Janus and PI3-kinases mediate glucocorticoid resistance in activated chronic leukemia cells

Glucorticoids (GCs) such as dexamethasone (DEX) remain important treatments for Chronic Lymphocytic Leukemia (CLL) but the mechanisms are poorly understood and resistance is inevitable. Proliferation centers (PC) in lymph nodes and bone marrow offer protection against many cytotoxic drugs and circulating CLL cells were found to acquire resistance to DEX-mediated killing in conditions encountered in PCs including stimulation by toll-like receptor agonists and interactions with stromal cells. The resistant state was associated with impaired glucocorticoid receptor-mediated gene expression, autocrine activation of STAT3 through Janus Kinases (JAKs), and increased glycolysis. The JAK1/2 inhibitor ruxolitinib blocked STAT3-phosphorylation and partially improved DEX-mediated killing of stimulated CLL cells in vitro but not in CLL patients in vivo. An automated microscopy-based screen of a kinase inhibitor library implicated an additional protective role for the PI3K/AKT/FOXO pathway. Blocking this pathway with the glycolysis inhibitor 2-deoxyglucose (2-DG) or the PI3K-inhibitors idelalisib and buparlisib increased DEX-mediated killing but did not block STAT3-phosphorylation. Combining idelalisib or buparlisib with ruxolitinib greatly increased killing by DEX. These observations suggest that glucocorticoid resistance in CLL cells may be overcome by combining JAK and PI3K inhibitors.


INTRODUCTION
High-dose glucocorticoids (HDGCs) are a useful treatment modality for CLL even in the era of novel agents [1,2]. HDGCs sometimes yield impressive remissions [3] but are not curative and eventually select for resistance [2]. In circulating CLL cells, glucocorticoids (GCs) activate a transcriptional program that leads to atrophy and death subsequent to decreased glycolysis from down-regulated pyruvate kinase (PK) M2 expression and function [4].
Circulating CLL cells arise from proliferation centers (PCs) in lymphoid organs and bone marrow [5]. CLL cells in these microenvironments divide in response to stimulation from signals such as antigens, costimulatory molecules, toll-like receptor (TLR) ligands, chemokines, and cytokines produced by T, nurse, stromal, and other CLL cells. These supportive microenvironments are sanctuaries that allow CLL cells to resist cytotoxic drugs [6,7] but little is known about their effects on responses to GCs.

Research Paper www.impactjournals.com/oncotarget
Surrogate models are required to study the behavior of cells in PCs as they are not readily accessible for experimental studies. Such models include coculturing CLL cells with stromal cells that express costimulatory molecules and/or activating CLL cells with mitogenic factors such as IL4 or ligands for TLRs, particularly TLR9 [8][9][10][11][12]. TLR9-and TLR7/8-signaling are similar and we have used the TLR-7/8 agonist resiquimod along with IL2 to mimic T cell signals in order to capture important features of PCs encountered by CLL cells in vivo [7,[10][11][12]. In this system CLL cells increase in size, proliferate, up-regulate expression of activation markers, make factors that activate the STAT3 transcription factor [13], and acquire resistance to drugs such as vincristine, fludarabine, and venetoclax [14,15,16]. In common with most in vitro PC models, the limitations include higher oxygen and glucose concentrations [17] and fewer intercellular interactions than exist in vivo. Nevertheless, results with this system suggest that microenvironmental signals from cytokines and phosphatidylinositol-3-kinase (PI3K) mediate resistance of CLL cells to glucocorticoids. Concomitant inhibition of both pathways may improve therapeutic outcomes with glucocorticoids in CLL.

