Genomically amplified Akt3 activates DNA repair pathway and promotes glioma progression

To gain a broad perspective of the Akt isoforms in human cancer, we analyzed their copy number variation (CNV) across multiple cancer types. Akt3 comprised the most frequently amplified isoform in many cancers, including GBM, ovarian, melanoma, endometrial, and breast cancer (Fig. 1A). Moreover, Akt3 was the least frequently deleted isoform. In contrast, Akt1 was the least frequently amplified and most deleted isoform. These CNV data suggest that Akt3 may be the most critical isoform in an array of cancer types. Indeed, a prominent role for Akt3 has been well-documented in melanoma, playing a critical role in tumor progression (18) and rendering cells resistant to apoptosis (19). Akt3 has also been reported to be the dominant isoform in hormone unresponsive breast (20) and prostate cancer cells (21), and a recurrent MAGI3-Akt3 fusion gene was recently discovered in triple-negative breast cancer (22). Given that Akt3 represents 50% of total Akt in the normal brain, has been documented to be essential for normal brain development (23), and was preferentially amplified in GBM, we hypothesized that Akt3 is the most critical isoform involved in glioma.

We directly tested the ability of Akt1, Akt2, and Akt3 to promote glioma progression in the RCAS/Ntv-a glial-specific mouse system. As previously reported, injection of the RCAS plasmid carrying PDGFB transduces normal, nestin-positive glial cells that are engineered to express the RCAS receptor, tv-a, leading to formation of low-grade glioma (LGG) in more than 80% of mice (24–26). The introduction of Akt1 in conjunction with PDGFB led to only 16% incidence of high-grade glioma (HGG), which was not statistically significant from PDGFB alone. However, when Akt2 or Akt3 was injected with PDGFB, the incidence of HGGs rose to 78% and 97%, respectively (Fig. 1B), which is significantly higher than PDGFB alone (P < 0.001, Fisher’s exact test) and Akt1 (P < 0.001, Fisher’s exact test). In addition, Akt3 produced a higher incidence of HGGs compared with Akt2 (P < 0.05, Fisher’s exact test). Akt3 tumors were particularly striking, as they frequently exhibited bilateral patterns and were very large, with pseudopalisading necrosis and marked vascular proliferation. Even the high-grade tumors differed among the Akt isoforms, with the Ki67 proliferation index again reflecting the Akt3 > Akt2 > Akt1 hierarchy (P < 0.001, Fisher’s exact test) (Fig. 1B and Fig. S1). The striking invasiveness, as evidenced by the number of bilateral tumors (Fig. S1), is consistent with a prior study that found Akt3 to promote anchorage independent growth in transformed astrocytes (17). Additionally, the cellular localization pattern of the activated form of Akt (pAkt-473) isoforms appeared to be distinct, perhaps providing clues of differential pathways being activated. Tumors derived from Akt1 displayed mostly nuclear staining of pAkt; Akt2 tumors displayed primarily membranous staining; whereas Akt3 tumors displayed strong nuclear and cytoplasmic staining (Fig. 1B). These findings are consistent with a recent phosphoproteomics screen in immortalized lung fibroblasts, which revealed more evenly distributed nuclear and cytoplasmic substrates in cells expressing Akt3, compared with those in Akt1- and Akt2-expressing cells (27). This study highlighted Akt1/3-mediated phosphorylation of an RNA processing regulator, which led to alternative splicing of FGFR2. Additional studies are reporting potential nuclear functions of Akt3, including Ago2-mediated translational repressive activities (28), transcriptional control of IGFBP3 in lung cancer cells (29), and ERBB2, ERBB3, and ERα in breast tumor cells from a mouse transgenic model (30). Although these effects are likely mediated by multiple complex systems, Akt3 has been reported to control the stability of the major nuclear export protein, chromosome maintenance region-1 (CRM-1), leading to altered mitochondrial biogenesis and autophagy (31). It is plausible that Akt3 may regulate other biological processes via similar mechanisms.

