SGK1 Is a Critical Component of an AKT-Independent Pathway Essential for PI3K-Mediated Tumor Development and Maintenance

Activation of PI3K signaling through a variety of mechanisms, including PIK3CA-activating mutations or amplification, and PTEN mutation, deletion, or silencing, is found in about 40% of all human cancers (1, 2). Some specific tumor types, such as breast, prostate, colorectal, and endometrial cancer, as well as glioblastoma and anaplastic thyroid carcinoma, show even higher mutation rates for members of this pathway (2, 3). PI3K-dependent generation of phosphatidylinositol-3,4,5-triphosphate (PIP3) triggers membrane recruitment of proteins containing a PH-domain, such as the three AKT isoforms and the PDK1 kinase. Furthermore, it leads to the activation of the mTORC2 complex by relieving an autoinhibitory mechanism involving its component, SIN1 (4). mTORC2 and PDK1 act then in concert to phosphorylate and activate AKT (5, 6).

Although, traditionally, AKT is considered the primary effector of PI3K-transforming activity, there is accumulating evidence that additional PI3K-dependent pathways may play a critical role during neoplastic transformation (7) or in driving adaptive resistance to PI3K inhibition (8, 9).

In fact, mTORC1 and mTORC2, two key PI3K effectors, activate, in cooperation with PDK1, several members of the AGC family of kinases, including S6K, PKC, and SGK (10). These targets, in turn, are known to control pathways with direct bearing on the transformation process (11).

In view of these notions, it is noteworthy that a formal, systematic in vivo genetic approach to address the questions of (i) whether AKT activation is sufficient for tumor development and maintenance, and (ii) what is the role in these processes, if any, of the other PI3K/PDK1–dependent pathways, has never been reported.

As PI3K activation is a common feature of advanced and aggressive thyroid cancer (3), we have used the thyroid gland as a relevant model system to address this issue and have generated a series of mouse models in which different critical components of the PI3K signaling cascade have been selectively manipulated in the thyroid epithelial cells.

We have found that although AKT activation is essential for PI3K-dependent neoplastic transformation, it is clearly not sufficient. In fact, we show that activation of PDK1-dependent AGC kinases is essential for tumor development. Furthermore, we provide evidence that the AGC kinase family member SGK1 plays a critical role downstream of PI3K activation for tumor maintenance and that it represents a critical therapeutic target.

The Pten^thyr−/−, [Pten^thyr−/−,Kras^G12D], and [Pten,Tp53]^thyr−/− strains have been described previously (12–14). Pdk1^L155E mice (15) were kindly provided by Dr. Dario Alessi (University of Dundee, Dundee, Scotland). All strains were backcrossed in the 129S6/SvEv background for at least 10 generations. All experimental procedures involving mice were conducted in accordance with IACUC-approved protocols.

Formalin-fixed samples were paraffin embedded and processed for routine hematoxylin and eosin staining. Six-micron sections were subjected to antigen retrieval, incubated with antibodies against Ki67 (Dako), and counterstained with hematoxylin.

Alternatively, mice were injected intraperitoneally with bromodeoxyuridine (BrdUrd; 10 mg/kg, Sigma) 2 hours before sacrifice. Anti-Ki67 and anti–BrdUrd-stained sections were analyzed using the ImageJ software.

Cells, thyroids, and tumor tissue were homogenized on ice in RIPA buffer (Thermo Fisher Scientific) supplemented with Halt Protease and Phosphatase Inhibitor cocktail (Thermo Fisher Scientific). Protein concentration was determined using the Pierce BCA Protein Kit (Thermo Fisher Scientific). Western blot analysis was carried out using 30 to 50 μg proteins run on Bis-Tris SDS precast gels (GenScript).

All the primary antibodies used were purchased from Cell Signal Technology and used at a dilution of 1:1,000 in 5% BSA in TBS-T. Signals were detected with HRP-conjugated secondary antibodies (Thermo Fisher Scientific) and the chemiluminescence substrate Luminata Crescendo (EMD Millipore). Equivalent loading was confirmed with anti–β-actin (Sigma-Aldrich).

