The Tumor Microenvironment Strongly Impacts Master Transcriptional Regulators and Gene Expression Class of Glioblastoma
2012; Elsevier BV; Volume: 180; Issue: 5 Linguagem: Inglês
10.1016/j.ajpath.2012.01.040
ISSN1525-2191
AutoresLee Cooper, David A. Gutman, C Chisolm, Christina Appin, Jun Kong, Yuan Rong, Tahsin Kurç, Erwin G. Van Meir, Joel Saltz, Carlos S. Moreno, Daniel J. Brat,
Tópico(s)MicroRNA in disease regulation
ResumoThe Cancer Genome Atlas (TCGA) project has generated gene expression data that divides glioblastoma (GBM) into four transcriptional classes: proneural, neural, classical, and mesenchymal. Because transcriptional class is only partially explained by underlying genomic alterations, we hypothesize that the tumor microenvironment may also have an impact. In this study, we focused on necrosis and angiogenesis because their presence is both prognostically and biologically significant. These features were quantified in digitized histological images of TCGA GBM frozen section slides that were immediately adjacent to samples used for molecular analysis. Correlating these features with transcriptional data, we found that the mesenchymal transcriptional class was significantly enriched with GBM samples that contained a high degree of necrosis. Furthermore, among 2422 genes that correlated with the degree of necrosis in GBMs, transcription factors known to drive the mesenchymal expression class were most closely related, including C/EBP-β, C/EBP-δ, STAT3, FOSL2, bHLHE40, and RUNX1. Non-mesenchymal GBMs in the TCGA data set were found to become more transcriptionally similar to the mesenchymal class with increasing levels of necrosis. In addition, high expression levels of the master mesenchymal factors C/EBP-β, C/EBP-δ, and STAT3 were associated with a poor prognosis. Strong, specific expression of C/EBP-β and C/EBP-δ by hypoxic, perinecrotic cells in GBM likely account for their tight association with necrosis and may be related to their poor prognosis. The Cancer Genome Atlas (TCGA) project has generated gene expression data that divides glioblastoma (GBM) into four transcriptional classes: proneural, neural, classical, and mesenchymal. Because transcriptional class is only partially explained by underlying genomic alterations, we hypothesize that the tumor microenvironment may also have an impact. In this study, we focused on necrosis and angiogenesis because their presence is both prognostically and biologically significant. These features were quantified in digitized histological images of TCGA GBM frozen section slides that were immediately adjacent to samples used for molecular analysis. Correlating these features with transcriptional data, we found that the mesenchymal transcriptional class was significantly enriched with GBM samples that contained a high degree of necrosis. Furthermore, among 2422 genes that correlated with the degree of necrosis in GBMs, transcription factors known to drive the mesenchymal expression class were most closely related, including C/EBP-β, C/EBP-δ, STAT3, FOSL2, bHLHE40, and RUNX1. Non-mesenchymal GBMs in the TCGA data set were found to become more transcriptionally similar to the mesenchymal class with increasing levels of necrosis. In addition, high expression levels of the master mesenchymal factors C/EBP-β, C/EBP-δ, and STAT3 were associated with a poor prognosis. Strong, specific expression of C/EBP-β and C/EBP-δ by hypoxic, perinecrotic cells in GBM likely account for their tight association with necrosis and may be related to their poor prognosis. See related Commentary on page 1768 See related Commentary on page 1768 Glioblastoma (GBM) (World Health Organization, grade IV) is the most common and highest grade astrocytoma.1Louis D.N. Ohgaki