Sonic Hedgehog Medulloblastoma Cancer Stem Cells Mirnome and Transcriptome Highlight Novel Functional Networks

Molecular classification has improved the knowledge of medulloblastoma (MB), the most common malignant brain tumour in children, however current treatments cause severe side effects in patients. Cancer stem cells (CSCs) have been described in MB and represent a sub population characterised by self-renewal and the ability to generate tumour cells, thus representing the reservoir of the tumour. To investigate molecular pathways that characterise this sub population, we isolated CSCs from Sonic Hedgehog Medulloblastoma (SHH MB) arisen in Patched 1 (Ptch1) heterozygous mice, and performed miRNA- and mRNA-sequencing. Comparison of the miRNA-sequencing of SHH MB CSCs with that obtained from cerebellar Neural Stem Cells (NSCs), allowed us to obtain a SHH MB CSC miRNA differential signature. Pathway enrichment analysis in SHH MB CSCs mirnome and transcriptome was performed and revealed a series of enriched pathways. We focused on the putative targets of the SHH MB CSC miRNAs that were involved in the enriched pathways of interest, namely pathways in cancer, PI3k-Akt pathway and protein processing in endoplasmic reticulum pathway. In silico analysis was performed in SHH MB patients and identified several genes, whose expression was associated with worse overall survival of SHH MB patients. This study provides novel candidates whose functional role should be further investigated in SHH MB.


Introduction
Medulloblastoma (MB) is the most common malignant brain tumour in childhood. Multimodal aggressive treatments include surgical resection and chemo-and radio-therapy, and are able to