To further explore the potential basis behind the functional differences, we compared Grantham polarity plots among the Akt isoforms, which reflect charge distribution across amino acids (32). Substantial differences were observed in two regions in the P domain (amino acids 40–55 and 120–150) and the R domain (amino acids 445–460), but the kinase domain (K domain) was the most consistent among the Akt isoforms (Fig. 2A and Dataset S1). Differences in polarity may affect protein interactions, leading to divergent cell signaling pathways. Notably, the K domain, containing one of the activating phosphorylation sites, Thr-308, had very little variation in polarity, indicating that this domain is likely not responsible for the functional differences among the isoforms. A complementary measurement of hydropathy via Kyte–Doolittle plots was consistent with the polarity plots, revealing similar variable regions (Fig. S2 and Dataset S2).

We hypothesized that the variable regions in polarity and hydropathy could be responsible for the functional differences observed in glioma progression. To test this, we performed domain swapping to generate chimeras in which the P or R domains of Akt2 or Akt3 were swapped with Akt1. The resulting chimeras were referred to as 2P and 2R or 3P and 3R, respectively (Fig. 2B and Table S1). We injected each chimera with PDGFB into Ntv-a mice to determine its ability to induce progression. The data revealed that the P or the R domain of Akt2 or Akt3 strongly initiated progression. These results may have important implications for the development of future inhibitors. Prior attempts to inhibit the Akt pathway in solid tumors have involved pan Akt inhibitors, most notably perifosine, a lipid-based compound that prevents Akt translocation to the cell membrane. Phase II clinical trials with this inhibitor have not yielded favorable outcomes (33, 34). More recently, the allosteric inhibitor MK-2206, which inhibits Akt1/2 more selectively than Akt3, was evaluated in Phase I and II clinical trials, with limited single agent activity (35). Our data suggest that future attempts to inhibit Akt should specifically target Akt2 and/or Akt3.

Given that there was a definitive hierarchy among the isoforms in promoting glioma progression, we sought to analyze the gene expression patterns of representative tumors to determine if distinct pathways were being activated. We performed microarray analysis on RNA isolated from high-grade tumors derived from PDGFB alone and each PDGFB/Akt combination. Gene expression patterns from Akt1/2/3 tumors were compared with that of PDGFB tumors. Consistent with the inability of Akt1 to induce glioma progression, there were strikingly no differentially expressed genes in Akt1 tumors compared with PDGFB. Conversely, tumors derived from Akt2 and Akt3 had 306 and 145 unique differentially expressed genes, respectively (Fig. 3A and Dataset S3). Moreover, Akt2 and Akt3 shared only 67 differentially expressed genes, indicating that they likely activate distinct pathways. To assess the specific pathways these differentially expressed genes were associated with, we performed gene ontology (GO) enrichment analysis using the WebGestalt software (36). The significantly enriched (false discovery rate < 0.05) GO categories were organized as a tree with broad/specific categories at the top and bottom. The broad categories at the top of the Akt2 tree were phosphate metabolic and protein modification processes, and the most specific category at the bottom of the tree was amino acid phosphorylation (Fig. 3B). The corresponding genes from the Akt2 tree are depicted in a heatmap, and demonstrate that indeed these genes have separate patterns from Akt1 and Akt3 (Fig. 3C). The Akt3 GO analysis was very different, with broad categories including cellular macromolecule metabolic processes, response to cellular stress, and response to DNA damage stimulus. The most specific categories were DNA-related, including DNA metabolic processes, DNA replication, and DNA repair (Fig. 3D). These results reflect the strong nuclear pattern of p-Akt staining in the mouse tumors (Fig. 1B). Many DNA repair machinery genes were significantly up-regulated, including PolD1, PolE, PolE2, EXO1, PCNA, and FEN1 (Fig. 3E). Given the striking pattern and potential implications of up-regulated DNA repair pathways, we next analyzed the human gene expression patterns in the TCGA GBM dataset to assess whether human tumors showed similar pathways that were discovered in the mouse model. Partial correlations were calculated between the expression of each Akt isoform and a panel of 356 DNA repair genes across 166 GBM samples. In accordance with the mouse expression data, Akt3 expression was significantly associated with a panel of 76 DNA repair genes (false discovery rate < 0.05 and |partial correlation| > 0.2) (Fig. 3F and Table S2). This pattern was not observed with either Akt1 or Akt2. These data highlight that the mouse model is a relevant and accurate representation of the human disease, and further, that Akt3 activation may contribute to therapeutic resistance (37).

