Monoallelic Expression Determines Oncogenic Progression and Outcome in Benign and Malignant Brain Tumors

LOH of cancer-associated genes is a common and important mechanism in carcinogenesis. However LOH at the level of the RNA, or monoallelic gene expression (MAE), is a much less understood phenomenon. MAE is defined as expression of a gene that is restricted to one allele in the presence of a diploid, heterozygous genome and is observed for imprinted genes and random X inactivation (lyonization). MAE has also been observed in the odorant receptor, immune receptor, and immunoglobulin classes of genes (1, 2). For most other genes, expression is typically expected to occur from both parental alleles without preference. Recent studies, however, have uncovered a potential role for random MAE, as well as differential allele-specific expression (DAE; i.e., a skew in allele-specific expression rather than complete monoallelic expression) in normal tissues as well as in some diseases, including cancer (3–9).

Several studies in normal tissues or derived cell lines have shown random MAE/DAE to be relatively frequent (5%–20%), depending on the number of genes included in the study and the gene selection bias (3, 8, 10). These data together reveal the previously unappreciated random nature of MAE/DAE among nonimprinted genes and provide a mechanism that may help to explain phenotypic differences between individuals, as well as their susceptibility to particular diseases and the disease course.

Cancer is a complex disease, the result of interplay among genetic, epigenetic alterations, and environmental cofactors. We chose brain tumors as a platform to examine MAE for several reasons. Brain tumors are a major cause of morbidity and mortality in pediatric and adult oncology (11). The genetic basis of brain tumors has been studied in depth over the past decade, leading to the identification of mutations/genetic alterations central to brain tumor development (12–15). However, although the role of alterations in the TP53, RB, and RAS/MAPK pathways in the development of specific brain tumors has been well characterized, research considering the widespread impact of MAE in these cancers is in its infancy. Expression of tumor suppressors and oncogenes is tightly regulated during the life of an individual, and it has been established that aberrant gene expression, such as that observed from loss of imprinting, can predispose individuals to cancer development (16). It is therefore reasonable to propose that aberrant gene expression caused by lack of one allele, MAE, could also lead to the development of cancer; indeed preliminary evidence exists to support this notion (7–9).

Previous studies have used Affymetrix and Illumina genome-wide single-nucleotide polymorphism (SNP) arrays to identify MAE in normal and tumor tissues (8, 9), however, the majority of studies investigating MAE/DAE have interrogated either cell lines or germline material from patients, and few have examined primary tumor tissues. Thus the role of MAE/DAE in cancer, and more specifically in brain tumor development, is as yet undetermined. Furthermore, the mechanisms and functional consequences of MAE in primary tumors have not been studied. To examine the role of MAE in brain tumors, we chose specific tumors (pediatric gliomas and choroid plexus tumors) that have distinct clinical and biologic subgroups, benign and malignant, and do not routinely transform from the low- to high-grade state. In addition, we analyzed adult gliomas, as these cancers tend to transform from low- to high-grade and, therefore, represent a clinically mixed group. We also included multiple phases of tumor development, such as autopsy samples, to examine changes in MAE over time. We achieved this by genotyping DNA and RNA (cDNA) samples and identifying SNPs in which DNA genotypes were heterozygous and corresponding RNA genotypes were homozygous. We used Affymetrix SNP 6.0 arrays and high-throughput genotyping techniques to interrogate the extent of MAE and to uncover novel genes displaying MAE. We then examined the role of MAE in specific brain cancer–related genes. We find MAE to be a frequent event in brain tumors; furthermore, it is gene and tumor specific and may precede LOH in mutant genes and predict tumor behavior.

