Inhibition of IRE1α-driven pro-survival pathways is a promising therapeutic application in acute myeloid leukemia

Survival of cancer cells relies on the unfolded protein response (UPR) to resist stress triggered by the accumulation of misfolded proteins within the endoplasmic reticulum (ER). The IRE1α-XBP1 pathway, a key branch of the UPR, is activated in many cancers. Here, we show that the expression of both mature and spliced forms of XBP1 (XBP1s) is up-regulated in acute myeloid leukemia (AML) cell lines and AML patient samples. IRE1α RNase inhibitors [MKC-3946, 2-hydroxy-1-naphthaldehyde (HNA), STF-083010 and toyocamycin] blocked XBP1 mRNA splicing and exhibited cytotoxicity against AML cells. IRE1α inhibition induced caspase-dependent apoptosis and G1 cell cycle arrest at least partially by regulation of Bcl-2 family proteins, G1 phase controlling proteins (p21cip1, p27kip1 and cyclin D1), as well as chaperone proteins. Xbp1 deleted murine bone marrow cells were resistant to growth inhibition by IRE1α inhibitors. Combination of HNA with either bortezomib or AS2O3 was synergistic in AML cytotoxicity associated with induction of p-JNK and reduction of p-PI3K and p-MAPK. Inhibition of IRE1α RNase activity increased expression of many miRs in AML cells including miR-34a. Inhibition of miR-34a conferred cellular resistance to HNA. Our results strongly suggest that targeting IRE1α driven pro-survival pathways represent an exciting therapeutic approach for the treatment of AML.


INTRODUCTION
Acute myeloid leukemia (AML) is an aggressive hematological malignancy characterized by a small population of self-renewing leukemic stem cells (LSCs) giving rise to a large population of immature leukemic blasts [1][2][3][4]. LSCs are relatively insensitive to current therapies [3]. Many AML cells initially respond to treatment; however, relapse is often caused by LSCs that are intrinsically resistant to chemotherapy [5][6][7].
The overall long-term survival of AML patients remains extremely disappointing at approximately 30% to 50% [1,8,9]. Novel therapeutic approaches are clearly needed.
Hematopoietic cells, including LSCs, are exposed to low levels of oxygen in the bone marrow, which may cause accumulation of misfolded proteins in the endoplasmic reticulum (ER), thereby stimulating ER stress and activating the unfolded-protein-response (UPR) pathway [5,6,[10][11][12]. ER stress and its UPR are properly compensated in normal marrow hematopoietic cells [13,14]. In contrast, leukemic cells proliferating in a hostile environment of low oxygen and limited nutrients accumulate misfolded proteins in the ER, causing continuous ER stress with initiation of UPR [15,16]. Furthermore, leukemic cells produce mutant proteins at a high rate resulting in misfolded proteins [17]. Sustained UPR initiates cellular defense mechanisms rescuing leukemic cells from extreme cellular stress by limiting de novo entry of proteins into the ER which in turn enhances both protein folding capacity and degradation activity [16]. The ability of leukemic cells to handle ER stress may allow them to escape apoptosis and continue their growth [16,18].
The functions of the ER and its associated stress pathways in AML have been studied including recent studies that found approximately 25% of AML samples had detectable XBP1s, indicating activation of UPR and an increase of chaperone proteins [26,27]. Many studies have reported that perturbing the UPR with proteasome inhibitors, such as bortezomib, can enhance apoptosis of AML cells [28][29][30][31][32]. In this study, we explored whether AML cells have activated IRE1α which can be therapeutically targeted.

XBP1 and XBP1s are up-regulated in AML
IRE1 signaling pathway through XBP1 and XBP1s is strongly linked with ER stress and UPR [33]. To examine if XBP1 has a crucial role in AML, we first analyzed the AML methylation database (27k Illumina methylation version) from TCGA. Compared with normal samples, XBP1 was highly hypomethylated on its CpG island in AML cases ( Figure 1A). Consistent with the methylation status, XBP1 expression was significantly up-regulated in AML cases [5 previously published microarray databases ( Figure 1B) and our QRT-PCR results ( Figure 1C)]. A combination analysis of the 5 published databases showed that XBP1 ranked No. 679th of the most highly expressed genes in AML ( Figure 1B). Results were calculated by online analysis engine Oncomine (https://www.oncomine. org/resource/login.html). Interestingly, XBP1s was detectable in 85% (22 of 26) of the leukemia cell lines and 71% (17 of 24) of AML patient samples ( Figures  1D, 1E). Normal purified CD34+ myeloid stem cells did not have detectable XBP1s ( Figure 1E). XBP1s was also significantly elevated in AML samples from patients compared to CD34+ normal myeloid stem cells (p=0.0043, n=28) as measured by QRT-PCR ( Figure 1F). To investigate correlations between XBP1 expression and AML clinical features, we first performed statistical analysis to correlate the expression of XBP1/XBP1S with French-American-British (FAB) subtypes in our own dataset (Table S2 and Figure 1C, 1E, 1F). However, probably due to the limited numbers of cases, we did not observe a significant association between XBP1/ XBP1s and FAB subtypes among the 24 AML samples (data not shown). We next performed similar statistical analysis using TCGA AML dataset. Since XBP1s was not discernable from total XBP1 in the dataset, we only tested total XBP1 level. Interestingly, XBP1 expression was significantly increased in FAB M3 subgroup compared with M0, M1 and M2 but significantly decreased in M4-M7 subgroup ( Figure S1). The biological significance of these correlations requires further investigations.

