Abstract

Integrin inhibitors targeting αv series integrins are being tested for their therapeutic potential in patients with brain tumors, but pathologic studies have been limited by lack of antibodies suitable for immunohistochemistry (IHC) on formalin-fixed, paraffin-embedded specimens. We compared the expression of αv integrins by IHC in brain tumor and normal human brain samples with gene expression data in a public database using new rabbit monoclonal antibodies against αvβ3, αvβ5, αvβ6, and αvβ8 complexes using both manual and automated microscopy analyses. Glial tumors usually shared an αvβ3-positive/αvβ5-positive/αvβ8-positive/αvβ6-negative phenotype. In 94 WHO (World Health Organization) grade II astrocytomas, 85 anaplastic astrocytomas WHO grade III, and 324 glioblastomas from archival sources, expression of integrins generally increased with grade of malignancy. Integrins αvβ3 and αvβ5 were expressed in many glioma vessels; the intensity of vascular expression of αvβ3 increased with grade of malignancy, whereas αvβ8 was absent. Analysis of gene expression in an independent cohort showed a similar increase in integrin expression with tumor grade, particularly of ITGB3 and ITGB8; ITGB6 was not expressed, consistent with the IHC data. Parenchymal αvβ3 expression and ITGB3 gene overexpression in glioblastomas were associated with a poor prognosis, as revealed by survival analysis (Kaplan-Meier logrank, p = 0.016). Together, these data strengthen the rationale for anti-integrin treatment of glial tumors.

Astrocytic brain tumors often recur because of their rapid invasion and extensive migration into the surrounding tissue. The cell-cell and extracellular matrix (ECM)-cell interactions in these tumors that drive these processes are mediated in part by a group of integrin transmembrane receptors (1, 2). These proteins are obligate α/β heterodimers; each chain has a large extracellular domain and, with the exception of αvβ4, a short cytoplasmic domain. Integrin ligation initiates signals that activate cytoplasmic kinase cascades, thus regulating diverse biologic functions, including cell attachment, differentiation, migration, wound healing, growth, and survival; hence, they play fundamental roles in many neoplasms (2–4). To date, 24 integrin heterodimers have been identified, at least 10 of which can be expressed by 1 cell type (5, 6). Five in-tegrins contain the αv subunit (αvβ1, αvβ3, αvβ5, αvβ6, and αvβ8), suggesting a pivotal role for the αv subfamily (5, 7). Chemokines and growth factors can modulate cellular integrin expression. Expression of αv complexes is often altered in tumors, and the expression pattern varies with the tumor type (8–10). The αv integrins bind to Arg-Gly-Asp (RGD) sequences in members of the provisional ECM in tumors (11, 12).

Integrins αvβ3 and αvβ5 are expressed in endothelial cells and tumor parenchyma of several types of cancer, suggesting a role in tumor initiation and progression (10, 12). In-tegrin β3-knockout mice display enhanced tumor growth and a proangiogenic phenotype (6, 13), which has led to the concept that αvβ3 expression may mediate a balance between protumor and antitumor effects (14). Expression of αvβ3 has been reported in neoplastic tumor cells and in brain tumor vasculature (15, 16). The ECM protein osteopontin, which is often overexpressed in gliomas (17), is a ligand of αvβ3 (18). It is noteworthy that integrins are overexpressed in high-grade gli-omas (15, 19, 20) but seem to be absent in low-grade tumors (20, 21). Inhibitory monoclonal antibodies and low-molecular-weight synthetic inhibitors of αv integrins are in clinical trials (22, 23), and mixed αvβ3/αvβ5 inhibitors can inhibit the growth of orthotopic brain tumor models (24–26). The αv in-tegrin inhibitors may also reduce tumor vascular burden and enhance the effects of other therapeutic modalities such as radiotherapy (24, 27, 28).

Diffuse astrocytomas and glioblastomas (GBMs), the most common glial neoplasms, are most often located in the subcortical or deep white matter of the cerebral hemispheres (29). Most gliomas develop de novo as primary GBMs, but tumors initially diagnosed as diffuse astrocytomas tend to progress to a more malignant phenotype within 6 to 8 years, ending finally as genetically distinct secondary GBMs (30). Such tumors have many mutations in the isocitrate dehydrogenase 1 and 2 genes (31), and IDH1 mutations are associated with a better overall outcome in astrocytic neoplasms compared with wild-type tumors (32). Finally, methyl guanine methyltransferase (MGMT) promoter methylation status has a prognostic role and is an indicator of treatment response in GBM (33). In combination with radiochemotherapy, the αvβ3/αvβ5 inhibitor cilengitide is currently in a phase III trial (6) in GBM, the most malignant form of primary brain tumor.

Studies of integrin expression in human tumors have been limited by lack of antibodies capable of reacting with targets in formalin-fixed, paraffin-embedded (FFPE) tissue, the method routinely used in surgical pathology. Thus, data on expression of target integrins αvβ3 and αvβ5 in brain tumors is limited to a relatively few frozen tumor samples (15, 16, 20) and vital imaging studies (34). Moreover, longitudinal studies are a profound logistic challenge. One of us (Simon Goodman) therefore developed a set of recombinant rabbit monoclonal antibodies (RabMabs) that 1) recognize integrins in FFPE material, 2) bind intact extracellular domains of the targets, and 3) bind reproducibly using a standard automated immunohis-tochemistry (IHC) system (35). For this study, we used these RabMabs against integrins αvβ3, αvβ5, αvβ6, and αvβ8 to analyze their expression in a large longitudinal series of archival FFPE normal brain, astrocytomas, and GBMs. Our data indicate that expression of αvβ3 in tumor cells and vessels, but not the other αv integrins, is dependent on tumor grade, and that its expression in tumor cells may have a prognostic impact.

Materials and Methods

Antibodies

Matched RabMabs directed against intact extracellular domains of human αvβ3, αvβ5, αvβ6, αvβ8 complexes, the common αv subunit, and of the isolated β3-cytoplasmic domain (cytoβ3) were generated and characterized as described (35). Commercial antibodies to the integrins and ligands fibronectin and fibrinogen were also used (Table 1).

TABLE 1

Antibodies

Antibody to Clone, Species Dilution (concentration) Pretreatment, Primary Antibody Incubation Time (Duration) Source 
αvβ3 EM227-03, rabbit 1:500 (2 μg/mL) Protease 12 minutes (0.1 U/mL), 32 minutes (35
Cytoplasmicβ3 EM002-12, rabbit 1:500 (2 μg/mL) SCC1, 32 minutes + amplification (35
αvβ5 EM099-02, rabbit 1:800 (1.25 μg/mL) Protease 12 minutes (0.1 U/mL), 32 minutes (35
αvβ6 EM052-01, rabbit 1:1000 (1 μg/mL) Protease 12 minutes (0.1 U/mL), 32 minutes (35
αvβ8 EM133-09, rabbit 1:1000 (1 μg/mL) Protease 12 minutes (0.1 U/mL), 32 minutes (35
Common αv chain EM013-09, rabbit 1:1000 (1 μg/mL) SCC1, 32 minutes (35
Fibronectin 568, mouse 1:100 (not supplied) Trypsin 30 minutes (0.2 g), 32 minutes Novocastra, Newcastle, UK 
Fibrinogen 1F2, mouse 1:1000 (10 μg/mL) SCC1, 32 minutes AbD Serotec, Düsseldorf, Germany 
IgG IgG1 isotype control 1:500 (2 μg/mL) Pretreatment and incubation times matched to primary antibody Genetex, San Antonio, TX 
Antibody to Clone, Species Dilution (concentration) Pretreatment, Primary Antibody Incubation Time (Duration) Source 
αvβ3 EM227-03, rabbit 1:500 (2 μg/mL) Protease 12 minutes (0.1 U/mL), 32 minutes (35
Cytoplasmicβ3 EM002-12, rabbit 1:500 (2 μg/mL) SCC1, 32 minutes + amplification (35
αvβ5 EM099-02, rabbit 1:800 (1.25 μg/mL) Protease 12 minutes (0.1 U/mL), 32 minutes (35
αvβ6 EM052-01, rabbit 1:1000 (1 μg/mL) Protease 12 minutes (0.1 U/mL), 32 minutes (35
αvβ8 EM133-09, rabbit 1:1000 (1 μg/mL) Protease 12 minutes (0.1 U/mL), 32 minutes (35
Common αv chain EM013-09, rabbit 1:1000 (1 μg/mL) SCC1, 32 minutes (35
Fibronectin 568, mouse 1:100 (not supplied) Trypsin 30 minutes (0.2 g), 32 minutes Novocastra, Newcastle, UK 
Fibrinogen 1F2, mouse 1:1000 (10 μg/mL) SCC1, 32 minutes AbD Serotec, Düsseldorf, Germany 
IgG IgG1 isotype control 1:500 (2 μg/mL) Pretreatment and incubation times matched to primary antibody Genetex, San Antonio, TX 

SCC1 (standard cell conditioner 1) and protease are reagents in the Benchmark Immunohistochemistry System (Ventana Medical Systems, Strasbourg, France). Cytoβ3, β3-integrin cytoplasmic domain.

Tissue Samples

Tumor and normal human brain samples were retrieved from the neuropathology archives of the Department of Pathology and Neuropathology, Institute of Pathology and Neuropathology, University of Tübingen, Tübingen, Germany (Table 2). Tissue handling was performed according to the ethical guidelines of the University of Tübingen using a protocol approved by the ethics committee (Permission No. 249/ 2010BO1). Histopathologic designation and grading of the tumors were done by at least 2 experienced neuropathologists according to criteria of the 2007 World Health Organization (WHO) classification of tumors of the CNS (36).

