo CML cells further corroborates the importance of KDM6A in regulation of TRKA expression in CML cells.
Additionally, we found that YY1 is responsible for transcriptional upregulation of TRKA by KDM6A. Notably, YY1-mediated histone modifications (acetylation and deacetylation) are involved in its roles in repressing and activating promoters [47,48]. We showed that YY1 interacts with KDM6A and appears to be responsible for recruiting KDM6A to the NTRK1 enhancer site. Furthermore, since the expression of KDM6A was required to maintain histone acetylation at the NTRK1 enhancer site, it can be speculated that YY1 maintains this chromatin region in a permissive epigenetic state as a result of its interaction with KDM6A. Irrespectively, our results indicate that YY1-mediated activation of TRKA plays an important role in promoting resistance of CML against TKIs through the NGF/TRKA axis. These findings, together with previous results showing that YY1 is upregulated in CML [49], points to an oncogenic role of YY1 in CML.
In conclusion, we have provided evidence that KDM6A-regulated, YY1-mediated transcriptional upregulation of TRKA is an important resistance mechanism against imatinib in CML. The practical relevance of our results was corroborated by the findings that KDM6A together with TRKA were upregulated in patient-derived CML cells, and in particular, KDM6A was further increased in CML samples from imatinib-resistant individuals. Moreover, our results demonstrated that NGF-induced resistance to imatinib was abandoned in CML cells deficient of KDM6A. Further preclinical evaluation of targeting KDM6A as a strategy to overcome resistance of CML to TKIs is therefore warranted. Similarly, further details of how KDM6A is upregulated in CML cells need to be determined.

CML patients
Archival bone marrow samples from diagnostic, imatinib-sensitive or resistant CML patients along with ITP (idiopathic thrombocytopenic purpura) control samples were obtained from The Affiliated Huaian first People's Hospital of Nanjing Medical University (Huai'an, China). The study was approved by the Institutional Ethics Committee of the hospital. All samples were obtained under written (signed) informed consent. The clinical and demographic details of all patients including those who all received imatinib therapy and were divided into subgroups according to their clinical response (sensitive versus resistant) are shown in Table S1.

Lentiviral transduction and clone derivation
CRISPR-CAS9 lentiviral particles were produced in 293FT cells after transfecting cells with a 2:1:2 mixture of psPAX2, pCMV-VSV-G and pLentiCrispr V2 plasmids using the Lipofectamine 2000 Reagent (Thermo) according to the manufacturers instructions. Alternatively for knockdown or overexpression studies, the pLVshRNA-puro or pLenti-EFs-FLAG-BSD-based vectors were substituted as required. Supernatants containing the viral particles were harvested at 48 h and centrifuged at 500 g for 5 min before filtering through a 0.45 μm PES filter. To increase viral titres the supernatants were precipitated by adding 4 ℃ PEG Virus Precipitation Solution (ab102538, abcam, MA, USA) at a 1:4 v/v ratio, refrigerating overnight before centrifuging the PEG mixture at 1500 × g for 30 min at 4 ℃. The precipitated lentiviral particles were resuspended in PBS and added to the culture medium for 36 h before selecting with 1 μg/mL puromycin (pLentiCrispr V2) or 10 μg/mL blasticidin (pLenti-EFs-FLAG-BSD). Selected cells were single cell cloned in 96 well plates and genomic DNA was amplified using the screening primers listed in Table S3. PCR products were TA-cloned and at least six fragments subjected to Sanger sequencing for each clone. Western blot analysis was then used to verify knockout or overexpression. For the fresh bone marrow samples, isolated mononuclear cells were cultured for 24 h before the addition of shRNA lentivirus for a further 24 h and selected in 1 μg/mL puromycin for a further 48 h.

Real-Time quantitative Polymerase Chain Reaction (qPCR)
Total RNAs were isolated using RNeasy Kits (Qiagen, 74104, Hilden, Germany) according to the manufacturer's instructions. cDNA was synthesized using the Primescript RT-reagent kit (Takara, RR047A, Shiga, Japan) containing gDNA eraser (DNAse) according to the manufacturer's instructions. The qPCR reactions were performed using the LightCycler 480 (Roche, Basel, Switzerland) with relative expression determined using the ΔΔCt method with normalization against β-actin. The primers which span exon-exon junctions to prevent gDNA signal are provided in Table S3.

Histone extraction
Histones were acid extracted as previously described [53]. Briefly, 2×10 6 cells were collected and resuspended in 200 μL NETN lysis buffer (20 mM Tris-HCl pH 8.0, 500 mM NaCl, 0.5% NP-40 (v/v), 1 mM EDTA, 2 mM phenylmethylsulfonyl fluoride (PMSF)) at 4 °C. After centrifugation, the pellet was resuspended in 0.2 N HCl at a cell density of 2×10 7 cells per mL. Histones were extracted by rotating at 4 °C overnight and then the protein concentration was determined using the BCA kit.

