PDK1 Signaling Toward PLK1–MYC Activation Confers Oncogenic Transformation, Tumor-Initiating Cell Activation, and Resistance to mTOR-Targeted Therapy

The phosphoinositide 3-kinase (PI3K)–AKT pathway is one of the most commonly deregulated signaling pathways in human cancers (1). Genetic aberrations affecting this pathway, such as activating mutations of PIK3CA or inactivation of PTEN, have been identified in virtually all epithelial tumors (2). The 3-phosphoinositide–dependent protein kinase-1 (PDK1) is known to be activated as a result of the accumulation of the PI3K product phosphatidylinositol-3,4,5-trisphosphate (PIP3), and thus considered an important component of the PI3K pathway. PDK1 is a master regulator of AGC kinase members, including AKT, p70 ribosomal S6 kinase (S6K), serum- and glucocorticoid-induced protein kinase (SGK), and protein kinase C (PKC) family members, thus having multiple roles in various physiologic processes such as metabolism, growth, proliferation, and survival (3). In human cancers, PDK1 is thought to be constitutively activated upon elevation of PIP3 owing to the loss of PTEN or gain of PIK3CA activity. In addition, PDK1 deregulation in human malignancy can also be caused by gene amplification or abnormal phosphorylation in the cytosol and nucleus, as in colon cancer and invasive breast cancer (4, 5).

One of the most defined PDK1 targets relevant in human cancer is AKT. Specifically, PDK1 directly phosphorylates AKT on T308, but requires mTOC complex 2 (mTORC2)-induced AKT phosphorylation on S473 to confer a full activation (6). Given its connection to AKT, PDK1 has been pursued as a critical anticancer target (7). However, in view of the diversity of PDK1 substrates, additional downstream targets of PDK1 may confer aberrant signaling heterogeneity and complexity in human malignancy. Indeed, it has been recently shown that inhibition of PDK1 has no significant effect on AKT signaling in a PTEN-deficient transgenic tumor mouse model (8) or breast tumor growth (9), and oncogenic functions of PDK1 through substrates other than AKT, such as SGK3 (10), mitogen-activated protein kinase (MAPK; ref. 11), or PKCα (12), have also been reported. In addition, our recent work has shown that PDK1 is required for MYC protein accumulation in colon cancer cells treated with the mTOR inhibitor rapamycin (5), indicating a potential functional link of PDK1 with MYC in oncogenesis.

MYC is implicated in both cancer and stem cell self-renewal. The relationship between stem cells and human cancers has become an important issue in cancer research given that self-renewal is a hallmark of both cell types (13). Genes associated with embryonic stem cell (ESC) identity, including pluripotency transcription factors, Polycomb targets, and MYC targets, have been observed in aggressive human cancers and are associated with poor disease outcome (14). Moreover, the MYC-associated molecular network is strikingly similar between ESC and human cancer transcription programs (15), and ectopic overexpression of MYC in differentiated somatic cells can induce both an ESC gene signature and properties of cancer stem cells (CSC; ref. 16). These findings suggest that activation of an ESC-like gene expression program in adult cells may confer self-renewal to cancer cells or CSCs. Notably, although the cancer-associated ESC-like gene regulation by transcription factors has been well documented, its regulation by a druggable kinase-driven signaling pathway has yet to be identified.

In the present study, we investigated PDK1-evoked key signaling events required for oncogenic transformation. We identified that the PDK1–Polo-like kinase 1 (PLK1)–MYC pathway is a major driver pathway conferring PDK-induced transformation, and its existence is readily evident in human cancers. We further show that PDK1–PLK1–MYC signaling drives an ESC-like gene expression signature relevant in human cancers and is robust in inducing a CSC phenotype. It is also involved in resistance to mTOR inhibitor in colorectal cancer cells. These findings provide important insights into cancer, CSC biology, and potential new treatment for targeting MYC dependency in human cancers.

As the first step in investigating the differential signaling pathways activated by PDK1 or PI3K in tumorigenesis, we compared the transforming capacity of PDK1 and PI3K by using the in vitro transformation assay that measures anchorage-independent growth in soft agar. We began with semitransformed human embryonic kidney epithelial cells (HEK) that express a low level of activated HRASV12 (HEK-TERV; ref. 17) and infected them with retroviral vectors expressing PDK1, MYC, a constitutively activating mutant of PIK3CA (E545K), or PTEN short hairpin RNA (shRNA), resulting in stable cell lines designated as HEK-PDK1, HEK-MYC, HEK-E545K, or HEK-shPTEN cells, respectively. The transformation assay results showed that they were all able to induce cellular transformation, although PDK1- and MYC-induced colonies appeared to be larger in size as compared with that of E545K- or shPTEN-expressing cells (Fig. 1A and Supplementary Fig. S1A). Consistent with our previous report showing a posttranslational MYC induction by PDK1 (5), we detected a marked protein accumulation of MYC in HEK-PDK1 cells but not in HEK-E545K or HEK-shPTEN cells (Fig. 1B), which was not due to the induction of Myc mRNA levels (Supplementary Fig. S1B). We also showed that the kinase activity of PDK1 is required for transformation as well as MYC protein induction, as a kinase-dead mutant of PDK1 (PDK1 K100N; ref. 9) induced neither the transformation nor the MYC accumulation (Supplementary Fig. S1C).

