Conceived and designed the experiments: TK JHS DB CSM. Performed the experiments: LC. Analyzed the data: LC DAG QL BAJ SRC DB CSM. Contributed reagents/materials/analysis tools: JHS. Wrote the paper: LC DB CSM.
The authors have declared that no competing interests exist.
The Cancer Genome Atlas Project (TCGA) has produced an extensive collection of ‘-omic’ data on glioblastoma (GBM), resulting in several key insights on expression signatures. Despite the richness of TCGA GBM data, the absence of lower grade gliomas in this data set prevents analysis genes related to progression and the uncovering of predictive signatures. A complementary dataset exists in the form of the NCI Repository for Molecular Brain Neoplasia Data (Rembrandt), which contains molecular and clinical data for diffuse gliomas across the full spectrum of histologic class and grade. Here we present an investigation of the significance of the TCGA consortium's expression classification when applied to Rembrandt gliomas. We demonstrate that the proneural signature predicts improved clinical outcome among 176 Rembrandt gliomas that includes all histologies and grades, including GBMs (log rank test p = 1.16e-6), but also among 75 grade II and grade III samples (p = 2.65e-4). This gene expression signature was enriched in tumors with oligodendroglioma histology and also predicted improved survival in this tumor type (n = 43, p = 1.25e-4). Thus, expression signatures identified in the TCGA analysis of GBMs also have intrinsic prognostic value for lower grade oligodendrogliomas, and likely represent important differences in tumor biology with implications for treatment and therapy. Integrated DNA and RNA analysis of low-grade and high-grade proneural gliomas identified increased expression and gene amplification of several genes including GLIS3, TGFB2, TNC, AURKA, and VEGFA in proneural GBMs, with corresponding loss of DLL3 and HEY2. Pathway analysis highlights the importance of the Notch and Hedgehog pathways in the proneural subtype. This demonstrates that the expression signatures identified in the TCGA analysis of GBMs also have intrinsic prognostic value for low-grade oligodendrogliomas, and likely represent important differences in tumor biology with implications for treatment and therapy.
Glioblastoma (GBM) is the most common primary brain tumor, with 8700 new cases per year in the United States
The analysis of gene expression patterns in GBMs suggests that this histologic category may include distinct subtypes. Several groups have developed approaches for subtyping GBMs by gene expression signatures
All human subjects data was publicly available de-identified data from The Cancer Genome Atlas project and the Rembrandt database, and thus, not designated as human subjects research. No Institutional Review Board approval was required.
Microarray and clinical data were acquired in un-normalized form from the Rembrandt
To classify Rembrandt samples within the TCGA classification schema, Rembrandt data for the probes from the Affymetrix U133 Plus 2.0 GeneChip were mapped to the TCGA class signature genes using HUGO gene symbol and Entrez gene ID number. This comparison yielded an intersection of 1486 Affymetrix probe sets. Classification of the Rembrandt samples was then performed using prediction analysis for microarrays using the signature gene class centroids
Comparisons of survival between different classes were performed using the log-rank test
Differences in gene expression between subtypes were determined using the comparative selection marker module of GenePattern 3.0. Cutoffs for statistical significance were a Benjamini- Hochberg corrected False Discovery Rate (FDR) <0.05 and a minimum fold change >1.8. Statistical significance of Gene Ontology overrepresentation were determined by hypergeometric distribution using the DAVID database
Cox proportional hazards (PH) models were used to examine the association between patient survival and four subtypes after adjusting for patient age at diagnosis as well as 1p/19q status whenever applicable. We note that 1p/19q deletion is present only in Oligodendrogliomas; hence, 1p/19q deletion status was not adjusted for in cases with Astrocytoma or GBM.
Unsupervised hierarchical cluster analysis of gene expression data derived from TCGA GBM samples resulted in four distinct gene expression subtypes: Neural, Proneural, Classical, and Mesenchymal. This classification utilized an integrated analysis of three gene expression platforms to identify a set of genes that is consistently and variably expressed among the TCGA samples. Expression measurements from the Affymetrix HT-HG-U133A, the Affymetrix Human Exon 1.0 ST, and a custom Agilent array, were combined by mapping to a transcript database 11,861 total genes. Of these, 1740 were reliably expressed across platforms, with 840 identified as class signature genes by ClaNC analysis
Total gene expression and clinical survival data were downloaded from the Rembrandt website (
(
True\Predicted | Classic | Mesenchymal | Neural | Proneural | Class Error rate |
Classic | 43 | 4 | 0 | 0 | 0.085 |
Mesenchymal | 2 | 40 | 1 | 0 | 0.070 |
Neural | 0 | 1 | 28 | 1 | 0.067 |
Proneural | 1 | 0 | 2 | 37 | 0.075 |
Overall prediction accuracy was 148/160 = 92.5% correct.
