mTOR complex 1 controls the nuclear localization and function of glycogen synthase kinase 3β

Glycogen synthase kinase 3β (GSK3β) phosphorylates and thereby regulates a wide range of protein substrates involved in diverse cellular functions. Some GSK3β substrates, such as c-Myc and Snail, are nuclear transcription factors, suggesting the possibility that GSK3β function is controlled through its nuclear localization. Here, using ARPE-19 and MDA-MB-231 human cell lines, we found that inhibition of mTOR complex 1 (mTORC1) leads to partial redistribution of GSK3β from the cytosol to the nucleus and to a GSK3β-dependent reduction of the levels of both c-Myc and Snail. mTORC1 is known to be controlled by metabolic cues, such as by AMP-activated protein kinase (AMPK) or amino acid abundance, and we observed here that AMPK activation or amino acid deprivation promotes GSK3β nuclear localization in an mTORC1-dependent manner. GSK3β was detected on several distinct endomembrane compartments, including lysosomes. Consistently, disruption of late endosomes/lysosomes through a perturbation of RAS oncogene family member 7 (Rab7) resulted in loss of GSK3β from lysosomes and in enhanced GSK3β nuclear localization as well as GSK3β-dependent reduction of c-Myc levels. These findings indicate that the nuclear localization and function of GSK3β is suppressed by mTORC1 and suggest a link between metabolic conditions sensed by mTORC1 and GSK3β-dependent regulation of transcriptional networks controlling cellular biomass production.