TABLE 2

Tumor Samples and Number of Cases With Positively Stained Tumor Cells for Each Integrin Complex*

Tumor WHO Grade No. Samples (No. Cases, Female/Male) Mean age (range), years αvβ3 Cytoβ3 αvβ5 αvβ6 αvβ8 αv-Common Chain 
Diffuse astrocytoma II 94 (38/56) 47.5 (18–81) 1/66 3/66 48/68 0/68 66/69 60/68 
Anaplastic astrocytoma III 85 (31/54) 47.6 (10–83) 3/56 2/63 45/67 0/63 57/65 29/65 
Glioblastoma IV 324 (176/148) 60.3 (19–91) 86/160 80/150 121/147 0/153 145/150 123/134 
Normal brain  90 (3/6)† 69 (37–100) 0/78 8/78 0/79 0/79 13/80 78/81 
Tumor WHO Grade No. Samples (No. Cases, Female/Male) Mean age (range), years αvβ3 Cytoβ3 αvβ5 αvβ6 αvβ8 αv-Common Chain 
Diffuse astrocytoma II 94 (38/56) 47.5 (18–81) 1/66 3/66 48/68 0/68 66/69 60/68 
Anaplastic astrocytoma III 85 (31/54) 47.6 (10–83) 3/56 2/63 45/67 0/63 57/65 29/65 
Glioblastoma IV 324 (176/148) 60.3 (19–91) 86/160 80/150 121/147 0/153 145/150 123/134 
Normal brain  90 (3/6)† 69 (37–100) 0/78 8/78 0/79 0/79 13/80 78/81 
*

Data from manual assessment.

Ten samples from each autopsy case, including gray and white matter from frontal and occipital lobes, hippocampus, basal ganglia, cerebellum, mesencephalon, pons, brainstem. Cytoβ3, β3-cytoplasmic domain.

Construction of Tissue Microarrays

Of the total of 503 tumors, 407 (81%) were available as tissue microarrays (TMAs) (Table 2). Two 1,000-μm cylindrical tissue core biopsies containing representative tumor tissue sections were punched from each paraffin donor block and transferred into prepunched holes on recipient paraffin blocks at defined array coordinates. For each of the 9 autopsy cases diagnosed as normal human brain, a total of 10 different regions (including gray and white matter of frontal and occipital lobes, hippocampus, basal ganglia, cerebellum, mesencephalon, pons, brainstem) were transferred to a recipient TMA block using a TMA machine (Beecher Instruments, Inc., Sun Prairie, WI). After a short period of sealing at 37°C, the TMA blocks were cut with a microtome (4-μm-thick sections) and mounted on Super-Frost Plus slides (Microm International, Walldorf, Germany).

Immunohistochemistry

Immunohistochemistry was performed on FFPE full-slide tissue sections and microarrays after deparaffinization on an automated IHC system (BenchMark, Ventana Medical Systems, Strasbourg, France), essentially as previously described (35). This system uses an indirect biotin-avidin system with standard cell conditioner 1 and EDTA pretreatment protocol, a universal biotinylated immunoglobulin secondary antibody, and diaminobenzidine as chromogen. To improve signal quality, the tissue sections were incubated with a copper enhancer (all reagents from iView DAB IHC Detection Kit; Ventana Medical Systems) and counterstained with hematoxylin. Protocols were adapted to each primary antibody to achieve optimal signal-to-noise ratio after initial antibody titration (Table 1).

Positive controls included normal kidney and malignant melanoma for αvβ3 and cytoβ3, normal colon tissue and the HT-29 colon cancer cell line for αvβ5, and normal kidney and the HT-29 colon carcinoma cell line for αvβ6 (35). Positive controls for αvβ8 included normal human peripheral nerve and Ovcar-3 ovarian cancer cell line. For the antibody against the αv chain, which reacts with all of the αvβx complexes, normal colon tissue, HT-29 colon carcinoma cell line, and DU-145 prostate cancer cell line served as positive controls. Positive controls for fibronectin and fibrinogen included clear-cell renal carcinoma and GBM samples. Negative control slides were processed parallel to each batch of staining by replacing the primary antibody with the appropriate rabbit or murine polyclonal immunoglobulin G (IgG) isotype control (Genetex, San Antonio, TX) at the same concentrations of IgG.

Mutation Analysis and Methylation-Specific Polymerase Chain Reaction

DNA was extracted from GBM tissue sections from paraffin blocks (microscopically controlled for tumor content) using a BlackPREP FFPE kit (Analytik, Jena, Germany), according to the manufacturer's instructions. The IDH1 mutational status was assessed in WHO grade III and IV tumors by a mutation-specific monoclonal antibody for the IDH1 R132H mutation in tumor samples (32, 37). Negative cases (i.e. lacking the R132H mutation) were assessed by direct sequencing of the relevant exon for other rare IDH1 mutations (31). Bisulfite treatment of DNA and analysis for MGMT promoter methylation by methylation-specific polymerase chain reaction in WHO grade IV tumors were performed as previously described (38).

Assessment of Clinical and Publicly Available Gene Expression Data

Data on tumor localization, survival, age, sex, tumor type (primary or recurrent), edema, and Karnofsky score were retrieved from the anonymized clinical files of all gliomas available as TMA or full slides.

The NIH REMBRANDT database (Release 1.5.5, release date: 07-27-2010, URL: https://caintegrator.nci.nih.gov/rembrandt/) was analyzed for possible effects of variations in ITGB3, ITGB5, ITGB6, ITGB8, and ITGAV gene expression on overall patient survival. REMBRANDT samples were divided into 3 groups based on mean expression levels of all reporters (high, more than 2-fold; intermediate, 0.5- to 2.0-fold; and low, less than 0.5-fold) for median using the Li CancerRes 2009 probe sets with Affymetrix chipsets (39). Survival analysis was performed directly with the database analysis tool (40).

Data Analysis and Statistical Evaluation

Stained TMA slides were scanned with a digital camera (DFWX710; Sony, Japan) using a Mirax Scan system (Zeiss, Göttingen, Germany). Digitized data were transferred to a workstation (Definiens Tissue Studio 2.0, Munich, Germany). After defining the tumor regions to be examined, staining thresholds for nucleus detection and quantitative membrane and cytoplasmic intensity were defined on 4 subsets for subsequent automatic analysis. Processed data (including number of tumor cells analyzed and their staining intensity) were then exported into statistical analysis software.

In addition, the digitized stained slides were scored manually by 2 independent observers (Jens Schittenhelm, Esther Schwab). Expressions of integrin complexes and their ligands in vessels were semiquantitatively recorded as follows: 0 (staining absent); 1 (staining in <50% of vessels); and 2 (staining in ≥50% vessels). Cytoplasmic and membranous parenchymal integrin expression intensity in tumor cells was recorded as follows: 0 (absent); 1+ (weak expression); 2+ (moderate expression); and 3+ (strong expression).

Statistical analysis using JMP 7.0 (SAS Institute, Cary, NC) for automated analysis included analysis of variance of the score calculated for intensity and distribution (i.e. a histoscore calculated as the percentage of tumor cells with weak expression + double the percentage of tumor cells with moderate expression + triple the percentage of tumor cells with strong expression; the system default setting provided by De-finiens Tissue Studio 2.0, Munich, Germany) followed by non-parametric testing (Wilcoxon for 2 pairs or Kruskal-Wallis for >2 pairs). Logistic regression was used to compare staining data from the manual analysis with those from automated analysis and to compare IHC results with age. Correlation between tumor type, edema, sex, localization, IDH1 mutation status or MGMT promoter methylation status, and integrin complex scores was assessed using the Wilcoxon signed-rank test.

For the survival analysis, integrin subunit expression was split at the medians for the expression for each integrin complex in GBM, with subsequent Kaplan-Meier analysis. Significances were determined using the logrank test. Survival data were available for 130 patients (mean, 1.3 years; maximum follow-up, 10.5 years).

Results

αv Integrin Subunit Expression

Numbers of positive tumor cases for each integrin complex determined by manual evaluation of tumor cell distribution are shown in Table 2. Integrin αvβ3 was consistently expressed on cell membranes in endothelial cell proliferations within GBM (Fig. 1A). There was also αvβ3 cytoplasmic/ membranous staining in tumor cells that sometimes had a distinctly perivascular pattern (Fig. 1B). Vascular αvβ3 expression was observed in endothelial cells and pericytes but was absent in perivascular smooth muscle cells and macrophages. Similarly, cytoβ3 staining was observed in capillaries and in endothelial cell proliferations in GBM (Fig. 1C).

FIGURE 1

Immunohistochemical expression patterns of integrin complexes in glioblastoma samples. (A, B) Expression of αvβ3 in tumor vessels ([A] arrows) and with additional parenchymal αvβ3 immunoreactivity ([B] asterisks). (C, D) Expression of β3 cytoplasmic domain (cytoβ3) in vessels is stronger than in the surrounding tissue (C) and is similar to that in perinecrotic areas (D). Tumors with αvβ5 expression restricted to tumor vessels (E) and with additional parenchymal αvβ5 immunoreactivity (F). (G) Expression of αvβ8 is absent in vessels (arrows) but usually is strong in tumor parenchyma. (H) Integrin αvβ6 is absent in glioma cells and vessels. Scale bar = 100 μm.

FIGURE 1

Immunohistochemical expression patterns of integrin complexes in glioblastoma samples. (A, B) Expression of αvβ3 in tumor vessels ([A] arrows) and with additional parenchymal αvβ3 immunoreactivity ([B] asterisks). (C, D) Expression of β3 cytoplasmic domain (cytoβ3) in vessels is stronger than in the surrounding tissue (C) and is similar to that in perinecrotic areas (D). Tumors with αvβ5 expression restricted to tumor vessels (E) and with additional parenchymal αvβ5 immunoreactivity (F). (G) Expression of αvβ8 is absent in vessels (arrows) but usually is strong in tumor parenchyma. (H) Integrin αvβ6 is absent in glioma cells and vessels. Scale bar = 100 μm.

