AMBRA1 is differentially expressed in MB subtypes and its levels correlate with patient survival
AMBRA1 is known to be crucial for both nervous system development and autophagy. Indeed, in mice at embryonic day (E)8.5, strong AMBRA1 staining is detected throughout the neuroepithelium and, later in development, it becomes abundant in the entire developing nervous system, as well as in other tissues [29, 50, 58]. Indeed, in adult mice cerebellum, Ambra1 expression depends on the neuronal subtype: for instance, in the cerebellar cortex, Ambra1 is highly expressed in the Purkinje cell layer, whereas it is almost excluded from the granule cell layer . Interestingly, by analysing Ambra1 protein levels in various neural stem cell types, we found a positive expression in E14.5 neuronal stem cells (NSCs), in cerebellar NSCs (cNSCs) as well as in granule neuron precursors (GNPs) from postnatal P4 mice (Supp. Fig. 1a, b, online resource). Altogether, these results confirm the prominent role of AMBRA1 in early nervous system formation and highlight its importance in cerebellar development.
Genetic programs involved in early development of the nervous system are often hijacked in brain cancers. To investigate the role of AMBRA1 in MB, we collected tumour samples of 47 cases (Supp. Table S1), including classic (n = 30), desmoplastic/nodular (DN) (n = 7), and LCA (n = 10) MB, and AMBRA1 expression was evaluated by immunohistochemistry (IHC). As shown in Fig. 1a, b, AMBRA1 is abundantly expressed in both classic and LCA subtypes, when compared with DN. Interestingly, when we examined MB samples in a genomic subgroup-specific fashion, we found that AMBRA1 expression is much higher in MBWNT and MBGroup3 than in the other two groups (Fig. 1c). Real-time PCR (qPCR) in patient-derived samples confirmed that AMBRA1 mRNA levels are increased in MBWNT and, at a less extent, in the more heterogeneous MBGroup3 tumors (Suppl. Fig. 1c, online resource). To further validate the pattern of AMBRA1 expression observed in MB subgroups, we analysed samples from two public MB datasets (deposited at http://r2.amc.nl; Pfister n = 167 samples and Cavalli n = 763 samples) [13, 48, 64]. Consistent with our data, we found that AMBRA1 transcriptional levels are significantly higher in both MBWNT and MBGroup3 compared to both MBSHH and MBGroup4 (Fig. 1d) (Supp. Fig. 1d, online resource).
In the light of the recent recognized heterogeneity across Group 3 and 4, we found that AMBRA1 is significantly upregulated in both G3α and G3β compared with other MB subtypes (SHHα,γ,δ–G4α,β) (Fig. 1e). Moreover, we decided to inspect the eight molecular subtypes of MBGroup3 and MBGroup4 described by Northcott et al. , that are highly congruent with the eight groups reported by Sharma et al. [39, 84]. As reported in Supp. Fig. 1e (online resource), we observed that AMBRA1 is significantly more expressed in subtype III (a bona fide Group3 subtype) and MBGroup3 than in MBGroup3/4 and MBGroup4.
Of the highest importance in MB studies, and given these findings, we decided to stratify our cohort of MB patients into two groups according to AMBRA1 immunostaining intensities (negative/weak expression versus median/strong positive). Intriguingly, patients with higher AMBRA1 levels (n = 27) have poorer survival than those with lower AMBRA1 expression (n = 19, p = 0.01; Fig. 1f). This result is fully in line with what we found by analysing the overall survival of MBGroup3 patients listed in the Cavalli dataset (n = 113; p = 0.026), and even more significant when analysing G3α subtype (n = 55; p = 0.010) (Fig. 1g,h).
Of note, it is known that MYC family members play a critical role in each of the MB subgroups . c-MYC expression was found to be very high in both MBWNT and MBGroup3 (Supp. Fig. 1f, online resource) [13, 39, 64]. However, MBWNT and MBGroup3 are characterized by a completely different overall survival, and when enhancing c-MYC expression in mouse models of MB, none of these results in the formation of WNT subgroup . MBWNT patients, indeed, are characterized by excellent prognosis, which may be partly due to alterations in the brain vasculature that could increase chemotherapy penetration .