Classic | 42 | 6 | 3 | 51 |
Mesenchymal | 21 | 0 | 5 | 26 |
Neural | 12 | 5 | 8 | 25 |
Proneural | 26 | 32 | 16 | 74 |
101 | 43 | 32 | 176 |
The Proneural subtype dominates oligodendriogliomas and there are no mesenchymal subtype in the oligodenroglioma samples.
To examine if there were any differences in survival for the entire set of 176 classified Rembrandt samples, we performed a Kaplan-Meier analysis (
We next performed a multivariate Cox Proportional Hazard analysis of expression subtypes, adjusting for age and 1p/19q deletion status (
152 | 101 | 32 | 36 | |
1.96 (CI: 1.23, 3.13); |
1.17 (CI: 0.69, 1.96); p = 0.563 | 8.48 (CI: 1.85, 38.85); |
6.25 (CI: 1.55, 25.19); |
|
2.68 (CI: 1.51, 4.76); |
1.56 (CI: 0.85, 2.85); p = 0.150 | 1.04 (CI: 0.29, 3.76); p = 0.9537 | N/A | |
2.38 (CI: 1.38, 4.11); |
1.57 (CI: 0.78, 3.17); p = 0.207 | 1.91 (CI: 0.72, 5.07); p = 0.1963 | 23.28 (CI: 4.63, 116.98); |
|
3.14 (CI: 2.12, 4.64); |
2.90 (CI: 1.84, 4.55); |
6.79 (CI: 2.77, 16.66); |
4.97 (CI: 1.58, 15.70); |
|
0.42 (CI: 0.16, 1.05); p = 0.0634 | N/A | N/A | 2.23 (CI: 0.59, 8.46); p = 0.2380 |
Shown are Hazard Ratios (HRs) plus 95% confidence intervals (CI) and associated p-values. Significant p-values (<0.05) are shown in bold font. For the Cox PH models, Proneural was used as the reference group for the subtype variable, patient age was dichotomized as <55 or ≥55 with <55 as the reference group, and no deletion in 1p/19q status was used as the reference group. Cox PH multivariate analysis was performed for all cases, cases with Astrocytoma, cases with GBM and cases with Oligodendrogliomas, respectively. We note that 1p/19q deletion is present only in Oligodendrogliomas; hence, 1p/19q deletion status was not adjusted for in cases with Astrocytoma or GBM.
To make sure that none of the classic or neural oligodendrogliomas were misclassified small cell GBMs, we carefully examined patient age, chromosome 10 loss, and EGFR amplification status. Of the six classic and five neural oligodenrogliomas, none of the neural and only two of the classic oligodendrogliomas had high patient age (>70), chr10 loss, and/or EGFR amplification. We reanalyzed the survival data excluding these two cases (HF1150 and HF0510), and none of the statistical findings were affected, except that the p-value for HR comparing classic and proneural oligodendrogliomas increased to 0.0595 (marginally significant) from 0.0100 (significant), which is not too surprising given that the sample size of the oligodendrogliomas was reduced by two.
To investigate mechanisms underlying the difference in survival associated with the Proneural subtype, we performed several differential gene expression analyses. First we compared the set of proneural lower grade gliomas (both oligodendrogliomas and astrocytomas; PN-OA) to proneural GBMs (PN-GBM). In addition, because of the markedly improved survival of proneural oligodendrogliomas (PN-Oligo), we compared the gene expression of PN- Oligo to PN-GBM, proneural astrocytomas (PN-Astro), neural oligodendrogliomas (N-Oligo), and classic oligodendrogliomas (C-Oligo). We also compared proneural oligodendrogliomas to the combined set of classic and neural oligodendrogliomas. Using minimum fold-change cutoff of 1.8 and a Benjamini- Hochberg corrected False Discovery Rate (FDR) <0.05, we identified 779 probes differentially regulated between PN-Oligo and PN-GBM (
(
DNA Copy number data for the Rembrandt samples was calculated using GenePattern version 3.2.0. Segmentation of raw copy numbers was calculated using the GLAD module (version 2). Significance analysis of amplifications and deletions was performed using the GISTIC method