Cell lines were established from tumors developed in Pten^thyr−/−, [Pten, Tp53]^thyr−/− and [Pten^thyr−/−,Kras^G12D,Pdk1^L155E] mice as described previously (16, 17). All mouse cell lines, as well as the human cell lines 8505c, ACT1, and CAL62 were grown in DMEM (HyClone) supplemented with 10% FBS (HyClone) and Mycozap Plus-CL (Lonza Biologics). The human cell lines TAD2, N-Thy-Ori, T235, T238, THJ21T, SW1736, C643, Ocut2, T243, and THJ16T were grown in RPMI (HyClone) supplemented with 10% FBS (HyClone) and Mycozap Plus-CL. All cell lines were maintained at 37°C with 5% CO₂. Cell identity was validated by STS profiling as well as by amplifying and sequencing genomic fragments encompassing their known mutations.

Chemical inhibitors GSK650394, EMD638683, MK2206, and GDC-0941 were purchased from Selleck Chemicals and dissolved in DMSO.

For drug sensitivity experiments on parental cell lines, all inhibitors were added 24 hours after plating. After 72 hours of treatment, cells were detached and counted using a Z2 Coulter counter (Beckman Coulter). Final DMSO concentration in media was always below 0.1%. EC₅₀ value calculation and statistical analysis were done using GraphPad Prism (GraphPad Software).

In experiments on lentivirally transduced cell lines, cells were pretreated with 250 ng/mL (THJ16T) or 500 ng/mL (T4888M) doxycycline hyclate (Sigma-Aldrich) for 48 hours prior to treatment with GDC-0941. Drugs were added immediately after cell seeding.

For long-term doubling rate experiments, 5 × 10⁴ cells were plated in the presence of chemical inhibitors or doxycycline hyclate. Every 3 days, cells were harvested, counted, and 5 × 10⁴ cells were reseeded.

Statistical analysis of drug synergy was evaluated from the results of the Wst-1 assays and calculated using the Chou–Talalay method (18) and the CalcuSyn Software (Biosoft). To determine synergy between two drugs, the software uses a median-effect method that determines whether the drug combination produces greater effects together than expected from the summation of their individual effects. The combination index (CI) values are calculated for the different dose-effect plots (for each of the serial dilutions) based on the parameters derived from the median-effect plots of the individual drugs or drug combinations at the fixed ratios. The CI was calculated on the basis of the assumption of mutually nonexclusive drug interactions. CI values significantly >1 are antagonistic, not significantly different than 1 are additive, and values <1 are synergistic.

A total of 5 × 10³ cells were plated in 96-well round bottom, ultralow attachment plates (Corning). Spheroid volume was measured after 3 and 10 days using ImageJ.

Cells were transduced with lentiviruses encoding shRNAs against SGK1 (from the TRC Genome-Wide shRNA Collection). After preliminary experiments with a battery of previously fully validated clones (9), shRNA TRCN0000040175 was chosen for subsequent experiments due to its superior targeting efficiency.

ShRNA TRCN0000040175 was cloned into the Tet-pLKO-puro (Addgene plasmid #21915). Lentiviral constructs were packaged in HEK293 cells transfected using Lipofectamine 2000 (Invitrogen). Viral supernatant was collected 48 to 72 hours after transfection, combined, and filtered using 0.45-μm Nalgene SFCA syringe filters (Thermo Fisher Scientific). Target cells were plated 24 hours before infection and then exposed to viral supernatant supplemented with 20 μg/mL polybrene (Santa Cruz Biotechnology). Mouse cell lines were then selected for at least 72 hours with puromycin (Corning) 5 μg/mL, whereas a concentration of 2 μg/mL was used for human cell lines.

Transduced cell lines were pretreated with 250 to 500 ng/mL doxycycline hyclate for 48 hours prior to plating for growth curve experiments. After seeding, cells were counted every 24 hours, and cell number was determined as described previously. In all experiments performed on transduced cell lines, doxycycline hyclate was replaced every 24 hours.

Mass spectrometry was used to measure inositol lipid levels essentially as described previously (19), using a QTRAP 4000 (AB Sciex) mass spectrometer and employing the lipid extraction and derivitization method described for whole tissue, with the modification that frozen tissue samples were ground under liquid N2 with a mortar and pestle prior to analysis (instead of Dounce homogenizer), and that 10 ng C17:0/C16:0 PtdIns(3,4,5)P₃ ISD (internal standard) and 10 ng C17:0/C16:0 PtdIns ISD was added to primary extracts. Final samples were dried in a speedvac concentrator rather than under N2. Measurements were conducted in triplicate per experiment, 0.5 mg per sample.