This study was approved by the Research Ethics Board at The Hospital for Sick Children. Tumor and blood samples from pediatric brain tumor patients were collected from the Pathology Department or through the Arthur and Sonia Labatt Brain Tumor Research Centre at The Hospital for Sick Children. Adult brain tumor tissues were collected from the Canadian Virtual Brain Tumor Bank. The number of tissues analyzed in each group is shown in Table 1. To define the differences between low- and high-grade brain tumors, we analyzed 2 tumor subtypes which are known to have these subgroups: gliomas and choroid plexus tumors. We also compared low- and high-grade gliomas between children and adults, as these exhibit different clinical behaviors. In total, we analyzed 102 samples from 51 tumors by SNP array and 236 brain tumor, blood, and cell line samples (Table 1) for presence of MAE, by genotyping 34 SNPs across 21 candidate genes by Sequenom genotyping (Supplementary Table S1). We also used direct sequencing on a cohort of patients (Table 1) to assess the TP53 (all exons) and IDH1 (exon 4) genes for mutations (Supplementary Table S2).

DNA and RNA were extracted from fresh frozen tumor tissues using standard phenol/chloroform extraction for DNA and TRIzol (Invitrogen) for RNA (see Supplementary Methods for details). The concentration of each nucleic acid was obtained using a NanoDrop spectrophotometer (Thermo Fisher Scientific). In addition, DNA samples were analyzed using a 1% TBE-agarose gel to assess DNA integrity, and RNA samples were analyzed using the Agilent 2100 Bioanalyzer (Agilent Technologies). Nuclear and cytoplasmic RNA were extracted from cell lines using a commercial kit (Norgen Biotek Corporation), following the manufacturers instructions.

All nucleic acids were assessed using Affymetrix Human SNP 6.0 arrays (Affymetrix) and all microarrays were processed by The Centre for Applied Genomics (TCAG) at The Hospital for Sick Children, Toronto. DNA was processed according to standard Affymetrix protocols. RNA samples were processed according to the GeneChip Whole Transcript (WT) Double Stranded Target Assay protocol, with the following exceptions: (i) the starting amount of RNA was 2 μg and (ii) the cRNA sample generated by the first synthesis cycle was divided into 2 reactions rather than 3, to ensure the volume of double stranded cDNA would be sufficient for hybridization.

Cell lines were maintained as follows: Med8a and D283 cells in DMEM (Dulbecco's modified Eagle's medium; Wisent), UW228, Daoy and fibroblasts 158FB and 1604FB in AMEM (Wisent), D425 and D458 in IMEM with 20% FBS and 1.1 g bicarbonate (Wisent), ONS76 and RES256 (2% FBS) in DMEM/F12 (Wisent) and U87 and U251 in DMEM media (Wisent) with 1% l-glutamine, 1% sodium pyruvate and 1% nonessential amino acids. All media were supplemented with 10% FBS (Invitrogen) unless indicated. Fetal neural stem HF240 cells and glioma tumor-initiating cell lines G179 and GLINS1 were maintained as previously described (17).

Affymetrix Genotyping Console 4.0 was used to genotype the DNA and cDNA samples, using default parameters. Genotypes for untranslated region (UTR) and exonic SNPs were then exported to allow comparison between DNA and cDNA pairs from the same samples. The genotypes for paired gDNA and cDNA samples at each SNP were then compared using a php script, which scored the SNP, depending on whether the gDNA/cDNA genotypes were identical, the cDNA was homozygous when the gDNA was heterozygous, or if there was no result for either gDNA or cDNA. This data was used to determine genome-wide rates of homozygous and MAE genotypes. In addition, gene-based analysis of the SNP data was also carried out for individual samples and groups of samples; genes were only included in this analysis if they were genotyped by at least 2 SNPs. We also compared our overall MAE rates with previously published data about genome-wide predictions of MAE in normal tissues (3). To determine whether the MAE we observed in the genome-wide data was potentially due to allele-specific copy number changes, we examined whether any of our short listed genes were located within regions of copy number change. We found this not to be the case for these samples and genes.

