Transcriptome analysis reveals distinct gene expression profiles in astrocytoma grades II-IV

Background. Astrocytoma is the most prevalent form of primary brain cancer categorized into four histological grades by the World Health Organization. Investigation into individual grades of astrocytoma by previous studies has provided some insight into dysregulation of regulatory networks associated with increasing astrocytoma grades. However, further understanding of key mechanisms that distinguish different astrocytoma grades is required to facilitate targeted therapies. Methods. In this study, we utilized a large cohort of publicly available RNA sequencing data from patients with diffuse astrocytoma (grade II), anaplastic astrocytoma (grade III), primary glioblastoma (grade IV), secondary glioblastoma (grade IV), recurrent glioblastoma (grade IV), and normal brain samples to identify genetic similarities and differences between these grades using bioinformatics applications. Results. Our analysis revealed a distinct gene expression pattern between grade II astrocytoma and grade IV glioblastoma (GBM). We also identified genes that were exclusively expressed in each of the astrocytoma grades. Furthermore, we identified known and novel genes involved in key pathways in our study. Gene set enrichment analysis revealed a distinct expression pattern of transcriptional regulators in primary GBM. Further investigation into molecular processes showed that the genes involved in cell proliferation and invasion were shared across all subtypes of astrocytoma. Also, the number of genes involved in metastasis, regulation of cell proliferation, and apoptosis increased with tumor grade. Conclusions. We confirmed existing findings and shed light on some important genes and molecular processes that will improve our understanding of glioma biology.


INTRODUCTION
Astrocytoma is the most prevalent form of brain cancer 1 .Previous studies have suggested that astrocytoma originates from cancer cells with stem-cell-like properties [2][3] .The World Health Organization (WHO) classified astrocytoma into four histological grades (Grade I to IV) of increasing malignancy 4 .Grade II diffuse astrocytoma (AS II) is a slowly growing, invasive, semi-benign astrocytoma that is frequently diagnosed in younger patients, aged between 20 and 45 years and with an average age of 35 years 5 .Histologically diffuse astrocytoma is not clearly distinguishable from surrounding normal tissue making it difficult for surgical resection 6 .Recurrence in most patients is observed after a few years, often progressed to more malignant grade III or grade IV Glioblastoma [7][8][9] .Grade III Anaplastic astrocytoma (AS III) is a fast growing, invasive, malignant astrocytoma.Tumor cells in AS III also vary in size and shape in comparison to AS II and were previously described to have an increased mitotic activity 4 .Mean age of patients diagnosed with AS III is 45 years.Patients undergo surgical resection, followed by chemotherapy and/or radiotherapy, as preferred means of treatment.Around 24% of the patients with AS III have an overall survival rate of five years 10 with a median survival between one to four years 11 .The majority of AS III cases progress into grade IV glioblastoma after a few years [7][8] .
WHO grade IV Glioblastomas (GBM IV) are very fast growing and the most malignant type of astrocytoma.GBM IV can be characterized by the presence of necrosis and/or vascular proliferation.Glioblastoma IV can be genetically distinguished into two different types: 1).Primary GBM (pGBM IV), which occur rapidly without prior occurrences of lower grade astrocytoma 8,12 and constitutes the majority of the GBM IV. 2).Secondary GBM (sGBM IV), constitute only 5% of the total GBM IV cases and progress gradually from lower grade astrocytoma over several years 13 .Primary GBM is histologically indistinguishable from secondary GBM.However, sGBM are distinguishable from pGBM by the presence of IDH1 and IDH2 (Isocitrate Dehydrogenase 1/2 (NADP+)) gene mutations which are also present in AS II and AS III (ref. 14).Poor survival rate is observed in patients with GBM with less than 5% of patients surviving longer than five years 13 .
However, the majority of these studies have either used no or a small number of normal brain samples in describing gene expression differences between astrocytoma grades 20,[26][27][28] .In this study, with the help of publicly available RNA sequencing data from patients with AS II, AS III, and GBM IV, and normal brain samples, we aimed to describe genetic similarities and differences between these subtypes and also to identify key regulatory networks and signaling pathways that they share and also those that set them apart.

