PDK1 Attenuation Fails to Prevent Tumor Formation in PTEN-Deficient Transgenic Mouse Models

Abnormal regulation of the PI3K/AKT signaling pathway is one of the most frequent epidemiological observations in human malignancy. Loss of the tumor suppressor PTEN (1, 2), mutations in receptor tyrosine kinases, as well as activating PI3K mutations all result in elevated levels of PIP₃ that activates the AKT-survival pathway (3). PIP₃ causes PDK1 and AKT to colocalize at the plasma membrane via their pleckstrin homology domains (4, 5). Here, PDK1 phosphorylates the kinase T-loop residue AKT^T308, an event that is absent in PDK1 knockout embryonic stem cells (6) and tissue-specific PDK1 ablation mouse models (7–9). Full activation of AKT also requires phosphorylation of AKT^S473 by mTORC2 (10). The role of PDK1 in phosphorylating the T-loop of numerous AGC kinases makes it an attractive target for anticancer therapy (11). Several proteins phosphorylated by PDK1, including AKT, S6K, RSK, and certain PKC isoforms, are themselves kinase drug targets (12–15).

Several strategies have been used to evaluate the therapeutic potential of inhibiting PDK1 function in cancer. These include in vitro and in vivo pharmacological and genetic approaches. Kinase selectivity represents a key challenge when using pharmacological inhibitors to study the involvement of a particular kinase in a cellular process (16, 17). For example, the indolinone-based PDK1 inhibitor BX-795 and its analogues inhibit cell cycle progression, induce apoptosis, and show efficacy in a xenograft model (18). However, BX-795 causes G2/M cell cycle arrest in both PDK1 wild-type (WT) and PDK1 knockout cells suggesting that the effects on cell cycle are unrelated to PDK1 inhibition (19). At present, analysis of small-molecule PDK1 inhibitors, covering a dozen structural classes (11), has not pinpointed any selective compound suitable for in vivo efficacy studies.

In contrast to pharmacological agents, genetic approaches are valued for the inherent specificity by which point mutations, gene deletion, or knockdown perturb protein function. In particular, manipulation of the mouse genome has been utilized for modeling of in vivo tumorigenesis and for drug target validation (20). For example, PTEN deficiency in mice faithfully models many aspects of human tumor development. While the constitutive PTEN knockout (PTEN^−/−) is embryonic lethal, heterozygous PTEN^+/− mice spontaneously develop a broad spectrum of tumors consistent with epidemiological observations in humans (21–23). In addition, the Cre-loxP technology, which enables conditional gene modulation (24), has allowed for the homozygous PTEN deletion in adult mice circumventing embryonic lethality. The loss of both alleles of PTEN results in a more rapid and reproducible onset of tumorigenesis compared to the constitutive PTEN^+/− model as exemplified in the PTEN^fl/flPBcre prostate cancer model (25). Importantly, some kinase inhibitors show antitumor activity in mouse models of PI3K/AKT pathway activation, highlighting the utility of engineered mice as a platform to evaluate response to targeted therapies and/or genetic perturbations (20, 26).

It was reported that PDK1^−/fl hypomorphic mice, which express low levels of PDK1, when crossed to PTEN^+/− mice, suppress PTEN-driven tumorigenesis (27). This observation has put PDK1 to the forefront as an anticancer target in context of PTEN deficiency (11). However, the constitutive knockdown of PDK1 impairs mouse development (27, 28), and therefore does not ascertain whether acute knockdown of PDK1 in adult mice, a situation mimicking more closely therapeutic intervention, will be efficacious.

To address this question, inducible, in vivo RNAi-mediated knockdown represents a promising technology for modulation of gene expression in adult mice (29, 30). An inducible gene silencing approach is particularly suited for functional studies of genes like PDK1, where constitutive models lead to strong developmental phenotypes or embryonic lethality (6, 27). Several methods for in vivo short hairpin (sh)RNA expression have been described (31–34). Here, we report on an RNAi mouse line where the Rosa26 locus has been specifically targeted to achieve reversible, doxycycline-dependent expression of a shRNA targeting mouse PDK1 (Rosa26^PDK1-RNAi). To genetically assess target efficacy the PDK1-RNAi mice were crossed with 3 distinct PTEN-deficient tumor models. We observed potent knockdown of PDK1, but found no prevention of tumor development or efficient pathway modulation. This result is further supported by pharmacological PDK1 inhibitor studies in primary ex vivo cultures. In contrast to AKT inhibitor treatment, small-molecule PDK1 inhibitors were not capable of completely reversing elevated phospho-AKT^T308 levels caused by PTEN deficiency, challenging a pharmacological intervention potential.