Total RNA was extracted with TRIzol reagent (Invitrogen) according to the manufacturer's protocol. RNA yields were assessed using a NanoDrop 1000 (Thermo Fisher Scientific). RNA (1 μg) was reverse transcribed using the Maxima First Strand cDNA Synthesis Kit (Thermo Fisher Scientific) according to the manufacturer's instructions. qRT-PCR was conducted on a StepOne Plus apparatus using the Absolute Blue qPCR Rox Mix (Thermo Fisher Scientific) and TaqMan expression assays (Applied Biosystem) following the manufacturer's instructions. The TaqMan expression assays used were Mm00441380_m1 (Sgk1), Mm00449845_m1 (Sgk2), Mm00453797_m1 (Sgk3), Mm02026778_g1 (Akt2), and Mm00446973_m1 (Tbp). Each sample was run in triplicate, and Akt2 and Tbp were used to normalize the input RNA. Relative quantities were calculated using the 2^−ΔΔC_t method.

Transduced cell lines were seeded in 10-cm plates, cultured overnight at 37°C, and incubated for 72 hours with doxycycline hyclate 250 ng/mL. Cells were harvested by trypsin treatment and fixed in 70% ethanol in ice for 30 minutes. After treatment with RNase (Genentech/Roche) for 5 minutes at room temperature, cells were stained with propidium iodide (Genentech/Roche) overnight, and DNA content was measured using a BD FACSCanto II system (BD Biosciences).

Cell lines were seeded in 100-mm plates, cultured overnight at 37°C, and incubated for 72-96-120 hours with doxycycline hyclate. Cells were harvested by trypsinization, and supernatant was also collected. Cells were stained with Annexin V FITC and propidium iodide (BD Biosciences) for 15 minutes at room temperature in the dark. Samples were analyzed by flow cytometry within 1 hour using a BD FACSCanto II system (BD Biosciences). Flow cytometry analysis was performed on the FloJo platform.

Eight-week-old NSG mice, from the Einstein Shared Facility in Stem Cell Research, were injected subcutaneously with 5 × 10⁶ T4888M or THJ16T cells previously transduced with an inducible shRNA targeting SGK1. When tumors reached a size between 100 and 150 mm³, mice were randomized to control and doxycycline hyclate treatment groups. Doxycycline hyclate was dissolved in water and administered via oral gavage (100 mg/kg) daily, and tumor volume was calculated from two-dimensional measurements using the equation: tumor volume = (length × width²) × 0.5. Tumor weight was measured at the end of the experiment.

Pten^thyr−/− mice are born with mildly hyperplastic thyroid glands (12) and slowly develop solid, nodular lesions that progress to locally invasive follicular carcinomas after one year of age (Fig. 1A and B; ref. 20). To test to what extent unrestrained PI3K relies on AKT activation to induce hyperproliferation and transformation of mouse thyrocytes, we generated a mouse model in which Akt1 and Akt2 are simultaneously deleted in the thyroid epithelial cells, alone or in combination with the loss-of-function Pten allele. Akt3 was excluded from this analysis, as its expression is undetectable in the mouse thyroid (Supplementary Fig. S1).

Although codeletion of Akt1 and Akt2 did not cause any overt phenotype, and did not alter thyroid size or histology at any time point analyzed, it drastically inhibited the development of both follicular hyperplasia and nodular lesions in Pten^thyr−/− mice (no triple-mutant mice with nodules at 48 weeks, vs. 66.7% of Pten^thyr−/− mice; Fig. 1A and B).

Thus, based on in vivo genetic data, AKT activation appears to be an absolute requirement for the oncogenic activity of unrestrained PI3K. It remains to be seen, however, whether it is sufficient to drive thyrocyte transformation.

Notably, we found that the size of the thyroid in the triple mutants was still slightly but significantly larger than that of wild-type mice (Fig. 1A), suggesting that AKT is not the only player downstream of active PI3K to participate in the thyroid hyperplastic response.

In addition to AKT, PI3K controls, via mTORC1 and mTORC2, the activation of several members of the AGC family of protein kinases, including S6K, SGK, and a number of PKC isoforms (10). Similar to AKT, activation of these kinases requires two phosphorylation events, one in the activation loop, mediated by PDK1, and the other in the hydrophobic motif, mediated by mTORC1 or mTORC2. Other AGC kinases such as RSK, instead, require ERK1/2 activity for the hydrophobic motif phosphorylation. In stark contrast with AKT, phosphorylation of the AGC kinase activation loop is absolutely dependent on prior phosphorylation of the hydrophobic motif, which creates a docking site for the PIF-pocket of PDK1 (21). Consequently, a point mutation in the PIF-pocket, L155E, completely abolishes PDK1 ability to activate all the AGC kinases, without impacting its ability to phosphorylate AKT (22).

