Analysis of copy number variation profiles in brain tumors in the context of a methylation based classifier

Brain tumour’s range from benign neoplasm such as pilocytic astrocytoma to malignant ones e.g glioblastoma. Histopathological diagnosis of these entities is frequently challenged with inter-observer variability. Moreover, the used genome wide methylation patterns cannot grade tumour severity which is key in patient management. Although specific copy number variation (CNV) profiles such as 1p/19q co-deletions is known to characterise oligodendroglioma and joint gain of chr 7 and loss of chr 10 characterise glioblastoma, other CNV profiles have not been well integrated in brain tumour diagnosis. Therefore, it seems promising to achieve improvements in methylation-based diagnostics and disease prognosis by establishing an approach to systematically include CNV information in classification of brain tumours. With the aim of addressing this issue, in the first phase of my study, I evaluated whether methylation data (450K and 850K epic) could inform about the presence of CNVs. I used 61 paired data sets processed from microarray based comparative genomic hybridization (aCGH) and Epic 450K/850K methylation arrays respectively. Copy number plots of the methylation data set were generated from the “conumee” R-package while aCGH data set plots were inferred from the “DNA copy” package. I observed >80% percent agreement between the two methods. To rule out chance agreement and check the extent agreement, I calculated Kappa statistics. I observed moderate (0.54) to substantial (0.61) Kappa statistic values. In conclusion I provided evidence that the methylation data is reliable in determining CNVs. In the second phase, I evaluated the CNV profiles and survival times using Kaplan Meier analysis between WHO classified astrocytoma grade II and III data (n=117) obtained from the cancer genome atlas (TCGA). Before clustering, I observed no significant difference in survival in WHO grade II and III. After hierarchical clustering (Pearson coefficient correlation ward linkage) using the log2 CNV values, I was able to identify 7 clusters which had different survival rates. The clusters had both unique and shared alteration between them. For example, cluster 4 (n=10) showed better survival with deletions at Chr3q, 4q, 5p/q, 11p, 12q, 13q and gain in Chr12p. These regions carry genes such as ANO2, CD4, LRRC23, VWF and GALNT8 genes. Cluster 3 had poor survival and increased deletions at chr 1q, 2q, 3q, 4q, 5p/q, 6q, 7q, 11p, 13q and chr gain at 9p (n=54). Some key genes altered in these loci, included C2orf88, CDKN2A/B, RB1, SORBS2, POLD1, MYBPC2 and TP63. These genes play critical roles in cell cycle regulation, growth and tumour suppressions. Cluster 7 had losses at chr 4p/q, 13p/q and 19q (n=8) which contained genes like LRBA, FBXW7, MARCHF1, SPOCK3, MTUS2 and RFC3. Moreover, CDH12 gene and Long noncoding RNA (LINC005) regulating CCND2 at 5p and 13q respectively were also deleted in >75% of samples. I further noticed that glioblastoma recurrent cases and primary tumor could be differentiated by presence of chr7p/q gain, 9p, 10p/q and 13p/q deletions using a total of n= 1500 cases and n= 1400 controls data set retrieved from TCGA. The 9p and 10p/q loci are already known to encode cell survival and apoptotic genes such as CDK2A/B, MDM2, EGFR and PTEN which are common in high grade glioma. These results therefore promise better tumour diagnosis and patients stratification approach which would help in both patient management and treatment outcomes predictions by use of CNV profiles. In the third phase of my study, I evaluated the methylation classes and pathways associated with genes in the altered regions. I observed a different frequency in the distribution of Isocitrate dehydrogenase (IDH) mutation and the 06- methylguanine DNA methyltransferase (MGMT) in the 7 clusters. In specific clusters 1 and 6 were A_IDH 100% and 70% respectively. A_IDH_HG dominated the other clusters as follows: cluster 5 (50%), cluster 4 (33%), cluster 3 (13%) and cluster 7 (12%). This indicates that methylome classes can be aligned with the CNV profiles. Using ingenuity pathway-based knowledge, I was able to identify canonical pathways associated with altered genes per group. I observed that fairly unique signaling pathways were associated with the disease. Notably, PTEN, ERK/MAPK, P53, IL-3, Glioblastoma multiforme, glioma invasiveness and axonal guidance signaling which are associated with glioma formation are featured in most clusters. Key altered genes included adenomatous polyposis coli which is a tumor suppressor, Glycogen synthase kinase 3 beta (GSK-3β) which affects cell proliferation, retinoblastoma (Rb,) which codes a tumor suppressor rb protein while Platelet-derived growth factor (PDGF) and Phosphoinositide 3-kinases (P13K) both regulate cell growth and other cellular functions. Proto-oncogen Rat sarcoma (Ras), WNT, Son of Sevenless (SOS), Auditory processing deficit (APD) and beta catenin (CTNNB1) were also lost. The WNT pathway activation aids in cellular differentiation which promotes brain tumour formation while Ras /PI3K/RTK pathway contributes to tumour growth deregulation. These findings show that multiple pathways dysregulated by CNVs can help in establishing novel brain tumour stratification, diagnostics and consequently identification of novel drug targets