Foxm1 controls a pro-stemness microRNA network in neural stem cells

Cerebellar neural stem cells (NSCs) require Hedgehog-Gli (Hh-Gli) signalling for their maintenance and Nanog expression for their self-renewal. To identify novel molecular features of this regulatory pathway, we used next-generation sequencing technology to profile mRNA and microRNA expression in cerebellar NSCs, before and after induced differentiation (Diff-NSCs). Genes with higher transcript levels in NSCs (vs. Diff-NSCs) included Foxm1, which proved to be directly regulated by Gli and Nanog. Foxm1 in turn regulated several microRNAs that were overexpressed in NSCs: miR-130b, miR-301a, and members of the miR-15~16 and miR-17~92 clusters and whose knockdown significantly impaired the neurosphere formation ability. Our results reveal a novel Hh-Gli-Nanog-driven Foxm1-microRNA network that controls the self-renewal capacity of NSCs.

Hh-Gli-driven expression of Nanog was shown to control a set of microRNAs (miRNAs) whose target genes encode proteins that inhibit the proliferation and diminish the self-renewal of these cells 9 .
In the present study, we used next-generation mRNA-and miRNA-sequencing to further explore this Hh-Gli-driven molecular network in early post-natal NSCs from the murine cerebellum. Our goal was to expand the map of this network by identifying other molecular players involved in the Hh-regulated self-renewal of these NSC pools.

Results
High-throughput transcriptome profiling of cerebellar NSCs. Cerebellar NSCs from postnatal day 4 (P4) mice were grown in stem-cell-selective medium, as described elsewhere 3 . As expected, under these conditions, the cells displayed high-level expression of stemness genes (Nanog, Nestin) and of Gli1 (Supplementary Figure 1A). Transfer of these NSCs to differentiation medium 3 was followed by significant increases in the expression of genes encoding astrocytic, neuronal, Purkinje, and oligodendrocytic cell markers (Supplementary Figure 1B).
Paired-end polyA + RNA-sequencing was used to profile the transcriptomes of NSCs grown in stem cell and differentiation media (NSCs and Diff-NSCs, respectively; three replicates of each). A total of 988 genes were differentially transcribed by the cells under these two conditions. NSCs and Diff-NSCs were clearly segregated, as observed in the hierarchical clustering of the 988 differentially expressed transcripts (DETs) (Fig. 1A).
Functional analysis of the DETs using the DAVID platform (Database for Annotation, Visualization and Integrated Discovery) revealed significant enrichment (Bonferroni-corrected P < 0.05) for the four Gene Ontology categories reported in Table 1 and detailed in Supplementary Figure 2. The most interesting clue that emerged from this analysis was the over-representation of genes involved in p53 signalling. This pathway is a well-known negative regulator of NSC self-renewal 9,10 , whose activity is modulated by signalling through the Hh-Gli-Nanog axis 3,11 .

Hedgehog-Gli pathway components enriched in NSCs.
To identify other molecular players with potential roles in Hh-Gli-driven self-renewal of cerebellar NSCs, we compiled a list of 53 genes known to be regulated by Hh-Gli signalling in settings (physiologic or pathologic) other than NSCs (Supplementary Table 1). Nine of these genes were differentially transcribed in NSCs before and after differentiation (Supplementary Table 2, Supplementary Figure 3A). Six of the nine genes encode cyclins (Ccnb2, Ccnb1, Ccna2, Ccnd2, Ccne1, Ccnd1) known to be involved in the regulation of cell cycle and cell division in NSCs. The seventh, Sema6a, is involved in nervous system development, in particular, in axon guidance 12 , and the eighth gene, Insm1, has been reported to be involved in mouse cerebellar development 13 .