The MassARRAY iPlex system (Sequenom) was used to genotype DNA and cDNA samples. The Sequenom software was used to determine gDNA and cDNA genotypes, and a trained technician then confirmed individual sample genotypes to ensure genotype accuracy. Samples for which the allele ratio was greater than 0.7:0.3 were considered to show differential allelic expression. The Sequenom system is also able to show changes in gDNA copy number; however, we observed no such changes in the SNPs analyzed using this system.

All statistical analyses were done using SPSS v15.

To assess the magnitude of MAE in brain tumors in an unbiased manner and to uncover genes that use MAE and contribute to tumorigenesis, we carried out a genome-wide analysis of MAE using Affymetrix SNP 6.0 arrays in a series of 51 low- and high-grade brain tumors (Table 1). For this analysis, we hybridized paired DNA and cDNA from primary tumors to Affymetrix SNP 6.0 chips, to carry out genome-wide genotyping of the DNA and cDNA. To detect MAE, we only considered SNPs for which the DNA genotype was heterozygous, and we removed all homozygous DNA genotypes prior to any analysis. Thus, any cDNA genotype that was subsequently homozygous was considered to display MAE.

To confirm the validity of the Affymetrix SNP 6.0 array in detecting MAE in a genome-wide manner, we examined heterozygous DNA genotypes from the X chromosome of female patients. As expected from the lyonization principle, we observed complete LOH in the cDNA samples, indicating X inactivation (Supplementary Fig. S1). To further validate the utility of this platform to diagnose MAE, we analyzed the frequency of MAE in a subset of imprinted and olfactory receptor genes, which are known to display MAE (1, 2). We observed a median of 82% of informative olfactory receptor genotypes and 50% of informative imprinted gene genotypes were MAE. We confirmed by sequencing the genotypes for SNPs within the CPA4 gene, which is known to be imprinted, with 100% accuracy.

To determine the role of MAE in low- and high-grade tumors, we investigated the genome-wide rates of DNA homozygosity and MAE within each brain tumor type. As expected, the percentage of SNPs with homozygous DNA genotypes was very high (>80%) in most tumors and higher (albeit not significantly) in high-grade tumors versus low-grade tumors (Table 2). Importantly, for the remaining informative (heterozygote) SNPs, MAE was significantly greater in the pediatric high-grade gliomas (pHGG) versus pediatric low-grade gliomas (pLGG; P = 0.012) and the choroid plexus carcinomas (CPC) versus choroid plexus papillomas (CPP; P = 0.0011; Table 2). Genome-wide rates of SNP MAE varied between 28% to 45% of informative genotypes. In addition, the gene-based analysis showed that the number of genes involved in MAE was higher in the pediatric high-grade tumors, though this was only significant in the choroid plexus tumors (CPC vs. CPP, P = 0.006; pHGG vs. pLGG, P = 0.063, Fig. 1A). Taken together, homozygous expression of alleles (either through DNA LOH or MAE) was greater in pediatric high-grade tumors compared with low-grade tumors, suggesting an important role for MAE in identifying genes involved in the clinical and biologic differences in these tumor subtypes.

In contrast, the frequency of MAE genes was not significantly different between the adult LGG and HGG, which in addition have a very broad range of MAE rates (Supplementary Table S4). This is consistent with the nature of adult gliomas, which invariably progress from low to high grade, whereas pediatric tumors rarely do so.

As the global, genome-wide MAE rate was greater than the MAE rate observed in the gene-based analysis, we hypothesized that MAE is gene specific. To further investigate this hypothesis, we compared MAE rates in particular gene subgroups (tumor suppressors and oncogenes) and also employed a gene-specific approach to analyze known brain tumor–associated genes (see below).

To determine whether MAE plays a differential role in tumor suppressors versus oncogenes, we investigated the relative frequency of MAE in known tumor suppressors (TS) and oncogenes, as described on the CancerGenes website and via literature searches (Supplementary Table S3). Although the rate of MAE in oncogenes and TS was similar within the low- and high-grade tumor cohorts (Fig. 1B), oncogenes had significantly higher MAE in high-grade tumors than in low-grade tumors (31% MAE vs. 18% MAE, respectively; P = 0.010). There was no significant difference in the rate of MAE in tumor suppressors between low- and high-grade tumors (P = 0.122). This suggested that high-grade tumors use allele-specific expression of oncogenes during transformation, potentially augmenting mutant or altered alleles.