Interestingly, despite the highly heterogeneous nature, at both molecular and histological levels, of the samples collected in public databases, we found a moderately positive correlation between AMBRA1 and c-MYC across the entire Cavalli dataset (r = 0.316, p = 3.7 × 10–19) and subtypes I-VIII (merged analysis of Group 3 and Group 4) described by Northcott et al. (r = 0.338, p = 4.23 × 10–4) (Supp. Fig. 1 g, h, online resource). Intriguingly, in a combined evaluation within this recent MB subclassification, AMBRA1 and c-MYC levels highly correlate among them in Group 3/4 (subtypes I, V, VII; r = 586, p = 6.71 × 10–4) (Supp. Fig. 1i, online resource). Further, although univariate survival analyses display significantly shorter progression-free survival associated with both high AMBRA1 and c-MYC levels in MBGroup3 and even in G3α (Cavalli dataset), multivariate Cox modelling shows that both AMBRA1 and c-MYC are independent risk factors for progression-free survival (see Supp. Fig. 1j, online resource), in both MBGroup3 and G3α subtype. Additionally, by analysing AMBRA1 levels in non-amplified c-MYC versus amplified c-MYC patients (G3γ-Cavalli dataset), no differences in AMBRA1 levels are found (Supp. Fig. 1k, online resource).
Next, AMBRA1 levels were investigated among normal cerebellum (NC) and different MB cell lines belonging to different MB subgroups and having distinctive c-MYC levels: DAOY = MBSHH low c-MYC, D283-Med = MBGroup3 high c-MYC, CHLA-01-Med = MBGroup3 c-MYC amplified [44, 90]. In agreement with our data on MB patients, we found that AMBRA1 mRNA levels were higher in both D283-Med and CHLA-01-Med (Fig. 1i).
These findings prompted us to look for a molecular correlation between AMBRA1 and c-MYC expression in MB.
AMBRA1 expression in MBGroup3 depends on c-MYC interaction with MIZ-1
To test the interplay between c-MYC and AMBRA1, we silenced c-MYC in both D283-Med and CHLA-01-Med cells. Upon c-MYC downregulation (for siRNA efficacy see Supp. Fig. 2a, online resource), we found that, in c-MYC-downregulated cells, AMBRA1 decreases at both mRNA and protein levels (Fig. 2a,b). Conversely, when we exogenously expressed c-MYC in DAOY cells, we observed a significant upregulation of AMBRA1 (Fig. 2c,d), this result supported a role for c-MYC in regulating AMBRA1 expression. To further investigate how c-MYC controls AMBRA1 levels, we next analysed the involvement of MIZ-1, a c-MYC cofactor known to directly regulate AMBRA1 transcription [98, 99] in both D283-Med and CHLA-01-Med cells. Similarly to c-MYC, when we down-regulated MIZ-1 by RNA interference (after selection of the most efficient anti-MIZ-1 siRNAs), we observed a significant decrease of AMBRA1 at both mRNA and protein levels (Fig. 2e) (Supp. Fig. 2b, online resource). By contrast, MIZ-1 overexpression was not able per se to increase AMBRA1 protein levels (Supp. Fig. 2c, online resource), this impinging on the c-MYC-dependence of MIZ-1 regulatory effect on AMBRA1. Given the crucial role of the Pox virus and Zinc finger (POZ) domain for the transactivation and repression activity of MIZ-1, to assess whether MIZ-1 transcription activity is necessary to control AMBRA1 expression, we performed rescue experiments in MIZ-1-silenced cells. As shown in Fig. 2f and Supp. Fig. 2d (online resource), we found that rescue with full length MIZ-1 (MIZ-1WT) but not with POZ-depleted MIZ-1 (MIZ-1∆POZ) restored AMBRA1 protein levels, this implying the requirement of MIZ-1 activity to AMBRA1 transcriptional upregulation. Given that c-MYC and MIZ-1 are able to act as a complex and that they represent a hallmark in MBGroup3 , we examined whether c-MYC-MIZ-1 interaction was necessary for AMBRA1 upregulation. First, by chromatin immunoprecipitation (ChIP) experiments, we found that both c-MYC and MIZ-1 bind the AMBRA1 promoter region (Fig. 2g), although c-MYC recruitment depends epistatically on the presence of MIZ-1 (Fig. 2h). To better investigate the relationship among these two transcriptional factors in AMBRA1 regulation, we used a specific c-MYC mutant construct (c-MYCV394D) deficient for MIZ-1 binding . As reported in Fig. 2i, exogenous expression of c-MYCWT increases AMBRA1 protein levels, at variance with that of c-MYCV394D, supporting the idea that c-MYC-MIZ-1 association is necessary for AMBRA1 regulation. Finally, to test whether or not MIZ-1/c-MYC complex activated AMBRA1 transcription, we engineered a fragment of the AMBRA1 promoter encompassing the MIZ-1-binding site , and performed transient reporter assays. Expression of MIZ-1 and c-MYC alone stimulates expression of the AMBRA1-promoter reporter gene, whereas Miz-1∆POZ and c-MYCV394D are ineffective in its transactivation (Fig. 2j). Interestingly, co-expression of both MIZ-1 and c-MYC significantly increases AMBRA1 promoter efficacy, relative to both single transfections and control vector. In sum, these data demonstrate that c-MYC-MIZ-1 interaction is required for the regulation of AMBRA1 expression in MBGroup3 stem cells.