H. Wiestler O.D. Cavenee W.K. WHO classification of tumours of the central nervous system.4th ed. Intl. Agency for Research, Lyon2007Google Scholar, 2CBTRUSCBTRUS statistical report: primary brain and central nervous system tumors in the United States in 2004–2006. Central Brain Tumor Registry of the United States, Hinsdale, IL2010Google Scholar Currently incurable, it has a mean survival that only slightly exceeds 1 year following standard surgical and adjuvant therapies.3Stupp R. Mason W.P. van den Bent M.J. Weller M. Fisher B. Taphoorn M.J. Belanger K. Brandes A.A. Marosi C. Bogdahn U. Curschmann J. Janzer R.C. Ludwin S.K. Gorlia T. Allgeier A. Lacombe D. Cairncross J.G. Eisenhauer E. Mirimanoff R.O. Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma.N Engl J Med. 2005; 352: 987-996Crossref PubMed Scopus (12735) Google Scholar Analyses of large scale gene expression and genomic datasets have indicated that this disease represents multiple molecular subclasses, raising the possibility that future therapies could be directed at underlying class-specific mechanisms. Phillips et al4Phillips H.S. Kharbanda S. Chen R. Forrest W.F. Soriano R.H. Wu T.D. Misra A. Nigro J.M. Colman H. Soroceanu L. Williams P.M. Modrusan Z. Feuerstein B.G. Aldape K. Molecular subclasses of high-grade glioma predict prognosis, delineate a pattern of disease progression, and resemble stages in neurogenesis.Cancer Cell. 2006; 9: 157-173Abstract Full Text Full Text PDF PubMed Scopus (2114) Google Scholar and Verhaak et al5Verhaak R.G. Hoadley K.A. Purdom E. Wang V. Qi Y. Wilkerson M.D. Miller C.R. Ding L. Golub T. Mesirov J.P. Alexe G. Lawrence M. O'Kelly M. Tamayo P. Weir B.A. Gabriel S. Winckler W. Gupta S. Jakkula L. Feiler H.S. Hodgson J.G. James C.D. Sarkaria J.N. Brennan C. Kahn A. Spellman P.T. Wilson R.K. Speed T.P. Gray J.W. Meyerson M. Getz G. Perou C.M. Hayes D.N. Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA. IDH1, EGFR, and NF1.Cancer Cell. 2010; 17: 98-110Abstract Full Text Full Text PDF PubMed Scopus (4325) Google Scholar have each shown that unsupervised clustering of GBM gene expression profiles results in three or four distinct transcriptional classes. Epigenetic changes and genetic alterations, including mutations, amplifications, and deletions of established tumor suppressors and oncogenes, account for at least some transcriptional class identity of GBM. For example, among The Cancer Genome Atlas (TCGA) tumors, which contain proneural, neural, classical, and mesenchymal transcriptional classes, IDH1 mutations and the CpG island methylator phenotype (G-CIMP+) are seen almost exclusively in the proneural transcriptional class, whereas nearly all tumors with NF1 mutations or deletions are within the mesenchymal class.5Verhaak R.G. Hoadley K.A. Purdom E. Wang V. Qi Y. Wilkerson M.D. Miller C.R. Ding L. Golub T. Mesirov J.P. Alexe G. Lawrence M. O'Kelly M. Tamayo P. Weir B.A. Gabriel S. Winckler W. Gupta S. Jakkula L. Feiler H.S. Hodgson J.G. James C.D. Sarkaria J.N. Brennan C. Kahn A. Spellman P.T. Wilson R.K. Speed T.P. Gray J.W. Meyerson M. Getz G. Perou C.M. Hayes D.N. Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA. IDH1, EGFR, and NF1.Cancer Cell. 2010; 17: 98-110Abstract Full Text Full Text PDF PubMed Scopus (4325) Google Scholar, 6Cancer Genome Atlas NetworkComprehensive genomic characterization defines human glioblastoma genes and core pathways.Nature. 