Stimulated CLL cells resist glucocorticoidmediated transcriptional death
Unstimulated CLL cells are killed by glucocorticoids such as DEX following a program of atrophy involving down-regulation of PKM2 activity and decreased glycolysis [4]. Cells can survive this death program by upregulating expression of genes such as PDK4 and PPARA Figure 1: Effect of Dexamethasone on stimulated CLL cells. CLL cells from 10 patients were purified and cultured in the presence (2S) or absence (unstimulated (US)) of IL-2 and resiquimod with or without (control, CTR) dexamethasone (30 μM, DEX) in serum free media. a. Cells were stained with 7AAD 48 h after DEX treatment and analyzed by flow cytometry. Specific death (i.e. the difference between the percentages of 7AADcells in the control and DEX treated samples) and median forward scatter (mfi) (indicating cell size) for individual patient samples are shown on the left and right panels, respectively. b. mRNA levels of PPARA (left) and PDK4 (right) (relative to HPRT) were determined by quantitative PCR in 20 different samples 18 h after treatment. c. Immunoblots of Ser211 phosphorylated glucocorticoid receptor (P-GR) in unstimulated CLL cells (-) and 2S-stimulated cells (+) with and without dexamethasone (DEX) treatment. Protein lysates for patient cells were collected 18 h after DEX treatment and immunoblots were probed with antibodies to pGR and actin used as a loading control. Representative results for patients P5, P6 and P7 are shown. *, p < .05; **, p < .01; ns, not significant. and switching to fatty acid oxidation [4]. However, CLL cells stimulated with IL2 and resiquimod (2S) [11,12,19] are resistant to DEX-mediated killing at a dose of 30 μM that approximates plasma levels following HDGCs (Figure 1a, left panel) [2,4]. While unstimulated CLL cells atrophied and became smaller, 2S-stimulated CLL cells were larger and did not shrink significantly in response to DEX (Figure 1a, right panel). Furthermore 2S-stimulated CLL cells did not increase their expression of PPARA and PDK4 mRNA transcripts, which are up-regulated in DEXtreated unstimulated cells (US) (Figure 1b).
The failure of DEX to kill 2S-stimulated cells could reflect altered glucocorticoid receptor (GR) expression or function. Ligand binding leads to phosphorylation and release of cytoplasmic GRs from heat shock proteins with subsequent entry into the nucleus to mediate gene transcription [4]. GR phosphorylation after DEX was not much different in US and 2S-stimulated cells, suggesting the initial steps of the response were intact (Figure 1c).

JAK inhibitors sensitize activated CLL cells to DEX in vitro but not in vivo
Cytokine signaling through Janus kinases (JAKs) inhibits transcriptional activity of the glucocorticoid receptor [20,21] and 2S-stimulated cells produce a number of cytokines, including IL10 and TNF-α [12]. These observations suggested JAK inhibitors might increase DEX-mediated killing of 2S-stimulated CLL cells.
Ruxolitinib is a selective inhibitor of JAK1 and JAK2 that is licensed to treat myelofibrosis [22] and being evaluated in CLL [23]. Ruxolitinib decreased cell size and IL10 production and increased TNFα production by 2S-stimulated CLL cells in vitro (Figure 2a and 2b). Consistent with a role for STAT3 and IL10 in promoting glucocorticoid-resistance, ruxolitinib significantly increased the killing of 2S-stimulated cells by DEX ( Figure 2c). A clinical trial of ruxolitinib as first-line treatment for CLL patients [23] provided an opportunity to assess the effect of this drug on glucocorticoid responses in vivo. Three subjects in this trial had been treated with HDGCs (187.5 mg/m 2 DEX for 4 days) several months before commencing ruxolitinib and required another course of HDGCs on ruxolitinib to treat cytopenias. Ruxolitinib was administered for all 4 days of HDGCs in patient 1 but the overlap was 1 day in patients 29 and 30 as ruxolitinib was held following initiation of DEX (Figure 2d). Circulating lymphocyte counts (WBC) prior to HDGCs and 4 weeks later were used to measure clinical response. The decrease in circulating lymphocyte counts after HDGCs was not enhanced by ruxolitinib, suggesting CLL cells had not been sensitized in vivo (Figure 2d).

