Virchows Arch (2008) 453:599–609 DOI 10.1007/s00428-008-0685-7
ORIGINAL ARTICLE
Gene expression profiling in glioblastoma and immunohistochemical evaluation of IGFBP-2 and CDC20 Gianluca Marucci & Luca Morandi & Elisabetta Magrini & Anna Farnedi & Enrico Franceschi & Rossella Miglio & Daniela Calò & Annalisa Pession & Maria P. Foschini & Vincenzo Eusebi
Received: 28 July 2008 / Revised: 29 September 2008 / Accepted: 6 October 2008 / Published online: 25 October 2008 # Springer-Verlag 2008
Abstract Thirty-nine glial tumours (28 glioblastomas (GB) and 11 low-grade gliomas) were investigated with DNA microarrays to reveal a possible specific gene expression profile. Unsupervised classification through hierarchical cluster analysis identified two groups of tumours, the first composed of low-grade gliomas and the second mainly composed of GB. Nine genes were identified as most informative: seven were over-expressed in low-grade gliomas and under-expressed in GB; on the contrary, two genes, insulin-like growth factor binding protein 2 (IGFBP2) and cell division cycle 20 homologue (CDC20), were over-expressed in GB and under-expressed in low-grade tumours. This same genetic profile was confirmed by reverse transcriptase polymerase chain reaction. Immunohistochemistry for IGFBP-2 was positive in 88.8% of the
This work was supported by the MIUR /FISR project 1509 (202). G. Marucci : L. Morandi : E. Magrini : A. Farnedi : A. Pession : M. P. Foschini : V. Eusebi Section of Pathology, Bellaria Hospital, University of Bologna, Bologna, Italy E. Franceschi Department of Medical Oncology, Bellaria Hospital, Bologna, Italy R. Miglio : D. Calò Department of Statistics “Paolo Fortunati”, University of Bologna, Bologna, Italy V. Eusebi (*) Sezione di Anatomia Patologica “M. Malpighi”, Ospedale Bellaria, Via Altura 3, 40139 Bologna, Italy e-mail:
[email protected]
cases of GB and in only one low-grade glioma, whilst CDC20 immunostained 74.1% of the cases of GB and none low-grade glioma. This was confirmed in an additional series of cases studied with immunohistochemistry only. In conclusion, over-expression of mRNA levels of IGFBP-2 and CDC20 is highly related to GB, IGFBP-2 and CDC-20 gene and protein expressions are strongly correlated, and IGFBP-2 and CDC20 immunopositivity can be useful for the identification of GB in small biopsies. Keywords Glioblastoma . High-grade glioma . Gene expression . IGFBP-2 . CDC20
Introduction Gliomas are the most common primary brain tumours. Amongst high-grade gliomas, glioblastoma (GB) shows the most aggressive clinical course with median overall survival of 10 to 12 months after diagnosis [1]. Despite advances in surgical techniques, post-operative supportive care, radiation and adjuvant systemic therapy, the life expectancy of patients with GB has remained essentially poor over the last several decades. On the other hand, glial tumours with oligodendroglial features, despite anaplastic characteristics, retain sensitivity to chemo and radiation therapy and show better overall survival. Thus, tumour classification [1] is the variable that most affects therapeutic decisions and prognostic estimation. Unfortunately, inter-observer variability can occur, resulting in limited diagnostic reproducibility. Coons et al. found that complete diagnostic concordance amongst four neuropathologists reviewing gliomas peaked at 69% espe-
600
cially in grade III lesions [2]. To develop more objective approaches to glioma classification, recent investigations have focused on molecular genetic analyses [3]. The use of DNA microarrays is set to change the development and use of tumour biomarkers. The few brain tumour biomarkers that are currently available include chromosomal loss of 1p and 19q for oligodendrogliomas (OL) [4], neurotrophic thyrosine kinase receptor type 3, N-MYC, C-MYC and v-erb-b2 erythroblastic leukaemia viral oncogene homologue 2 for medulloblastoma [5]. Analysis of the association between any of these single biomarkers and response to therapy requires a large number of patients to achieve sufficient statistical power. A single biomarker usually has limited predictive power especially if many other genes or proteins are important to determine the outcome. DNA microarrays can be used to detect groups of genes that, in aggregate, contain more predictive informations than any individual biomarker. The fact that an eight-gene model can be used to predict survival in medulloblastoma [6], that a six-gene model can be used to predict the outcome in patients with diffuse large-B-cell lymphoma [7] and that 70 genes can predict axillary metastasis in breast carcinoma [8] indicates that chemotherapy-response predictors can be modelled using relatively small panels of genes that can be screened by reverse transcriptase polymerase chain reaction (RT-PCR) or immunohistochemistry. Therefore, a series of 39 gliomas, 11 low grade and 28 high grade, was studied at gene and protein (immunohistochemistry) levels to disclose possible genetic portraits of malignancy.