Samples
RNA sequencing data used in this study was obtained from previously published studies as detailed in Table1.Normal samples included RNA sequencing data from 3 different studies that included, 2 epileptic patients, 5 Ambion Human Brain Reference RNA (HBRR), and 61 single cell (astrocytes) transcriptomes from epileptic patients.Astrocytoma grades II Diffuse astrocytoma, grade III anaplastic astrocytoma, and grade IV glioblastoma and corresponding normal brain samples were considered in this study (see supplementary Table 1 (S1)).No ethical approval was sought for this study as all the data used in this study was obtained from published studies that have obtained ethical approval.

RNA sequencing data analysis
Raw sequence reads were obtained from GEO (Gene Expression Omnibus; RRID: SCR_005012) in the form of FastQ files.Reads from FastQ files were aligned to human reference sequence hg19 (RRID: SCR_006553) using Rsubread aligner 29 .Mapped sequencing reads were assigned to hg19 refGene using featureCounts (RRID: SCR_012919).RefGene exon coordinates were obtained from UCSC table browser (RRID: SCR_013787).Differentially expressed genes were identified using edgeR (RRID: SCR_012802) (ref. 30).Genes with read count per million > 1 in at least half of the samples were considered for further analysis.The data was normalized using TMM (trimmed mean of M values) method.

Gene set enrichment analysis
Gene set enrichment analysis was done using Molecular Signatures Database (MSigDB; RRID: SCR_003199) to identify pathways and underlying biological themes.GSEA tool (RRID: SCR_003199) was used to investigate enrichment of 8 gene sets (tumor suppressor genes, oncogenes, translocated cancer genes, protein kinases, cell differentiation markers, homeodomain proteins, transcription factors/co-factors, and cytokine and growth factors) and the cancer census gene list 31 in differentially expressed genes in our study.

Pathway analysis
Pathways enriched in differentially expressed genes were investigated using Molecular Signatures Database (MSigDB; RRID: SCR_003199).MSigDB evaluates the overlap of our genes sets with its collections and provides an estimate of the statistical significance (P and FDR q-value).We also used cytokine-cytokine interaction pathway (hsa (Homo sapiens): 04060) and glioma pathway (hsa: 05214) maps from Kyoto Encyclopedia of genes and genomes (KEGG; RRID: SCR_012773) to highlight key genes from our study.

Statistical analysis
Statistical analysis was done using the R language and environment (RRID: SCR_003005).The Heatmap3 program was used to generate the heat map for differentially expressed genes in R (ref. 32).

S imilarities and differences in gene expression pattern in astrocytoma grades II-IV
I n our global gene expression analysis for grades II -IV astrocytoma, distinct gene expression pattern was observed between astrocytoma grade II and grade IV (F ig. 1).A total of 8936 over-expressed genes and 4699 under-expressed genes were observed in all AS II-IV subtypes when compared to all normal brain samples (log fold change (FC) ≥ 2, P ≤ 0.001 and FDR ≤ 0.002) (see supplementary Table 2