miRNA-Sequencing Determines miRNA Signatures in SHH MB CSCs
SHH MB CSCs were subjected to small RNA sequencing and miRNA expression levels were compared to our recently published NSC mirnome [26]. Hierarchical clustering of the differentially expressed miRNAs yielded two distinct clusters, clearly distinguishing SHH MB CSCs and NSCs ( Figure 1). In detail, SHH MB CSCs were characterised by 35 up-regulated and 133 down-regulated miRNAs (Tables 1 and 2 report the top 20 miRNAs, respectively; the entire datasets are in Tables S1 and S2). Results of the miRNA-sequencing were validated at the transcriptional level by quantitative PCR. Higher expression of miR-20a-5p and miR-193a-5p and lower expression of miR-222-5p, miR-34a-5p, miR-345-5p, miR-210-5p, and miR-200a-3p were confirmed in SHH MB CSCs compared to NSCs (marked in bold in Tables S1 and S2, Figure 2). The SHH MB CSC miRNA signature included several miRNAs that have been described in primary MBs [27][28][29][30]. Specifically, among miRNAs characterising SHH MB CSCs, we identified let-7a, miR-100, miR-132, miR-135a, miR-135b, miR-150, and miR-203, which we had previously described as deregulated miRNAs in primary human SHH MBs (originally named Gli high ) in respect to non-SHH MBs (Gli low ) [27]. Moreover, miR-135b and miR-203 were similarly down-regulated as in Gli high in respect to Gli low MBs, possibly representing the MB CSC compartment. Northcott et al. [30] performed a comparison between MB subgroups and identified a number of SHH MB-related miRNAs, several of which were deregulated in SHH MB CSCs, such as miR-20a, miR-17, miR-135a, miR-106a, miR-222, miR-135b, and miR-203. Of particular interest are the up-regulated miRNAs (miR-20a and miR-17) and the down-regulated miRNAs (miR-135b and miR-203) that were similarly expressed in both studies. Moreover, in the same study of Northcott et al. the expression of miRNAs in Shh-treated cerebellar granule neuron precursors (CGNPs) is reported and miR-19b, miR-714, and miR-709 showed a coherent deregulation with SHH MB CSCs. We previously reported miRNAs expressed in primary MBs compared to normal adult and fetal human cerebellum [28]. Interestingly, thirteen miRNAs deregulated in primary MB were also deregulated in SHH MB CSCs (miR-203, miR-361, miR-31, miR-17, miR-20a, miR-106a, miR-let-7a, miR-132, miR-150, miR-212, miR-330, miR-29a, and miR-135a) and of particular interest miR-203, miR-361, miR-31, miR-17, and miR-20a were expressed in a similar fashion in both primary MBs and SHH MB CSCs. Further studies have been performed focusing on miRNA expression in MB mouse models and as a result the collaboration of miR-17~92 cluster with the HH pathway in MB and its up-regulation in SHH MB [29,31]. Int expressed miRNAs yielded two distinct clusters, clearly distinguishing SHH MB CSCs and NSCs ( Figure 1). In detail, SHH MB CSCs were characterised by 35 up-regulated and 133 down-regulated miRNAs (Tables 1 and 2 report the top 20 miRNAs, respectively; the entire datasets are in Tables S1 and S2). Results of the miRNA-sequencing were validated at the transcriptional level by quantitative PCR. Higher expression of miR-20a-5p and miR-193a-5p and lower expression of miR-222-5p, miR-34a-5p, miR-345-5p, miR-210-5p, and miR-200a-3p were confirmed in SHH MB CSCs compared to NSCs (marked in bold in Tables S1 and S2, Figure 2). The SHH MB CSC miRNA signature included several miRNAs that have been described in primary MBs [27][28][29][30]. Specifically, among miRNAs characterising SHH MB CSCs, we identified let-7a, miR-100, miR-132, miR-135a, miR-135b, miR-150, and miR-203, which we had previously described as deregulated miRNAs in primary human SHH MBs (originally named Gli high ) in respect to non-SHH MBs (Gli low ) [27]. Moreover, miR-135b and miR-203 were similarly down-regulated as in Gli high in respect to Gli low MBs, possibly representing the MB CSC compartment. Northcott et al. [30] performed a comparison between MB subgroups and identified a number of SHH MB-related miRNAs, several of which were deregulated in SHH MB CSCs, such as miR-20a, miR-17, miR-135a, miR-106a, miR-222, miR-135b, and miR-203. Of particular interest are the up-regulated miRNAs (miR-20a and miR-17) and the down-regulated miRNAs (miR-135b and miR-203) that were similarly expressed in both studies. Moreover, in the same study of Northcott et al. the expression of miRNAs in Shh-treated cerebellar granule neuron precursors (CGNPs) is reported and miR-19b, miR-714, and miR-709 showed a coherent deregulation with SHH MB CSCs. We previously reported miRNAs expressed in primary MBs compared to normal adult and fetal human cerebellum [28]. Interestingly, thirteen miRNAs deregulated in primary MB were also deregulated in SHH MB CSCs (miR-203, miR-361, miR-31, miR-17, miR-20a, miR-106a, miR-let-7a, miR-132, miR-150, miR-212, miR-330, miR-29a, and miR-135a) and of particular interest miR-203, miR-361, miR-31, miR-17, and miR-20a were expressed in a similar fashion in both primary MBs and SHH MB CSCs. Further studies have been performed focusing on miRNA expression in MB mouse models and as a result the collaboration of miR-17~92 cluster with the HH pathway in MB and its up-regulation in SHH MB [29,31].

Enriched Pathway Analysis of SHH MB CSCs Transcriptome and Mirnome
SHH MB CSCs were next subjected to mRNA-sequencing and transcripts were ranked according to their expression (Table S3). mRNA-sequencing results were validated at the transcriptional level by quantitative PCR. Higher expression of Ccnd1, Hdac2, Atf4, Bcat1, Ccnd2, Myc and Hiflf1a was confirmed in SHH MB CSCs, whereas higher expression of Cdkn1a and Egfr was confirmed in NSCs ( Figure 3). In order to determine the enriched pathways of SHH MB CSCs, pathway enrichment analysis was performed. The statistically significant pathways with the highest gene count, and thus most affected, are reported in Figure 4A. Among them, the decrease on oxidative phosphorylation has already been described in MB and the combinatorial targeting of MB metabolism has been proposed for effective treatments [32]. Of equal interest is the protein processing in endoplasmic reticulum pathway since endoplasmic reticulum stress has been reported to promote angiogenesis, invasiveness, and MB tumour growth in a murine model [33]. Moreover, the enrichment of PI3K-Akt signalling pathway and pathways in cancer were identified. The PI3K-Akt/mTOR pathway has been reported to be activated in MB [34][35][36], involved in metastasis [37] and its targeting has been proposed to inhibit MB CSCs [38] and overcome resistance to SMO antagonists [39]. All enriched pathways along with their matching genes are listed in Table 3 and Table S4, findings are ranked according to observed gene count.