We have taken advantage of this differential activation mechanism to explore the role and relevance of AGC kinases activation downstream of PI3K.

To this end, Pten^thyr−/− mice were crossed to mice in which the Pdk1 L155E-mutant allele is expressed from its endogenous locus in a Cre-dependent manner (15). The resulting compound mutant progeny was then compared with sex- and age-matched wild-type and single-mutant mice. Western blot analysis of thyroid extracts confirmed that AKT was still fully phosphorylated in the double mutants and was able to phosphorylate its targets, GSK3α/β, while the PIF pocket–dependent S6K was, as predicted, unable to phosphorylate ribosomal protein S6 (Fig. 1C).

Analysis of control, Pten^thyr−/−, and compound mutant mice at 12, 48, and 60 weeks of age revealed that impairment of AGC kinase activation significantly mitigates the hyperplastic phenotype observed in the thyroid of Pten^thyr−/− mice. Both thyroid weight and thyrocyte proliferative index were reduced by at least 50% at each time point analyzed, strongly suggesting that AKT activation is only partially responsible for the hyperproliferative phenotype, and that AGC kinase-driven pathways are critical for thyroid proliferation (Fig. 1D–F).

Histologic analysis of these thyroids revealed striking differences: at 48 weeks of age, when over 60% of Pten^thyr−/− female mice display nodular lesions, and a few of them (3%) already have progressed to invasive carcinomas, 97% of Pten^thyr−/−,Pdk1^L155E compound mutant mice displayed only mild hyperplasia (Fig. 2A and B). At 60 weeks, when over 46% of Pten^thyr−/− female mice display invasive carcinomas, and 43% have nodular lesions, 94% of Pten^thyr−/−,Pdk1^L155E compound mutant mice still displayed simple hyperplasia (Fig. 2A and B). We have previously reported (20) that male Pten^thyr−/− mutants show a delay in the development of thyroid neoplastic lesions. At 60 weeks of age, 7% of Pten^thyr−/− male mice display invasive carcinomas, and 64% have nodular lesions; on the other hand, 94% of Pten^thyr−/−,Pdk1^L155E compound mutant males displayed simple hyperplasia (Supplementary Fig. S2). These data strongly suggest that, despite constitutive AKT activity, the simple thyroid hyperplasia developed by Pten^thyr−/−,Pdk1^L155E compound mutants does not progress to neoplastic lesions.

The thyroid lesions developed by Pten^thyr−/− mice progress very slowly to invasive tumors. This prompted us to analyze the effect of ablating AGC kinase activation in more aggressive models of thyroid cancer. We have reported that simultaneous thyroid-specific activation of Kras and PI3K leads to the rapid development of aggressive carcinomas (14). When we introduced the Pdk1 L155E-mutant allele into this strain, we found that the median survival of the triple compound mutants was increased from 8 to 29 weeks (Fig. 2C). Furthermore, histologic analysis revealed that the lesions developed by the triple mutants were not solid carcinomas, as in the Pten^thyr−/−,Kras^G12D mice, but rather massive hyperplasias ultimately leading to mouse death via airway compression (Fig. 2D).

We have generated stable cell lines from thyroid lesions developed by Pten^thyr−/−,Kras^G12D (20) as well as by Pten^thyr−/−,Kras^G12D,Pdk1^L155E mice. In agreement with the in vivo data, cell lines carrying the Pdk1 L155E-mutant allele proliferated dramatically slower than their Pdk1 wild-type counterpart, both in standard culture conditions and as spheroids in the more physiologic 3D growth conditions (Fig. 2E and F).

Finally, we tested the effect of the Pdk1 L155E-mutant allele on the development of anaplastic thyroid carcinoma, the most aggressive and undifferentiated subtype of thyroid cancer, using the [Pten, Tp53]^thyr−/− model (16). [Pten, Tp53]^thyr−/− mice develop follicular carcinomas early on, which progress to anaplastic carcinomas in over 70% of cases by 48 weeks of age.

Impairment of AGC kinase activation reduced the weight of the compound mutant glands by 75%, to the same level observed in single-mutant Pten^thyr−/− mice (Fig. 3A). Histologic analysis of these lesions showed that only 5% of cases had nests of undifferentiated cells, while 65% of cases showed only well-differentiated follicular carcinomas, and 30% nodular hyperplasia (Fig. 3B and C).

