AXL-TBK1 driven AKT3 activation promotes metastasis

AXL activation induces TBK1-dependent AKT3 activation

For these studies, we used PDA cells where AXL activation has been shown to promote tumor cell migration and invasion (22). We and others have shown that AKT is activated downstream of AXL (14). To determine which AKT isoform is most associated with AXL-driven epithelial plasticity, we first evaluated the expression of AXL, AKT isoforms, and EMT markers in publicly available GEO RNA sequencing data of PDA cell lines (Fig. 1A). This analysis revealed a stronger correlation between AXL and AKT3 compared with the other AKT isoforms. This was corroborated by the results of gene expression profiling interactive analysis (GEPIA) for AKT isoforms and AXL in human PDA using the TCGA database (Fig. 1B) (23), where AKT3 and AXL expression showed the best correlation. Considering that increased AXL expression has been shown in other types of cancer, we examined expression profiles of AXL, EMT markers, and AKT isoforms in breast cancer. Up-regulated AXL expression correlated with expression of EMT markers N-cadherin and Twist2 by immunohistochemistry (IHC) in primary breast tumors (fig. S1A). Similar to PDA cells, AXL expression strongly correlated with AKT3, but not AKT1 or AKT2, expression in breast cancer cell lines and breast tumors (fig. S1, B and C). Moreover, in MCF10a breast epithelial cells, forced expression of Slug increased AKT3 expression and pAKT in an AXL-dependent manner with no effect on AKT1 expression (fig. S1D). These data suggest that AXL expression is essential for accumulation and activation of AKT3.

Next, we tested the accumulation and phosphorylation of AKT isoforms downstream of AXL in human PDA cells (Panc1). Because Panc1 cells naturally express Gas6 and Gas6 activity is vitamin K dependent, the cells were pretreated with vitamin K inhibitor warfarin before stimulation (9). We evaluated phosphorylation of AXL [by enzyme-linked immunosorbent assay (ELISA) or immunoblotting] and TBK1 (by immunoblotting) (Fig. 1C and fig. S2A). Gas6 treatment increased accumulation of pAXL and pTBK1. Control samples were incubated with an AXL inhibitor (AXLi, BGB324/bemcentinib) or a TBK1-specific inhibitor (TBK1i, GSK8612) in combination with Gas6, which reduced accumulation of pAXL or pTBK1, respectively, to unstimulated (Control/UN) levels (Fig. 1C and fig. S2A). To test the phosphorylation of each AKT isoform, the samples were immunoprecipitated with AKT isoform–specific antibodies (Abs) and subsequently immunoblotted for pAKT. The results showed a 2.7-fold increase in phosphorylation of AKT3 with a molecular weight of ~90 kDa (Fig. 1C, “IP,” right) and a modest increase in the phosphorylation of the predicted size AKT3 band (~60 kDa; fig. S2A) in Gas6-stimulated cells. Unstimulated, AXLi-, and TBK1i-treated cells had similar basal levels of pAKT3. To determine the potential posttranslational modification of the ~90-kDa AKT3 band and confirm its identity, we tested AKT3 sumoylation status in Panc1 cells. Previous studies have shown the importance of AKT sumoylation for cell proliferation and tumorigenesis (24) and identified a SUMO-consensus site (K272) on AKT3 (25). AKT3 was immunoprecipitated with three AKT3-specific Abs and blotted with Abs specific for SUMO1 or SUMO2/3. Figure 1D shows ~90-kDa sumoylated AKT3 bands in each reaction, suggesting that the 90-kDa pAKT3 band detected in Fig. 1C is likely SUMO-modified AKT3. For pAKT1 and pAKT2, only bands of predicted molecular weight (~60 kDa) were detected (fig. S2A). AKT1 showed a less strong AXL-induced response but was not inhibited by TBK1i, suggesting that AKT1 is being activated by unidentified components of AXL signaling (fig. S2A). Basal levels of pAKT2 were elevated by Gas6 stimulation but were unaffected by AXLi (fig. S2A). We also tested the phosphorylation status of GSK3β, a known AKT substrate, but did not see any changes in its phosphorylation or total levels in response to Gas6 stimulation (fig. S2A).

To confirm the contribution of TBK1 to AXL-induced phosphorylation of AKT3, we used cells from a mouse genetic model (KIC Tbk1^Δ/Δ) that harbors a deletion of TBK1 kinase domain, which affects TBK1 activity and reduces its accumulation to undetectable levels (26). Kras^LSL-G12D; Cdkn2a^lox/lox; Ptf1a^Cre/+ (KIC) mice develop PDA that presents as low-grade ductal lesions at 3 weeks of age and progresses to pancreatic adenocarcinoma by 7 to 8 weeks of age. Stimulation with Gas6 induced pAXL in Tbk1^+/+ and Tbk1^Δ/Δ cell lines (fig. S2B). TBK1 phosphorylation was detectable only in KIC Tbk1^+/+ cells and increased with Gas6 stimulation. AXLi efficiently inhibited phosphorylation of AXL in each cell line and TBK1 phosphorylation in Tbk1^+/+ cells. Similar to Panc1 cells, AKT3 phosphorylation was up-regulated in an AXL-dependent manner by Gas6 stimulation (2.2-fold) in Tbk1^+/+ but not in Tbk1^Δ/Δ cells (fig. S2B). AKT1 phosphorylation was induced by Gas6 in Tbk1^+/+ and Tbk1^Δ/Δ cells. However, AKT2 phosphorylation levels dropped below unstimulated levels in Gas6-treated Tbk1^+/+ cells and were restored in the presence of AXLi (fig. S2B). In Tbk1^Δ/Δ cells, pAKT2 levels were nearly undetectable and did not change with Gas6 or Gas6/AXLi treatment (fig. S2B). In addition, no changes in total or phosphorylated GSK3β levels were detected in Gas6-treated cells in the presence or absence of AXLi and TBK1i in Tbk1^+/+ cells. In Tbk1^Δ/Δ cells, GSK3β phosphorylation was minimal and unchanged by stimulation with Gas6. Together, these results suggest that AKT isoforms respond differently to AXL stimulation and therefore may have distinct contributions to AXL-induced EMT. Only AKT3 showed increased phosphorylation and activation by AXL in a TBK1-dependent manner. AKT1 activity was modestly induced by AXL independent of TBK1, whereas Gas6-induced AKT2 phosphorylation was cell line dependent given that it was elevated in an AXL-independent manner in Panc1 cells but reduced by AXL activation in KIC cells.

