FOXO Transcription Factors Both Suppress and Support Breast Cancer Progression

The PI3K–AKT/PKB pathway is commonly hyperactivated in cancer and negatively regulates nuclear localization and thereby the activity of Forkhead Box O family of transcription factors (FOXO1, FOXO3, FOXO4). When active, FOXOs control a plethora of target genes involved in redox homeostasis, cell proliferation, differentiation, metabolism, and apoptosis (1).

FOXOs are putative tumor suppressors as combined loss of Foxo1, Foxo3, and Foxo4 in adult mice results in increased hematopoietic lineage–specific tumors and FOXO activation arrests the cell cycle and induces apoptosis or anoikis in multiple types of tumor cells (2–5). Despite their tumor-suppressive roles, mutational loss of all FOXO alleles is statistically unlikely and has not been reported in cancer. Apparently, AKT/PKB–mediated inhibition of FOXO is either sufficient to avoid tumor suppression, or the maintenance of a certain level of FOXO activity is beneficial to tumor cells. It remains, however, to be established what level of activity and under what conditions FOXO activity then would be required.

It has been suggested that FOXOs may also contribute to tumorigenesis, and a more active mutant form of FOXO1 has been reported in cancer (3, 5–10). In addition, FOXOs have been implicated in a drug resistance signaling feedback loop that reestablishes RTK–PI3K–AKT signaling after pharmacologic inhibition of AKT/PKB (11–13). In line with these contrasting roles for FOXOs in cancer, studies aimed to use FOXO expression levels and localization as prognostic markers for cancer patient survival yielded diametrically opposing results (14).

The current understanding of the role of FOXOs in cancer is clearly incomplete, scattered over many different studies, in various types of cancer and limited to only the effects of FOXO loss or gain of function in isolated aspects of cancer (3). Studies designed to characterize the effects of both acute FOXO gain and loss of function within one cancer model covering the complete disease spectrum from tumor initiation to metastasis are lacking.

To clarify the role of FOXOs in cancer, we characterized inducible FOXO activation in a mouse model for metastatic invasive lobular carcinoma (mILC) that allows live monitoring of both primary tumor growth and metastasis (15, 16). Our study reveals that FOXOs both support and suppress metastatic breast cancer progression and hence are not classical tumor suppressors. FOXO activity is actually required for efficient proliferation and dissemination, which suggests that FOXO inhibitors might be of use as cancer therapeutics.

The luciferase-expressing mouse invasive lobular carcinoma cell line (mILC1) has been described before and was originally derived from tumors that developed in female K14cre;Cdh1^F/F;Trp53^F/F mice and cultured as described previously (15). The identity of these cells was regularly checked by luciferase expression and PCR for Cdh^Δ/F and p53 and these cells behaved as expected in the transplantation studies. mILC1 cells used for the transplantation assays were in culture for two weeks prior to surgery. Cells virally transduced to express iFOXO3, iFOXO3.A3, or shFOXOs were cultured and selected for three weeks prior to freezing. MCF7, MDA-MB-231, and SKBR3 cell lines were acquired from the ATCC and verified by short tandem repeat analysis before freezing (2013–2014). Cells were passaged two times a week for about two months, after which fresh vials were thawed. All cells used tested negative for mycoplasma (tests were performed routinely every ∼3 months).

Cells were treated with 100 nmol/L doxycycline. Trolox (200 μmol/L), EUK-134 (20 μmol/L), or N-acetyl cysteine (1 mmol/L) was used for daily antioxidant treatment.

Three-dimensional spheroid cultures were established in 20-μL drops containing 40% Matrigel, and 1.5 mg/mL rat tail collagen I (Ibidi). For colony formation assays, 2,000 cells/well were plated, subsequently fixed with methanol, and stained with 0.5% crystal violet. Fresh tumor tissue digestion into single cells was performed with collagenase and dispase. Neomycin (50 μg/mL) and primocin (100 μg/mL; Invivogen) were used to dispense of stromal cells and avoid infection. NADP⁺/NADPH ratio was determined using Amplite Fluorimetric NADP⁺/NADPH Ratio Assay Kit (AAT Bioquest).

Third-generation packaging vectors and HEK293T cells were used to generate lentiviral particles (17). Lentiviral cDNA expression vectors expressing FOXO3 and FOXO3.A3 were generated using Gateway cloning in the pINDUCER20 and selected with neomycin (Addgene #44012; ref. 18). Foxo1 and Foxo3 knockdown was done using the lentiviral FH1tUTG doxycycline-inducible vector (19, 20). The lentiviral constructs pLV-CMV-PIK3CA-IRES-puromycin, pLV-CMV-PIK3CA^H1047R-IRES-Puromycin, pCDH-myr-HA-PKB-puromycin (Addgene #46969) pLV-CMV-RICTOR-IRES-Hygromycin, and pINDUCER20-GFP-FOXO3^mt were used to express PI3K, PI3K^H1047R, ^myrPKB, RICTOR, and FOXO3^mt, respectively.

