Whole Exome- and mRNA-Sequencing of an AT/RT Case Reveals Few Somatic Mutations and Several Deregulated Signalling Pathways in the Context of SMARCB1 Deficiency

Background. AT/RTs are rare aggressive brain tumours, mainly affecting young children. Most cases present with genetic inactivation of SMARCB1, a core member of the SWI/SNF chromatin-remodeling complex. We have performed whole exome- and mRNA-sequencing on an early onset AT/RT case for detection of genetic events potentially contributing to the disease. Results. A de novo germline variant in SMARCB1, c.601C>T p.Arg201∗, in combination with somatic deletion of the healthy allele is likely the major tumour causing event. Only seven somatic small scale mutations were discovered (hitting SEPT03, H2BFM, ZIC4, HIST2H2AB, ZIK1, KRTAP6-3, and IFNA8). All were found with subclonal allele frequencies (range 5.7–17%) and none were expressed. However, besides SMARCB1, candidate genes affected by predicted damaging germline variants that were expressed were detected (KDM5C, NUMA1, and PCM1). Analysis of differently expressed genes revealed many dysregulated pathways in the tumour, such as cell cycle, CXCR4 pathway, GPCR-signalling, and neuronal system. FGFR1, CXCR4, and MDK were upregulated and may represent possible drug targets. Conclusion. The loss of SMARCB1 function leads to AT/RT development and deregulated genes and pathways. Additional predisposing events may however contribute. Studies utilizing NGS technologies in larger cohorts will probably identify recurrent genetic and epigenetic alterations and molecular subgroups with implications for clinical practice and development of targeted therapies.


Introduction
Atypical teratoid/rhabdoid tumours (AT/RTs) are rare and highly malignant neoplasms of the central nervous system (CNS) usually occurring in early childhood. According to World Health Organization (WHO), AT/RT is classified as a grade IV neoplasm and constitutes 1 of 3 major embryonal tumour entities. The median age for presentation is 20 months with slightly higher frequency in males (M : F, 1.6 : 1) [1]. AT/RT represents approximately 10% of CNS tumours in infants and is most often fatal. Unlike most pediatric brain neoplasms, AT/RTs are in terms of genetic alterations intersimilar and nearly all cases present with deletion and/or mutation of the tumour suppressor gene SMARCB1 (INI1/hSNF5) located on chromosome 22 which results in loss of SMARCB1 protein expression [2][3][4][5]. This recurrent event seems to be the most common and often the sole genetic aberration reported in AT/RTs [3,6]. SMARCB1 protein is a core subunit of the ATP-dependent SWI/SNF chromatin remodelling complex that functions as a transcriptional regulator [7]; consequently the loss of its function introduces epigenetic alterations that affect gene expression and might contribute to oncogenesis [8]. An increasing number of mutations detected in epigenetic regulatory genes including additional components of the SWI/SNF complex are also being reported in several other cancer forms, highlighting the importance of alterations in chromatin structure in the development of many malignancies [9]. Due to the aggressiveness of AT/RT tumours, a correct diagnosis is important to ensure the indication of a proper intensive treatment. However, the morphological/histopathological examination of the specimen can as a result of cell heterogeneity be difficult; therefore molecular analysis can have an important role in assisting diagnosis. The determination of SMARCB1 inactivation with loss of protein expression, mainly assessed by immunohistochemical methods, is now the gold standard procedure applied to confirm diagnosis of suspected AT/RT cases and to differentiate those from morphologically similar tumours.
With the development of next generation sequencing techniques, it is now possible to screen the entire exome/genome for DNA mutations and to profile the complete transcriptome in human biopsies. Such progress has in recent years led to the detection of many novel altered genes and pathways in specific cancers and demonstrated that various molecular subgroups, with differences in prognosis and possibly treatment targets, exist within histologically similar tumours, such as medulloblastomas, glioblastomas, and breast cancer [10][11][12]. The application of these new methods for the analysis of pediatric brain tumours has then the potential to further increase our knowledge regarding the underlying genetic events involved in the development of these malignancies. Here we report a comprehensive investigation, applying whole exome-and mRNA-sequencing of an early-onset AT/RT tumour diagnosed in a boy during the first year of life.