AXL-induced AKT3 activation promotes EMT

Considering that AXL-induced stimulation of TBK1 is necessary for AXL-driven EMT in PDA (14) and AXL stimulation leads to TBK1-dependent AKT3 activation, we examined whether AKT3 is required for initiation of EMT. For this, AKT3 CRISPR-Cas9 KO was generated in a primary PDA cell line derived from Kras^{LSL − G12D/+}Trp53^fl/flPdx1^Cre/+ (KPfC) mice. Next, we tested the expression of EMT markers in empty vector control (Cas9-EV) and AKT3 KO A and AKT3 KO B cells (Fig. 1E). Minor variations in protein expression were detectable between KO A and B, likely because of heterogeneity of the wild-type cells (WT) (Westermann et al., 2022). However, compared with Cas9-EV both KO lines showed reduced expression of mesenchymal markers vimentin and Slug and increased expression of epithelial marker E-cadherin, confirming that AKT3 is necessary for the expression of EMT markers in KPfC cells in vitro (Fig. 1E).

To test whether AKT3 phosphorylation induces EMT markers in a TBK1-dependent manner, we analyzed expression of EMT markers in KIC Tbk1^+/+ and Tbk1^Δ/Δ cells. Basal levels of pAKT3 were sufficient to induce the expression of vimentin and Slug in Tbk1^+/+ cells. In Tbk1^Δ/Δ cells, vimentin and Slug expression were reduced, whereas E-cadherin levels increased dramatically compared with KIC Tbk1^+/+ cells. To determine the importance of TBK1, we transformed KIC Tbk1^Δ/Δ cells with a construct expressing constitutively active myristoylated AKT3 (27), which was sufficient to reverse the expression of EMT markers to the levels detected in Tbk1^+/+ cells (Fig. 1F). In parallel, we tested these cell lines in a three-dimensional (3D) invasion assay. Unlike Tbk1^+/+ cells, Tbk1^Δ/Δ cells formed tight spheroids and did not show any invasiveness (Fig. 1G). In correlation with the EMT expression data, myrAKT3 expression in Tbk1^Δ/Δ cells reversed the phenotype to a more mesenchymal-like one displayed by KIC Tbk1^+/+ cells (Fig. 1G). These results demonstrate that TBK1-dependent AKT3 activation contributes to AXL-mediated EMT.

TBK1 interacts with and phosphorylates AKT3

TBK1 can directly activate AKT (15, 16). We hypothesized that TBK1 directly binds to and activates AKT3 to drive EMT downstream of AXL. First, we examined localization of TBK1 and AKT3 in untreated Panc1 cells and determined that there is a considerable overlap in TBK1- and AKT3-specific signals suggesting that the proteins colocalize mostly in the cytoplasm but also in the nucleus (Fig. 2A). Similarly, in KIC Tbk1^+/+ cells, TBK1 and AKT3 colocalized in the cytoplasm and the nuclei of unstimulated and Gas6-stimulated cells (fig. S2, C and D). Next, we tested TBK1 and AKT3 interactions in a series of pull-down assays. Using human embryonic kidney (HEK) 293T and Panc1 cells transfected with epitope-tagged TBK1 or AKT3, we demonstrated that exogenous epitope-tagged TBK1 and AKT3 coimmunoprecipitate with endogenous AKT3 and TBK1, respectively (Fig. 2, B and C). Second, we immunoprecipitated endogenous TBK1:AKT3 complexes from KIC, KPfC, and Panc1 cells using AKT3- and TBK1-specific Abs (Fig. 2D). These results demonstrate that TBK1 interacts with AKT3.

To investigate whether TBK1 can directly bind to and activate AKT3, we performed an in vitro kinase activity assay with human recombinant TBK1 and AKT3 using cold adenosine triphosphate (ATP). Kinase reactions were separated via SDS–polyacrylamide gel electrophoresis (SDS-PAGE) and analyzed by mass spectrometry (MS), which showed phosphorylation of AKT3 at T144, T193, T298, T302/309, and S472 (Fig. 2E and fig. S2D). The latter two sites correspond to most commonly detected phosphorylation sites on AKT isoforms, AKT1 T308/AKT2 T309 and AKT1 S473/AKT2 S474 (28–31), as shown by the amino acid alignment (fig. S2D). Phosphorylation at both sites is necessary for maximal activation of AKT [reviewed in (32)]. These results confirm that AKT3 is a direct substrate of TBK1. Given that these proteins colocalize and coimmunoprecipitate and AKT3 phosphorylation upon AXL-stimulation is TBK1 dependent, these data strongly suggest that AKT3 is phosphorylated by TBK1.