For Western blot analysis, cells were lysed in sample buffer containing 0.2% m/v SDS, 10% v/v glycerol, 0.2% v/v β-mercaptoethanol, 60 mmol/L Tris pH 6.8. Protein concentration was determined using Bradford (Bio-Rad) and equal concentrations of protein were used for Western blot analysis. Proteins were detected using 6%–15% SDS-PAGE gels and subsequent Western blot analysis with primary antibodies (Supplementary Table S1) detected by secondary horseradish peroxidase conjugated. Unless stated otherwise, nonspecific bands were used as a loading control. For immunofluorescence, cells were fixed with 3.7% paraformaldehyde in PBS, permeabilized with 0.1% v/v Triton in PBS, and blocked with 2% m/v BSA (Sigma) and 2% v/v NGS (Sigma). Cells were subsequently incubated with primary antibody and visualized with Alexa-488 secondary antibody and Hoechst 33342 (Sigma).

Cells (10⁴) were orthotopically transplanted in the inguinal (4th) mouse mammary gland of Hsd:Athymic Nude-Foxn1^nu (Harlan) recipient mice as described previously (21). Primary tumors were allowed to develop to a volume of 50–100 mm³ (∼100 mm³), at which point expression or knockdown of FOXOs was induced by feeding doxycycline-containing chow (230 mg/kg; Ssniff). Alternatively, FOXO expression or loss was induced 2 weeks after transplantation. Tumor volumes and lung metastases were followed in time as described using a Biospace ϕ imager (Biospace) and digital caliper (Mitotyo; ref. 21). All animal experiments have been conducted in accordance with the approval of the Utrecht University Institutional Animal Care and Use Committee (DEC-ABC no. 2012.III.12.135).

Tissues were fixed in 4% formaldehyde for 24 hours, followed by dehydration in ethanol and paraffin embedding. Rehydrated slides were blocked for endogenous peroxidase activity in phosphate buffer (pH 5.8) containing 1.5% hydrogen peroxide. Antigen retrieval was performed by 20 minutes boiling in 10 mmol/L citrate buffer (pH 6). Primary and secondary horseradish peroxidase–conjugated antibodies were incubated overnight or 1 hour at 4°C, respectively. Staining of slides was performed using diaminobenzidine (DAB) and hematoxylin. Relative DAB intensity was determined using color deconvolution in ImageJ and optical density calculation.

mRNA was isolated with the Qiagen RNeasy Kit (Qiagen). cDNA synthesis was performed using the iScript cDNA Synthesis Kit (Bio-Rad). Real-time PCR was performed using SYBR Green FastStart Master Mix (Roche) in the CFX Connect Real-time PCR detection system (Bio-Rad). Target genes were amplified using specific primer pairs (Supplementary Table S2) and specificity was confirmed by analysis of the melting curves. Target gene expression levels were normalized to Gapdh, Pbdg, and Hrnpn1a levels. Genomic DNA from tumors and cell lines was isolated using the DNeasy kit (Qiagen).

Quality control of total mRNA samples for microarray analysis was performed using a 2100 Bioanalyzer (Agilent). The mRNA of the parental mILC1 cell line was used as reference pool. Microarrays used were Mouse Whole Genome Gene Expression Microarrays V1 (Agilent Technologies) and analyzed using MAANOVA. Raw microarray data will be made publicly available (GSE#). P value is calculated after family-wise error correction (FWER) and 10,000 permutations; P values <0.05 were considered significant. Pathway activity prediction was performed with all significantly changed genes with a log₂ change of >0.5 using the Ingenuity Pathway Analysis platform. Z-scores indicate the predicted pathway activity based on gene expression levels of genes annotated to be functionally involved in the pathway (qiagenbio-informatics.com/products/ingenuity-pathway-analysis; ref. 22). The microarray data have been deposited at the GEO repository under accession number GSE108842. The GEO repository can be accessed at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi.

