Abstract
Glioblastoma multiforme lacks effective therapy options. Although deregulated kinase pathways are drivers of malignant progression in glioblastoma multiforme, glioma cells exhibit intrinsic resistance toward many kinase inhibitors, and the molecular basis of this resistance remains poorly understood. Here, we show that overexpression of the protein phosphatase 2A (PP2A) inhibitor protein PME-1 drives resistance of glioma cells to various multikinase inhibitors. The PME-1–elicited resistance was dependent on specific PP2A complexes and was mediated by a decrease in cytoplasmic HDAC4 activity. Importantly, both PME-1 and HDAC4 associated with human glioma progression, supporting clinical relevance of the identified mechanism. Synthetic lethality induced by both PME-1 and HDAC4 inhibition was dependent on the coexpression of proapoptotic protein BAD. Thus, PME-1–mediated PP2A inhibition is a novel mechanistic explanation for multikinase inhibitor resistance in glioma cells. Clinically, these results may inform patient stratification strategies for future clinical trials with selected kinase inhibitors in glioblastoma multiforme. Cancer Res; 76(23); 7001–11. ©2016 AACR.
Introduction
Glioblastoma multiforme remains extremely difficult to treat (1, 2). The failure of targeted therapies in glioblastoma multiforme can be linked to inherently high apoptotic resistance of glioblastoma multiforme cells, especially glioblastoma multiforme cancer stem cells (GSC; refs. 3, 4). Particularly puzzling is dichotomy between the fact that glioblastoma multiforme is driven by receptor tyrosine kinase signaling (1, 2), but specific small-molecule kinase inhibitors have failed to provide any clinical benefit in numerous clinical trials (3–5).
The serine/threonine protein phosphatase PP2A dephosphorylates large fraction of cellular serine/threonine phosphorylated proteins and acts as a major tumor suppressor (6, 7). PP2A is a trimeric protein complex in which substrate specificity of the complex is determined by up to 15 different regulatory (B) subunits, which together with the scaffolding (A) and catalytic (C, PP2Ac) subunits generate a large number of alternative PP2A trimers (7, 8). PP2A inhibition by overexpression of endogenous inhibitory proteins, or by genetic mutations, results in hyperphosphorylation and activation of various cancer drivers. Consequently, PP2A inhibition contributes to the malignant transformation of human cells and promotes many malignant phenotypes (6, 9).
The endogenous PP2A inhibitor protein, protein phosphatase methylesterase 1 (PME-1; ref. 10), regulates PP2Ac activity by demethylating the highly conserved carboxyl-terminal tail and by directly binding to the PP2Ac catalytic site (7, 11). Notably, PME-1 selectively regulates activity of only certain B-subunit containing PP2A complexes (7). Consequently, as PME-1 does not regulate ubiquitous PP2A activity, only few PP2A phosphoprotein targets have thus far been identified to be affected by PME-1 in cancer cells. Clinically, PME-1 is expressed in about 50% of human astrocytic gliomas, and its expression levels correlate with the tumor cell proliferation and glioma disease progression (12).
Mammalian histone deacetylases (HDAC) can be classified into 4 classes (I–IV). The class IIa HDACs (HDAC 4, 5, 7, and 9) shuttle between the nucleus and cytoplasm and have potential deacetylation targets in both compartments (13). Phosphorylation of HDAC4 facilitates its nuclear export and conversely dephosphorylation by PP2A promotes HDAC4 nuclear import (14, 15). More specifically, PP2A trimers including PPP2R2A (B55α) and PPP2R5A (B56α) interact with HDAC4 (14, 15). Very interestingly, it is the cytoplasmic HDAC4 activity that is important for cell survival particularly in cell types of neuronal origin (16, 17).
Here, we identify PME-1 as a novel determinant of glioma cell kinase inhibitor resistance. Mechanistically, PME-1 effects are mediated by reactivation of specific PP2A complexes and by consequent decrease in the cytoplasmic HDAC4. The relevance of the findings is validated across several human glioma cell lines and in glioma stem cells, (GSC) as well as in in vivo xenograft model and human glioma tissue samples.