Cytoplasmic/membranous cytoβ3 expression was preferentially found in large anaplastic GBM cells and perinecrotic tumor regions (Fig. 1D). In approximately one fourth of the tumors, αvβ5 expression was limited to vessels (Fig. 1E). In the remainder of the tumors, the distribution of cytoplasmic/ membranous αvβ5 was pronounced in perivascular tumor cells (Fig. 1F). Integrin αvβ5 expression was always observed in endothelial cells, pericytes, perivascular monocytes/macrophages, and occasionally, in perivascular smooth muscle cells. Cytoplasmic/membranous αvβ8 was strongly expressed in most glioma tumor cells; in many cases, only sparing the tumor vessels (Fig. 1G). Gliomas lacked αvβ6 (Fig. 1H).

Cytoplasmic/membranous αv expression, although usually strong, was absent in many tumors (Table 2). Ten regions of 9 normal human brain autopsy cases were evaluated manually for αvβx integrin expression (Table 2). No distinct αvβ3 expression was found in normal vessels or CNS parenchyma (Fig. 2A). There was homogeneous weak cytoβ3 expression in the basal ganglia in 2 cases (Fig. 2B), but expression was absent in other brain regions. Integrin αvβ5 was seen in most capillaries of normal brain but not in the CNS parenchyma (Fig. 2C, arrows). As in gliomas, αvβ6 was absent from normal human brain (Fig. 2D). In some cases, there was prominent and homogeneous immunoreactivity for αvβ8 in neocortical neuropil, whereas adjacent vessels and white matter were unstained (Fig. 2E). The αv complex was homogeneously expressed in all human CNS regions and was also observed to a weaker extent in CNS vessels (Fig. 2F).

FIGURE 2

Immunohistochemical expression patterns of integrin complexes in normal human brain samples. (A) Lack of αvβ3 expression in basal ganglia (inset: melanoma cell line as a positive control). (B) Weak and diffuse β3 cytoplasmic domain (cytoβ3) expression in basal ganglia. (C) Expression of αvβ5 in basal ganglia is restricted to microvessels (arrows). (D) Integrin αvβ6 is absent in normal brain (inset: adenocarcinoma of the intestine as a positive control). (E) Neocortical αvβ8 expression is slightly enhanced in subpial neuropil regions but is absent in vessels. (F) Neocortical αv complex expression in neuropil and in vessels (arrow). Scale bar = 100 μm.

FIGURE 2

Immunohistochemical expression patterns of integrin complexes in normal human brain samples. (A) Lack of αvβ3 expression in basal ganglia (inset: melanoma cell line as a positive control). (B) Weak and diffuse β3 cytoplasmic domain (cytoβ3) expression in basal ganglia. (C) Expression of αvβ5 in basal ganglia is restricted to microvessels (arrows). (D) Integrin αvβ6 is absent in normal brain (inset: adenocarcinoma of the intestine as a positive control). (E) Neocortical αvβ8 expression is slightly enhanced in subpial neuropil regions but is absent in vessels. (F) Neocortical αv complex expression in neuropil and in vessels (arrow). Scale bar = 100 μm.

Integrin αv Complexes Staining Intensity Scores and Tumor Grade

Integrin αvβ3 expression intensity, as determined manually, was significantly higher in GBM than in anaplastic astrocytomas, diffuse astrocytomas WHO grade II, and normal brain (p < 0.0001) (Table 3A). The difference in αvβ3 expression between grade III and grade II neoplasms was not significant (Table 3A). Likewise, the mean intensity of expression of the cytoplasmic β3 domain was significantly higher in GBM (p < 0.0001); expression was almost absent in astrocytomas and normal brain, with no significant differences between grades II and III and nontumor tissue.

TABLE 3

Manual Evaluation of Integrin Complexes and Ligands

 αvβ3 Cytoβ3 
No. Tumors Mean SD No. Tumors Mean SD 
Tumor grade         
WHO II 66 0.02 0.12  66 0.05 0.21  
WHO III 56 0.11 0.56  63 0.03 0.18  
WHO IV 159 0.50 0.68 <0.0001* 129 0.74 0.67 <0.0001* 
Tumor type         
Primary 133 0.14 0.55  128 0.25 0.53  
Recurrent 40 0.03 0.16 0.43 37 0.32 0.58 0.15 
Sex         
Female 111 0.34 0.61  100 0.37 0.56  
Male 166 0.28 0.62 0.30 152 0.39 0.63 0.98 
Localization         
Both     
Left 62 0.15 0.51  62 0.32 0.59  
Right 95 0.07 0.33 0.55 89 0.27 0.54 0.75 
Age† 270 0.31 0.61 0.0007 244 0.39 0.60 0.0002 
Karnofsky† 134   0.0337 129   0.0105 
Edema         
No 0.00   0.00   
Low 10 0.60 0.70  10 0.40 0.52  
High 32 0.44 0.56 0.57 33 0.45 0.56 0.70 
IDH1 mutation status‡         
Absent 137 0.37 0.68  128 0.52 0.65  
Present 33 0.15 0.37 0.084 33 0.18 0.46 0.0053 
MGMT promoter methylation        
Methylated 14 0.85 0.77 0.87 11 1.09 0.70  
Unmethylated 16 0.81 0.75  11 1.00 1.00 0.80 
 αvβ5 αvβ6 
B No. Tumors Mean SD p No. Tumors Mean SD p 
Tumor grade         
WHO II 68 0.85 0.68  68  
WHO III 64 0.88 0.72  63  
WHO IV 146 1.16 0.78 0.0042* 151 1.0 
Tumor type         
Primary 135 0.94 0.69  131  
Recurrent 39 0.87 0.66 0.88 38 1.0 
Sex         
Female 109 1.06 0.77  111  
Male 165 1.01 0.74 0.69 167 1.0 
Localization         
Both     
Left 65 0.95 0.67  61  
Right 94 0.89 0.65 0.81 93 1.0 
Age† 267 1.02 0.75 0.25 268  
Karnofsky† 129   0.63 131  
Edema         
No 2.00     
Low 10 1.30 0.82  10  
High 29 0.79 0.68 0.08 33 1.0 
IDH mutation status‡         
Absent 133 1.01 0.77  105  
Present 34 0.97 0.62 0.75 35 1.0 
MGMT promoter methylation        
Methylated 13 1.30 0.48  14  
Unmethylated 14 1.57 0.85 0.32 16 1.0 
 αvβ8 αv Chain 
C No. Tumors Mean SD p No. Tumors Mean SD p 
Tumor grade         
WHO II 69 1.68 0.78  68 2.19 0.63  
WHO III 65 1.66 0.94  65 1.02 1.29  
WHO IV 149 2.53 0.73 <0.0001* 133 1.65 1.38 <0.0001* 
Tumor type         
Primary 136 1.90 0.93  135 1.90 1.14  
Recurrent 37 1.86 0.92 0.43 39 1.72 1.26 0.42 
Sex         
Female 113 2.14 0.93  104 1.66 1.28  
Male 166 2.11 0.89 0.61 158 1.66 1.27 0.90 
Localization         
Both  1.5 2.12132034  
Left 63 1.86 0.98  65 1.95 1.10  
Right 93 1.99 0.89 0.76 92 1.83 1.22 0.93 
Age† 271 2.12 0.90 <0.0001 244 1.64 1.28 0.81 
Karnofsky† 133   0.0031 131   0.32 
Edema         
No 3.00   0.00   
Low 10 2.80 0.42  10 0.00 0.00  
High 33 2.61 0.66 0.63 32 0.16 0.63 0.70 
IDH1 mutation status‡         
Absent 137 2.35 0.88  133 2.00 1.16  
Present 36 1.75 0.77 0.0002 35 1.06 1.30 <0.0001 
MGMT promoter methylation        
Methylated 13 2.23 1.01  12 2.66 0.49  
Unmethylated 15 2.40 0.98 0.65 11 2.63 0.50 0.88 
 Fibrinogen Fibronectin 
D No. Tumors Mean SD p No. Tumors Mean SD p 
Tumor grade         
WHO II 66 0.21 0.73  66 0.02 0.12  
WHO III 67 0.04 0.27  67 0.39 0.63  
WHO IV 140 1.06 0.91 <0.0001* 140 0.68 0.79 <0.0001* 
Tumor type         
Primary 133 0.33 0.70  133 0.29 0.58  
Recurrent 38 0.39 0.75 0.06 38 0.42 0.64 0.03 
Sex         
Female 106 0.63 0.88  105 0.50 0.74  
Male 162 0.60 0.91 0.67 164 0.41 0.68 0.27 
Localization         
Both   
Left 63 0.33 0.74  62 0.32 0.62  
Right 92 0.40 0.74 0.59 92 0.33 0.63 0.72 
Age† 260 0.60 0.89 <0.0001 261 0.45 0.70 0.00017 
Karnofsky† 129   0.0266 129   0.0022 
Edema         
No    0.00   
Low 10 0.90 0.88  10 0.40 0.52  
High 32 0.84 1.02 0.74 33 0.42 0.66 0.77 
IDH1 mutation status‡         
Absent 131 0.73 0.86  133 0.51 0.68  
Present 31 0.12 0.42 0.0002 31 0.35 0.35 0.24 
MGMT promoter methylation        
Methylated 1.11 1.16  10 0.50 0.52  
Unmethylated 11 1.25 0.62 0.75 12 0.91 1.08 0.25 
 αvβ3 Cytoβ3 
No. Tumors Mean SD No. Tumors Mean SD 
Tumor grade         
WHO II 66 0.02 0.12  66 0.05 0.21  
WHO III 56 0.11 0.56  63 0.03 0.18  
WHO IV 159 0.50 0.68 <0.0001* 129 0.74 0.67 <0.0001* 
Tumor type         
Primary 133 0.14 0.55  128 0.25 0.53  
Recurrent 40 0.03 0.16 0.43 37 0.32 0.58 0.15 
Sex         
Female 111 0.34 0.61  100 0.37 0.56  
Male 166 0.28 0.62 0.30 152 0.39 0.63 0.98 
Localization         
Both     
Left 62 0.15 0.51  62 0.32 0.59  
Right 95 0.07 0.33 0.55 89 0.27 0.54 0.75 
Age† 270 0.31 0.61 0.0007 244 0.39 0.60 0.0002 
Karnofsky† 134   0.0337 129   0.0105 
Edema         
No 0.00   0.00   
Low 10 0.60 0.70  10 0.40 0.52  
High 32 0.44 0.56 0.57 33 0.45 0.56 0.70 
IDH1 mutation status‡         
Absent 137 0.37 0.68  128 0.52 0.65  
Present 33 0.15 0.37 0.084 33 0.18 0.46 0.0053 
MGMT promoter methylation        
Methylated 14 0.85 0.77 0.87 11 1.09 0.70  
Unmethylated 16 0.81 0.75  11 1.00 1.00 0.80 
 αvβ5 αvβ6 
B No. Tumors Mean SD p No. Tumors Mean SD p 
Tumor grade         
WHO II 68 0.85 0.68  68  
WHO III 64 0.88 0.72  63  
WHO IV 146 1.16 0.78 0.0042* 151 1.0 
Tumor type         
Primary 135 0.94 0.69  131  
Recurrent 39 0.87 0.66 0.88 38 1.0 
Sex         
Female 109 1.06 0.77  111  
Male 165 1.01 0.74 0.69 167 1.0 
Localization         
Both     
Left 65 0.95 0.67  61  
Right 94 0.89 0.65 0.81 93 1.0 
Age† 267 1.02 0.75 0.25 268  
Karnofsky† 129   0.63 131  
Edema         
No 2.00     
Low 10 1.30 0.82  10  
High 29 0.79 0.68 0.08 33 1.0 
IDH mutation status‡         
Absent 133 1.01 0.77  105  
Present 34 0.97 0.62 0.75 35 1.0 
MGMT promoter methylation        
Methylated 13 1.30 0.48  14  
Unmethylated 14 1.57 0.85 0.32 16 1.0 
 αvβ8 αv Chain 
C No. Tumors Mean SD p No. Tumors Mean SD p 
Tumor grade         
WHO II 69 1.68 0.78  68 2.19 0.63  
WHO III 65 1.66 0.94  65 1.02 1.29  
WHO IV 149 2.53 0.73 <0.0001* 133 1.65 1.38 <0.0001* 
Tumor type         
Primary 136 1.90 0.93  135 1.90 1.14  
Recurrent 37 1.86 0.92 0.43 39 1.72 1.26 0.42 
Sex         
Female 113 2.14 0.93  104 1.66 1.28  
Male 166 2.11 0.89 0.61 158 1.66 1.27 0.90 
Localization         
Both  1.5 2.12132034  
Left 63 1.86 0.98  65 1.95 1.10  
Right 93 1.99 0.89 0.76 92 1.83 1.22 0.93 
Age† 271 2.12 0.90 <0.0001 244 1.64 1.28 0.81 
Karnofsky† 133   0.0031 131   0.32 
Edema         
No 3.00   0.00   
Low 10 2.80 0.42  10 0.00 0.00  
High 33 2.61 0.66 0.63 32 0.16 0.63 0.70 
IDH1 mutation status‡         
Absent 137 2.35 0.88  133 2.00 1.16  
Present 36 1.75 0.77 0.0002 35 1.06 1.30 <0.0001 
MGMT promoter methylation        
Methylated 13 2.23 1.01  12 2.66 0.49  
Unmethylated 15 2.40 0.98 0.65 11 2.63 0.50 0.88 
 Fibrinogen Fibronectin 
D No. Tumors Mean SD p No. Tumors Mean SD p 
Tumor grade         
WHO II 66 0.21 0.73  66 0.02 0.12  
WHO III 67 0.04 0.27  67 0.39 0.63  
WHO IV 140 1.06 0.91 <0.0001* 140 0.68 0.79 <0.0001* 
Tumor type         
Primary 133 0.33 0.70  133 0.29 0.58  
Recurrent 38 0.39 0.75 0.06 38 0.42 0.64 0.03 
Sex         
Female 106 0.63 0.88  105 0.50 0.74  
Male 162 0.60 0.91 0.67 164 0.41 0.68 0.27 
Localization         
Both   
Left 63 0.33 0.74  62 0.32 0.62  
Right 92 0.40 0.74 0.59 92 0.33 0.63 0.72 
Age† 260 0.60 0.89 <0.0001 261 0.45 0.70 0.00017 
Karnofsky† 129   0.0266 129   0.0022 
Edema         
No    0.00   
Low 10 0.90 0.88  10 0.40 0.52  
High 32 0.84 1.02 0.74 33 0.42 0.66 0.77 
IDH1 mutation status‡         
Absent 131 0.73 0.86  133 0.51 0.68  
Present 31 0.12 0.42 0.0002 31 0.35 0.35 0.24 
MGMT promoter methylation        
Methylated 1.11 1.16  10 0.50 0.52  
Unmethylated 11 1.25 0.62 0.75 12 0.91 1.08 0.25 