Patient Material.
A three-month-old boy, previously healthy, with no siblings, presented at the local hospital with rapidly growing head circumference, irritability, and inability to raise his head. Clinical examination revealed irritability, bulging fontanel, and head circumference of 44 cm (37 cm at birth) with no neurological deficits. MRI of the head showed a 9 × 8 × 8 cm supratentorial tumour on the left side (Figure 1(a)). The patient was referred to the pediatric intensive care unit and operated on the day after, achieving tumour subtotal resection. Two more operations followed within three weeks for resection of bleeding and residual tumour. MRI of the spine was normal and analysis of the cerebrospinal fluid showed no tumour cells. X-ray of the lungs and ultrasound of the abdomen were both normal.
Fresh tumour tissue and blood samples were collected at the initial surgery at Karolinska University Hospital, Stockholm, Sweden, frozen, and kept at −70 ∘ C in a local biobank.
The histopathological examination of the primary specimen revealed a highly cellular neoplasm mainly consisting of rhabdoid cells, relatively rich in eosinophilic cytoplasm and containing globular eosinophilic inclusions. The nuclei were vesicular and peripherally located with prominent nucleoli. Undifferentiated neuroectodermal tumour cells were also observed in some parts. The tumour was highly mitotic, 17/10 HPF; some apoptotic cells and areas of necrosis were also seen. There were foci of inflammatory cells, mainly consisting of lymphocytes. IHC stainings for Vimentin and MAP-2 were positive for most cells. There was patchy positivity for NFP, NSE, CD56, Synaptophysin, GFAP, MNF116, EMA, and EGFR E30. Staining for INI1 was negative in the tumour cells but positive in endothelial and inflammatory cells (Figure 1(b)). Most cells were negative for S-100 and PGR, CD34, Desmin, IDH1, and NeuN. IHC for p21 and p53 showed similar number of stained nuclei, indicating no probable p53 mutation. Ki-67 staining was variable, but ∼ 50% of tumour cells were in proliferative phase. Intraventricular AT/RT, WHO grade IV, was the concluding diagnosis (Table 1).
Intensive chemotherapy treatment (systemic and intraventricular) was started 4 weeks after initial operation according to the EU-RHAB protocol (Prof. Michael Fruhwald, Augsburg). A subdural shunt had to be inserted three months later. Because of local relapse, a reoperation was performed eight months after the first operations and second line chemotherapy was started. The boy stayed on low intensive chemotherapy inclusive intraventricular therapy for one year. During this time, no visible tumour was observed by MRI, but potential malignant cells were detected occasionally BioMed Research International 3

DNA Sequence Data
Analysis. Quality control of the sequencing reads was conducted using FastQC (http:// www.bioinformatics.babraham.ac.uk/projects/fastqc/). The fastq sequence reads were aligned to the human reference genome build GRCh37.p5 using CLC Genomics workbench (CLC, Aarhus, Denmark). The alignment settings allowed for successful mapping if at least 70% of reads length had >95% sequence similarity with the reference genome when the mismatch, insertion, and deletion costs were set to 2, 3, and 3, respectively. Reads matching to multiple locations were discarded. Removal of PCR duplicates was performed with Picard (http://picard.sf.net). Further processing of reads including trimming for Illumina adapters, low quality, and short length (below 30 bases) and variant calling were conducted within CLC Genomics workbench. The following criteria were applied for variant calling: (i) a maximum of 2 gaps/mismatches within a 21 bp window, with a minimum base quality of 30, minimum read count of 4, and minimum allele frequency of 5% for tumour and (ii) a maximum of 3 gaps/mismatches in a 21 bp window, with a minimum base quality of 25, minimum read count of 1, and minimum allele frequency of 2% for blood. The variants were annotated according to their overlap with genes and transcripts (UCSC, refSeq GRCh37/hg19 at http://genome.ucsc.edu/) and Sanger cancer census gene (http://cancer.sanger.ac.uk/cancergenome/projects/census/) conservation scores (UCSC), segmental duplications (UCSC), exon number, splice sites, amino acid change, cosmic database v63, The Human Gene Mutation Database (public version, http://www.hgmd.org/), ClinVar (database of mutations and their clinical relevance http://www.ncbi .nlm.nih.gov/clinvar/) and dbSNP v137. Additionally, predictions from SIFT (http://sift.jcvi.org) and Polyphen (http:// genetics.bwh.harvard.edu/pph2/bgi.shtml) were added to the list of variants.