Materials and Methods
Cell culture and transfections
Established human glioblastoma cell lines T98G (VTT Technical Research Centre, Turku, Finland in 2010) and U251MG (from Pirjo Laakkonen, University of Helsinki, Helsinki, Finland in 2011) were cultured in Eagle MEM (Sigma-Aldrich) and U87MG-luciferase (gift from Ari Hinkkanen, University of Eastern Finland, Joensuu, Finland in 2011; ref. 18) in DMEM (Sigma-Aldrich). All growth mediums were supplemented with 10% heat-inactivated FBS (Gibco), 2 mmol/L l-glutamine, and penicillin (50 units/mL)/streptomycin (50 μg/mL; pen/strep) in a humidified atmosphere of 5% CO2 at 37°C. T98G and U87MG-luciferase cells lines were authenticated by 16 loci short tandem repeat (STR) profiling (LGC Standards) in 2013. Normal rat astrocytes isolated from the hippocampus region of new born (P0–P1) rat pup brains were kindly provided by Dr. Eleanor Coffey (Turku Centre for Biotechnology, Turku, Finland). Astrocytes were cultured on BD Matrigel–coated dishes in DMEM supplemented with 10% FBS, l-glutamine, and pen/strep.
E2, G7, R10, R15, and R24 primary glioblastoma cell lines were derived from freshly resected glioblastoma multiforme specimens as previously described (19) and generously provided by Dr. Colin Watts (University of Cambridge, Cambridge, United Kingdom). Tissue collection protocols were compliant with the UK Human Tissue Act 2004 (HTA License 12315) and approved by the local regional Ethics Committee (LREC 04/Q0108/60). Informed consent was obtained from each patient before surgery. Their culture conditions have been described (19) and are included in Supplementary Materials and Methods.
siRNA transfections were performed with Lipofectamine RNAiMAX (Invitrogen, Life technologies). siRNA sequences are in the Supplementary Table S1. HDAC4-GFP plasmid (gift from Prof. Stuart L. Schreiber, Howard Hughes Medical Institute, Chevy Chase, MD; ref. 20) was transfected using Fugene HD (Roche) 1 day after the siRNA transfection.
Chemical inhibitors and drugs
Following chemicals were purchased from indicated distributors: H7, H8, H89, chelerythrine chloride (Chl Cl), sunitinib, tandutinib, lapatinib, and vandetanib from Biaffin GmbH & Co KG; UO126, LY294002, RO-31-8220, GÖ 6976, and SB218078 from Calbiochem; staurosporine, CEP-701, UCN-01, PKC412, Z-VAD-FMK, okadaic acid, sodium selenite, xylulose-5-phosphate (X5P), and SB431542 from Sigma-Aldrich; temozolomide, arcyriaflavin-A, and K252c from Tocris Bioscience; K252a and rebeccamycin from Enzo Life Sciences; enzastaurin (LC laboratories); FTY720 (Cayman chemicals); panobinostat and FK228 (MedChem Express); SAHA and ABT263 (Selleckchem). DHPCC-9 (21) was a kind gift from Dr. Päivi Koskinen (University of Turku, Turku, Finland). ABL127 (22) kindly provided by Prof. Benjamin Cravatt (The Scripps Research Institute, La Jolla, CA).
Apoptosis assay (nuclear fragmentation assay)
The percentage of the sub-G0/G1 fraction of fragmented nuclei stained with propidium iodide (PI) was taken as a measure of apoptotic cells. Approximately 30,000 cells were seeded on 24-well plates, transfected with siRNA for 48 hours, and then treated with test compounds in fresh media. After 24 hours of treatment, both floating and adherent cells were harvested by centrifugation. Cell pellets were resuspended in 400 μL of hypotonic PI buffer [40 mmol/L Tri-sodium citrate (Merck), 0.3% Triton X-100 (Sigma-Aldrich), and 50 μg/mL propidium iodide (Sigma-Aldrich) in PBS] and incubated at room temperature for 10 minutes in dark. The flow cytometric analysis was performed using FACSCalibur (Becton Dickinson), and the recorded data were analyzed with CellQuest Pro software.