For each antibody, the number of tumors, mean intensity scores, and SD are shown. Results of the multivariate analysis are from Wilcoxon testing if not otherwise specified.

*

Kruskal-Wallis test.

Logistic fit.

High-grade WHO III and IV gliomas.

IDH1, isocitrate dehydrogenase 1; MGMT, methyl guanine methyltransferase.

Mean αvβ5 expression intensity in GBM was significantly greater than that in anaplastic astrocytomas (p = 0.005) and diffuse astrocytomas WHO grade II (p = 0.01) (Table 3 B). The mean intensity in all tumors was significantly higher than in normal brain (p < 0.0001), but the difference between grade II and III tumors was not significant. The mean αvβ8 expression intensity was strongest in GBMs, with an expression score significantly higher (p < 0.0001) than expression levels in anaplastic astrocytomas and WHO grade II diffuse astrocytomas (Table 3 C). In all tumor groups, the mean αvβ8 intensity scores were significantly higher than those in normal brain (p < 0.0001); there were no significant intensity differences between grade II and III tumors (Table 3C). Integrin αvβ6 was absent from gliomas and normal brain (Table 3B).

As expected from the results of the staining intensity of αvβx complexes, mean αv expressions were strong in GBMs and anaplastic astrocytomas, with the highest levels in WHO grade II diffuse astrocytomas (Table 3C). The differences in expression between all 3 groups were highly significant (p < 0.0001 to p = 0.0032). The expression of the integrin ligands matched the pattern seen in the αvβx complexes. Whereas there was a significant and progressive increase of mean fibronectin intensity from grade II diffuse astrocytomas (p = 0.001, II vs III; p < 0.0001, II vs IV) over anaplastic astrocytomas to GBM (p < 0.0028 for III vs IV), the mean expression intensity of fibrinogen was significantly lower in grade II and III tumors (no statistical differences between tumor grades II and III) than in GBM (p < 0.0001) (Table 3D).

Association Between Staining Intensities of αv Integrin Complexes and Ligands

Multivariate analysis of all tumors followed by pairwise correlations of manual staining intensity revealed a significant positive association between integrin complexes αvβ3, αvβ5, αvβ8 and cytoβ3 (p < 0.0001 to p = 0.0002). There was a significant positive correlation with the αv subunit for αvβ5 (p = 0.0306) and cytoβ3 (p = 0.0009) but not for αvβ3 (p = 0.441) and αvβ8 (p = 0.729). The positive association of the β complexes with the staining intensity of the ligands fibrinogen (p < 0.0001) and fibronectin (p = 0.0020) was also highly significant.

Integrin αvβ3 and αvβ8 Staining and Other Tumor Features

The mean αvβ3 (p = 0.084), β3 cytoplasmic domain (p = 0.0053), αvβ8 (p = 0.0002), αv (p < 0.0001), and fibronectin (p = 0.0002) staining intensities were less in high-grade tumors carrying the prognostic IDH1 R132H mutation (Table 3A, C, D); the mean αvβ5 intensity was similar in both groups (Table 3B). There was no difference in αv integrin expression intensity between GBMs expressing methylated or unmethylated MGMT promoters (Table 3A–D). The intensity of staining of integrin complexes in relapsing tumors was the same as that in primary neoplasms (Table 3A–D). The mean intensity significantly increased with patient age for αvβ3 (p = 0.0007), cytoplasmic β3 (p = 0.0002), αvβ8 (p < 0.0001), fibrinogen (p < 0.0001), and fibronectin (p = 0.00017), whereas αvβ5 and αv expression levels were not significantly different between age groups (Table 3A–D). These results did not retain significance after the tumors were separated according to their WHO grade and age was correlated with mean integrin expression intensity levels in these subgroups; however, there was an age-dependent increase in expression of αvβ8 (p = 0.0249) and fibrinogen (p = 0.003) in GBM (data not shown).

Patients with lower Karnofsky scores exhibited slightly higher mean αv integrin intensity than those with higher scores (Table 3A–C), except for αvβ5. After separating tumors according to their WHO grade, the results were no longer significant, except for αvβ8 in diffuse astrocytoma (p = 0.0273) and fibronectin (p = 0.0268) in GBM (Table 3D). No significant associations were found between sex, tumor localization, or edema and integrin intensity.