AMBRA1 strictly controls stem cell potential in various MB-derived cell lines
Given the differential AMBRA1 expression in MB subgroups, among which MBGroup3 is the most aggressive, we decided to check for the presence of an AMBRA1-dependent phenotype. D283-Med cells were selected as the best cell model to represent MBGroup3 in this study of the mechanism of action of AMBRA1, because of higher AMBRA1 levels compared to CHLA-01-Med cells. To this aim, we first performed an RNA-seq analysis, characterizing the global gene expression changes induced by AMBRA1 siRNA-mediated downregulation. Indeed, using gene set enrichment analysis (GSAA), we observed that AMBRA1 loss reduces primarily the expression of gene sets that have been linked to represent hallmarks of stemness properties of cerebellar neuronal stem cells and MB stem cells (MBSCs) [59, 72] (Fig. 3a) (Supp. Fig. 3a, online resource). Subsequently, based on this finding and on the knowledge that MBGroup3 cells display features of partially committed neural stem and progenitor cells [36, 47], we checked for the expression of the CD133 and NESTIN stem-cell markers among MB cell lines, and detected higher expression in c-MYC+ cells (D283-Med and CHLA-01-Med), when compared with DAOY (Supp. Fig. 3b,c, online resource). Of the highest importance, depletion of AMBRA1 by siRNAs strongly reduces expression of stem-cell markers at both mRNA and protein levels (Fig. 3b-c) (Supp. Fig. 3d, online resource). Of note, when we reintroduced AMBRA1 in stably AMBRA1-interfered (shAMBRA1) D283-Med cells, we were able to rescue CD133 protein levels, with this implying AMBRA1 requirement for CD133 upregulation (Fig. 3d).
Then, to further investigate the effects of AMBRA1 modulation on MB cell clonogenicity, we performed a soft-agar assay for D283-Med cells and a clonogenic assay for CHLA-01-Med cells. As shown in Fig. 3e,f, we found that loss of AMBRA1 results into a significant reduction in clonogenic ability [measured as the number of medullospheres (MS) formation] of all MB cells studied. To strengthen our findings, we also used MBSCs from human MB (hum MBSCs) , confirming a significant decrease of both their clonogenic potential and stemness marker expression levels after AMBRA1 silencing (Fig. 3g) (Supp. Fig. 3e, online resource). Then, to investigate the correlation between AMBRA1 expression and patient stem cell potential, we analysed (i) CD133 levels in 43 MB primary samples previously analysed for AMBRA1 by IHC, and (ii) CD133 expression in correlation with AMBRA1 in both Cavalli and Pfister datasets, by looking at different subtypes. Indeed, 18/19 samples highly expressing AMBRA1 are also positive for CD133 (Supp. Fig. 3f, online resource); moreover, we found that AMBRA1 and CD133 highly and significantly correlate in Group 3/4 (subtypes I–V–VII Pfister dataset; r = 579, p = 8.94 × 10–4) and in G3α (Cavalli dataset, r = 416 p = 4.69 × 10–4) (Fig. 3h) but not in the other subtypes analysed (G3β, G3γ, II–III–IV and VI–VIII subtypes) (Supp. Fig. 3g,h, online resource). We finally confirmed our results by culturing DAOY cells that usually express low levels of stemness markers in stem-cell media. Indeed, they give rise to MS and upregulate stem markers, including SOX2 and OCT4 (Supp. Fig. 3i, online resource) in addition to what previously described for c-MYC, NESTIN, NANOG, CD133 . Interestingly, we could detect, in this context, an increase in AMBRA1 expression (Fig. 3i). By contrast, knockdown of AMBRA1 strongly inhibits the ability of DAOY cells to form MS, which also resulted to be significantly smaller in size (Fig. 3j), as well as the expression of stemness markers (Supp. Fig. 3j, online resource). In agreement with previous results, using retinoic acid to induce D283-Med cells differentiation, we were able to induce a decrease in AMBRA1 expression (Supp. Fig. 3 k,l, online resource), this supporting the evidence of a close link between AMBRA1 and MB stem cell potential.