ABT263 treatment was performed in combination with staurosporine for 24 hours, starting from one day after seeding cells on 24-well plates. Treatments with PP2A activators FTY720, sodium selenite, and xylulose-5-phosphate were performed throughout the experiment starting from one day after seeding cells. Pretreatment with PP2A inhibitor okadaic acid was performed for 24 hours before staurosporine treatment. PME-1 inhibitor ABL127 treatment was performed for 24 hours prior to staurosporine treatment. Z-VAD-FMK treatment was performed one day after siRNA transfection for 24 hours followed by combined treatment with staurosporine and fresh Z-VAD-FMK inhibitor treatment for another 24 hours before analysis. HDAC inhibitor treatments were performed for 24 hours alone, followed by combined HDAC and staurosporine or UCN-01 treatment for another 24 hours before staining and fluorescence-activated cell sorting (FACS) analysis.
Colony formation assay
Optimized numbers of cells (2 × 103 to 8 × 103) were seeded in 12-well plates until formation of colonies. siRNA transfections were performed using Lipofectamine RNAiMAX Reagent (Invitrogen), and after 48 hours, colonies were treated with indicated concentration of chemical drugs for another 48 hours. Cell colonies were fixed with 3.7% formaldehyde and stained with 0.2% crystal violet solution (made in 10% ethanol) for 15 minutes at room temperature each. Excess stain was removed by repeated washing with PBS. Plates were dried and scanned with Epson perfection V700 scanner. Quantifications were performed with ColonyArea (23) ImageJ (24) plugin and graphs were plotted using the area % values.
Statistical analysis
The significance level of differences between the mean values of 2 groups of data was evaluated using the unpaired Student t test assuming unequal variances among the sample means. All P values were 2-tailed. Parameter differences with probability value P < 0.05 were depicted as statistically significant (*) and P < 0.001 as highly significant (**).
Correlation between sensitivity of PME-1- or HDAC4-depleted T98G cells toward staurosporine derivatives was analyzed by Pearson correlation method, where relationships are determined by the correlation coefficient. R +0.4 to +0.69 indicate strong positive correlation.
Results
PME-1 depletion induces glioma cell synthetic lethality in conjunction with kinase inhibitors
Given the capacity of PP2A to negatively regulate phosphoproteins, we explored whether reactivation of PP2A activity in PME-1–depleted cells would further predispose glioma cells to apoptosis induction by kinase inhibitors. We employed a series of kinase inhibitors targeting different classes of kinases, mainly grouped under AGC/CAMK (protein kinases A, G, C and calcium/calmodulin-regulated kinases) and tyrosine kinase (TK) inhibitor categories, and one broad-spectrum kinase inhibitor, staurosporine. All tested inhibitors showed efficient downregulation of known phosphorylated target proteins (Supplementary Fig. S1A). Consistent with the inherent kinase inhibitor resistance of glioblastoma multiforme (3–5, 25), no apparent induction of apoptosis was observed by treatment with any of the tested inhibitors in the control siRNA–transfected cells (Fig. 1A, grey bars). Also, PME-1 depletion alone did not induce strong apoptotic phenotype (Fig. 1A, black bar in control). In contrast, PME-1–silenced glioblastoma multiforme cells displayed enhanced sensitivity to several of kinase inhibitors, most notably observed with H7, sunitinib, and LY294002 (Fig. 1A, black bars). However, the strongest synthetically lethal activity was observed in PME-1–deficient cells treated with multikinase inhibitor staurosporine (Fig. 1A). The staurosporine-induced synthetic lethality in PME-1–silenced cells was dose-dependent and not due to siRNA off-target effects (Supplementary Fig. S1B–S1F). Furthermore, the combination of PME-1 inhibition and staurosporine resulted in very efficient inhibition of clonogenicity of both T98G and U251MG glioblastoma cells (Fig. 1B and C). Whereas the synthetic lethality was not observed in normal astrocytes isolated from the rat hippocampus (Fig. 1D and Supplementary Fig. S1G) or in other types of cancer cell lines than in gliomas (Supplementary Fig. S1H).