We identified a number of regions with significant amplifications or deletions in the PN-GBM and PN-Oligo subtypes (
To gain insight into the progression of proneural samples, we performed an integrated analysis of DNA copy number and mRNA gene expression within the proneural subtype of the Rembrandt samples, specifically examining those changes between PN-OA and PN-GBM samples. We observed 66 probe sets corresponding to 52 genes that showed significant amplification or deletion, as well as significant differences in gene expression between these subtypes. A comprehensive list of the mRNA probes that intersect with GISTIC copy number analysis is given in
GBM-Amp | 12p13.32 | 1.21E-03 | C1R, C1RL, C1S |
GBM-Amp | 9p23 | 1.34E-02 | GLIS3, HAUS6 |
GBM-Amp | 9p23 | 1.34E-02 | PLIN2 |
GBM-Amp | 6p21.31 | 4.30E-02 | CLIC1, F13A1 |
GBM-Amp | 6p21.31 | 4.30E-02 | HISTH1C, HISTH2BK, HLA-DQA1, HLA-DQB1, HLA-DRA, HLA-DRB1 |
OA-Amp | 22q11.1 | 4.69E-04 | BCR |
OA-Amp | 22q11.1 | 4.69E-04 | MN1, RASL10A, SEZ6L |
OA-Amp | 22q11.1 | 4.69E-04 | SLC25A18 |
OA-Amp | 8q24.22 | 8.23E-04 | FAM84B |
OA-Amp | 19p13.3 | 3.10E-03 | SHD, SLC1A6 |
OA-Amp | 6p21.2 | 8.78E-03 | DAAM2 |
OA-Del | 19q13.31 | 1.19E-06 | EMP3 |
OA-Del | 4q21.23 | 8.58E-03 | CCDC109B, CFI |
OA-Del | 4q21.23 | 8.58E-03 | ENPEP, HERC5, IBSP, MLF1IP, SEC24D, SPP1, TDO2 |
Loci with amplification or deletion and corresponding changes in gene expression in Proneural GBM or Proneural Oligodendrogliomas/Astrocytomas are shown.
Secondly, to gain insight into differences in histologic differentiation and survival, we also performed an integrated analysis of DNA copy number and mRNA gene expression between PN-Oligo samples and PN-GBM samples in the Rembrandt dataset. We observed 323 probe sets corresponding to 240 genes that showed significant amplification or deletion, as well as significant differences in gene expression between these subtypes. A comprehensive list of the mRNA probes that intersect with GISTIC copy number analysis is given in
We performed four separate types of pathway analysis using gene set enrichment analysis (GSEA), the Database for Annotation, Visualization and Integrated Discovery (DAVID), Literature Lab Analysis (LLA), and Ingenuity Pathway Analysis (IPA). The comparison of PN-GBM to PN-OA subtypes identified a number of gene ontologies and pathways that were significantly enriched in multiple analyses (
(
Response To Wounding | 47 | 1.89E-13 | 11 | 5.97E-05 | ||
Collagen Fibril Organization | 14 | 1.02E-11 | 10 | 1.88E-10 | ||
Extracellular Matrix Organization | 19 | 8.33E-09 | 23 | 0.16288991 | 3 | 1.86E-03 |
Skeletal System Development | 29 | 2.58E-07 | 30 | 1.59E-09 | ||
Inflammatory Response | 28 | 1.90E-06 | 120 | 1.37E-06 | ||
Cell Adhesion | 40 | 3.90E-05 | 43 | 1.11E-05 | ||
Vasculature Development | 21 | 5.88E-04 | 16 | 7.30E-09 | ||
Regulation Of Cell Proliferation | 38 | 0.00586091 | 109 | 6.41E-11 | ||
ECM_Receptor_Interaction | 79 | 0.0247256 | ||||
Cell_Communication | 107 | 0.06453185 | ||||
Blood_Coagulation | 38 | 0.20890287 | ||||
Regulation_Of_MAPKKK_Cascade | 18 | 0.14662187 | ||||
Cancer | 166 | 1.78E-27 | ||||
Apoptosis | 61 | 2.54E-10 | ||||
Development | 30 | 1.59E-09 | ||||
Invasion | 37 | 1.75E-09 | ||||
Brain Cancer | 25 | 4.04E-08 | ||||
Cell Cycle Progression | 51 | 5.69E-08 | ||||
Mitosis | 29 | 1.12E-06 | ||||
Glioma | 13 | 3.05E-05 | ||||
Astrocytoma | 8 | 4.23E-04 |
Proneural GBM samples have increased expression of genes associated with enhanced vascularization, proliferation, invasion, and inflammation. Count = number of genes identified with each annotation by the three analyses. FDR = false discovery rate. P-value for IPA is based on the hypergeometric distribution.