Materials and methods We retrieved for this study 100 randomly selected cases of gliomas. All the cases had been received unfixed. A sample of neoplastic tissue had been fresh frozen over liquid nitrogen and stored at −80°C into the frozen tissue bank of the Section of Pathology of the University of Bologna at Bellaria Hospital (Bologna), between 1990 and 2002. The remaining (specular) tissues were formalin fixed and paraffin embedded and stained with haematoxylin eosin for routine histological diagnosis. Informed consent was obtained from the subjects. Only samples with rRNA 28S/18S ratio >1.5, measured by Bioanalyzer 2100 (Agilent, see below for details) and no evidence of ribosomal degradation, were included. Some cases were excluded as the tissue samples were composed only by necrosis; other cases did not show neoplastic proliferation in the specimen stored; in other cases, tissue was not enough to obtain a frozen section necessary to check the presence of tumour. Therefore, only 39 glial neoplasms were suitable for the study, i.e. 28 GB and 11 low-grade glial tumours, namely four OL, five pilocytic
Virchows Arch (2008) 453:599–609
astrocytomas (PA) and two fibrillary astrocytoma (FA), showed good quality RNA. All the tumours were re-staged and graded according to 2007 WHO [1] criteria at time of gene expression analysis by two pathologists (GM and VE). Any disagreement was discussed at a double head microscope and a consensus was reached. RNA extraction and labelling Tissue samples were treated with RNAlater™ (Qiagen, Hilden, Germany) and homogenised. Total RNA was extracted (RNAeasy Protect mini kit, Qiagen, Hilden, Germany) and analysed by Bioanalyzer 2100 using RNA 6000 Nano kit (Agilent Technologies, Palo Alto, CA, USA). Agilent Human 1A oligomicroarrays (Agilent Technologies, Palo Alto, CA, USA) were used in this study containing 60-mer DNA probes synthesised in situ in a 22-k format. Of 19,061 spots, 18,086 are non-controls and there are 17,086 unique transcript sequences, matching to 15,989 unique human genes (TIGR Resourcerer 10.0 July 2004 Release; http://www.tigr.org/tigr-scripts/magic/r1.pl). The labelling of complementary RNA (cRNA), hybridisation to 22K-gene arrays and assessment of expression ratios were all performed as previously described [8]. The cRNA was generated by in vitro transcription with the use of T7 RNA polymerase (low RNA input fluorescent linear amplification kit Agilent cod. 5184–3523) and labelled with Cy3-CTP (Perkin Elmer, Waltham, MA, USA) for the reference RNA or Cy5-CTP (Perkin Elmer, Waltham, MA, USA) for the test RNA. RNA reference pool consisted of a total RNA mixture of four oligodendrogliomas, five pilocytic astrocytomas and two fibrillary astrocytomas enrolled in this study. Labelled test (0.75 μg) and reference cRNA were fragmented and hybridised at 60°C for 17 h. After hybridisation, the slides were washed and scanned with a confocal laser scanner (Agilent Catalogue, Number G2565BA). Fluorescent intensities on scanned images were quantified by the feature extraction software 7.5, and LogRatio data were available for further statistical analysis. Microarray data analysis The scan data were analysed with Agilent Feature Extraction Software, which performs spots localisation (find spot algorithm), outlier pixels rejection based on the interquartile range method (cookie cutter algorithm) and flagging of saturated features (a feature is considered saturated when more than 50% of its pixels had an intensity of above 65,502). Thirty-nine specimens (28 GB and 11 low-grade gliomas) passed routine quality control analysis.