(S2))
. There was high overlap between grade IV subtypes in our study.Furthermore, we also observed a common expression pattern for grades II-IV samples for a group of genes (Fig. 1) (list of genes can be seen in supplementary Table 3 (S3)).Differential gene expression analysis of astrocytoma grades AS II, AS III, pGBM, sGBM and rGBM revealed 773, 149, 3963, 280, and 304 over-expressed and 889, 200, 1859, 377, and 390 under-expressed genes, respectively (log FC ≥ 2, P ≤ 0.001, and FDR ≤ 0.002) (see sup plementary Table 2 (S2)).The number of under-expressed genes was higher than the over-expressed genes in AS II, III, sGBM and rGBM, whereas in pGBM the number of overexpressed genes was more than that of under-expressed genes.The number of differentially expressed genes per sample in primary GBM was higher (3 fold for over-expressed genes) than the rest of the astrocytoma grades.137 over-expressed genes and 191 under-expressed genes were common to all astrocytoma grades (Suppleme ntary Fig. 1).The number of shared over-expressed and underexpressed genes in grade IV GBM was 247 and 330, respectively (Supplementary Fig. 2 and the list of genes can be seen in supplementary Table 3 (S3)).
51 over-expressed genes and 42 under-expressed genes were found to be unique to AS II.Over-expressed genes include transcription factor genes (THRA, KLF15, and FHL1); protein kinase gene DCLK2; Cytokine and growth factor genes (IL17D, CMTM4 and SEMA6A); and Oncogenes (TFRC, MDM2, ETV5).Transcription factor gene ENO1 and cytokine and growth factor gene PTN were exclusively under-expressed in AS II.Overexpression of the ITM2C (BRI3) gene and under-expression of 3 genes; ADAM20, FAM133CP and LINC00504 were exclusive to AS III (see sup plementary Table 3 (S3)).3236 over-expressed and 1017 under-expressed genes were unique to grade IV glioblastoma subtypes.Genes previously reported in brain cancer, such as BIRC3, BRCA1, EGFR, ERBB2, PDGFR1B, and VEGFA were over-expressed and FGFR2 and NTKR3 were under-expressed exclusively in grade IV GBM, albeit, at varying degrees of significance (see supplementary Table 3 (S3)).In primary GBM, 3192 and 994 genes were exclusively over-expressed and under-expressed, respectively (see supplementary Table 3 (S3)).SOX4 and SALL3 were only genes overexpressed in secondary GBM in our study.No underexpressed genes were found in our study.Out of 3 genes, unique to recurrent GBM, FTL and CTSL genes were over-expressed and MT1JP gene was under-expressed.