Somatic variants were called if they were not present in the blood sample and a minimum read coverage of 8 was achieved (in the blood). Among these, somatic variants presenting with a probably functional effect were selected on the basis of: being previously reported as clinical SNPs (as reported in ClinVar and/or dbSNP v137), present in Cosmic database, determined to be probably damaging/damaging according to Polyphen/SIFT, overlapping a splice sites, resulting in a stop codon or were indels giving rise to amino acid change. The variations were further filtered out if detected in any of 97 pieces of control exomes data obtained from noncancer patients (including samples from the 1000 genome project) or were listed in dbSNP v137, with a population frequency higher than 1%. Moreover, mutations located within segmental duplications or in the TTN, MUC, and OR genes were removed based on the known risk of introducing false positives due to high sequence similarities and associated alignment problem. The resulting list of somatic mutations was visually inspected across the read alignments. Germline variants were processed in a similar way with the exception that the variants were required to be present in both tumour and blood samples with a minimum variant frequency of 20% and a minimum read count of 4 to be considered.

RNA Extraction and Next Generation Sequencing.
Total RNA was extracted from the tumour tissue according to Trizol protocol followed by Qiagen RNeasy Mini cleanup kit and DNase treatment. RIN values were checked on a Bioanalyser (Agilent) and polyA-tailed mRNA isolation according to Illumina library preparation instructions was conducted. The enriched mRNA was sequenced in a 1/8 lane of an Illumina Hiseq as 2 × 100 bp paired reads at the Karolinska Institutet Science for Life Laboratory.
We used two public control data sets, SRS151250 and SRS173568, that represent mRNA-seq data from equally pooled amounts of total RNA from the frontal part of the superior frontal gyrus of postmortem brains at different age points. The age in SRS151250 was from 1 to 35 days with 3 males and 2 females and the age in SRS173568 was from 182 to 274 days with 3 males and 2 females. Raw data generated and used here were 76 bp paired reads sequenced on Illumina GAII (http://trace.ddbj.nig.ac.jp/index e.html).

Gene Ontology and Pathway
Analysis. Gene Ontology (GO) and gene set overlap calculations were performed on the significantly upregulated and downregulated genes separately for determination of overrepresentation of specific biological functions, including cell signaling pathways. Gene sets used include curated collections from pubmed and online databases: canonical pathways, KEGG, BIOCARTA, and REACTOME (http:// www.broadinstitute.org/gsea/msigdb/annotate.jsp) [17]. GO analysis for overrepresented biological processes was performed using DAVID (http://david.abcc.ncifcrf.gov/ home.jsp) [18].

Results
The AT/RT patient investigated in this study presented with symptoms already at the age of three months (see case description in Section 2). Imaging studies revealed a large supratentorial tumour (Figure 1(a)) and the histopathological examination confirmed an AT/RT diagnosis including negative INI1 staining (Figure 1(b)). Despite surgical treatment and intense chemotherapy, local relapse occurred after eight months. Additional chemotherapy treatment was then started. Potential malignant cells were however found in cerebrospinal fluid and the boy has been reoperated for a second relapse 20 months after initial diagnosis (Table 1).