To further investigate the potential role of Akt3 in DNA damage and repair process, we generated U87 stable cell lines that overexpress Akt1, Akt2, or Akt3 (Fig. S3A). We first examined the cells for evidence of DNA damage response with γ-irradiation by assessing micronuclei (MN), a well defined readout of damaged nuclei. Strikingly, Akt3-expressing cells had the highest level of MN among them (Fig. 4A). We further characterized the role of Akt3 in DNA damage response by evaluating a key marker of DNA double strand break (DSB), p-γ-H2AX. We observed a significant increase in p-γ-H2AX foci following γ-irradiation only in Akt3 cells, compared with empty vector (Fig. 4B) (P = 0.01). We observed that overexpression of Akt3 following irradiation led to a strong phosphorylation of Ataxia telangiectasia mutated (ATM) (Fig. S3B), a major first-responder of DNA damage that mediates activation of a plethora of transducers and effectors of the DDR. It should be noted that the functional classification of 1,077 putative ATM substrates is very similar to the pathways that were enriched in Akt3-derived tumors, including chromatin organization, stress response, DNA metabolism, and responses to DNA damage (38), providing more evidence that ATM is a key player in Akt3-mediated pathways.

Next we performed experiments that directly tested the ability of Akt3 to promote nonhomologous end joining (NHEJ) and homologous recombination (HR), two major pathways for DSB repair. To assess NHEJ, we performed a cell-free assay in which linearized plasmid was incubated with nuclear extracts of U87 cells stably expressing Akt isoforms. The relative amount of religated plasmid was monitored with semiquantitative PCR. Consistent with our hypothesis, Akt3 nuclear extracts had significantly higher levels of religated plasmid, indicating Akt3 promotes NHEJ (Fig. 4C). HR was monitored using a well characterized cell-based repair system, where the Dr-GFP cassette, which contains two nonfunctional GFP genes and SceI sequences were stably transfected into U87 cells stably expressing myristoylated Akt1, Akt2, or Akt3. Upon DNA double strand break via the SceI endonuclease, cells that undergo HR produce a functional GFP allele and may be monitored via flow cytometry. The stable U87 cells were transfected with SceI and collected 48 h following transfection for flow cytometry assay. Akt3 overexpression led to a significant increase in the population of GFP cells, validating its role as a mediator of HR (Fig. 4D). These results were further corroborated in HeLa cells (Fig. S3C). Furthermore, we showed that Akt3 rendered cells resistant to irradiation and temozolomide (TMZ) treatments via cell viability assays (Fig. 4 E and F). To assess the potential level of DNA repair in human GBM, we analyzed the number of mutations according to Akt3 copy number in TCGA samples, and found that the Akt3 high group had significantly fewer mutations compared with the Akt3 low group (Fig. 4G), indicating increased DNA repair and potential therapeutic resistance. The proposed model mechanism by which Akt3 promotes DNA repair and glioma progression is depicted in Fig. S4.

GBMs are notoriously resistant to therapies and are among the most difficult cancers to treat (39). Identifying and blocking key pathways which mediate resistance could positively impact the successful treatment of this disease. We have demonstrated that Akt3 strongly promotes glioma progression and mediates activation of DNA repair pathways. We believe that this work has potential to be highly impactful, because Akt3 is operative in multiple cancer types and may be a key factor in therapeutic resistance.