2008; 455: 1061-1068Crossref PubMed Scopus (5223) Google Scholar, 7Noushmehr H. Weisenberger D.J. Diefes K. Phillips H.S. Pujara K. Berman B.P. Pan F. Pelloski C.E. Sulman E.P. Bhat K.P. Verhaak R.G. Hoadley K.A. Hayes D.N. Perou C.M. Schmidt H.K. Ding L. Wilson R.K. Van Den Berg D. Shen H. Bengtsson H. Neuvial P. Cope L.M. Buckley J. Herman J.G. Baylin S.B. Laird P.W. Aldape K. Identification of a CpG island methylator phenotype that defines a distinct subgroup of glioma.Cancer Cell. 2010; 17: 510-522Abstract Full Text Full Text PDF PubMed Scopus (1601) Google Scholar However, some of the best characterized genomic alterations in GBM, including TP53 and PTEN mutation, EGFR and PDGFR amplification and CDKN2A deletion are noted in multiple transcriptional classes, indicating that cell-intrinsic genetic defects only partially explain the class-specific gene expression patterns. A recent analysis of GBM expression classes used a novel algorithm to reconstruct transcriptional interactions and uncovered a small set of transcription factors that regulate the transition to the mesenchymal class, including C/EBP-β, C/EBP-δ, STAT3, FOSL2, bHLHE40, and RUNX1. Among these, C/EBP-β, C/EBP-δ, and STAT3 were found to be master transcriptional regulators, controlling the expression of other key regulators, and accounting for the majority of downstream signaling events and the mesenchymal gene signature.8Carro M.S. Lim W.K. Alvarez M.J. Bollo R.J. Zhao X. Snyder E.Y. Sulman E.P. Anne S.L. Doetsch F. Colman H. Lasorella A. Aldape K. Califano A. Iavarone A. The transcriptional network for mesenchymal transformation of brain tumours.Nature. 2010; 463: 318-325Crossref PubMed Scopus (821) Google Scholar Underlying genetic alterations or pathophysiological triggers of these master transcriptional regulators were not uncovered. It remains possible that elements of the tumor microenvironment, including tumor hypoxia, necrosis, angiogenesis, or inflammatory cell infiltrates, could strongly impact both transcriptional regulators and gene expression class. Microenvironmental contributions to expression class, as well as the tissue sampling considerations that are intimately related, will need to be carefully considered as molecular profiles are used to direct therapies. To address these issues, we performed an integrated morphological and molecular analysis of microenvironmental factors as they relate to GBM transcriptional class. We analyzed gene expression and genetic correlates of angiogenesis and necrosis in GBM using molecular data and the digitized images from corresponding frozen sections used for quality assurance by the TCGA. We found that the mesenchymal class of GBM was enriched with samples displaying a high degree of necrosis, and that the expression of transcriptional regulators of the mesenchymal transition, C/EBP-β, C/EBP-δ, STAT3, FOSL2, bHLHE40, and RUNX1, were tightly correlated with the extent of necrosis. Nonmesenchymal GBMs became more transcriptionally similar to the mesenchymal class with increasing levels of necrosis. Using human GBM tissue sections, we demonstrated that C/EBP-β and CEBP-δ were specifically expressed by hypoxic, peri-necrotic pseudopalisading cells, accounting for the association of these factors with necrosis. Our finding that the high expression of C/EBP-β, C/EBP-δ, and STAT3 portends a poor prognosis suggests that these key signaling nodes may hold potential for targeted therapies. A complete description of samples, and related imaging, molecular, and pathology data are provided in Supplemental Tables 1–6 (available at http://ajp.amjpathol.org). Whole-slide digitized images of GBM frozen sections were obtained from the TCGA portal (http://tcga-data.nci.nih.gov/tcga/tcgaHome2.jsp; last accessed June 6, 2011). Frozen section slides were scanned and digitized at 20× resolution on an Aperio scanner at the TCGA Biospecimen Core Resource located at the International Genomics Consortium (Intgen, Phoenix, AZ). The frozen section slides were those reviewed by Intgen contracted neuropathologists for quality assurance to