Analysis of this list revealed that Foxm1 was the Hh-Gli-regulated transcription factor most markedly expressed in NSCs prior to differentiation. Foxm1 is a transcriptional activator 14 whose role as a downstream mediator of Hh-Gli signalling has thus far been documented exclusively in human cancer cells [15][16][17][18] . The relation between Foxm1 and Hh-Gli signalling was also evident in our murine cerebellar NSC model. As shown in Supplementary Figure 3B,C, mRNA and protein levels in our NSCs displayed good correlation with Hedgehog pathway activation, which is known to peak in these cells on post-natal days 4-6 7,19 . In addition, suppression of Hh signalling with a Smoothened inhibitor cyclopamine-KAAD (KAAD) significantly diminished Foxm1 mRNA levels in P4 cerebellar NSC (Fig. 1B), and similar effects were observed in human neural progenitors (NHNP) (Fig. 1C). Consistent with their high-level expression of Gli1 (Supplementary Figure 1A), the NSCs displayed strikingly higher levels of Foxm1-at both the transcriptional ( Fig. 2A) and protein levels (Supplementary Figure 3C)-prior to their differentiation. Review of our mRNA-seq data confirmed this NSC-associated upregulation for two of the four known protein-coding Foxm1 transcript isoforms (ENSMUST00000073316 [Foxm1-201] and ENSMUST00000112148 [Foxm1-202]) (Supplementary Table 3). As shown in Fig. 2B, this finding was validated by PCR performed with isoform-specific primers and by immunoblot analysis, which revealed clear predominance in the NSCs of the 757-amino-acid Foxm1-201 protein isoform.
Foxm1 mediates Hh-Gli-driven self-renewal of the NSCs. To explore the functional relevance of this upregulation, we transfected NSCs with siRNA directed against Foxm1 (siFoxm1) and evaluated their self-renewal capacity, as reflected by their ability to form neurospheres. As shown in Figs. 2C,D, Foxm1 knockdown was associated with significantly impaired neurosphere formation. Interestingly, when we prevented Hh activation with the use of cyclopamine-KAAD (KAAD), (Fig. 2D) the silencing of Foxm1 did not exert any additional effect. We then investigated whether the knockdown of Foxm1 had an effect on differentiation, proliferation and apoptosis. As shown in Fig. 2C, the silencing of Foxm1 led to the up-regulation of differentiation markers βIII-tubulin and S100b. We did not observe any significant modulations in proliferation and apoptosis markers (Supplementary Figure 4). Comparison of the human FOXM1 and murine Foxm1 promoter regions revealed a of βIII-tubulin (*P < 0.05: 0.043) and S100b (*P < 0.05: 0.031) in NSCs transfected with siCtrl and siFoxm1.  Table 4). Consistent with recent findings on its human ortholog 17,18 , the murine Foxm1 promoter was found to harbour eight putative Gli-binding sites (s1-s8) ( Fig. 3A and Supplementary Information). We performed quantitative PCR-ChIP assays in order to determine whether both transcriptional activators of the Hh-Gli pathway could occupy the Foxm1 promoter in these putative binding sites. Experiments were performed in NSCs both before and after differentiation to quantitatively assess Gli recruitment and histone H3 acetylation (AcH3, a marker of transcriptional activation). Promoter occupancy was reported in all putative Gli binding sites (s1-5) and (s6-8) as evidenced by the higher percentage in NSCs when compared to Diff-NSCs. In particular, Gli2 reported a significantly higher percentage of occupancy in the Gli (s1-5) binding sites of the Foxm1 promoter in NSCs in respect to Diff-NSCs (Fig. 3B). Similarly, Gli1 promoter occupancy in the Gli (s6-8) binding sites of the Foxm1 promoter was significantly higher in NSsC when compared to Diff-NSCs (Fig. 3C). No binding was observed in an unrelated chromatin region (Supplementary Figure 5A). Our data allow us to conclude that Gli-binding sites were bound by both transcriptional activators of the Hh-Gli pathway in NSCs.
Foxm1 modulates stemness through the activation of specific miRNAs in NSCs. As previously noted, the Hh-Gli-regulated stemness marker, Nanog, modulates the proliferation and self-renewal of murine cerebellar NSCs via miRNA-mediated suppression of genes promoting cell-cycle arrest and differentiation 3,9 . We wondered whether miRNAs might also play a role in Foxm1's effects on NSC self-renewal. As shown in Fig. 4A, miRNA-sequencing studies identified 80 miRNAs that were differentially expressed in NSCs and Diff-NSCs.