Copy number data from various cancer types studied by TCGA were downloaded from Synapse (https://synapse.prod.sagebase.org) on March 26, 2013. Only cancer types with more than 200 samples were considered. The data were produced by the Affymetrix Genome-Wide Human SNP Array 6.0 platform (Affymetrix) and recurrent copy number changes were identified by the GISTIC software (40) as described (4). The copy number amplification and deletion thresholds were set to be 1 and −1, respectively.

Microarray experiments were carried out using the Mouse Whole Genome Oligo Microarray kit from Agilent Technologies, following the manufacturer’s protocol (the detailed protocol can be found on the Agilent website: www.agilent.com). In brief, 500 ng of total RNA from each sample was used and labeled with Cy3- or Cy5-CTP. After 17 h of hybridization at 65 °C, the arrays were washed and scanned with Agilent’s dual-laser-based scanner. Feature extraction software GE2-v4_91 was used to link a feature to a design file and determine the relative fluorescence intensity between the samples. The raw data were analyzed using R 2.15.2 with the Limma package (3). Background correction, within- and between-array normalization were performed with parameters: bc.method = “normexp”, method = “loess”, and method = “Aquantile”, respectively. The normalized data were then log2 transformed. Selection of differentially expressed probes between conditions was carried out by the lmFit and eBayes functions included in the Limma package (41), which perform a linear model analysis and an empirical Bayes estimation of the variance, taking into account the small sample size of microarray experiments. P values were adjusted to obtain the q value, i.e., False Discovery Rate (FDR), by the Benjamini–Hochberg multiple testing correction (42). FDR is controlled at a level of 0.1. The probes were then mapped to genes (when multiple probes mapped to the same gene, the probe with the best q value was selected to represent the gene). Heatmap plots were generated using the R software package to visualize the normalized gene expression across samples. Genes were grouped by hierarchical clustering.

Differentially expressed genes from each experimental condition were used for the pathway analysis. We performed gene ontology (GO) enrichment analyses using the WebGestalt software (36) to identify the top enriched GO categories, which were organized as Directed Acyclic Graphs (DAG, which is a tree-like structure). The most broad categories at the top of the DAG are biology process, molecular function and cellular component, respectively. Going down the DAG, the categories become more specific. The top 10 enriched categories of each tree were highlighted in red (if q value < 0.05) or maroon (otherwise).

DNA repair genes were downloaded from the Gene Ontology (GO) category “DNA repair” (GO:0006281), including 356 human DNA repair genes. RNA sequencing data (Illumina HiSEq. 2000, Illumina) of 166 TCGA GBM samples was downloaded from the Broad Institute Firehose data standardization run on March 9, 2013. Partial correlation measures the degree of association between two variables, with influences from a third set of variables removed. To specify, the partial correlation between X and Y given a third set of variables Z is the correlation between the residuals obtained from the linear regression of X on Z and of Y on Z, respectively. We calculated the partial correlation between the expression of an Akt isoform and a DNA repair gene, with effects from the other Akt isoforms removed. The expression of 76 DNA repair genes were significantly associated with Akt3, but not with Akt1 or Akt2 (Benjamini–Hochberg adjusted false discovery rate < 0.05 and |partial correlation| >0.2).

The mutation data (Illumina Genome Analyzer DNA Sequencing, Illumina) of 291 TCGA GBM samples was downloaded from the Broad Institute Firehose data standardization run on March 9, 2013. Total number of mutations was calculated for each sample. The samples are then divided into two groups according to Akt3′s copy number. Specifically, the Akt3 low (high) copy number group consists of samples which are two MAD (Median Absolute Deviation) below (above) the median copy number value of Akt3. One-tailed t test is used to test the hypothesis that Akt3 high copy number group has lower mutation counts than the other group.

Differences in HGG and tumor proliferation index were analyzed with the one-sided Fisher’s exact test; H2AX foci and HR assay were analyzed via the one-sided t test with multiple testing correction. Please refer to SI Methods for detailed descriptions of bioinformatics analyses.

All animal experiments were performed in accordance with The University of Texas MD Anderson Cancer Center Institutional Animal Care and Use Committee guidelines.