Given the potency of staurosporine and its selectivity for gliomas, we focused on staurosporine as a model compound to study the role of PME-1 in mediating kinase inhibitor resistance in glioma cells. Consistent with the nuclear fragmentation assay, we found significantly increased caspase-3/7 activity upon combination treatment (Fig. 1E). Furthermore, pan-caspase inhibitor Z-VAD-FMK, or co-depletion of proapoptotic BH3 family protein BAD, completely blocked the apoptosis induction (Fig. 1F and G and Supplementary Fig. S1I). However, BAD activation by a small-molecule Bcl-2/-xL inhibitor, ABT263, was not sufficient to synergistically induce apoptosis in combination with staurosporine treatment (Fig. 1H). These results indicate that PME-1 inhibits apoptotic response even in glioma cells with activated BAD.
PME-1 drives resistance to clinically relevant indolocarbazole derivatives in glioblastoma multiforme cell lines and in patient-derived glioma stem cells
Many indolocarbazole analogues of staurosporine have more specific kinase inhibition profiles than staurosporine and are well-tolerated in patients (Supplementary Table S2; refs. 26, 27). Especially CEP-701 (lestaurtinib) and PKC412 (midostaurin) have progressed to phase III clinical trials in hematologic cancers (Supplementary Table S2). Thus, we screened a collection of staurosporine analogues for synthetic lethality in PME-1–depleted glioblastoma cells. Screening identified PKC412, CEP-701, K252a, and UCN-01 as highly potent synthetically lethal compounds in PME-1–depleted T98G cells (Fig. 2A). These results were validated by colony growth assays using T98G, U251MG, and U87MG glioma cell lines (Fig. 2B and C, Supplementary Fig. S2A and S2B). Moreover, identification of several effective staurosporine derivatives enabled us to perform a structure–activity relationship (SAR) analysis to uncover the structural determinants of indolocarbazole compounds relevant for synthetic lethality in PME-1–depleted cells (Supplementary Table S3).
Importantly, PME-1–driven resistance to active indolocarbazoles is not restricted to in vitro cell culture but occurs also in in vivo environment. PME-1 silenced U87MG xenograft tumors showed markedly enhanced responses to UCN-01 as compared with the scrambled SiRNA–transfected tumors (Fig. 2D and E). In parallel cell cultures, PME-1 siRNA was found to silence PME-1 expression for up to 10 days posttransfection (Supplementary Fig. S2C). Preferential antitumorigenic activity of PME-1 siRNA and UCN-01 combinatorial therapy was further validated by intratumor delivery of PME-1 siRNA within established U87MG-luc xenograft tumors in mice subsequently treated with UCN-01 (Supplementary Fig. S2D and S2E).
Cancer stem cells (GSCs) are important determinants of therapy resistance in glioblastoma (1, 28). Therefore, we evaluated whether the results of this study would also be relevant to GSCs (19). The stem cell identity of GSC population was demonstrated by expression of several stem cell markers such as CD133, SOX2, Olig2, and CD15 (Supplementary Fig. S2F). Moreover, we have previously shown that these GSCs display characteristic infiltrative growth pattern in intracranial in vivo model (29, 30). Indeed, a significant inhibition of colony growth was observed in PME-1–depleted GSCs (E2 and R10 cell lines) treated with either staurosporine or UCN-01 (Fig. 2F and G and Supplementary Fig. S2G and S2H).
These results demonstrate that PME-1 promotes resistance to multiple clinically applicable indolocarbazole kinase inhibitors in several glioma cell lines and in the therapy-resistant GSCs.
Reactivation of specific PP2A complexes is required for PME-1 depletion–mediated synthetic lethality
As expected, we found elevated phosphatase activity in PP2A immunoprecipitates from PME-1–silenced cells, as compared with those from control cells (Fig. 3A). The inhibition of PP2A catalytic activity by semiselective chemical inhibitor okadaic acid in turn rescued the cells from staurosporine-induced synthetic lethality in a dose-dependent manner (Fig. 3B). However, the chemical inhibition of PME-1 methylesterase activity by ABL127 (22) failed to show any synthetic lethality with staurosporine in glioblastoma multiforme cells (Supplementary Fig. S3A). On the basis of these results, it is likely that PME-1–mediated direct inhibition of PP2A catalytic activity (11), rather than its methylesterase activity (10), is relevant for resistance to apoptosis by kinase inhibitors.