In addition, to investigate molecular mechanisms underlying improved survival of the PN-Oligo subtype relative to other types of oligodendrogliomas, we performed pathway analysis of the 129 genes corresponding to the 190 probe sets differentially expressed between PN-Oligo and the Classic and Neural Oligodendrogliomas. As expected, the differential genes were indicative of the proneural subtype, with some of the most significant annotations including Differentiation of Astrocytes, Neurogenesis, Cell Death, Development of Blood Vessels, and Benign Tumor (
Here we demonstrate that the Proneural gene expression signature as defined by TCGA is enriched in gliomas with oligodendrogliomal differentiation. This signature also predicts improved outcome for oligodendrogliomas. Analysis of the copy number data from the Rembrandt dataset demonstrated a high frequency of large losses of chromosomes 1p and 19q in oligodendrogliomas, but not in astrocytomas or GBMs. Twelve of the 42 oligodendrogliomas showed co-deletion of at least 85% of chromosomes 1p and 19q, and eleven of those twelve samples were PN-oligos. The other sample with co-deletion of 1p/19q was a classic oligodendroglioma. Nevertheless, our findings regarding the improved outcome of proneural subtype remained significant in multivariate analyses controlling for 1p/19q status and age at diagnosis. Multivariate analysis also showed that the proneural subtype of astroctyoma has significantly better outcome than the classic subtype of astrocytoma, suggesting that the proneural gene expression signature carries prognostic significance across histologic types in the diffuse gliomas.
Our integrated analysis of expression patterns and DNA copy number of low-grade and high-grade proneural gliomas in the Rembrandt dataset identified genes and pathways associated with progression of this disease. Genes increased in high-grade proneural GBMs (PN-GBM) at both the DNA and RNA level included GLIS3, TGFB2, TNC, AURKA, and VEGFA. Network analysis of these gene expression changes identified several regulatory hubs, including GLI1, RUNX2, MYC, BMP2, and NOTCH1. The GLI transcription factors are effectors of the Hedgehog pathway and have been strongly implicated as key regulators of glioblastoma behavior since their discovery. GLI1 regulates stem cell renewal and tumorigenicity of gliomas
Both Ingenuity Pathway Analysis and Literature Lab Analysis identified the Notch pathway as being differentially regulated in low grade and high-grade proneural gliomas. Several components of the Notch pathway, including DLL3 and HEY2 are reduced in PN-GBM, while NOV/CCN3, which is associated with Notch inhibition
As gliomas progress from lower grade (grades II and III) to GBM, hypoxia and necrosis develop centrally, while angiogenesis emerges peripherally. These processes are related, with hypoxia-inducible factors (e.g. VEGFA) secreted by hypoxic, perinecrotic tumor cells, resulting in the development of new vessels. It has been suggested that vascular pathology, including vascular endothelial apoptosis, vascular occlusion and thrombosis, initiates the development of central hypoxia and necrosis. Angiopoetin 2 has been implicated in initiating endothelial apoptosis by Tie2 receptor in this setting
In summary, our
Complete clinical data downloaded from the Rembrandt public data repository (
(0.15 MB XLS)
Survival and PAM subtype classification data for Rembrandt samples used in the survival analysis.
(0.05 MB PDF)
Complete Comparative Marker Selection Results generated from Rembrandt data comparing low- and high-grade proneural samples. Data were generated using GenePattern 3.0.
(7.50 MB TXT)
Complete Comparative Marker Selection Results generated from Rembrandt data comparing proneural oligodendroglioma and proneural GBM samples. Data were generated using GenePattern 3.0.
(7.50 MB TXT)
Complete Comparative Marker Selection Results generated from Rembrandt data comparing proneural oligodendroglioma and neural oligodendroglioma samples. Data were generated using GenePattern 3.0.
(7.50 MB TXT)
Complete Comparative Marker Selection Results generated from Rembrandt data comparing proneural oligodendroglioma and classic oligodendroglioma samples. Data were generated using GenePattern 3.0.
(7.50 MB TXT)
Significant Comparative Marker Selection Results generated from Rembrandt data comparing proneural oligodendroglioma and all other oligodendroglioma samples. Data were generated using GenePattern 3.0.
(0.15 MB XLS)
Integrated RNA and DNA Copy number analysis comparison of low- and high-grade proneural gliomas.
(0.19 MB XLS)
Integrated RNA and DNA Copy number analysis comparison of proneural oligodendrogliomas and proneural GBM samples.
(0.15 MB XLS)
Gene Ontology analysis of the 129 genes differentially expressed between proneural oligodendrogliomas and the all other oligodendrogliomas.
(0.02 MB XLS)
Kaplan-Meier Analysis of Low-grade gliomas.
(0.13 MB TIF)
The authors recognize the support of the Emory In Silico Center of Research Excellence (ISCRE) for performance of these studies.