Vascular αvβ3 and Cytoplasmic β3 Expressions Increase With Tumor Grade

Vessel staining was analyzed only in the 63 tumor samples available as full slides because not all TMA punches contained scorable vasculature. Integrins αvβ6 and αvβ8 were absent in tumor vessels (Fig. 1G, H). There was significant upregulation of αvβ3 (p < 0.0001) and cytoplasmic β3 (p < 0.0001) distribution in tumor vessels that correlated with the grade of malignancy, whereas the distribution of other inte-grin complexes remained stable (i.e. no significant differences among tumor grades). In addition, the expression of the ligands fibronectin (p < 0.0001) and fibrinogen (p = 0.0007) in GBM vessels showed a significant increase with tumor grade (Fig. 3D).

FIGURE 3

(A) Immunohistochemical staining for αvβ5 of a glioblastoma tissue microarray sample (left) and an example of a detailed computed analysis after segmentation of the same sample into cell-based fields (right). The colors indicate quantification of expression within each field: (white, no staining score; yellow, weak staining, 1+; orange, moderate staining, 2+; Bordeaux red magenta, strong staining, 3+). (B) Mean percentage of cells scored as 0 to 3+ of all tissue microarrays stratified for tumor grade and analyzed for each integrin complex. αv-, pan-αv integrin; cytoβ3, β3 integrin cytoplasmic domain.

FIGURE 3

(A) Immunohistochemical staining for αvβ5 of a glioblastoma tissue microarray sample (left) and an example of a detailed computed analysis after segmentation of the same sample into cell-based fields (right). The colors indicate quantification of expression within each field: (white, no staining score; yellow, weak staining, 1+; orange, moderate staining, 2+; Bordeaux red magenta, strong staining, 3+). (B) Mean percentage of cells scored as 0 to 3+ of all tissue microarrays stratified for tumor grade and analyzed for each integrin complex. αv-, pan-αv integrin; cytoβ3, β3 integrin cytoplasmic domain.

Correlations of Digital and Manual Staining

The logistic fit of manual intensity scoring with the histoscores calculated by automated intensity and distribution analysis with the Definiens image processing software was analyzed where available (n = 55; only TMA). The 2 methods showed a significant association for αvβ3 (p < 0.0001), cytoβ3 (p = 0.0005), αvβ5 (p < 0.0001), αvβ8 (p = 0.0001), αv (p < 0.0001), fibrinogen (p < 0.0001), and fibronectin (p = 0.0009) expression. Integrin αvβ6 was not expressed.

Staining Intensities and Tumor Grade

After the overall concordance between automatic and manual data was confirmed, 386 brain tumor TMA cores were evaluated by detailed automated cell-based analysis for the staining intensity and distribution of each integrin complex (Fig. 3). Each TMA core was segmented into cell fields depending on the nuclei present in the immunostained specimen. An average of 3,853 fields on a single TMA core with a mean diameter of 70.3 μm for each field were available for a detailed analysis of staining intensity (Fig. 3A). Mean percentages of tumor cell segments showing absence or weak, moderate, and strong expression in all fields were calculated (Fig. 3B).

The digitized αvβ3 microarray samples of high-grade gliomas (WHO III and WHO IV) showed a higher mean percentage of tumor cells exhibiting moderate (17% in WHO III; 13% in WHO IV) and strong staining intensity (20% in WHO III; 14% in WHO IV) than in grade II tumors (p = 0.037, 11% moderate; p = 0.013, 12% strong), whereas the frequency of low expression intensities (68% in WHO II; 58% in WHO III; 69% in WHO IV) was similar. This analysis included some capillary endothelial cells that could not always be excluded; however, the mean percentage of tumor cells negative for αvβ3 was higher in GBM (22%) than that in grade II (7%) and III neoplasms (4%) (p < 0.0001; Fig. 3B), indicating that a subset of GBM loses αvβ3 during tumorigenesis/progression. Analysis of αvβ5 showed that the distribution of staining intensity was unchanged between tumor grades.

In GBM, the mean percentage of αvβ8 high-expressing cells was significantly higher (44%; p < 0.001) than that in grade II (8%) and grade III (8%) tumors, indicating an increase in intensity accompanying tumor progression. By contrast, fewer GBM showed a low mean αvβ8 staining intensity (to 28%, p < 0.001; Fig. 3B) than grade II (71%) and grade III (67%) tumors. There was also a slightly increased mean percentage of tumor cells with moderate αvβ8 immunoreactivity in GBM (26%) compared with those in astrocytoma grade II (17%) and grade III (16%).

Analysis of the αv integrin subunit showed that the numbers of αv high-expressing cells were lower in grade II (17%) and III (9%) tumors than in grade IV tumors (51%, p < 0.0001), whereas αv low-expressing and immunonegative cells were more frequent in grade II (low, 19%; absent, 36%) and III tumors (low, 10%; absent, 68%) than in GBMs (low, 8%; absent, 9%; p = 0.0013 to p < 0.0001), independently indicating that there was a general increase in αv integrin expression during tumor progression.

Combined Intensity and Distribution Histoscore Generated by Automated Analysis

The intensity and distribution results of the automated analysis (Fig. 3B) were combined into a histoscore for each complex and tumor grade (Fig. 4). The significance of the mean histoscores was divergent for αvβ3, possibly representing an activation epitope (p = 0.02 between grades II and III; nonsignificant between grades III and IV), and cytoplas-mic β3 (nonsignificant between grades II and III; p = 0.0039 between grades III and IV), representing the overall protein (35). In contrast, the mean histoscores for αvβ5 did not differ significantly between the tumor grades, and, as expected, αvβ6 was completely absent. The mean αvβ8 and αv histoscores were significantly higher in GBM (p < 0.0001) than those in grade II and III tumors (Fig. 4).

FIGURE 4

Combined histoscore results of an automated analysis of staining intensity and distribution of immunoreactive parenchymal tumor cells for each integrin complex differentiated according to WHO tumor grade. The number of tumor samples is shown for each grade. Means are depicted by bars.

FIGURE 4

Combined histoscore results of an automated analysis of staining intensity and distribution of immunoreactive parenchymal tumor cells for each integrin complex differentiated according to WHO tumor grade. The number of tumor samples is shown for each grade. Means are depicted by bars.

ITGB3 and ITGB5 Gene Expression in an Independent Data Set

We attempted to validate the results of the IHC analysis independently at the level of gene expression. We analyzed the gene expression of the relevant integrin chains in the Unified Gene data set NIH REMBRANDT (40). This analysis revealed that ITGB3 expression increased during progression from normal human brain (median, 39.3) to astrocytoma tissue (median, 59.0), with the highest levels in GBM (median, 100.3). Expression of ITGB5 in tumors (median GBM, 1,048; median astrocytoma, 853.5) was also higher than that in normal brain (median, 532). Expression of ITGB8 gene increased during the progression from nontumor tissue (median, 167.4) to astrocytoma (median, 294.7) and then to GBM (median, 351.1). The ITGB6 gene expression intensity in astrocytomas (median, 26.7) and GBMs (median, 23.8) did not exceed baseline levels found in normal brain (median, 34.1). Thus, the gene expression data from an independent sample of normal and tumor patients closely correlated with the protein-based longitudinal IHC data from our brain tumor FFPE archive.

Survival Analysis

Statistical analysis of the survival plots of patients from the current IHC study split at the median histoscore from the automated analysis of GBM (Fig. 4) showed a significantly better overall survival for those with no or low expression of the αvβ3 complex (logrank, p = 0.016; median histoscore used for splitting, 67; number of samples analyzed, 39) than for those with high expression (i.e. above the median). In contrast, no significant association was seen between patient survival and integrin αvβ5 (logrank, p = 0.46; median, 82; n = 47), β3 cytoplasmic domain (logrank, p = 0.07; median, 49; n = 57), αv (logrank, p = 0.2225; median, 155; n = 30), or αvβ8 (logrank, p = 0.93; median, 211; n = 58) expression in GBM (Fig. 5).

FIGURE 5

Kaplan-Meier survival analysis of patients expressing αvβ3 (upper left, low [red]/high [blue], n = 17/22), β3 cytoplasmic domain (cytoβ3) (upper right, low/high, n = 35/22), αvβ5 (middle left, low/high, n = 14/33), αvβ8 (middle right, low/high, n = 26/ 32), and αv (lower left, low/high, n = 12/18) assessed by immunohistochemistry using rabbit monoclonal antibodies in glioblastomas split at the median histoscore. Integrin αvβ6 was excluded because no positively stained tumors were available (scale bars are tumor dependent because not all tissue microarray tumor punches were available for all integrin complex stains).

FIGURE 5

Kaplan-Meier survival analysis of patients expressing αvβ3 (upper left, low [red]/high [blue], n = 17/22), β3 cytoplasmic domain (cytoβ3) (upper right, low/high, n = 35/22), αvβ5 (middle left, low/high, n = 14/33), αvβ8 (middle right, low/high, n = 26/ 32), and αv (lower left, low/high, n = 12/18) assessed by immunohistochemistry using rabbit monoclonal antibodies in glioblastomas split at the median histoscore. Integrin αvβ6 was excluded because no positively stained tumors were available (scale bars are tumor dependent because not all tissue microarray tumor punches were available for all integrin complex stains).