Glycogen synthase kinase 3␤ (GSK3␤) phosphorylates and thereby regulates a wide range of protein substrates involved in diverse cellular functions. Some GSK3␤ substrates, such as c-Myc and Snail, are nuclear transcription factors, suggesting the possibility that GSK3␤ function is controlled through its nuclear localization. Here, using ARPE-19 and MDA-MB-231 human cell lines, we found that inhibition of mTOR complex 1 (mTORC1) leads to partial redistribution of GSK3␤ from the cytosol to the nucleus and to a GSK3␤-dependent reduction of the levels of both c-Myc and Snail. mTORC1 is known to be controlled by metabolic cues, such as by AMP-activated protein kinase (AMPK) or amino acid abundance, and we observed here that AMPK activation or amino acid deprivation promotes GSK3␤ nuclear localization in an mTORC1-dependent manner. GSK3␤ was detected on several distinct endomembrane compartments, including lysosomes. Consistently, disruption of late endosomes/lysosomes through a perturbation of RAS oncogene family member 7 (Rab7) resulted in loss of GSK3␤ from lysosomes and in enhanced GSK3␤ nuclear localization as well as GSK3␤-dependent reduction of c-Myc levels. These findings indicate that the nuclear localization and function of GSK3␤ is suppressed by mTORC1 and suggest a link between metabolic conditions sensed by mTORC1 and GSK3␤-dependent regulation of transcriptional networks controlling cellular biomass production.
GSK3␤ activity is indeed regulated by phosphorylation on Ser-9, mediated by kinases such as Akt, protein kinase C, and p90RSK, resulting in negative regulation of GSK3␤ activity (7)(8)(9)(10)(11)(12). Phosphorylation of other sites on GSK3␤ may also suppress GSK3␤ activity, such as that of Ser-389 by p38 MAPK (13). In addition to GSK3␤ phosphorylation, control of GSK3␤ action may be achieved by localization of GSK3␤ or some of its substrates into distinct cellular compartments, such as the nucleus, such that GSK3␤ may have limited and regulated access to certain substrates (4,14,15).
Several GSK3␤ substrates are transcription factors (4) localized largely to the nucleus, including c-Myc (16), Snail (17), C/EBP␣ and -␤ (18,19), and CREB (20). c-Myc controls genes important for proliferation, metabolism, and biomass production and stem-cell self-renewal (for reviews, see Refs. [21][22][23]. Moreover, c-Myc is an oncogene altered in many cancers (23), highlighting the need for precise regulation of its function. c-Myc protein levels are controlled by GSK3␤-dependent phosphorylation of Thr-58 on c-Myc (16), leading to ubiquitin-dependent proteosomal degradation (24). Control of phosphorylation and/or degradation of these nuclear substrates by GSK3␤ may involve modulation of GSK3␤ nuclear localization. However, the identity of the cellular compartments in which GSK3␤ is localized and how it moves from various cellular compartments to the nucleus is not well-defined.
GSK3␤ localizes in part to membrane compartments in the cytoplasm. It is recruited to the plasma membrane via association with Axin (25), impacting Wnt signaling to ␤-catenin (26). GSK3␤ is also detected on APPL1 early endosomes (27). APPL1 acts as an adaptor protein to recruit Akt, facilitating GSK3␤ phosphorylation and inactivation on these early endosomes, thus impacting clathrin-mediated endocytosis (28) and cell survival (27). GSK3␤ also localizes to lysosomes (29) and controls lysosomal acidification (30). Hence, GSK3␤ may localize to multiple distinct endomembrane compartments including the plasma membrane, early endosomes, and lysosomes, with distinct functions at each locale.
GSK3␤ exhibits nuclear localization under certain conditions including in response to apoptotic signals induced by heat shock or staurosporine treatment (31), S-phase of the cell cycle (32), replicative senescence in fibroblasts (33), and loss of phosphatidylinositol 3-kinase (PI3K)-Akt signaling in embryonic stem cells (15). Site-directed mutagenesis studies revealed that nuclear localization of GSK3␤ requires a bipartite nuclear localization sequence (NLS) contained within residues 85-103 on GSK3␤ and that nuclear localization was also modulated by the N-terminal 9 amino acids on GSK3␤ (14).
Collectively, these observations raise the question of whether the control of GSK3␤ nucleocytoplasmic shuttling could be an important mechanism to control its function by modulating access to nuclear substrates. Indeed, nuclear localization of GSK3␤ induced by inhibition of PI3K or Akt leads to GSK3␤dependent phosphorylation of c-Myc, leading to its degradation (15). However, it is not clear how PI3K-Akt signals impact GSK3␤ localization. Because this N-terminal region contains the Ser-9 phosphorylation site, it is possible that Akt or other kinases capable of phosphorylation of this residue impact nuclear localization of GSK3␤, although a GSK3␤ mutant (S9A) did not show obvious differences in nuclear localization (14).
Whereas Akt may control GSK3␤ localization via direct phosphorylation of GSK3␤ on Ser-9, this may be indirect and result from Akt-dependent activation of other signals, downstream of Akt, such as the mTOR complex 1 (mTORC1). Mitogenic activation of PI3K-Akt signals leads to inhibition of the tuberous sclerosis complex 1/2 (TSC 1/2) (34), activation of the GTPase Rheb (35), and thus mTORC1 activation (36,37). mTORC1 in turn controls many processes, including metabolism, protein synthesis, cell growth, and autophagy (recently reviewed by Saxton and Sabatini (38)). In addition to mitogenic control, mTORC1 is also strongly controlled by metabolic cues. Amino acid levels are sensed by a mechanism involving the lysosomal V-ATPase and other sensors, leading to the recruitment and activation of mTORC1 at the lysosome under conditions of amino acid sufficiency (39,40). Further, activation of AMPactivated protein kinase (AMPK) by energy insufficiency, resulting from an increase in the cellular levels of AMP and ADP relative to ATP, leads to phosphorylation and activation of TSC2, thus suppressing mTORC1 (41,42). Hence, mitogenic and metabolic signals control mTORC1 activation.
Although it is not known whether PI3K-Akt signaling regulates GSK3␤ nuclear localization via engagement of mTORC1, several studies have reported that GSK3␤ enhances mTORC1 activity. GSK3␤ phosphorylates the mTORC1 subunit Raptor (43), resulting in enhanced mTORC1 activity (30,43). GSK3␤ also negatively regulates mTORC1 signaling by binding (44) and phosphorylation of TSC2 (42). Moreover, GSK3␤ binds to and regulates AMPK (45). Hence, GSK3␤ controls the mTORC1 and AMPK metabolic and mitogenic sensors. However, the possibility of a reciprocal regulation of GSK3␤ by signals from mTORC1 and AMPK, impacting GSK3␤ nuclear localization and thus access to substrates therein, such as c-Myc, has so far been unexamined.
Here, we examine mTORC1 regulation of GSK3␤ nuclear localization and function. To do so, we use pharmacological and other approaches to manipulate mitogenic or metabolic signals and examine GSK3␤ localization to various endomembrane compartments and nucleus as well as GSK3␤-dependent functions associated with nuclear GSK3␤. We find a novel regulatory axis sensing mitogenic signals, metabolic cues, and membrane traffic at the late endosome/lysosome that modulates GSK3␤ nuclear localization and function.

