Transcriptional profiles of pilocytic astrocytoma are related to their three different locations, but not to radiological tumor features

Background Pilocytic astrocytoma is the most common type of brain tumor in the pediatric population, with a generally favorable prognosis, although recurrences or leptomeningeal dissemination are sometimes also observed. For tumors originating in the supra-or infratentorial location, a different molecular background was suggested, but plausible correlations between the transcriptional profile and radiological features and/or clinical course are still undefined. The purpose of this study was to identify gene expression profiles related to the most frequent locations of this tumor, subtypes based on various radiological features, and the clinical pattern of the disease. Methods Eighty six children (55 males and 31 females) with histologically verified pilocytic astrocytoma were included in this study. Their age at the time of diagnosis ranged from fourteen months to seventeen years, with a mean age of seven years. There were 40 cerebellar, 23 optic tract/hypothalamic, 21 cerebral hemispheric, and two brainstem tumors. According to the radiological features presented on MRI, all cases were divided into four subtypes: cystic tumor with a non-enhancing cyst wall; cystic tumor with an enhancing cyst wall; solid tumor with central necrosis; and solid or mainly solid tumor. In 81 cases primary surgical resection was the only and curative treatment, and in five cases progression of the disease was observed. In 47 cases the analysis was done by using high density oligonucleotide microarrays (Affymetrix HG-U133 Plus 2.0) with subsequent bioinformatic analyses and confirmation of the results by independent RT-qPCR (on 39 samples). Results Bioinformatic analyses showed that the gene expression profile of pilocytic astrocytoma is highly dependent on the tumor location. The most prominent differences were noted for IRX2, PAX3, CXCL14, LHX2, SIX6, CNTN1 and SIX1 genes expression even within different compartments of the supratentorial region. Analysis of the genes potentially associated with radiological features showed much weaker transcriptome differences. Single genes showed association with the tendency to progression. Conclusions Here we have shown that pilocytic astrocytomas of three different locations can be precisely differentiated on the basis of their gene expression level, but their transcriptional profiles does not strongly reflect the radiological appearance of the tumor or the course of the disease. Electronic supplementary material The online version of this article (doi:10.1186/s12885-015-1810-z) contains supplementary material, which is available to authorized users.


Background
Pilocytic astrocytoma (PA) is the most common type of brain tumor in the pediatric population, comprising approximately 25 % of all primary tumors, with the most frequent occurrence taking place between 5-10 years of age. Fortunately this tumor has a generally good outcome, however recurrences or leptomeningeal dissemination are also sometimes observed. PAs can affect various anatomical structures, but there are three most common locations: cerebellum, optic tract with hypothalamus, and cerebral hemispheres. Those tumors are mainly sporadic, except for cases occurring in patients with neurofibromatosis type 1 and, less frequently, with Frasier and Noonan syndromes [1][2][3]. Molecularly, pilocytic astrocytoma is characterized by a relatively small number of chromosomal abnormalities with the most common alteration located at chromosome 7q34 comprising the BRAF oncogene [4,5]. In the highthroughput analysis era, limited reports of this type of tumor were made using expression profiling. The presupposition concerning the molecular heterogeneity of pilocytic astrocytomas was defined previously by Wong et al. as the result of unsupervised hierarchical clustering. Without inference from a clinical outcome, they identified two subgroups of tumors. Such results could be a consequence of including two cases of subtotally resected tumors and, more importantly, the more aggressive variant of astrocytoma with pilomyxoid features [6]. Later, an assumption describing different expression profiles for PAs of various locations was given by Sharma et al., who showed the LHX2 gene expression to be connected with supratentorial location [7]. A following analysis made by Tchoghandjian et al. showed upregulation of LHX2 together with SIX6 in tumors originated from the hypothalamo-chiasmatic region [8]. At the same time, MATN2 and ALDH1L1 genes were assumed to be connected with plausible PAs progression despite total surgical resection [9,10]. Children affected by this tumor usually have a good prognosis, although in some cases recurrence or leptomeningeal dissemination may be observed [11][12][13][14]. Thus there is an ongoing need to search for molecular markers influencing the clinical behavior of this tumor. On the basis of observations made to date we verified the hypothesis that the location of pilocytic astrocytomas is the major cause of their genomic differences, and tried to find genes connected with patient outcome and tumor appearance. The aim of this study was to identify gene expression profiles related to the most frequent locations, radiological features, and the clinical course of the disease in a representative group of Polish children with PAs.