Correlation of ITGB3 gene expression from the NIH REMBRANDT database and patient survival revealed a significantly poorer outcome for those showing high ITGB3 levels (i.e. >2-fold gene expression changes; patients, n = 53) than for those with intermediate and low ITGB3 transcript levels (<2-fold expression change; n = 203; logrank, p = 0.0022) (Fig. 6). Patients with more than 2-fold ITGB5 upregulation (n = 77) showed a trend toward poorer survival compared with the ITGB5 intermediate and low groups (n = 177; logrank, p = 0.06). No significant differences were observed for the ITGB6 upregulated and intermediate/low ITGB6-expressing cohort (p = 0.98). By contrast, significantly poorer survival was noted in the cohort with high ITGB8 transcript levels (n = 188) compared with those with intermediate and low ITGB8 transcript levels (n = 68; logrank, p = 0.0006). Likewise, the cohort with high ITGAV transcript levels (n = 94) showed significantly poorer survival than the intermediate/low ITGAV transcript group.

FIGURE 6

Kaplan-Meier survival analysis of glioma samples from an NIH REMBRANDT data set (x axis: days of observation) with mean upregulated ITGB3 gene expression (blue) compared with those with intermediate (red) and low ITGB3 expression. Logrank p values are shown for each gene inside the graph.

FIGURE 6

Kaplan-Meier survival analysis of glioma samples from an NIH REMBRANDT data set (x axis: days of observation) with mean upregulated ITGB3 gene expression (blue) compared with those with intermediate (red) and low ITGB3 expression. Logrank p values are shown for each gene inside the graph.

Discussion

Several agents can modulate cell attachment, differentiation, and migration through inhibition of αvβ3 and αvβ5 integrins (1, 23, 41, 42), which are current therapeutic targets in human gliomas (42, 43). In this study, we examined the expression of αv integrins by IHC in archival FFPE brain tumor specimens and compared this with historical gene expression data in a different patient population recorded in the NIH REMBRANDT expression database (39). Our principal findings are 1) that expressions of αvβ3 and αvβ8 protein and ITGB3 and ITGB8 mRNA increase during glioma progression; 2) that patients with above median protein expression at diagnosis for αvβ3 have a poorer prognosis than those with below the median, but this does not hold for αvβ5 or αvβ8; 3) that the same poorer prognosis for ITGB3 overexpression is seen in the NIH REMBRANDT sample (in addition to a poor prognosis associated with ITGB8); and 4) that automated sample processing and data collection by IHC using a novel set of RabMabs against αv integrins can provide a robust prognostic signal in glioma patients. This cross validation of protein and mRNA expression data strongly indicates that αv integrins, and especially αvβ3, are independent prognostic indifcators in gliomas. An αvβ3/αvβ5 inhibitor, cilengitide, is currently in a phase III clinical trial for glioma (6, 42, 44), and it will be interesting to compare the outcome of this trial with the αv integrin expression profiles of the patients.

To date, analysis of integrins in human pathology specimens has been hampered because most of the available antibodies stain only frozen tissue, which is rarely accessible for longitudinal study, whereas antibodies detecting integrins in FFPE specimens are generally lacking. Thus, the type of study of integrin profile in archival materials described here was essentially not possible in the past. Newly developed RabMabs against integrin αv complexes (35) show easily interpretable staining patterns in FFPE tissue, including samples of normal brain, diffuse astrocytomas, and GBMs. They can be used on standard IHC autostainers using standardized validated protocols.

In this study, there was excellent correlation between the scores performed independently by a pathologist and an automated computer analysis for staining intensity and distribution. This validates the analysis algorithm used (45) and indicates that the methods are suitable for evaluating large numbers of specimens. The automated method also provided detailed data for each field, approximately representing an analysis of individual tumor cells. This allowed us to detect shifts of expression within tumor cohorts that are often hidden within an overall staining score, for example, what is routinely applied in immunopathology to regions of heterogeneous expression. A pathologist must still instruct the software during automated analysis about which regions of the section are to be analyzed because the analyzer cannot discriminate between normal, necrotic, or neoplastic tissue and does not fully exclude smaller vessels. The current drawback is the time-consuming step of digitizing individual slides and selecting regions for analysis. We have found that construction of TMAs is the method of choice to optimize the automated analysis and save valuable time.

The integrin αvβ3 is a promiscuous receptor for components of the provisional ECM, including vitronectin, fibrin-ogen, and fibronectin, and is found in normal smooth muscle cells, osteoclasts, monocytes, and platelets (46). Integrin αvβ3 is upregulated in gliomas, melanomas, and carcinomas (2, 11, 19), and its expression is deregulated in autoimmune diseases and transplant rejection (6, 12). In glioma cell lines, upregulation of αvβ3 can increase their sensitivity to anticancer radiotherapy and to drug-induced apoptosis (14, 47). Endothelial αvβ3 enhances the activity of vascular endothelial growth factor and platelet-derived growth factor, which are frequently expressed in malignant gliomas (48, 49). Using relatively small numbers of tumor frozen sections, Gladson (16) and Bello et al (15) demonstrated that αvβ3 expression occurs in glioma endothelial cells. We were now able to confirm these observations on a large number of FFPE tumors, that is, αvβ3 is indeed expressed in glioma tumor vessels and is upregulated with grade of malignancy, further supporting a role of αvβ3 for vascular development in GBM. Using the monoclonal antibody LM609, Bello et al (15) reported αvβ3 to be highly expressed in 35% of tumors (14 cases), moderately in 18% (7 cases), and weakly or sporadically in 30% (12 cases). In our αvβ3 analysis of FFPE specimens, however, we obtained results more similar to a later study (20): 53% of our GBM expressed αvβ3-immunopositive cells. Furthermore, we determined that parenchymal expression of αvβ3, although present even in some low-grade tumors, is indeed correlated with increased tumor grade both in immunostains and gene expression analyses and is also associated with a poorer prognosis compared with αvβ3 parenchymal-negative tumors. There was no such correlation at the αvβ5 and αvβ8 protein level. Moreover, expression of αvβ3 was reduced in IDH1-mutated gliomas, probably further affecting the prognostic differences between these groups (32, 37). This is important because integrin αvβ3 and αvβ5 inhibitors affect both tumor vessels and parenchymal cells (16, 21, 24). In contrast to αvβ3, expression of αvβ5 in GBM vessels and parenchymal cells remained constant in our series, indicating a major role for αvβ3 in gliomas. Surprisingly, we did not find an association between any of the integrin complexes and MGMT promoter methylation, as GBM patients carrying the MGMT methylation have a better prognosis under combined temozolomide, radiotherapy, and cilengitide treatment (50). Further studies on the possible prognostic relationship of MGMT and αvβ3 are needed.

In principle, the antibody for the cytoplasmic domain of β3 (EM002-12) and the anti-αvβ3 antibody (EM227-03) should have similar tissue staining patterns outside the vasculature. Whereas the intensity results were similar (Table 3A), the combined intensity and distribution scores were divergent (Fig. 4). The αvβ3 antibody (EM227-03) can recognize an epitope expressed more strongly on ligated integrins, whereas the cytoplasmic β3 reagent (EM002-12) should see all available αvβ3. Thus, the difference between the poorer prognosis of αvβ3 (i.e. EM22703 high expressers) and the significantly better prognosis for low expressers and the similar survival of the high- and low-expressing cytoβ3 population is of interest. Perhaps, ligated αvβ3 integrin in patients with poor prognosis is related to their outcome, which would justify a targeted αvβ 3 therapy (43). However, it is unclear whether EM22703 reacts with activated αvβ3 in situ, so this remains a speculation. Publicly available data on integrin gene expression from the REMBRANDT files also reveal a poor outcome for glioma patients in whom ITGB3 was upregulated, which substantiates our data on patients with high αvβ3 protein expression as assessed by IHC.

Integrin αvβ5, a vitronectin receptor, has been reported in several types of cancer and established cell lines (35, 51–53). In carcinomas, interaction between EGFR and integrin αvβ5 affects metastatic and invasive cell potential (54). Recent data indicate that there is parenchymal αvβ5 expression in gliomas and in angiogenic endothelial cells (27). Previous studies indicate that the number of integrin-positive cells in frozen glioma tissue is between 30% and 85%, depending on the antibodies and evaluation methods used (15, 20, 55). The observed distribution heterogeneity in tumor samples may result from hypoxia-dependent induction of αvβ5 in gliomas (56). Overall, our data indicate that the role of αvβ5 in glioma is minor compared with that of αvβ3.

The epithelial fibronectin receptor αvβ6 is upregulated during wound healing and carcinoma progression and can modulate matrix metalloproteinases and transforming growth factor-β1 activation (7). Expression has been previously reported in gastrointestinal, pancreatic, and esophagus adenocarcinomas (9). Upregulation of αvβ6 in non-small-cell lung carcinoma correlates with shorter survival (57). There was no αvβ6 immunoreactivity in our glioma series or in normal brain, although the EM05201 antibody strongly stained controls (35) (Fig. 2D), which we confirmed on more than 100 carcinoma samples (data not shown). Because integrin αvβ8 was expressed in most gliomas (∼95%),, regardless of WHO stage, a combined αvβ8-positive/αvβ6-negative immunoprofile might help determine whether a tumor is of glial origin—when the glial fibrillary acidic protein immunophenotype in late-stage dedifferentiated GBM is equivocal (58). Integrin αvβ6 has not been reported in the brain, and it is classically an epithelial integrin that is upregulated in carcinoma (59). Therefore, our confirmation of its absence in glioma and normal brain is not surprising, but the brain phenotype of αvβ8 compared with the epithelial phenotype of αvβ6 is notable particularly because both can assist activation of transforming growth factor-β1 (60).