mTORC1 is important for AXL-induced AKT3 phosphorylation

AKT was previously shown to be activated by mammalian target of rapamycin complex 2 (mTORC2) in a positive-feedback loop (33, 34). To determine mTOR’s contribution to AXL/TBK1-induced AKT3 phosphorylation, KIC TBK1^+/+ cells were pretreated with warfarin and inhibitors of mTORC1 (mTORC1i, rapamycin) or mTORC2 (mTORC2i, JR-AB2–011) and stimulated with Gas6. mTor inhibition did not affect AXL, TBK1, or AKT3 accumulation or AXL phosphorylation induced by Gas6 (fig. S2E). However, mTORC1i, but not mTORC2i, reduced total pTBK1 and pAKT3 (71 and 69%, respectively) compared with “No Drug” control in Gas6-treated cells (fig. S2E) suggesting that mTORC1 is involved in AXL/TBK1-induced AKT3 activation. Similar to mTORC2, mTORC1 is an AKT substrate, and it might activate AKT3 isoform in a positive-feedback loop. In addition, mTORC1 was shown to be a TBK1 target (35). Further, TBK1 coimmunoprecipitated with mTOR in No Drug and mTORC2i-treated cells. However, mTORC1i inhibited mTOR:TBK1 interactions (fig. S2E). Therefore, it is possible that mTORC1 is stimulated by AXL/TBK1 signaling and then phosphorylates AKT3.

AXL-TBK1 is required for accumulation of pAKT3 and Snail in the nucleus

To quantify accumulation levels of total and pAKT3 in the nucleus and the cytoplasm, Panc1 cells were pretreated with warfarin and stimulated with Gas6 in the presence or absence of AXLi or TBK1i. Lysates were fractionated and analyzed by immunoblotting. Tubulin and lamin A/C were used as controls for cytoplasmic and nuclear fractionation, respectively (Fig. 3A). TBK1 and AKT3 were found in the nuclear and cytoplasmic fractions. AKT3 was immunoprecipitated from the fractionated lysates and blotted for pAKT. Gas6-stimulation increased phosphorylation of AKT3 in the cytoplasm and the nucleus (Fig. 3A). However, pAKT:total AKT3 ratio was twofold higher in the nucleus compared with the cytoplasm. AXLi and TBK1i reduced accumulation of total AKT3 in cytoplasmic fraction and pAKT3 in both fractions. This demonstrates that (i) AXL-induced activation of AKT3 in the nucleus is TBK1 dependent and (ii) total nuclear AKT3 levels do not change upon AXL stimulation or treatment with AXLi and TBK1i. This suggests that, in Panc1 cells, AXL/TBK1 signaling does not induce additional AKT3 translocation to the nucleus. However, AKT3 has a nuclear export signal (NES) (36) and may shuttle out of the nucleus. Therefore, future experiments characterizing subcellular localization of AKT3 ΔNES mutant in Gas6-stimulated cells are necessary.

Because we have previously shown that AXL-driven TBK1 activation increases expression of EMT inducing TFs (EMT-TFs) Slug and Snail (14), we evaluated their expression in fractionated lysates. We also found that these EMT-TFs displayed different localization patterns. Slug was detected only in the cytoplasm, and its expression did not seem to respond to any treatment. Snail mainly accumulated in the nucleus, and its nuclear levels were increased 2.4-fold by Gas6 stimulation (Fig. 3A). Consistent with our prior data, AXLi and TBK1i reduced Snail levels but only in the nucleus.

To clarify the contribution of TBK1 to AKT3 nuclear localization, we compared AKT3 protein levels in fractionated KIC Tbk1^+/+ and Tbk1^Δ/Δ cells: Gas6 increased nuclear AKT3 in each cell line, but, unfortunately, nuclear AKT3 levels were too low to detect phosphorylation by immunoblotting. These results suggested that levels of nuclear AKT3 may be a cell line–specific feature, and in KIC cells, this process is TBK1 independent (fig. S3, A and B). In contrast, in MDA-MB-231 breast cancer cells and MCF10a normal breast epithelial cells, nuclear localization of AKT3 is readily detectable in all cells by immunofluorescence and is inhibited by AXL knockdown (KD) with AXL short hairpin RNA (shRNA) (fig. S4, A and B), indicating that in these cells, AKT3 is transported to the nucleus during AXL signaling.

Next, we tested Snail accumulation in fractionated KIC Tbk1^+/+ and Tbk1^Δ/Δ cell lysates. Similarly to Panc1 cells, AXL stimulation increased nuclear Snail ~1.3-fold in Tbk1^+/+ cells (fig. S3A). In Gas6-treated Tbk1^Δ/Δ cells, however, nuclear Snail accumulation did not change, whereas cytoplasmic Snail dropped ~12-fold compared with the unstimulated control (fig. S3B). These results support that TBK1 participates in Snail accumulation and nuclear localization during AXL signaling.

Nuclear localization of AKT3 is essential for cell invasion

AKT3 is found in the nucleus (37), but the mechanisms of targeting AKT3 to the nucleus are unclear. Proteins more than 40 kDa must be actively transported through the nuclear membrane by importins that recognize and bind a nuclear localization signal (NLS) (38). We used a web-based NLS mapper (39) and identified a weak bipartite NLS in the AKT3 amino acid sequence (accession number: Q9Y243) located in a flexible linker region between the PH domain and kinase domain (fig. S4C). On the basis of these in silico findings, we created two AKT3 mutant expression constructs: AKT3-NLS1 and AKT3-NLS2. AKT3-NLS1 carries two point mutations (K141R and R142A) that alter the leucine-rich NLS region to the sequence that resembles the linker area in AKT2 (fig. S4C). For AKT3-NLS2, a 10–amino acid sequence flanking the NLS was replaced to mimic a longer part of the linker region as coded in AKT2. WT AKT3 and the mutants were stably expressed in human breast HMLER cells and analyzed by fluorescence microscopy. Parental HMLER cells do not express endogenous AKT3 (40); therefore, no AKT3 signal was detectable in control HMLER cells stably expressing green fluorescent protein (HMLER:GFP; fig. S4D). WT AKT3 was predominantly nuclear (87%); however, AKT3-NLS1 and AKT3-NLS2 mutants were largely restricted to the cytoplasm with 18 and 29% nuclear localization, respectively (fig. S4E).

