PTEN regulates glioblastoma oncogenesis through chromatin-associated complexes of DAXX and histone H3.3

Glioblastoma (GBM) is the most lethal type of human brain cancer, where deletions and mutations in the tumour suppressor gene PTEN (phosphatase and tensin homolog) are frequent events and are associated with therapeutic resistance. Herein, we report a novel chromatin-associated function of PTEN in complex with the histone chaperone DAXX and the histone variant H3.3. We show that PTEN interacts with DAXX and, in turn PTEN directly regulates oncogene expression by modulating DAXX-H3.3 association on the chromatin, independently of PTEN enzymatic activity. Furthermore, DAXX inhibition specifically suppresses tumour growth and improves the survival of orthotopically engrafted mice implanted with human PTEN-deficient glioma samples, associated with global H3.3 genomic distribution changes leading to upregulation of tumour suppressor genes and downregulation of oncogenes. Moreover, DAXX expression anti-correlates with PTEN expression in GBM patient samples. Since loss of chromosome 10 and PTEN are common events in cancer, this synthetic growth defect mediated by DAXX suppression represents a therapeutic opportunity to inhibit tumorigenesis specifically in the context of PTEN deletion.

G lioblastoma (GBM) is the most common and aggressive form of cancer of the central nervous system. The TCGA (The Cancer Genome Atlas) data indicate that about 50% of GBMs harbour somatic alterations in the phosphatidylinositol 3-OH kinase pathway 1,2 . One of the essential regulators of this pathway that is significantly altered in GBMs (30-40%) is the PTEN tumour suppressor gene 1,3 , which encodes a phosphatase responsible for the removal of phosphate from the 3 0 position of the phospholipid second messenger phosphatidylinositol-3,4,5trisphosphate, thus opposing mitogenic signalling mediated by class 1 phosphatidylinositol 3-OH kinases 4 . The loss of PTEN function has been mechanistically linked to metastasis 5 , and lack of radio-therapy 6 and chemo-therapy 7,8 response in brain and breast cancer patients, indicating that PTEN is a key regulator of tumour sensitivity to multiple therapeutic approaches. It has been well established that different epigenetic, transcriptional and post-translational mechanisms control the level and function of PTEN. Moreover, PTEN protein-protein interactions can also affect its tumour suppressor properties [9][10][11] .
GBMs undergo genetic lesions that affect the epigenomic machinery that controls histone modifications, DNA methylation and gene expression. One such target is the histone H3 variant, H3.3, which is incorporated into chromatin in a cell cycle independent manner and is associated with transcriptionally active and silent chromatin in somatic and embryonic cells 12,13 . Two distinct and mutually exclusive H3.3 mutations (K27M and G34R/V) have been identified in paediatric GBMs 14,15 , associated with global downregulation of the repressive histone mark H3K27me3, DNA hypomethylation [16][17][18] , ALT (Alternative Lengthening of Telomere) phenotype 14 and upregulation of the MYCN pathway 19 . In adult GBMs, H3.3 expression is repressed by MLL5 (Mixed Lineage Leukemia 5), leading to chromatin reorganization and self-renewal 20 . Histone chaperones that are involved in the recruitment of H3.3 to chromatin are DAXX (death-domain associated protein), ATRX (alpha-thalassaemia/ mental retardation X-linked syndrome protein) and HIRA (histone cell cycle regulator) 12,[21][22][23][24] . Somatic mutations of DAXX and ATRX have been reported in adult pancreatic neuroendocrine tumours 25 , low grade gliomas 26, 27 and paediatric high grade gliomas 14 ; however, genetic alterations of DAXX are rare in other types of cancers.