Few Somatic Mutations with Low Allele Frequencies Not
Detected in RNA. To discover tumour specific mutations, we performed whole exome-sequencing on DNA derived from the patient's tumour and blood, which generated 4.9 × 10 7 and 8.6 × 10 7 paired reads for each sample. The average read coverage for the targeted region (62.1 Mb) after trimming of adapters, alignment, and duplicate read removal was 19x and 49x for tumour and blood, respectively (Supplementary Table 1A in Supplementary Material available online at http://dx.doi.org/10.1155/2015/862039). To identify somatic mutations, we compared the variants present in blood and tumour samples and selected those found only in the tumour, with sufficient read coverage in blood; see Section 2. After filtering out common and nondamaging variants (see Section 2), seven somatic mutations remained. This indicated a mutation rate of as low as 0.145 mutations per Mb (Tables 1 and 2). We further inspected the allele frequencies of these mutations and could observe that all these variants were found at a subclonal frequency levels (range 5.7-17%), indicating that they were present only in a fraction of tumour cells and consequently they probably represent passenger or later events that occur after tumour initiation.
We also performed in depth transcriptome analysis. For that, isolated polyA enriched mRNA from the tumour sample was paired end sequenced, producing 2.1 × 10 7 aligned reads after duplicates removal (Supplementary Table 1B). The mRNA seq data allowed us to investigate the variant specific expression of transcribed genes. Interestingly, none of the seven somatic mutations determined by exome-sequencing (though present at low DNA allele frequency) were found in tumour RNA (Table 2).

Germline Candidate Variants Including the Mutation and
Somatic Loss of SMARCB1. Furthermore, we investigated the presence of possible germline disease causing mutations. After applying appropriate filtering steps (see Section 2), 205 uncommon and predicted damaging germline variations affecting 200 genes were found (Table 1 and Supplementary  Table 2). Among these, the most likely tumour causing event in this patient seems to be a germline nonsense mutation in SMARCB1 gene, c.601C>T p.Arg201 * . This C>T substitution which results in a truncated SMARCB1 protein is not reported in the general human population (dbSNP v137) but is the second most common SMARCB1 mutation reported in the catalogue of somatic mutations in cancer, Cosmic database (June 2013) ( Figure 2). Moreover, the allele frequency observed for the germline mutation (c.601C>T p.Arg201 * ) raised from 57.8 in blood to 87.5 in the tumour, indicating a somatic loss of the healthy allele (however not reaching 100% frequency probably due to the existence of some normal cells in the tumour sample). In addition, we confirmed the absence of SMARCB1 protein expression by IHC staining (Figure 1(b)). Consequently, the presence of a germline mutation and the acquired somatic loss of the healthy allele lead to 2 hits inactivation and complete deficiency of SMARCB1 function.
Among the remaining genes affected by uncommon germline variations predicted to be deleterious for  We used the mRNA seq data to investigate the expression of these potentially important variants in the tumor. Interestingly, several of these germline variants were found to be expressed in the tumour, including the nonsense mutation in SMARCB1 ( Table 3). The allele frequency of the SMARCB1 mutation in mRNA seq data was 75%, also supporting the loss of the wild type allele and the consequently biallelic inactivation of SMARCB1 in the tumour (but as in exome data not reaching a frequency of 100% due to some nontumor cells contamination, Supplementary  some reads do not align to this region and instead continue to exon 3 (Supplementary Figure 2). Concerning KDM5C transcription, the detected germline variant was expressed with an allele frequency of 100%, which is the expected value as the patient investigated here is a boy and the gene is lying on chromosome X. The germline variants observed in the cancer related genes NUMA1 and PCM1 were also transcribed and detected in the RNA data (Table 3). These genes have functions related to mitotic spindle establishment and centrosome assembly, respectively. In total 81 of the 205 uncommon and predicted damaging germline variants were found in the RNA data; for additional 29, only the common allele was detected and for the remaining neither allele was expressed (Figure 3 and Supplementary Table 2).  Table 2). Among the nine selected cancer associated genes with germline variants (Table 3) SMARCB1 was downregulated (Supplementary Figure 1) and so was CNTN6, whereas MYCL was upregulated (Supplementary Table 3). Statistical gene set overlap evaluation using the comprehensive Molecular Signature data base (MSigDB) on all upregulated genes in the tumour revealed that immune system, cell cycle, CXCR4 pathway, pathways in cancer, GPCR signaling, regulation of actin cytoskeleton, and p53 downstream pathway were among the top 20 significantly activated gene sets (Supplementary Table 4). A heat map with all upregulated genes involved in pathways of cancer is shown in Figure 4. Many of these genes, for example, PIK3R3, PIK3R5, ITGB1, ITGA2, ITGA6, STAT1, STAT5A, and STAT3, are also involved in the above mentioned CXCR4 pathway and some of them also contribute to the regulation of actin cytoskeleton and GPCR signaling, for example, CXCR4, RHOC, and RGS1 (Figure 4).