Gene expression pattern previously reported in glioma
We found significant overlap (P ≤ 0.001) of differentially expressed genes in AS II-IV between our study and to those of a previously published study using microarray data 28 (see sup plementary Table 4 (S4)).36 genes were upregulat ed in grades II-IV astrocytoma in both studies.Among these 36 genes known cancer genes such as HIP1, HLA-A, MED12, and NOTCH1 genes were observed.AKT2 and EGFR genes were upregulated in grade IV astrocytoma in both studies.We observed 21 downregulated in grades II-IV astrocytoma in both studies which also include a cancer census gene WIF1.CALM2 (Calmodulin 2) gene was shared across all subtypes in both studies.Furthermore, we found 8 macrophage/microglia markers (TGFBI, CTSC, CD14, CD4, TLR2, ITGB2, TGFBR2, and HEXB) upregulated in grade IV astrocytoma in our study (pGBM) and Seifert et al. study 28 .Macrophage markers were absent in shared genes for lower grade astrocytoma (AS-II and III) in these two studies (see supplementary Table 4 Furthermore, a literature review of the differentially expressed genes in our study revealed a similar expression pattern in genes (highlighted in bold letters in Table 2) that were previously implicated in glioma-genesis.Among these genes, MKI67 (Marker Of Proliferation Ki-67), EGFR (Epidermal Growth Factor Receptor), RTKs (Receptor tyrosine kinases), collagens, G-coupled receptor proteins genes and ligands, home box genes, chemokine receptors and ligands, and PDGF/VEGF growth factor family genes were among the most significantly over-expressed genes in AS II-IV (see Table 2 and supplementary Table 2 (S2)).EGFR gene is a well-known diagnostic marker for gliomas which is a receptor tyrosine kinase of the ErbB family, known to activate at least 4 major downstream signaling cascades including the RAS-RAF-MEK-ERK, PI3 kinase-AKT, PLC gamma-PKC and STATs modules.This gene was previously reported to be over-expressed in GBM and was significantly overexpressed in all GBM subtypes in our study (pGBM (4.06 log FC), sGBM (1.95 log FC), and rGBM (2.57log FC).Also, SMARC4 (involved in controlling cell proliferation, migration, and invasion) which has been reported to be over-expressed in gliomas was significantly over-expressed in our study, (pGBM (1.87 log FC), sGBM (2.46 log FC), and rGBM (2.05 log FC) (ref. 33).
MALAT1 (Metastasis-associated lung adenocarcinoma transcript 1) was under-expressed in all astrocytoma grades in our study.It is a type of long noncoding RNA and its over-expression was reported to be associated with decreased metastasis and is a favorable prognostic factor as it plays a vital role in the reduction of extracellular *Genes that were previously reported in astrocytoma were highlighted in bold letters.** All Up-regulated genes in this table has log FC of > 3.7, P ≤ 0.000002 and FDR ≤ 0.00002) ***All downregulated genes in this table has log FC of < -5, P ≤ 0.00002and FDR ≤ 0.000009) signal-regulated kinase/mitogen-activated protein kinase (ERK/MAPK) signaling activity and expression of matrix metalloproteinase 2 (MMP2) (ref. 34).GJB6 (gap junction beta-6) gene was significantly under-expressed in all astrocytoma grades in our study and was previously reported to be deleted and/or mutated in glioma 35 .WIF1 (Wnt inhibitory factor 1) gene is a candidate tumor suppressor gene in glioblastoma and was reported to be downregulated in most glioblastomas 36 .WIF1 gene was under expressed in all AS grades in this study (Table 2).SLC1A2 (Solute Carrier Family 1 Member 2) gene was under-expressed in all astrocytoma grades in our study.Downregulated SLC1A2 gene assists in metastasis in glioma 37 .Underexpressed BMPR1B (bone morphogenetic protein receptor IB subunit) gene may reduce the malignancy of glioblastoma cells by upregulation of p21 and p27Kip1 (ref. 38).BMPR1B was under-expressed in all astrocytoma grades in this study.LAMP1 (Lysosomal-associated membrane protein-1) gene was significantly over-expressed in all astrocytoma subtypes in our study (Table 2).This gene may play a role in metastasis in astrocytoma 39 .PFN1 (Phosphorylation of profilin-1) gene directs the angiocrine expression and glioblastoma progression through HIF-1α accumulation 40 and was over-expressed in grade IV glioblastoma subtypes in our study.YBX1 (Y-box binding protein-1) gene was over-expressed in all astrocytoma grades in our study (Table 2).It is a member of the cold shock protein family and functions in transcription and translation and is reported to be highly expressed in tumor cells and as a marker for tumor aggressiveness and clinical prognosis in various types of cancers, including glioblastoma 41 .CDH4 gene, which encodes for a cell adhesion protein involved in neural outgrowth and brain segmentation was under-expressed in AS II (log FC -1.97) (ref. 42).Notch ligand DLL3 known for its function in neurogenesis and glioma biology was over-expressed (log FC 8.99) (ref. 43).Apoptotic gene GAS2 (log FC -2.41) and SHROOM2 (log FC -2.21), involved in cell spreading, were under-expressed.Gene NR2E1, involved in anterior brain differentiation, was under-expressed (log FC -1.57).
Under-expression of NR2E1 gene was previously reported to be associated with cancer stem cell death and longer survival of G-CIMP glioma patients 44 .