To date, various approaches have been used to target chromatin deregulation in cancer cells [28][29][30] . Here, we report a novel chromatin-associated function of the PTEN tumour suppressor that represses oncogene expression and tumour growth in patientderived glioma xenografts through DAXX-H3.3 association. We show that DAXX physically interacts with PTEN, and PTEN regulates H3.3 loading on chromatin by limiting DAXX interactions with this histone, and thereby controls expression of several tumour-promoting genes. Moreover, DAXX inhibition affects global H3.3 deposition and gene expression, specifically suppresses intracranial tumour growth and significantly improves the survival of PTEN-null glioma-bearing mice. These results demonstrate a synthetic growth defect that occurs due to loss of these two tumour suppressor genes.

PTEN interacts with DAXX and controls oncogene expression.
Several reports have shown that PTEN can control tumorigenesis independent of its enzymatic activity, through its interaction with specific nuclear proteins [9][10][11] . To uncover further PTEN-nuclear interactions, an in silico analysis was performed using the Human Interactome Map (HiMAP) 31 bioinformatics site. From this analysis four interactome complexes were predicted for PTEN and for DAXX, that included previously reported interactions PTEN-TP53 (ref. 10), -MCRS1 (ref. 11) and -PML 9 and DAXX-TP53 (ref. 32), -MCRS1 (ref. 33) and -PML 34 , and a new interaction consisting of PTEN-DAXX (Fig. 1a). Since DAXX is a key regulator of gene expression 35,36 and has been shown to indirectly regulate PTEN stability 37 , we determined whether PTEN and DAXX could be physically associated. Pulldown assays using purified recombinant proteins (Fig. 1b) and total protein lysates from glioma cells that expressed different endogenous PTEN and DAXX levels (Fig. 1c) demonstrated that PTEN can physically interact with DAXX. To map the interaction domains for each protein, various GFP-PTEN and Flag-DAXX deletion constructs were co-transfected with full length Flag-DAXX or GFP-PTEN, respectively. These experiments showed that DAXX bound to the unfolded PTEN-hinge domain (amino acids 186-202) (Fig. 1d), while PTEN bound DAXX through its histone-binding domain (Fig. 1e).
DAXX is a histone chaperone protein that directly interacts with the histone H3.3 variant and facilitates its deposition on chromatin 23,24,38 . Using protein lysates from patient-derived glioma neurospheres (GBM-PDX) that express endogenous PTEN, DAXX and H3.3, we examined whether the PTEN-DAXX complex identified above interacts with H3.3. As shown in Fig. 2a and Supplementary Fig. 1a,b, endogenous PTEN co-immunoprecipitated with H3.3 and DAXX in GBM-sphere samples. We corroborated this result by performing sequential immunoprecipitation from nuclear extracts of a PTEN-null glioma-sphere that over-expresses exogenous PTEN (HK281-PTEN) (Fig. 2b), and also by quantitative immunofluorescence of endogenous proteins in GBM-spheres ( Fig. 2c and Supplementary Fig. 1c,d Pearson's coefficient in co-localized region equal to 0.4168 for TS576, 0.4526 for GBM39, 0.7665 for TS543 and 0.7344 for TS528 GBM-spheres). These results confirm the presence of a nuclear PTEN-DAXX-H3.3 tripartite complex in patient-derived GBM neurospheres.
We then asked if the PTEN-DAXX complex modulated the histone H3.3 chaperone function of DAXX. By immunonprecipitating DAXX from nuclear protein extracts of Pten-wt and Pten-null MEF (mouse embryonic fibroblast) cells we observed that the levels of H3.3 that co-immunoprecipitated with DAXX increased twofold in Pten-null cells when compared with Pten-wt cells (Fig. 2d, left panel and Supplementary Fig. 1e). Conversely, upon reconstitution of PTEN expression in a PTEN-deficient glioma cell line (U87), we observed a threefold reduction in the levels of H3.3 bound to DAXX (Fig. 2d, right panels and Supplementary Fig. 1e). These results indicate that DAXX-H3.3 interaction could be regulated by PTEN expression.