Patient samples
Eighty-six children with pilocytic astrocytoma who were operated on at the Department of Neurosurgery, Polish Mother's Memorial Hospital, Research Institute in Lodz were included in this study. The group was comprised of 55 males and 31 females. The median age of patients at the time of diagnosis was 7 years (ranging from 14 months to 17 years). There were 40 cerebellar, 23 optic tract and hypothalamic, 21 cerebral hemispheric, and 2 brainstem tumors (Fig. 1). All specimens were diagnosed at the Department of Molecular Pathology and Neuropathology, Medical University of Lodz, according to the WHO criteria [1].
In all patients, preoperative MRI scans with and without contrast administration were obtained. For assessing the radiological features of tumors, we adopted the classification of radiological subtypes of PA proposed by Pencalet et al., commonly used in analyses of this tumor [15,16]. According to the radiological features presented on MRI, all tumors were divided into four subtypes: cystic with a non-enhancing cyst wall, cystic with an enhancing cyst wall, solid with central necrosis, and solid or mainly solid tumors (Fig. 2).
In 81 cases primary surgical resection was the only and curative treatment, while in five cases progression of the disease, requiring additional treatment, was noted. In two cases clinical manifestation of neurofibromatosis type 1 (NF1) was observed. The clinical data of patients included in this study are presented in Table 1. All samples were collected using the protocols approved by the Bioethics Medical University Committee (Approval No. RNN/154/06/KE).
Written informed parental consent was obtained from all patients under 16 (75 patients). In eleven older patients the participants gave their own consent according to the Polish law. All data were processed and stored in compliance with the Helsinki Declaration.

RNA isolation
Total RNA was extracted from the snap-frozen tumor tissues stored at-80°C after excision, using the acid phenol-guanidinum extraction method, purified using commercially available sets (RNeasy Mini Kit, Qiagen) and treated with DNAase (Qiagen) [17]. In order to obtain a high amount of RNA, macrodissection was used in all cases. Specimens were visually assessed by the pathologist to confirm that at least 50 % of tumor cells within the sample and areas with highest content of neoplastic tissue were used for direct RNA extraction. The quantity of RNA was measured using the NanoDrop 1000 (Thermo Scientific). RNA samples' quality was analysed using 2000 Bioanalyzer (Agilent Technologies), and after capillary electrophoresis the RNA integrity number (RIN) was generated by the software for each specimen.
cRNA synthesis and hybridization 250 ng RNA of each sample selected for array analysis (50 cases) was used for cDNA and subsequent cRNA synthesis (GeneChip® 3′ IVT Expression Kit, Affymetrix). The amplified and biotinylated complementary RNA (cRNA) was purified and fragmented using heat and Mg 2+ , and then underwent hybridization (45°C, 16 hours) with GeneChip Human Genome U133 Plus 2.0 Array (Affymetrix), followed by array staining (Streptavidin, Alexa Fluor 610-R-phycoerythrin conjugate, Molecular Probes). All procedures were performed according to the manufacturer's instructions (Affymetrix). Arrays were scanned using the GeneChip Scanner 3000 (Affymetrix).

Microarray analysis
Quality control of microarray data was carried out according to standard protocols, based on R/Bioconductor packages (ver. 2.3.5). Data were pre-processed using the GC-Robust Multi-array Average (GC-RMA) procedure, Normalized Unscaled Standard Error (NUSE) and Relative Log Expression (RLE) measures were calculated to verify the technical homogeneity of the dataset. On the basis of quality control, 47 out of 50 microarrays were then classified to the bioinformatic analyses.
