Paper IV

Despite the indisputable role of FBXW7 in carcinogenesis, detailed knowledge of how the FBXW7 pathway is altered in specific malignancies and in response to therapy is limited.

A combined approach to identify novel substrates of FBXW7, based on protein expression in FBXW7-WT and KO HCT116 cells and prediction of GSK3β phosphorylation and CPD motifs, shortlisted SOX9 as a putative substrate[28]. Using immunoprecipitation assays, we confirmed that FBXW7 and SOX9 interact and that this interaction depends on an evolutionarily conserved CPD motif phosphorylated by GSK3β. Consistently, mutagenesis of

this CPD motif, resulted in loss of interaction with FBXW7 and stabilisation of SOX9.

Furthermore, depletion of FBXW7α but not the other isoforms resulted in marked stabilisation of SOX9 protein while over-expression of FBXW7α promoted turn-over of SOX9 in a proteasome-dependent manner. Accordingly, FBXW7 ubiquitylated SOX9 in vitro and cell lysate-based ubiquitylation assays. In contrast, mutagenesis of the FBXW7 substrate recognition domain, in this case R465A, or deletion of the F-box abolished SOX9 ubiquitylation. Regulation of SOX9 by FBXW7 was found in both untransformed cells and various cancer cell lines, including colorectal carcinoma, breast cancer, glioma and MB cell lines.

As SOX9 has previously been shown to determine MB sub-type, promote neural stem cell renewal and tumour development[249], we wondered if deregulation of FBXW7-mediated degradation affects these processes in MB. Therefore, we re-analysed whole-exome sequencing data from 133 SHH-MB tumours and identified FBXW7 mutations in about 11%

of this cohort[307]. Many of these mutations were situated within the WD40 domains affecting three arginines crucial for substrate interaction, which suggests deregulation of substrates as the mechanistic basis for the presence of FBXW7 mutations in MB. Additionally, FBXW7 mRNA levels are low in MB, particularly in SHH-MB, as compared to normal cerebellum, and low expression levels are associated with poorer MB patient outcome. Although higher FBXW7 expression was linked to a significant survival advantage in group 3 MB, separation of low and high-risk groups based on FBXW7 expression was greatest in SHH-MB. In line with FBXW7-mediated SOX9 degradation as a putative determinant of MB malignancy, increased SOX9 protein levels were observed in tumours with lower FBXW7 mRNA expression and vice versa.

To determine if indeed SOX9 was the decisive target for FBXW7 in MB, presumably explaining its effect on patient survival, we orthotopically transplanted the human MB cell line DAOY overexpressing SOX9 either alone or together with FBXW7α into athymic nude Foxn1^nu mice and found that FBXW7α expression significantly prolonged survival and, strikingly, completely abolished metastasis formation. FBXW7 expression also reduced ability of DAOY cells to migrate in trans-well migration assays in vitro. The in vivo metastasis phenotype was reproduced using MB-inducing cells (MICs) isolated from the MYCN^T58A model. In this case, overexpression of SOX9 resulted in markedly increased metastatic spread as compared to MIC control cells. Furthermore, MB patients were significantly more likely to present with metastasis when tumours expressed high SOX9 protein, further supporting a clinical significance of the discovered mechanism.

To explore differential gene regulation as a result of increased SOX9 stabilisation in greater detail, we stably expressed SOX9-WT or the SOX9-CPDmt (T236/240A) mutant in MB002 cells, a human MB cell line, and profiled these cells by RNA sequencing. Among the gene sets most clearly upregulated in SOX9-WT as compared to control cells and further elevated upon expression of SOX9-CPDmt was the EMT gene set, encompassing upregulation of genes such as SLUG or vimentin. Gene set enrichment analysis further revealed that metastasis signatures were enriched upon induction of SOX9 overexpression, in line with clinical and in vitro migration and metastasis data above.

Based on the link between EMT, metastasis and drug resistance, we next asked whether stabilisation of SOX9 confers resistance to cisplatin, commonly used for the treatment of MB.

Overexpression of SOX9 in MB002, DAOY and MICs increased cell viability upon cisplatin exposure which could be reverted by additional FBXW7α induction. Moreover, mice engrafted with DAOY-SOX9-WT or SOX9-CPDmt did not respond to cisplatin therapy, in contrast to mice transplanted with DAOY cells featuring endogenous SOX9 expression.

In the light of increased cisplatin resistance due to SOX9 upregulation, we reassessed our RNA sequencing datasets. We found DUSP2 and ATP7A, pointed out in previous reports as determinants of cisplatin resistance[308, 309], among the most differentially expressed genes between empty vector control and SOX9-CPDmt. Thus, these genes may underly increased cisplatin resistance in response to SOX9 overexpression.

These findings prompted us to explore ways to sensitise MBs exhibiting high SOX9 levels.

Intriguingly, hyperactivation of PI3K/AKT/mTOR signalling in human MB is associated with poor survival[307] and AKT phosphorylates and inhibits GSK3[310], which in turn may mitigate SOX9 degradation. We hypothesized that pharmacological inhibition of PI3K/AKT/mTOR signalling might promote SOX9 turn-over. Indeed, treatment with AZD2014, AZD5363 or AZD8186 inhibiting mTOR, AKT and PI3K, respectively, enhanced FBXW7α-dependent turn-over of SOX9. Furthermore, AZD2014 and cisplatin synergistically reduced viability in MB cell lines but not upon expression of SOX9-CPDmt, which supports the idea that targeting SOX9 through pharmacological upregulation of GSK3β-FBXW7α-dependent degradation sensitises MBs to cisplatin chemotherapy.