H3.3 deposition has been associated with both active and repressive chromatin 12,13 , while DAXX has been associated exclusively with gene repression 35,36 . This dichotomy led us to investigate the role of PTEN in the regulation of H3.3 deposition on chromatin and consequent gene expression. Four genes, either involved in neural stem cell proliferation and/or neuronal activity that have been reported to be regulated by PTEN [39][40][41][42][43] and DAXX 35,38 , were analysed (CCND1, MYC, FOS and BCL2). Regulation of gene expression by PTEN has been reported previously 9 . To determine if changes in the amount of H3.3 bound to chromatin were dependent on the lipid phosphatase activity of PTEN, ChIP assays were performed with anti-H3.3 in PTEN-deficient cells reconstituted with PTEN-wild type (PTEN-WT) or a PTEN-lipid and protein phosphatase inactive mutant (PTEN-G129R). As shown in Supplementary Fig. 2f,g, PTEN-G129R increased the amount of H3.3 bound to chromatin to the same level as PTEN-WT and also reduced the levels of H3.3 bound to DAXX ( Supplementary Fig. 2h), indicating that H3.3 deposition on chromatin was regulated by PTEN independently of its phosphatase activity.
DAXX has been shown to compete with DNA for H3.3-H4 tetramer formation 44,45 and we found that PTEN competes with H3.3 for the DAXX-histone binding domain (Fig. 1e). Based on these results, the effect of PTEN on DAXX cellular compartmentalization was investigated using cell fractionation. As shown in Fig. 2i, a reduction in the DAXX-chromatin fraction and a reciprocal increase in the DAXX soluble nuclear fraction occurred in Pten-null cells compared with Pten-wt cells. Upon PTEN reconstitution in PTEN-deficient cells, the amount of DAXX associated with chromatin increased from 37 to 61% (Supplementary Fig. 3a). Corroborating these results, DAXX immunofluorescence quantification in PTEN-WT cells displayed a specific signal associated with nuclear bodies (Fig. 2j-l and Supplementary  Fig. 3b); however, the DAXX signal was highly diffuse in the nucleus and cytosol in PTEN-deficient cells (Fig. 2j-l and Supplementary Fig. 3b). No changes in nuclear body distribution of ATRX and HIRA were observed in Pten-wt and Pten-null cells ( Supplementary Fig. 3c,d). In order to interrogate if ATRX or an ATRX mutant (R1426*) that is present in gliomas affect DAXX-PTEN and/or DAXX-H3.3 association we overexpressed these constructs and analysed their effect. No changes in DAXX-PTEN and DAXX-H3.3 associations were observed after overexpressing ATRX-WT or ATRX-R1426* mutant ( Supplementary Fig. 3e Fig. 5e), explained in part by their quiescent or slowly proliferative nature 48,49 . Together, these results illustrate that DAXX inhibition in PTEN-deficient cells restores the deposition of H3.3 on chromatin, promotes oncogene repression and compromises cellular proliferation.
In order to confirm that DAXX-knockdown directly affects oncogene expression through H3.3 and to eliminate the possibility of off-target effects, we reconstituted DAXX-kd/ PTEN-deficient cells with a DAXX-shRNA-resistant construct (HA-DAXX). As shown in Supplementary Fig. 6a,b protein and mRNA levels of CCND1, MYC, FOS and BCL2 genes were restored after DAXX re-expression in DAXX-kd cells. H3.3 enrichment on the promoter of those genes was also restored to normal levels compared with shControl cells after reconstitution of HA-DAXX or PTEN re-expression in DAXX-kd/PTENdeficient cells ( To test if these cellular and molecular effects observed after DAXX knockdown were PTEN dependent, DAXX expression was knocked down in glioma cell lines and GBM-spheres that express wild-type PTEN. No significant changes in total protein levels of CYCLIN-D1, MYC, FOS and BCL2 were detected after DAXX knockdown in PTEN-WT cells compared with shControl/PTEN-WT cells ( Fig. 3b and Supplementary Fig. 7a,e). Additionally, no or faint changes in the enrichment of H3.3 ( Fig. 3d and Supplementary Fig. 7b) or in cell cycle progression (Supplementary Fig. 7c,f) were observed in Daxx-kd/PTEN-WT cells in comparison with shControl cells. DAXX knockdown in PTEN-WT cells slightly affected cell proliferation in glioma cells ( Supplementary Fig. 7d,g), which is perhaps related to other molecular mechanisms associated with DAXX 50 . These data indicate that PTEN is a critical component for H3.3 chromatin deposition-associated gene repression, while DAXX suppression, in the context of PTEN deficiency, re-establishes H3.3-mediated oncogene repression ( Fig. 6h and Supplementary  Fig. 21).