Expression of αvβ8 integrin has been reported for human glioma cell lines (61), in neuronal progenitor cells (61), and in neuronal and glial cells in rodent brain (62, 63). In EGFR/PDGFRA-overexpressing gliomas, αvβ8 is upregulated, often in a perinecrotic or perivascular pattern (64). Parenchymal αvβ8 plays a role in vascular development and in the pathogenesis of brain arteriovenous malformations (65, 66). Here, automated analysis of αvβ8 showed a transition from weak to strong expression in tumor cells with increasing grade of malignancy. These tumor cells were often perivascular. We further corroborated the reported high expression of β8 mRNA in primary GBM (64) using the REMBRANDT gene data set. Because, in the postnatal brain, αvβ8 integrin has been reported in neural stem cells and in progenitor cells (67), the consistent β8 expression in glioma was possibly caused by reactivation of a glioneuronal phenotype. The lack of αvβ8 in human tumor vessels parallels the lack of αvβ8 in rodents (63). In none of these studies were the reagents capable of detecting intact αvβ8 complexes, however. This complicates the interpretation of the data sets, unlike the unequivocal staining of our αvβ8 antibody (EM13309), which binds the intact extracellular domains of the integrin αvβ8 heterodimer. There is evidence that the tumor vascular pathology and hemorrhage seen in high-grade gliomas may be related to loss of parenchymal αvβ8 (61, 63, 65). Inasmuch as we observed increased αvβ8 expression in tumor cells in GBM, αvβ8 either overexpression or underexpression may affect vessel formation. Our observations of αvβ8 expressed in tumors derived from astrocytes apparently conflict with some cellular data on the expression of αvβ8 integrin. Using the antibodies used here, Tchaicha et al and one of us found little αvβ8 on U87 glioma (35, 60, 61). Some GBM lines clearly can express αvβ8, which may reflect their invasive capability (61). It has also been reported that αvβ8 in U87 glioma can be suppressed by miRNA93 transfection (68). Because expression of miRNA93 is also significantly different in GBM compared with U87MG cells (69), this might also affect αvβ8 expression in the cell line. Furthermore, cell lines can switch their protein expression profile during adaptation to culture conditions, whereas IHC reveals the primary status of the tumor in situ. Provided that the antibodies are well characterized and the staining conditions are well validated, IHC data on expression in situ should take precedence over data derived from cell lines in vitro. In addition, the models used by Tchaicha et al (61) lack the typical pseudopalisading necrosis and glial fibrillary acidic protein expression of human GBM and were genetically more closely related to secondary GBM. This may indicate different roles for αvβ8 in different tumor subpopulations that need further study. Finally, αvβ8 data from cell lines need to be interpreted cautiously because the origin of several “glial” lines has now been challenged (70).

In conclusion, we used newly developed rabbit monoclonal antibodies for fast automated histopathologic determination of αv integrin complexes in long-term archival glioma specimens. Overexpression of αvβ3 integrin overtly signifies a poor prognosis in glioma and might help in stratifying patients for anti-integrin treatment by IHC, whereas other integrins of the αv series do not. Finally, the detailed semi-automated quantification of parenchymal αv expression in paraffin blocks will allow us to easily correlate integrin expression data from individual patients with their therapeutic response to integrin inhibitor treatment.

Acknowledgements

The authors thank Katrin Trautmann and Karen Petersen for help with additional immunostaining. Research antibodies EM227-03, EM002-12, EM099-02, EM052-01, EM133-09, and EM013-09 were kindly provided by Merck KGaA, Darmstadt, Germany.

References

1.
Arnaout
MA
Goodman
SL
Xiong
JP
.
Structure and mechanics of integrin-based cell adhesion
.
Curr Opin Cell Biol
 
2007
;
19
:
495
507
2.
Hynes
RO
.
Integrins: Bidirectional, allosteric signaling machines
.
Cell
 
2002
;
110
:
673
87
3.
Wang
W
Luo
BH
.
Structural basis of integrin transmembrane activation
.
J Cell Biochem
 
2010
;
109
:
447
52
4.
Worthington
JJ
Klementowicz
JE
Travis
MA
.
TGFA: A sleeping giant awoken by integrins
.
Trends Biochem Sci
 
2011
;
36
:
47
54
5.
Switala-Jelen
K
Dabrowska
K
Opolski
A
et al
.
The biological functions of beta3 integrins
.
Folia Biol
 
2004
;
50
:
143
52
6.
Tabatabai
G
Tonn
JC
Stupp
R
et al
.
The role of integrins in glioma biology and anti-glioma therapies
.
Curr Pharm Des
 
2011
;
7
:
2402
10
7.
Bandyopadhyay
A
Raghavan
S
.
Defining the role of integrin alphavbeta6 in cancer
.
Curr Drug Targets
 
2009
;
10
:
645
52
8.
Rathinam
R
Alahari
SK
.
Important role of integrins in the cancer biology
.
Cancer Metastasis Rev
 
2010
;
29
:
223
37
9.
Sipos
B
Hahn
D
Carceller
A
et al
.
Immunohistochemical screening for beta6-integrin subunit expression in adenocarcinomas using a novel monoclonal antibody reveals strong upregulation in pancreatic ductal adenocarcinomas in vivo and in vitro
.
Histopathology
 
2004
;
45
:
226
36
10.
Max
R
Gerritsen
RR
Nooijen
PT
et al
.
Immunohistochemical analysis of integrin alpha vbeta3 expression on tumor-associated vessels of human carcinomas
.
Int J Cancer
 
1997
;
71
:
320
4
11.
Marshall
JF
Hart
IR
.
The role of alpha v-integrins in tumour progression and metastasis
.
Semin Cancer Biol
 
1996
;
7
:
129
38
12.
Desgrosellier
JS
Cheresh
DA
.
Integrins in cancer: Biological implications and therapeutic opportunities
.
Nat Rev Cancer
 
2010
;
10
:
9
22
13.
Taverna
D
Moher
H
Crowley
D
et al
.
Increased primary tumor growth in mice null for beta3- or beta3/beta5-integrins or selectins
.
Proc Natl Acad Sci U S A
 
2004
;
101
:
763
8
14.
Hodivala-Dilke
K
.
alphavbeta3 integrin and angiogenesis: a moody integrin in a changing environment
.
Curr Opin Cell Biol
 
2008
;
20
:
514
19
15.
Bello
L
Francolini
M
Marthyn
P
et al
.
Alpha(v)beta3 and alpha(v)beta5 integrin expression in glioma periphery
.
Neurosurgery
 
2001
;
49
:
380
9
16.
Gladson
CL
.
Expression of integrin alpha v beta 3 in small blood vessels of glioblastoma tumors
.
J Neuropathol Exp Neurol
 
1996
;
55
:
1143
9
17.
Yan
W
Qian
C
Zhao
P
et al
.
Expression pattern of osteopontin splice variants and its functions on cell apoptosis and invasion in glioma cells
.
Neuro Oncol
 
2010
;
12
:
765
75
18.
Miyauchi
A
Alvarez
J
Greenfield
EM
et al
.
Recognition of osteopontin and related peptides by an αvβ3 integrin stimulates immediate cell signals in osteoclasts
.
J Biol Chem
 
1991
;
266
:
20369
74
19.
Mizejewski
GJ
.
Role of integrins in cancer: Survey of expression patterns
.
Proc Soc Exp Biol Med
 
1999
;
222
:
124
38
20.
Schnell
O
Krebs
B
Wagner
E
et al
.
Expression of integrin alphavbeta3 in gliomas correlates with tumor grade and is not restricted to tumor vasculature
.
Brain Pathol
 
2008
;
18
:
378
86
21.
Gladson
CL
Cheresh
DA
.
Glioblastoma expression of vitronectin and the alpha v beta 3 integrin
.
Adhesion mechanism for transformed glial cells. J Clin Invest
 
1991
;
88
:
1924
32
22.
Cox
D
Brennan
M
Moran
N
.
Integrins as therapeutic targets: Lessons and opportunities
.
Nat Rev Drug Discov
 
2010
;
9
:
804
20
23.
Goodman
SL
Picard
M
.
Integrins as therapeutic targets
.
Trends Pharmacol Sci
 
2012
;
33
:
405
12
24.
Mikkelsen
T
Brodie
C
Finniss
S
et al
.
Radiation sensitization of glioblastoma by cilengitide has unanticipated schedule-dependency
.
Int J Cancer
 
2009
;
124
:
2719
27
25.
MacDonald
TJ
Taga
T
Shimada
H
et al
.
Preferential susceptibility of brain tumors to the antiangiogenic effects of an alpha(v) integrin antagonist
.
Neurosurgery
 
2001
;
48
:
151
7
26.
Bäuerle
T
Komljenovic
D
Merz
M
et al
.
Cilengitide inhibits progression of experimental breast cancer bone metastases as imaged noninvasively using VCT, MRI and DCE-MRI in a longitudinal in vivo study
.
Int J Cancer
 
2011
;
128
:
2453
62
27.
Rüegg
C
Alghisi
GC
.
Vascular integrins: Therapeutic and imaging targets of tumor angiogenesis
.
Recent Results Cancer Res
 
2010
;
180
:
83
101
28.
Abdollahi
A
Griggs
DW
Zieher
H
et al
.
Inhibition of alpha(v)beta3 integrin survival signaling enhances antiangiogenic and antitumor effects of radiotherapy
.
Clin Cancer Res
 
2005
;
11
:
6270
9
29.
CBTRUS Central Brain Tumor Registry of United States
.
Statistical Report Table
 .
2011
. Available at: http://www.cbtrus.org/2011-NPCR-SEER/WEB-0407-Report-3-3-2011.pdf. Accessed September 24, 2012
30.
Ohgaki
H
Kleihues
P
.
Genetic pathways to primary and secondary glioblastoma
.
Am J Pathol
 
2007
;
170
:
1445
53
31.
Hartmann
C
Meyer
J
Balss
J
et al
.
Type and frequency of IDH1 and IDH2 mutations are related to astrocytic and oligodendroglial differentiation and age: A study of 1,010 diffuse gliomas
.
Acta Neuropathol
 