ensure that the tissue between the top section and bottom section were adequate for molecular studies. Importantly, the tissue used for molecular analysis by TCGA was immediately adjacent to the tissue used to create slides for this investigation. We used these digitized images as our primary source of data and annotated necrosis on 177 slides from 99 samples corresponding to 91 patients. Angiogenesis was annotated on 168 slides from 95 samples corresponding to 88 patients. As a secondary source of data on necrosis in GBM samples, we used semi-quantitative annotations of 293 GBMs that were recorded by TCGA neuropathologists after reviewing frozen section slides for quality assurance. For these images, the percent necrosis was recorded as a visual estimate of total tissue area involved. Angiogenesis was recorded only as "present" or "absent" and was not useful in our analysis. There were 70 samples with overlap between the two sources of data. Regions of necrosis and angiogenesis were manually outlined using ImageScope software (Aperio, Vista CA) by two pathologists working together and reaching a consensus on the features to include (CA, DJB).9Cooper L.A. Kong J. Gutman D.A. Wang F. Cholleti S.R. Pan T.C. Widener P.M. Sharma A. Mikkelsen T. Flanders A.E. Rubin D.L. Van Meir E.G. Kurc T.M. Moreno C.S. Brat D.J. Saltz J.H. An integrative approach for in silico glioma research.IEEE Trans Biomed Eng. 2010; 57: 2617-2621Crossref PubMed Scopus (39) Google Scholar The boundaries of tissue sections were outlined to calculate total tissue area. Duplicate adjacent sections within the same slide were not analyzed. Regions identified as either necrosis or angiogenesis were exhaustively outlined within the tissue boundaries (Figure 1). Angiogenic regions were identified as those vascular regions departing from normal, and exhibiting characteristics of cellular hypertrophy, cellular hyperplasia, or microvascular proliferation.1Louis D.N. Ohgaki H. Wiestler O.D. Cavenee W.K. WHO classification of tumours of the central nervous system.4th ed. Intl. Agency for Research, Lyon2007Google Scholar, 10Perry A. Brat D.J. Practical surgical pathology: a diagnostic approach. Elsevier, Philadelphia2010Google Scholar, 11Brat D.J. Prayson R.A. Ryken T.C. Olson J.J. Diagnosis of malignant glioma: role of neuropathology.J Neurooncol. 2008; 89: 287-311Crossref PubMed Scopus (67) Google Scholar In addition to the endothelial compartment, the perivascular cells, including pericytes, fibroblasts, and any inflammatory infiltrates, were included within marked angiogenic regions. Luminal areas were automatically subtracted from angiogenic regions using computer-based color segmentation. The extent of necrosis and the extent of angiogenesis were calculated as a percentage of whole tissue by taking the ratios of necrosis or angiogenesis surface areas to the total tissue section area. Areas were summed over multiple slides before ratio calculation to determine per-sample and per-patient percentages. Transcriptional class labels for TCGA patients were obtained from the TCGA Advanced Working Group. This labeling extends the original set of samples labeled by Verhaak et al5Verhaak R.G. Hoadley K.A. Purdom E. Wang V. Qi Y. Wilkerson M.D. Miller C.R. Ding L. Golub T. Mesirov J.P. Alexe G. Lawrence M. O'Kelly M. Tamayo P. Weir B.A. Gabriel S. Winckler W. Gupta S. Jakkula L. Feiler H.S. Hodgson J.G. James C.D. Sarkaria J.N. Brennan C. Kahn A. Spellman P.T. Wilson R.K. Speed T.P. Gray J.W. Meyerson M. Getz G. Perou C.M. Hayes D.N. Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA. IDH1, EGFR, and NF1.Cancer Cell. 2010; 17: 98-110Abstract Full Text Full Text PDF PubMed Scopus (4325) Google Scholar using Affymetrix HT_HG-U133A data to classify previously unlabeled TCGA samples given centroids derived from the original labeled set. Patient-reduced percentages for necrosis and angiogenesis were tested for association with transcriptional class with one-way analysis of variance tests. Copy number data were obtained from the GBM Pathway analysis at the Memorial Sloan Kettering Cancer Genomics Portal (http://www.cbioportal.org/public-portal; last accessed June 6, 2011). Level 3 sequence data obtained from the TCGA portal was filtered to remove silent mutations. Necrosis was compared between amplified/normal, deleted/normal, and mutant/wild-type samples using two-way t-tests. Level 2 robust multichip average normalized gene expression data from the Affymetrix U133A platform was averaged over samples with multiple arrays to create patient-reduced expression profiles. Probes with an unlogged expression range of <20 and a fold change of <1.5 were removed. Percentage necrosis and angiogenesis were correlated with expression profiles using the Cox proportional hazards within the significance analysis of microarray procedure.12Tusher V.G. Tibshirani R. Chu G. Significance analysis of microarrays applied to the ionizing radiation response.Proc Natl Acad Sci USA. 2011; 98: 5116-5121Crossref Scopus (9446) Google Scholar Gene-centric profiles of C/EBP-β, C/EBP-δ, and STAT3 were obtained by averaging over the corresponding probes. Methylation phenotype G-CIMP status was calculated, as previously described.7Noushmehr H. Weisenberger D.J. Diefes K. Phillips H.S. Pujara K. Berman B.P. Pan F. Pelloski C.E. Sulman E.P. Bhat K.P. Verhaak R.G. Hoadley K.A. Hayes D.N. Perou C.M. Schmidt H.K. Ding L. Wilson R.K. Van Den Berg D. Shen H. Bengtsson H. Neuvial P. Cope L.M. Buckley J. Herman J.G. Baylin S.B. Laird P.W. Aldape K. Identification of a CpG island methylator phenotype that defines a distinct subgroup of glioma.Cancer Cell. 2010; 17: 510-522Abstract Full Text Full Text PDF PubMed Scopus (1601) Google Scholar, 13Cooper L.A. Gutman D.A. Long Q. Johnson B.A. Cholleti S.R. Kurc T. Saltz J.H. Brat D.J. Moreno C.S. The proneural molecular signature is enriched in oligodendrogliomas and predicts improved survival among diffuse gliomas.PLoS ONE. 2010; 5: e12548Crossref PubMed Scopus (111) Google Scholar Ingenuity Pathway Analysis (IPA) (Ingenuity Systems, Redwood City, CA) was applied to gene lists generated by significance analysis of microarray analysis of necrosis and angiogenesis, as previously described.13Cooper L.A. Gutman D.A. Long Q. Johnson B.A. Cholleti S.R. Kurc T. Saltz J.H. Brat D.J. Moreno C.S. The proneural molecular signature is enriched in oligodendrogliomas and predicts improved survival among diffuse gliomas.PLoS ONE. 2010; 5: e12548Crossref PubMed Scopus (111) Google Scholar Both direct and indirect relationships were included. Data sources were restricted to human species and cell lines. Network significance was assigned by hypergeometric scoring. We obtained TCGA neuropathologists ratings of pathological categories for 112 GBM samples with either gene expression or transcriptional class assignments (http://tcga-data.nci.nih.gov/tcga/tcgaHome2.jsp; last accessed June 6, 2011). Association between transcriptional class and the presence of macrophages (0, 1+, or 2+) was performed using Fisher's test. Expression differences for C/EBP-β, C/EBP-δ, and STAT3 were examined using two-way t-tests with unequal variance. The mesenchymal signature was calculated using the classification to nearest centroid procedure.14Dabney A.R. Classification of microarrays to nearest centroids.Bioinformatics. 