Several Upregulated and Downregulated Genes and Pathways with Implications in
FGFR1 was another upregulated gene acting in pathways of cancer and also playing a role in the actin cytoskeleton signaling. Interestingly, ligands for this receptor, FGF2, FGF17, FGF8, and FGF19, were also overexpressed in the tumour. Moreover, genes in the WNT, SHH, and BMP pathways such as WNT5A, FZD7, and FZD5, SHH, SMO, and GLI2 as well as BMP4 were likewise among the upregulated genes and were included in pathway of cancer gene set (Figure 4). Overexpression of BMP4 and several members of the BMP pathway have previously been correlated to bad prognosis in AT/RT [19]. The search among all the DE genes, for the 87 known BMP members, according to Birks et al. [19], revealed that 25 additional BMP genes were significantly dysregulated in the tumour; of those, 14 were among the upregulated ones and 11 were among the downregulated ones (Supplementary Table  3). Gene ontology analysis carried out with all the upregulated genes showed an enrichment of partly similar biological Immune system

Cell cycle
Interferon signaling Interferon alpha beta signaling Interferon gamma signaling CXCR4 pathway Adaptive immune system Cell cycle mitotic Hemostasis Cytokine signaling in immune system processes: mitosis, chromosome segregation, immune system development, embryonic development, and cell proliferation among others (Supplementary Table 6).
Among the top overexpressed genes, MDK, S100A4, and HMGA2 are worth mentioning. MDK, which ranked 25th amongst all 1661 upregulated genes (log2 FC: 4.3 and FPKM: 706.7), acts within the activated integrin pathway. MDK is known to promote proliferation and migration and to inhibit apoptosis. Moreover, it has been correlated to bad prognosis in glioblastoma [20]. S100A4 and HMGA2 are also interesting as both have been implicated in cancer. They ranked 3rd (log2 FC: 8.5 and FPKM: 874.3) and 6th (log2 FC: 8.1) among most upregulated genes, respectively.
MSigDB gene set overlap calculations performed with the significantly downregulated genes revealed that signaling by GPCR, developmental biology, axon guidance, neuronal system, and ion channel transport were among the top 20 overrepresented pathways (Supplementary Table 5). Very similar biological processes were observed as a result of gene ontology analysis (Supplementary Table 7). SMARCB1 was as mentioned above downregulated in the tumour and so were DOCK4, SPOCK1, PTN, and ATP1B1 which are known to have expression levels that correlate with SMARCB1 [21][22][23]. These genes play roles in neuronal system development and have previously been shown to be downregulated in rhabdoid tumours [22,23]. Neuronal developmental markers SOX11 and SNAI2 had also significantly lower expression compared to controls as well as the GNAI1 gene. Another potential interesting gene is MEIS2 which was the second most underexpressed gene. MEIS2 is involved in transcriptional regulation and it has also been shown to have decreased expression in poor prognosis prostate cancer [24].