Cells were stained for apoptosis with Annexin-V-FITC (IQ Products) and propidium iodide (Sigma-Aldrich) and quantified using either a FACSCalibur or a FACSCelesta flow cytometer (BD Biosciences). Anoikis was assessed by culturing 5 × 10⁴ cells/mL for 24 hours in 6-well ultralow cluster polystyrene culture dishes (Corning) prior to apoptotic analysis as above. Oxidative stress was assessed by treatment for 30 minutes with 5 μmol/L CellRox deep red (Thermo Fisher Scientific) or CM-H₂DCFA (Thermo Fisher Scientific) and cytometric analysis. Cell-cycle profiles were generated on the basis of DNA content as measured by propidium iodide staining after fixation with ethanol and RNAse treatment.

Cells were cultured in Lab-Tek II 8-well imaging chambers and treated with doxycycline 72 hours prior to imaging. mILC1-iFOXO3.A3 cells were treated with doxycycline 8 hours prior to imaging. Differential interference contrast imaging was performed on an Olympus Real-Time imager microscope for 48 hours at 37°C and 5% CO₂. Cell tracking and quantification of cell migration was performed using ImageJ.

To determine how FOXO levels and activity affect invasive breast cancer, we established mILC1 cell lines constitutively expressing luciferase, combined with either doxycycline-inducible wild-type FOXO3 (iFOXO3) or the AKT/PKB resistant, constitutively active mutant FOXO3.A3 (iFOXO3.A3). As individual FOXOs function is largely redundant and FOXO3 is most ubiquitously expressed we opted for this paralog. High FOXO3 and FOXO3.A3 expression was observed in mILC1-iFOXO3 and mILC1-iFOXO3.A3 cells after doxycycline treatment. As expected, phosphorylation of the threonine-32 PKB/AKT site in FOXO3 was only detected on FOXO3 and not on FOXO3.A3 (Fig. 1A). FOXO3 was largely sequestered in the cytoplasm and FOXO3.A3 was localized in the nucleus, confirming that Akt/Pkb negatively regulates FOXO3 under basal conditions in mILC1 cells (Fig. 1B). Colony outgrowth was impaired in both mILC1-iFOXO3 and mILC1-iFOXO3.A3 cells (Fig. 1C). No significant increase in apoptosis was measured in both mILC1-iFOXO3 and mILC-iFOXO3.A3 cells after 24 hours (Fig. 1D). Induction of FOXO3.A3 for 48 hours did, however, induce a G₁ cell-cycle arrest (Fig. 1E). A small increase in G₁ cells was observed in mILC1-iFOXO3 cells. Expression of the FOXO target Cdkn1b/p27 was detected after both FOXO3 and FOXO3.A3 induction (Fig. 1F). Combined, these results indicate that colony formation is restricted upon FOXO3 expression through cell-cycle inhibition. In case of iFOXO3, the cell cycle effect appears temporary despite p27 induction after 16 hours. Colony outgrowth is more efficient compared with mILC1-iFOXO3.A3 cells and only a mild G₁ cell-cycle arrest is observed after 48 hours. This could be attributed to the observed high Akt/Pkb–mediated phosphorylation and inactivation of FOXO3. These results confirm that the well-described roles of FOXOs in regulation of the cell cycle are conserved in mILC1 cells.

Orthotopic transplantation of mILC1 cells is a robust model for mILC and the aggressive nature of this tumor type allows studying both primary tumor growth and formation of metastasis. To verify the observed tumor-suppressive effects of FOXO3 and FOXO3.A3 expression in vivo, we orthotopically transplanted 10⁴ cells into the inguinal mammary glands of recipient mice. Expression of FOXO3 and FOXO3.A3 was subsequently induced by feeding mice doxycycline-containing chow either early in tumor growth (2 weeks after surgery) or at a later stage at which tumors reached a size of approximately 100 mm³ (measured weekly). Doxycycline alone has been shown not to influence mILC1 tumor growth in this model (4).

No effect on primary tumor growth was observed upon induction of FOXO3 (Fig. 1G), whereas early expression of FOXO3.A3 did hamper initial primary tumor outgrowth. Late expression of FOXO3.A3 resulted in collapse of the tumor (Fig. 1H). However, primary tumor growth was reestablished upon late FOXO3.A3 expression, despite continuous doxycycline administration. IHC analysis revealed that reduction of tumor volume correlated with inhibition of cell proliferation (Ki67) and induction of apoptosis (cleaved caspase-3; Supplementary Fig. S1A).

Detection of luciferase activity in the mouse outside the primary tumor location, primarily the lungs, was used to determine metastasis of mILC1 tumors. No significant difference in metastasis-free survival was detected for mice bearing tumors in which FOXO3 or FOXO3.A3 was induced during early tumor growth. In contrast, late FOXO3 or FOXO3.A3 induction significantly delayed metastasis detection (Fig. 1I and J). The average primary tumor volumes were significantly larger when metastases were first detected in doxycycline-treated mice (Supplementary Fig. S1B and S1C). This suggests that FOXO3 activation not only restrained primary tumor growth and thereby delayed the onset of metastasis, but that tumors also needed larger volume and cell numbers for metastasis to arise. Together, these results show that AKT/PKB-mediated FOXO inhibition facilitates both cancer growth and metastasis.