Further analyses were performed overlapping the H3.3 DiffBind peaks from ChIP-seq with DiffExp genes from RNA-seq. In total 1,390 genes were overlapping between the two data sets, where a subset of 133 genes were upregulated and enriched for H3.3, and 37 genes were downregulated and enriched for H3.3 (Po0.05 and log2foldChange41) (Fig. 4f). Gene ontology analysis confirmed enrichment in pathways associated with nervous system development, regulation of MAPK cascade and neuronal differentiation ( Supplementary  Fig. 11c,d). These data show that DAXX inhibition robustly alters H3.3 genomic distribution leading to affects on gene expression in PTEN-null/GBM neurospheres.
We next studied whether DAXX inhibition can compromise the oncogenic behaviour of GBM-PDX cells grown in neurosphere conditions (Supplementary Fig. 12a). Immunoblot analysis of transcription factors involved in the development of gliomas (OLIG2, MYC, SOX2 and PAX6) 51-53 showed a specific decrease in expression in PTEN-deficient cells (GSC11, GSC23, HK281) compared with PTEN-WT cells (GBM6, TS675, GBM39, TS543) ( Supplementary Fig. 12b,c), suggesting that DAXX disruption in PTEN-null/GBM-PDXs affects the expression of transcription factors implicated in gliomagenesis. A downregulation of ATRX expression was also observed after DAXX inhibition in both conditions ( Supplementary Fig. 12c), PTEN-expressing and PTEN-null GBM-spheres, as has previously been reported by other groups 12,23 .
Additionally, we determined if DAXX knockdown affects GBM-PDX proliferation. Knocking down DAXX resulted in a specific reduction of number and size of spheres formed by PTEN-null (GSC11 and GSC23) cells, compared with PTENexpressing (GBM6 and TS576) cells (Fig. 4g,h), Po0.001. Furthermore, using an in vitro limiting dilution assay, DAXX knockdown resulted in a three-to eightfold reduction in the self-renewal capacity of PTEN-null GBM-PDXs compared with DAXX-kd/PTEN-WT spheres (Fig. 4i, Supplementary Fig. 13). In order to investigate if DNA replication was affected after DAXX disruption, since it has been reported that PTEN regulates DNA regulation and repair 9,54 , we quantified the percentage of positive cells relative to two components of the fork replication complex (RPA32-P and ATRIP-P). As is shown in Supplementary Fig. 14 DAXX inhibition increases the percentage of RPA32-P-and ATRIP-P-positive cells (GSC23, 17% and 12% increase, respectively) specifically in PTEN-null GBM-spheres. We next evaluated the effect of DAXX suppression on the differentiation capacity of GBM-PDX neurospheres.
Promoter  In contrast to shControl cells, there was a 30-50% decrease in OLIG2 ( Supplementary Fig. 15a, Po0.0010.0001) and a 20-30% decrease in GFAP-positive cells ( Supplementary Fig. 15b, Po0.001) when DAXX knockdown cells were incubated in differentiation conditions. In general, these results illustrate that DAXX inhibition, independent of ATRX, disrupts GBM-PDX oncogenic properties selectively in a PTEN-deficient genetic background, in part through downregulation of transcription factors that preserve glioma proliferation and upregulation of tumour suppressor genes.