Identification of PI3K/AKT inhibitors as glucocorticoid-sensitizing agents by automated high-content confocal fluorescent microscopy
Other signaling pathways activated in CLL cells following stimulation with IL2 and resiquimod might mediate resistance to glucocorticoids and insensitivity to JAK inhibitors. To identify these pathways, an unbiased high-content image-based microscopic approach [15,24] was used to screen a library of 320 kinase inhibitors (KIs) (Supplementary Table S2) on 2S-stimulated CLL cells from three different patients. The KIs (1 µM) were evaluated alone and in combination with DEX (30 µM) (Figure 3a) with the hypothesis that common hits that enhanced DEX-mediated death should reveal the protective pathways. Cell stress and death were assessed by automated image analysis of 2 fluorescent channels. Cells were stained with a red lipophilic TMRE dye that measures loss of mitochondrial membrane potential in dying cells and Draq5 (blue) to obtain total cell counts. Drug responses were assessed by applying intensity and area thresholds established from DMSO treated controls [15l]. KI efficacy was assessed by z-score analysis of the TMRE signal and compounds with a z-score ≥ 3 were identified as hits for DEX-enhanced kill (i.e. an increase in dead cells above KI treatment alone). Stimulated cells from all three patients were resistant to DEX (Figure 3a, bottom) and exhibited variable responses to DEX plus KI treatment ( Figure 3a). Compounds were ranked by average z-score across all 3 patients, and the most common targeted pathways indicated by a color scheme ( Figure  3, bottom). To identify the most important pathways that mediated glucocorticoid resistance, the top 38 KIs with z-scores ≥ 3 for at least 2/3 patients were further distinguished from compounds with z-scores from ≤ 2 to ≥7 as shown in the magnification heatmap ( Figure 3b, Supplementary Table S2).
A JAK inhibitor was among the top 38 KIs, but the major classes of inhibitors that sensitized 2S cells to DEX were inhibitors of receptor tyrosine kinases (EGFR, PDGFR, VEGFR, 6/38 inhibitors, orange), cell cycle regulators (7/38 inhibitors, green), and members of the PI3K/AKT/GSK3 pathway (8/38 inhibitors, purple) ( Figure 3b). The high-frequencies of these pathways could not be explained simply by the proportion of pathwayspecific inhibitors in the library. While compounds targeting the first two pathways are not used presently to treat CLL, the latter pathway is an important target [25] and was chosen for more detailed studies.

Inhibition of glycolysis restores glucocorticoidsensitivity in stimulated CLL cells
Activation of the PI3K/AKT pathway is associated with increased glucose transport and glycolysis [26] and marked by phosphorylation of substrate FOXO proteins [27]. Consistent with increased glycolysis, PKM2 transcript levels were increased in 2S-stimulated cells ( Figure 4a). Inhibition of AKT can be achieved by inhibiting glycolysis with 2-deoxyglucose (2-DG) [28] and sensitizes acute leukemia cells to DEX [29,30] Based on these results, 2-DG should improve glucocorticoid-efficacy in CLL patients but it is unfortunately too toxic for clinical use [31]. The AKT pathway can also be targeted by clinically relevant PI3K inhibitors such as idelalisib (IDE), a specific inhibitor of the PI3Kδ isozyme, and buparlisib (BKM), a pan-class I PI3K inhibitor [25,32,33,34]. These agents were evaluated for their ability to sensitize 2S-stimulated CLL cells to DEX.

PI3K inhibitors combined with ruxolitinib enhance DEX-mediated death in stimulated CLL cells
Single agent buparlisib and idelalisib were not toxic to 2S-stimulated CLL cells at concentrations up to 3 μM (Supplementary Figure 1). A dose of 0.5 μM, below plasma concentrations obtained with these drugs [32,33], was chosen for further studies and found to increase DEXmediated killing of 2S-stimulated cells (Figure 5b; left 3 panels). However, the increases, albeit significant, were of the order of 10-20% and similar to those obtained with ruxolitinib in vitro that proved not to have therapeutic efficacy in vivo (Figure 2c, 2d). These observations suggested a need to inhibit additional compensatory signaling pathways and achieve greater killing in vitro to expect to overcome glucocorticoid resistance in CLL cells in patients [35]. www.impactjournals.com/oncotarget   To identify signaling pathways that remained active in the presence of PI3K inhibitors, 2S-stimulated CLL cells from four patients were treated with idelalisib or buparlisib, analyzed with a phosphokinase array (not shown), and the results confirmed by immunoblotting (Figure 5a). Regardless of the presence of DEX, phospho-STAT3 levels were found to remain high in the face of PI3K inhibition (Figure 5a). The effect of adding ruxolitinib was then evaluated due to its ability to inhibit JAK-mediated STAT3 activation [23]. Treatment with both ruxolitinib and burparlisib or idelalisib eliminated pSTAT3 almost entirely, especially in the presence of  (Figure 5a). Importantly, DEX-mediated death was enhanced considerably when either PI3K inhibitor was used together with ruxolitinib (Figure 5b, right 3 Figure 1).