Although induction of FOXO3.A3 in mILC1 initially led to reduced tumor growth and delayed metastasis, these effects were temporary, which suggests that mILC1-iFOXO3.A3 cells either adapt to high FOXO3.A3 levels or that there is negative selection for cells expressing high FOXO3.A3 levels. To test this, we established cell lines from mILC1-iFOXO3.A3 cells isolated from untreated, early, and late treated tumors. mILC1-iFOXO3.A3 cell lines established from doxycycline-treated mice overcome the FOXO3.A3-mediated inhibition of colony formation. In contrast, colony formation was still abolished upon FOXO3.A3 induction in cells isolated from mILC1-FOXO3.A3 tumors from untreated mice. All these cell lines had retained neomycin resistance, indicating that the pINDUCER20-FOXO3.A3 construct was still present (Fig. 2A). This was further confirmed by quantification of the relative amount of viral integrations of the construct in the genome of cell lines established from the tumors from all groups. Strong selection for cells carrying low amounts of viral integrations of the FOXO3.A3 construct was observed in doxycycline-treated tumors (Fig. 2B), but the construct was indeed not completely lost and super-endogenous levels of FOXO3.A3 protein were still observed after doxycycline treatment (Fig. 2C). Similar selection for low, but super-endogenous levels of FOXO3.A3 expression was observed in mILC1-iFOXO3.A3 cells cultured in the presence of doxycycline for 3 weeks (Fig. 2D). Although expression levels of FOXO3.A3 are low in cells isolated from mILC1 tumors continuously treated with doxycycline, FOXO3.A3 localization was still capable of nuclear accumulation (Fig. 2E).

Taken together, we observed a strong selection for cells expressing low (but higher than endogenous Foxo3) levels of FOXO3.A3, we did however not find tumors that lost inducible FOXO3.A3 expression altogether.

As outlined above, our data confirmed that hyperactivation of FOXOs suppresses tumor growth and metastasis in vivo, but also suggests that complete loss of FOXO activity is not selected for during tumor progression. We therefore set out to determine the effects of induced FOXO loss in mILC. As FOXO transcription factors are highly redundant in their function, we first determined the endogenous expression levels of Foxo1, Foxo3, and Foxo4 in mILC1. Both Foxo1 and Foxo3 were expressed but Foxo4 mRNA was hardly detected (Supplementary Fig. S2A). Conveniently, Foxo1 and Foxo3 share sufficient sequence overlap that they can be targeted by the same shRNA (Supplementary Fig. S2B; refs. 20, 23, 24). As shRNA#3 is predicted to be the most preferred shRNA based on targeting of both mouse and human FOXOs with the least off-targets, we expressed shRNA#3 in a doxycycline-inducible fashion in mILC1 cells (shFOXOs; portals.broadinstitute.org/gpp/public). Efficient knockdown of both Foxo1 and Foxo3 was observed upon activation of shFOXOs expression (Fig. 3A; Supplementary Fig. S2B and S2C). Several phenotypes described further in this study using shRNA#3 below were corroborated using shRNA#1 and shRNA#2 (Supplementary Fig. S2D–S2G). Although Foxo knockdown reduced colony outgrowth (Fig. 3B; Supplementary Fig. S2D), it did not induce apoptosis or lead to a cell-cycle arrest (Fig. 3C and D; Supplementary Fig. S2E). However, we observed an overall increase in cell-cycle time as measured by video time-lapse microscopy (Fig. 3E). Whereas Cdkn1b/p27 was induced upon FOXO3 activation, inactivation of FOXOs in mILC1 cells resulted in the expression of Cdkn1a/p21, indicating that FOXO loss and FOXO activation block cell-cycle progression through different mechanisms (Fig. 3F).

To determine the effect of FOXO loss in vivo, we orthotopically transplanted mILC1-shFOXOs cells into the inguinal mammary gland of recipient mice. As it was previously reported that shLuc expression does not affect tumor growth in our mILC1 model, we did not include this in the current study (21). Early FOXO loss hampered tumor initiation strongly, delaying primary tumor growth by approximately 4 weeks. Late expression of shFOXOs in primary tumors of approximately 100 mm³ resulted in efficient but temporary growth inhibition (Fig. 3G). IHC analysis of Foxo1, Ki67, phospho-Histone H3, and cleaved caspase-3 levels in mILC1-shFOXOs tumors treated with doxycycline for 7 days after the tumors had reached approximately 100 mm³, revealed that the inhibition of primary tumor growth correlates with reduced proliferation (Supplementary Fig. S3A).