2009
;
118
:
469
74
32.
Schittenhelm
J
Mittelbronn
M
Meyermann
R
et al
.
Confirmation of R132H mutation of isocitrate dehydrogenase 1 as an independent prognostic factor in anaplastic astrocytoma
.
Acta Neuropathol
 
2011
;
122
:
651
2
33.
Hegi
ME
Diserens
AC
Gorlia
T
et al
.
MGMT gene silencing and benefit from temozolomide in glioblastoma
.
N Engl J Med
 
2005
;
352
:
997
1003
34.
Schnell
O
Krebs
B
Carlsen
J
et al
.
Imaging of integrin alphavbeta3 expression in patients with malignant glioma by [18F]galacto-RGD positron emission tomography
.
Neuro Oncol
 
2009
;
11
:
861
70
35.
Goodman
SL
Grote
HJ
Wilm
C
.
Matched rabbit monoclonal antibodies against αv-series integrins reveal a novel αvβ3-LIBS epitope, and permit routine staining of archival paraffin samples of human tumors
.
Biol Open
 
2012
;
1
:
329
40
36.
Louis
DN
Ohgaki
H
Wiestler
OD
, eds.
World Health Organization Classification of Tumours. Pathology and Genetics of Tumours of the Nervous System
 .
Lyon, France
:
IARC Press
; 2007
37.
Capper
D
Sahm
F
Hartmann
C
et al
.
Application of mutant IDH1 antibody to differentiate diffuse glioma from nonneoplastic central nervous system lesions and therapy-induced changes
.
Am J Surg Pathol
 
2010
;
34
:
1199
204
38.
Martinez
R
Schackert
G
Yaya-Tur
R
et al
.
Frequent hypermethylation of the DNA repair gene MGMT in long-term survivors of glioblastoma multiforme
.
J Neurooncol
 
2007
;
83
:
91
93
39.
Li
A
Walling
J
Ahn
S
et al
.
Unsupervised analysis of transcriptomic profiles reveals six glioma subtypes
.
Cancer Res
 
2009
;
69
:
2091
99
40.
Madhavan
S
Zenklusen
JC
Kotliarov
Y
et al
.
REMBRANDT: Helping personalized medicine become a reality through integrative translational research
.
Mol Cancer Res
 
2009
;
7
:
157
67
41.
Dimitropoulos
K
Giannopoulou
E
Argyriou
AA
et al
.
The effects of anti-VEGFR and anti-EGFR agents on glioma cell migration through implication of growth factors with integrins
.
Anticancer Res
 
2010
;
30
:
4987
92
42.
Reardon
DA
Neyns
B
Weller
M
et al
.
Cilengitide: An RGD pentapeptide αvβ3 and αvβ5 integrin inhibitor in development for glioblastoma and other malignancies
.
Future Oncol
 
2011
;
7
:
339
54
43.
Tabatabai
G
Weller
M
Nabors
B
et al
.
Targeting integrins in malignant glioma
.
Target Oncol
 
2010
;
5
:
175
81
44.
Dechantsreiter
M
Planker
E
Mathä
B
et al
.
N-methylated cyclic RGD peptides as highly active and selective alphavbeta 3 integrin antagonists
.
J Med Chem
 
1999
;
42
:
3033
40
45.
Baatz
M
Zimmermann
J
Blackmore
CG
.
Automated analysis and detailed quantification of biomedical images using Definiens Cognition Network Technology
.
Comb Chem High Throughput Screen
 
2009
;
12
:
908
16
46.
Takada
Y
Ye
X
Simon
S
.
The integrins
.
Genome Biol
 
2007
;
8
:
215
47.
Kim
JH
Zheng
LT
Lee
WH
et al
.
Proapoptotic role of integrin β3 in glioma cells
.
J Neurochem
 
2011
;
117
:
494
503
48.
Soldi
R
Mitola
S
Strasly
M
et al
.
Role of alphavbeta3 integrin in the activation of vascular endothelial growth factor receptor-2
.
EMBO J
 
1999
;
18
:
882
92
49.
Woodard
AS
García-Cardeña
G
Leong
M
et al
.
The synergistic activity of alphavbeta3 integrin and PDGF receptor increases cell migration
.
J Cell Sci
 
1998
;
111
:
469
78
50.
Stupp
R
Hegi
ME
Neyns
B
et al
.
Phase I/IIa study of cilengitide and temozolomide with concomitant radiotherapy followed by cilengitide and temozolomide maintenance therapy in patients with newly diagnosed glioblastoma
.
J Clin Oncol
 
2010
;
28
:
2712
18
51.
Nabors
LB
Mikkelsen
T
Hegi
ME
.
A safety run-in and randomized phase 2 study of cilengitide combined with chemoradiation for newly diagnosed glioblastoma (NABTT 0306)
.
Cancer
 
2012
;
118
:
5601
7
52.
Coene
E
Schelfhout
AM
De Ridder
L
et al
.
Generation of a monoclonal antibody directed against a human cell substrate adhesion molecule and the expression of the antigen in human tissues
.
Hybridoma
 
1997
;
16
:
77
83
53.
Fabricius
EM
Wildner
GP
Kruse-Boitschenko
U
et al
.
Immunohistochemical analysis of integrins αvβ3, αvβ5 and α5β1 and their ligands, fibrinogen, fibronectin, osteopontin and vitronectin in frozen sections of human oral head and neck squamous cell carcinomas
.
Exp Ther Med
 
2011
;
2
:
9
19
54.
Ricono
JM
Huang
M
Barnes
LA
et al
.
Specific cross-talk between epidermal growth factor receptor and integrin alphavbeta5 promotes carcinoma cell invasion and metastasis
.
Cancer Res
 
2009
;
69
:
1383
91
55.
Maurer
GD
Tritschler
I
Adams
B
et al
.
Cilengitide modulates attachment and viability of human glioma cells but not sensitivity to irradiation or temozolomide in vitro
.
Neuro Oncol
 
2009
;
11
:
747
56
56.
Skuli
N
Monferran
S
Delmas
C
et al
.
Alphavbeta3/alphavbeta5 integrins-FAK-RhoB: A novel pathway for hypoxia regulation in glioblastoma
.
Cancer Res
 
2009
;
69
:
3308
16
57.
Caccavari
F
Valdembri
D
Sandri
C
et al
.
Integrin signaling and lung cancer
.
Cell Adh Migr
 
2010
;
4
:
124
9
58.
Lolait
SJ
Harmer
JH
Auteri
G
et al
.
Expression of glial fibrillary acidic protein, actin, fibronectin and factor VIII antigen in human astrocytomas
.
Pathology
 
1983
;
15
:
373
8
59.
Sheppard
D
.
Epithelial integrins
.
Bioessays
 
1006
;
18
:
655
60
60.
Tchaicha
JH
Reyes
SB
Shin
J
et al
.
Glioblastoma angiogenesis and tumor cell invasiveness are differentially regulated by β8 integrin
.
Cancer Res
 
2011
;
71
:
6371
81
61.
Tchaicha
JH
Mobley
AK
Hossain
MG
et al
.
A mosaic mouse model of astrocytoma identifies alphavbeta8 integrin as a negative regulator of tumor angiogenesis
.
Oncogene
 
2010
;
29
:
4460
72
62.
Nishimura
SL
Boylen
KP
Einheber
S
et al
.
Synaptic and glial localization of the integrin alphavbeta8 in mouse and rat brain
.
Brain Res
 
1998
;
791
:
271
82
63.
McCarty
JH
Lacy-Hulbert
A
Charest
A
et al
.
Selective ablation of alphav integrins in the central nervous system leads to cerebral hemorrhage, seizures, axonal degeneration and premature death
.
Development
 
2005
;
132
:
165
76
64.
Riemenschneider
MJ
Mueller
W
Betensky
RA
et al
.
In situ analysis of integrin and growth factor receptor signaling pathways in human glioblastomas suggests overlapping relationships with focal adhesion kinase activation
.
Am J Pathol
 
2005
;
167
:
1379
87
65.
Proctor
JM
Zang
K
Wang
D
et al
.
Vascular development of the brain requires beta8 integrin expression in the neuroepithelium
.
J Neurosci
 
2005
;
25
:
9940
8
66.
Su
H
Kim
H
Pawlikowska
L
et al
.
Reduced expression of integrin alphavbeta8 is associated with brain arteriovenous malformation pathogenesis
.
Am J Pathol
 
2010
;
176
:
1018
27
67.
Mobley
AK
Tchaicha
JH
Shin
J
et al
.
Beta8 integrin regulates neurogenesis and neurovascular homeostasis in the adult brain
.
J Cell Sci
 
2009
;
122
:
1842
51
68.
Fang
L
Deng
Z
Shatseva
T
et al
.
MicroRNA miR-93 promotes tumor growth and angiogenesis by targeting integrin-β8
.
Oncogene
 
2011
;
30
:
806
21
69.
Costa
PM
Cardoso
AL
Pereira de Almeida
LF
et al
.
PDGF-β-mediated downregulation of miR-21: New insights into PDGF signaling in glioblastoma
.
Hum Mol Genet
 
2012
,
21
:
5118
30
70.
Ishii
N
Maier
D
Merlo
A
et al
.
Frequent co-alterations of TP53, p16/CDKN2A, p14ARF, PTEN tumor suppressor genes in human glioma cell lines
.
Brain Pathol
 
1999
;
9
:
469
79

Author notes

Jens Schittenhelm is supported by a grant from the Ludwig-Hiermaier Foundation for Applied Cancer Research, Tübingen, Germany. This study was funded in part by Merck KGaA. Merck KGaA did not influence the selection of the patients, evaluation and acquisition of data, or the academic interpretation of the data set.