Rosa26^PDK1-RNAi and Rosa26^PDK1-RNAi(2) mice were generated by targeting the Rosa26 locus (30). Targeted ES cell lines were treated with doxycycline for 2 days (1 g/kg), followed by PDK1-specific reverse transcriptase PCR (RT-PCR) to identify shRNA constructs with efficient knockdown (shPDK1: 5′-GGCTAGAGATCTTGTGGAA-3′; shPDK1(2): 5′-GAATTTGCACCAGCAGACA-3′). PDK1⁻ and PDK1^fl alleles (mixed genetic background) were obtained from Dario Alessi (University of Dundee), rederived with C57BL/6NTac and then intercrossed to generate hypomorphic PDK1^−/fl mice. Out of 103 offspring, 2 hypomorphic PDK1^−/fl mice were obtained (2%) versus the expected 25%. PTEN^+/− mice were generated by intercrossing PTEN^−/fl mice with a ubiquitous Cre-deleter strain [C57BL/6NTacGt(ROSA)26Sortm16(cre)Arte]. PTEN^fl and PB-cre lines were obtained from Hong Wu (25) and Rosa26^creERT2 mice from Taconic (30). Primary mammary epithelial cells were prepared from virgin female mice as described (35). All studies were approved by the Merck Institutional Animal Care and Use Committee (IACUC).

Blood glucose was assayed with a glucometer (Lifescan). Plasma insulin was determined by ELISA (ALPCO immunoassays). Glucose tolerance tests (GTT) were performed on overnight fasted mice. d-glucose was injected intraperitoneally (2 g/kg). Insulin tolerance tests (ITT) were performed on fasted (4h) animals by intraperitoneal injection of human insulin (0.75 U/kg; Eli Lilly).

Tissue and cell homogenates were prepared using radioimmunoprecipitation assay buffer (Pierce) plus protease and phosphatase inhibitor cocktails (Roche). Immunoblots were probed with antibodies against: PDK1, p-PDK1^S241, AKT, p-AKT^T308, p-AKT^S473, S6K1, p-S6K1^T389, p-S6K1^S229, S6, p-S6^S235/236, p-ERK1/2^T202/204, and β-actin (Cell Signaling Technologies); p-RSK^S221 (Invitrogen); and p-70S6K^T229 (Abcam). The specificity of the PDK1 antibodies is evaluated in Figure S1. Fluorescence activated cell-sorting (FACS) analysis was conducted as described previously (36).

Tumors were fixed in formalin and embedded in paraffin blocks. Sections were hematoxylin and eosin (H&E) stained (Leica ST5020 multistainer). Ki67-positive cells (clone SP6; 1:2,000; Thermo Scientific) were scored using an Ariol automatic scan system (Applied Imaging). All histopathologic analyses were performed at the Rodent Pathology Core, Harvard Medical School. Incidence of hyperplasia in lymph nodes, thyroids, adrenal glands, and uteri were scored on blinded samples (Table 1).

To study PDK1 function in adult mice, an inducible approach to modify PDK1 expression was chosen as traditional knockout mice (PDK1^−/−) are embryonic lethal (6) and PDK1^−/fl hypomorphic mice, which display constitutively low levels of PDK1, show strong developmental phenotypes including a 30% to 40% reduction in body size (27). For these reasons, and to more closely mimic a cancer treatment setting, we pursued the approach that affords reversible and temporal control of doxycycline-induced shRNA expression (30). The Rosa26 locus in mouse ES cells was gene-targeted to encode a single construct containing a PDK1-shRNA-driven by the tet-inducible H1 promoter and tetR-driven by the CAGGS promoter (Fig. 1A). The tetR regulatory cassette is flanked by loxP sites, which in addition to doxycycline-regulated RNAi, allows for Cre-recombinase–mediated expression of the shRNA in vivo in an irreversible manner.

Two targeted mouse lines (Rosa26^PDK-RNAi/+ and Rosa26^{PDK-RNAi(2)/+}) expressing nonoverlapping shRNAs [shPDK1 or shPDK1(2)] were generated with these ES cells and analyzed at 5 weeks of age to evaluate knockdown efficiency following 10 days of doxycycline treatment (Fig. 1B and C). Mice expressing a luciferase targeting shRNA (shFluc) were used as a control (30). In response to doxycycline, efficient knockdown was observed at both mRNA and protein level. In the absence of doxycycline, 9 out of 10 tissues had normal PDK1 mRNA levels compared to the shFluc controls revealing tight tetR-mediated control of gene silencing. Only brain showed a significant reduction in PDK1 expression in the absence of doxycyclin suggesting incomplete tetR-mediated suppression of the shRNA in this organ. In summary, both PDK1-RNAi mouse lines show robust drug-inducible knockdown of PDK1 in all tissues tested.

