The precise cell of origin for MB remains to be determined, however data from both human and mice studies suggest that some MB sub-types arise from NSCs 
. Several research groups have investigated deregulated miRNA expression in MB pathogenesis and have evaluated miRNA expression levels relative to normal human cerebellum. However, normal cerebellum consists of a heterogenous population of cell types and may not represent the optimal normal control tissue 
. This is the first study to perform global miRNA profiling of human MB primary specimens relative to a putative cell of origin, CD133+ NSCs.
miRNA profiling has been used previously to classify tumors according to their molecular and histological sub-type 
. In this study, clustering analysis based on miRNA expression profiling identified two distinct sub-groups of primary MB specimens, a finding reminiscent of sub-types identified on the basis of mRNA expression profiles. Primary MB specimens in cluster three represented Shh-driven sub-type B, whilst specimens in cluster four represented sub-types C, D and E (Bertram et al., manuscript in preparation and 
). Additionally, our clustering analysis based on miRNA expression profiling revealed MB cell lines clustering as a distinct subgroup from primary MB specimens. This is consistent with previous reports in bladder cancer 
, cholangiocarcinoma 
and germ cell tumors 
highlighting underlying differences in miRNA expression between immortalized cell lines grown in culture and primary tumor specimens, in addition to sample heterogeneity due to the absence of stroma in cell lines. Although based on small sample sizes, our findings suggest that miRNA expression profiling is capable of classifying primary MB specimens into different tumor sub-groups. Larger sample numbers are required to further characterize these sub-groups at the miRNA level.
The majority of differentially expressed miRNAs in MB were up-regulated in comparison to normal CD133+ NSCs. This is in contrast to previous reports of general down-regulation of miRNA gene expression in primary human MB compared to normal adult and fetal cerebellum 
. A possible explanation for this discrepancy lies with the varying differentiation status of each of the normal populations used for comparison. Accumulating evidence indicates that miRNA expression profiles reflect the state of cellular differentiation, due to the critical role of miRNAs in the regulation of the terminal differentiation of numerous cell types 
. Consistent with this, miRNA profiling studies performed in embryonic stem cells, embryoid bodies and day 11 mouse embryos indicated that both miRNA expression levels and the number of different miRNAs expressed increased as the maturation process progressed towards terminal differentiation 
. Furthermore, recent studies have indicated that normal human adult cerebellum has higher expression levels of miRNAs when compared to pediatric cerebellum, consistent with developmentally regulated temporal expression of miRNAs in the brain 
. Therefore, we suggest that the over-expression of miRNAs in MB observed in this study reflects our comparison of miRNA levels in MB to multipotent NSCs.
Clustering of human miRNA genes is common, with 38% of known miRNA genes residing in clusters of two or more separated by less than 5000 nucleotides 
. We found multiple miRNAs with significantly altered expression in MB mapped to identical chromosomal regions, including 14q32, which contains one of the largest known bipartite miRNA clusters in vertebrates, consisting of 47 miRNA genes in total 
. Previous studies reported that miRNAs mapping to this region are transcribed as a long polycistronic transcript, spanning approximately 210 kb of the mouse genome 
. Evolutionary conservation of clustered miRNA genes suggests an important common biological function, with clustered miRNAs shown to co-regulate identical targets or other components of the same pathway 
. Consistent with this, we identified several miRNAs mapping to 14q32 that were predicted to target the same down-regulated target genes. Thus, the identification of aberrantly expressed miRNA genes clustered in identical chromosomal locations suggests they may have a critical role in the coordinate regulation of target gene expression in MB.
Up-regulation of miRNAs mapping to 14q32 has been reported in other cancers including acute myeloid leukemia 
and gastrointestinal stromal tumors 
. Additionally, findings obtained in this study are consistent with those described by Northcott et al
, where the up-regulation of several miRNAs mapping to 14q32 was observed in metastatic MB specimens with a neuronal differentiation signature. Given that a subset of primary MB specimens utilized in this study shared this signature, this suggests that the up-regulation of miRNAs mapping to 14q32 may be representative of metastatic disease. Despite this, the precise function of miRNAs mapping to 14q32 has not been elucidated. Expression of several miRNAs mapping to this region has been demonstrated throughout embryonic development in mice, progressively decreasing during postnatal development and ultimately becoming restricted to the adult brain 
. Gene ontology (GO) enrichment analysis performed on predicted targets of these miRNAs pointed to their roles in neurogenesis, cell motility and nervous system development 
. Moreover, several genes of the MAPK signaling pathway predicted to be targeted by miRNAs mapping to 14q32 have been previously shown to be aberrantly expressed in Group C and D MB primary specimens 
; for instance, the MAPK signaling genes, FGF13
, were down-regulated in primary MB specimens. Thus, our findings suggest that the up-regulation of miRNAs mapping to 14q32 in MB may regulate signaling pathways that have previously implicated in MB pathogenesis.