2005; 21: 4148-4154Crossref PubMed Scopus (94) Google Scholar A set of 250 probes that distinguish mesenchymal samples were selected by ranking t-statistics computed by comparing sample-reduced expression profiles of mesenchymal and nonmesenchymal samples. The mesenchymal signature was calculated as the average expression of these 250 probes of all mesenchymal sample profiles. Transcriptional distance of each nonmesenchymal sample to the mesenchymal signature was calculated as the Euclidean distance normalized by the SD of each probe. Correlation between transcriptional distance and percent necrosis were calculated using Spearman rank correlation. Ten archived surgically resected GBM specimens were retrieved from Emory University Hospital Department of Pathology. These GBMs had not been previously treated with radiation or chemotherapy and blocks contained regions of high-grade glioma with foci of necrosis, infiltrating glioma, and adjacent non-neoplastic brain. GBM tissues were fixed in 10% buffered formalin, routinely processed, and paraffin-embedded. Immunohistochemical (IHC) studies were performed on 6-μm sections as described.15Rong Y. Belozerov V.E. Tucker-Burden C. Chen G. Durden D.L. Olson J.J. Van Meir E.G. Mackman N. Brat D.J. Epidermal growth factor receptor and PTEN modulate tissue factor expression in glioblastoma through JunD/activator protein-1 transcriptional activity.Cancer Res. 2009; 69: 2540-2549Crossref PubMed Scopus (96) Google Scholar, 16Rong Y. Post D.E. Pieper R.O. Durden D.L. Van Meir E.G. Brat D.J. PTEN and hypoxia regulate tissue factor expression and plasma coagulation by glioblastoma.Cancer Res. 2005; 65: 1406-1413Crossref PubMed Scopus (177) Google Scholar Sections were deparaffinized and subjected to heat-induced epitope retrieval by steaming for 15 minutes. Slides were then incubated with antibodies directed toward Stat3 (rabbit polyclonal, 1:100; Cell Signaling, Beverly, MA) and phospho-Stat3 (rabbit polyclonal, 1:100; Abcam, Cambridge, MA), C/EBP-β (monoclonal, 1:100; Abcam), C/EBP-δ (polyclonal, 1:100, Novus Biologicals, Littleton, CO), CD68 (monoclonal, 1:100; Biocare Medical, Concord, CA), and CD163 (polyclonal, 1:100; Novus Biologicals). Antibodies were detected using the avidin-biotin-peroxidase complex method using 3,3′-diaminobenzidine as the chromogen. Normal sera served as the negative control. Sections were counterstained with hematoxylin. Human GBM cell line U87MG cell culture conditions have been previously described.15Rong Y. Belozerov V.E. Tucker-Burden C. Chen G. Durden D.L. Olson J.J. Van Meir E.G. Mackman N. Brat D.J. Epidermal growth factor receptor and PTEN modulate tissue factor expression in glioblastoma through JunD/activator protein-1 transcriptional activity.Cancer Res. 2009; 69: 2540-2549Crossref PubMed Scopus (96) Google Scholar, 16Rong Y. Post D.E. Pieper R.O. Durden D.L. Van Meir E.G. Brat D.J. PTEN and hypoxia regulate tissue factor expression and plasma coagulation by glioblastoma.Cancer Res. 2005; 65: 1406-1413Crossref PubMed Scopus (177) Google Scholar Cells used in experiments were grown to 80% confluence in 100-mm culture dishes, placed in serum free media in conditions of 21% O2 (normoxia) or 1% O2 (hypoxia)17Rong Y. Hu F. Huang R. Mackman N. Horowitz J.M. Jensen R.L. Durden D.L. Van Meir E.G. Brat D.J. Early growth response gene-1 regulates hypoxia-induced expression of tissue factor in glioblastoma multiforme through hypoxia-inducible factor-1-independent mechanisms.Cancer Res. 2006; 66: 7067-7074Crossref PubMed Scopus (82) Google Scholar, 18Post D.E. Sandberg E.M. Kyle M.M. Devi N.S. Brat D.J. Xu Z. Tighiouart M. Van Meir E.G. Targeted cancer gene therapy using a hypoxia inducible factor dependent oncolytic adenovirus armed with interleukin-4.Cancer Res. 2007; 67: 6872-6881Crossref PubMed Scopus (80) Google Scholar. For experiments in 1% O2, culture dishes were placed in incubators that