Transcripts showing minimal variation of expression across the set of arrays were excluded from the analysis. Genes with expression differed by at least 1.5 times from the median in at least 10 % of the arrays, with variance significantly larger than the median variance (p ≤ 0.01) retained. For the selection of genes' differentiating subgroups, the Welch t-test with false discovery rate (FDR) estimation was used. A global test was applied to test whether the expression profiles differed between the classes by permuting the labels of which arrays corresponded to which classes. Biological relevance and contribution in cellular processes of obtained sets was analyzed by Gene Ontology classification on the basis of the Gene Ontology Consortium database (http:// www.geneontology.org). For selected genes the gene set enrichment analysis, with curated and motif gene set collections, was performed to analyze the signaling pathways (Molecular Signatures Database v. 3.0, http:// www.broadinstitute.org/gsea/msigdb/index.jsp). These analyses were performed using Kolmogorov-Smirnov, the Least Squares test, and the Gene Set Analysis method (p ≤ 0.001). Statistical analysis was carried out

Validation of the microarray data
Correlation analysis of RT-qPCR and microarray expression values were carried out for 39 independent samples equally diversified according to the three tumor locations: cerebral hemispheric tumors (M1), optic tract and hypothalamic tumors (M2), cerebellar tumors (M3). TaqMan® Gene Expression Assays by TaqMan® real time PCR with TaqMan® Universal PCR Master Mix (Applied Biosystems, UK) was used following the manufacturer's instructions on a Rotor Gene 6000 instrument (Qiagene-Corbett Life Science, Sydney, Australia) for selected genes (Additional file 1: Table S1). The PCR reactions for each assay were run in triplicate and the results were averaged. The normalized relative expression level of the genes of interest was calculated according to the method described by Pfaffl and Vandesompele et al., with GAPDH used as a reference gene [18,19]. Statistical comparison of three subgroups was made on the basis of the Kruskal-Wallis nonparametric test, with post hoc pairwise comparisons using the Dwass-Steel-Critchlow-Fligner test. Statistical significance was assumed for p ≤ 0.05.

Results
We performed bioinformatic analysis of the global gene expression of 47 childhood pilocytic astrocytoma with respect to the selected clinical features. After pre-processing of the data 21,910 probesets showed significant variance and were further analysed. For the purposes of bioinformatic analysis, all analyzed samples were divided, on the basis of pivotal clinical data, into eight subgroups: cerebral hemispheric tumors (M1); optic tract and hypothalamic tumors (M2); cystic cerebellar tumors with an nonenhanced cyst (M3R1); cystic cerebellar tumors with an enhanced cyst (M3R2); solid cerebellar tumors with central necrosis (M3R3); solid or mainly solid cerebellar tumors (M3R4); tumors linked to the neurofibromatosis type 1 (NF1); and progressive tumors (P2).
During the comparison of these eight subgroups using the parametric Welch t-test and post hoc class comparison test, we found 345 probesets with significantly changed expression (p < 0.001). The observed differences were also strongly significant in the global test (p < 0.007) (Additional file 2: Table S2).
The evaluation of biological processes represented within the selected genes was done on the basis of the gene ontology over-representation analysis. The most significantly represented ontology classes were connected with neuronal cells building proteins, adhesion molecules, cell junctions, and hormone and neuropeptides activity ( Table 2). Within genes with significantly changed expression, there were some that connected with transcriptional processes and acting during embryogenesis and central nervous system differentiation.
The analysis of selected genes' contribution in the signaling pathways revealed changed regulation of 77 within 2131 curated gene sets, and 14 within 179 motif gene sets. The highest statistical significance was obtained for genes functionally connected with immune response pathways, pathways engaged in silencing suppressors during histone methylation and activation of the NFkB pathway. Interesting group consisted of targets for miR324-5p, miR432, miR299-3P, miR486 and miR499 and genes located near promoter regions of NR6A1, POU3F2, CUTL1, PAX8 and AHR transcription factors (Table 3).
In the next step we analyzed the expression values of genes differentiating clinical subgroups of PA. Genes with highest amplitude were chosen for hierarchical clustering of samples (Fig. 3a). On the basis of such analysis we obtained three distinct clusters showing almost perfect classification of samples, which revealed that the main source of variability is related to the location of the tumors. The cerebellar tumors consist of a homogenic cluster, while the supratentorial samples showed single outlier specimens (Fig. 3b). Within tumors of optic tract and hypothalamus there was also a sample of brain stem PA, which presented a low correlation of gene expression with the supratentorial cases (r = 0,4). This sample was excluded from statistical analyses because of its low RNA quality, and as a consequence both cases of PA located within the brain stem were used only during data visualization.
Bioinformatic analysis of our dataset revealed that 32 probesets showed different expression pattern according to radiological subclasses (p < 0.005). Unfortunately these genes demonstrated weak transcriptome differences ( Fig. 4a), with borderline significance in the global test of association (p = 0,88). Hierarchical clustering and PCA analyses taking into account the radiological features of tumors did not show a specific gene expression signature correlated with the radiological features of analyzed PA (Fig. 4b, c).
