Glioblastomas require integrin αvβ3/PAK4 signaling to escape senescence

Abstract

Integrin αvβ3 has been implicated as a driver of aggressive and metastatic disease, and is upregulated during glioblastoma progression. Here we demonstrate that integrin αvβ3 allows glioblastoma cells to counteract senescence through a novel tissue-specific effector mechanism involving recruitment and activation of the cytoskeletal regulatory kinase PAK4. Mechanistically, targeting either αvβ3 or PAK4 led to emergence of a p21-dependent, p53-independent cell senescence phenotype. Notably, glioblastoma cells did not exhibit a similar requirement for either other integrins or additional PAK family members. Moreover, αvβ3/PAK4 dependence was not found to be critical in epithelial cancers. Taken together, our findings established that glioblastomas are selectively addicted to this pathway as a strategy to evade oncogene-induced senescence, with implications that inhibiting the αvβ3/PAK4 signaling axis may offer novel therapeutic opportunities to target this aggressive cancer.

Introduction

Glioblastoma multiforme, or GBM, is the most aggressive and malignant form of astrocytoma characterized by highly invasive tumor cells. Although these tumors are treated using a combination of surgery, radiotherapy, and chemotherapy, only 5% of patients survive for longer than 5 years after diagnosis. Large-scale efforts have recently provided new clues into gliomagenesis and alterations that characterize this disease (1). Despite identification of new biomarkers and important molecular pathways (2), targeted therapies have not yet elicited durable clinical responses (3).

The expression of integrin αvβ3 (ITGAV/ITGB3) and its ligand vitronectin increase during the transition from low-grade astroglial-derived tumors to advanced glioblastoma (4, 5), and we and others have identified αvβ3 as a driver of an aggressive and metastatic tumor phenotype (6, 7). In GBM biopsy samples, αvβ3 expression is prominent in both tumor microvessels and glial tumor cells, and is the most prevalent in highly proliferating and infiltrating areas (8). Cilengitide, designed to target the ligand binding properties of αvβ3 and other αv integrins (9), was tested in combination with temozolomide chemoradiotherapy in a randomized phase III trial for patients with newly diagnosed glioblastoma with methylated MGMT promoters (ClinicalTrials.gov NCT00689221). While some patients responded, Cilengitide failed to meet its primary endpoint of a significant survival advantage (10).

In addition to its ligand-dependent signaling role, recent studies suggest αvβ3 has non-canonical cell biological functions that are ligand-independent (6, 7, 11). Since αvβ3 expression correlates with glioblastoma progression, we silenced β3 in a variety of human glioblastoma cells and assessed their growth in vivo and in vitro to evaluate the net contribution of this integrin’s ligand-dependent and –independent functions to glioblastoma biology. To our surprise, glioblastoma cells demonstrated an addiction to αvβ3 as a means to avoid p21 (CDKN1A)-dependent cellular senescence, whereas β3 knockdown did not trigger this effect in a range of histologically distinct epithelial cancers. Loss of αvβ3 led to a concomitant decrease in PAK4 activation, while PAK4 knockdown increased p21 and senescence. These findings reveals a new cell type-specific function for integrin αvβ3, and highlights a particular vulnerability of glioblastoma cells for components of this pathway.

Materials and Methods

Cell lines

All cells were purchased from American Type Culture Collection (ATCC) within the past 5 years: glioblastoma (U87MG, LN229, LN18, U373, U118, U251), medulloblastoma (DAOY), renal (7860), colorectal (SW620), pancreatic (PANC1), breast (MDA-MB-231, BT20), and lung (A549, H23). Cell line authentication was performed by the ATCC using short tandem repeat DNA profiles. Upon receipt, each cell line was expanded, cryopreserved as low-passage stocks, and tested routinely for mycoplasma.

RNA interference and expression constructs

For transient knockdown, cells were transfected using the HiPerFect (Qiagen) with AllStars siRNA (Qiagen) for negative control (1027280), ITGB3 (SI00004585), ITGB5 (SI02780617), PAK4 (SI00082341), CDKN1A (SI00008547), or TP53 (SI00011655). For stable knockdown, cells were infected with shRNA targeting ITGB3 (Open Biosystems; TRCN0000003234) using a lentiviral system.