Next, to test whether AKT3 nuclear localization is important for EMT, we transduced pancreatic KPfC AKT3 KO cell clones A and B (KO A and KO B) with WT AKT3 or AKT3-NLS1 expression constructs and evaluated 3D invasion. Analysis of tumor cell morphology and invasion revealed that whereas AKT3 KO resulted in a reduced invasive phenotype compared with control, rescuing AKT3 expression with WT AKT3 restored the invasive phenotype (Fig. 3, B and C). However, AKT3-NLS expression in AKT3 KO cells did not restore invasiveness, suggesting that nuclear AKT3 localization is necessary to drive EMT.

AKT3 interacts with Snail in the nucleus

To validate cell fractionation data, we analyzed Snail expression and localization by confocal microscopy. In unstimulated Panc1 cells (UN) treated with warfarin to “silence” AXL stimulation, Snail expression was barely detectable. However, Gas6-stimulation upregulated nuclear levels of Snail, whereas its cytoplasmic levels remained the same (Fig. 4A). Similar to cell fractionation data, AXLi treatment reduced nuclear Snail levels to background levels.

Knowing that pAKT3 and Snail accumulate in the nucleus upon AXL activation, we assessed nuclear localization of AKT3 and Snail in Panc1 cells by confocal microscopy. Imaging (Fig. 4B) indicates that AKT3 and Snail overlap in the nucleus, suggesting that the proteins interact. Interaction of AKT3 and Snail was confirmed by co-immunoprecipitation. Snail was detected in the AKT3 IP from whole-cell lysate and in the nuclear fraction of untreated Panc1 cells. Snail was detected at the predicted molecular weight and at a higher molecular weight likely indicating that Snail is posttranslationally modified (Fig. 4C). Snail was detectable in the cytoplasmic fraction of the lysate, but it was not pulled down with AKT3. This suggests that AKT3:Snail interactions occur in the nucleus but not in the cytoplasm. Because AKT3 colocalizes and interacts with TBK1, we probed AKT3 pull downs for TBK1. TBK1 coprecipitated with AKT3 from both fractions; however, most of the TBK1 signal was detected in the nuclear fraction (Fig. 4C). Similar to Snail, we detected bands corresponding to the predicted TBK1 molecular weight and slower migrating TBK1-specific bands indicative of its potential posttranslational modifications. This suggests that AKT3 interacts with TBK1 and Snail and that all three proteins might function as part of the multiprotein signaling complex located in the nucleus.

AXL activation increases Snail half-life in an AKT3-dependent manner

To investigate the mechanism regulating Snail accumulation in AXL-stimulated cells, we knocked down AKT3 expression with shRNA (Fig. 4D) and tested Snail protein stability by cycloheximide (CHX) chase assay (Fig. 4E). Panc1 cells were treated with CHX in the presence or absence of Gas6 over a time course of 8 hours. Consistent with previous findings (41), Snail had a half-life of ~2 hours and was difficult to detect by 8 hours of chase when treated with CHX (Fig. 4E). However, AXL stimulation with Gas6 extended Snail half-life. Further, we found that Snail expression was nearly undetectable in shAKT3 Panc1 cells, and Gas6 treatment no longer had a stabilizing effect on Snail (Fig. 4E). These results demonstrate that AXL-induced accumulation of Snail in the nucleus is likely due to AKT3-dependent stabilization of the protein.

Selective targeting of AKT3 with an allosteric small-molecule inhibitor

Several drugs targeting pan-AKT activity (e.g., GDC0068, AXD5363, and MK-2206) are currently in various stages of clinical testing. However, many of these trials report toxicity, such as hyperglycemia and hyperinsulinemia, because of the essential functions of AKT1 and AKT2 in tissue homeostasis (42–45). An AKT3-selective inhibitor has the potential to overcome these issues. The similarity between AKT1, AKT2, and AKT3 in the kinase domain precludes selective kinase inhibition. However, an allosteric site located in a cleft between the PH domain and the kinase domain has been used to identify AKT1-, AKT2-, and AKT1/2-selective inhibitors (46–48). Sequence alignment around this allosteric site suggested that there are exploitable differences in this region (Fig. 5A). A structural model produced by comparison of the crystal structures of AKT1 (48) and the AKT2 kinase domain suggested that a single amino acid deletion in AKT2 and AKT3 compared with AKT1 leads to a change in the path that the protein backbone follows, opening a pocket at the front of the allosteric binding site (Fig. 5B). This pocket is small in the case of AKT2 because of the protrusion of the large side chain of Arg²⁶⁹, but the pocket is larger in AKT3 because of the presence of a glycine at this site. A series of allosteric small-molecule inhibitors of AKT3 were developed (BerGenBio ASA, patent number WO/2016/102672) with backbones that bind to the allosteric site on AKT3 specifically. One of these compounds, a potent and selective allosteric AKT3 inhibitor, BGB214 (N- (5- (4- (1-aminocyclobutyl)phenyl)- 4-pheny lpyridin-2-yl)-2-((1r,4r)-4-(N-methylacetamido) cyclohexyl) acetamide), referred to as AKT3i, was chosen for our studies (Fig. 5C). In biochemical assays using purified tag-free enzymes, AKT3i had a median inhibitory concentration (IC₅₀) of 13 nM for AKT3 with approximately 1000-fold selectivity against AKT1 and >35-fold selectivity against AKT2 (Fig. 5D).