Enriched Pathway Analysis of SHH MB CSCs Transcriptome and Mirnome
SHH MB CSCs were next subjected to mRNA-sequencing and transcripts were ranked according to their expression (Table S3). mRNA-sequencing results were validated at the transcriptional level by quantitative PCR. Higher expression of Ccnd1, Hdac2, Atf4, Bcat1, Ccnd2, Myc and Hiflf1a was confirmed in SHH MB CSCs, whereas higher expression of Cdkn1a and Egfr was confirmed in NSCs ( Figure 3). In order to determine the enriched pathways of SHH MB CSCs, pathway enrichment analysis was performed. The statistically significant pathways with the highest gene count, and thus most affected, are reported in Figure 4A. Among them, the decrease on oxidative phosphorylation has already been described in MB and the combinatorial targeting of MB metabolism has been proposed for effective treatments [32]. Of equal interest is the protein processing in endoplasmic reticulum pathway since endoplasmic reticulum stress has been reported to promote angiogenesis, invasiveness, and MB tumour growth in a murine model [33]. Moreover, the enrichment of PI3K-Akt signalling pathway and pathways in cancer were identified. The PI3K-Akt/mTOR pathway has been reported to be activated in MB [34][35][36], involved in metastasis [37] and its targeting has been proposed to inhibit MB CSCs [38] and overcome resistance to SMO antagonists [39]. All enriched pathways along with their matching genes are listed in Table 3 and Table S4, findings are ranked according to observed gene count.  Accordingly, pathway enrichment analysis was performed for the SHH MB CSC miRNAs. In detail, up-regulated and down-regulated miRNAs were used as inputs and 25 pathways were reported as statistically significant for each input (Tables 4 and 5 report the top 20 pathways. The entire tables are reported in Tables S5 and S6). Since we were interested in the identification of the pathways characterising SHH MB CSCs, we focused on the intersection of the enriched pathways between the  Accordingly, pathway enrichment analysis was performed for the SHH MB CSC miRNAs. In detail, up-regulated and down-regulated miRNAs were used as inputs and 25 pathways were reported as statistically significant for each input (Tables 4 and 5 report the top 20 pathways. The entire tables are reported in Tables S5 and S6). Since we were interested in the identification of the pathways characterising SHH MB CSCs, we focused on the intersection of the enriched pathways between the SHH MB CSC miRNAs and their transcriptome. Pathways in cancer, PI3K-Akt pathway, and protein processing in endoplasmic reticulum were the three common pathways identified by the intersection between the enriched pathways of the up-regulated miRNAs and those of the SHH MB CSC transcriptome ( Figure 4B). Whereas, only two pathways were in common between the down-regulated miRNAs in SHH MB CSCs and their transcriptome, as shown in Figure 4B, pathways in cancer and PI3K-Akt pathway.   Tables 3-5 and Tables S4-S6, respectively).

Novel Functional Networks in SHH MB CSCs
The binding of miRNA to mRNA is a type of regulation that can control both the translation of the mRNA and the availability of miRNA for other targets [40], indeed miRNAs post-transcriptional regulation can amount to controlling 50% of the protein-coding genes, due to the ability of miRNAs to target multiple genes [41]. On this basis, we focused on the identification of networks that could be involved in the perpetuation of SHH MB CSCs using the putative targets of the up-regulated and down-regulated miRNAs that belong to the common enriched pathways. As reported in Figure 5, we firstly focused on the networks comprising the putative targets of the up-regulated miRNAs. Networks in Figure 5A,B show several genes in common, such as Mapk1, Mapk3, Raf1, Fgfr1, Ptk2,  Tables 3-5  and Tables S4-S6, respectively).