Gene set enrichment analysis of differentially expressed genes in AS II-IV
Gene set enrichment analysis of 8 selected MSigDB gene family gene sets (as described in methods) and cancer census gene list revealed significant enrichment of all gene sets except tumor suppressor genes (P ≤ 0.001) for over-expressed genes in astrocytoma grades II-IV using.Significant depletion (P ≤ 0.001) of cell differentiation markers, homeodomain proteins, transcription factors/cofactors, and cytokine and growth factors was observed in the under-expressed genes (Fig. 2).Oncogenes and cancer census gene sets were significantly enriched (P < 0.05) in both over-expressed and under-expressed genes in each astrocytoma subtype (Supplementa ry Fig. 3 and 4).Tumor suppressor genes were not significantly enriched/depleted in any subtype except in over-expressed genes of pGBM (P Fig. 2. Functional annotations for all astrocytoma genes.Enrichment of tumor suppressor genes, oncogenes, translocated cancer genes, protein kinases, cell differentiation markers, homeodomain proteins, transcription factors and cytokine and growth factors; and Cancer Census Genes in differentially expressed genes in astrocytoma.Significant enrichment of genes in a category within a tumor type is represented by '*' (P < 0.05), '**' (P < 0.01) and '***' (P < 0.001) (Fisher's exact test).= 0.049) (Supplement a ry Fig. 3).Homeodomain proteins were not significantly enriched/depleted in any subtype except in pGBM, where they were significantly depleted in over-expressed genes (P ≤ 0.001) (Supplementary Fig. 3).Significant enrichment of translocated cancer genes was observed in all AS subtypes except in the under-expressed genes of recurrent GBM.Transcription factors/cofactors were significantly enriched in all subtypes except in the over-expressed genes in AS III and under-expressed genes in sGBM).Significant depletion of transcription factors/cofactors was observed in over-expressed genes of pGBM (Supplementary Fig. 3 and 4).Protein kinases were significantly enriched in under-expressed genes in all GBM, however, enrichment was also observed in the overexpressed genes of pGBM and AS II.Enrichment of cell differentiation markers was seen in under-expressed genes of AS II and pGBM, as well as in the over-expressed genes of AS II and all GBM.Cytokine and growth factors were significantly enriched in under-expressed genes of pGBM only (Supplementary Fig. 4).Significant enrichment of tumor suppressor genes, oncogenes, protein kinases, and translocated cancer genes was observed in both overexpressed and under-expressed genes unique to pGBM.Significant enrichment (in over-expressed genes) and de-pletion (in under-expressed genes) of cell differentiation markers was observed in pGBM and vice versa for transcription factors/cofactors (P < 0.05).Significant depletion of homeodomain proteins (in over-expressed genes) and of cytokine and growth factors (in under-expressed genes) unique to pGBM was observed (P < 0.05).

Expression of transcription factors in astrocytoma grades II-IV
Transcriptio n factors/co-factors was the most abundantly expressed group of genes in astrocytoma grades our study (Fig. 2).Distinct gene expression patterns were observed between AS II and all AS IV GBM for the 1446 transcription factor (TF) genes (Fig. 3).Primary GBM samples showed a different expression pattern compared to that of AS II and secondary GBM.A group of secondary GBM samples showed a similar gene expression pattern to that of AS II (Fig. 3).TFs in AS II were significantly enriched in cell cycle control, DNA repair, TGF BETA signaling, hypoxia, proliferation and apoptosis.Known oncogenes such as BRD4, CIC, ETV1, PATZ1, and PML and tumor suppressor gene KDM5C were upregulated, whereas oncogenes such as ETV5 and NFIB were downregulated in AS II.In AS III, TF genes were enriched in cell cycle (E2F) targets and chromatin binding.Oncogenes JUN (upregulated) and NFIB (downregulated) were observed in AS III subtype.Significant enrichment of TF genes involved in cell cycle control, DNA repair, TGF BETA signaling, hypoxia, proliferation, invasion, IL2-STAT5 pa thway and apoptosis was seen in primary GBM.Secondary GBM showed enrichment of TF genes in pathways similar to those of primary GBM in our study.TF genes in recurrent GBM significantly enriched in cell cycle control, TGF BETA signaling, hypoxia, proliferation and apoptosis.In grade IV GBM, TF genes include tumor suppressor genes (ERCC2, KDM5C, *This is a summary from 100 most significantly enriched pathways in AS II-IV (P ≤ 0.0013 and FDR ≤ 0.0018).*Y-indicates significant enrichment of that pathway.