DAXX expression is upregulated in gliomas. Having established that DAXX disruption inhibits tumour growth and increases survival in GBM-PDX models, we next examined if DAXX gene expression was altered in different human gliomas. By using the REMBRANDT and TCGA databases we found a statistically significant (Po0.0001) upregulation of DAXX expression in GBMs, oligodendrogliomas and astrocytomas in comparison with normal brain (Fig. 6a). This was apparent in the classical, mesenchymal and proneural GBM subtypes (Fig. 6b). However, no significant changes in DAXX protein signal were observed within PTEN-positive and PTEN-negative adult GBM samples (Fig. 6c, Chi-square 0.7639, P ¼ 0.6825, n ¼ 68); but a significant anticorrelated expression (cor ¼ À 0.298, P ¼ 0.001, n ¼ 166) between DAXX and PTEN-WT was found in the GBM-TCGA database (Fig. 6d) and confirmed by immunohistochemistry in adult GBM tissues (Fig. 6e, cor ¼ À 0.3721 P ¼ 0.0025, n ¼ 67). A similar anticorrelated DAXX/PTEN expression pattern was also observed in invasive breast carcinoma (BRCA, cor ¼ À 0.325, P ¼ 1.26e-27, n ¼ 1,096) from the TCGA data set ( Supplementary  Fig. 18a); suggesting a more general cancer-related gene expression regulatory mechanism. No significant correlation was observed between ATRX/PTEN and HIRA/PTEN ( Supplementary  Fig. 18b,c), indicating a specific PTEN-DAXX regulatory mechanism.
We next investigated the biological consequence of DAXX upregulation in human GBMs. Gene set enrichment analysis showed that the three most overrepresented gene sets that positively correlated with DAXX expression in GBM patients were E2F targets, G2M checkpoint components and MYC targets (Fig. 6f,g); in concordance with our data showing that DAXX knockdown affects cell cycle progression, cellular proliferation ( Supplementary Figs 4 and 5) and tumour growth ( Fig. 5 and Supplementary Fig. 16). The prognostic impact of DAXX genetic alterations (mutations) in GBMs was also interrogated using the TCGA data set. We found that 1% of GBMs have DAXX alterations (missense mutations); however, a difference in overall survival rate compared with non-altered cases was not apparent ( Supplementary Fig. 19a, P value 0.544). The overall survival rate in patients with both DAXX and PTEN alterations was not significant (P value 0.976) in GBM cases ( Supplementary  Fig. 19b). Our analysis indicates that DAXX expression is upregulated in gliomas and inversely correlated with PTEN.
Finally, we analysed gene expression levels in the normal human brain using the Allen human brain database 55 . Here, it was determined that PTEN and H3F3B (H3.3) have similar expression profiles in comparison with CCND1, MYC, FOS and BCL2 genes in the same brain structure regions ( Supplementary  Fig. 20). Particularly in the metencephalon (MET), the intensity of PTEN expression was 3-6 times higher than expression of the other analysed genes, using different probes. These analyses are consistent with our general model, proposing that PTEN controls gene expression through the regulation of H3.3 and DAXX.

Discussion
Our study suggests that PTEN is part of a chromatin complex with DAXX and H3.3, and negatively regulates genes involved in oncogenesis ( Fig. 6h and Supplementary Fig. 21). Since DAXX recruits proteins like H3.3 to PML-nuclear bodies (PML-NBs) 24,56 , and PML-NBs have been shown to regulate PTEN 37 , we propose that DAXX, H3.3, PML and PTEN may form a chromatin complex that regulates gene transcription. We suggest that in the absence of PTEN an unincorporated H3.3-chromatin fraction is recruited to PML-NBs in a DAXX-dependent manner 56 , leading to an increase in a DAXX-H3.3 soluble fraction. Upon DAXX inhibition, we speculate that H3.3 is liberated from PML-NBs and is hence restored for chromatin binding. We propose a model that in PTEN-deficient tumour cells, DAXX removes H3.3 from chromatin (Fig. 6h, Supplementary  Fig. 21a), probably by competing for chromatin binding, as has been reported by other groups 45 . Therefore, inhibition of DAXX restores H3.3 on the chromatin and inhibits oncogene expression (Fig. 6h, Supplementary Fig. 21b). The anti-tumorigenic effect mediated by DAXX inhibition does not work in PTEN-expressing cells because PTEN can also bind to H3.3 and we speculate that this attenuates H3.3 chromatin binding ( Supplementary Fig. 21c).