Thus, despite constitutive PI3K and AKT activation, neoplastic lesions developed by [Pten, Tp53]^thyr−/−, Pdk1^L155E thyroids are unable to progress to aggressive, undifferentiated anaplastic carcinomas.

Genetic manipulation of PI3K downstream effectors might lead to unexpected feedbacks that could impact PI3K catalytic activity. To assess the activation status of PI3K in our genetic models, we measured the levels of phosphatidylinositol-3,4,5-triphosphate (PIP3) in wild-type, single, double, and triple mutant thyroids, using a well-established mass spectrometry approach and found that PIP3 levels were increased, as expected, in Pten^thyr−/− mice, but did not change upon introduction of additional mutant alleles (Fig. 3D).

Thus, despite a fully active PI3K/AKT axis, Pten^thyr−/− thyrocytes absolutely need AGC kinase activity to reach their full transformation potential, which is dictated by the additional genetic alterations introduced (Kras, Tp53, etc.).

To identify the critical AGC kinase that is required for full PI3K oncogenic activity, we used a large panel of mouse and human thyroid cancer cell lines encompassing the spectrum of mutations most commonly found in advanced thyroid tumors and performed a sensitivity screen using small-molecule inhibitors of PI3K and mTORC1 (as controls), and of the PI3K-regulated AGC kinase families, S6K, PKC, and SGK.

Strikingly, we found that PKC and S6K inhibitors had little to no effect on the proliferation of thyroid cancer cells at concentrations commonly used in cell culture assays. On the other hand, GSK650394, an SGK inhibitor, had a dramatic growth-suppressive effect, especially in cells carrying driver mutations that activate PI3K signaling (Fig. 4A).

Given the lack of antibodies specifically recognizing active SGK, we tested the phosphorylation level of NDRG1, a bona fide SGK target, in the thyroids of wild-type, Pten-mutant, and double-mutant mice, and found that, indeed, SGK activation is increased in Pten^thyr−/− thyroids and returns at baseline levels in the glands of Pten^thyr−/−,Pdk1^L155E compound mutant mice (Fig. 4B).

These data suggest that SGK is the critical AGC kinase that is required for PI3K full oncogenic activity.

Commercially available SGK inhibitors are notoriously suboptimal in terms of target selectivity. To verify that SGK is indeed the critical target affecting thyroid cancer cell proliferation, we first reduced the dose of GSK650394 by 40% and found that a 3 μmol/L concentration was still able to profoundly inhibit proliferation of PI3K-active cell lines (Fig. 4C). Then we compared the response of nine thyroid cancer cell lines to both GSK650394 (3 μmol/L) and to another SGK inhibitor, EMD638686, again used at a lower concentration (2.5 μmol/L) than usually found in published studies (10 μmol/L). We found a strong correlation between the sensitivity of these cells to the two inhibitors, further suggesting that they are indeed targeting SGK (Fig. 4D).

Next, we used the four Pten^−/−,Kras^G12D cell lines described above, two of which also carry the Pdk1^L155E mutation, to demonstrate that these two inhibitors were able to decrease cell proliferation in a dose-dependent manner in the AGC kinase-proficient cells, but not in the AGC kinase-defective lines (Fig. 4E and F). Furthermore, we tested whether GSK650394 could affect the ability of SGK to phosphorylate NDRG1 in T4888M cells and found that it significantly reduced NDRG1 phosphorylation. Interestingly, concomitant AKT inhibition was required for full ablation of phosphorylation, suggesting that, at least in specific cellular contexts, NDRG1 is a common target of SGK and AKT (Fig. 4G).

Finally, we performed a time course analysis of the effect of 3 μmol/L GSK650394 on the proliferation of PI3K-active thyroid cancer cells and found that SGK inhibition profoundly affected cell growth, with inhibition rates ranging between 63% and 95% after 4 days (Fig. 4H), thus formally validating SGK as a critical component of the PI3K-dependent oncogenic machinery.