Immunoblotting

Lysates made in 4% SDS were quantified using the Pierce BCA kit (ThermoScientific) and 25–50 μg protein loaded onto a denaturing SDS-polyacrylamide gel, transferred to polyvinylidene difluoride membranes, blotted with HRP-conjugated secondary antibodies (Bio-Rad), and bands detected by enhanced chemiluminescence (Advansta). Antibodies include β3 (Abcam); β-Actin (Sigma); FAK-pY861 (Invitrogen); FAK and p130Cas (BD Trandsduction Laboratories); β5, p21 Waf1/Cip1, p27 Kip1 (D69C12), p53-pS392), p53 (7F5), Rb-pS795, Rb (D20), PAK4, p130Cas-pY410, PAK4-pS474/PAK5-pS602/PAK6-pS560, PAK1, PAK2, and PAK1-pS144/PAK2-pS141 (Cell Signaling).

Proliferation and cell cycle

The colorimetric BrdU Cell Proliferation Assay kit (Millipore) was used with absorbance at 450/550 nm relative to control. Cells were stained using propidium iodide and subjected to flow cytometry analysis for cell cycle.

Animals

Animal protocols were approved by the UCSD Institutional Animal Care and Use Committee. 6–8 week old female athymic nu/nu mice were purchased from the UCSD Animal Care Program.

Flank tumor xenografts

Mice were injected subcutaneously with 10⁶ tumor cells in 200 μl PBS. Tumor size was measured weekly with calipers.

Orthotopic brain tumor xenografts

Mice were anesthetized by intramuscular injection of ketamine, dexmedetomidine, and buprenorphine. Using a stereotaxic frame (Stoelting Co.), a small burr hole was made in the skull 2 mm anterior and 2 mm lateral to the bregma. A 31-gauge Hamilton needle/syringe was inserted 3 mm, and 0.25 μl/minute was dispensed (10⁵ tumor cells in 2 μl media). Animals were monitored daily and those exhibiting signs of morbidity were euthanized.

Tumor spheroids

Multicellular spheroids were prepared by seeding 10⁵ cells per 24-well pre-coated with heated 1% Seaplaque agarose (Lonza) in serum-free medium. After 7–10 days, spheroids were collected, fixed in 4% paraformaldehyde, and embedded in paraffin.

SA-β-galactosidase staining

Cell senescence was measured by the Senescence β-Galactosidase Staining Kit (Cell Signaling).

Immunofluorescence microscopy

Cells on glass coverslips were fixed with 4% paraformaldehyde and processed for immunofluorescence as previously described (6) for imaging on a Nikon Eclipse C1 confocal microscope with 1.4 NA 60× oil-immersion lens. Antibodies include integrin αvβ3 (LM609), p21, phospho-PAK4, pan-methyl-histone H3 (Lys9) (D54) (Cell Signaling).

Statistics

One way ANOVA or t-tests with P<0.05 were performed using Prism software (GraphPad).

Results

Glioblastomas depend on integrin αvβ3 for growth in vitro and in vivo

Since we and others have established integrin αvβ3 as a mediator of glioblastoma progression (4, 9, 12), we assessed the effect of β3 knockdown on growth of U87MG human glioblastoma cells. U87MG cells with siRNA-mediated knockdown of β3 (si-β3) produced smaller 3-D spheroids with less cell proliferation, a lower mitotic index, but no difference in apoptosis compared with control (si-CTRL) or β5 (ITGB5) (si-β5) siRNA (Fig. 1A). When implanted as intracranial orthotopic xenografts in mice, U87MG cells with stable shRNA-mediated knockdown of β3 failed to form detectable tumors by 75 days, whereas knockdown of β5 accelerated tumor growth and decreased survival time (Fig. 1B). This β3 dependence for in vivo tumor growth was also observed for LN229 glioblastoma cells, but not for PANC1 pancreatic or A549 lung carcinoma cells expressing similar levels of β3 (Fig. 1C, Supplemental Fig. 1). Furthermore, β3 knockdown suppressed proliferation of multiple glioblastoma cell lines and the DAOY medulloblastoma cell line, but not β3-positive carcinomas from pancreas, breast, and lung (Fig. 1D). Together, these findings suggest that glioblastomas and perhaps other brain tumors are unique in their requirement for integrin αvβ3.