Next, we tested whether the acute knockdown of PDK1 was sufficient to have physiological consequences. Because the PI3K pathway plays an integral part in glucose homeostasis we characterized the PDK1-RNAi mice for a diabetic phenotype. Female mice carrying the PDK1 knockdown allele (Rosa26^PDK1-RNAi/+) or WT littermates were administered doxycycline for 30 days and blood glucose and plasma insulin levels were monitored as well as PDK1 knockdown levels in peripheral blood mononuclear cells (PBMC) (Fig. 2A).

Of note, we tested several anti-PDK1 antibodies by Western blotting and observed that they often depicted, in a tissue-dependent manner, nonspecific signals close to the molecular weight of PDK1 complicating evaluation of knockdown efficiency (Fig. S1). By contrast, the anti-phospho-PDK1^S241 antibody was highly specific and since this phosphorylation event is absolutely required for PDK1 activity (15) it was routinely employed to evaluate knockdown efficiency (Fig. 2B). The use of phospho-PDK1^S241 to evaluate PDK1 knockdown levels was further supported by quantitative mass-spectrometric analysis that identified greater than 99% phospho-occupancy at PDK1^S241 (data not shown).

PDK1 protein levels were rapidly downregulated in PBMCs after feeding doxycycline diet and reversed within 5 weeks after withdrawal of the inducer (Fig. 2B). Similar kinetics for target knockdown have been reported for the insulin receptor RNAi model (30) and is consistent with reports that saturation of a system with doxycycline is an intrinsically faster process than depletion (37, 38). Notably, systemic PDK1 knockdown in vivo resulted in hyperinsulinemia as well as hyperglycemia (Fig. 2C and D). Thus, GTT and ITT were performed to further investigate a potential diabetes phenotype. Mice with PDK1 knockdown exhibited considerably higher fasting blood glucose levels and impaired ability to clear glucose during intraperitoneal GTT (Fig. 2D and E). Moreover, Rosa26^PDK1-RNAi/+ mice had decreased insulin sensitivity compared to WT littermates during ITT (Fig. 2F). The efficiency of PDK1 knockdown in peripheral insulin-responsive tissues (fat, liver, and muscle) was analyzed and found to be greater than 90% in each tissue (Fig. 1C and data not shown). Subsequently, we removed doxycycline from the diet to test whether the diabetic phenotype would be reversible. Similar to the recovery of PDK1 protein levels (Fig. 2B) fasted blood glucose returned to normal levels within 6 to 7 weeks after doxycycline withdrawal (Fig. 2D). Altogether, we conclude that Rosa26^PDK1-RNAi/+ mice enable an efficient and reversible knockdown of PDK1 in adult, mature tissues, sufficient to impact physiological functions.

To better understand the role PDK1 inhibition could play in preventing PI3K-pathway-dependent tumorigenesis, we crossed both PDK1-RNAi mouse lines with a conditional PTEN-deficient prostate tumor model (Fig. 3A). It is well established that the PTEN^fl/flPB-cre mice develop prostatic intraepithelial neoplasia shortly after probasin cre-mediated PTEN deletion (i.e., 4 weeks of age) and that invasive adenocarcinomas can be seen by 9 to 12 weeks of age (25). To model systemic PDK1 inhibition (analogous to a small-molecule kinase inhibitor) the mice were treated with doxycycline starting at 4 weeks of age. Dissection at 14 to 17 weeks revealed that anterior prostate lobes were visibly enlarged and histological analysis showed massive hyperplasia in the PTEN-deficient prostates irrespective of PDK1 knockdown (Fig. 3B and C). PDK1 deficiency limited the weight increase of the anterior prostate to a small extent, however a similar weight reduction was also found in the nontumorigenic PTEN-WT context (Fig. 3C). Thus, it is unlikely antitumor efficacy but rather due to PDK1's role in cell size regulation (28).

Strikingly, both RNAi models (PTEN^fl/flPB-creRosa26^PDK1-RNAi/+ and PTEN^fl/flPB-creRosa26^{PDK1-RNAi(2)/+}) could not prevent the formation of microinvasive tumors (Fig. 3B). Consistent with the observed hyperplasia in PTEN-deficient tissue, the number of Ki67-positive cells was increased compared to PTEN-WT prostate (Fig. 3B and D). PDK1 knockdown had no significant effect on tumor cell proliferation based on this marker (Fig. 3B and D).