Discussion
AT/RT is primarily a disease of children under three years of age [1] and represents approximately 1.3% of paediatric brain tumours and 20% of all embryonal tumours [25,26]. Treatment approaches include resection, high dose chemotherapy, and radiation, the latter being used if the patient is over 18 months of age. The patient studied here was operated on for both primary tumour and relapses and suffered from the disease and the intense treatment received.
Based on our findings, the disease causing mutation is a de novo germline variant in SMARCB1, c.601C>T p.Arg201 * occurring in the patient, that when combined with the somatic loss of the healthy allele leads to the complete loss of SMARCB1 function in the tumour cells. The germline mutation resides in the conserved SNF5 domain of the protein, which is a relatively common hotspot for alterations. There are more than 20 cases reported in Cosmic database with the same nonsense mutation and the majority of them are AT/RTs. The mutation in the patient was also confirmed by Sanger sequencing (clinical analysis, data not shown). Moreover, sequence analysis of parents' blood DNA sequencing (clinical analysis, data not shown) could not identify the presence of this SMARCB1 mutation. This supports a de novo origin of the constitutional mutation in the child and excludes the parents as carriers, with the exception of the very unusual presence of gonadal mosaicism. Germline alterations in SMARCB1 give rise to the Rhabdoid Tumour Predisposition Syndrome (RTPS) which is manifested by development of malignant rhabdoid tumours in infancy and early childhood [27,28]. This also explains the early disease onset, with the patient being only three months old. Due to the relatively high frequency (up to 35%) of germline mutations in SMARCB1 in AT/RT and the incomplete penetrance of inherited mutations in some cases [29], the screening of constitutional mutations in SMARCB1 in patients diagnosed with AT/RT is now a routine procedure at Karolinska University Hospital. Results from such testing can bring out important information used in genetic counselling for families with an affected child and importantly discovery of a SMARCB1 mutation also confirms diagnosis.
In the present case, loss of the healthy SMARCB1 allele in the tumour led to biallelic inactivation of the gene that is reflected by the increased allele frequency of the germline mutation in the tumour compared to blood and by the negative IHC staining for SMARCB1 protein. Another striking observation was the limited number of somatic variants detected that were predicted to be damaging. Only seven, and all of them found at subclonal frequency level were the highest allele frequency was 17%. Moreover, none of them were expressed. Furthermore, none of these genes are known as cancer drivers neither have the identified mutations been reported in Cosmic or in other AT/RT tumours [3]. Therefore, these somatic changes are not likely the ones driving tumour development. Similar trends have been previously reported by Lee et al. [3] who investigated 32 AT/RT cases for somatic changes and found only 172 mutations; many of them were detected at low allele frequency resulting in an overall very low mutation incidence compared to that in the present case (0.19 versus 0.14 mutations per Mb, resp.). The number of somatic mutations in AT/RT then appears to be even lower than that found in other malignant paediatric brain tumours such as medulloblastomas and much lower than that in adult malignant brain tumours as anaplastic astrocytomas and glioblastomas [10,30,31]. It might be that in the case of AT/RT the biallelic inactivation of SMARCB1, despite being the main recurrent somatic aberration, is also a sufficient event to promote tumour development. This is further supported by a study where 115 selected cancer associated genes were sequenced in 25 AT/RT samples. All cases presented with SMARCB1 alterations, apart from one, where a mutation in NRAS was found instead [32]. DNA copy number profiling has also shown that often there is only a single gross genomic aberration present in almost all tumours, which is loss of chromosome 22, where SMARCB1 is located [6].