To identify miRNAs likely to be direct targets of Foxm1, we examined the promoter regions of the 40 miRNAs that were upregulated in NSCs (Supplementary Table 5) and found putative Foxm1 binding sites in 20. To increase our chances of identifying targets with biological relevance to NSC self-renewal, we restricted our subsequent analyses to the 15 miRNAs on this list with the most statistically significant upregulated expression in the NSCs (Table 2 Quantitative PCR-ChIP assays were then performed on NSCs before and after differentiation to quantitatively assess Foxm1 recruitment and histone H3 acetylation (AcH3) at the promoter region of each miRNA gene putatively targeted by Foxm1. The 15 miRNAs were transcribed singly or as part of a cluster ( Table 2). In detail, in case of a cluster, for further experiments expression levels of one representative cluster member are reported. For all miRNAs tested, Foxm1 recruitment was higher in the NSCs that in the Diff-NSCs. We focused our attention on those miRNAs where promoter occupancy, as evidenced by AcH3, was significantly more intense in the NSCs (Fig. 4B,C). No binding was observed in an unrelated chromatin region (Supplementary Figure 6B). Foxm1 recruitment indeed was statistically significant for miR-130b, miR-301a, and miRNAs belonging to the miR-15~16 and miR-17~92 clusters (Fig. 4B).
The results of the previous experiments point to miR-130b, miR-301a, and miRNAs of miR-15~16 (n = 4) and miR-17~92 clusters (n = 2) as particularly important mediators of Foxm1's effects in NSCs. Of note, expression levels of the above mentioned miRNAs were down-modulated after silencing of Foxm1 (Supplementary Figure 6C). Our claim was also supported by the effects observed in the cells after locked nucleic acid (LNA)-mediated depletion of these miRNAs. As shown in Fig. 5A, the NSCs' capacity for neurosphere formation was not significantly reduced by anti-miR knockdown of any single miRNA or miRNA cluster. However, it was significantly impaired by combined depletion of miR-130b, miR-301a, and miR-19a (3 LNA combination), and the impairment was not significantly greater when miR-15b was also depleted (4 LNA combination). We investigated whether the knockdown of these miRNAs had an effect on differentiation, proliferation and apoptosis. As observed after the silencing of Foxm1 (Fig. 2C), also the knockdown of these downstream miRNAs resulted in the upregulation of differentiation markers, but had no effects on proliferation and apoptosis markers (Fig. 5B).
To identify possible mechanisms underlying the stemness-promoting effects of this miRNA network, we explored the genes targeted by the ChIP-confirmed miRNAs listed in Table 2. A miRTarBase (http://mirtarbase. mbc.nctu.edu.tw/) search returned validated murine targets for only three of these miRNAs: miR-15b-5p, miR-130b-3p, and miR-92a-3p (Supplementary Table 6). We therefore extended our search to the literature on each miRNA, focusing specifically on validated or putative target genes (in any species) whose downregulation could explain the combined effect of these miRNAs in NSC self-renewal.
The results that emerged reiterated the importance of p53 signalling, whose loss/suppression is essential for the maintenance of embryonic stem-cell pluripotency 20,21 . Of particular interest was a report showing that miR-130b-3p regulates CD133 + tumour-initiating cells in human hepatocellular carcinoma by targeting

No.
MicroRNA log 2 fc* adj. P-value   22 , which encodes a downstream component of the p53 signalling pathway. Our previous work showed that Trp53inp1 expression in murine cerebellar NSCs is also suppressed by microRNAs of the miR-17~92 cluster, and the upregulated expression of these miRNAs was attributed to signalling through the Hh-Gli-Nanog axis 9 .
In light of our current findings, p53 signalling in these cells also appears to be under the control of a second miRNA network, this one regulated by Foxm1. In support of this hypothesis, mirSVR prediction scores provided by microRNA.org (http://www.microrna.org/microrna/home.do) indicate that murine Tp53inp1 is a likely target of miR-130b (mirSVR score: −0.0029). The same applies to miR-301a, another Foxm1-regulated microRNA belonging to the miR-130b family (mirSVR score: −0.0030). The Foxm1-regulated miRNA network that modulates p53 signalling might also comprise miR-92a-3p, whose validated targets include Trp63, another member of the p53 family of transcription factors. Because of the high-level sequence homology that characterizes these transcription factors, p63 and p73 are capable of transactivating p53-responsive genes, thereby causing cell cycle arrest and apoptosis 23 .
We cannot exclude of course other possible mechanisms of action for the Foxm1-mediated miRNAs. There are many ways in which these miRNAs can exert their function and an example is the report of high expression of miR-15a in MeCP2-deficient neural progenitors 24 , indicating its possible role in maintaining the undifferentiated state.