IRE1α RNase inhibitors blocked splicing of XBP1 mRNA and exhibited cytotoxicity against AML cells
Recently, a novel small-molecule RNase inhibitor of IRE1 (MKC-3946) was noted to have potent anti-proliferative activity in multiple myeloma (MM) [34]. The compound was found to be very unstable in vitro; however, one of the two major hydrolyzed precursors, A-106 (2-hydroxy-1-naphthaldehyde, HNA) retained the IRE1α RNase inhibitory activity [35]. Tunicamycin (TM) induces ER stress and XBP1 splicing in many cells [36]. Following TM treatment, increased expression of XBP1s mRNA and decreased XBP1u (unspliced, transcriptional inactive form of XBP1) were observed in 293T and K562 myeloid leukemia cells ( Figure S2A). Compared with MKC-3946, HNA showed either the same or more potent ability to inhibit the activity of IRE1α to cleave XBP1 into the active XBP1s after TM induced activation of NB4 cells ( Figure S2B). STF-083010 is a newly developed IRE1α endonuclease specific inhibitor which has shown cytotoxic activity against human multiple myeloma [37,38]. Treatment of AML cells with increasing drug dosage showed slightly enhanced potency of HNA compared to STF-083010 ( Figures  S3A-D). HNA dose-dependently inhibited XBP1s expression induced by TM in AML cell lines and AML patient samples (Figures 2A-2C). HNA significantly decreased cellular viability of both AML cell lines (mean GI 50 =31 μM, n=8) and AML patient samples (mean GI 50 =35 μM, n=18) compared to untreated patient samples (mean GI 50 =154 μM, n=5, Figures 2C-2E). Importantly, HNA caused a significant inhibition (mean GI 50 =6 μM, n=6) of clonogenic growth in soft agar of AML cells from patients ( Figure 2F). In contrast, HNA had very low toxicity against normal human marrow mononuclear cells (mean GI 50 =123 μM, n=4) ( Figure  2E). We conducted western blotting assay on BALL1, (n=58) and AML samples (n=140) was analyzed using TCGA level 3 database. The p-values were calculated by student t test. B. 5 publicly available microarray databases showed XBP1 was highly expressed in AML samples compared with normal BM samples. 1. Andersson Leukemia [84]; 2. Haferlach [85]; 3. Stegmaier [86]; 4. TCGA [87]; 5. Valk [88]. The rank for a gene is the median rank for that gene across each of the analyses. The p-value refers to the median-ranked analysis. C. QRT-PCR analysis of AML blast cells from patients (n=22) compared with normal human CD34+ cells (n=6) showed significant up-regulation of XBP1, using GAPDH as an internal control (p<0.01). REH and K562 cell lines, and confirmed that the XBP1s protein levels were correlated with their mRNA levels. Specifically, K562 cells showed expression of both XBP1s mRNA and protein, whereas BALL1 and REH cells expressed neither mRNA nor protein of XBP1s ( Figures 1D and S2D). Furthermore, we confirmed that the ER stress inducer thapsigargin successively induced XBP1s expression ( Figure S2D).

IRE1α induced apoptosis and G1 cell cycle arrest in AML
Inhibition of UPR blocks pro-survival pathways is implicated in apoptosis and cell-cycle arrest in several model systems [39]. In a dose-dependent manner, HNA treatment of AML cell lines and primary samples significantly increased the percent annexin V positive cells  Figures 2C and 2D. "Con", untreated control. The GI 50s were calculated by Graphpad software. F. Soft agar clonogenic assays of 6 AML patient samples exposed to HNA. Figures are representative example of 3 replicates. GI 50s were calculated. Data represent mean ± SD, n=3.