Strikingly, the detection of metastasis was significantly delayed when FOXO was knocked down early in tumor growth compared with control tumors. Late FOXO knockdown in established tumors delayed detectable metastasis in mice that had not presented with metastasis up to that point (Fig. 3H). Average primary tumor volumes were significantly larger when metastases were first detected in doxycycline-treated mice, indicating that loss of Foxo1 and Foxo3 expression not only restrained primary tumor growth but also metastasis formation (Supplementary Fig. S3B). Together, these results show mILC1 tumors initially depend on the presence of Foxos for both efficient growth and metastasis.

Similar to mILC1-FOXO3.A3 tumors, the suppressive effects on tumor growth observed in mILC1-shFOXOs tumors were temporary. We quantified the relative amount of viral integrations of the FH1tUTG-shFOXOs construct in the genome of mILC1-shFOXOs cells isolated from untreated, early and late treated tumors by real-time qPCR. In contrast to mILC1-iFOXO3.A3 tumors, quantification of viral integrations in the genome of mILC1-shFOXOs tumor cells showed that only in early treated tumors selection for cells with significantly lower amounts of viral integrations occurred (Fig. 4A). GFP expression levels corresponded to the relative amount of viral integrations (Fig. 4B). Efficient Foxo1 and Foxo3 knockdown was still induced in doxycycline-treated mILC1-shFOXOs cells derived from the doxycycline-treated tumor groups; however, slightly higher residual FOXO levels were detected in cell lines derived from early treated tumors (Fig. 4C). Similar to the cell lines isolated from tumors, mILC1-shFOXOs cells cultured in the constant presence of doxycycline for 3 weeks still showed efficient FOXO knockdown (Fig. 4D) Interestingly, colony-forming capacity of tumor cell lines carrying lower amounts of viral integrations was no longer impaired, suggesting that a slight reduction in knockdown efficiency, and therefore residual Foxo1 and Foxo3 expression, is already sufficient to overcome the tumor-suppressive effect of FOXO loss (Fig. 4E). To test whether our findings apply also to human breast cancer, we established MCF7, MDA-MB-231, and SKBR3 breast cancer cell lines carrying the shFOXOs and iFOXO3.A3-inducible constructs and found that similar to mILC1 cells, human breast cancer cells are also sensitive to both loss and overactivation of FOXO (Fig. 4F–I).

As our results suggest that mILC1 cells could rely on balancing FOXO activity to thrive, we determined whether there is indeed a window of FOXO expression that supports the growth of mILC1, MCF7, SKBR3, and MDA-MB-231 cells. To this end, we titrated doxycyline in shFOXOs and iFOXO3.A3 cells, to gradually change FOXO levels. Indeed, we find the highest colony formation capacity when FOXO levels are near or slightly above the endogenous level (Fig. 4J–L). In contrast to the common idea that FOXOs are tumor suppressors, our results now establish that breast cancer cells rely on the presence of FOXOs for efficient tumor growth and metastasis.

As our data indicate that cancer cells preferably retain FOXOs, we sought to determine in which processes and steps of tumor progression FOXOs are most important. For cells to become metastatic, they need to be motile and able to survive dissemination from the primary tumor. FOXOs have been suggested to be involved in regulating cell motility and migration in breast and colon cancer cells (6, 8). In contrast, others and we have previously shown that FOXO activation induces anoikis in cancer cells (4, 25). We therefore determined whether the observed decrease in metastatic potential upon both loss and gain of FOXO activity correlates with changes in cell migration or prevention of anchorage-independent survival.

Live tracking microscopy of mILC-shFOXOs and mILC1-iFOXO3.A3 in 2D culture showed a significant inhibition of cell motility upon loss but not upon activation of FOXO (Fig. 5A). Large multicellular structures that invade into the matrix are formed when mILC1, MDA-MB-231, and SKBR3 cells grow as spheroids in a mixture of Matrigel and collagen. Loss of FOXO results in a marked reduction of this invasive growth, whereas forced expression of FOXO3.A3 abolishes spheroid formation or leads to the collapse of the 3D structure when induced after the formation of invasive protrusions (Fig. 5B; Supplementary Fig. S4).