SGK is a family of three closely related kinases. To assess their relative contribution to the transformation process, we analyzed the RNA level of the three isoforms in the thyroid of control, Pten-mutant, and double-mutant mice and found that Sgk1 is the most highly expressed of the three genes, while Sgk2 expression levels are several orders of magnitude lower than those of Sgk1 and Sgk3 (Fig. 5A). These results were also confirmed at the protein level, in thyroid extracts from wild-type and Pten-mutant mice (Fig. 5B). More variability, instead, was observed in mouse thyroid cancer cell lines, where, although Sgk1 is still the most expressed gene, the relative expression of Sgk2 and Sgk3 is different in each line (Fig. 5C). The almost undetectable expression of Sgk2 suggests that this isoform is unlikely to play a critical role in the control of thyroid cell proliferation. On the other hand, Sgk3 is unique among the SGK genes, as it has an N-terminal PX-domain targeting the protein to the endosomes, where it is activated in response to the class III PI3K Vps34 (23). On the basis of these notions, as well as on preliminary shRNA-based experiments showing very limited effects of Sgk3 depletion on cell proliferation (Supplementary Fig. S3), we focused on Sgk1 as the primary conduit of this AKT-independent, PI3K-driven pathway.

Preliminary experiments revealed that constitutive and sustained shRNA-mediated Sgk1 depletion is not tolerated by PI3K-active thyroid cancer cells and that no stably transduced pools could be maintained for more than a few passages. Thus, we generated doxycycline-inducible mouse and human thyroid cancer cell lines, using a previously validated shRNA (9) that targets both mouse and human SGK1 (Fig. 5D). Four days of SGK1 depletion resulted in reduced NDRG1 phosphorylation and in a slight increase in AKT activation, accompanied by a minor increase in GSK3β phosphorylation (Fig. 5D). However, no changes in S6 phosphorylation were observed.

Analysis of cell proliferation upon shSgk1 induction revealed dramatic growth suppression in both mouse (Fig. 5E–G) and human (Fig. 5H) thyroid cancer cells harboring driver mutations that activate PI3K signaling, with inhibition rates ranging between 68% and 98% after 6 days of doxycycline treatment. The growth-suppressive effect, conversely, was significantly lower in human thyroid cancer cell lines without mutations activating PI3K (44%; Fig. 5I; Supplementary Fig. S4).

Flow cytometric analysis of control and shSgk1-induced thyroid cancer cells showed that SGK depletion caused arrest in G₁ at early time points (Fig. 5J), followed, 24 to 48 hours later, by induction of a significant degree of apoptosis (Fig. 5K).

Taken together, these data show that PI3K-active thyroid cancer cells absolutely need SGK1 activity to support sustained proliferation and survival.

To test the long-term effect of SGK1 depletion on the growth of PI3K-active thyroid cancer cells, we performed a 3-week long analysis of cell proliferation using the THJ16T cell line. Induction of shSGK1 drastically reduced cell proliferation, which reached its lowest level by day 6 and then remained rather constant for the remainder of the follow-up (Fig. 6A). SGK1-depleted cells, however, did not undergo irreversible growth inhibition or senescence, as doxycycline removal after 12 days promptly restored the normal proliferation rate for these cells (Fig. 6A, green line).

To test the effect of SGK1 depletion on tumor growth in vivo, we generated xenograft models using the murine T4888M cell line (Pten^−/−, Tp53^−/−) and the human THJ16T cell line (PIK3CA^E545K, TP53^−/−). Doxycycline-mediated induction of the shRNA targeting SGK1 severely impaired tumor growth in both models (Fig. 6B and C). In line with the higher knockdown level (see Fig. 5D), the effect was more dramatic in the THJ16T xenograft: SGK1 depletion resulted in tumor regression in almost 90% of cases, with complete tumor disappearance in 25% of the mice (Fig. 6D). These results were further validated by tumor weight analysis at endpoint, demonstrating that, despite continuous PI3K signaling, SGK1 depletion reduced mean tumor size by over 80% (Fig. 6E).

IHC analysis of these tumors showed no significant changes in the number of apoptotic cells. However, the fraction of proliferating cells was strikingly reduced in SGK1-depleted cells (Fig. 6F).

The notion that SGK1 activation is fully dependent on PI3K signaling leads to the prediction that simultaneous targeting of both kinases should not result in a significant advantage over PI3K inhibition alone.

We tested this postulate using the doxycycline-inducible shSGK1 T4888M and THJ16T cell lines. As expected, the pan-PI3K inhibitor GDC0941 was clearly effective in reducing thyroid cancer cell proliferation over a 3-day period. When comparing each time point to the proliferation of untreated cells, simultaneous depletion of SGK1 dramatically increased the overall growth-suppressive effect (Fig. 7A). However, when we normalized cell proliferation at each GDC0941 concentration to the basal proliferation in the presence or absence of SGK1, thus obtaining a measure of the “net effect” of GDC0941 treatment, it became clear that SGK1 depletion paradoxically reduced the net ability of GDC0941 to inhibit cell proliferation, in particular in the THJ16T cells (Fig. 7B). These data demonstrate that SGK1 is a key PI3K downstream effector for the induction of proliferation, and thus, its absence reduces the net ability of PI3K inhibitors to impact cell proliferation, despite complete abolishment of AKT and S6 phosphorylation (Fig. 7C).