Autophagy enhances stemness in MB cells in vitro
Given the key roles of AMBRA1 in the upstream regulation of autophagy [18, 19, 63, 92], we decided to study the role of this process in MB stemness. First, we analysed the expression levels of autophagy regulators, such as BECLIN 1, ATG5, P62, ULK1, LC3, ATG7, ATG13 and WIPI2 among MB cell lines (Fig. 4a). Among the genes known to regulate autophagy, many of them are more expressed in both D283-Med and CHLA-01-Med than in DAOY cells, with AMBRA1 showing the most evident and consistent differential expression. Then, to determine whether MBGroup3 cells also exhibited increased autophagy activity, we used chloroquine (CQ: an inhibitor able to block late stages of autophagy, commonly used also to assess the autophagy flux) on both D283-Med and DAOY cell lines. As shown in Fig. 4b, we detected an increased lipidation of the autophagy marker LC3 in D283-Med than in DAOY cells, suggesting an increased basal autophagy. In addition, when two autophagy-related factors (BECLIN 1 and ATG7) were silenced in D283-Med cells (see Supp. Fig. 4a online resource for siRNA selection), a clear decrease in the stem phenotype was recapitulated (Fig. 4c,d) (Supp. Fig. 4a, online resource). Interestingly, by both acute (single treatment for 48 h) (Fig. 4e) or chronic (multiple treatments every 48 h) (Fig. 4f) administration of CQ to D283-Med cells, we obtained a decrease in the expression of stem cell markers, this implying a key role for autophagy in the maintenance of a stem-like phenotype. Consistent with this hypothesis, analysis of RNA-seq in D283-Med ATG7-silenced cells show that loss of ATG7 is sufficient to decrease the expression of genes involved in stemness (Fig. 4g) (Supp. Fig. 4b,c, online resource). To extend these findings to D283-Med cells, autophagy was also measured in DAOY cells after MS induction and we found that autophagy is induced when DAOY cells are cultured as MS (Fig. 4h). Last, to dissect whether the AMBRA1 pro-autophagic activity was essential for the regulation of stem potential, we took advantage of an AMBRA1 mutant construct (AMBRA1AA) impaired in its autophagic function . Indeed, as illustrated in Fig. 4i, AMBRA1AA is able to rescue at least in part CD133 protein levels.
Altogether, these data demonstrate that autophagy positively regulates stemness in MB cells and that pharmacological inhibition of autophagy reduces MB stem cell potential in vitro.
AMBRA1 downregulation or autophagy inhibition affect MB cell survival and migration, and induce their differentiation
Given the strong effect of AMBRA1 loss on stemness, we decided to investigate whether AMBRA1 downregulation could lead to MBSCs differentiation. As a proof of concept, we cultured shAMBRA1 D283-Med cells and human primary MBSCs  for 10 days and analysed the expression of differentiation markers such as the Glial Fibrillary Acidic Protein (GFAP) and Synaptophysin (SYP). As reported in Fig. 5a, shAMBRA1 cells show typical morphological changes of differentiated cells in both D283-Med and human primary MBSCs, together with significant upregulation of the two differentiation markers analysed (Fig. 5b). Similarly, chronic CQ treatment led to increase of both SYP and GFAP levels in D283-Med, suggesting that also autophagy inhibition can trigger MBSCs differentiation (Fig. 5c, d).
Next, we investigated the role of AMBRA1 on MB cell proliferation. In AMBRA1-silenced D283-Med cells, we found a significant decrease of cell viability and cell number (Fig. 5e, f) and an increase in cell death (Fig. 5g). By contrast, no significant differences were found upon ATG7 silencing (Supp. Fig. 5a, online resource).