are dedicated to hypoxia (94% N2, 5% CO2, and 1% O2 at 37°C). Exposure to hypoxia lasted 24 hours. Immunoblots were performed on proteins from cell lysates of the indicated cell lines. The NE-PER Nuclear and Cytoplasmic Extraction Reagents (Pierce Biotechnology, Rockford, IL) was used for separation of nuclear and cytoplasmic protein fractions. Protein concentrations were determined by a Bradford assay (Bio-Rad Laboratories, Hercules, CA). Equal amounts of protein (30 μg) were resolved on a 10% SDS-PAGE and transferred to nitrocellulose membranes. Blots were incubated in blocking solution (PBS containing 0.02% Tween-20 and 5% nonfat milk) and incubated overnight at 4°C with antibodies specific for Stat3 (monoclonal, 1:2000, Cell Signaling) and phospho-Stat3 (monclonal, 1:1000; Cell Signaling), C/EBP-β (polycolonal, 1:1000, Cell Signaling), C/EBP-δ (polyclonal, 1:1000, Novus Biologicals), and HIF-1α (monoclonal, 1:1000, BD Transduction Lab, Research Triangle Park, NC). Blots were washed and incubated with horseradish peroxidase conjugated to goat anti-mouse or goat anti-rabbit antibodies (1:2000, Bio-Rad, Hercules, CA) for 1 hour at room temperature and developed by enhanced chemiluminescence reagents (Pierce Biotechnology). Histone H1 (monoclonal, 1:4000; Santa Cruz Biotechnology, Santa Cruz, CA) was used as loading control for the nuclear compartment. Association between C/EBP-β, C/EBP-δ, and STAT3 expression and survival were examined using the log rank test to compare the upper and lower quartiles of gene-centric profiles. Survival was taken as "days to death" for uncensored patients and "days to last follow-up" for right-censored patients. All qualitative analysis were repeated three times. Quantitative data are expressed as mean ± SEM. Significance was defined as P < 0.05. We investigated potential correlation of microenvironmental features in GBM with gene expression and genomic patterns, and we primarily focused on the two dominant pathological findings of GBM: necrosis and angiogenesis.11Brat D.J. Prayson R.A. Ryken T.C. Olson J.J. Diagnosis of malignant glioma: role of neuropathology.J Neurooncol. 2008; 89: 287-311Crossref PubMed Scopus (67) Google Scholar, 19Rong Y. Durden D.L. Van Meir E.G. Brat D.J. "Pseudopalisading" necrosis in glioblastoma: a familiar morphologic feature that links vascular pathology, hypoxia, and angiogenesis.J Neuropathol Exp Neurol. 2006; 65: 529-539Crossref PubMed Scopus (357) Google Scholar, 20Kaur B. Tan C. Brat D.J. Post D.E. Van Meir E.G. Genetic and hypoxic regulation of angiogenesis in gliomas.J Neurooncol. 2004; 70: 229-243Crossref PubMed Scopus (121) Google Scholar Two sources of data were used as measures of the degree of necrosis and angiogenesis in frozen section slides of TCGA samples analyzed for quality assurance before molecular analysis. As a primary source, we downloaded digitized images from all 177 available frozen section slides, corresponding to 99 samples and 91 patients, and marked up images for degree of necrosis and angiogenesis using a computer-human interface (Figure 1). The percentage necrosis varied from 0 to 84.7% with a mean of 13.6 ± 1.9%. The degree of angiogenesis varied from 0 to 13.7% with a mean of 1.4 ± 0.2%. There was a weak positive relationship between the percent necrosis and angiogenesis within each sample (Spearman's rho = 0.29). As a secondary source, we used annotations of 293 GBMs, which were recorded by TCGA neuropathologists who reviewed frozen section slides for quality assurance. Percent necrosis was recorded as a visual estimate of total tissue area involved. Angiogenesis was recorded only as "present" or "absent," and was not useful in our analysis. Percent necrosis in these cases varied from 0 to 87.5% with