Here we verified the hypothesis that the location of PAs is the major cause of their genomic differences. Analyses of three main anatomical subclasses (M1, M2, M3) using the parametric Welch t-test were very prominent and revealed statistically significant differences for 862 probesets based on the false discovery rate (FDR adjusted p-value of 0.001).
In the global test the differences were also strongly significant (p < 0.001). The comparisons of pairs (M1vsM3; M2vsM3; M1vsM2) using the post hoc test (BRB Array-Tools) revealed that the majority of genes showed different expression for the M2vsM3 and M1vsM3 (847 and 323 genes respectively), while 105 genes showed differences both for M1vsM3 and M2vsM3 tumors. The most   similarities (24 strongly differentiating genes) were noted for two analyzed supratentorial subgroups (Table 4, Fig. 3d, Additional file 3: Table S3 and Additional file 4: Table S4). These comparisons were also repeated with more restricted statistical criteria using the Benjamini-Hochberg multiple comparisons correction, with the criterion of FDR < 1 %. After comparison of the M3 and combined M1 and M2 subgroups, a list of 348 probesets was obtained. The probability of proper classification of tumors on the basis of gene expression profile reached a range of 80 % accuracy. In order to exclude the potential influence of other clinical variables on the obtained results, additional analysis was also performed for infratentorial cases, which included all four radiological types of PA. This approach confirmed our observations.
Our analyses of the transcriptome profile of five cases with progressive disease did not show any correlation with a worse outcome. Only five genes (SIX3, RGS8, FAM82, KIF9, WDR63) reached statistical significance (p = 0.001) when the univariate model was used, but the global test revealed that this association did not meet the criteria of statistical significance (p = 0.83) (Fig. 5a). Cases with neurofibromatosis type 1 had no connection with expression profile.
In the final stage of analysis we applied an unsupervised method (Principal Component Analysis, PCA) to The number of genes changed in each category was compared with the number of expected occurrences. Only GO classes and parent classes with at least five observations in the selected subset and with an 'observed vs. expected' ratio of at least two were shown determine the sources of variability in our group of samples according to the clinical data, and that analysis also indicated that the gene expression profile of pilocyticastrocytomas highly depends on the tumor's location (p = 0.001), but not on other clinical features (Figs. 3c, 4c and 5b). As the analysis of the most important determinants of gene expression (including location and radiological appearance) include the potential for multivariate associations, we carried out a two-way analysis of variance. In this multivariate approach two variables were taken into account (location: supratentorial vs infratentorial, i.e. M1 + M2 vs M3 + M4, and radiology: cystic vs solid, i.e. R1 + R2 vs R3 + R4). In the multivariate analysis only location seemed to be associated with gene expression (702 probesets with non-corrected p-value < 0.001, 1007 probesets with FDR < 10 %), while radiological appearance was almost without influence on gene expression (5 probesets with non-corrected p < 0.001, no probesets with FDR below 10 %).
During our analysis the most prominent differences connected with the location of the tumor were noted for the IRX2, PAX3, CXCL14, LHX2, SIX6, CNTN1 and SIX1 genes. For all these genes we performed validation results by independent RT-qPCR on 39 cases of PA, which confirmed the data obtained during microarray analysis with similar expression differences. For the PAX3, LHX2, CNCT1 and CXCL14 genes we obtained results which converged with the microarray results and these discriminants were the best to differentiate M1 and M2 from M3 tumors. For PAX3, LHX2, IRX2 and CNTN1 genes similar as in microarray analysis, the major expression differences for M1/M2 and M3 tumors was showen. The gene with a statistically significant level of expression for all three subgroups of tumors was SIX1 (Fig. 6).
The data discussed in this publication have been deposited in the NCBI's Gene Expression Omnibus and are accessible through the GEO Series accession number GSE73066 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cg i?acc=GSE73066) [20].