Although PP2A has been previously considered as a nonspecific housekeeping phosphatase, recent evidence suggests that PP2A complexes differentially regulated by endogenous inhibitor mechanisms selectively dephosphorylate specific targets (Fig. 3C; refs. 7, 8). Thereby, we wanted to study whether the observed synthetic lethality could be achieved by other means of PP2A reactivation. Notably, neither depletion of another PP2A inhibitor protein CIP2A (31), nor chemical compounds previously used to reactivate PP2A in cancer cells (32), induced significant synthetic lethality (Fig. 3D and E and Supplementary Fig. S3B).
Next, we wanted to uncover the involvement of specific B-subunit containing PP2A trimers in PME-1–regulated kinase inhibitor resistance. To this end, we employed RNAi-mediated codepletion of PME-1 with PP2A B-subunits from PPP2R2 and PPP2R5 families, as they have previously been implicated in PME-1–mediated processes (7). Interestingly, depletion of only some members of these B-subunit families, namely, PPP2R2A, PPP2R5A, and PPP2R5B significantly reduced synthetic lethality (Fig. 3F and Supplementary Fig. S3C). Together, these results indicate that the observed synthetic lethality is mediated by specific PME-1–regulated PP2A complexes. Notably, on the basis of cBioPortal The Cancer Genome Atlas (TCGA) database (33, 34), patients with glioblastoma multiforme do not show genetic alterations in any of the PP2A subunits (Fig. 3G). As PP2A seems to be genetically intact in human gliomas, this indicates that PME-1 overexpression could be a disease-relevant mechanism of PP2A inhibition in this disease. Furthermore, inhibition of overexpressed PME-1 in gliomas (12) might result in therapeutic reactivation of genetically intact PP2A phosphatase complexes.
HDAC4 mediates PME-1–driven kinase inhibitor resistance
To discover potential downstream targets of PP2A responsible for PME-1–regulated kinase inhibitor resistance, the existing literature was studied to create a list of proteins known to interact with B-subunits PPP2R2A, PPP2R5A, and PPP2R5B (Supplementary Table S4; ref. 35). We inhibited some of these proteins with siRNA or with specific chemical inhibitors and subsequently treated the cells with the model compound staurosporine (Fig. 4A and B). Interestingly, depletion of MYC and HDAC4 selectively phenocopied the PME-1 depletion–induced synthetic lethality (Fig. 4A and B). Given that HDAC4 was downregulated in cells transfected with MYC siRNA, but HDAC4 siRNA did not affect MYC expression (Fig. 4C), we postulate HDAC4 is the potential PME-1 target and that the synthetic lethality seen in the MYC silenced cells was also due to HDAC4 inhibition.
Similar to PME-1, also HDAC4 depletion reduced clonogenicity of staurosporine-treated glioma cells (Fig. 4D and E). Importantly, HDAC4- and PME-1–depleted cells showed correlative sensitivity profiles to different indolocarbazole analogues, both in nuclear fragmentation (Fig. 4F and Supplementary Fig. S4A) and in clonogenicity assays (Fig. 4G). HDAC4 depletion also significantly sensitized GSC cultures to staurosporine- and UCN-01–induced inhibition of colony growth (Fig. 4H and I). These results were confirmed in another patient-derived GSC cell line BT3 and its CD133-enriched subpopulation (Supplementary Fig. S4B–S4D). In addition, codepletion of PME-1 and HDAC4 induced apoptosis as efficiently as depletion of either of them alone (Fig. 4J and Supplementary Fig. S4E). Furthermore, similar to PME-1 (Fig. 1G), simultaneous depletion of BAD and HDAC4 rescued the cells from synthetic lethality (Fig. 4K and Supplementary Fig. S4F). Thus, identical synthetic lethality response profile in PME-1- and HDAC4-depleted cells (Fig. 4F and K), and absence of any additive or synergistic effects upon depletion of PME-1 and HDAC4 (Fig. 4J), strongly indicates that HDAC4 mediates PME-1–driven kinase inhibitor resistance in glioblastoma multiforme.
Importantly, codepletion of PPP2R2A B-subunit, implicated in PME-1 depletion–induced synthetic lethality (Fig. 3F), did not block synthetic lethality in HDAC4-depleted cells (Fig. 4L). This further indicates that PME-1 and PP2A function upstream of HDAC4. Furthermore, proximity ligation assays (PLA) using anti-HDAC4 together with either anti-PP2A Cα subunit or anti-PPP2R2A antibodies revealed a physical association between HDAC4 and this specific B-subunit containing PP2A complex (Fig. 4M).