DISCUSSION
Our study was focused on describing similarities and differences in gene expression in adult astrocytoma samples by transcriptome analysis.Our investigation revealed a distinct gene expression pattern between grade II and astrocytoma grade IV.This observation was similar to Fig. 4. Gene expression changes in glioma pathway.Key KEGG glioma pathway genes in our study.In this figure we have highlighted key KEGG glioma pathway genes (hsa: 04060) that were observed in pGBM.Upregulated genes (in red colored rectangles) and downregulated genes (in blue colored rectangles) from our study were highlighted in this figure.Genes that were not highlighted by colored rectangles were either absent or unchanged in our study.
previous studies which described that AS-II grade is less proliferative and less aggressive compared to grade IV astrocytoma 4 .IL2-STAT5 pathway genes were significantly enriched in grade IV astrocytoma alone and previous studies have reported that STAT5 regulates glioma cell invasion by pathways dependent and independent of STAT5 DNA binding 45 .The number of differentially expressed genes per sample in pGBM (n=58) and AS -II (n=34) were higher than those of AS-III (n=13), sGBM (n=19), and rGBM (n=19), however this difference could partly due to higher number of samples in these two subtypes.
Our data has confirmed gene expression pattern for genes that were previously reported to characterize the GBM IV, with an exception of expression pattern in genes MDM2, MDM4, AXL, FGFR3, and MERTK, which were reported to be over-expressed in glioma and were significantly under-expressed in all GBM subtypes in our study (pGBM (-2.92 log FC), sGBM (-2.14 log FC), and rGBM (-2.57log FC).MDM2 and MDM4 are known to inhibit a p53-dependent growth control [46][47][48] .However, amplification of the CDK4 gene and over-expression of cyclin D1 are suggested to have similar effects to p16 or pRb inactivation and these mechanisms may provide additional alternatives to subvert cell-cycle control and facilitate progression to GBM (ref. 49).CDK4, CCND1 ( CyclinD1) and E2F were over-expressed in primary GBM samples in our study, which may have contributed to the evasion of cell cycle control and facilitate G1/S progression (Fig. 4).
BIRC3 (CIAP2) encodes a member of the IAP family of proteins that inhibit apoptosis by binding to tumor necrosis factor receptor-associated factors TRAF1 and TRAF2.This gene was reported to be a transcriptional target of NF-κB/p65 and reduced levels of p65 renders glioma cells more sensitive to the cytotoxic effects of TNF-α, suggesting a good therapeutic target for the treatment of malignant gliomas [50][51] .
NTRK3 was reported to be under-expressed in highgrade gliomas due to DNA methylation changes, and was also significantly under-expressed in our study (pGBM (-2.09 log FC), sGBM (-1.22 log FC), and rGBM (-1.71 log FC) (ref. 52).TGFBI (transforming growth factor beta induced) gene was significantly over-expressed in our primary GBM samples.It encodes an RGD-containing protein that binds to type I, II and IV collagens.Collagens were also significantly over-expressed in our pGBM samples.In our study, pGBM samples were enriched in the genes involved in TGFBI module, namely, TGFBI, Collagens, MMP2 (matrix metalloproteinase 2), SPARC (secreted protein, acidic and rich in cysteine), and fibronectin (FN).TGFBI module may present as an alternative signaling pathway (to the canonical SMAD-mediated) TGF-beta signaling pathway in GBM (ref. 53).All the genes that are associated with this alternative pathway were significantly over-expressed in our AS grade IV pGBM samples (see supplementary Table 2