Discussion
In the current study we performed expression analysis of PAs diversified according to the main clinical discriminants, from which the tumor location was the most important. We also tried to verify if and how the transcriptional profile is connected with the radiological appearance of the tumor. Moreover we attempted to identify the expression profiles related to the clinical course of the disease. Unsupervised clustering analysis revealed differences between PAs located in three anatomical regions, which was also well confirmed by decomposition into principal components (Fig. 3b, c). The specific transcriptional profile for hemispheric tumors was mainly characterized by upregulation of FOXG1, NEDD4L, L1CAM and CXCL14 genes. The FOXG1 and NEDD4L have a functional connection with the TGF-beta signaling pathway which is involved in cells' proliferation and differentiation irregularities noted in other brain tumors, including gliomas, medulloblastomas and supratentorial ependymomas [21][22][23]. For the following L1CAM, acting during brain development and having an unknowable role in the adult nervous system, the correlation with tumor grade in several other solid human cancers was noted [22,24]. In turn, expression of CXCL14 chemokine with enigmatic physiological function is connected with the tendency of tumor infiltration described on the basis of in vitro studies [25,26]. Taken together, these genes are closely related not only to brain development but also to brain tumor growth and expansion. In this subgroup we also found the moderate overexpression of the LHX2 gene. Alterations of LHX2 expression were previously found in all supratentorial or hypothalamo-chiasmatic region  [8,9]. Here we confirmed significantly higher levels of gene expression for PAs located within the hypothalamo-chiasmatic region. The validation of array data reached statistical significance and allowed us to distinguish these two subgroups of supratentorial tumors on the basis of LHX2 activity. The LHX2 gene has a known function in brain development during embryogenesis among different organisms. Previous functional studies on model organisms showed the relationship between lhx2 and two six3 genes in forebrain development. Recent analyses of human embryonic stem cells demonstrated that it also has a crucial role in human brain morphogenesis [27,28]. There is evidence that LHX2 is expressed in the early neural lineage, which affects signaling pathways fundamental for early lineage differentiation in the human brain via regulation of the neural transcription factors from the SIX gene family [28,29]. Our results constitute confirmation of this observation. The unique transcriptional profile of our hypothalamo-chiasmatic region tumors was characterized by the overrepresentation of several SIX homeobox transcription factors, SIX1, SIX3 and SIX6.
The upregulation of other homeobox transcription factors was also observed by us in cases of infratentorial tumor, which was shown to be connected with Iroquois homeobox transcription factors (IRX1, IRX2, IRX3, IRX5) and the PAX3 gene. A similar trend was noted by Sharma et al. [9]. These genes play a crucial role in early brain regionalization, including patterning of the anterior-posterior and dorso-ventral axis, as well as the differentiation of its specific regions. On model organisms its homologs overexpression induces the occurrence of ectopic neural tissue and causes inhibition of neuronal differentiation [30,31].
Contactin family genes (CNTN1, CNTN3), which function as cell adhesion molecules with an essential role in later stages of cerebellar morphogenesis and differentiation, were also significantly overrepresentated in our cohort of cerebellar tumors [32,33]. Moreover, recent experimental analysis of developing mice cerebellum suggests the cooperation of CNTN proteins in modulating SHH-induced neuronal precursor proliferation and its connection with foliation of the cerebellum [32]. The deregulation of the SHH signaling pathway in cerebellar granule neuron precursors is one of the crucial causes of the SHH subgroup of medulloblastoma, but recent observations also associate this signaling pathway with other central nervous system tumors, including gliomas [34]. This is the first report where CNTN1 and CNTN3 gene expression were shown to be connected with cerebellar pilocytic astrocytomas biology. Also our following observation concerning misregulation of miR-324-5p, engaged in the tumorigenesis of various cancers including glial tumors, suggests SHH contribution in PA biology as probably the consequence of changed progenitor differentiation [35][36][37].