Combining kinase inhibitors overcomes glucocorticoid resistance in CLL cells cocultured with CD40L-expressing stromal cells
These results, obtained in a specific in vitro model using 2S-stimulated CLL cells, suggested that PI3K inhibitors combined with ruxolitinib might be an effective strategy for overcoming glucocorticoid-resistance in CLL microenvironments. To show the results were not simply model-dependent, the experiments were repeated in a stromal cell coculture model. CLL cells from four patients were cocultured with murine bone marrow stromal cells (OP9) expressing human CD40L [9,15,36,37,38]. Consistent with results from 2S-stimulated cells, cocultured CLL cells acquired resistance to DEX (30 µM) and were not killed by ruxolitinib, buparlisib or idelalisib alone ( Figure 6). Neither buparlisib nor idelalisib reversed DEX-resistance (average kill of 9% and 6%, respectively, Figure 6a and 6b). However, combining the PI3K inhibitors with ruxolitinib strikingly increased DEXmediated cell death (average cell kill of 77% and 71% for buparlisib and idelalisib, respectively (Figure 6a and 6b, n = 4)), confirming the importance of blocking multiple kinase-dependent pathways.

Effect of ibrutinib on glucocorticoid-resistance
The Burton's tyrosine kinase (BTK) inhibitor ibrutinib was not included in the KI library used for screening (Supplementary Table S2). Based on its importance in the clinical management of CLL [1] and ability to inhibit AKT-signaling indirectly [39], studies were carried out to determine if ibrutinib might also combine with ruxolitinib to sensitize resistant CLL cells to glucocorticoids.
In the 2S model, no toxicity was seen with concentrations of ibrutinib alone up to 3 µM. In contrast to buparlisib and idelalisib, no toxicity occurred when 3 µM of ibrutinib was combined with ruxolitinib (Supplemental Figure 1).
In the OP9-CD40L stromal cell coculture model, ibrutinib had minimal effects on cell death, either as a single agent or in combination with DEX ( Figure 6) (average cell death of 12% and 9% respectively, n = 4). Combining ibrutinib and ruxolitinib enhanced DEXmediated death in 2 out of 4 patients (average total kill of 54% for all four patients (Figure 6)).