Interestingly, through RNA-seq analyses performed following AMBRA1 silencing, we found that loss of AMBRA1 induces gene sets representing inactivation of TGF-β (Supp. Fig. 5b, online resource). While previous reports have associated the TGF-β pathway with pro-migratory potential of MB , a strict correlation between stemness and motility has been found to be crucial for MBGroup3 dissemination capability [28, 33, 46]. Moreover, in ATG7-silenced cells, two migration-related pathways (proteoglycans and adherent junctions, respectively, see Supp. Fig. 4b, online resource) are found significantly downregulated. Prompted by this evidence, we decided to analyse the expression of epithelial-to-mesenchymal transition (EMT) markers, such as Snail (SNAI1) and Vimentin (VIM), which are usually highly expressed in MBGroup3 cells . Indeed, individual silencing of AMBRA1, BECLIN 1 or ATG7 (and after 40 μM CQ treatment) impairs the expression of the EMT markers analysed in this experiment (Fig. 5h). Moreover, we also detected a decreased migration and invasion capabilities of siAMBRA1-, siATG7- and treated D283-Med cells (Fig. 5i) (Supp. Fig. 5c, online resource), this indicating that both AMBRA1 and autophagy downregulation influence cell migration in MB. Last, since EMT is known to induce chemo-resistance in MB , we checked whether the combination of autophagy inhibition (mediated by CQ) and cisplatin (CIS, a commonly used chemotherapeutic agent in MB ) could affect MB-cell survival. Both cell death and viability were analysed after either individual CQ and CIS treatments, or their combination. As shown in Fig. 5j, k, the combinatory approach decreases D283-Med cell survival, this implying that such an approach might have future therapeutic implications. In summary, our data demonstrated that AMBRA1 manipulation or autophagy inhibition negatively impact on the major phenotypical hallmarks of MBGroup3 cells.
Downregulation of both AMBRA1 and autophagy inhibits MBGroup3 aggressiveness in vivo
Due to the high levels of AMBRA1 expression in MBGroup3, we decided to determine the effect of AMBRA1 silencing on MB growth in vivo. The peritoneal metastatic D283-Med cell line represents an excellent model to study MB growth and metastatic dissemination in an orthotopic xenograft approach . After engineering D283-Med cells for the constitutive expression of GFP and luciferase (shCTRL), we downregulated AMBRA1 by lentiviral infection (shAMBRA1) and orthotopically injected these cells into cerebella of athymic nude mice. We then followed tumour growth by weekly bioluminescent imaging (Fig. 6a) for 41 days. Notably, tumour masses arising from shAMBRA1 cells were significantly smaller than those deriving from control cells, as reported in Fig. 6b. Interestingly, when mice were analysed for a longer period, we found that shAMBRA1 cells-injected animals survive much longer (median survival: 62 days) than controls (p = 0.03; Fig. 6c). Both IHC and qPCR of primary tumour tissues from xenografted mice show inhibition of stem potential and proliferation (decreased NESTIN (NES) and Ki67 expression) and induction of cell differentiation (increased SYP expression) (Fig. 6d, e) (Supp. Figure 6a, online resource). Furthermore, staining for p62 and LC3 indicates that, in this context, autophagy is inhibited (Supp. Fig. 6a, online resource).
In light of these findings, we next wanted to investigate the effect of pharmacologic inhibition of autophagy by CQ on MB growth and migration in a second immune-compromised mouse model, NSG (NOD scid gamma mouse), which represents the most consistent model for studying cancer metastases . First, we injected D283-Med-Luc cells into the cerebellum of NSG mice and, after confirmation of tumour engraftment by bioluminescent imaging (after 4 days), mice were randomized into two groups: those treated with either vehicle or CQ (60 mg/kg intraperitoneally every day). Animals were treated until they displayed tumour-associated morbidity and then euthanized (Fig. 6f). Although both groups generated similar sizes of primary cerebellar tumours (Supp. Fig. 6b, online resource), CQ-treated mice produced less detectable spinal metastases (Fig. 6g). Furthermore, IHC of the brain tumour tissues from these mice showed a decrease in CD133 and an increase in p62, LC3 puncta and β-III TUBULIN (β-III TUB) expression, respectively (Fig. 6h, i). Additionally, IHC of spinal cords dissected from the same mice displayed MB seeding cells that strongly express Ki67 (Fig. 6j). In sum, our data suggest that autophagy maintains stem cell potential and increases MBGroup3 invasion ability, highlighting this pathway as a suitable and druggable target for MB.