Four tumour-related genes associated with neural stem cell proliferation and neuronal activity, and regulated by PTEN [39][40][41][42][43] and DAXX 35,38 were studied. Particularly, MYC and CCND1 have been reported to be upregulated after PTEN disruption in progenitor cells 40,42 , and associated with brain hyper-proliferation in a Pten knockout mouse model 40 . Here we demonstrate that PTEN impinges upon MYC and CCND1 expression at the transcriptional level by increasing the loading of a repressive DAXX-H3.3 complex on the chromatin. In contrast, MYC and CCND1 overexpression that occurs in the context of PTEN deficiency can be abrogated by DAXX inhibition, which restores chromatin loading of repressive H3.3. Furthermore, genomic analysis in DAXX-knockdown/PTEN-deficient GBM samples display a genome-wide H3.3 distribution change (1,751 genes with H3.3 different binding signal) and upregulation of several tumour suppressor genes and downregulation of various oncogenes (1,403 genes with differential expression), including CCND1, MYC, FOS, SOX2 and OLIG2, compared with shControl/PTEN-deficient GBM samples. In concordance with the literature 12,38,57 , we show that H3.3 enrichment correlates with an upregulation and downregulation of genes involved in nervous system development and neuronal differentiation.
It is well documented that the H3.3 variant is regulated and incorporated in the chromatin by distinct proteins. Histone H3.3 is preferentially loaded at euchromatic regions by HIRA 12,22 and at heterochromatic regions by DAXX 23,24 . Furthermore, the ATRX/DAXX complex is required for targeting H3.3 to telomeric chromatin 12 . However, DAXX and ATRX have a distinct chromatin-binding profile, where DAXX preferentially binds to promoter regions 58 and regulates H3.3 loading of immediate early genes after neuronal stimulation 38 . More recently, it was reported that MLL5 (Mixed Lineage Leukemia 5) represses H3.3 expression in adult GBMs allowing global reorganization of chromatin and self-renewal 20 . These data suggest that several histone chaperons and chromatin regulator proteins, including PTEN, can be involved in H3.3 deposition and its expression, and consequently chromatin regulation. In this study, we did not observe changes in the expression or localization of ATRX and HIRA in PTEN-deficient cells, no changes in DAXX-PTEN and DAXX-H3.3 association after ATRX overexpression, nor a correlated expression with PTEN-WT. However, we did detect a downregulation of expression of ATRX after DAXX inhibition, as has been reported previously 12,23 . Our results indicate that DAXX disruption specifically affects GBM-PDX oncogenesis in PTEN-null models independently of ATRX.
DAXX and ATRX mutations have also been correlated with an alternative lengthening of telomeres (ALT) phenotype in pancreatic neuroendocrine tumours 59 and paediatric GBMs 14 ; however, intact telomeres have been observed in PTEN-deficient cells 9 , where chromosomal translocations and centromeres breakages are mainly affected. We reason that because genetic alterations of DAXX are uncommon in adult GBMs, oncogene transcription and chromosomal instability may drive cellular transformation mediated by PTEN disruption and DAXX deregulation through nuclear functions.