First, AKT3i efficacy was evaluated in cell viability assays. The calculated AKT3i IC₅₀ for Panc1 cells (Fig. 5E) and MDA-MB-231 (fig. S5A) cells were 13.5 and 9.3 μM, respectively. AKT3i specificity was validated using HMLER cells transduced with vectors expressing GFP (control) or AKT3. Treatment of HMLER cells (GFP and AKT3) with increasing concentrations of AKT3i did not affect the accumulation or phosphorylation of AKT1 or AKT2 (fig. S5B). Because HMLER-GFP cells do not express endogenous AKT3, AKT3 signal was undetectable. In HMLER-AKT3 cells, AKT3i inhibited AKT3 phosphorylation at 0.3 μM concentration but did not affect total levels of AKT3 (fig. S5B), highlighting its specificity. AKT3i efficacy was validated in a 3D cell migration assay. MDA-MB-231 cells were plated in collagen/Matrigel and treated with Gas6 in the presence or absence of 3 μM AXLi or 2 μM AKT3i for 48 hours. Invasion more than 50 μm was determined and quantified, revealing that inhibition of AXL or AKT3 substantially reduced cell migration/invasion (fig. S5, C and D). However, cell viability was not noticeably affected by the drugs used at indicated concentrations (fig. S5E).

Last, we used AKT3i to confirm AKT3’s contribution to Snail upregulation in Gas6-stimulated Panc1 cells. AKT3 expression and phosphorylation and Snail expression levels were up-regulated in Gas6-treated cells compared with untreated (UN) cell levels (Fig. 5F). AXLi and TBK1i, used for control, reduced levels of pAKT and Snail. Strikingly, AKT3i treatment reduced AXL activity and potently reduced AKT3, pAKT3 and Snail levels. AKT3i also decreased phosphorylation of GSK3β. On this basis, we conclude that selective inhibition of AKT3 with the allosteric inhibitor (AKT3i) blocks AKT3-mediated up-regulation of Snail and inhibits migration/invasion properties of the cells.

AKT3 expression is associated with poorly differentiated tumors and increased metastasis

To assess the biological consequence of AKT3 activity, control KPfC cells (CAS9-EV), AKT3 KO KPfC cells (AKT3 KO), or AKT3 KO KPfC cells transduced with AKT3 (rescue) were injected orthotopically into the pancreases of C57BL/6J mice (Fig. 6, A to H). Primary tumor and metastatic burden was evaluated 19 days after injection. Although primary tumor weight did not differ between the three groups (Fig. 6B), gross metastatic burden was reduced in tumors lacking AKT3, an effect rescued by AKT3 reexpression (Fig. 6A). Hematoxylin and eosin (H&E) analysis as well as IHC for CK19, a PDA tumor cell marker (49), confirmed reduced metastasis to livers of AKT3 KO tumor–bearing mice (Fig. 6, C and D). Consistent with this observation, tumors lacking AKT3 were more differentiated and expressed higher levels of E-cadherin and lower levels of vimentin (Fig. 6, E to H). The expression of E-cadherin and vimentin in vivo was consistent with the expression of these proteins in vitro (Fig. 6F). In addition, we observed that the expression level of AKT3 correlated with the number of gross metastases (Fig. 6, A and F).

Nuclear AKT3 is associated with aggressive cancer and worse survival in patients

To investigate the importance of nuclear AKT3 in cancer development, we assessed AKT3 localization in pancreatic and breast tumors from patients by IHC. In AXL-positive tumors, nuclear AKT3 was detected in single tumor cells outside epithelial ducts (Fig. 7A and quantification in Fig. 7B and fig. S6A). However, in AXL-negative tumors, AKT3 was cytoplasmic, supporting our findings that (i) AXL activity induces nuclear accumulation of AKT3 and (ii) nuclear AKT3 results in a less differentiated (more mesenchymal-like) tumor cell phenotype.

Next, to assess the influence of AKT3 signaling and its downstream targets on patient survival, genes that were found to be differentially expressed after AKT3 overexpression in MCF10A cells were used to generate an AKT3 score (table S1), which was then mapped against probes in the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) database. The results showed that high AKT3 expression is associated with an EMT signature and worse survival outcome (fig. S6B). Further, we found a prominently different distribution between breast cancer subtypes (based on PAM50 intrinsic subtypes) and AKT3 scores (fig. S6C). In estrogen receptor (ER)–negative tumors and basal-like tumors, high levels of AKT3-associated gene expression correlates with more aggressive forms of breast cancer, worse overall outcome, and a higher hazard ratio. This is consistent with previous reports showing high AKT3 copy number alterations in patients with triple negative breast cancer (TNBC) (50, 51) and a correlation between AKT3 expression levels and higher grade breast tumors (52). To assess the effect of nuclear AKT3 localization on patient survival, we correlated IHC data for AKT3 in clinical breast cancer samples (fig. S6A) with the clinical outcomes. The results revealed that nuclear AKT3 predicted a worse overall outcome in this cohort of patients with breast cancer (fig. S6D).

Last, to validate our in vitro data showing nuclear AKT3:Snail interactions, we examined localization of these proteins in human PDA tissues using IHC. Nuclear AKT3 and Snail expression strongly correlated and completely overlapped in all tested samples (Fig. 7C).