A survey of known AGC substrates of PDK1 revealed that PDK1 also induced a strong phosphorylation of PKCδ and a modest increase of phosphorylated AKT (T308). The other known PDK1 substrates, including SGK1/3 and S6K, were not activated, nor AKT phosphorylated at S473, which is required for a full activation of AKT. In contrast, E545K overexpression induced strong phosphorylation of AKT (at both T308 and S473) as well as the downstream substrates FOXO1 and FOXO3A (Fig. 1B). Thus, the remarkable observation that PDK1 induces transformation in the presence of a weak AKT activation suggests a potential more functional role of MYC in this process. Indeed, RNA interference–mediated knockdown of MYC resulted in much-reduced transformation of HEK-PDK1 cells, but not of HEK-E545K cells (Fig. 1C), showing a MYC dependency for PDK1-induced transformation. Moreover, in a series of dose–response analysis (see Supplementary Fig. S1D and S1E), HEK-PDK1 cells, compared with HEK-E545K cells, were much more sensitive to small-molecule PDK1 inhibitors BX795 and BX912 (Fig. 1D, left and Supplementary Fig. S1D). In contrast, E545K-transformed cells were much more sensitive to the PI3K inhibitor GDC-0941 and the AKT inhibitors MK2206 and GSK690693 (Fig. 1D, left and Supplementary Fig. S1E). Consistent with these effects, BX795 reduced MYC accumulation but had only a modest effect on AKT. In contrast, GDC-0941 or MK2206 easily abolished phosphorylations of AKT in HEK-E545K cells, but had no such effects on MYC inhibition (Fig. 1D, right). These results showed the differential pathway dependency for the two transformed cell systems. Interestingly, MYC-transformed cells were also sensitive to BX795 (Fig. 1D, left), which is consistent with the observation that BX795 was able to eliminate the exogenous MYC in these cells (Fig. 1D, right). Altogether, these data show that PDK1-induced transformation depends more on MYC, but less on AKT signaling, when compared with E545K-driven transformation. The data also suggest that MYC-dependent cells become sensitive to the PDK1 inhibitor, regardless of PDK1 status, which reveals a PDK1 dependency in MYC-driven cells. PDK1-induced MYC activation upon transformation was also observed in immortalized human mammary epithelial cells (HMEC) and prostate epithelial cells (RWPE-1; Fig. 1E and F), suggesting that MYC activation by PDK1 is not restricted to HEK cells but occurs in multiple epithelial lineages.

To show the physiologic relevance of the PDK1–MYC connection in human cancers, we showed that PDK1 knockdown was able to eliminate MYC expression in a variety of human cancer cell lines (Fig. 1G). Moreover, in a panel of breast cancer cell lines in which the MYC-dependent viability has been previously characterized (18), BX795 treatment resulted in similar MYC depletion in all these cells (Fig. 1H) but preferentially reduced the cell viability of MYC-dependent breast cancer cell lines (MDA-MB-231, Hs578T, and SUM159PT) as compared with the MYC-independent breast cancer cell lines (T47D and BT474; Fig. 1I). Of note, in these cell lines BX795 seemed to inhibit AKT and FOXO3A phosphorylations in a cell-dependent manner (Fig. 1H). Taken together, these results show a potential role of PDK1 activity toward MYC regulation, which is therapeutically implicated for MYC-driven tumors.