(S2)).
Genes exclusively over-expressed in AS II include the transcription factors such as THRA (Thyroid Hormone Receptor) gene which is a nuclear hormone receptor that can act as a repressor or activator of transcription and play a role in development (NOTCH1-mediated pathway for NF-KB activity modulation), KLF15 (which is a negative regulator of TP53 acetylation and also inhibits NF-kappa-B activation through repression of EP300dependent RELA acetylation), and FHL1 genes, Protein kinase gene, DCLK2, and Cytokine and growth factors, IL17D, CMTM4 and SEMA6A (ref. 54).
Interestingly, IL-17D gene expression was reported to be suppressed in more advanced, higher grade gliomas (grade III and grade IV) relative to less advanced, lower stage gliomas (grade II oligodendroglioma).IL-17D expression was also suppressed in Grade IV GBM when compared to Grade III astrocytoma, high expression of IL-17D in tumor biopsies correlated with a greater survival time for a subset of patients with grade IV GBM (ref. 55).Oncogenes TFRC, MDM2, ETV5, and Transcription factor ENO1, and cytokine and growth factor PTN are exclusively under-expressed in the AS II subtype in our study.Interestingly, MDM2 gene which was reported to be over-expressed in gliomas has under or un-altered expression in our study 56 .TFRC was postulated to be expressed in the normal brain and may be dysregulated in brain cancers 57 .Previous studies suggested ENO1 overexpression is associated with tumor progression in highgrade gliomas 58 .ENO1 was exclusively under-expressed in astrocytoma subtype AS II and non-significant (P < 0.05) over-expression of ENO1 gene was observed in AS-III and grade IV in our study.ITM2C (BRI3) was the only gene that was exclusively over-expressed in AS III subtype.ITM2C is a negative regulator of beta-amyloid peptide production, which inhibits the processing of APP by blocking its access to alpha-and beta-secretase.ITM2C was suggested to play a role in TNF-induced cell death and neuronal differentiation 59 .SOX4 and SALL 3 genes were exclusively over-expressed in secondary GBM in our study.SOX4 (sex determining region Y-box 4) was reported to be over-expressed in GBM (ref. 53).SOX4 was shown to transcriptionally activate EGFR by binding to its promoter 60 .Other growth factors that are reported to be targets of SOX4 are FGFRL1 and IGF2R (see supplementary Table 4 (S4).
Gene set enrichment analysis of differentially expressed genes in each AS subtype revealed important cancer-relevant signaling processes.Transcription factors/co-factors in astrocytoma grades were enriched in cell cycle control, DNA repair, TGF BETA signaling, hypoxia, proliferation and apoptosis, with the exception of grade III where enrichment of cell cycle (E2F) targets and chromatin binding was observed.Pathway analysis of each astrocytoma subtype revealed a significant enrichment of hallmark pathways in cancer such as evasion of apoptosis (e.g., Class I MHC pathway), growth and survival (e.g., Phosphatidylinositol signaling pathway), and proliferation and invasion (e.g., EGFR signaling, collagens, ECM, CAM, Chemokines, ERBB signaling) (see Table 3).Mapping of the genes that were differentially expressed genes in AS subtypes on KEGG glioma pathway revealed key genes involved in glioma-genesis (Fig. 4).

CONCLUSION
In conclusion, this study revealed similarities and differences in gene expression levels between grades II-IV AS subtypes by comparing their gene expression levels to those of normal brain samples, especially between grade II and grade IV astrocytoma.We identified genes that were exclusively expressed in each of the AS subtype.Pathway analysis of differentially expressed genes in each AS subtype revealed important cancer-relevant signaling processes.Furthermore, we may have identified an alternative signaling pathway (to the canonical SMAD-mediated) TGF-beta signaling pathway in GBM.A literature review of the genes identified in this study revealed that many were implicated in glioma pathogenesis and other cancers.We confirmed existing findings and shed light on some important genes and molecular processes that will improve our understanding of glioma biology.However, an integrated study involving much larger cohort augmented with methylation, whole genome and gene expression data from both tumor and normal brain samples will provide a greater understanding of glioma development and aid in personalized therapy.We believe our findings will provide new insights into the mechanisms of glioma pathogenesis and provide a valuable resource for future studies.

Fig. 1 .
Fig. 1.Differential gene expression in AS II -IV.Heat map of differentially expressed genes in astrocytoma subtypes.Heat map of 13635 genes expressed across different subtypes showing a distinct gene expression pattern between grade II and grade IV astrocytoma samples.Shared gene expression between all grades can be seen as well.

Fig. 3 .
Fig. 3. Expression of transcription factors in AS II -IV.Heat map of transcription factors in different grades of astrocytoma.Heat map was generated from a list of 1446 transcription factor genes in different astrocytoma grades.A clear difference between grade II and grade IV was observed.Primary GBM samples showed a different expression pattern when compared to grade II astrocytoma.A cluster of transcription factor genes were shared among grade IV subtypes.

Table 1 .
List of samples used in this study.

Table 2 .
List of 25 most significant differentially expressed genes in AS II-IV.

Table 3 .
Pathways implicated in astrocytoma grade II -IV.