Together, these results confirm that AXL-mediated signaling induces nuclear accumulation of pAKT3 and Snail, which promotes EMT and metastasis (Fig. 8). This suggests that (i) AKT3 may serve as a therapeutic target that avoids toxicity associated with pan-AKT inhibition and (ii) nuclear AKT3 can be used as a prognostic biomarker for predicting worse overall survival in aggressive cancers, including metastatic pancreas and breast tumors. Further investigation is needed to determine whether nuclear AKT3 is an important feature of additional types of aggressive cancer.

Drugs and Abs

The drugs used in this study are as follows: warfarin (S4545, Selleckchem), BGB324/bemcentinib (AXLi, BerGenBio), GSK8612 (TBK1i, Selleck Chemicals), BGB214 (AKT3i, BerGenBio), rapamycin (mTORC1i, MedChemExpress), JR-AB2–011 (mTORC2i, MedChemExpress), and CHX (01810, Sigma-Aldrich). All Abs and reagents used for protein staining are listed in table S2.

Plasmid constructs

pRP(Exp)-CMV-huGas6 was generated by VectorBuilder by inserting a human Gas6 coding sequence into a pRP(Exp)–cytomegalovirus (CMV) expression vector under a CMV promoter. pMFlag-TBK1 (pWZL Neo Myr Flag TBK1, Addgene plasmid # 20648; http://n2t.net/addgene:20648; RRID: Addgene_20648) and pMFlag-AKT3 (pWZL Neo Myr Flag AKT3, Addgene plasmid # 20423; http://n2t.net/addgene:20423; RRID: Addgene_20423) were gifts from W. Hahn and J. Zhao (86). pMHA-AKT3 was a gift from W. Sellers (1236 pcDNA3 Myr HA Akt3, Addgene plasmid # 9017; http://n2t.net/addgene:9017; RRID: Addgene_9017). pHA-AKT3 was created using a pMHA-AKT3 construct by removing the N-terminal myristoylation signal (Gly-2) with overlap extension polymerase chain reaction (PCR) (87) and primers set 1 (5′-CCTACTTGGCAGTACATCTACGTATTAGTCATC-3′; 5′-GGCTTGCTCTTGCTGCTCGCCATGGTGGTG-3′) and set 2 (5′-CACCACCATGGCGAGCAGCAAGAGCAAGCC-3′; 5′-CAGTGCAATTTCAGGTGCAATTGGGAATTC-3′). CRU5–internal ribosomal entry site (IRES)–GFP retroviral vectors for expression of hSLUG, AKT3, shLuc, and shAXL were prepared as described (11). CRU5-IRES-GFP retroviral vectors expressing AKT3-NLS and AKT3-NLS were generated by site-directed mutagenesis (Quick change, #2200519). Constructs expressing AKT3 targeting single-guide RNAs (sgRNAs) for CRISPR-Cas9 KO of AKT3 were generated using oligo sets “A” (5′-CACCGAATGGTAACATCGCTCATGA-3′; 5′-AAACTCATGAGCGATGTTACCATTC-3′) and “B” (5′-CACCGCCTCTGCAATCGGTCGGCTA-3′; 5′-AAACTAGCCGACCGATTGCAGAGGC-3′) that were annealed, phosphorylated, and cloned into the dual-expression vector PX458 at BbsI sites. All vectors were confirmed by DNA sequencing.

Cell culture, transfections, retroviral, and lentiviral transductions

Human cell lines PANC1 (CRL-1469), MDA-MB-231 (HTB-26), and MCF10A (CRL-10317) and murine 4T1 (CRL-2539) cells (American Type Culture Collection, Rockville, MD) were cultured as described (11). HMLE and HMLER (CVCL_DG85) cells (a gift from R. Weinberg) were maintained as previously described (88). TBK1 WT and deficient KIC and KPfC murine cell lines were established from single-cell clones isolated from minced and digested tumors. Cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum (FBS) (14). Retroviruses were harvested 48 hours after viral vector plasmid transfection of Phoenix A retroviral packaging cells (89). Lentiviruses were collected 48 hours after cotransfection of viral vector and lentiviral packaging plasmids in HEK293 T cells (14). Expi293F cells (A14527, Thermo Fisher Scientific), suspension-adapted HEK 293 cells, were maintained in Expi293 expression medium (A1435102, Thermo Fisher Scientific).

CRISPR KO

Plasmid constructs expressing AKT3 sgRNA or empty vector control were transfected into KPfC cells with Lipofectamine 2000 (11668027, Thermo Fisher Scientific). Positive cells expressing GFP were sorted as single clones and expanded. Each expanded clone was subjected to validation through PCR (AKT3 KO A: 5′-AAACCCTAAAACTGACCTGCAA-3′; 5′-AGGAAAGACCAACTCTCAGCAC-3′; AKT3 KOB: 5′-GACATTATTTGCATTCATCCCA-3′; 5′-GACGCATCCATCTCTTCTTCTC-3′) and immunoblotting.

Human Gas6 preparation and cell stimulation

Human Gas6 from conditioned media was prepared by transfecting Expi293F cells with pRP(Exp)–CMV-huGas6 using ExpiFectamine 293 transfection kit (A14525, Thermo Fisher Scientific). Gas6 expression and carboxylation were tested by immunoblotting. Gas6 concentration was measured by ELISA. For stimulation with Gas6, Panc1 cells were plated in 2 μM warfarin–1% FBS–DMEM, and KIC and KPfC cells were plated in 2 μM warfarin–10% FBS–DMEM 24 hours before stimulation. The cells were preincubated with indicated drugs (2 μM AXL, 10 μM TBK1i, or 10 μM AKT3i for 2 hours; 10 μM mTORC1i or 10 μM mTORC2i for 24 hours before stimulation) and then stimulated with Gas6 (200 ng/ml) in the presence or absence of the inhibitors for 30 min. After stimulation, the cells were washed with phosphate buffered saline (PBS), scraped, spun down, and either immediately lysed and fractionated or stored at −80°C.