To identify novel tumor-associated genes, we looked specifically at genes with high rates of MAE in each of the tumor types. Using the genome-wide data generated from the Affymetrix SNP 6.0 arrays, we initially restricted our gene list by including only those genes which were informative in more than 75% of tumors within at least one sample group; this yielded a short list of 56 genes. We then identified genes that displayed MAE in 100% of these samples, to identify the genes most affected by MAE. We identified 9 genes that displayed MAE in all informative samples (Table 3). Strikingly, 7 of these 9 genes were reported in the literature to be associated with cancer. Although the proportion of informative cases for these genes in each tumor group varied, all informative samples showed complete MAE at these loci.

Having identified genome-wide differences between different tumor types, we investigated the extent of MAE in known brain tumor–associated genes. To achieve this, we used a semiquantitative genotyping method (Sequenom) to analyze a large cohort of tumor and normal samples for multiple coding SNPs within genes of interest (Supplementary Table S1).

We observed a global increase in MAE/DAE with increasing tumor phase, from normal brain (∼7%), through tumors at time of biopsy/resection (∼11%), tumors sampled at autopsy (∼15%) to cell lines (∼27%; Fig. 2A). Interestingly, MAE was observed in only a minority of genes (Fig. 2B) and only 5 genes had more than 10% MAE/DAE; TERT, the gene with the highest rate of MAE, was recently identified in genome-wide association studies as a risk factor for development of gliomas (18, 19). All other genes, except MSH3 (P = 0.2) and cMYC (P = 0.06), had significantly lower MAE/DAE rates compared with TERT (Fig. 2B).

The highest rate of MAE/DAE at 35% was observed in hTERT. MAE/DAE at hTERT varied according to tumor type and phase (Fig. 2C). Specifically, a striking difference was observed between adult and pediatric tumors; more than 50% of adult gliomas showed aberrant hTERT expression compared with approximately 30% of pediatric gliomas (P = 0.0224, Fisher's exact test). Similarly, the cell line/autopsy tumor samples which represent late phases of tumor evolution had even higher MAE/DAE, in which more than 80% of informative cell lines had aberrant hTERT expression. The presence of MAE in the tumors and cell lines was confirmed by direct sequencing. Interestingly, we observed a tendency toward one allele in samples displaying MAE (67% of MAE samples expressed the G allele rather than the A allele for rs2736098); however, there was no difference in allele expression according to tumor grade.

In an attempt to elucidate the mechanisms underlying the observed allele-specific expression at the hTERT locus, we first sought to localize the changes in expression, to either pre- or posttranscription. We examined nuclear and cytoplasmic RNA extracts from informative cell lines and found no differences in allelic expression between the 2 cellular compartments, suggesting that the allele-specific alterations in hTERT expression were due to pretranscriptional mechanisms.