To investigate whether or not there are downstream kinases of PDK1 that are crucial for MYC induction and cell transformation, we conducted a screen for kinases that, when pharmacologically inhibited, selectively kill PDK1-transformed cells. Among 60 small-molecule protein kinase inhibitors we have screened, we found that two PLK1 inhibitors (BI2536 and GW843682X), one mitogen-activated protein (MAP)-extracellular signal-regulated kinase (ERK) (MEK) inhibitor (PD0325901), one ALK inhibitor (NVP-TAE684), one BRC–ABL inhibitor (PD180970), and one tyrosine kinase inhibitor (sunitinib) showed a preferential inhibitory effect on the viability of HEK-PDK1 cells as compared with the control cells (Supplementary Fig. S2A). The two PLK1 inhibitors were further validated in a secondary screen and thus were chosen for further study (data not shown). Further analyses in all three epithelial systems showed that PDK1-transformed cells were much more sensitive to the PLK1 inhibitors compared with vector control or E545K-transformed cells (Fig. 2A and B). This finding reveals a possible role of PLK1 in PDK1-induced transformation. Indeed, Western blot analysis showed an induction of PLK1 phosphorylation in all three PDK1-transformed cell lines, but not in E545K- or MYC-transformed cells (Fig. 2C). Similar to the PDK1 inhibitor BX795, BI2536 treatment resulted in strong colony growth inhibition in both PDK1- and MYC-transformed cells, but not in E545K-transformed cells (Fig. 2D and Supplementary Fig. S2B). Furthermore, like BX795, the PLK1 inhibitors BI2536 or GW843682X were able to eliminate endogenous MYC in HEK-PDK1 cells but also exogenous MYC in HEK-MYC cells (Fig. 2E). This finding suggests that exogenous MYC is also sensitive to the perturbation of the basal level of PDK1–PLK1 signaling. Accordingly, in both HEK-PDK1 and HEK-MYC cells, but not in HEK-E545K cells, BI2536 treatment resulted in strong apoptosis, as shown by both fluorescence-activated cell sorting (FACS) analysis (Fig. 2F) and increased caspase-3 activity (Supplementary Fig. S2C), whereas E545K-transformed cells mainly displayed G₂–M arrest, a typical feature related to a mitotic effect following PLK1 inhibition (Supplementary Fig. S2D). Furthermore, to confirm the PLK1-specific effect of BI2536, PLK1 depletion by three independent siRNAs gave rise to similar effects on endogenous and exogenous MYC and apoptosis in PDK1- or MYC-driven cells (Fig. 2G). These findings suggest a crucial role for PDK1–PLK1 signaling in regulating MYC and cancer cell survival. Consistent with the in vitro data, xenograft tumors derived from HEK-PDK1 cells were highly sensitive to BI2536 treatment and displayed a strong tumor regression following just two dosages, whereas the same treatment induced only tumor growth inhibition in E545K-derived xenograft tumors (Fig. 2H).

We further showed that the PLK1 regulation of MYC was not limited to transformed cells but was also physiologically relevant in human cancers, as either PLK1 knockdown or BI2536 treatment resulted in endogenous MYC protein depletion in various cancer cell lines without changing Myc mRNA levels (Supplementary Fig. S3A–S3C). In addition, a time-course analysis indicated that BI2536 treatment resulted in MYC depletion as early as 8 hours, concomitant with an early G₂–M arrest (Supplementary Fig S3D), indicating that MYC downregulation is unlikely to be a result of the secondary effect of cell-cycle change. BI2536 treatment also resulted in more effective growth inhibition in MYC-dependent breast cancer cell lines compared with MYC-independent cells (except MDA-MB-231; Fig. 2I). These results further support a role of PDK1–PLK1 signaling in supporting MYC-driven tumorigenesis.

We next sought to determine whether or not the PLK1 activation by PDK1 seen in transformed cells represents a finding that is physiologically relevant in human cancer cells. To achieve this, we first used the colon cancer HCT116 and DLD1 cells in which PDK1 is genetically knocked out (19). To facilitate the detection of phosphorylation of PLK1, which is a mitotic kinase, cells were first synchronized by double-thymidine block and then released into the cell cycle progressively (Fig. 3A and Supplementary Fig. S4A). In PDK1 wild-type cells, we noticed progressive induction of PDK1 phosphorylation upon cell-cycle progression into mitosis as indicated by the elevated levels of phosphor-histone H3, which was accompanied by a similar pattern of PLK1 phosphorylation; whereas in PDK1^−/− counterparts, we detected a much more deficient PLK1 phosphorylation and MYC accumulation but not the phosphorylation of the PLK1-related kinase Aurora A (refs. 20, 21; Fig. 3A; Supplementary Fig. S4A).

We also probed the changes in the AKT–mTOR pathway in these cellular contexts. Of note, compared with p-PLK1, p-AKT (T308) was only modestly changed in this condition in PDK1^−/− cells, whereas p-AKT (S473) and p-FOXO3A were even enhanced in both PDK1^−/− cell lines (Fig. 3A). This could be due to the inhibition of S6K in PDK1^−/− cells, leading to a feedback activation of p-AKT (S473). In contrast, in a different condition where cells were serum-starved and then stimulated with growth factors for early time points, we saw a clear p-AKT-(T308) inhibition in HCT116 PDK1^−/− cells (Supplementary Fig. S4B). Thus, PDK1 regulates p-PLK1 and p-AKT (T308) in different growth conditions. To further consolidate the data, we also conducted PDK1 knockdown by shRNA in MDA-MB-231 cells. The result showed again that PDK1 knockdown resulted in ablation of PLK1 phosphorylation and MYC accumulation (Fig. 3B), as well as deficient entry into mitosis (Fig. 3C). These data consolidated a role of PDK1 in driving PLK1 and MYC activation, not just in a confined system but also in cancer cells.