Virchows Arch (2008) 453:599–609
601
Loess normalisation and between-array scale normalisation [9] were performed using the “LIMMA” package [10] of Bioconductor (www.bioconductor.org). Hierarchical clustering was performed using Ward’s method of linkage and Euclidean distance as a metric. The nearest shrunken centroid classification method was used to identify genes useful to classify and best characterise high- and low-grade gliomas [11]. This method consists in “shrinking” each of the class centroids toward the overall centroid by a given amount, with the aim of reducing the effect of noisy genes. This shrinkage does automatic gene selection. Tenfold balanced cross validation was used to choose the optimal amount of shrinkage and the corresponding optimal subset of genes. The analysis was performed using the PAMR package of Bioconductor. A list of differentially expressed genes was performed through LIMMA package [12], using a t statistic with pooled variance; adjustment for multiple comparisons was made using false discovery rate (0.001). Overall survival time was defined as the time elapsed between surgery and death from the disease. Gene expression levels were categorised in order to obtain two groups: low and high expression level, the value 0 was chosen as cut-point. Univariate survival analysis was performed for each gene, selected from the shrunken centroid method using the log-rank test, and survival curves were estimated using the Kaplan–Meier method. Cox multivariate regression model was estimated considering all genes and age at diagnosis as covariates. Variables were selected using a backward stepwise selection procedure forcing age at diagnosis to be present in the model. Statistical analyses were performed using STATA (V.9).
complex, locus H (LY6H), NADH dehydrogenase (ubiquinone) Fe-S protein 2 (NDUFS2) and transmembrane 4 super-family 2 (TM4SF2)) in all the cases. Amongst the genes selected by the nearest shrunken centroid method, only these five genes were studied as these were the most discriminating genes between high- and low-grade tumours in view of the paucity of residual RNA material (see the “Results” section). In brief, an aliquot of 1 μg of the same RNA used for microarray analysis was reverse-transcribed using SuperScript™ First-Strand Synthesis System for RT-PCR (Invitrogen), according to the manufacturer’s instructions. PCR was carried out in a 25-μl reaction volume containing 10 pmol of each primers, 1 U of FastStart Taq DNA Polymerase™ (Roche), 1× GC rich solution (Roche), 4 mM MgCl2, 1× Rox (Invitrogen) and 1× GelStar in DMSO (Lonza). Real-time RT-PCR was carried out by the Applied Bio systems SDS-7000 thermal cycle (Applied Bio systems, CA, USA). Cycling parameters consisted of an initial dye normalisation at 50°C for 2 min, a denaturation step at 95°C for 4 min, followed by 40 cycles of denaturation at 95°C for 30 s, primer annealing at 58°C, extension at 72°C for 30 s with data acquisition, followed by melt curve analysis. Primer sequences are listed in Table 1 and were retrieved from the RTPrimerDB database (http://medgen. ugent.be/rtprimerdb/) [13]. Four housekeeping genes were chosen for data normalisation (ACTB, B2M, TBP and HPRT1). Relative quantification were performed using qBase (http://medgen.ugent.be/qbase/; Vandesompele et al.) [14]. A Wilcoxon two group test has been used to compare the two pools of ΔCt values, using Bonferroni correction for multiple comparisons.