More than 50% of cancers and a significant number of advanced gliomas (20) and CPC (21) harbor either a heterozygous or monozygous somatic TP53 mutation with LOH. We examined the rate of TP53 MAE using the common codon 72 polymorphism (rs1042522) and TP53 mutation status in peripheral blood lymphocytes and brain tumors from different phases (Fig. 3A). We found no MAE in blood samples [including those from Li–Fraumeni syndrome (LFS) patients, who harbor germline heterozygous TP53 coding mutations]. Furthermore, pLGGs that do not harbor TP53 mutations did not reveal MAE. In contrast, adult LGGs had an MAE rate of approximately 20%, in addition to the 43% of samples with homozygous TP53 mutations, resulting in a total of 63% LOH at expressed alleles. For high-grade tumors, which showed the same trend in pediatric and adult samples, approximately 20% of heterozygous samples showed MAE, in addition to the 58% of tumors with homozygous TP53 mutations, for a total of 78% LOH of expressed alleles. Furthermore, we observed a high rate of homozygous TP53 mutations in both autopsy tumors (83%) and established long-term cell lines (100%). Finally, to determine whether MAE in a mutant TP53 tumor could be induced over time, we examined the effect of implanting a TP53 double heterozygote (c. C455T and G797A) patient-derived glioblastoma cell line into mouse cortex. The xenograft tumor, which previously expressed both alleles, displayed a shift in the allele ratios revealing DAE toward the mutant G797A allele and the normal C455T allele (Fig. 3B). Together, these findings showed association between increasing frequency of homozygous TP53 mutation at the expression level with increasing glioma grade and disease stage.

The phenomenon of MAE at TP53 mutation loci was tumor specific because paired blood and tumor DNA/RNA from LFS patients who had MAE in their tumor did not reveal MAE in their blood samples. Furthermore, adult low-grade tumors with MAE predominantly expressed only the normal allele (66% vs. 37% with MAE to the mutant allele), whereas the high-grade tumors predominantly expressed the mutated allele (66% vs. 37% with MAE to the normal allele; data not shown).

To identify the mechanism for MAE at the TP53 locus, we considered 2 explanations: (i) the presence of promoter polymorphisms that affect transcription factor binding sites, thus leading to allele-specific expression; and (ii) polymorphisms within the 3′-UTR that might affect transcript degradation by altering microRNA attachment, leading to reduced degradation of one allele over the other. We sequenced the reported 2,024 bp of the promoter region (Fig. 3C) in 6 samples with and without MAE. Although we identified a number of known polymorphisms within some of the samples, we did not identify promoter polymorphisms in any MAE sample. We also sequenced the 3′-UTR of known microRNA attachment sites but did not find any mutation or allelic change.

We then investigated the role of MAE at another frequently mutated gene in adult gliomas, the IDH1 (exon 4). We sequenced exon 4 of IDH1 in an unbiased cohort of 31 adult low-grade tumors and 46 adult high-grade tumors. Consistent with previous data (22), we found a mutation rate of 71% (n = 22) in the adult low-grade tumors and 23% (n = 11) in the adult high-grade tumors. Of the mutated tumors, 5 (15%) showed MAE. We then investigated the role of MAE at the IDH1 locus in a second cohort of IDH1 mutated adult gliomas (42 LGG and 25 HGG), with clinical data available for analysis. Within the clinical cohort, we identified IDH1 MAE in 12 of 67 tumors (17.9%). In 10 of 12 cases, MAE was specific for the normal allele, which is associated with reduced survival. Clinical correlation within the low-grade glioma cohort revealed significantly higher patient survival (P = 0.04) for mutated tumors that did not have MAE, compared with patients whose tumors had MAE toward the normal allele (Fig. 4). This data agrees biologically with previous data, which has shown a survival advantage in patients with mutations, compared with those without IDH1 mutations (23, 24).

Monoallelic expression of genes by allele-specific silencing has been described and is a common mechanism in imprinted genes, immune receptors, and olfactory receptors. The existence of genome-wide random MAE in normal tissues has recently been described (3, 8, 10). However, the extent of MAE in brain tumors, and its biologic and clinical significance are still unknown. This study shows that MAE is relatively common in brain tumors. It is a gene- and tumor-specific event rather than a general mechanism and may have a cardinal role in tumor progression, as well as clinical implications.