A role for Nanog in Foxm1 regulation. Interestingly, members of the miR-17~92 cluster are components of both the Nanog-regulated 9 and Foxm1-regulated miRNA networks, but this was the only commonality observed. This evidence indicates that Nanog and Foxm1 activate largely non-overlapping cohorts of miRNAs to ensure suppression of p53 signalling in cerebellar NSCs. How these two networks interact to achieve this goal is unclear. Interestingly, Nanog has been identified as a target of Foxm1 14,25,26 . To determine whether this relation might be reciprocal, we re-examined the Foxm1 promoter region for evidence of Nanog binding sites. As shown in Fig. 6A, four putative binding sites for Nanog were found −3790 to −3277 bp upstream from the Foxm1 TSS. (For details, see Supplementary Information).
ChIP experiments (Fig. 6B) demonstrated endogenous Nanog at these four sites in both NSCs and Diff-NSCs. However, pre-and post-differentiation occupancy rates were significantly different only at s2-s3, where the higher Nanog occupancy in NSCs was associated with greater activation of Foxm1. No binding was observed in an unrelated chromatin region (Supplementary Figure 7A). To further elucidate the relation between these two transcription factors, we performed dual luciferase reporter assays in NSCs transfected with Foxm1 wild-type promoter or an s2-or s3-defective mutant promoter (Fig. 6C). Ectopic expression of Nanog in these cells resulted in substantial activation of the Foxm1 wild-type promoter, whereas this induction was significantly reduced by deletion of critical nucleotides of s2 and s3, suggesting that these sites are required for Nanog binding onto the Foxm1 promoter. Interestingly, Foxm1 promoter luciferase reporter was significantly activated by Nanog also in 293 T cells (Supplementary Figure 7B).
Taken together, these results indicate that the Hh-Gli-driven miRNA networks regulated by Nanog and Foxm1 are characterized by bidirectional crosstalk, which might conceivably allow more finely tuned, combinatorial regulation of cerebellar NSC self-renewal.

Discussion
The main goal of our study was to identify new molecular mechanisms involved in the Hh-Gli driven regulation of cerebellar NSC functions. As shown in Fig. 7, our findings indicate that Hh-Gli signalling controls self-renewal of murine NSCs in the P4 cerebellum via p53-pathway targeting miRNAs (i.e., those of the miR-17-92 cluster, that are also crucial for the expansion of cortical NSCs in vivo 27 ), which are regulated not only by Nanog, as previously reported 9 , but also by a second transcriptional activator, Foxm1.
Our findings show that Foxm1 expression in these cells is controlled by Gli1 and Gli2, directly but also indirectly via their modulation of the expression of Nanog, which was shown to transcriptionally activate Foxm1. This dual mechanism of control is rendered even more complex by the fact that has been reported that Nanog is also a target of Foxm1 14,25,26 . As for the functions of Foxm1 in cerebellar NSCs, we also provide intriguing preliminary evidence suggesting that, aside from its above-mentioned co-regulation with Nanog of the miR-17-92 cluster, Foxm1 also controls other miRNAs (miR-130b and miR-301a) in these cells that putatively target the p53-signaling pathway. We used mRNA-and miRNA-sequencing to identify Foxm1-regulated miRNAs (other than those of the miR-17-92 cluster) with high-level expression in NSCs that drops significantly after these cells undergo differentiation. Knockdown of these miRNAs significantly impairs cerebellar NSC neurosphere formation by NSCs, suggesting these cells' capacity for self-renewal is sustained by a Foxm1-regulated miRNA network.
Signals transmitted through the Hh-Gli-Foxm1 and Hh-Gli-Nanog axes clearly converge on miR-17~92 cluster members to suppress the expression of Trp53inp1. However, Trp53inp1 is also a putative target of two of the microRNAs regulated by Foxm1 alone, miR-130b 22 , which has already been associated with maintenance of embryonic NSCs 30 , and miR-301a. This observation suggests the possibility of combinatorial and partially redundant control of the p53 signalling cascade, which underlines the importance of this pathway in NSCs. Combinatorial repression of Trp53inp1, and consequently of p53, by both the Hh-Gli-Foxm1-and Hh-Gli-Nanog-driven miRNA networks, might be part of a hypothetical "bypass axis" that serves to maintain control of NSC self-renewal even in the presence of gene mutations affecting key components of the regulatory network. It is important to stress that the ability of miR-130b and miR-301a to downregulate Trp53inp1 expression requires verification in future studies. Nonetheless, self-renewal of the cerebellar NSCs was significantly reduced by combined knockdown of all four Foxm1-regulated miRNAs.