Figure 1. Knockdown of integrin αvβ3 impairs glioblastoma growth in vitro and in vivo.

A) U87MG human glioblastoma cells transiently expressing siRNA against integrin β3, β5, or vector control were grown in vitro as 3D spheroids for 10 days. Spheroid area for phase-contrast images was measured in pixels (n=6, *P<0.0001 vs. sh-CTRL). Cell proliferation, mitotic index, and apoptosis were assessed using immunostaining for Ki-67, p-Histone H3, and cleaved Caspase-3 respectively (n=9, *P<0.001 vs. sh-CTRL). B) RFP-expressing U87MG cells were implanted intracranially in immunocompromised mice and tumor growth monitored using IVIS200 imaging (n=9, *P<0.001 for bioluminescence; n=10, *P<0.005 for survival curve). C) Human tumor cells were grown as subcutaneous tumors in nude mice. Tumor volume was measured weekly with calipers (n=5–12 each group). D) Integrin β3 was transiently (siRNA) or stably (shRNA) knocked down in human cancer cell lines and their ability to proliferate in vitro was assessed by BrdU incorporation, and expressed as fold-difference relative to control si/shRNA (n=3+ each group, *P<0.01).

To understand why β3-knockdown suppressed glioblastoma growth, we examined their growth properties and morphology in vitro. Following β3-knockdown, the cells stopped proliferating well before reaching confluence, yet they remained attached to the plate and adopted a large flat appearance consistent with cells undergoing senescence. Accordingly, silencing β3 but not β5 resulted in G0/G1 arrest with no increase in the sub-G1 population (Fig. 2A and Supplemental Fig. 2), suggesting cell cycle arrest without apoptosis. β3 knockdown also led to enhanced methylation of histone H3 in senescence-associated heterochromatin foci (Fig. 2B), increased γ-H2AX positive nuclei indicative of DNA double strand breaks (Fig. 2C), and accumulation of senescence-associated acidic β-galactosidase (SA-β-gal) activity (Fig. 2D). Interestingly, no changes in SA-β-gal were observed following knockdown of β5 integrin in glioblastomas, nor was senescence induced following β3 knockdown for multiple breast, colorectal, lung, and pancreatic carcinomas (Fig. 2D). Together, these findings suggest that β3, but not β5, may function as a suppressor of senescence in glioblastoma cells. Indeed, increased SA-β-gal was observed for sh-β3 U87MG subcutaneous tumors 7 days after implantation in mice (Fig. 2D), providing a snapshot of senescence that precedes the eventual exponential growth of control tumors (Fig. 1C).

Figure 2. β3 knockdown induces senescence in glioblastoma cells.

The effects of siRNA- or shRNA-mediated knockdown of β3 expression was compared in multiple glioblastoma and epithelial cancer cell lines in vitro. A) Cell cycle arrest in glioblastoma cell lines was analyzed using flow cytometry. B) The presence of senescence-associated heterochromatin foci was evaluated by immunofluorescence staining for pan-methylated Histone H3K9 (green). Cell nuclei were counterstained with Hoechst dye (blue). (U87: n=10 fields, *P<0.0001; LN18: n=5 fields, *P<0.05) C) DNA double-strand breaks were detected using immunofluorescence staining for γ-HA2X (green). Cell nuclei were counterstained with Hoechst dye (blue). D) Senescence-associated β-galactosidase staining was used to identify the frequency of senescent cells in vitro or in vivo. (n=5 fields counted per group in at least 2 independent experiments)

αvβ3 ligand-binding function appears relevant for its suppression of senescence, as a β3-D119A mutant unable to bind ligand could not rescue the β3-knockdown phenotype (Supplemental Fig. 3A). Furthermore, treatment with the function blocking αvβ3 antibody LM609 induced senescence in U87MG glioblastoma cells (Supplemental Fig. 3B). Together, these findings suggest that integrin αvβ3 may serve a specialized function in glioblastoma cells to support tumor growth by suppressing pathways governing cellular senescence.