DISCUSSION
The studies in this manuscript provide insights into why glucocorticoids are only partially effective in treating CLL. They suggest that signals from the PC microenvironment prevent glucocorticoids from initiating cell death transcriptional programs in CLL cells ( Figure  1). These signals appear to involve multiple pathways. Cytokine-signaling through Janus Kinases is one important mediator of glucocorticoid-resistance ( Figure  2). Our high-throughput screen of more than 300 kinase inhibitors also identified PI3K/ATK, receptor tyrosine kinases and cell cycle regulators as potential targets (Figure 3). Taken together, the observations suggest that blocking protective signals from the microenvironment with kinase inhibitors may be a useful strategy to improve the therapeutic efficacy of glucocorticoids in CLL. However, combinations of kinase inhibitors are needed to deal with the complexity of the protective signals ( Figure  5, 6).
The precise mechanisms that cause stimulated CLL cells to become resistant to DEX are not known and may be specific for individual patients. While there was some variability in resistance of 2S-stimulated CLL cells to DEX-mediated killing, it did not correlate with the clinical characteristics of the patients (Supplementary  Table S1). There was also some variability in the responses of 2S-stimulated CLL cells to different kinase inhibitors (Figure 3). Combining PI3K and JAK inhibitors is expected to improve the outcomes of most CLL patients who require high-dose glucocorticoids but there may be some who will benefit from unique kinase inhibitors. High-content microscopy-based screening of kinase inhibitor libraries, as described here (Figure 3) and elsewhere [15], may potentially identify such inhibitors.
The initial steps of glucocorticoid signaling are intact in stimulated CLL cells (Figure 1c) and the explanation for glucocorticoid-resistance appears to lie further downstream at the transcriptional level [40]. Promoter regions of genes that are subject to regulation by glucocorticoids are often complex and the outcome may be affected significantly by cooperating transcription factors such as members of the FOXO and PPAR gene families [41]. Activated signaling pathways in stimulated CLL cells affect the localization or expression of these factors [42], preventing glucocorticoids from mediating the transcriptional responses responsible for cellular atrophy and death [4]. For example, PPARα is significantly down-regulated in stimulated CLL cells [11] and phosphorylation prevents FOXO1 from entering the nucleus to support transcriptional programs with the GR [27]. The association of high-levels of pFOXO1 in stimulated DEX-resistant CLL cells and its absence in DEX-sensitive cells (Figure 4c) is consistent with a role in glucocorticoid-mediated resistance of CLL cells as in other cell-types [42,43]. The role of cytokines ( Figure 2b) and receptor tyrosine kinases (Figure 3, yellow) may be linked to their ability to activate the PI3K/AKT pathway (Figure 3, 5, 6) as well as drive CLL cells through the cellcycle [12]. Cell cycle proteins (Figure 3, green) have also been shown to negatively regulate transcription by nuclear receptors [44].
Despite increased killing of stimulated CLL cells in vitro and prevention of STAT3-phosphorylation in response to exogenous cytokines in vitro ( Figure  5), clinical responses to HDGCs were not noticeably improved by ruxolitinib (Figure 2d). A number of reasons may account for the apparent inability of single-agent ruxolitinib to increase therapeutic efficacy of HDGCs in vivo. The sensitizing effects ascribed to ruxolitinib were based on a specific in vitro microenvironmental model (Figure 1) and used a 10% increase in DEX-mediated killing as evidence for potential clinical importance (Figure 2c). Absence of sensitization to glucocorticoids in vivo may reflect the limitations of this model or that more in vitro activity (eg. as seen with ruxolitinib and PI3K inhibitor combinations ( Figure 5 Combined use of PI3K inhibitors with ruxolitinib to potentiate therapeutic activity of HDGCs in CLL is supported by the experiments in Figures 5 and 6. Information about the potential toxicity of combining buparlisib and ruxolitinib will be obtained from on-going clinical trials in myelofibrosis (NCT01730248) [22]. The selective PI3Kδ-inhibitor idelalisib may be less toxic than buparlisib due to the restricted expression of this isozyme. Either of these combinations may improve the therapeutic efficacy of glucocorticoids for most patients. Individual drug screening may optimize the identification of effective combinations for others ( Figure 3) [15].

CLL patients and cells
CLL cells were isolated from patients who were otherwise untreated for at least 3 months [4,15,18]. Clinical characteristics of the patients are shown in Supplementary Table S1. Protocols were approved by the Sunnybrook Ethics Review Board. The clinical trial of ruxolitinib [23] was registered as NCT02015208 and the high-dose glucocorticoid regimen administered to some patients consisted of oral dexamethasone (187.5 mg/m 2 ) for 4 days [2].

Cell culture
Unless otherwise specified, purified CLL cells were cultured at a concentration of 1 x 10 6 cells/mL in RPMI-1640 medium supplemented with Transferrin and 0.02% AlbuMAX II in 6-or 24-well plates (BD Labware) at 37°C in 5% CO 2 for the times indicated. Resiquimod and IL2 were used at 1 µg/ml and 500 U/ml. CLL cells stimulated in this manner are designated "2S" cells [11,12]. OP9-CD40L cells were generated and cultured in alpha-MEM supplemented with 20% FBS (Gibco) as described before [15].

CLL-OP9 coculture experiments
CD40L-expressing OP9 bone marrow stromal cells were seeded into 24-well plates at a concentration of 5x10 4 cells/well the day before initiation of CLL cocultures. CLL cells were added onto the stromal layer at a final concentration of 1x10 5 cells/ml. For comparison, suspension cells of the same CLL sample were cultured at a density of 1x10 5 cells/ml. Coculture and suspension cells were treated with dexamethasone (30 µM, DEX), ruxolitinib (1 µM, RUX), ibrutinib (3 µM, IBR), idelalisib (3 µM, IDE) or buparlisib (3 µM, BKM) alone or in combination as indicated. Cells were collected by pipetting after 48 h, leaving the intact adherent stromal layer behind.