Another interesting finding concerns the RUNX1T1 transcript upregulation observed in silico study performed for pilocytic astrocytomas by Deshmukh et al. On the basis of the transcriptional regulatory network, Fig. 4 Results of association of radiological features with global gene expression profile. a probesets expression showing a relatively small amplitude of differences radiological subgroups. b unsupervised hierarchical clustering did not show specific gene expression signature correlated with radiological features (32 probesets, FDR adjusted p < 0.005). c radiological-related subgroups are not distinct in fully unsupervised multidimensional scaling analysis. R1, cystic tumor without an enhanced cyst wall; R2, cystic tumor with enhanced cyst; R3, solid tumor with central necrosis; R4, solid or mainly solid tumor they indicated that RUNX1T1, together with four other transcription factors, is within alteration common for pilocytic astrocytomas and glioblastomas [38]. This is not the only relationship between those histologically different glial tumors. Our group of cerebellar PAs was also characterized by elevated levels of PROM1 expression, which in glioblastoma is considered as a molecular factor of poor prognosis, and in proneural glioblastoma seems to be the potential target of anti-angiogenic therapy [5,39]. One could consider such observations as the molecular explanation of poor outcomes observed in some cases of PAs, but in light of findings showing gene expression also in normal adult brain astrocytes, such connection should be rather be regarded as additional indication of a gene with a nearly unknown function. Aggressive clinical behavior is more frequently observed in pilocytic astrocytoma located within the hypothalamochiasmatic region, and up to now is regarded as the result of limited surgical excision. However, the relation of an unfavorable disease outcome and overexpression of genes contributes to cell growth, proliferation, and differentiation, and hence it could not be excluded that it also promotes the cancerogenesis and tumor metastasis observed in PAs [40][41][42][43]. On the other hand, the frequently observed (in our study) alteration of genes involved in brain structures' development like homeobox transcription factors, contactins and miRNAs inducing neurogenesis, together with the generally benign behavior of PAs, lead to the statement that these tumors could be rather recognized by some scientists as a neurodevelopmental disorder [7,27,28,32,44]. Another argument for this is the coexistence of neurodevelopmental disorders and pilocytic  [45][46][47]. Such observations, which suggest the plausible association of autism and cerebellar PAs, also raise the question of their similar genetic background. Until the problem of the biological characteristics of pilocytic astrocytoma will be solved, and the degree of surgical resection remains the only clinically confirmed factor affecting recurrence or dissemination, there is still a need to identify the molecular markers influencing the clinical behavior of this tumor. Taking this into account, we tried to search how gene expression profiling is correlated with the clinical outcomes of children with PA. An additional incentive for us to perform such analyses was the limited studies on this subject [48][49][50]. Unfortunately in our study we didn't observe a correlation between a tumor's global gene expression profile and the clinical course of the disease. In the analysis of the univariate test of progressive and low-risk disease, only five genes differentiated between the subgroups with progressive disease and the others, but in the global test there were no real differences between the analyzed classes.
Hence our unsupervised analysis didn't disclose any differences between PA of different outcomes. The problem with revealing genes connected with the progressive disease during expression analyses was also noted in other studies, where such correlation were excluded or postulated only for an extremely limited number of genes. Sharma et al. and Rodriguez et al. assumed that the MATN2 and ALDH1L1 genes are connected with the progression of PA, but it seems to be more an accidental occurrence than crucial molecular changes. This observation was confirmed in the following studies [9,10]. In our opinion this could be an effect of the low biological aggressiveness of Pas, and finding molecular factors usable in clinical practice seems very unlikely. Moreover, malignant transformation of PAs is really a very rare phenomenon and possible histological misdiagnosis cannot be excluded.
We also tried to find any correlations between molecular profile and radiological features, a subject which was completely unknown for PAs. Such plausible connections were noted in other glial brain tumors such as glioblastoma and oligodendroglioma, but this was the first attempt to find them for PAs [8,9,51,52]. Major attention was directed on tumor borders and enhancement patterns after contrast administration. Here we adapted the division of predominant imaging patterns of pilocytic astrocytoma first proposed by Pencalet et al. and modified by others [15,16]. In our study radiological features presented by pilocytic astrocytoma were not associated with global gene expression profile. In the univariate test only four genes were differentiated between the subgroups, with no differences observed in the global test. Moreover, the genes which best discriminated between different radiological features of tumors had a relatively small amplitude of differences and were not good discriminators between the groups in clustering (Fig. 4). Previous studies indicated that radiological appearance could be a derivative of tumor location; solid tumors were frequently observed infratentorially, and optic tract tumors rarely consist of cystic elements [14][15][16]53]. Therefore we performed additional analyses for a cohesive subgroup of infratentorial lesions, which also didn't reveal any correlations. Similarly, analysis of two subgroups of tumors with mainly solid and mostly cystic features showed no associations with transcriptional pattern. Our results show that different radiological patterns of PA are not determined by trancriptional changes.

Conclusions
In our comprehensive study we showed the differences between pilocytic astrocytoma of different locations and pointed out the heterogeneity of a tumor coming from the supratentorial region (Table 5). We may state that gene expression profile in pilocytic astrocytomas is connected only with tumor location, which suggests a different origin of PA arising within various anatomical brain structures. This observation is complementary to current knowledge of pilocytic astrocytoma's biology and indicates that for targeted therapy three molecular subgroups of the tumor must be considered, particularly in the case of optimization of the treatment of optic nerve PAs. Our observation, combined