Next we evaluated whether extended PDK1 inhibition (>10 months) could achieve antitumor efficacy during the final stages of prostate tumor progression. Mice were analyzed at an age of 47 ± 3 weeks for PTEN^fl/flPB-cre and 45 ± 4 weeks for PTEN^fl/flPB-crePDK1-RNAi, respectively. Irrespective of PDK1 knockdown, tumors further progressed resulting in enlarged anterior prostates compared to younger animals (Fig. 3C and F). Histological examination confirmed no difference in tumor stage and invasiveness despite potent and long-term PDK1 knockdown (Fig. 3F).

Both PDK1-RNAi models achieved efficient PDK1 knockdown in the PTEN-WT (PTEN^fl/fl) and PTEN-deficient (PTEN^fl/flPB-cre) context (Fig. 3E). Surprisingly, long-term knockdown of PDK1 had no measurable effect on AKT^T308 phosphorylation in the PTEN-deficient context but impacted modestly phosphorylation of p90RSK^S221 and S6^S235/236 (Fig. 3E). Notably, we observe 2 feedback mechanisms. First, p-ERK1/2^T202/Y204 levels increase in response to PDK1 knockdown (Fig. 3E) possibly due to inhibition of RSK. A similar effect was observed in vitro with BI-D1870, a selective pharmacological inhibitor of p90RSK (39). Second, total AKT protein expression is downregulated in response to PTEN deletion suggesting a compensatory mechanism to the observed activation/hyperphosphorylation of AKT (Figs. 3E, 4D, and 5). In conclusion, PDK1 knockdown (>90%) is not able to prevent tumor formation in PTEN-deficient prostate consistent with the lack of strong PI3K/AKT-pathway modulation.

PTEN deletion occurs with high frequency in many human cancers, including prostate, brain, breast, and leukemia (1). To model the leukemic event, we generated a conditional mouse model with the capacity for inducible PTEN deficiency in hematopoietic stem cells by combining the PTEN^fl allele (25) with the tamoxifen controllable Rosa26^creER allele (ref. 30; Fig. 4A). These mice (PTEN^fl/flRosa26^creER/+) developed typical characteristics of leukemia [Acute lymphoblastic leukemia (ALL), acute myelogenous leukemia (AML)] after tamoxifen-induced PTEN deletion. At 5 weeks post Cre-induction, mice started to show significant weight loss, became hunched, and developed a scruffy coat in comparison to PTEN^fl/fl controls (Fig. 4A). These phenotypes are similar to a previously described interferon-driven inducible PTEN deficiency model (40, 41). PTEN^fl/flRosa26^creER/+ mice displayed consistent changes in the composition of white blood cell fractions detectable as early as 2 weeks post PTEN deletion. The B220⁺ B cell fraction was reduced while the fraction of CD11b⁺ cells was strongly increased (data not shown). The extent of these changes however did not correlate with disease progression or length of survival in the study. Thus, we conclude that these are systemic changes linked to PTEN deficiency, not reflecting a leukemic disease.

Disease progression was fast and overall body condition deteriorated such that more than 90% of animals needed to be sacrificed at 5 to 12 weeks post PTEN deletion (Fig. 4B). At sacrifice, most PTEN^fl/flRosa26^creER/+ animals presented with enlarged lymph nodes and spleens (Fig. 4A and C). Leukemic disease was heterogeneous: some mice had a transplantable T-ALL with an outgrowth of CD4⁺CD8⁺ or CD4⁺ T cells in the periphery, while others had an outgrowth of CD11b⁺ cells in lymphoid organs and bone marrow (Fig. S2).

Next, we investigated whether a strong reduction of PDK1 expression in the PTEN-deficient leukemia model could effectively modulate activity of the PI3K pathway and prevent or slow disease onset. PTEN^fl/flRosa26^{PDK1-RNAi/cre-ER} mice were fed doxycycline diet 4 days prior to the induced PTEN deletion to ensure effective PDK1 knockdown before heightened PI3K pathway activity would occur. Interestingly, these mice all developed disease with comparable characteristics as their littermates without the PDK-RNAi allele (Fig. 4C and data not shown). At sacrifice, greater than 90% PDK1 knockdown was observed in the PTEN^fl/flRosa26^{PDK1-RNAi/creER} diseased animals (Fig. 4D). PDK1 knockdown had a small effect on spleen weights irrespective of PTEN status (Fig. 4C). However, despite PDK1 knockdown most PTEN^fl/flRosa26^{PDK1-RNAi/creER} mice still displayed significantly enlarged spleens when compared to control mice (Fig. 4C). Disease onset and progression was not significantly delayed by PDK1 knockdown (Fig. 4B). Median survival was 65 days for PTEN^fl/flRosa26^{PDK1-RNAi/creER} mice versus 57 days in PTEN^fl/flRosa26^+/creER littermates.