Plasma inhibitor assay for ruxolitinib levels
Plasma samples were obtained before each cycle of ruxolitinib and stored at -80°C. To determine if plasma levels of ruxolitinib were sufficient to inhibit IL6-mediated STAT3-phosphorylation, 2x10 6 cells were incubated with 1 ml of thawed plasma for 30 min and then stimulated with IL6 (40 ng/ml). Protein extracts were collected after 30 min and subjected to immunoblotting.

Flow cytometry
Cell viability was determined by staining cells with 3.5 mL of 7AAD for 10 min and washing in phosphatebuffered saline (PBS). Flow cytometric detection of Annexin V and PI stained cells was performed using the Alexa Fluor 488 Annexin V/dead cell apoptotis kit (Thermofisher Scientific) according to the manufacture's protocol. Cells were co-stained with Alexa Fluor 488 Annexin V and propidium iodide (PI). Flow cytometry was performed using the BD LSRII flow cytometer. Annexin V and PI were excited using the 488 nm blue laser and 505 LP filter and the 545 nm laser and 635 LP filter, respectively. Twenty-five thousand counts were measured and data analyzed using Flowjo 9.6 software (Tree Star, San Carlos, CA).

Immunoblotting
Protein extraction and immunoblotting were performed as previously described [18].

Real-time PCR
RNA was prepared with the RNeasy mini kit (Qiagen, Valencia, CA, USA) and cDNA was synthesized from 2 g of RNA using Superscript III reverse transcriptase (Life Technologies, Invitrogen), according to the manufacturer's instructions. The following primers were used: PKM2 forward, GTCGAAGCCCCATAGTGAAG; reverse, ATGTCCTTCTCCGACACAGC, PPARA forward, CTGGAAGCTTTGGCTTTACG; reverse, ACCAGCTTGAGTCGAATCGT, PDK4 forward, CATACTCCACTGCACCAACG; reverse, CCTGCTTGGGATACACCAGT, and HPRT forward, GAGGATTTGGAAAGGGTGTT; reverse ACAATAGCTCTTCAGTCTGA. Polymerase chain reactions were carried out in a DNA engine Opticon System (MJ Research, Waltham, MA, USA) and cycled 34 times after denaturation (95°C, 15 min) with the following parameters: denaturation at 94°C for 20 sec, annealing at 58°C for 20 sec, and extension at 72°C for 20 sec. mRNA abundance was evaluated by a standard amplification curve. Copy numbers were determined from two independent cDNA preparations for each sample and expressed as the relative fold change of the target to HPRT.

Cytokine measurements
Cytokine levels in culture supernatants were measured with commercial ELISA kits for human IL10 and TNFα according to the manufacturer's instructions (R&D Systems, Minneapolis, MN).

High content kinase inhibitor screening
CLL cells were seeded at 1.5-2x10 6 cells/ml with IL-2 and resiquimod in a 384-well plate (Cell Carrier; PerkinElmer, Waltham, MA) and cultured at 37 ο C and 5% CO 2 . Individual members of the kinase inhibitor library were dissolved in DMSO and added to the cultures at final concentrations of 1 μM. After 72 h, fluorescent dyes were added for 30 min prior to image acquisition. Tetramethylrhodamine ethyl ester (TMRE; Life Technologies) at 10 nM and Draq5 (Biostatus, Leicestershire, UK) at 5 μM were used to assess mitochondrial membrane potential and detect cell nuclei for cell segmentation, respectively.

High content confocal fluorescent microscopy and image analysis
Confocal microscopy was performed at 37 º C and 5% CO 2 using an Opera QEHS high-content screening system (PerkinElmer) as described [15]. TMRE intensity thresholds were established for each sample from untreated control cells compared to a positive control for dead cells treated with staurosporin.
Based on the TMRE signal, drug combinations that enhanced DEX-mediated cell death were identified by z-score analysis using the mean and standard deviation of eight in-plate DMSO control replicate wells. Enhanced DEX-mediated cell kill was calculated by the total cell kill induced by the combination of DEX and KI above cell kill induced by kinase inhibitors alone. Kinase inhibitors with z-scores ≥3 were scored as positive hits.