Along the same lines, simultaneous pharmacologic cotargeting of PI3K and SGKs did not result in strong synergistic activity (Supplementary Fig. S5).

Our model also predicts that targeting in combination the two major PI3K effectors, AKT and SGK, should be more effective than targeting each kinase separately. In fact, we found that the AKT inhibitor MK2206 strongly synergized with SGK inhibition in both T4888M and THJ16T cells, with a CI at EC₉₀ of 0.58 and 0.45, respectively (Fig. 7C and D).

Prolonged inhibition of certain signaling pathways can elicit adaptive responses that result in the development of resistance to the inhibitor. These responses are usually not evident in the standard 3-day proliferation assays. To assess the effect of long-term pathway inhibition on cell proliferation, we measured the doubling rate of T4888M cells over a 2-week period and found that the growth-inhibitory effect of the PI3K inhibitor GDC0941, used at EC₉₀, was rapidly and progressively lost, beginning after 6 days of continuous treatment (Fig. 7E, red line). On the other hand, single-agent treatment with the AKT inhibitor MK2206 or the SGK inhibitor GSK650394, at their EC₈₀, had a more steady growth-suppressive effect. Strikingly, when administered in combination, these two inhibitors dramatically synergized, leading to almost complete inhibition of cell proliferation (Fig. 7E).

Thus, combined inhibition of the two critical PI3K effectors, AKT and SGK, has a clearly increased and prolonged therapeutic potential, compared with direct PI3K inhibition.

Three decades of research have formally established the critical role of the PI3K/AKT/mTOR pathway in tumor development and progression (24). At the same time, the development of genetic tools for accurate pathway dissection and of small-molecule inhibitors of most members of this signaling cascade have uncovered the existence of an intricate network of signals and feedbacks, which strongly underlines the need for a deeper understanding of this pathway, if we want to effectively target it for therapeutic purposes (25).

One of the tenets that have been recently challenged by several studies is the universal role of AKT activation as the critical conduit of PI3K-transforming signals (26). As an example, activation of SGK3 has been shown to be essential for the growth of a subset of PIK3CA-mutant breast tumors exhibiting only minimal AKT activation (7).

Our data address for the first time, in a clinically relevant in vivo model, the question of AKT sufficiency in a tumor type that instead relies on AKT activity as an effector of PI3K activation. In fact, we show that simultaneous genetic ablation of Akt1 and Akt2 completely suppresses thyroid cancer development in mice with thyroid-specific loss of the Pten tumor suppressor gene, thus formally proving that AKT activity is essential for tumorigenesis downstream of PI3K in this tissue. At the same time, the residual hyperplasia observed in [Pten, Akt1, Akt2]^thyr−/− mice suggests that additional PI3K-dependent, AKT-independent pathways contribute to thyrocyte hyperproliferation.

The Pdk1 L155E genetic model that uncouples AKT-dependent and AKT-independent pathways downstream of PI3K has been previously utilized to show that PI3K-dependent, AKT-independent pathways are required for T-cell progenitor proliferation (27), leucine-induced mTORC1/S6K activation in the cardiac muscle (28), and proper brain patterning (29). Our data further extend these concepts by providing clear genetic evidence that activation of AGC kinases controlled by PDK1 via its PIF pocket is critical not only for the proliferative response to PI3K activation, but also for the ability of PI3K to drive neoplastic transformation and tumor progression in three well-characterized mouse models of follicular, poorly differentiated, and anaplastic thyroid cancer, respectively.

The AGC kinases cocontrolled by PI3K and PDK1 are grouped in three major families, S6K, PKC, and SGK, each with multiple members. This renders impractical any in vivo genetic approach to pinpoint the critical kinases that act downstream of PI3K to facilitate tumor initiation.

We elected instead to use a small-scale inhibitor-based screen, combined with shRNA approaches in established thyroid cancer cell lines, to identify which AGC kinase is required for the maintenance of the transformed phenotype in cells that depend on PI3K activity. This approach has resulted in the identification of SGK1 as the primary candidate for this role.