Loss of cytoplasmic HDAC4 predisposes to kinase inhibitor–induced synthetic lethality
In unperturbed cells, majority of HDAC4 is localized to the cytoplasm, and particularly in certain neuronal cells, the loss of cytosolic HDAC4 has been associated with induction of cell death (16, 17, 36). Supportive of survival role of cytoplasmic HDAC4 also in glioma cells, we observed predominantly cytoplasmic localization of endogenous HDAC4 in the T98G cells (Fig. 5A), as well as in glioma patient tumor tissue (Fig. 5B). Notably, as shown in Fig. 4M, also the HDAC4–PPP2R2A association was predominantly localized in the cytoplasm. Consistent with prosurvival activity of cytosolic HDAC4 (16, 17, 36), PME-1 depletion induced translocation of HDAC4-GFP (20) from cytoplasm to the nucleus (Fig. 5C). Moreover, loss of cytosolic HDAC4 activity upon PME-1 depletion was supported by increased acetylation of several cytosolic proteins, whose acetylation was induced also in HDAC4-depleted cells (Fig. 5D).
On the basis of previous studies, PME-1 depletion could promote HDAC4 nuclear translocation by dephosphorylation of Ser 298 (14). Unfortunately, because of lack of phospho-Ser 298 HDAC4 antibody, we could not confirm this assumption. Furthermore, we did not observe reduction in the phosphorylation of cytosolic 14-3-3 binding residues, Ser 246 or Ser 632 of HDAC4 by PME-1 depletion (Fig. 5E and Supplementary Fig. S5A). However, in line with previous demonstration of nuclear translocation of HDAC4 upon staurosporine-mediated inhibition of HDAC4 phosphorylating kinases (37, 38), HDAC4 phosphorylation at these residues was reduced upon 6-hour treatment with staurosporine (Fig. 5E and Supplementary Fig. S5A). Importantly, total HDAC4 expression was downregulated by combination of PME-1 depletion and staurosporine treatment (Fig. 5E). Therefore, the combined effects of HDAC4 nuclear translocation by PME-1 depletion, and staurosporine-elicited inhibition of Ser 632 and Ser 246 phosphorylation synergistically inhibit cytoplasmic HDAC4 expression and thus trigger synthetic lethality under staurosporine exposure (Fig. 4B–I).
To address whether existing small-molecule HDAC inhibitors could be used to induce synthetic lethality together with staurosporine and its derivatives in glioma cells, we compared effects of 3 inhibitors, out of which both FK228 (romidepsin) and panobinostat inhibit HDAC4 activity in the nanomolar range, whereas SAHA does not (Fig. 5F; refs. 39, 40). Indeed, inhibition of HDAC activity by FK228 or panobinostat showed similar synthetic lethality in combination with staurosporine or UCN-01 treatment as with PME-1 or HDAC4 silencing (Fig. 5G and H and Supplementary Fig. S5B). However, no evidence for synthetic lethal activity was observed in cells cotreated with SAHA (Fig. 5I), even though the tested concentrations of all 3 HDAC inhibitors were found to inhibit deacetylation of histone H3 (Supplementary Fig. S5C, yellow bars). Notably, cells treated with FK228 and panobinostat also showed reduced levels of HDAC4 protein (Supplementary Fig. S5C, green bars), correlating again HDAC4 expression status to sensitivity toward staurosporine-elicited synthetic lethal activity.
Therapeutic vulnerability in human glioblastoma based on PME-1/HDAC4/BAD status
Previously, by IHC analysis, it has been shown that 50% of primary glioblastoma multiforme tumors express elevated levels of PME-1 protein (12). Similarly, gene expression analysis using TCGA glioblastoma multiforme cohort suggests that about 50% of the patient tumors also transcribe PME-1 (https://genome-cancer.ucsc.edu/). We speculate that PME-1 protein might be stabilized by posttranslational mechanisms in this subpopulation of PME-1 mRNA–expressing glioblastoma multiforme tumors.