We also searched for additional germline mutated genes other than SMARCB1, which could contribute to tumour predisposition. Experiments carried out in conditional SMARCB1 knock outs in mouse fibroblast cells have shown that the silencing of SMARCB1 results in growth arrest and p53-mediated cell death transformation [33]. This might suggest that additional genomic events or the specific cellular environments in determined cell types are required for SMARCB1 associated malignant transformation. To note, there are also many studies showing that reintroduction of wild type SMARCB1 in human AT/RT cells leads to cell cycle arrest and its tumour suppressor function is firmly established [34][35][36][37]. It is though interesting that we found a few possibly damaging variants in cancer associated genes in the patient's constitutional DNA. KDM5C, for example, is an interesting gene with epigenetic implications. It is situated on chromosome X and subsequently only the uncommon allele is expressed in this male patient. KDM5C encodes a histone demethylase specific for histone 3 lysine 4 that functions as a transcriptional repressor. This gene has been reported to be mutated in cancer including medulloblastoma (Cosmic). NUMA1 and PCM1 are additional genes for which the predicted damaging germline variants were expressed in the tumour. Both NUMA1 and PCM1 codes for proteins with important roles in cell division, so alterations in these proteins may have an impact on the normal cell cycle. It is important to document the presence of these variants; however, the possible synergistic effect of any additional genetic event(s) beside SMARCB1 aberrations responsible for AT/RT development is difficult to evaluate from a mutational screen of a single case.
Transcriptome analysis revealed several deregulated genes and pathways in the AT/RT compared to control samples. In accordance with a highly proliferative tumour, cell cycle was among the top significant enriched processes for upregulated genes. Additionally, p53 pathway and pathways in cancer, the latter including several SHH pathway members (SHH, SMO, and GLI2), seemed to be activated. Interestingly, and in agreement with our results, cell cycle and p53 and SHH pathways were shown to be enriched among upregulated genes in an expression microarray study performed on 20 AT/RT and 10 kidney rhabdoid tumours (KRT) [23]. Moreover, aberrant activation of SHH pathway has also been observed in SMARCB1-deficient cells [38]. WNT pathway members (such as WNT5A, FZD7, and FZD5) were also upregulated in the AT/RT case studied here, which is of interest as WNT pathway has been associated with a cluster of AT/RTs with significantly shorter survival [19]. Both WNT and SHH are important developmental pathways known to have disturbed activation in other paediatric malignancies [10]. Furthermore, the upregulation of BMP4 in the tumour, which is a signaling cytokine growth factor, acting within the here enriched pathways of cancer, is also a noteworthy finding since BMP4 has been reported to be the top DE gene between clusters of AT/RT tumours with different survival outcome, showing a clear overexpression in tumours with short survival [19]. The BMP pathway is a crucial signaling cascade acting during embryo growth, particularly important for CNS development, but it is also essential in maintaining homeostasis in adult tissues [39,40]. BMP4 overexpression has also been shown to promote invasion in different cancer cells [41][42][43]. Therefore, the upregulation of BMP4 in this case with several tumour relapses might be supportive of its role as a marker for bad prognosis.
We also found a strong upregulation of the cancer associated gene FGFR1 and several FGF ligands, all of them acting within the pathways of cancer and regulation of actin cytoskeleton. Of note, FGFR1 was overexpressed in AT/RT in the recent expression array study mentioned above [23]. The gene is also known to be amplified in squamous cell lung and breast cancer (so is FGF2) and to contain activating mutations in low grade astrocytoma [44][45][46][47]. Signaling via FGFRs activates signaling transduction pathways including Ras-MAPK, PI3-AKT, and PLC -PKC, mediating cell survival, proliferation, differentiation, migration, and drug resistance. Importantly, these pathways could be modulated in AT/RT as several FGFRs-targeted therapies have been developed which have been shown to reduce proliferation of cancer cells [48]. Another upregulated pathway which is worth mentioning is CXCR4. CXCR4 mediates cascades such as PI3K/AKT, JAK/STAT, and Rac/Rho, which are involved in cancer and promote tumour progression. Various cancers including glioma, neuroblastoma, and leukemia show overexpression of CXCR4 and since their blockade may have implications for therapy several inhibitors have been developed and are, at least for hematopoietic cancers, being tested in clinical trials [49,50]. Not much attention has been given to this pathway in AT/RT, even though the downstream signaling effectors such as AKT and Rho have been seen to be upregulated [21,51]. CXCR4 may be therefore a promising candidate for drug targeting in AT/RT cases which show such aberrant signaling activation.