Furthermore, in multiple cancer cell lines treated with BX795, GDC0941, or MK2206 upon double-thymidine block and release (Supplementary Fig. S4C), we saw that BX795 always blocked PLK1 phosphorylation and MYC accumulation, but inhibited AKT phosphorylation in a cell line–dependent manner (for example, AKT phosphorylation is not affected by BX795 in MDA-MB-231 cells). In contrast, GDC0941 and MK2206 consistently inhibited AKT phosphorylation in each of these cell lines, but had little effect on PLK1 and MYC. Together, these data show the physiologic relevance of AKT-independent PDK1–PLK1–MYC signaling in cancer cells.

We next investigated whether or not PLK1 is a potential substrate of PDK1. PDK1 is known to regulate AGC kinases. Protein domain analysis indicated that the kinase domain of PLK1 is part of the AGC kinase family (Fig. 3D). Interestingly, the amino acid sequence surrounding the Thr210 contains a consensus motif for PDK1, which is found in many known PDK1 substrates (Fig. 3D), thus enhancing the possibility that PLK1 could be a potential substrate of PDK1. Indeed, cotransfection of PDK1 and PLK1 into 293T cells, followed by PLK1 immunoprecipitation, showed that PDK1 enhanced the phosphorylation of both the endogenous and exogenous PLK1, which was abolished when cells were treated with BX795 or BX912, but not the Aurora A inhibitor VX680 (Fig. 3E). This suggests that PDK1-induced PLK1 phosphorylation was unlikely to be an indirect effect of Aurora A, which might be coimmunoprecipitated with PLK1. Furthermore, in an in vitro kinase assay using endogenous PLK1 immunoprecipitated from DLD1 cells as a substrate, recombinant PDK1 added in the kinase assay induced PLK1 phosphorylation at T210, which was markedly reduced in cells treated with BX795 (Fig. 3F), indicating that the recombinant PDK1 can directly induce endogenous PLK1 phosphorylation in vitro. Importantly, in cells treated with VX680, where the PLK1 phosphorylation was greatly reduced as expected, recombinant PDK1 still boosted the PLK1 phosphorylation in the in vitro kinase assay (Fig. 3F). This further excludes the possibility that PDK1 may induce PLK1 phosphorylation indirectly through Aurora A kinase. Finally, in in vitro kinase assays using both PDK1 and PLK1 as recombinant proteins, we showed that PDK1 induced a strong PLK1 phosphorylation, which was blocked by BX795 and BX912, but not by VX680 (Fig. 3G). Collectively, these experiments provided evidence that PDK1 directly regulates PLK1 in human cancer cells.

We next investigated whether PLK1 directly regulates MYC. Through coimmunoprecipitation (co-IP) assay, we showed an interaction between both exogenous PLK1 and MYC in 293T cells (Fig. 4A). We also showed that the endogenous interaction between the two proteins occurs in HEK-PDK1 cells as well as in various cancer cell lines (Fig. 4B and C). We further showed that PLK1 kinase activity is required for MYC protein accumulation, as the wild-type PLK1, but not the kinase-dead mutant, induced strong MYC accumulation (Fig. 4D). Crucially, in vitro kinase assays using either recombinant PLK1 (Fig. 4E) or endogenous PLK1 pulled down from the cancer cells (Fig. 4F) showed a robust induction of S62 phosphorylation of recombinant MYC but not T58 phosphorylation, which was reduced in the presence of BI2536. Importantly, the endogenous PLK1 kinase activity toward MYC phosphorylation was strongly abolished in cells treated with BX795 (Fig. 4G) or in PDK1^−/− cells (Fig. 4H). Thus, these results not only showed a direct phosphorylation of MYC by PLK1, but also showed that PLK1 activity on MYC is crucially dependent on PDK1. Together, these data reiterate the operation of PDK1–PLK1–MYC signaling in cancer cells.

During culturing of these transformed cells, we noticed that HEK-PDK1 cells, and to a lesser extent HEK-MYC cells, displayed distinct morphologies from HEK-E545K cells and once they became confluent in culture, started to form semiattached three-dimensional (3D) clusters on the plate (Fig. 5A, top), suggesting that they displayed tumorigenic and stem cell–like properties. This feature, however, was not observed in E545K-transformed cells (Fig. 5A). Given the known role of MYC in inducing ESC- or CSC-like phenotypes in differentiated somatic cells (16), it raises a possibility that PDK1, which activates MYC, may have a similar capacity in inducing CSC-like behavior. This hypothesis was first tested using an in vitro assay for spheroid formation in serum-free suspension culture, a property associated with cancer stem/progenitor cells (22). We observed that PDK1- or MYC-transformed HEK or HMEC cells formed large and abundant nonadherent tumorspheres after 7 days of growth in suspension culture, whereas E545K-transformed cells were able to generate only a low number of small spheres (Fig. 5A, bottom and Supplementary Fig. S5A). These spheres were able to reform a monolayer when placed back to a tissue culture plate containing serum-rich medium (Fig. 5B). Furthermore, after dispersion into single cells, PDK1- or MYC-transformed HEK or HMEC cells reformed spheres with increasing enrichments for at least four passages (Fig. 5C and Supplementary Fig. S5B), indicating a gain of self-renewal capacity that resembles a stem cell–like property.