To examine clinical relevance of the PME-1–mediated HDAC4 regulation, we studied HDAC4 status by IHC from the same primary astrocytic glioma patient material as above (12). Importantly, we found a significant correlation between HDAC4 expression and cytoplasmic PME-1 staining (P = 0.049; Supplementary Fig. S6A), and similar to PME-1, HDAC4 expression was absent or low in nearly half (41%) of the astrocytic gliomas (Fig. 6A and B). A significant correlation (P = 0.000) between HDAC4 immunopositivity and the grade of astrocytic tumors was also observed (Fig. 6B).
On the basis of these data, approximately 50% of patients with glioma could display indolocarbazole-sensitive phenotype on the basis of their low PME-1 or HDAC4 expression. To mimic glioma cells with constitutively low PME-1 expression, we generated clones of T98G cells stably expressing either GFP (control) or PME-1 targeting shRNA. In line with the requirement for high PME-1 expression for supporting the growth of high-grade gliomas (12), the maximal tolerated inhibition of PME-1 expression was about 50% (Fig. 6C). Importantly, the isogenic cell lines with modestly reduced PME-1 expression (shRNA PME #1 and #2 clones) showed robust sensitivity to K252a and CEP-701 treatment in both the colony growth (Fig. 6D and E) and the cell viability assays (Supplementary Fig. S6B).
Preferential kinase inhibitor sensitivity of PME-1 and HDAC4 low glioma cells is dependent on the BAD expression (Figs. 1G and 4K). Thereby, we further assessed BAD immunopositivity from the same patient material. The BAD IHC analysis revealed that about 48% of gliomas exhibited moderate-to-high BAD expression (Fig. 6A, F and G). Thereby, following the patient tumor PME-1low/BADhigh, or HDAC4low/BADhigh phenotypes, we were able to identify about 17% and 11% patients with glioma, respectively, who constitute the potential responder subgroup to the indolocarbazole kinase inhibitor monotherapies (Fig. 6F–H). Alternatively, 37% of patients with glioma with HDAC4high/BADhigh phenotype could potentially benefit from combination therapy with small-molecule HDAC4 inhibitors and selected kinase inhibitors (Fig. 6G and H).
Discussion
Glioma cells exhibit intrinsic resistance to most therapies, which is reflected in dismal clinical response rates (4, 5, 25). In particular, failure to treat patients with glioblastoma multiforme with various kinase inhibitors that show clinical activity in other cancer types has been confusing; especially in the light of genetic information that glioblastoma multiforme particularly is a disease driven by receptor tyrosine kinase signaling (1, 2, 34). Here we report PME-1–mediated inhibition of PP2A as an unprecedented explanation of glioblastoma multiforme cell resistance to at least reasonable spectrum of different kinase inhibitors.
Inhibition of PP2A tumor suppressor activity is required for malignant transformation of various types of normal cells and promotes malignancy in many types of human cancers (6, 9). However, the functional relevance of PP2A as a tumor suppressor in glioblastoma multiforme has not been extensively studied. Interestingly, on the basis of TCGA analysis, glioblastoma multiforme harbors very few genetic alterations that would directly affect PP2A complexes (Fig. 3G). Thus, our previous (12) and current identification of PME-1 as a clinically relevant human glioma oncoprotein provide thus far the best understood mechanism how PP2A is inhibited in human gliomas and in glioblastoma multiforme. PME-1 is known to affect the activity of only certain PP2A complexes (7). Consistently, we found remarkable selectivity among PP2A B-subunit containing complexes that mediate PME-1–driven kinase inhibitor resistance (Fig. 3F and Supplementary Fig. S3C).