Results
The PI3K-Akt signaling pathway controls GSK3␤ nucleocytoplasmic shuttling and thus access of GSK3␤ to nuclear targets either directly or via activation of the downstream kinase mTORC1. mTORC1 integrates both mitogenic (PI3K-Akt) and metabolic cues and is localized to the lysosome once activated. Using a variety of strategies to manipulate mitogenic and metabolic signals converging on mTORC1 and lysosomal membrane traffic, we examined how mTORC1 regulates GSK3␤ nuclear access and function.

mTORC1 controls GSK3␤ nuclear localization and c-Myc levels
To determine whether mTORC1 regulates GSK3␤ localization and function downstream of PI3K/Akt, we first examined the effect of the mTORC1 inhibitor rapamycin on c-Myc levels. Treatment of ARPE-19 cells (henceforth referred to as RPE cells) with 1 g/ml rapamycin caused a time-dependent decrease in c-Myc levels, reaching 57 Ϯ 4.8% after 2 h of rapamycin treatment (n ϭ 6, p Ͻ 0.05; Fig. 1A). Importantly, cotreatment with a 10 M concentration of the GSK3␤ kinase inhibitor CHIR99021 blunted the decrease in c-Myc levels elicited by rapamycin treatment (Fig. 1A). Consistent with this result, rapamycin treatment also elicited a reduction in levels of the transcription factor Snail, an effect also blunted by co-treatment with CHIR99021 (Fig. 1B).
We next used siRNA gene silencing of GSK3␤, which resulted in a 91 Ϯ 4.7% reduction of GSK3␤ protein levels (n ϭ 3, p Ͻ 0.05; Fig. S1A). Whereas RPE cells also express the paralog GSK3␣, silencing of GSK3␤ was specific and did not impact expression of GSK3␣ (Fig. S1B). Cells subjected to silencing of GSK3␤ exhibited no change in c-Myc upon inhibition of sequential signals in the PI3K-Akt-mTORC1 axis, achieved by treatment with either LY294002, Akti-1/2, or rapamycin, respectively (Fig. 1C). In contrast, each inhibitor effectively reduced c-Myc levels in cells subjected to nontargeting (control) siRNA treatment (Fig. 1C). Taken together, these results indicate that PI3K-Akt signals converge on mTORC1 to enhance c-Myc levels in a manner that requires the regulation of GSK3␤.
Thus, under conditions in which mTORC1 is active, GSK3␤ and c-Myc are compartmentalized separately within the cytosol and nucleus, respectively.
We next determined how PI3K-Akt-mTORC1 signaling regulates GSK3␤ localization. Treatment of RPE cells with either LY294002, Akti-1/2, or rapamycin to perturb PI3K, Akt, or mTORC1, respectively, resulted in a robust and significant (n ϭ 3, p Ͻ 0.05) increase in nuclear GSK3␤, measured by the ratio of nuclear to cytosolic mean fluorescence intensities of GSK3␤, which we term the GSK3␤ nuclear localization index (Fig. 2B). Importantly, the effect of rapamycin treatment on GSK3␤ nuclear translocation and Snail protein levels was also observed in MDA-MB-231 breast cancer cells (Fig. S1, D and E), demonstrating that the mTORC1-dependent control of GSK3␤ is not unique to RPE cells. Similar observations were made when detecting GSK3␤ cytoplasmic and nuclear localization by subcellular fractionation (Fig. 2C). Furthermore, inhibition of the PI3K-Akt-mTORC1 axis also resulted in robust nuclear localization of GSK3␣ (Fig. 2D), a paralog of GSK3␤ with highly similar kinase domains but unique terminal motifs (3,49). These results indicate that PI3K-Akt signals act via control of mTORC1 to regulate GSK3␤ nuclear localization as well as that of GSK3␣.
To test the importance of Ran in mTORC1-dependent GSK3␤ nuclear translocation, we examined the impact of Ran GTP-binding mutants on GSK3␤ localization. We expressed WT Ran or one of two Ran mutants, Ran T24N and G19V, which are constitutively GDP-or GTP-bound, respectively (50). Cells expressing WT Ran exhibited little nuclear GSK3␤ in the control condition, but a robust localization of GSK3␤ in the nucleus was observed upon treatment with rapamycin ( Fig. 3, top panels; quantification shown in the bottom panel). In contrast, cells expressing Ran T24N exhibited nuclear GSK3␤ in both control and rapamycin-treated conditions (Fig. 3), consistent with Ran-GDP acting to facilitate nuclear import (50). Further, cells expressing Ran G19V exhibited mostly cytosolic GSK3␤ in both control and rapamycin-treated conditions, consistent with this mutant blocking Ran-dependent nuclear import (Fig. 3). These experiments confirm that the change in localization of GSK3␤ observed upon treatment with rapamycin indeed corresponds to nuclear translocation. Further, these results indicate that GSK3␤ undergoes Ran-dependent nucleocytoplasmic shuttling and Ran-dependent nuclear import that is regulated by mTORC1.