a mean of 14.5 ± 1.6%. Among 70 samples with overlap between the two sources of data, there was strong agreement on percent necrosis (Spearman's rho = 0.71). To determine whether there was a correlation between transcriptional class and extent of necrosis or angiogenesis, we examined each class for the distributions of these features. In the set of 85 GBMs manually marked for necrosis with associated molecular data, we found that the mesenchymal class was significantly enriched with samples that had a high degree of necrosis (31% of cases with >25% necrosis) compared to the other three tumor subtypes (0 cases with >25% necrosis) (Figure 1C). The mesenchymal class also had a higher mean percent necrosis (21.5 ± 4.5%) than the other three classes combined (6.9 ± 0.9%; one-way analysis of variance; P = 8.7 × 10−4). Extent of angiogenesis was not significantly associated with GBM subtype, yet all extreme outliers (angiogenesis >8%) were mesenchymal (Figure 1F). We validated these findings using the set of GBMs, which were visually estimated for necrosis by TCGA neuropathologists (291 samples; not shown). In this set, the mesenchymal class was also significantly enriched in samples with high necrosis compared to other classes (one-way analysis of variance; P = 6.9 × 10−5). Because genetic alterations may play a role in the gene expression class and could be related to the extent of necrosis and angiogenesis, we determined if the level of necrosis was related to TP53, PTEN, EGFR, NF1, or IDH1 mutations, EGFR or PDGFRA amplification, CDKN2A deletion or G-CIMP+ status. In analysis of 189 samples with DNA methylation data, we found that low necrosis was significantly associated with G-CIMP+ status (two-way t-test; P = 0.0064), but not with the proneural class as a whole. Analysis of a more limited set of 114 cases with mutation data did not reveal a significant association among IDH1 mutants and wild-type cases. No other significant correlations were noted between angiogenesis and necrosis, and frequent genetic alterations in GBM (TP53, PTEN, EGFR, or NF1 mutations, EGFR or PDGFRA amplification, CDKN2A deletion). Next we compared extremely necrotic mesenchymal samples to GBMs from other transcriptional classes to determine whether they had a distinctive spectrum of genomic alterations.6Cancer Genome Atlas NetworkComprehensive genomic characterization defines human glioblastoma genes and core pathways.Nature. 2008; 455: 1061-1068Crossref PubMed Scopus (5223) Google Scholar Because those mesechnymal GBMs with >40% necrosis represented statistical outliers (Figure 1C), we compared this subset to all nonmesenchymal GBMs. These outlying necrotic mesenchymal GBMs did not harbor a particular set of genetic alterations when compared to samples in the other three transcriptional classes (Table 1). Thus, there was not a defining set of genetic alterations associated with the highly necrotic mesenchymal cases.Table 1Mutation and Copy Number Alterations in Highly Necrotic GBMs Compared to Low Necrosis, Nonmesenchymal GBMsGenes mutatedHigh necrosis, MS; n (%)Low necrosis, non-MS; n (%)TP531/4 (25)15/70 (21.4)PTEN0/4 (0)14/70 (20.0)NF11/4 (25)3/70 (4.3)EGFR1/4 (25)14/70 (20)ERBB20/4 (0)5/70 (7.1)RB10/4 (0)4/70 (5.7)PIK3R10/4 (0)10/70 (14.3)PIK3CA1/4 (25)3/70 (4.3)IDH10/4 (0)7/70 (10.0)Copy number alterationsn (%)n (%) EFGR amplification3/4 (75)50/66 (75.8) PDGFRA amplification1/4 (25)12/66 (18.2) CDKN2A deletion2/4 (50)2/66 (63.6)Frequency of mutations and copy number variation from the TP53, RB, and receptor tyrosine kinase pathways in MS samples with more than 40% necrosis and samples from all other transcriptional classes.GBMs, glioblastomas; MS, mesenchymal; RB, retinoblastoma. Open table in a new tab
Referência(s)