AMBRA1 depletion suppresses STAT3 signalling in MBGroup3
Importantly, one key factor regulating both the maintenance and migration of CSCs is the signal transducer and activator of transcription 3 (STAT3), a factor involved in the development of multiple cancer types [8, 45]. RNA-seq of AMBRA1 silencing in both D283-Med and CHLA-01-Med cells showed a shared downregulation pattern of c-MYC validated targets and STAT3 transcriptional activation (Supp. Fig. 7a, online resource). Given that (i) MBGroup3 is characterized by upregulation of active STAT3 [32, 100]; (ii) STAT3 directly regulates markers of pluripotent stem cells (e.g., CD133, NANOG, SOX2, c-MYC) ; and, of the highest importance, (iii) AMBRA1 is a critical regulator of the activity of CULLINs [4, 5, 16], an E3-ubiquitin ligase family indirectly controlling STAT3 activity, we next hypothesized that STAT3 signalling was also involved in AMBRA1-dependent or autophagy-dependent MBGroup3 stem potential.
Indeed, in agreement with others [9, 32], we found that MBGroup3 cells rely on STAT3, since genetic inhibition of STAT3 by gene knockdown (after selection of the most efficient siRNAs) results into a significant decreased expression of stemness markers, such as c-MYC and CD133 (Supp. Fig. 7b,c online resource), decreased proliferation (Supp. Fig. 7d, online resource), increased cell death (Supp. Fig. 7e, online resource) and decreased migration capabilities (Supp. Fig. 7f, online resource). Interestingly, we observed that AMBRA1 downregulation in D283-Med cells decreases STAT3 transcriptional targets, such as CD133, NANOG, FOXO3, CYCLIN D2 and, of the highest importance, c-MYC (Fig. 7a). Intriguingly, only a few of them (CD133, NANOG) decreased in ATG7-depleted cells (Supp. Fig. 7g, online resource). Moreover, in AMBRA1-deficient cells, we found a decrease of STAT3 activation by tyrosine-phosphorylation (Tyr 705)  and of the expression of STAT3 target genes in both D283-Med and CHLA-01-Med cells (Fig. 7b) (Supp. Fig. 7 h–k, online resource), including c-MYC, which does not occur at all after ATG7 silencing (Supp. Fig. 7 g, k, online resource). To verify whether STAT3 was responsible for the decrease in both the expression of stemness markers and cell viability after AMBRA1 downregulation, we overexpressed STAT3 in AMBRA1-deficient cells. As shown in Fig. 7c, d and in Supp. Fig. 7l (online resource), STAT3 is able of significantly restoring its-target genes expression levels and cell viability after AMBRA1 downregulation, this suggesting a crucial role for STAT3 in determining AMBRA1- function in MB context.
To unravel the mechanism behind this AMBRA1-mediated regulation, we set up to investigate the molecular axis that may lead to STAT3 activation. In more details, AMBRA1 has been shown to interact with the CRL4-DDB1 complex [that involves CULLIN 4 (CUL4)] to target Elongin C (ELOC), an essential adaptor protein that mediates the assembly of the CRL5-SOCS3 complex, which includes CUL5 . This complex in turn influences the stability of Suppressor of Cytokines signalling 3 (SOCS3) [16, 34]. Indeed, SOCS3 inhibits STAT3 activation by acting on the JAK2 kinase-dependent phosphorylation on Tyr 705 . We thus hypothesized that AMBRA1, besides regulating autophagy, could also control STAT3 activation by mediating the stability of its inhibitor SOCS3 in MB. Intriguingly, SOCS3 protein levels increase in AMBRA1-silenced cells, while a decrease was found at the mRNA level (Fig. 7e) (Supp. Fig. 7m, online resource). We then determined whether preventing SOCS3 accumulation in AMBRA1-deficient cells was sufficient to rescue STAT3 activation. After testing three different siRNAs against SOCS3 (see Supp. Fig. 7n, online resource), we generated cells in which AMBRA1 and SOCS3 were co-depleted; as expected p-STAT3 is significantly decreased in cells upon AMBRA1 downregulation, while siRNA-mediated silencing of SOCS3 restores STAT3 phosphorylation, confirming the existence of an epistatic relationship among AMBRA1, SOCS3 and STAT3 (Fig. 7f). Finally, since AMBRA1 effect on CRL5 depends on the association between CRL4-DDB1 and AMBRA1 , we investigated whether the binding between AMBRA1 and DDB1 was crucial for SOCS3 regulation. To this aim, we used two specific AMBRA1 mutant constructs, deficient for its interaction with the CUL4-DDB1 complex (AMBRA1∆WD40 and AMBRA1S113A) ; as shown in Fig. 7g, after their overexpression in AMBRA1-silenced cells, an increased stability of SOCS3 was observed compared to wild-type form of AMBRA1. Altogether, we thus found that AMBRA1 controls STAT3 activity via SOCS3, independently of autophagy and via CRL4-DDB1, thereby increasing the levels of c-MYC and other stem cell markers.