Metabolic cues regulate GSK3␤ nuclear localization via mTORC1
As mTORC1 is regulated by both mitogenic (PI3K-Akt) signals as well as metabolic cues, we next examined how each of these signals contributes to the control of GSK3␤ nuclear localization. AMPK is activated via ATP insufficiency and negatively regulates mTORC1 signaling through phosphorylation and activation of TSC2 (41,42). Consistent with the effects of mTORC1 inhibition by rapamycin, treatment with the AMPK activator A769662 resulted in robust GSK3␤ nuclear localization (Fig. 4A). Importantly, AMPK and mTORC1 exhibit reciprocal negative regulation (51). As such, GSK3␤ nuclear localization could conceivably be the direct result of loss of antibodies. Also shown are mean c-Myc levels Ϯ S.E. (error bars) (n ϭ 6; *, p Ͻ 0.05) (A) or mean Snail levels (n ϭ 3; *, p Ͻ 0.05) (B) relative to that in the control conditions (absence of CHIR99021 and rapamycin). C, RPE cells were transfected with siRNA targeting GSK3␤ or nontargeting siRNA (control) and then treated with either 10 M LY294002, 5 M Akti-1/2, or 1 M rapamycin for 1 h. Shown are representative immunoblots of whole-cell lysates probed with anti-c-Myc or anti-actin (load) antibodies as well as mean c-Myc levels (n ϭ 4). *, p Ͻ 0.05 relative to that in the control conditions (absence of LY294002, Akti1/2, and/or rapamycin). All Western blotting quantifications shown have been normalized to loading controls (clathrin or actin).

mTORC1 regulates GSK3␤
mTORC1 signals or an increase in AMPK activation, both of which would be expected to occur upon treatment with either rapamycin or A769662. To dissect a role for mTORC1 versus AMPK in control of GSK3␤ nuclear localization, we used the AMPK inhibitor compound C (52). Cells treated with compound C exhibited a rapamycin-dependent increase in GSK3␤ nuclear localization comparable with that observed in cells treated with rapamycin but not compound C (Fig. 4A). This indicates that AMPK activity is dispensable for GSK3␤ nuclear localization induced by mTORC1 inhibition. As GSK3␤ forms a complex with AMPK (45), we also tested whether AMPK may have a kinase-independent, structural role in regulation of GSK3␤. However, silencing of AMPK did not impact GSK3␤ nuclear localization (Fig. S2). Collectively, these results indicate that whereas AMPK activation also triggers an accumulation of nuclear GSK3␤, this occurs as a result of AMPK-dependent inhibition of mTORC1 signals, and not as a result of direct action of AMPK on GSK3␤ nuclear localization. mTORC1 is activated by abundance of amino acids in a manner that requires the V-ATPase (39). To determine how amino acid-dependent activation of mTORC1 impacted control of GSK3␤ localization, we treated cells with the V-ATPase inhib-