Novel Functional Networks in SHH MB CSCs
The binding of miRNA to mRNA is a type of regulation that can control both the translation of the mRNA and the availability of miRNA for other targets [40], indeed miRNAs post-transcriptional regulation can amount to controlling 50% of the protein-coding genes, due to the ability of miRNAs to target multiple genes [41]. On this basis, we focused on the identification of networks that could be involved in the perpetuation of SHH MB CSCs using the putative targets of the up-regulated and down-regulated miRNAs that belong to the common enriched pathways. As reported in Figure 5, we firstly focused on the networks comprising the putative targets of the up-regulated miRNAs. Networks in Figure 5A,B show several genes in common, such as Mapk1, Mapk3, Raf1, Fgfr1, Ptk2, and Itgb1. All of them play a pivotal role in cancer cell growth and maintenance. Specifically, Mapk1 and Mapk3 activation have been associated with metastatic MB and poor outcomes [42]. Also, resistance to SMO inhibition along with metastasis in SHH dependent tumours has been attributed to the activation of RAS/MAPK pathway [43]. Fgfr1 is a member of the fibroblast growth receptors family and activation of Fgf signalling has been proposed as a target of MB and other SHH dependent tumours [44]. Similarly, Ptk2, a focal adhesion kinase, has been reported to induce MB cells proliferation and mediate c-Met migration and invasion [45]. Recently, another study investigated the activation of Itgb1 by Map4k4 and its association with the infiltration of MB cells [46]. Interestingly, Hsp90b1 was involved in all three networks ( Figure 5A-C). Hsp90b1 is an essential molecular chaperone [47] and it has recently emerged that inhibitors of Hsp90b1, together with inhibitors of the other proteins of the Hsp90 family, are promising classes of anti-cancer drugs in both solid and hematologic malignancies [48]. and Itgb1. All of them play a pivotal role in cancer cell growth and maintenance. Specifically, Mapk1 and Mapk3 activation have been associated with metastatic MB and poor outcomes [42]. Also, resistance to SMO inhibition along with metastasis in SHH dependent tumours has been attributed to the activation of RAS/MAPK pathway [43]. Fgfr1 is a member of the fibroblast growth receptors family and activation of Fgf signalling has been proposed as a target of MB and other SHH dependent tumours [44]. Similarly, Ptk2, a focal adhesion kinase, has been reported to induce MB cells proliferation and mediate c-Met migration and invasion [45]. Recently, another study investigated the activation of Itgb1 by Map4k4 and its association with the infiltration of MB cells [46]. Interestingly, Hsp90b1 was involved in all three networks ( Figure 5A-C). Hsp90b1 is an essential molecular chaperone [47] and it has recently emerged that inhibitors of Hsp90b1, together with inhibitors of the other proteins of the Hsp90 family, are promising classes of anti-cancer drugs in both solid and hematologic malignancies [48]. On the other hand, from the putative targets of the down-regulated miRNAs in SHH MB CSCs, we observed the genes involved in both networks and found some crucial genes involved in MB progression, including Myc, Grb2, Egfr, and Rac1, as shown in Figure 6A,B. In detail, Myc has been identified as a marker of tumour aggressiveness and SHH activation in the SHH MB subgroup [4,49,50]. Next, we found two components of the Egfr pathway, whose function consists of the adaptor molecule Grb2 binding to Egfr with the subsequent activation of the pathway. ERK activation has been associated with MB progression and control of cell-cycle related proteins' translation, possibly through mTOR activation [51]. Moreover, SHH induced Rac1 activation and migration of fibroblasts [52], a mechanism that was also reported in highly invasive MB cells [53]. On the other hand, from the putative targets of the down-regulated miRNAs in SHH MB CSCs, we observed the genes involved in both networks and found some crucial genes involved in MB progression, including Myc, Grb2, Egfr, and Rac1, as shown in Figure 6A,B. In detail, Myc has been identified as a marker of tumour aggressiveness and SHH activation in the SHH MB subgroup [4,49,50]. Next, we found two components of the Egfr pathway, whose function consists of the adaptor molecule Grb2 binding to Egfr with