It was previously shown that FOXO hyperactivation can induce anoikis (4), but the effect of FOXO loss on anoikis has not been studied thus far. Anoikis was increased in mILC1-shFOXOs cells when transferred to suspension after 72 hours of FOXO knockdown (Fig. 5C; Supplementary Fig. S2F). Similar to mILC1, anoikis resistance of MDA-MB-231 and SKBR3 was abolished upon loss of FOXOs (Fig. 5D). Anoikis induction in response to FOXO3 hyperactivation involves the upregulation of Bim and Bmf expression. These FOXO targets are indeed downregulated upon FOXO knockdown in mILC1, MCF7, MDA-MB-231, and SKBR3 cells and hence does not explain the concurring increase in anoikis (Fig. 5E–G). The expression levels of the antiapoptotic Bcl2 and Bcl-xl were, however, also reduced after FOXO knockdown, which could underlie the observed sensitivity to anoikis (Fig. 5E–G). Indeed, exogenous expression of BCL2 or BCL-XL in mILC1-shFOXOs and mILC1-iFOXO3.A3 reestablished anoikis resistance in these cells (Fig. 5H–J). Collectively, these results show that mILC1 cells rely on the presence of FOXOs to invade and circumvent anoikis, providing a rationale for the observed reduction in metastasis upon loss of FOXO in vivo. How FOXOs influence invasion and migration at the molecular level remains to be elucidated. Changes in canonical EMT and migration markers (5) in a FOXO-dependent manner were not observed as discussed below.

Anchorage-independent survival in cancer is dependent on active growth factor signaling and the ability to maintain redox homeostasis (26, 27). FOXOs play key roles in redox signaling and redox homeostasis (9, 28–30). FOXO knockdown led to induction of stress-activated protein kinase (JNK and p38) activity, suggesting that FOXO loss leads to elevated receiver operating characteristic (ROS) levels (Fig. 6A). Next, we determined whether stress kinase activation correlates with increased levels of ROS. CellRox fluorescence intensity was increased after FOXO knockdown, indicating that ROS levels increase upon FOXO loss (Fig. 6B). Similar to mILC1, loss of FOXOs resulted in increased ROS levels in MCF7, MDA-MB-321, and SKBR3 cells (Supplementary Fig. S5A). The NADPH/NADP⁺ ratio decreased after FOXO knockdown supporting the notion that an overall more oxidative environment prevails in cells after loss of FOXO (Fig. 6C). Interestingly, no changes in the expression of FOXO target genes Catalase and Sod2 were detected, suggesting that the redox imbalance is not caused by reduced expression of proteins involved in ROS detoxification (Supplementary Fig. S5B–S5D).

Although FOXOs clearly affect redox homeostasis in mILC1 cells, compounds that act to boost antioxidant function in various ways, for example, Trolox (free radical species scavenger), N-acetyl cysteine (GSH precursor), or EUK134 (Superoxide and hydrogen peroxide scavenger), did not rescue cell proliferation or anoikis induction by FOXO loss (Fig. 6D and E). Taken together, these data indicate that although FOXOs are essential for cellular homeostasis in mILC1, the role of FOXOs in cancer is evidently more complex than just providing tumor cells with sufficient reductive capacity.

Growth factor signaling is fundamental to tumor progression and anchorage-independent survival is strongly dependent on RAS and PI3K pathway activity (4, 25, 26, 31). Although the precise regulation of growth factor feedback signaling by FOXO remains to be fully characterized, it is clear that FOXOs are key regulators of PI3K pathway activity in response to pharmacologic inhibition of PI3K (11–13, 32). We therefore set out to determine the effects of FOXO activation and loss on PI3K pathway activity.

Expression of FOXO3.A3 increases PI3K–PKB activity, whereas loss of FOXO correlated with a marked reduction in Pkb-S308 and Pkb-S473 phosphorylation (Fig. 7A; Supplementary Fig. S2G). To validate these observations in vivo, we stained tissue sections from mILC1-iFOXO3.A3 and mILC1-shFOXOs tumors for Pkb and Pkb-S473 phosphorylation. Induction of FOXO3.A3 or shFOXOs correlated with increased and reduced levels of Pkb-S473, respectively (Fig. 7B). FOXO activation induced mRNA expression of a panel of growth factor signaling feedback components including Met, Erbb2, Erbb3, Insr, Igf-1r, Pdgfra, Fgfr1, Fgfr3, Irs1, Irs2, Rictor, Sesn1, and Sesn3 (Fig. 7C). FOXO knockdown indeed resulted in decreased mRNA expression of most of these gene transcripts. (Fig. 7C). Western blot analysis confirmed that FOXOs control the protein expression levels of Met, Igf-1r, Insr, Erbb2, Erbb3, Irs1, Irs2, Rictor, and Pdgfra (Fig. 7D). The role of FOXOs as important mediators of PI3K–PKB feedback signaling was confirmed in MCF7, MDA-MB-231, and SKBR3 human breast cancer cells (Supplementary Fig. S6A and S6B).