Validation by real time RT-PCR Immunohistochemistry Real time RT-PCR was used to confirm the gene expression microarray results for five selected genes (insulin-like growth factor binding protein 2 (IGFBP-2), cell division cycle 20 homologue (CDC20), lymphocyte antigen 6
The immunohistochemical study was performed in 37 cases (27 GB and 10 low-grade tumours) of the series submitted to microarray analysis, as in two cases there was no residual
Table 1 Primer sequences used for RT-qPCR Gene name
Forward primer
Reverse primer
ACTB B2M TBP HPRT IGFBP2 CDC20 LY6H NDUFS2 TM4SF2
TTGCCGACAGGATGCAGAAGGA ATGAGTATGCCTGCCGTGTGA TGCACAGGAGCCAAGAGTGAA TGACACTGGCAAAACAATGCCA GCCCTCTGGAGCACCTCTACT AGACCTGCCGTTACATTCCTTC GTGTGTGCCAGTGTCCGAATC CAGGCCTATTCTCTAGCTGTGGA CTGCTGCATGAACGAAACTGAT
AGGTGGACAGCGAGGCCAGGAT GGCATCTTCAAACCTCCATG CACATCACAGCTCCCCACCA GGTCCTTTTCACCAGCAAGCT CATCTTGCACTGTTTGAGGTTGTAC GCCAGTACATTCCCAGAACTCC AGTCACAGGAGGAGGCACACAT GCCATGATGTGGTTCAACAAAC CAGCGATGATTCCCATGTTAGT
602
material available. Furthermore, additional ten cases of grade II gliomas (five FA and five OL) and ten cases of grade III gliomas (six anaplastic astrocytomas, three anaplastic oligodendrogliomas and one anaplastic oligoastrocytomas) were randomly selected in order to enhance the immunohistochemical study. Cases were immunostained with anti-GFAP (Dako, clone 6F2, dilution 1:1,200), anti-Ki 67 (Novocastra, clone Mib1, dilution 1:200), anti-IGFBP-2 (Cell Signalling Technology, polyclonal antibody, dilution 1:25) and antiCDC20 antisera (Santa Cruz Biotechnology, clone p55 CDC (E-7), dilution 1:50). Standard avidin–biotin immunoperoxidase procedures were performed on paraffinembedded sections. For IGFBP-2, semi-quantitative scores were recorded for cytoplasmatic staining, using normal brain as positive control. For CDC20, semi-quantitative scores were recorded for nuclear staining, using endothelial cells as positive control. The absence of primary antibody was used as negative control. Counting was performed on ten areas in each case (×400 magnification). Negative is 11 of positive cells. For statistical purposes, tissue samples were considered positive with score + and ++. For Ki 67, a percentage value was attributed to each case evaluating ten fields (×400 magnification). Statistical analysis of immunohistochemical results The associations between tumour grade and the immunoistochemical expression of IGFBP-2 and CDC 20 were evaluated using the Fisher exact test. A one-way ANOVA was conducted to compare the mean IGFBP2 and CDC20 gene expression levels amongst the corresponding immunohistochemical groups. Levene’s test was used to examine the homogeneity of variance assumption. The association of IGFBP-2 and CDC20 with overall survival was also evaluated in the cases studied for genetic expression. The survival curves were estimated using the standard KaplanMeier method and compared with log-rank test.
Results Gene expression data were obtained from 39 samples (28 GB and 11 low-grade gliomas) and are available on the GEO web site (http://www.ncbi.nlm.nih.gov/geo/) with the accession number GSE9885. Clinical and pathological variables are reported in Table 2. Hierarchical clustering, based on gene expression data and performed using Ward’s method of linkage and Euclidean distance as a metric (Fig. 1), separated two groups showing good concordance with histology: 100%
Virchows Arch (2008) 453:599–609
for OL (four of four clustered in group 1), 100% for PA (five of five PA clustered in group 1) and 50% for FA (one of two in group 1). GB clustered mainly in group 2 (25 of 28, 89.3%) together with one FA. Median overall survival time for group 1 was not reached after a median follow-up of 4 years. Median overall survival time for group 2 was 10 months. Three GB clustered with group 1 that included most of the low-grade glioma cases. However, survival in these three patients was 2, 11 and 13 months, respectively, consistent with GB. The nearest shrunken centroid method identified nine genes (IGFBP-2, CDC20, LY6H, NDUFS2, TM4SF2, SNURF, GABARAPL1, AZGP1, BC016828) as the most important genes for distinguishing between low- and highgrade gliomas. Amongst them, seven genes were overexpressed in low-grade gliomas but under-expressed in GB; on the contrary, two genes (IGFBP-2 and CDC20) were over-expressed in GB but under-expressed in low-grade tumours (Fig. 2). All nine genes are included amongst the 50 significantly expressed top-ranked genes (Table 3). The rate of misclassification on the basis of tenfold balanced cross validation was 0% for low-grade gliomas, 10.7% (three of 28) for GB and 7.7% (three of 39) for all cases. Adequate mRNA quantities of five out of the nine genes, identified with nearest shrunken centroid method, were tested by RT-PCR for validation (see Supplementary informations for details). Four reference genes were used for normalisation as indicated by Vandesompele et al. [14]. We found significant expression differences between the two groups in three out of the five genes: IGFBP2 (p< 0.0005), CDC20 (p