the subsequent activation of the pathway. ERK activation has been associated with MB progression and control of cell-cycle related proteins' translation, possibly through mTOR activation [51]. Moreover, SHH induced Rac1 activation and migration of fibroblasts [52], a mechanism that was also reported in highly invasive MB cells [53]. MiRNAs have been associated not only with tumorigenesis but also as prognostic biomarkers with the first study performed by Calin et al. describing the association of loss of miR-15 and miR-16 expression with a more favourable prognosis in chronic lymphocytic leukemia patients [54]. Unfortunately, even though treatment of MB patients has improved, 30% of patients still die from the disease [55]. Extending our research, we questioned if any of the putative targets of the SHH MB CSC miRNAs could have a role in the overall survival of MB patients [2]. Following the description of Cavalli et al. [2] regarding the four subtypes of SHH MB (α, β, γ, δ) being enriched for different age groups, we decided to take into consideration the factor of age in overall survival. To this end, in silico analysis was performed and survival curves were estimated in association with each gene expression level in MB patients divided in age groups. Firstly, infant (under 3 years of age) and pediatric (between 4 and 17 years of age) SHH MB patients together, then infant SHH MB, pediatric SHH MB, and finally adult SHH MB patients. Gene expression was associated with overall survival (OS) for each category and detailed results are reported in Table 6. Interestingly, expression levels of several genes were associated with worse overall survival of SHH MB patients. In detail, worse overall survival in infant and pediatric SHH MB patients was reported with high expression of RHEB and UBQLN1 and low expression of CHUK, CTNNB1, and DDIT4. MBTPS1, RAF1, TXNDC5, and VCP high expression and SAR1B and UBE4B low expression were associated with worse overall survival in infant SHH MB patients. Additionally, high expression of PPP2R1A, RAD23B, RHEB, and UBQLN1 and low expression levels of CHUK and CTNNB1 were observed in pediatric SHH MB patients with worse overall survival. The absence of common genes among infant SHH MB patients and the other two categories is noteworthy. Infant MB patient treatment is extremely challenging [56] and this singularity of infant SHH MB patients could provide useful information since only RAF1 is part of the PI3K-Akt pathway, whereas MBTPS, TXNDC5, VCP, SAR1B, and UBE4B are part of the protein processing in endoplasmic reticulum pathway. On the other hand, in pediatric SHH MB patients, RHEB is involved in the PI3K-Akt pathway, CHUK and CTNNB1 in pathways in cancer and MiRNAs have been associated not only with tumorigenesis but also as prognostic biomarkers with the first study performed by Calin et al. describing the association of loss of miR-15 and miR-16 expression with a more favourable prognosis in chronic lymphocytic leukemia patients [54]. Unfortunately, even though treatment of MB patients has improved, 30% of patients still die from the disease [55]. Extending our research, we questioned if any of the putative targets of the SHH MB CSC miRNAs could have a role in the overall survival of MB patients [2]. Following the description of Cavalli et al. [2] regarding the four subtypes of SHH MB (α, β, γ, δ) being enriched for different age groups, we decided to take into consideration the factor of age in overall survival. To this end, in silico analysis was performed and survival curves were estimated in association with each gene expression level in MB patients divided in age groups. Firstly, infant (under 3 years of age) and pediatric (between 4 and 17 years of age) SHH MB patients together, then infant SHH MB, pediatric SHH MB, and finally adult SHH MB patients. Gene expression was associated with overall survival (OS) for each category and detailed results are reported in Table 6. Interestingly, expression levels of several genes were associated with worse overall survival of SHH MB patients. In detail, worse overall survival in infant and pediatric SHH MB patients was reported with high expression of RHEB and UBQLN1 and low expression of CHUK, CTNNB1, and DDIT4. MBTPS1, RAF1, TXNDC5, and VCP high expression and SAR1B and UBE4B low expression were associated with worse overall