Coexpression of an shRNA-insensitive mutant of wild-type FOXO3 (FOXO3^mt) in mILC1-shFOXOs both restored PKB activity and prevented stress kinase activation (Supplementary Fig. S7A). Colony formation was, however, not rescued, which is in line with our observations in Fig. 1C that FOXO overexpression blocks proliferation. This suggests that in the context of high FOXO expression the ensuing high PI3K/PKB activity is not sufficient to bypass the FOXO-mediated arrest.

To test whether loss of active PI3K/PKB underlies the inhibition of colony formation upon shFOXO expression, constitutively active variants of PI3K (PI3K^H1047R), PKB, (^myrPKB), or RICTOR were coexpressed. This could, however, not rescue shFOXO-mediated anoikis and proliferation, even in the presence of antioxidants (Supplementary Fig. S7B–S7G). These results suggest that FOXO controls other crucial survival pathways PI3K/PKB signaling and the cellular redox state.

Indeed, microarray analysis of mRNA expression changes in mILC1-shFOXOs and mILC1-iFOXO3.A3 revealed that the expression levels of thousands of genes are influenced by FOXOs, including the most well-established FOXO target genes (Supplementary Fig. S8A; refs. 33, 34). Subsequent pathway prediction analysis using all significantly changed genes including non-FOXO target genes, underscores that many pathways and processes additional to the above studied cell proliferation, apoptosis, stress, and RTK pathways in cancer cells are attenuated upon aberrant FOXO expression (Supplementary Fig. S8A–S8E; Supplementary Table S3). Of note, and in line with the observed loss of viability in the context of both shFOXOs and iFOXO3.A3, several pathways are regulated in the same direction upon loss and overexpression of FOXO. Despite the earlier described roles for FOXO in cell migration and EMT (5), we found no evidence of transcriptional control of these programs by FOXO (Supplementary Fig. S8F)

Our findings suggest that carefully tuning FOXO activity is essential for tumor cells to thrive, as ectopic hyperactivation, complete loss of FOXO or hyperactivation of GFR signaling all hamper tumorigenesis due to imbalances in proliferation, survival, migration, redox homeostasis, and PI3K pathway activity (Fig. 7E).

This study for the first time characterizes the effects of both induced FOXO activation and loss during tumor progression in a model for epithelial cancer (i.e., metastatic breast cancer). We show that FOXO3 has tumor-suppressive potential at various stages of cancer in this model. Elevated FOXO3 levels had no effect on primary tumor growth but did delay metastasis. The expression of constitutively nuclear and hyperactive FOXO3.A3 strongly affected both primary tumor growth and metastasis, underlining that attenuation of FOXO activity by AKT/PKB is essential for tumor cells to efficiently proliferate and survive. Cells expressing high levels of FOXO3.A3 were negatively selected for, but cells with levels higher than endogenous FOXO3 were retained in the tumor meaning that FOXO activity is not tumor suppressive per se and could contribute to tumor growth or survival.

Indeed, induced loss of FOXOs revealed that mILC1 tumors are also highly dependent on endogenous FOXOs for tumor growth, metastasis, and cellular homeostasis. These observations strongly argue against a general role for FOXOs as classical tumor suppressors. Interestingly, it takes longer for early treated mILC1 tumors to grow out upon FOXO loss than upon FOXO3.A3 overexpression, suggesting that loss of FOXO is a more complex problem for mILC cells to overcome. On the basis of these observations, we conclude that FOXOs can both suppress and support tumor progression and that tuning FOXO activity is essential for tumor growth and metastasis. The key observations in this manuscript were corroborated in human breast cancer cell lines, suggesting that the conclusions can be extended to human cancer.