To show the tumor-initiating capacity of these transformed cells in vivo, we next injected these cells at different numbers into the flanks of BALB/c nude mice. Strikingly, 1,000 HEK-PDK1 cells were sufficient to generate tumors in all six mice as early as 2 weeks (Fig. 5D, left). HEK-MYC cells seemed to be less tumorigenic and required 10,000 cells to generate a similar size of tumors. In contrast, 10,000 HEK-E545K cells were unable to induce tumors in the mice (Fig. 5D). A further experiment showed that as few as 100 PDK1 cells were sufficient to give rise to xenograft tumors, whereas 3 × 10⁶ E545K cells were required to generate observable tumors by 28 days (Fig. 5D, right). Importantly, PDK1-associated primary xenograft tumors were self-renewable, as determined by the ability to form secondary and tertiary tumors using as few as 100 xenograft tumor cells (Fig. 5E). These in vitro and in vivo data showed a strong tumorigenicity of PDK1-transformed cells with self-renewal capacity.

We also tested the ability of PDK1 in inducing mouse embryonic fibroblast (MEF) reprogramming. To achieve this, we used p53-deficient MEFs, as immortalization by p53 inactivation has been shown to enhance MEF reprogramming efficiency (23, 24). Again, PDK1 but not E545K was also able to induce MYC activation, as well as tumorsphere formation in immortalized Trp53^−/− MEFs (Supplementary Fig. S5C and S5D). In the PDK1 sphere populations, we detected strongly increased expression of embryonic stem pluripotency factors SOX2 and OCT4 as assessed by both quantitative PCR (qPCR) and confocal imaging compared with the monolayer growth (Supplementary Fig. S5E and S5F). When these MEF-PDK1 cells were cultured in ESC medium containing the differentiation inhibitor LIF (leukemia inhibitory factor), MEF-PDK1 cells formed colonies resembling the ESC-like morphology and were alkaline phosphatase-positive (Supplementary Fig. S5G), though we found that these colonies were unable to maintain the ESC-like morphology in the subsequent passages, probably due to an incomplete reprogramming. Thus, in both human epithelial cells and MEFs, PDK1 is able to induce PLK1 and MYC activation and ESC-like properties.

Aberrant high PDK1 activity has been shown in invasive and metastatic breast tumor samples (25). To show the capacity of the PDK1–PLK1–MYC pathway in regulating CSCs, we used the highly invasive breast cancer MDA-MB-231 and SUM159PT cells that contain a high percentage of CD44⁺/CD24^−/low CSC-like cells (26). Intriguingly, PDK1–PLK1–MYC signaling was found to be enriched in CD44⁺/CD24^−/low cells compared with the non-CD44⁺/CD24^−/low cells (Fig. 5F). Knockdown of PDK1/PLK1 or treatment with their corresponding inhibitors BX795/BI2536 resulted in marked reduction of the CD44⁺/CD24^−/low population (Fig. 5G–I). In contrast, PI3K/AKT inhibitors GDC-0941 and MK-2206 were unable to do so (Fig. 5I). Corresponding to the reduced CD44⁺/CD24^−/low cells, PDK1 or PLK1 inhibition either by gene knockdown or inhibitor treatment resulted in marked inhibition of tumorsphere formation in MDA-MB-231 cells (Fig. 5J).

It is known that MYC is able to activate ESC-like transcriptional programs in adult epithelial cells, resulting in a CSC-like phenotype in the appropriate genetic context (16). To characterize the transcriptional program underlying the PDK1-induced CSC-like behavior, we compared the gene expression profiles in HEK-PDK1, -MYC, or -E545K cells. Significant analysis of microarray identified 1,750, 1,080, and 297 differentially expressed genes in these transformed cells when compared with nontransformed control cells, respectively (false discovery rate < 0.05; P < 0.01; Supplementary Tables S1–S3). Gene Venn Diagram analysis shows that HEK-PDK1 cells shared a robust transcriptional program with HEK-MYC cells, but had little overlap with HEK-E545K cells (Fig. 6A). In addition to the PDK1 and MYC common gene set, PDK1 also regulates a unique set of 784 genes. We further stratified the PDK1- or MYC-regulated genes into 889 upregulated and 1,151 downregulated genes via gene cluster analysis (Fig. 6B). Notably, a number of well-known genes implicated in ESC pluripotency or maintenance, including SOX2, LIN28B, SALL4, and EZH2 (27), or CSCs, including EPCAM, ALDH1A, and S100A4 (28), were upregulated in both HEK-PDK1 and HEK-MYC cells, but not in HEK-E454K cells (Fig. 6B). JAG2, which was recently shown to be a MYC target (29) with a role in modulating CSCs, was also markedly induced by PDK1 and MYC but not by E545K. In addition, a set of genes encoding secreted inhibitors of autocrine signaling, including DKK1, SFRP1, and BMP4, whose reduction has been recently shown by Scheel and colleagues (30) to enable self-renewal of epithelial cells, were strongly repressed in PDK1 and MYC cells but not in E545K cells. CD24, a negative selection marker for CSCs (31), was also selectively repressed in PDK1 and MYC cells. The array results of selected genes were further validated by quantitative real-time PCR (qRT-PCR; Fig. 6C) and Western blotting (Fig. 6D). Notably, many genes coregulated by PDK1 and MYC, including SOX2, EPCAM, JAG2, and S100A4, were more affected by PDK1 than MYC. In total, we identified 668 genes showing such a pattern (Fig. 6E and Supplementary Table S4), which is consistent with a more robust role of PDK1 than MYC in tumorigenesis.