Additionally, our studies show that DAXX inhibition in GBM-PDX neurospheres suppresses tumour growth and increases survival, specifically in a PTEN-deficient background, in part by negatively regulating the expression of oncogenes implicated in gliomagenesis. It has been shown that SOX2, MYC and OLIG2 are required to maintain proliferation in progenitor cells and they have been implicated in different types of cancer 28,[51][52][53]60 . We suggest that downregulation of expression of these GBM-TFs can be associated with the PTEN-DAXX-H3.3 complex, since H3.3 is enriched near the transcription binding sites of these genes, as we observed from the genomic data and in concordance with the literature 12 , and DAXX disruption only affects their expression in PTEN-deficient cells.
In a therapeutic context, DAXX-H3.3 interaction can be disrupted in PTEN-null cells using staple peptides as reported by Kim and collegues 61 for the disassociation of an EZH2-EDD complex in a leukaemia model; or by using small molecules which have been efficient at antagonizing chromatin associated proteins and their interactions with other proteins 62 . Another strategy to target DAXX is by inhibiting its expression at the transcriptional level which can be attempted by performing high-throughput gene expression modulation by small molecules (GEMS) screening of compounds 28,63 .
In summary, we propose that PTEN-DAXX-H3.3 is a chromatin complex that regulates gene transcription ( Fig. 6h and Supplementary Fig. 21). Our study nominates DAXX as a new therapeutic target to revert tumorigenesis caused by PTEN loss of function in GBMs. Additionally, a DAXX-inhibition strategy offers an opportunity for other PTEN-null tumours where MYC and CYCLIN-D1 are upregulated, including medulloblastomas, endometrial cancer, breast cancer and melanoma 8,46,47 .     Cellular fractionation. Subcellular protein fractions were extracted according to the manufacturer's instruction (Thermo Scientific) and 10 mg of proteins were resolved by SDS-PAGE followed by immunoblotting.
Immunofluorescence microscopy. Cells were plated on poly-D-lysine-coated glass coverslips (Thermo Scientific), fixed with 10% formalin (Sigma-Aldrih), blocked with 2% of BSA IgG-free (Jackson ImmunoResearch) and stained with primary antibodies overnight at 4°C. Secondary antibody was added for 1 h at room temperature. Coverslips were mounted on microscope glass slides using Fluro-Gel with DAPI (Electron Microscopy Science) followed by visualization using confocal microscopy (Leica SP5 confocal with resonant scanner). A detailed description of immunofluorescence acquisition and analysis is in Supplementary Methods. Tumour size measurement and survival analysis. Animals were observed for neurological signs and the relative fluorescence signal of the xenografts were analysed by fluorescence molecular tomography (PerkinElmer) and quantified using TrueQuant 3.1 software (PerkinElmer). For survival analysis, animals were killed when they showed signs of distress and morbidity.
Densitometry quantification. Immunoblots were acquired with ChemiDocMP (Bio-Rad) and the intensity signal was quantified by densitometry analysis with Image Lab software.
Immunohistochemistry and tissue microarray. Slides were deparaffinized and rehydrated by washing steps of 3 min in xylene, xylene:ethanol 1:1, 100% ethanol, 95% ethanol, 70% ethanol, 50% ethanol and water. After deparaffinization, sections were boiled in citrate buffer (pH 6.0) for 25 min. Sections were then treated with 5% serum-blocking solution for 20 min. A detailed description of the method is in Supplementary Methods.
In silico protein-protein interactions. New PTEN nuclear interacting complexes were simulated using the bioinformatics site Human Interactome Map 31  TCGA, REMBRANDT and Allen Brain analysis. Gene expression, genetic alterations and survival rate analysis from TCGA, REMBRANDT and Allen Human Brain Atlas are described in Supplementary Methods.
Statistical analysis. Data sets were analysed by unpaired t-test or multiple comparisons one-way ANOVA or two-way ANOVA according to the experiment using GraphPad Prism software. *Po0.05, **Po0.001 and ***Po0.0001. Kaplan-Meier curves and comparison of survival were analysed using Long-rank (Mantel-Cox) test.
Data availability. Data generated during the study have been deposited in Sequence Read Archive (SRA) SRP090820.