An unexpected finding in this study is that both loss and gain of FOXO activity hampers metastasis formation. A potential explanation might be that metastasis is delayed in both situations because primary tumor growth is also delayed and that a tumor needs to reach a certain size or stage before it can metastasize. However, primary tumor volume at the time of the first detected metastasis was larger both upon FOXO loss and gain of activity (Supplementary Figs. S1 and S3), arguing against this explanation. Another rationale could be that these observations may stem from the different roles of FOXO at different stages of metastasis. It could be that that both the anoikis promoting as well as the antiproliferative effects of FOXO3.A3 hyperactivation may hamper both dissemination of tumor cells and efficient outgrowth of distant metastasis to form detectable tumors, whereas loss of FOXO prevents efficient survival during EMT due to its roles in motility, invasion,-anchorage independent growth, PI3K feedback signaling, and cellular redox homeostasis. Adverse conditions like high levels of reactive oxygen species or nutrient depletion (as has been reported to occur during EMT) have been reported to activate FOXOs even in the presence of growth factor signaling (35–37) to maintain homeostasis. It is therefore not unlikely that loss of these FOXO functions delay the metastasis in the mILC1-shFOXOs tumors. An observation that somewhat supports this idea is that late induction of loss of FOXO in mILC1-shFOXOs tumors seems to delay metastasis efficiently, but in only about half of the mice (Fig. 2H). It could be that in the other mice EMT and establishment of (not yet detectable) micrometastasis had already occurred and that these are less dependent on FOXO activity for survival. Interestingly, expression of genes involved in canonical EMT and invasion in response to FOXO activity did not change in a way that could explain the observed effects on invasion and migration (5, 8, 38). This further strengthens the idea that the role of FOXO in metastasis lies more in regulation of survival during anchorage-independent growth, PI3K–PKB feedback signaling and proliferation and outgrowth of distant metastasis. It could, however, be that FOXOs regulate EMT at earlier stages of tumorigenesis. After all, the mILC1 breast cancer model is already prone to metastasis before FOXO3.A3 or shFOXO expression is induced.

Because of the role of FOXO in a wide variety of cellular processes, it can be expected that our attempts to reestablish proliferation and anoikis resistance after FOXO knockdown with antioxidants and/or growth factor signaling components were unsuccessful in tissue culture experiments. Likely, the right balance of induction of processes otherwise regulated by FOXO must be found, a complex process that is difficult, if not impossible, to faithfully mimic using experimental conditions. It could for instance be that tumor cells benefit most from oscillating or regulatable levels of FOXO activity to maintain (redox) homeostasis and PI3K–PKB feedback signaling without the induction of growth arrest.

FOXOs regulate a variety of vital cellular processes and it therefore may seem not surprising that tumor cells retain FOXO activity. However, most, if not all, tumor cells have loss of function of the p53, also a transcription factor that besides its role as a tumor suppressor also has a multitude of functions in cellular homeostasis, many of which overlap with FOXO (39). It would therefore be interesting to study to what extent tumor cells rely on FOXO activity when, unlike in mILC1 cells, p53 function is intact. It is tempting to speculate that in the absence of p53, tumor cells could rely more on FOXO activity. As, unlike p53, mutational loss of all FOXOs has not been reported in cancer (40), this would make FOXO an interesting therapeutic target. Our observations that loss of FOXO is detrimental to tumor growth and metastasis were unexpected and might go against the rationale for the development of several small molecules that aim to promote FOXO activity either directly (41) or through inhibition of PI3K/AKT for cancer therapy (42, 43). Although we do not currently know how our observations on the requirement for FOXO extend to other cancer types, it is tempting to speculate that compounds that can inhibit FOXOs could be effective in cancer treatment. Of note, recently efforts have been made to make such FOXO-directed compounds for the treatment of type II diabetes, illustrating that the development of FOXO inhibitors is certainly possible (44, 45).

No potential conflicts of interest were disclosed.

Conception and design: M. Hornsveld, L.M. Smits, D.E. Kloet, P.W.B. Derksen, B.M.T. Burgering, T.B. Dansen

Development of methodology: M. Hornsveld, L.M. Smits, D.E. Kloet, T.B. Dansen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Hornsveld, L.M. Smits, M.J.G. Koerkamp, F.C. Holstege, P.W.B. Derksen, T.B. Dansen

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Hornsveld, M.J.G. Koerkamp, D. van Leenen, F.C. Holstege, P.W.B. Derksen, T.B. Dansen

Writing, review, and/or revision of the manuscript: M. Hornsveld, B.M.T. Burgering, T.B. Dansen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Hornsveld, M. Meerlo, M. van Amersfoort, D. van Leenen

Study supervision: M. Hornsveld, B.M.T. Burgering, T.B. Dansen

This work was supported by Dutch Cancer Society (KWF Kankerbestrijding) grants UU2014-6902 and UU2009-4490 (to T.B. Dansen) and UU2014-7201 and UU2017-10456 (to P.W.B. Derksen). B.M.T. Burgering, M. Hornsveld, L.M.M. Smits and M. Meerlo are part of the Oncode Institute, which is partly financed by the Dutch Cancer Society (KWF Kankerbestrijding) and was funded by the gravitation program CancerGenomiCs.nl from the Netherlands Organisation for Scientific Research (NWO). We are grateful for input from our colleagues from the Dansen and Burgering labs.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.