We also investigated the changes of microRNAs (miRNA, miR) that are MYC-associated and implicated in ESC self-renewal. LIN28B is a known MYC target and is able to inhibit the biogenesis of the let-7 family miRNAs (32). Inhibition of let-7 miRNAs has been shown to enhance reprogramming of somatic cells to induced pluripotent stem (iPS) cells (33). MYC also transactivates the miR-17-92 cluster, which is also implicated in ESC maintenance (34). Consistent with MYC and LIN28B elevation, we detected a marked upregulation of miR-17-92 and downregulation of let-7s in PDK1 and MYC cells, but not in E545K cells (Fig. 6F). Because let-7 suppresses its own negative regulator LIN28B (33), it is likely that PDK1 enforces a feedback loop via MYC-LIN28B–mediated let-7 downregulation to support the self-renewal program. Finally, we showed that BI2536 treatment of HEK-PDK1 cells resulted in reduced expression of some ESC- or CSC-related genes, including EPCAM, SOX2, SALL4, and JAG2 (Fig. 6G), validating a role of PLK1 in the PDK1-mediated CSC gene signature. These findings indicate that PDK1 is able to evoke multiple transcriptional programs that coordinately induced a remarkable reprogramming toward a state resembling CSCs. In this process, MYC is one important factor but not the only one that modulates the reprogramming.

Aberrant gene expression associated with embryonic stem cell identity, including ESC genes, MYC targets, and Polycomb targets, has been found in poorly differentiated tumors (14–16). By further interrogating several previously published datasets collectively, we showed that the above ESC-related genes were significantly enriched in the PDK1-dependent transcriptome, including upregulation of 97 ESC-expressed genes and downregulaton of 182 Polyccomb targets (PRC genes; Supplementary Fig. S6A and Supplementary Table S5). In contrast, the E545K-associated transcriptional program displayed a distinct gene set that is not significantly associated with ESCs (Supplementary Fig. S6A).

We next determined whether PDK1-driven ESC-like gene expression is of clinical relevance to human malignancy. Gene set enrichment analysis (GSEA; ref. 35) of several previously published datasets showed that PDK1-induced ESC-like genes were found to be significantly enriched in colon and lung tumor samples as compared with the normal controls, whereas the Polycomb targets were inversely correlated in these samples (Supplementary Fig. S6B). Moreover, in breast cancer, deregulation of these genes was significantly correlated with the high-grade tumors as compared with the low-grade tumors (Supplementary Fig. S6C). This indicates that aberrant expression of these PDK1-regulated ESC genes is associated with malignant progression from normal to aggressive tumors. In addition, we also showed that the PDK1-regulated ESC-like gene signature was associated with poor disease outcome as shown in the survival analysis of breast and lung cancer cohorts (Supplementary Fig. S6D), providing a prognostic value of these genes. Together, these findings show that the PDK1-activated ESC-like gene signature we identified from the in vitro culture system is clinically relevant to human cancers arising in distinct tissues and support a link of PDK1–MYC signaling to aggressive cancer behavior.

We have previously shown that mTOR inhibition by rapamycin or mTOR/Raptor knockdown induces MYC accumulation in colorectal cancer, which can be inhibited by PDK1 inhibition, resulting in rapamycin sensitization (5). As our data now indicate that PLK1 is required for PDK1–MYC signaling, together with our further observation that PLK1 is highly expressed in colorectal cancer tumors compared with adjunct normal regions (Supplementary Fig. S7A and S7B), we hypothesized that the PLK1 inhibitor could also sensitize colorectal cancer cells to mTOR inhibitors through abolishing mTOR inhibitor–induced MYC activation. Classical mTOR inhibitors such as rapalogs are known to induce compensatory feedback activation of PI3K–AKT due to S6K inhibition. BEZ235, a dual PI3K–mTOR kinase inhibitor, is able to overcome the feedback AKT activation and is currently being tested in several clinical trials either as a single agent or in combination with other therapeutics. Unlike rapamycin treatment, which induced both AKT and MYC activation in colorectal cancer cells, BEZ235 did not induce AKT activation but retained the ability to induce MYC (Fig. 7A). Of note, neither drug induced ERK activation in colorectal cancer, which is, however, often seen in breast cancer cells (36). As expected, BI2536 cotreatment effectively removed BEZ235-induced MYC induction (Fig. 7B). In these cells, BI2536 or BEZ235 alone failed to induce significant apoptosis, but their combination, which resulted in inhibition of both MYC and p-4EBP1, induced massive apoptosis, as evidenced by strong detection of PARP cleavage (Fig. 7B), cells in sub-G₁ (Fig. 7C), and caspase-3 activation (Supplementary Fig. S7C). The combinatorial effect was synergistic, as shown by combination index analysis (Supplementary Fig. S7D) and further confirmed by time-course analysis of cell viability (Fig. 7D) and long-term colony formation assay (Supplementary Fig. S7E). Finally, to assess the potential of the combination strategy in vivo, SW480 and HT15 cells were injected subcutaneously into nude mice to establish tumor xenografts. We showed that BEZ235 also induced MYC accumulation in the xenograft tumors, which can be inhibited via combination with BI2536 (Supplementary Fig. S7F). Accordingly, the combination treatment induced synergistic tumor growth inhibition compared with the single-agent treatment, validating the in vitro findings (Fig. 7E).