αvβ3 allows glioblastoma cells to evade senescence by suppressing p21

Senescence is a process triggered by cellular stress, leading to a permanent withdrawal from the cell cycle. Exposing cells to DNA damage/attrition, oxidative stress, or oncogene activity drives anti-proliferative signaling through regulatory proteins such as p16, p21, p27, p53, and Rb that dictate whether a cell will senesce in response to the assault (13). Silencing of β3 induced a marked increase in p21 protein expression for multiple glioblastoma cells, whereas p21 levels were unchanged or even decreased following β3 knockdown in various epithelial cancer cells (Fig. 3A–B). Knockdown of p21 rescued glioblastoma cells from senescence in the absence of β3, pointing to its requirement for the observed phenotype (Fig. 3C). Notably, other mediators of senescence such as p27 (CDKN1B) and p53 (TP53) did not show consistent changes in response to loss of β3, and senescence was not hindered by p53 knockdown in U87MG or LN229 glioblastomas (Fig. 3A, 3C). Together, these results implicate integrin αvβ3 in the protection of glioblastomas from senescence via its impact on p21.

Figure 3. Integrin αvβ3 suppresses p21 expression in glioblastoma cells.

A) The effect of β3 knockdown on expression and phosphorylation of p21, p27, Rb, and p53 was evaluated using immunoblotting. B) Immunofluorescence staining for p21 (red) confirms accumulation in glioblastoma cells with β3 knockdown compared with sh-CTRL. C) siRNA-mediated knockdown of p21 but not p53 significantly reduces the percent of SA-β-galactosidase positive cells (blue) in glioblastoma cells with β3 knockdown. (n=5 fields counted per group in at least 2 independent experiments, *P<0.01)

PAK4 implicated as mediator of senescence downstream from αvβ3

Integrin clustering on the cell surface regulates formation of focal adhesions and assembly of signaling complexes that, in turn, activate various signaling cascades (14). Accordingly, phosphorylation of FAK (focal adhesion kinase, PTK2) and p130 CAS (Crk-associated substrate, BCAR1) were both increased in β3-knockdown cells (Fig. 4A), consistent with enrichment of focal adhesions in senescence (15). Knockdown of β3, but not β5, selectively reduced Serine 474 phosphorylation of PAK4 (Fig. 4A), a type 2 p21-activated kinase. Serine 474 in the PAK4 kinase domain can be auto-phosphorylated or phosphorylated by a number of different kinases to constitute the “active” form of PAK4 (16, 17). PAK4 appears to play a functional role in suppressing senescence in glioblastoma, since treating cells with the ATP-competitive PAK4 inhibitor PF-03758309 increased SA-β-gal staining in U87MG cells in 2D culture and subcutaneous xenografts in vivo (Fig. 4B). This activity was observed for additional glioblastoma and medulloblastoma cell lines, but not for epithelial carcinomas (Fig. 4B). Like knockdown of β3 itself, loss of PAK4 expression induced p21 expression and SA-β-gal staining selectively in glioblastoma and medulloblastoma cells (Fig. 4C). These findings suggest β3 activates PAK4 to suppress p21 in a cell type-specific manner.

Figure 4. αvβ3 expression is required for PAK4 phosphorylation and evasion of senescence in glioblastoma.

A) The effect of β3 versus β5 knockdown in U87MG cells on focal adhesion-associated proteins FAK, CAS, PAK, RAF, and Src was evaluated using immunoblotting. B) Pharmacological inhibition of PAK using PF-03758309 causes cell senescence in vitro (10–1000 nM) and in vivo (25 mg/kg/day) as measured by SA-β-Gal staining for glioblastomas but not epithelial carcinomas. C) Knockdown of PAK4 selectively induces p21 expression and SA-β-gal staining in glioblastoma cells. (n=5 fields per group, *P<0.001) D) Integrin αvβ3 (green) and phosphorylated PAK4 (red) co-localize in glioblastoma and medulloblastoma cells, but not epithelial carcinoma cells.