Following PTEN deletion, PI3K-pathway activation was detectable in both peripheral white blood cells (Fig. S3) and the spleen from PTEN^fl/flRosa26^creER/+ mice (Fig. 4D). However, while knockdown of PDK1 in PTEN-WT context reduced phosphorylation of AKT^T308, p70S6K^T389, and S6^S235/236, PDK1 knockdown in PTEN-deficient context (PTEN^fl/flRosa26^{PDK1-RNAi/creER}), did not lower p-AKT^T308 or p-S6^S235/236 levels, which remained significantly above the basal levels seen in PTEN-WT controls (Fig. 4D).

We conclude that similar to the prostate tumor formation, inducible PDK1 knockdown (>90%) in adult mice is not sufficient to prevent development of leukemia following PTEN deletion.

One possible explanation for the missing p-AKT^T308 downregulation despite more than 90% PDK1 knockdown is that the remaining PDK1 protein is still sufficient to perform the catalysis. To test this hypothesis, we isolated mammary epithelial cells from PTEN-WT (PTEN^fl/fl) and PTEN-deficient (PTEN^fl/flRosa26^+/cre-ER) mice with or without the Rosa26^PDK1-RNAi/+ allele and treated them ex vivo with 2 structurally diverse and cell potent PDK1 inhibitors (Fig. S4). Both inhibitors display single digit nanomolar potency (EC₅₀) against purified PDK1 enzyme. To maximize inhibition of the residual PDK1 activity, we used high compound concentrations (10 μmol/L) only achievable in an in vitro setting. As controls we used an allosteric and highly selective AKT inhibitor (42) and the mTOR inhibitor rapamycin.

At the time of cell harvest, mammary glands had not developed tumors in the PTEN-deficient context but appeared hyperplastic. As shown in Figure 5, the PI3K pathway was strongly activated by PTEN loss (indicated by p-AKT and p-p70S6K) whereas the phosphorylation of the mitogen activated protein kinase (MAPK) pathway components p90RSK and extracellular signal regulated kinase (ERK) was unchanged (lanes 1 vs. 3). AKT inhibitor and rapamycin treatment resulted in potent downregulation of p-AKT (lanes 1 vs. 9 and 3 vs. 11) and p-p70S6K levels (lanes 1 vs. 5 and 3 vs. 7), respectively. Similar to our in vivo findings, knockdown of PDK1 by more than 90% was not able to counteract the elevated phosphorylation of AKT^T308 caused by PTEN-deletion (lanes 3 vs. 4). In contrast, PDK1-knockdown decreased phosphorylation of p90RSK^S221 about 2-fold across all treatment groups.

For each of the 4 genotypes, addition of pharmacological PDK1 inhibitors reduced the phosphorylation levels of the PDK1 substrates p70S6K, p90RSK, and AKT (Fig. 5, lanes 13–16 and 17–20) compared to dimethyl sulfoxide (DMSO) controls (Fig. 5, lanes 1–4). Moreover, consistent with a direct pharmacological inhibition of PDK1 kinase activity, only p-AKT^T308 was reduced whereas p-AKT^S473 was not altered in the primary mammary gland cultures. PDK1 inhibitor treatment did not alter p-PDK1^S241 or total PDK1 protein levels. In contrast to pharmacological PDK1 inhibition, rapamycin and the AKT inhibitor potently inhibited the phosphorylation of S6 and AKT phosphosites, respectively (Fig. 5). Although a modest inhibition of p-AKT^T308 was observed ex vivo by the 2 PDK1 inhibitors, p-AKT^T308 was still elevated when compared to PTEN-WT controls (lane 15 vs. 1 and 19 vs. 1). Combined PDK1 knockdown and pharmacological inhibition (lanes 14, 16, 18, 20) showed a further inhibitory effect on p-AKT^T308 phosphorylation, however, the effects remained small compared to direct AKT inhibition (lanes 9–12). An additive effect of combined knockdown and pharmacological inhibition was also observed for other pathway nodes (p90RSK^S221 and pS6^S235/236 phosphorylation).