Like BEZ235, a specific mTOR inhibitor, PP242 (37), also generated similar results on MYC, p-4EBP1, and apoptosis when combined with BI2536 (Supplementary Fig. S8A and S8B). In contrast, BI2536, though it also blocked rapamycin-induced Myc accumulation (Supplementary Fig. S8C), did not enhance apoptosis (Supplementary Fig. S8D), but only potentiated the antiproliferation effect and xenograft tumor growth inhibition (Supplementary Fig. S8E and S8F). This is probably due to the inability of rapamycin to block 4EBP1 phosphorylation (Supplementary Fig. S8C) as previously shown. 4EBP1, but not S6K, has been recently shown to be the key effector of the mTOR pathway responsible for cell proliferation and survival (38), and additional inhibition of 4EBP1 phosphorylation seems to be required for apoptosis induction in response to the AKT inhibitor (39). Thus, the simultaneous inhibition of both MYC and 4EBP1 phosphorylation upon combination of BI2536 and BEZ235 seemed to be crucial for apoptosis induction of colorectal cancer cells. Taking all the data together, we conclude that the combination of BI2536 and PI3K–mTOR dual inhibitors such as BEZ235 may represent a promising treatment strategy for colorectal cancer.

Although PI3K–AKT signaling has been considered to be the main signaling pathway associated with PDK1 in oncogenesis, our study uncovers another arm of signaling that routes to PLK1–MYC to confer malignant phenotypes. Importantly, the pathway we identified using a chemical genetic approach with a PDK1-transformed cell line has been validated to be relevant in human cancers, as shown in multiple cancer cell lines derived from various tissue types. Although in our system we detected AKT phosphorylation at T308 by PDK1, we did not see AKT phosphorylation at S473 that is required for a full AKT activation (6). This is in contrast to PI3K-transformed cells in which both AKT phosphorylations are strongly induced. This observation is consistent with a recent report showing that PDK1-deficiency in colon cancer cells has only a modest effect on p-AKT (T308) and has no effect on p-AKT (S473; ref. 19). Our data thus indicate that PDK1 signaling might be wired differentially in certain oncogenic contexts to confer a growth advantage that becomes less dependent on AKT. Indeed, previous reports have shown that PDK1 is not linked to PI3K signaling in a PTEN-deficient tumor model (8) and can route through the AKT-independent pathway for cell survival in some cancer cell lines (10). As PDK1 has attracted much attention as a potential therapeutic target in cancer, we propose that MYC can be an alternative pharmacodynamic marker for the evaluation of small-molecule PDK1 inhibitors under preclinical and clinical development.

An intriguing finding of this study is the identification of the crucial role of PDK1–PLK1–MYC signaling for cancer cell survival. We provide evidence that PDK1 induces PLK1 phosphorylation and PLK1 binds to and induces MYC phosphorylation and protein accumulation widely in cancer cells. A previous report has shown that PLK1-induced MYC phosphorylation is required for SCF^βTrCP-mediated MYC protein stabilization during late-stage cell-cycle progression (40). We further show here that PLK1 can directly bind to MYC. Regardless of whether or not PDK1–PLK1 signaling regulates MYC stability through a similar or distinct mechanism, the direct regulation of MYC by PDK1–PLK1 signaling immediately suggested a therapeutic approach targeting MYC-driven tumors. Indeed, our data show a preferential killing of small-molecule inhibitors of PDK1 or PLK1 in MYC-dependent breast cancer cells compared with MYC-independent breast cancer cells. Given that a clinical inhibitor of MYC is not available, small-molecule inhibitors such as BI2536, which is currently in late-stage clinical trials, may provide an alternative anti-MYC strategy. Given that PLK1 is often found to be overexpressed in human cancers (41), therapeutic targeting of PLK1 may yield a more favorable therapeutic index in MYC-associated tumors.