Absence of Xbp1 in murine myeloid cells produced resistance to Ire1α inhibitors
To assess whether Xbp1 is a major driver of Ire1α signaling pathway, we generated a model based on bone marrow progenitor cells from mice carrying a floxPflanked allele of Xbp1 (Xbp1 flox/flox ). After Cre activation,  Figures (A, B, E) are representative example of 3 replicates. Data represent mean ± SD, n=3. www.impactjournals.com/oncotarget more than 92% of Xbp1 was confirmed to be deleted in the floxed cells as evidenced by QRT-PCR ( Figure 4A). Ire1α-Xbp1 in vivo inhibition effect was confirmed as HNA decreased TM induced Xbp1s levels in bone marrow cells of mice ( Figure S6). Myeloid cells (94 % CD34+) with deleted Xbp1 showed slower proliferation ( Figure  4B) and increased resistance to both IRE1α inhibitors (toyocamycin and HNA) (Figures 4C, 4D). Treatment with TM (1 mg/ml) produced negligible cytotoxicity to the murine BM cells ( Figure S8B). Combination of TM (1 mg/ml) with HNA, increased cell viability in Xbp1 -/cells compared to vector control cells (Figures 4E), suggesting that Ire1α induced UPR is at least partially reliant on Xbp1.

Combination of HNA with either bortezomib or AS 2 O 3 synergistically inhibited growth of AML cells
Bortezomib is a potent 26S proteasome inhibitor which induces terminal UPR and apoptosis in many cells [42][43][44][45]. AS 2 O 3 is a leading therapy for treatment of acute promyelocytic leukemia (APL) [46] and has recently been shown to induce ER stress in cancer cells including leukemia cells [47,48]. Both drugs induced apoptosis associated with p-JNK activation in cancer cells [49,50]. We hypothesized that the combination of an IRE1α RNase inhibitor with either bortezomib or AS 2 O 3 , will enhance the inhibition of proliferation of AML cells. Indeed, various concentrations of either drug combined with HNA synergistically inhibited growth of both NB4 cells and an AML patient sample #19 (Figures 5A-5D). Also, bortezomib induced expression of p-JNK, as well as reduced levels of p-MAPK and p-PI3K in NB4 cells. (Figure 5E).

miR-34a triggered sensitivity of IRE1 inhibitor in vitro
Small RNA antagonist against miR-34a was transiently transfected into 3 AML cell lines (K562, NB4, U937), and cell viability was examined after treatment with HNA (12.5-100 μM). Knockdown efficiency of miR-34a by the antagonist was validated both at the preand mature miR level by QRT-PCR ( Figures 7A-7C, left panels). Inhibition of miR-34a conferred modest survival advantage compared to HNA alone in these AML cells ( Figures 7A-7C, right panels). In addition, HNA inhibited transcriptional levels of several targeted genes of miR-34a (c-Myc, cyclin D1, CDK4) in NB4 cells ( Figure 7D); Also in the AML cells, the miR-34a antagonist restored protein expression levels of c-Myc and cyclin D1 proteins, these had been inhibited in their expression in the presence of HNA alone. (Figure 7E).