survival in infant SHH MB patients. Additionally, high expression of PPP2R1A, RAD23B, RHEB, and UBQLN1 and low expression levels of CHUK and CTNNB1 were observed in pediatric SHH MB patients with worse overall survival. The absence of common genes among infant SHH MB patients and the other two categories is noteworthy. Infant MB patient treatment is extremely challenging [56] and this singularity of infant SHH MB patients could provide useful information since only RAF1 is part of the PI3K-Akt pathway, whereas MBTPS, TXNDC5, VCP, SAR1B, and UBE4B are part of the protein processing in endoplasmic reticulum pathway. On the other hand, in pediatric SHH MB patients, RHEB is involved in the PI3K-Akt pathway, CHUK and CTNNB1 in pathways in cancer and PPP2R1A, RAD23B, and UBQLN1 are part of the protein processing in endoplasmic reticulum pathway. Interestingly, two genes belonging in pathways in cancer, EGFR and MAPK3, were identified only in adult MB patients. Low expression of EGFR and high expression of MAPK3 were associated with worse OS in adult MB patients. We further investigated the putative targets of SHH MB CSCs taking into consideration the four SHH subtypes, pediatric SHHα, infant SHHβ, infant SHHγ, and adult SHHδ, as described by Cavalli et al. [2]. The OS analysis was performed for all putative targets of SHH MB CSC miRNAs for each SHH subtype and the statistically significant results are reported in Table 7. In detail, worse OS was associated with CTNNB1, PDIA3, and RAF1 low expression and RAD23B, RHEB, UBQLN1, and YWHAH high expression in pediatric SHHα patients. Notably, CTNNB1, RAD23B, RHEB, and UBQLN1 were also reported in pediatric MB patients ( Table 6). Infant SHHβ patients were associated with worse OS when low expression of UBE4B was observed and this gene was also reported in infant MB patients ( Table 6). None of the putative targets of the SHH MB CSC miRNAs was associated with worse OS in infant SHHγ, a subtype characterised by a good prognosis. Finally, in adult SHHδ patients low expression of CHUK and DNAJB11 were associated with worse OS, these genes were not reported in the analysis regarding adult MBs (Table 6).
Whether the genes of interest reported in Table 6 could be characteristic only of SHH MB patients led to further in silico analysis of the gene expression levels between WNT, SHH, Group 3, and Group 4 MB patients and the results are reported in Table 8. Interestingly, a statistically significant higher expression of CTNNB1, DDIT4 was observed in infant and pediatric SHH MB patients, whose low expression was associated with worse OS. Moreover, in the same age group higher expression of RHEB was observed in Group 4 MB patients compared to other MB subgroups, whose high expression reported a worse OS. Low expression of MBTPS1 was observed in infant SHH and WNT MB patients compared to Group 3 and Group 4 MB patients and its high expression was associated with a worse OS. When comparing only pediatric MB patients the high expression of CTNNB1 was observed in SHH MB patients compared to the other subgroups and its low expression was linked to worse OS. Of note, lower expression of RAD23B and UBQLN1 was observed in pediatric SHH MB when compared to other subgroups and their high expression was associated to worse overall survival. Moreover, in WNT pediatric patients low expression of RHEB was observed when compared to other subgroups and its high expression was reported in patients with worse OS. Finally, in adult SHH MB patients high expression of EGFR was observed when compared to other MB subgroups and its low expression was related with worse OS.
Recently, researchers have also focused their research in understanding the GLI crosstalk with other pathways in order to develop possible combinational treatments [57]. Specifically, EGFR and GLI crosstalk has been reported in MB [58], whereas in other tumour contexts such as anaplastic thyroid cancer, triple breast cancer, and colon cancer GLI crosstalk has been reported with RAS-BRAF-ERK [59], WNT [60], and WNT/β-catenin [61] signalling pathways, respectively.
Our results report changes in the enriched pathways and changes in the number of genes that participate in each pathway, shedding light on another aspect of SHH MBs and identifying novel molecules and pathways that should be taken into consideration in future studies.