Enrichment analysis of predicted mRNA targets of the miRNAs that were aberrantly expressed in MB identified several pathways of interest in MB pathogenesis including neuronal development pathways, axonal guidance and Reelin signaling in neurons. Axonal guidance signaling has been shown to regulate both neuronal migration and survival 
, with several studies identifying aberrant expression of this signaling pathway in a variety of different human tumor types (Reviewed in 
). Similarly, Reelin signaling has been shown to play a critical role during embryonic brain development in the positioning and migration of Purkinje cells and subsequent expansion of granule cell precursors (GCPs) during early cerebellum development 
. Transgenic mouse studies have also indicated a role for Reelin in NSC proliferation, where an absence of RELN
expression results in decreased NSC proliferation 
and migration 
. IPA highlighted several putative miRNA-regulated signaling pathways containing the genes PIK3CA
as prominent in MB. Hyperactivation of Phosphoinositide 3-kinase (PI3K) signaling in several types of human cancers including MB is widely reported 
, resulting in increased activity of the downstream effector Akt 
. In turn, this promotes cancer cell growth 
, invasion 
and apoptosis resistance 
and thus PI3K and Akt have emerged as key targets for the development of novel anti-cancer therapeutics 
expression is reportedly up-regulated in primary MB 
, and PI3K signaling is critical to normal GCP and MB cell proliferation 
. Moreover, activation of PI3K signaling is essential for the proliferation of MB BTSCs residing in the perivascular niche following irradiation 
, suggesting that its targeted inhibition might impair the tumor resistance to radiation therapy. Thus, our findings suggest that the aberrant expression of miRNAs in MB may be involved in the regulation of critical pathways involved in embryonic and neuronal development, in addition to cell proliferation pathways previously linked to MB pathogenesis.
Integrative analysis of miRNA and mRNA gene expression is a robust and valuable approach for the identification of computationally predicted target genes in MB. Computational prediction of “true” miRNA targets is challenging, largely due to the context-dependent nature of post-transcriptional regulation resulting in a significant proportion of false-positive predicted miRNA/mRNA interactions 
. The approach taken in our study aimed to reduce the proportion of false-positive predictions, by integrating target gene predictions with gene expression profiles to select for inversely-correlated, functional miRNA/mRNA relationships. Whilst little is known about the role of miR-935-regulated target genes KIAA0232
plays an important role in mammalian osmoregulation 
is a widely expressed sodium/myo-inositol (MI) co-transporter, functioning to maintain intracellular MI levels 
. Alterations in MI levels have been suggested to affect phosphatidylinositol synthesis 
, with altered levels of Phosphatidylinositol 3-phosphate (PI3P
) and Phosphatidylinositol-4,5-bisphosphate (PIP2
) possibly linked to aberrant PI3K signaling 
. Taken together, experimental validation of several miR-935 regulated target genes in MB cells confirmed the validity of this approach, suggesting that predicted miRNA/mRNA interactions obtained in our study may represent functional miRNA/mRNA networks in MB.
A comparison of previously published MB miRNA expression datasets reveals significant differences with respect to expression patterns of individual miRNAs. Previous studies compared miRNA expression in primary MB specimens to both adult and fetal normal human cerebellum, with discrepancies likely due to varied source and age of the cerebellum used for comparison as previously discussed 
. Not surprisingly, deregulated miRNAs identified in our study differ from those previously identified as deregulated in MB, likely due to our comparison to normal control reference CD133+ NSCs. More specifically, comparison of the available miRNA data sets relative to adult and fetal human cerebellum with our data identified no significantly deregulated miRNAs as overlapping between the three studies (Figure S3
). Additionally, miRNA expression levels in different studies were obtained using varied profiling platforms, previously identified as a source of miRNA expression variation in other reports 
. Despite this, the identification of deregulated miRNAs specific to individual MB sub-types, as performed by Northcott et al
, revealed comparable findings to those obtained in this study. In particular, several miRNAs mapping to 14q32, including miR-376c, miR-495 and miR-539, were identified as up-regulated in MB specimens defined by a neuronal differentiation signature (sub-type C as per 
. This is consistent with our findings in which the majority of MB specimens were also characterized by either the neuronal differentiation (sub-type C) or mixed neuronal and photoreceptor gene signatures (sub-type DE). In addition, two miRNAs of the miR-106/-363 cluster mapping to chromosomal region Xq26.2 were specifically down-regulated in tumours characterized by a neuronal differentiation signature 
, consistent with the down-regulation of Xq26.2 miRNA, miR-106a* observed in this study. Therefore, future MB miRNA profiling studies should take into account potential inconsistencies between dataset comparisons due to both the platforms used and the source of normal reference RNA.
In summary, using an integrative approach we have identified aberrant miRNA-regulated networks that may be involved in the transformation of normal NSCs to BTSCs. These data provide a platform for future investigations aimed at characterizing the functional significance of these networks in MB pathogenesis.