The main characteristic of the PDK1-induced transformation is that it is able to induce both the genotype and phenotype of CSCs that have been proposed to account for tumor initiation, progression, and chemoresistance (13, 31). We show that as few as 500 PDK1-transformed cells can induce robust tumorigenicity and that PDK1 activates clinically relevant transcriptional programs associated with poor disease outcome. In addition, PDK1 or PLK1 inhibition also resulted in disruption of both embryonic and adult stem cell self-renewal while inducing differentiation. Activation of an ESC-like signature and an ESC-like phenotype in differentiated somatic cells also indicates that the embryonic stem program can be reactivated during the course of tumor progression and is not necessarily inherited from a stem cell-of-origin. This notion is consistent with a recent report from Chaffer and colleagues (42) showing spontaneous CSC generation from mammary epithelial cells.

Furthermore, consistent with a previous report that PDK1 is hyperactivated in invasive and metastatic breast cancer (25), we show that PDK1 or PLK1 inhibition in highly invasive breast cancer MDA-MB-231 cells resulted in depletion of CSC-like CD44⁺/CD24^−/low populations and accordingly strongly reduced tumorsphere formation, whereas PI3K–AKT inhibition did not have such effects. Thus, small-molecule inhibition of PDK1–PLK1–MYC signaling for elimination of CSCs may provide a targeted therapy to overcome recurrence of aggressive breast tumors following chemotherapy.

An additional therapeutic application of our studies is the identification of strategies to overcome resistance to mTOR-targeted therapy in colorectal cancer. Drug resistance and tumor recurrence is the main cause of patient relapse, possibly owing to recurrence of CSCs. We have previously discovered that mTOR inhibition induces MYC activation, a compensatory effect mitigating the antiproliferative effect of mTOR inhibitors in colorectal cancer (5). We now show that PLK1 inhibitor blocks mTOR inhibitor–induced MYC activation, providing a rational approach to developing a new combination therapy for colorectal cancer. Specifically, a low dose of PLK1 inhibitor BI2536 plus PI3K–mTOR dual inhibitor BEZ235 induced massive apoptosis in colorectal cancer cells and a synergistic loss of colony formation, suggesting that this strategy might be useful in colorectal cancer. Given that both drugs are in late-stage clinical trials, we hope that our findings will spur clinical trials in colorectal cancer for the combination of BEZ235 and BI2536 to improve the therapeutic outcome.

Human full-length PDK1, MYC, PIK3CA-E545K, and PLK1 were cloned into PMN–IRES–GFP retroviral vector and introduced into human epithelial cells and MEFs. All kinase inhibitors used in this study were obtained from Axon Medchem. Information regarding plasmid DNA vectors and stable cell line construction is provided in the Supplementary Materials and Methods.

Cell cultures and various cellular assays are described in the Supplementary Materials and Methods. All cancer cell lines were purchased from American Type Culture Collection, and no authentication of cell lines was done by the authors.

All the experiments in xenografts are described in the Supplementary Materials and Methods.

Details are described in the Supplementary Materials and Methods.

The microarray hybridization was conducted using the Illumina Gene Expression Sentrix BeadChip HumanHT-12_V4 (Illumina), and the data were analyzed using the GeneSpring GX 11.0.2 (Agilent Technologies). Detailed information can be found in the Supplementary Materials and Methods. Primers used in RT-PCR analysis are described in Supplementary Table S6.

PDK1-regulated ESC-like and PRC gene signature definition is described in the Supplementary Experimental Procedures. GSEA (35) was conducted to assess the degree of correlation between PDK1-regulated gene signatures and cancer phenotypes on different human patients. Survival curves were calculated using the Kaplan–Meier survival analyses and the quantiles-rank test. Detailed statistical analysis is included in the Supplementary Data. Data are presented as mean ± SEM, unless otherwise stated. A Student t test was used to compare two groups for statistical significance analysis.

The microarray data are deposited into the Gene Expression Omnibus with the accession number GSE30669.

No potential conflicts of interest were disclosed.

Conception and design: J. Tan, Q. Yu

Development of methodology: J. Tan, P.L. Lee, P. Guan, M. Feng

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Z. Li, P. Guan, S.T. Lee, Z.N. Wee, Y.C. Lim, Q. Yu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Tan, P. Guan, M. Feng, C.Z. Lim, Y.C. Lim, R.K.M. Karuturi

Writing, review, and/or revision of the manuscript: J. Tan, Q. Yu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P.L. Lee, P. Guan, M.Y. Aau, S.T. Lee, C.Z. Lim, E.Y.J. Lee, Z.N. Wee

Study supervision: Q. Yu

The authors thank Dr. William C. Hahn for the HEK-TERV cells. The authors also thank Dr. Fu Zheng for the human PLK1 plasmids and Dr. Luca Primo for the PDK1 kinase-dead mutant construct.

This work was supported by the Agency for Science, Technology and Research of Singapore (A*STAR).