HDAC4 associates with PP2A B-subunits identified here to mediate PME-1 depletion–induced synthetic lethality (Fig. 4M; refs. 17, 41). On the other hand, loss of cytosolic HDAC4 has been shown to induce apoptosis in neuronal models (16, 17, 36). We extend these findings by showing for the first time that in human glioma patient samples, HDAC4 is predominantly cytosolic and that acetylation of yet-to-be-identified cytosolic proteins is regulated by PME-1 in glioma cells. Role for HDAC4 as a critical target for PME-1–driven kinase inhibitor resistance was further validated by significant correlation of PME-1 and HDAC4 function toward a library of staurosporine derivatives (Fig. 4F and G), as well as by induction of synthetic lethality by small-molecule inhibitors of HDAC4 activity in combination with UCN-01 and staurosporine. Naturally, on the basis of our data, we cannot exclude potential contribution of other PP2A target proteins in mediating at least partly the PME-1–driven kinase inhibitor resistance phenotype. Clearly highlighting the potential clinical applicability of our results for future glioblastoma multiforme therapy, panobinostat and FK228 have been approved for therapy of multiple myeloma and cutaneous T-cell lymphoma, respectively (42, 43), and with panobinostat clinical trials have been conducted in adult gliomas and are underway in pediatric gliomas (44, 45)
BAD was shown to be essential for synthetic lethality induced by either PME-1 or HDAC4 depletion. BAD acts as a node between upstream survival signaling pathways and downstream apoptosis signaling. However, these results do not rule of the possibility of other BH3-only proapoptotic proteins to be involved in this response. Importantly, results with BAD depletion allowed us to further scrutinize potential clinical applicability of our results for diagnostic stratification of patients with glioblastoma multiforme. Our results indicate that PME-1low/BADhigh or HDAC4low/BADhigh phenotype in diagnostic glioblastoma multiforme biopsy might serve as a biomarker for stratification of patients to treatment with kinase inhibitors such as CEP-701 (lestaurtinib) and PKC412 (midostaurin) that are currently being tested in phase III clinical trials (www.clinicaltrials.gov). The patients with PME-1high/BADhigh or HDAC4high/BADhigh tumor phenotype on the other hand could benefit from the combined PP2A reactivator compound or HDAC4 inhibitor and kinase inhibitor therapies. The principles of potential future clinical trials based on the proposed scheme are depicted in Fig. 6H.
In summary, results of this study reveal an unprecedented mechanism by which glioblastoma multiforme cells can resist cell-killing activity of kinase inhibitors targeting various survival kinase pathways. The work also identifies both PME-1 and HDAC4 as context-dependent survival factors in glioblastoma multiforme and as potential targets for future combination therapies.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Authors' Contributions
Conception and design: A. Kaur, S.U. Ahmed, H.K. Haapasalo, J. Eriksson, J. Westermarck
Development of methodology: A. Kaur, X. Qiao, E. Peuhu, S.U. Ahmed, O. Raheem, J. Eriksson, P.M. Laakkonen
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Kaur, O. Denisova, X. Qiao, E. Peuhu, S.U. Ahmed, H.K. Haapasalo, A.J. Chalmers, P.M. Laakkonen
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Kaur, M. Jumppanen, E. Peuhu, O. Raheem, H.K. Haapasalo, A.J. Chalmers, J. Westermarck
Writing, review, and/or revision of the manuscript: A. Kaur, O. Denisova, M. Jumppanen, O. Raheem, H.K. Haapasalo, A.J. Chalmers, P.M. Laakkonen, J. Westermarck
Study supervision: J. Eriksson, P.M. Laakkonen, J. Westermarck
Acknowledgments
Authors thank Prof. Ari Hinkkanen, Prof. Benjamin Cravatt, Prof. Stuart L. Schreiber, Dr. Eleanor Coffey, Dr. Päivi Koskinen, Dr. Panu Jaakkola, and Dr. Colin Watts for generous support of research materials. Taina-Kalevo Mattila and Inga Pukonen are acknowledged for their skillful assistance and help. Authors are very grateful to Prof. David Virshup and Prof. Johanna Ivaska for critical comments on the article. Turku Centre for Disease Modelling (TCDM) is acknowledged for expert help with mouse experiments.
Grant Support
J. Westermarck was supported by grants from Sigrid Juselius Foundation, Finnish Cancer Association, Academy of Finland (grant 252572), Jane and Aatos Erkko Foundation, and holds a Research Professor position supported by Foundation of the Finnish Cancer Institute. A. Kaur was supported by Turku Doctoral Programme of Biomedical Sciences (TuBS), Turku Doctoral Programme of Molecular Medicine (TuDMM), Orion Research Foundation, Cancer Society of Finland, Instrumentarium Science Foundation and K. Albin Johanssons Foundation. A.J. Chalmers received a grant from UK Medical Research Council.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.