Cell fractionation

Cells plated in 10-cm dishes were treated as described, washed with ice-cold PBS, and harvested. Cell fractionation was done as described elsewhere (90). Briefly, cells resuspended in 1 ml of ice-cold 0.1% NP-40–PBS–supplemented Halt protease and phosphatase inhibitor cocktail (PI78442, Thermo Fisher Scientific) and triturated five times with P1000 pipette and transferred to 1.5-ml microcentrifuge tube. An aliquot of the lysate was transferred to a fresh tube, resuspended in Laemmli sample buffer, and labeled as “whole cell lysate (WCL).” The remaining lysate was spun down at 9600g for 10 s. The supernatant was transferred to a fresh tube and labeled as “cytoplasmic fraction.” The pellet containing nuclei was briefly washed with 1 ml of ice-cold 0.1% NP-40–PBS–supplemented Halt protease and phosphatase inhibitor cocktail, spun down at 9600g for 10 s, resuspended in 1 ml of 0.1% NP-40–PBS–supplemented Halt protease and phosphatase inhibitor cocktail and Benzonase nuclease (E1014–5KU, MilliporeSigma), and labeled as “nuclear fraction.”

Immunoprecipitation

Cells plated in 10-cm dishes were treated as described and harvested. To determine phosphorylation status of AXL and AKT isoforms, the cells were lysed in 50 mM tris (pH7.4)–150 mM NaCl–5 mM EDTA–0.1% SDS–0.5% sodium deoxycholate–0.5% NP-40 (RIPA) buffer supplemented with Halt protease and phosphatase inhibitor cocktail and Benzonase nuclease. The lysates were immunoprecipitated with indicated Ab overnight at 4°C. To capture ~90-kDa pAKT3, the lysates were immunoprecipitated for 2 hours at 4°C. For pull-down TBK1:AKT3 complexes, the cells were lysed in RIPA buffer, and the lysates were subjected to immunoprecipitation overnight at 4°C. AKT3:Snail complexes were immunoprecipitated in 1% NP-40–PBS supplemented with Halt protease and phosphatase inhibitor cocktail and Benzonase nuclease. mTORC1/mTORC2 complexes were immunoprecipitated in 50 mM tris (pH7.4)–5 mM EDTA–0.3% CHAPS–150 mM NaCl (TECN) buffer supplemented with Halt protease and phosphatase inhibitor cocktail and Benzonase nuclease. The lysates (250 to 500 μg of protein) were normalized on the basis of total protein concentrations measured with bicinchoninic acid (BCA) protein assay kit (23225, Pierce). The samples were precleared by incubation with protein A/G plus agarose beads (Santa Cruz Biotechnology, sc-2003) and coimmunoprecipitated with indicated Ab and protein A/G plus agarose beads or Ab conjugated to agarose or Sepharose beads (table S2). The samples were washed with the lysis buffer and analyzed by immunoblotting with indicated Ab (table S2).

Immunoblotting

For protein accumulation assays, the cells were lysed in RIPA buffer supplemented with Benzonase nuclease and Halt protease and phosphatase inhibitor cocktail. Lysates were normalized on the basis of total protein concentrations measured with BCA protein assay kit. Otherwise, the samples were prepared as indicated. The samples were separated on 10% SDS-PAGE, transferred to a nitrocellulose membrane, blocked with 3% bovine serum albumin (BSA, A30075, Research Products International) diluted in 20 mM tris base–0.5 M NaCl–0.05% Tween 20 (TBST) buffer, and blotted with the indicated Ab diluted in 3% BSA–or 5% skim milk–TBST buffer.

Enzyme-linked immunosorbent assay

In vitro kinase assay

Recombinant GST-AKT3 (1.5 μg, BML-SE369–0005, Enzo Life Sciences) or His-AKT3 (NBP3–13999, Novus Biologicals), 0.1 μg of recombinant active human GST-TBK1 (T02–10G-05, SignalChem), and 1 μl of 10 mM ATP were combined in kinase reaction buffer [20 mM tris-HCl (pH 7.4), 500 mM β-glycerol phosphate, and 12 mM magnesium acetate] up to a total of 30 μl. The kinase reaction was carried out at 30°C, 500 rpm for 1 hour. After the reaction, AKT3 protein was resolved by SDS-PAGE and stained by Coomassie brilliant blue. AKT3 bands were cut out for MS analysis to identify phosphorylation site by the UT Southwestern Proteomics core. The purity of GST-TBK1 preparation was confirmed using MS analysis of complex mixtures.

Confocal microscopy

The cells were plated on coverslips or chambered coverglass and incubated overnight as described. The cells were fixed/permeabilized with 2% formalin or fixed with 4% formaldehyde and permeabilized with 0.2% Triton X-100, stained with indicated Abs, and imaged with Zeiss LSM780 or Zeiss LSM880 inverted microscopes.

Immunohistochemistry

3D culture experiments

Growth factor–reduced Matrigel (354230, Corning, 10 to 12 mg/ml stock concentration) and bovine (354231, Corning) or rat tail (354236, Corning) collagen I were used for organotypic culture experiments. Vertical invasion assays and experiments in 3D culture were performed and quantified as described previously (92) using a Matrigel/collagen I matrix (3 to 5 mg/ml Matrigel and 1.8 to 2.1 mg/ml collagen I). A 120-μm span on the z axis is shown for the vertical invasion assays.