Co-localization between αvβ3 integrin and phosphorylated PAK4 in glioblastoma

Although PAK family kinases are well-established effectors of integrin signaling, there are likely unappreciated distinctions between specific integrin αβ heterodimers and PAK family kinase members that could enact specialized functions. We reasoned that integrin β3 may influence the subcellular localization and hence activation status of PAK4 in a cell type-specific manner providing an explanation for its role in suppressing senescence in glioblastoma but not epithelial carcinoma cells. Indeed, αvβ3 and p-PAK4 co-localized within focal adhesions in glioblastoma, but not epithelial cancer cells (Fig. 4D). β3 did not co-localize with phospho-PAK1 in these cells (Supplemental Fig. 4), suggesting a specialized interaction between β3 and PAK4 in glioblastoma cells. These findings may explain how αvβ3 functions specifically via PAK4 in GBM and perhaps other brain tumors to protect them from senescence.

Discussion

Whereas senescence disables proliferation of damaged cells during development and aging, it also occurs in cells that have acquired a single activated oncogene (18). Oncogene-induced senescence locks cells into a permanent state of arrest to prevent uncontrolled proliferation and propagation of mutations (19). During this process, tumors acquire characteristic changes in tumor suppressor genes that collectively halt cell cycle progression, including activation of p53 and induction of p21 or p16. Progression to malignancy occurs when these cells alter gene expression or acquire additional mutations to evade senescence. We propose acquisition of αvβ3 during progression to malignancy provides glioblastomas a means to escape from senescence. This concept is supported by the fact that increased expression of αvβ3 and its ligand vitronectin are associated with glioblastoma progression (5).

Our findings raise the question of how glioblastomas and medulloblastomas develop an addiction to β3/PAK4. αvβ3 is absent on quiescent blood vessels, but becomes highly expressed on invasive endothelial cells during angiogenic remodeling (14). While β3 has been linked to metastasis in epithelial cancers (6, 7), glioblastomas typically show high β3 expression once tumors have become locally invasive (4), and PAK4 expression tracks with glioma grade (20). It is therefore tempting to consider that a major function of increasing αvβ3 during glioblastoma progression may be to continuously suppress oncogene-induced senescence to permit tumor progression.

Our findings also point to a β3/PAK4 interaction as a glioblastoma-specific addiction pathway. PAK proteins, serine/threonine p21-activated kinases that are substrates of the small GTP-binding proteins Cdc42 and Rac, are important for cell migration and proliferation. In glioblastoma cells, αvβ3 associates with activated PAK4 but not other PAK family members, and this interaction functions to suppress p21-dependent senescence. While epithelial cancers can express both αvβ3 and PAK4, they do not co-localize, which may explain why β3 knockdown fails to influence senescence.

Increased expression of αvβ3 and its ligand vitronectin have been associated with glioblastoma progression (5). The ligand-binding function of αvβ3 appears to be at least partially responsible for its role in suppressing senescence, since the β3-D119A mutant unable to bind ligand did not rescue the β3-knockdown senescent phenotype (Suppl. Fig. 3), and treating U87MG cells with the αvβ3 function-blocking antibody LM609 induced senescence (Suppl. Fig. 3). Despite its similarity to β3, knockdown of β5 accelerated orthotopic growth of U87MG glioblastomas. This could have interesting implications for cancer therapy, as the integrin antagonist Cilengitide targets both αvβ3 and αvβ5. While Cilengitide was effective in some patients, it did not produce a statistically significant survival benefit in a phase III trial for glioblastoma. A selective inhibitor of αvβ3 that does not target αvβ5 may represent a more appropriate targeted therapy for glioblastoma.

We have unveiled a unique vulnerability of glioblastomas to undergo senescence upon interference with β3/PAK4 signaling. Knockdown of β3 or PAK4, or targeting PAK4 with a pharmacological inhibitor, induced p21-dependent senescence in glioblastomas and medulloblastomas, while producing no effect for a range of histologically distinct epithelial cancers. Future studies could delineate why β3/PAK4 interact only in glioblastomas to account for the observed addiction to these signaling molecules. It will also be important to validate the function of this pathway in more sophisticated models of glioblastoma and medulloblastoma, including patient-derived xenografts or genetically engineered mouse models of these cancers. A better understanding of the signaling pathway triggered by β3/PAK4 may provide novel therapeutic strategies to exploit the addiction to these genes for the treatment of glioblastoma in the clinic.