DISCUSSION
The UPR is a defense mechanism activated by cells during stressful conditions in response to an accumulation of misfolded proteins in the ER [51]. Cancer cells are usually exposed to many stressful environments (e.g. hypoxia, nutrient starvation, oxidative stress and other metabolic dysregulation) resulting in continued ER stress. Furthermore, genomic mutations can also lead to accumulation of misfolded proteins. The UPR is often activated to help cancer cells escape from ER stressinduced cell death [52]. Reactive oxygen species (ROS) in AML cells can also stimulate ER stress and the UPR [11,53,54]. Studies have particularly focused on the functional roles of UPR proteins in multiple myeloma cells because chronic ER stress occurs in these cells  [37,55,56]. Activation of the unfolded protein response including XBP1 splicing has been noted in AML samples [26,27]. In our study, we confirmed that one of the major branches of the UPR (IRE1-XBP1s) is frequently activated in AML cell lines and AML patient samples, suggesting that targeting the UPR may be a promising adjunctive approach for treatment of AML.
XBP1 and its spliced form XBP1s have been reported to be upregulated in several types of cancers, and blockage of the IRE1/XBP1 pathway is considered as a promising therapeutic option [55,[57][58][59][60][61]. XBP1 is one of the most well studied genes modulated by IRE1α, which has important roles in the regulation of cell survival and UPR as a downstream target of IRE1α [55,[62][63][64][65]. However, XBP1 is not the only factor in response to IRE1α inhibition. In fact, as an RNase, IRE1α recognizes and cleaves a consensus element, CUGCAG, in target RNAs upon cellular stresses [66]. Therefore, IRE1α has the capacity to target directly a number of RNAs (both coding and noncoding) and regulate their expression levels through Regulated IRE1-Dependent Decay (RIDD). For example, in a recent study using MEF cells, IRE1α was found to cleave four anti-CASP2 miRs (miR-17, -34, -96 and -125) [22,23]. Based on our bioinformatic analysis of miRs and RNA microarray results, we found that many pre-miRs with IRE1α cutting consensus motif were upregulated upon HNA treatment. We focused on 5 of these candidate miRs and QRT-PCR results demonstrated that they were induced in the presence of IRE1α inhibitors ( Figure 6). Together, these results suggest that the antineoplastic effects of IRE1α inhibitors are the results of the enhanced nuclease activity that generates not only XBP1s, but also a number of other RNAs, including miRNAs.
We particularly focused on miR34a. Recent studies indicate that miR-34a is a tumor suppressor [67]. Downregulation of miR-34a causes resistance to chemotherapy [67][68][69]. Targets of miR-34a include caspase-2, c-Myc, Bcl-2, cyclin D1, MET and SIRT1. miR-34a dependent inhibition of SIRT1 can increase acetylation and activation of p53 resulting in up-regulation of p21 and PUMA [70]. By post-transcriptional blockade of these genes, miR-34a suppresses migration and induces apoptosis, G1 cycle arrest and senescence in cancer cells [71,72]. Our study showed that IRE1α inhibitors significantly increased pre-and mature miR-34a mRNA levels, associated with inhibition of CDK4, c-Myc, Bcl-2 and Cyclin D1 and induction of p21 cip1 and Silencing of miR-34a by small RNA antagonist significantly induced resistance of AML cells to IRE1α inhibitors and restored levels of miR-34a targeted oncogenes such as c-Myc and Cyclin D1. These results indicate that miR-34a plays an important role in IRE1αdependent UPR in AML. Through cleavage of miR-34a, IRE1α decreases miR-34a induced apoptosis and helps AML cells to escape death.
Bortezomib is a proteasome inhibitor which has been approved by FDA as the first example of UPRmodulating regimen for the treatment of Multiple Myeloma [73]. Recently, bortezomib was also approved for treatment of mantle cell lymphoma [74]. In the context of AML, several clinical studies have shown that bortezomib has potent anti-neoplastic activity, and the mechanisms include induction of apoptosis and transcriptional inactivation of several important AMLdrivers such as DNA methyltransferases (DNMTs) and receptor tyrosine kinases (RTKs) [29,38,75,76]. Moreover, bortezomib has shown greater therapeutic value when applied in combination with traditional cytotoxic chemotherapies to treat AML [31,77,78]. Therefore, although not yet approved by FDA, bortezomib has demonstrated its promising therapeutic merit and is considered as one of the candidate drugs for the treatment of AML. On the other hand, several studies have shown that IRE1α inhibitors in combination with bortezomib strongly impaired the growth of multiple myeloma cells both in vitro and in vivo [37,55]. Mechanistically, IRE1α inhibition can overcome the tumor cytoprotective effects conferred by bortezomib-induced UPR via activation of XBP1s [55]. In the present study, we also showed that combined addition of HNA with bortezomib synergistically increased apoptosis of AML cells associated with the up-regulation of CHOP and p-JNK. Therefore, we believe that our results will provide useful information to the investigators who are interested in developing bortezomib for the treatment of AML.
Treatment of acute promyelocytic leukemia (APL) with AS 2 O 3 has unique favorable cure rates [79], and the major mechanism of action is through the degradation of PML-RARα, the driver of APL [80]. Notably, AS 2 O 3 has been reported to enhance UPR by increasing the expression of GRP78, CHOP, phosphorylated eIF2α and ATF4 [81,82]. In our study, the combination of HNA and AS 2 O 3 synergistically increased cell apoptosis of the APL cells NB4, which might be due to corroborative inhibition of different branches of the UPR pathway. In the previous clinical trials, QT interval prolongation and APL differentiation syndrome are the most serious side-effects of AS 2 O 3 treatment [83]. The combined use of HNA and AS 2 O 3 may improve the toxicity profile.
In conclusion, we found that XBP1 and its spliced form (XBP1s) are often increased in AML. Inhibition of IRE1α RNase activity by small molecules inhibited AML cell growth. This inhibition of proliferation probably occurred by inducing ER stress by blocking the compensatory pathways including the prevention of cleavage of tumor suppressor miRs (e.g. miR-34a) which may augment the anti-proliferative effect of IRE1α inhibitors. Therefore, compounds which inhibit activation of IRE1α represent a novel pathway for cell kill and may be a useful compliment to chemotherapy.

MATERIALS AND METHODS
Information of reagents and vendors, and all methods are described in the Supplemental Methods.