The extent of MAE in normal tissues ranges from 5% to 20% (3, 8, 10), with approximately 10% of genes showing MAE when analyzed using Affymetrix SNP arrays. This is in agreement with the levels of MAE we observed for most genes in the candidate gene study and the levels of MAE found in blood samples. We observed a continuous trend of low to high MAE from normal tissue, low-grade and high-grade tumors originating from the same tissue, and late established tumor cells, such as autopsy biopsies and cell lines. This finding was consistent on a single-gene level (TP53 and hTERT, Figs. 2C and 3A) as well as on a genome-wide level, in which the average level of MAE within the brain tumor samples was approximately 20% (Fig. 1A) compared with approximately 10% in normal cell lines (3). Furthermore, because DNA LOH increases with tumor progression, the addition of MAE will be associated with a dramatic increase on the rate of allelic expression throughout the process of tumorigenesis. Vogelstein described the process of progression from premalignant lesions to high-grade tumors in colon cancer through stochastic acquisition of mutations in multiple cancer-related genes (25). Our observations suggest that MAE may precede or enhance a mutation by expression of only the mutant- or disease-related allele and by that may add a new dimension to Vogelstein's prediction.

Our investigation of MAE within specific classes of genes revealed a significant difference in the MAE frequency between low- and high-grade tumors, with fewer oncogenes showing MAE in low-grade tumors. These data suggest that MAE in the context of brain tumors is predominantly advantageous to the tumor, if one considers the biologic situation that predicts that low-grade tumors will have fewer oncogenes disrupted than their high-grade counterparts.

Recent studies have highlighted the potential role for microRNA in cancer development (26, 27), and TP53 has a number of microRNA binding sites within the 3′-UTR. Although we sequenced the entire promoter, 5′-UTR and 3′-UTR regions, we were able to discard polymorphisms in regulatory regions as the basis for MAE at the TP53 locus. However, in investigating MAE in the TP53 gene, we show that MAE can contribute to tumor progression within a single gene. We observed no MAE in tumor samples that do not have TP53 mutations and that do not progress but observed high rates (∼20%) of MAE in tumors that have TP53 mutations and are progressive. Although one could argue that the mutated transcript is quickly degraded, thus leaving only one allelic form to be detected, we found MAE of both the normal and mutated alleles. Indeed, we observed that low-grade tumors predominantly display MAE of the normal allele, whereas high-grade tumors predominantly express only the mutated allele. These data suggest a tantalizing notion that more malignant tumors are “using” MAE at the RNA level to generate homozygously expressed TP53 mutations, while still possessing both alleles at the DNA level. This hypothesis agrees with the genome-wide framework, in which we observe more MAE within high-grade tumors compared with nonprogressing low-grade tumors (i.e., pLGGs and CPPs).

The clinical implications of MAE are yet to be realized. Nevertheless, our preliminary findings suggest that MAE can be used to predict survival in tumors harboring mutations that are of prognostic relevance. Our analysis of IDH1 mutations revealed a clinically significant distinction between mutated MAE tumors and mutated non-MAE tumors. Specifically, mutated IDH1 gliomas with MAE toward the normal allele have worse survival than non-MAE tumors (Fig. 4). This adds a new dimension to the recent studies which showed that mutations at IDH1 are predictors of better survival in gliomas (24, 28) and can help tailor therapies for these patients.

In summary, we have shown that in brain tumors, a high level of MAE is present and is tumor and gene specific. Furthermore, MAE may be a mechanism employed by cancers to achieve a more aggressive phenotype. Finally, genome-wide search may identify novel genes involved in these cancers. Additional studies are required to determine the mechanisms that govern MAE and allow for a full appreciation of the role of MAE in disease development.

No potential conflicts of interest were disclosed.

The authors thank all of the patients and their families who were involved with the study, as well as the Canadian Virtual Brain Tumor Bank and its affiliate members for the adult tumor tissues and also thank staff at TCAG for assistance and advice with the microarray processing, and staff at AGTC for assistance with processing and analysis of the Sequenom genotype data.

This work was supported by funds received from The Hospital for Sick Children Comprehensive Cancer Centre to U. Tabori, D. Malkin, and E.J. Walker, and in part by funds received from a grant of the Canadian Institutes of Health Research (#MOP-97834) to D. Malkin.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.