In conclusion, the regulatory network here proposed highlights the importance of Hh-Gli signaling and its downstream effector Foxm1 in cerebellar NSC maintenance and provides further insight into the complex mechanisms involved in the fine-tuning of stemness and differentiation cues in these cells.

Methods
Unless otherwise specified, all commercial products were used in accordance with the manufacturer's protocol.
Cell models. Murine cerebellar NSC cultures. Cerebellar NSCs were isolated from P4 wild type black 6 /C56 (C57BL/6) mice (n = 8 per group) and cultured, as previously described 3 . In brief, freshly dissected cerebella were placed in HBSS supplemented with 0.5% glucose and penicillin-streptomycin and dissociated, mechanically and enzymatically. The cells were maintained as mycoplasma-free neurosphere cultures in selective stem-cell medium consisting of serum-free DMEM-F12 supplemented with 0.6% glucose, insulin 25 mg/ml, N-acetyl-L-cysteine 60 mg/ml, heparin 2 mg/ml, B27 supplement without vitamin A, EGF 20 ng/ml, and bFGF 20 ng/ml. To induce differentiation, NSCs were mechanically disaggregated and plated on poly-lysine coated support 48 h in differentiation medium consisting of the medium described above, prepared without the EGF/bFGF and supplemented with platelet-derived growth factor (PDGF; 10 ng/ml) (Sigma, P3076) 3   Neurosphere-forming assay. Neurospheres grown in the stem-cell-selective medium were dissociated to single cell through dissociation solution non-enzymatic buffer (Cod: C5789, Sigma-Aldrich). Viable cells were counted after Trypan-blue exclusion and re-plated at clonal density (1-2 cells/mm²) in 96-well plates containing stem-cell selective medium. Results were expressed as the percentage of cells that gave rise to neurospheres.
RNA isolation and quantitative RT-PCR. RNA was isolated from cells, as previously described 31 . The High Capacity cDNA reverse transcription kit (Applied Biosystems Life Technologies, ThermoFisher) was used to synthesize cDNA, as previously described 32 .
Quantitative reverse transcription (RT-PCR) analysis was performed using a High Capacity cDNA Reverse Transcription kit. mRNA expression was analysed on cDNAs using the ViiA ™ 7 Real-Time PCR System, SensiFAST ™ Probe Lo-ROX (Bioline), TaqMan gene expression assay according to the manufacturer's instructions (Applied Biosystems). Each amplification reaction was performed in triplicate, and the average of the three threshold cycles was used to calculate the amount of transcripts in the sample (SDS software, AB). mRNA quantification was expressed, in arbitrary units, as the ratio of the sample quantity to the calibrator or to the mean values of control samples. All values were normalized to three endogenous controls: ß-2-microglobulin, Hprt and Gusb.
Knockdown studies. Small-interfering-RNA knockdown of Foxm1 was performed in NSCs with ON-TARGETplus SMARTpool (L-057933-01-0005), control ON-TARGETplus Non-targeting siRNA knockdown (D-001810-02-05) (Dharmacon). Hiperfect reagent (Qiagen) was used for transfections. Cells were harvested for assay 72 h post-transfection. Transfection efficiency was > 80%, monitored using the fluorescent control SiGLO (Dharmacon) and calculated as the percentage of fluorescent cells. All experiments were performed using the using the best knockdown efficiency without off-target effects, controlled using three different housekeeping genes (ß−2-microglobulin, Hprt and Gusb) for RNA studies and using Actin or Hsp70 for immunoblots.
Putative miRNA target genes. MicroRNA.org (http://www.microrna.org/microrna/home.do) database was queried for putative miRNA target genes, where miRNA-target interactions were predicted with miRanda 3.3a and scores were calculated with mirSVR (good mirSVR score less than or equal to −0.1).
Data availability. All data generated or analysed during this study are included in this published article (and its' Supplementary Information files).