Biochemical assays for AKT activity

Akt isoforms were activated by phosphorylation with PDK1 kinase in the presence of indicated concentrations of AKT3i. The output was measured using the GSK3α-derived LANCE Ultra Ulight–labeled crosstide substrate peptide (TRF0106-M, PerkenElmer) and europium-labeled LANCE Ultra Anti-Phospho-GSK3a (Ser²¹) Ab (TRF0202-M, PerkinElmer) according to manufacturer instructions. Plates were read with an EnVision multilabel plate reader, excitation at 320 nm and emission at 665 and 615 nm. Results were converted to percent inhibition of phosphorylation by normalizing to positive and negative controls, and compound IC₅₀ was determined using a three-parameter equation (Prism, GraphPad).

Cell viability assay

Panc1 or MDA-MB-231 cells were plated on day 0. On day 1, the medium was replaced with a fresh one containing AKT3i in fourfold dilutions. Eight different drug concentrations were tested with eight replicates per concentration. Relative cell number was determined by an MTS assay: 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymet hoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium was added, and the cells were incubated for 1 to 3 hours at 37°C. Drug sensitivity curves and IC₅₀s were calculated using in-house software. Response was validated in replicate plates (n ≥ 4). Assays were repeated in four biological replicates, with 16 technical replicates each in total.

Gene expression analysis and RNA sequencing

The expression analysis of the breast cancer cell lines and human samples (cancer, normal) was performed from published and GEO-submitted Affymetrix data as described (93). Total RNA from fluorescence-activated cell sorting–enriched primary culture cells were isolated with TRIzol (Invitrogen) and RNeasy Mini column (Qiagen) and evaluated using Bioanalyzer (Agilent Technologies). Gene expression levels were measured using the Illumina HumanHT-12 v4 Expression BeadChip whole-genome expression array. The Illumina Bead Array data were quality controlled in Genome Studio, and both probe level and gene level data were imported into JExpress Pro (http://jexpress.bioinfo.no) for analysis. After quantile normalization, both datasets were log₂ transformed. Correspondence analysis (94) was performed on the datasets, together with hierarchical clustering of the samples using a Pearson correlation measure on a per gene mean-centered version of the data. Differentially expressed genes between AXL⁺ and AXL⁻ groups were identified using the rank product method on both datasets (95). The resulting lists of differentially expressed genes with a false discovery rate value q = 10% from these two analyses were considered differentially expressed between the two groups. The cells were plated on 10-cm dishes until cell densities of 70% were achieved. Total RNA was extracted from cells using QIAGEN RNeasy Mini kit and stored at −80°C. One microgram of total RNA per sample was subjected to library generation using the TruSeq stranded total RNA sample preparation kit, according to the manufacturer’s protocol (Illumina). The libraries were pooled and sequenced on a NextSeq 500 instrument (high output flowcell) at 1 × 75–base pair single-end reads (Illumina). Raw RNA sequencing reads were aligned against to the human genome release GRCh38/hg38 using HISAT2 (96), and exons were counted using RSubread.featureCounts (97). Libraries were filtered to remove gene counts of less than 1 CPM across all libraries and normalized. Differentially expressed genes between GFP control group and AKT3 overexpressing MCF10A cells were calculated using edgeR (98, 99). Genes were considered differentially expressed with a fold change > 2 and P < 0.05.

AKT3 score and METABRIC dataset

To assess the influence of AKT3 signaling and its downstream targets on survival of patients with breast cancer, genes that were found to be differentially expressed after AKT3 overexpression in MCF10A cells were used to generate an AKT3 score. The score essentially represented the sum of expression of 42 differentially expressed genes, adjusted for expected directionality. Initially, we examined 46 different genes, but only 42 of them were represented with probes on the expression array. For genes represented by multiple probes (the 42 genes mapped to 71 different probes), mean signal intensity was used. The influence on breast cancer–specific survival and the putative difference between molecular subtypes was investigated in the Metabric cohort, composed of 1980 patients with breast cancer enrolled at five different hospitals in the UK and Canada (100). Gene expression was assessed using the Illumina HT-12 v3 microarray, and normalized data were downloaded from the European Genome-phenome Archive (EGA) data portal. Missing values were imputed using the impute.knn function as implemented in the R library “impute” with default settings (101). The data were batch adjusted for hospital effect using the pamr.batchadjust function in the “pamr” library with default settings (102). Association between the score and molecular subtypes (100, 103) was tested using Kruskal-Wallis rank test, and correlations were estimated with Spearman’s rank correlation. Survival analyses were performed using a Cox proportional hazards regression model as implemented in the R library “rms” (104). Survival plots were generated using the survplot function, as implemented in the rms library. All analyses were performed using R version 3.3.1.

Animal studies

Syngenic PDA cells (KPfC: CAS9-EV, AKT3 KO, Rescue) were injected orthotopically (2.5 × 10⁵ cells) in 6- to 8-week-old female C57BL/6 mice (stock # 000664, the Jackson Laboratory). Nineteen days after tumor cell injection, the mice were euthanized and organs were harvested for analysis. All animals were housed in a pathogen-free facility with 24-hour access to food and water. Animal experiments in this study were approved by and performed in accordance with the institutional animal care and use committee at the UTSW Medical Center at Dallas. Before implantation, the cells were confirmed to be pathogen free.

Statistical analysis

Comparisons of histological staining index (SI) groups were performed by Pearson χ² test using cut-off values for SI categories based on median values. Fisher’s exact tests were used to investigate the association between nuclear and cytoplasmic AKT3 and AXL in human PDA. Kruskal-Wallis tests were used to determine whether there was a significant difference in continuous outcomes among more than two groups. If a Kruskal-Wallis test showed a significant difference, Dunn’s multiple comparisons tests were conducted to examine which pairs of groups yielded significant differences. GraphPad Prism 5.0 for PC and MATLAB were used for statistical analysis.