In summary, compared to direct AKT inhibition using an AKT1/2 specific inhibitor, combined genetic and pharmacological inhibition of PDK1 still remains inferior as a means to inhibit AKT activity.

Thus far, in 2 models of inducible homozygous PTEN loss, efficient PDK1 knockdown did not prevent tumor formation. By contrast, PDK1^fl/− hypomorphic mice, which display a constitutive reduction of PDK1 enzymatic activity by about 80% to 90% in all tissues, have been reported to prevent tumorigenesis in a heterozygous PTEN^+/− context (27).

First, to ensure that this discrepancy was not caused by significant differences in the remaining PDK1 protein levels between RNAi and hypomorphic knockdown models, we compared tissues from these models for PDK1 expression by Western analysis (Fig. 6A). Mice with a single hypomorphic allele (PDK1^fl/+) showed around 45% reduction in PDK1 expression as previously reported (27). Stronger reduction of PDK1 levels (>80%) was observed in the PDK1^fl/fl and PDK1^fl/− hypomorphic mice and these low levels were comparable to the RNAi-mediated PDK1 knockdown observed in prostate, T cells, mammary epithelial organoids, and liver (Fig. 6A). Thus, we conclude that the reduction in PDK1 protein is comparable between the 2 model systems.

Next, to test whether the difference in efficacy outcome could be explained by homozygous versus heterozygous PTEN deficiency, we combined the inducible Rosa26^PDK1-RNAi allele with heterozygous PTEN^+/− mice. Similar to the approach of Bayascas and colleagues, we tested whether PDK1 knockdown would have efficacy on tumor burden in multiple organs (Fig. 6B). Taking advantage of the inducible knockdown system, adult mice were continuously fed either doxycycline or control diet beginning at 6 weeks of age, which circumvents potential developmental issues of constitutive knockdown models. All animals were sacrificed within 1 week, at an age of 6 to 7 months. This was the stage when visible lymphomas became apparent in the PTEN^+/− background. Thyroid, uterus, lymph nodes, and adrenal glands were harvested and examined in a blinded comparative histopathologic analysis (Table 1). Western blot analysis confirmed that efficient PDK1 knockdown (>80%) was achieved in all tissues analyzed (Fig. 6C).

Tissue analysis of our PTEN^+/− mice showed phenotypes similar to previously reported PTEN^+/− heterozygous models, where animals develop a variety of tumors in multiple organs including the lymph nodes, mammary gland, uterus, prostate, adrenal glands, and thyroid (21–23). Strikingly, histopathologic examinations revealed that neoplastic lesions were present at high incidence in both PTEN^+/− and in PTEN^+/−Rosa26^PDK1-RNAi/+ animals, irrespective of PDK1 knockdown status (Table 1). For example, in the presence of doxycycline, 82% of the PTEN^+/− mice showed thyroid adenomas comparable to the 89% in PTEN^+/−Rosa26^PDK1-RNAi/+ animals. Neoplastic leasons in the adrenal gland were less frequent and we observed high variation among control groups (20%–53%). Thus the apparent lowered incidence (4 vs. 2 animals) in the doxycycline treated group may have occurred by chance. Alternatively, we noted that loss of PTEN in the adrenal gland had little effect on AKT upregulation, which could provide a setting for PDK1 knockdown to be efficacious (Fig. 6C).

In summary, substantial knockdown of PDK1 (>80%) is not able to prevent tumor development in PTEN^+/− tissues. This difference in phenotype from the PDK1 hypomorphic mice crossed to PTEN^+/− mice may be attributed to knockdown in adult versus embryonic tissues as the latter results in a small cell phenotype with significantly altered body size and nutrients uptake defects (27, 28).

PDK1 is thought to be a key modulator of PI3K-pathway signaling based on elegant studies using PDK1^−/− mouse ES cells and knockin mutant mice (15). It is an activator of numerous AGC kinases and more than a dozen companies have disclosed efforts on developing PDK1 inhibitors (11). However, to date, no inhibitors have emerged with suitable kinase selectivity and physiochemical properties to enable meaningful in vivo studies (11). In this light, genetic approaches to study PDK1 function become highly valuable. Such approaches directly complement preclinical drug discovery efforts as a means to identify pharmacodynamic biomarkers and explore patient responder hypotheses.