Specific additional upregulated genes in the AT/RT case that might be of importance and also possible candidates for biomarkers and/or drug targeting are MDK, S100A4, and HMGA2. MDK is a secreted growth factor known to promote proliferation, migration, antiapoptosis, transformation, and angiogenesis. Interestingly, previous reports have also shown high expression of MDK in advanced tumours which correlated to serum levels and tumour progression [20,52]. MDK could also be considered as target for therapy as its inhibition has shown to potentiate the effect of cytotoxic drugs [52]. S100A4 plays important roles in several cellular processes, including the regulation of cell cycle and differentiation. S100A4 is upregulated in many types of cancer in which it has been clearly linked to tumour progression and bad prognosis [53][54][55][56][57][58]. HMGA2 functions as a transcriptional regulator and as a component of the enhanceosome [59]. HMGA2 is expressed during early development but also in late stage or invasive cancer forms [60,61]. Both S100A4 and HMGA2 were found to be overexpressed in AT/RT [23].
Regarding downregulated transcripts in AT/RT, many genes involved in developmental biology, axon guidance, and neuronal system were found. Decreased expression of genes associated with neuronal development was seen in RT with SMARCB1 loss in comparison with other kidney tumours [22] and the same above mentioned processes were found to be downregulated in short survivor AT/RT patients [19], which indicates aberrant neuronal functions. DOCK4 and PTN, for example, were significantly downregulated genes in this AT/RT case. Both genes were recently shown to be underexpressed in AT/RT and/or RT [22,23] and to present with concordant differential downregulation with SMARCB1 [21]. GNAI1 is moreover included in the downregulated gene sets of GPCR, axon guidance, and neuronal development in our analysis. It regulates cell proliferation and differentiation [62,63] as well as cancer migration and invasion [64]. GNAI1 was also previously described as underexpressed in AT/RTs [23].
However, several early neural cell genes that were downregulated in KRT [22] were not seen to be significantly downregulated in our case. Similar results were observed in a recent gene expression array study of AT/RTs [23]. Specific genes seen to be dysregulated previously, such as the histone methyl transferase EZH2 that is a part of PRC2, were not upregulated in our study either. So, despite often only carrying a seemingly simple genetic alteration, leading to the inactivation of SMARCB1, AT/RT tumours are heterogeneous when it comes to the transcripts/pathways affected and this may be reflected in patient outcome. Predisposing genetic events, environmental factors, and/or different cell of origin may represent some features that influence this heterogeneity.

Conclusions
In summary, regardless of these gene expression differences observed between studies and samples, striking common abnormalities exist within AT/RTs, and several pathways and genes appear recurrently dysregulated. Among these cell cycles, p53 pathway and SHH as well as the specific genes mentioned above, FGFR1, S100A4, and HMGA2, were repeatedly upregulated, whereas neuron differentiation and neuronal development pathways as well as genes with neuronal functions, DOCK4, PTN, and GNAI1, were reiteratively downregulated. Consequently they should be taken into account as potential biomarkers and/or targets for therapy. The recurrence of these findings across studies is, beside its biological importance, a validation of our analysis. AT/RT is a relative unusual tumour and there is currently a lack of comprehensive genomic and epigenomic studies including larger set of patients. However, with the development and more extended application of next generation sequencing techniques as well as improvements in the collection of appropriate material in bio banks, more reports will likely be seen. Such efforts should reveal if there are more frequent genetic aberrations to be found that could define molecular subgroups with clinical implications. The increased knowledge will aid in the development of targeted therapies for improved quality of life and hopefully overall survival for these unfortunate children.