SGK1 was originally discovered as an immediate early gene, induced by glucocorticoids and serum (30) and has historically been associated with the control of ion transport, in particular in the kidney (31). Notably, AKT and SGK1 display a high level of target promiscuity, which has often complicated the attribution of specific functions to either of them (32). Accumulating data in recent years support the notion that SGK1 might play a distinct, AKT-independent role in cellular transformation as an effector of activated PI3K signaling. As an example, Sgk1^−/− mice develop less colon tumors than wild-type mice when subjected to chemical carcinogenesis (33), and Apc^Min/+, Sgk1^−/− mice display a 75% reduction in the number of tumors per mouse, compared with Apc^Min/+ mice (34).

However, the best characterized mechanism through which SGK1 has been described to contribute to transformation is via opportunistic compensation upon PI3K or AKT inhibition. In fact, in breast cancer cells expressing high levels of SGK1 and treated with a PI3K inhibitor, SGK1 was shown to stimulate mTORC1 activity by directly phosphorylating TSC2, thus inducing resistance to PI3K inhibition (8). Along the same lines, high levels of SGK1 have been found to predict resistance to AKT inhibitors in breast cancer cells (9).

Here, we have demonstrated that loss of SGK1 profoundly impacts thyroid cancer cell proliferation and survival, both in vitro and in vivo, despite intact PI3K and AKT activity. As such, our results support a new model in which SGK1 plays an essential role as an integral part of PI3K-transforming machinery.

Further in-depth work is of course warranted to identify the specific pathways controlled by SGK1 in its oncogenic role. This function might be attained by impinging upon classical AKT targets and contributing to their phosphorylation, or by controlling known SGK targets that impact cell proliferation, such as the Ca²⁺ release–activated Ca²⁺ channel (I_CRAC; ref. 35), potassium channels such as Kv1.3 (36), proton exchangers such as NHE3 (37), or the epithelial sodium channel, ENAC (38). Alternatively, SGK1 could phosphorylate still unknown targets, which may contribute to neoplastic transformation either directly or by controlling additional downstream players.

The notion that SGK1 is a critical component of the PI3K oncogenic pathway has obvious clinical consequences. Our data clearly show that, upon short-term exposure, SGK1 genetic depletion does not significantly increase the efficacy of PI3K inhibition, as SGK1 activation appears to be completely PI3K dependent. At the same time, however, we show that prolonged exposure to a pan-PI3K inhibitor elicits a rapid adaptive response that dramatically dampens the efficacy of the inhibitor. This effect correlates with and might be mechanistically linked to the limited and short-lived responses to different PI3K inhibitors observed in clinical trials (39–41).

Pharmacologic SGK inhibition, instead, appears to strongly cooperate with AKT inhibition, does not instigate the adaptive resistance response, and leads to sustained and powerful growth inhibition, a result of a magnitude rarely observed in anaplastic thyroid cancer models, and which might apply to additional PI3K-dependent aggressive tumor types. Of note, the absence of any major phenotype in mice lacking both Sgk1 and Sgk2 (42), or both Sgk1 and Sgk3 (43), suggests that SGK inhibition should cause limited toxicity in patients.

In conclusion, our results uncover an essential role of SGK1 in the PI3K-mediated neoplastic transformation process and warrant an increased effort in the development of small-molecule SGK inhibitors for clinical use, with the promise of more effective and long-lasting therapeutic effects.

No potential conflicts of interest were disclosed.

Conception and design: A. Orlacchio, A. Di Cristofano

Development of methodology: A. Orlacchio, V.A. Arciuch, A. Di Cristofano

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Orlacchio, M. Ranieri, M. Brave, T. Forde, D. De Martino, K.E. Anderson, P. Hawkins

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Orlacchio, K.E. Anderson, P. Hawkins, A. Di Cristofano

Writing, review, and/or revision of the manuscript: A. Orlacchio, V.A. Arciuch, A. Di Cristofano

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Orlacchio, M. Brave

Study supervision: A. Di Cristofano

We acknowledge the Animal Housing, Histology, shRNA, Analytical Imaging, and Flow Cytometry Core Facilities of Albert Einstein College of Medicine, which are partially supported by the NIH Cancer Center Support Grant to the Albert Einstein Cancer Center (P30CA013330). This work utilized a High-Speed/Resolution Whole Slide Scanner that was purchased with funding from an NIH SIG grant 1S10OD019961-01. The Einstein Shared Facility in Stem Cell Research is supported in part by the ESSCF, NYS-DOH, Contract #C029154A. Research reported in this publication was supported by NIH grants CA172012, CA128943, and CA167839 to A. Di Cristofano.

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.