Materials and Methods
Unless otherwise specified, all commercial products were used in accordance with the manufacturer's protocol.

SHH MB CSCs and NSCs
Murine SHH MB CSCs were derived, as previously described, from spontaneous tumours arisen in Ptch1 +/− mice [20], maintained as in Reference [23] and RNA extraction was performed as in Reference [62]. NSC mirnome data were obtained from our recently published manuscript [26]. Animal experiments were performed according to the European Community Council Directive 2010/63/EU and were approved by the local Ethics Committee for Animal Experiments (Prot. N 03/2013, 12/March/2013) of the Sapienza University of Rome.

miRNA-Sequencing
Three biological replicates of SHH MB CSCs were subjected to miRNA-sequencing, quality control, mapping, quantification and differential expression analysis was performed between SHH MB CSCs and NSCs, as previously described [26]. In detail, an average of 10.48 ± 0.748 million reads per biological replicate were obtained by the sequencing and quality control was performed using the FastQC tool (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/, accessed on 01/Februay/2017 before mapping where an average of 92% of reads with a quality score >30 were obtained. All sequenced single-end reads (31 bases) were aligned to the to the Mus musculus small RNA library , (miRBase V20) using the Genomatix Mining Station (GMS, Sesame 2.4, https://www.genomatix.de/, accessed on 19/February/2017) resulting in an average of 79% mapping. Differential expression analysis was performed using the miRNA-seq reads of SHH MB CSCs vs NSCs in the Genomatix Genome Analyzer (GGA, v3.30126, https://www.genomatix.de/, accessed on 21/February/2017) using the DESeq2 method. Differences characterised by a minimum log2 fold change of 1 and an adjusted p-value of < 0.05 (Benjamini-Hochberg correction for multiple testing) were statistically significant.

mRNA-Sequencing
Three biological replicates of SHH MB CSCs were subjected to mRNA sequencing, quality control, mapping and transcript abundance were obtained using the TopHat/Cufflinks pipeline, as previously described [26]. In detail, an average of 80.91 ± 5.41 million reads per biological replicate were obtained by the sequencing and quality control was performed using the FastQC tool (http://www. bioinformatics.babraham.ac.uk/projects/fastqc/, accessed on 1 February 2017) before mapping where we obtained an average of 86% of reads with a quality score >30. All sequenced 70mer paired-end reads were aligned to the Mus musculus genome (GRCm38 mm10 ENSEMBL) using TopHat (version 2.0.11, John Hopkins University, Baltimore, MD, USA) and resulted in an average of 80% mapping. Transcript assembly was performed with Cufflinks (version 2.2.2, University of Washington, Seattle, WA, USA) and transcript abundance was estimated using fragments per kilobase of exon per million fragments mapped (FPKM) values resulting in an average of 14,410 transcripts per biological replicate.
mRNA quantification was expressed in arbitrary units and each amplification reaction was performed in triplicate. All data were evaluated using the 2 -∆∆CT method and values were normalised to three endogenous controls.

Pathway Enrichment Analysis
DAVID (https://david.ncifcrf.gov/, accessed on 27 February 2017) [63] was used to perform pathway enrichment analysis using the KEGG (Kyoto Encyclopedia of Genes and Genomes) database for all the transcripts reported from the mRNA-sequencing and pathways enriched for more than 45 genes that were deemed of interest. Up-regulated miRNAs in SHH MB CSCs were used as input to obtain the enriched pathways and the same analysis was performed for the down-regulated miRNAs in SHH MB CSCs. Pathways with FDR < 0.05 and p < 0.05 were statistically significant and are reported in detail in Tables 3-5 and Tables S4-S6, respectively.

miRNA Putative Targets
Identification of miRNA putative targets was performed with miRSystem (http://mirsystem. cgm.ntu.edu.tw/, accessed on 12 January 2018)) [64] and target predictions for up-regulated and down-regulated miRNAs in SHH MB CSCs were obtained. The predicted targets were intersected with the genes included in the enriched pathways for the up-regulated (Tables S7-S9) and down-regulated miRNAs (Tables S10 and S11).

Functional Networks
The construction of the functional networks was performed with String DB (https://stringdb.org/, accessed on 30 January 2018)) [65]. The genes of interest reported in Tables S7-S9 for the up-regulated miRNAs and in Tables S10 and S11 for the down-regulated miRNAs in SHH MB CSCs were used as input after the exclusion of the common genes between the predicted targets of the up-regulated and down-regulated miRNAs of the same pathway. Medium confidence interaction score was selected and networks were exported from the database.

Conclusions
This study allowed us to focus on the CSC compartment of SHH MB and the data obtained from mirnome and transcriptome sequencing brought in evidence of the implicated pathways in SHH MB CSCs. The activation of PI3K-Akt pathway has been reported in MBs and new studies are focusing on the combinational treatment of MB with SMO and PI3K-Akt inhibitors. On this basis, this study presents other candidate pathways and genes whose inhibition or activation could be the foundation of future functional studies, leading to a shift in the landscape of survival in SHH MB patients.