Here, to provide a preclinical assessment of PDK1 target efficacy and safety, we generated mouse lines with efficient doxycycline-induced knockdown of PDK1. Compared with conventional gene-targeted knockout strategies, this approach has several advantages including the reversible and temporal control of genetic-alteration. While recombinase-based knockouts or mutational approaches (exemplified by the Cre/loxP system) enable temporal control, the genetic modification is irreversible. For modeling of many human diseases, such as PTEN deficiency-dependent tumorigenesis, conditional Cre/loxP systems are very powerful as illustrated by the advanced PTEN-null leukemia and prostate cancer models (25, 40, 41). However, as a genetic approach to mimic pharmaceutical intervention, gene knockout models are limited by the fact that they completely deplete the cell of the targeted protein. Pharmacologically, for an enzyme this would equal a 100% chronic inhibition of catalytic activity, which is an unrealistic benchmark for most drug targets. RNAi-mediated inactivation, however, typically lowers the amount of catalytically active drug target, which in this respect is more similar to a pharmacological intervention. However, inherent differences still remain between RNAi and small-molecule inhibitors. Frequently, small-molecule drugs specifically target enzymatic activity while knockdown impacts all protein functions (i.e., scaffolding and adaptor functions). In some cases, this can result in distinct biological phenotypes (43, 36). However, given the lack of suitable in vivo PDK1 tool compounds (11), the present RNAi-based approach is highly warranted.

Important for our inducible gene silencing approach is that substantial PDK1 knockdown is achievable and that shRNA expression is effectively suppressed prior to administration of the inducer (doxycycline). We report tight control of PDK1-shRNA expression and efficient knockdown, consistent with the initial report using this targeting strategy (30). Out of 10 tissues, brain is the only organ with a significant reduction of PDK1 transcripts in the absence of doxycycline. However, treatment with doxycycline resulted in further knockdown and lowered protein levels. The apparent leakiness of shRNA expression in the brain is likely an inherent property of the H1-tetR system, since we also observed it for other genes evaluated by the same RNAi approach (unpublished data).

For PDK1 we have demonstrated that efficient (90%) knockdown is neither sufficient to block PI3K pathway activation in various tissues nor prevent tumor formation and progression resulting from loss of PTEN. This genetically invalidates PDK1 as an attractive mono-therapy drug target in a PTEN-deficient tumor context. Our results clearly indicate that the residual PDK1 was sufficient to support high levels of p-AKT^T308 and maintain flux through the PI3K-pathway. Alternatively, other factors exist that can compensate for the low PDK1 levels. The latter scenario may be aided by adaptive cellular mechanisms on long-term PDK1 knockdown (i.e., feedback mechanism, subcellular compartmentalization, changes in phosphatase activities). Taken together, we conclude that PDK1 is not rate limiting for AKT phosphorylation in PTEN-deficient tissues.

While PDK1 is not the rate limiting factor for the pathway, it is still absolutely required for phosphorylation of AKT^T308 as recently shown in using conditional PDK1 knockouts. The total loss of PDK1 in PTEN-deficient context completely abolishes p-AKT^T308 and prevents formation of T-Cell leukemia (8, 9).

Our findings are in contrast to what was seen in PDK1 hypomorphic mice, where the neomycin selection cassette within the PDK1 locus reduced PDK1 activity and protein levels by 80% to 90% (27). When combined with PTEN^+/− mice, tumor development was completely absent until 8 months of age. In older mice (14 to 15 months), the authors also reported a severe suppression of tumor incidence with only 22% of hypomorphic PDK1^fl/−PTEN^+/− mice having detectable tumors compared to 72% in PDK1^+/−PTEN^+/− mice (27). This observation, and the fact that PDK1^−/− ES cells are unable to activate numerous AGC kinases (6), including AKT, was originally taken to validate PDK1 as a prime drug target for PI3K pathway–addicted tumors.

Notably, our constitutive PTEN^+/− animals display shorter tumor latency than the 8 months reported by Bayascas and colleagues (27) and a tumor spectrum that is more similar to that reported by other groups (21–23). Across different tissues, we observed tumor formation at 6 months of age or younger and report no significant antitumor correlation with PDK1 deficiency.

To investigate this discrepancy, we have directly compared PDK1 protein levels between the PDK1-RNAi and hypomorphic models and found comparable lowered protein levels as judged by Western blotting. We cannot formally exclude that a subtly better knockdown in the hypomorphic animals exists, which could lower PDK1 protein levels below a threshold that is required to sustain elevated PI3K signaling in tumor cells. Such a threshold, although in a developmental context, could explain why we observe a less than Mendelian ratio of PDK1^fl/− offspring when generating hypomorphic mice (Experimental Procedures). Apparently, a functional threshold for PDK1 hypomorphism exists that depends on the genetic background as backcrossing of the PDK1^fl/− mice onto a C57BL/6 background severely reduced viability. Another significant difference between the 2 models, which may contribute to the discrepancy in antitumor efficacy, is the strong developmental impact of PDK1 hypomorphism on body size (30%–40% reduction; ref. 28).