Combined targeting of autophagy and STAT3 signalling impairs MBGroup3 progression
Due to the role of AMBRA1 in controlling both autophagy and STAT3 activity, and since STAT3 inhibition has been shown per se to impact autophagy , we speculated that inhibition of both STAT3 and autophagy could be more effective than either alone in MB treatment. To this aim, we used two STAT3 inhibitors, NSC 74859 (also known as S3I-201) that inhibits STAT3 dimerization  and WP1066, a STAT3 inhibitor that readily crosses the blood–brain-barrier [26, 38] that prevents STAT3 phosphorylation. First, a strong effect is observed upon both NSC 74,859 and WP1066 treatment, with decreased p-STAT3 levels and reduced expression of its target genes (c-MYC and CYCLIN D2) (Fig. 8a) (Supp. Fig. 8a, online resource). Interestingly, after pharmacological combination of both STAT3 and autophagy inhibition (by CQ 40 μM), we observed a significant decrease of proliferation and increased cell death when compared to STAT3 or autophagy inhibition alone (Fig. 8b, c) (Supp. Fig. 8b, c, online resource). Moreover, we observed an increased amount of LC3B-II in STAT3i-treated cells compared to vehicle-treated controls both basally and in the presence of CQ to inhibit flux (Supp. Fig. 8d, online resource), in line with the anti-autophagic functions of STAT3 .
To extend these in vitro findings to in vivo tumour growth, we evaluated the combination of STAT3 and CQ inhibition after injection of D283-Med-Luc cells into the cerebellum of NSG mice. In this experiment, both WP1066 and CQ were given every second day to reduce stress on the mice. As shown in Fig. 8d, e and Supp. Fig. 8e (online resource), brain tumors harvested from mice treated with the combination of WP1066 (40 mg/kg) and CQ (60 mg/kg) were significantly smaller than tumors from mice treated with STAT3 or autophagy inhibition alone at day 28. Additionally, the combination of both drugs induces less detectable spinal metastases. Also, mice treated with the combination of WP1066 and CQ exhibited a 13-day increased median survival compared with both vehicle-treated mice and CQ-treated mice, and a 10-day increased compared with WP treatment alone (Fig. 8f). We can thus conclude that concurrent inhibition of STAT3 and autophagy suppresses MBGroup3 growth more efficiently than inhibition of STAT3 and CQ alone.
Since MB represents a very heterogeneous family of tumours requiring different therapeutic approaches to obtain reliable results, we decided to test the combination of both WP1066 and CQ in a MBGroup3 mouse model, in which tumours are induced by c-Myc and orthodenticle homeobox 2 (Otx2) overexpression in P0 CD1 mice cerebella, as previously described [10, 57]. Mice treated with the combination of WP1066 (30 mg/kg) and CQ (60 mg/kg) for 24 days present a higher survival rate, when compared with vehicle-treated mice (Fig. 8g). Intriguingly, western blotting analysis of primary tumour tissues show a significant upregulation of Ambra1 protein levels, at variance with normal cerebellum (Supp. Fig. 8f, online resource), supporting the role of c-MYC overexpression in enhancing AMBRA1 levels.
In sum, we can conclude that upon c-MYC/MIZ-1-mediated upregulation of AMBRA1 transcription, this factor enhances both autophagy and STAT3-c-MYC signalling in high-risk MB. Molecular dissection of this axis primed us to demonstrate that acting on both pathways may represent a valid combined approach to counteract high-risk MB growth and spread (Fig. 8h).