Immunofluorescence by confocal microscopy
K562 cells were centrifuged on glass coverslips, fixed with 4% paraformaldehyde in PBS for 20 min, and subsequently permeabilized with 0.1% Triton X-100 for 10 min. After blocking with 1% BSA for 1 h at room temperature , cells were incubated with primary antibody overnight at 4 °C, followed by secondary antibody conjugated with CoraLite488 (SA00013-1, Pro-teintech, IL, USA) for 1 h at room temperature. After three washes with PBS, cells were mounted with Aqueous Mounting Medium containing 4',6-diamidino-2-phenylindole (DAPI, Beyotime, C1005, Shanghai, China) and visualized with a confocal microscope.

Apoptosis assay
Analyses were performed using Annexin V-FITC/PI apoptosis detection kit (KeyGen, KGA107, Nanjing, China) according to the manufacturer's protocol. Briefly, CML cells were seeded into 12-well plates with 2.5 × 10 5 each well and treated with 0.5 µM imatinib and/or 100 ng/mL NGF for 24 h. Cells were then harvested and washed with PBS for three times before addition of 500 μL binding buffer, 5 μL Annexin V-FITC and 5 μL PI solution to the cell pellet for 15 min at room temperature in the dark before flow cytometric analysis to detect early apoptotic cells (Annexin-V positive and PI-negative) (Beckman Coulter, USA).

RNA-Seq analysis
Transcriptome analyses were performed comparing the control K562 or K562/G01 cells against the KDM6A deletion group as indicated. RNA-Seq was performed using an Illumina Hiseq platform (Novogene, Beijing, China). Differential expression analysis of two groups was performed using the ballgrown R package (1.16.1). Ballgown provides statistical routines for determining differential expression in digital gene expression data using a model based on the negative binomial distribution [54]. The resulting p-values were adjusted using the Benjamini and Hochberg's approaches to control false discovery rate. Genes with an adjusted p-value <0.05 were considered to be differentially expressed. The clusterProfiler R package was used to detect the statistical enrichment of differential expression genes in KEGG pathways (http://www.genome.jp/kegg/). The raw data of RNA-seq was deposited in the NCBI Sequence Read Archive (SRA) under accession numbers, SRR 11452004-SRR11452009 and SRR12755782-SRR12755787.

Luciferase assay
Empty pGL3-promoter or NTRK1-containing luciferase reporter plasmids were co-transfected with pRL-TK into U2OS cells along with control vector, KDM6A or the KDM6A-ED mutant as indicated. Luciferase activity after 48 h measured by the Dual Luciferase Reporter Gene Assay Kit (Beyotime, RG027, Shanghai, China) on the Hidex Sense instrument (Hidex, Finland). Firefly luciferase readings were normalized to the co-transfected Renilla Luciferase control.

Chromatin immunoprecipitation assay
Chromatin immunoprecipitation (ChIP) assays were performed using the Millipore ChIP kit according to the manufacturer's protocol [55]. Briefly, 3×10 7 cells were collected, fixed and sonicated with a Bioruptor sonicator (Diagenode) to generate DNA fragments of approximately 500 bp in length. Chromatin immunoprecipitates for proteins were amplified by quantitative PCR, normalized to input, and calculated as percentages of inputs. Fold enrichment levels indicate the fold changes over the negative control immunoglobulin G (IgG). The PCR primer sequences are described in Table S3.

Cell viability assay
Cells were grown in medium containing 10% FBS for 24/72 h in the absence or presence of β-NGF (PeproTech, 450-01, Suzhou, China) and/or imatinib (MCE, HY-15463, NJ, USA). The number of living cells was counted on Countless II FL (Life technology) using 0.4% Trypan Blue.

Immunoprecipitation
Cells were extracted using the IP lysis buffer (Pierce, 87787, IL, USA) plus EDTA free protease inhibitor cocktail and PMSF. The clarified cell lysates (1×10 7 cells) were incubated at 4 ℃ for 12 h with anti-KDM6A antibody preabsorbed on Protein A/G Magnetic Beads (Pierce, 88803, IL, USA). After washing, the immunoprecipitates were denatured and subjected to SDS-PAGE and Western blot analyses with the appropriate antibodies.

Statistical analysis of data
SPSS software version 19.0 (IBM Corp., Armonk, New York, US) and GraphPad Prism v6 (GraphPad Software, Inc., San Diego, California, US) were used for all statistical analyses. Data were first evaluated for normal distribution using the Shapiro-Wilk method and homogeneity of variance with the Levene method. Pairwise comparisons of normally distributed data were analysed using Student's t test or for multigroup comparisons, one-way analysis of variance (ANOVA) with post hoc Tukey's test. Data not meeting normal distribution/homogeneity of variance were compared using Kruskal-Wallis and Mann-Whitney non-parametric tests. *, p < 0.05; **, p < 0.01; and ***, p < 0.001 denote statistically significant changes.