In the current study, we evaluated the impact on glucose metabolism primarily to assess whether PDK1 knockdown levels were sufficient to have immediate physiological consequences. We found that PDK1 knockdown had an acute metabolic effect that coincided with the kinetics of protein knockdown. Consistent with this, tissue specific knockout of PDK1 in liver and pancreatic beta cells has been reported to result in glucose intolerance and diabetes, respectively (7, 44). Also, mutation of the PDK1 PH-domain leads to insulin resistance in mice due to impaired phosphorylation of AKT^T308 (45). A general concern of targeting the PI3K-pathway has been the potential impact on glucose homeostasis in terms of potentially limiting the therapeutic window. Importantly, long-term PDK1 knockdown was well tolerated and elevation of blood glucose levels was reversible on removal of doxycycline. PDK1 knockdown animals appeared overall healthy and maintained normal weight. This safety profile is in contrast to the liver specific PDK1 knockout mice, which die from liver malfunction after 4 to 6 weeks (7).

The side-by-side comparison of single digit nanomolar inhibitors targeting either AKT or PDK1 revealed a different ability to impact p-AKT^T308 levels. Even when combined with efficient PDK1 knockdown, treatment of primary PTEN-deficient cells with high concentrations of PDK1 inhibitor (1,000-fold more than enzymatic EC₅₀) was not capable of normalizing elevated p-AKT^T308. In contrast, treatment with the allosteric AKT inhibitor efficiently blocked AKT phosphorylation in this system. Of note, we used primary cells with exclusive deletion of PTEN to study the effect of PI3K-pathway inhibitors. In contrast to transformed cell lines, which have adapted to growth on tissue culture plastic and have acquired additional mutations, the advantage of the ex vivo primary cultures is that we can study a defined PTEN deficiency context without direct interference from other activated or suppressed pathways.

In summary, our mouse models predict that human cancer patients, who have elevated p-AKT^T308 as a biomarker, are at best likely to benefit only marginally, if at all, from PDK1 inhibition as a single agent. However, it is possible that PDK1 knockdown, in combinations with other targeted agents may show a synergistic antitumor effect. For example, the inhibition of p-p90RSK^S221 in ex vivo primary mammary epithelial cultures suggests inhibitory effects on MAPK-pathway signaling. p-p90RSK^S221 inhibition was unique to PDK1 inhibition and not observed with mTOR and AKT inhibitors. Paradoxically, RSK inhibition appears to be associated with potential compensatory ERK activation (Fig. 5) as described by others (39). In the leukemia model, we also noted that PTEN loss was inversely correlated with p-ERK status, highlighting PI3K/MAPK cross-pathway regulation (Fig. 4D). Moreover, we identified an adaptation or feedback mechanism that lowers AKT protein expression in response to PTEN deficiency (Figs. 3E, 4D, and 5) presumably to limit the activity of AKT.

Since we have observed physiological changes in response to PDK1 knockdown (i.e., reversible diabetes phenotype, effects on spleen size and trend toward prolonged survival in the leukemia model), the presented RNAi model is well positioned for continued evaluation of PDK1 as a potential drug target in other tumor and disease contexts. In this study, we have tested PDK1 target efficacy under a single agent treatment paradigm. Thus, it is possible that PDK1 inhibition may have significant beneficial impact on tumor inhibition either in a combination therapy or in tumors that have less strong activation of the PI3K pathway.

Finally, it remains unknown whether small-molecule PDK1 inhibitors would mimic the results from our genetic efficacy studies. However, based on our genetic efficacy studies in 3 PTEN-deficient tumor models, we conclude that in cancer patients where blocking elevated AKT activity is highly desirable (i.e., PTEN-deficient tumors), inhibiting AKT directly using an AKT specific inhibitor will be far superior than targeting the upstream phosphorylating kinase PDK1.

There are no conflicts of interest to disclose.

The authors thank Dario Alessi and the Division of Signal Transduction Therapy, University of Dundee, for generous sharing of reagents; Michele Cleary for shRNA design; Jason Laskey, Roseanne Wexler, Cara Lerner, Clay Efferson for technical support; and Gabriele Staples, Pradip Majumder, Thomas Rosahl for assistance with animal model generation.

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.