mTORC1 regulates GSK3␤
itor concanamycin A. Cells treated with concanamycin A exhibited a significant enhancement of nuclear GSK3␤ relative to control (Fig. 4B). Consistent with this result, amino acid deprivation achieved via incubation of cells in amino aciddepleted medium (Earle's balanced salt solution (EBSS)) also mimicked the effect of rapamycin treatment in RPE (Fig. 4C) as well as MDA-MB-231 ( Fig. S1C) cells. These results indicate that amino acid sensing by mTORC1 contributes to the regulation of GSK3␤ nuclear localization. mTORC1 inhibition also leads to induction of autophagy (53). We therefore tested whether autophagy is required for GSK3␤ nuclear localization upon mTORC1 inhibition with rapamycin. To inhibit autophagy induction, we treated cells with siRNA targeting endogenous ULK (54), which resulted in a robust 77 Ϯ 6.2% reduction of ULK expression (n ϭ 3, p Ͻ 0.05; Fig. S3A). Cells treated with siRNA to silence ULK1 exhibited cytosolic GSK3␤, which relocalized to the nucleus upon rapamycin treatment in a manner indistinguishable from cells treated with nontargeting siRNA (Fig. 4D). As autophagy induction has also been reported to lead to c-Myc degradation (55), we also tested the effect of ULK1 silencing on rapamycininduced c-Myc levels. Surprisingly, silencing of ULK1 on its own reduced c-Myc levels (Fig. S3B). Moreover, and in contrast to the findings of a previous study (55), impairment of autophagy induction by ULK1 silencing did not prevent the rapamycin-induced reduction in c-Myc levels (Fig. S3B). Thus, GSK3␤ nuclear translocation and c-Myc degradation observed upon mTORC1 inhibition are largely independent of autophagy induction. Instead, c-Myc degradation upon mTORC1 inhibition is mediated by regulation of GSK3␤ nuclear localization and function.
Using phos-tag acrylamide electrophoresis, a technique that exaggerates differences in apparent molecular weight of phosphorylated species of a protein (56), we observed two detectable species of GSK3␤, of which the higher molecular weight species probably corresponds to the Ser-9 -phosphorylated form given its sensitivity to PI3K and Akt inhibition (Fig. S4A). In contrast and as expected, rapamycin had no effect on GSK3␤ detected by this method. Collectively, these results indicate that regulation of GSK3␤ Ser-9 phosphorylation does not contribute to control of GSK3␤ nuclear localization by PI3K-Akt-mTORC1 signals.

GSK3␤ is localized to several distinct membrane compartments within the cytoplasm
Active mTORC1 is recruited to the surface of the lysosome (57). Together with our observations that mTORC1 controls GSK3␤ nuclear localization, this suggests that (i) mTORC1 control of GSK3␤ may occur at lysosomes and (ii) control of . Rapamycin-induced GSK3␤ nuclear localization is Ran-dependent. RPE cells were transfected with plasmids encoding HA-tagged WT, T24N, or G19V Ran and then treated with 1 M rapamycin for 1 h, followed by detection of endogenous GSK3␤ and exogenous HA-tagged Ran proteins. Shown (top) are micrographs obtained by wide-field epifluorescence microscopy representative of three independent experiments. Scale bar, 20 m. Also shown for each condition as GSK3␤ overlay are sample cellular and nuclear outlines and a box corresponding to a magnified image of a single cell. Also shown (bottom) is the mean GSK3␤ nuclear localization index Ϯ S.E. (error bars) (n ϭ 3, Ͼ30 cells/condition/experiment); *, p Ͻ 0.05 relative to control conditions (no rapamycin treatment).

mTORC1 regulates GSK3␤
GSK3␤ nuclear localization by mTORC1 may require lysosomal membrane traffic. To determine whether a pool of GSK3␤ is indeed localized to lysosomes concomitantly to GSK3␤ recruitment to other endomembrane compartments, we systematically examined the localization of endogenous GSK3␤ relative to APPL1 and EEA1 early endosomes and to lysosomes demarked by LAMP1. We observed punctate distribution of endogenous GSK3␤ within the cytoplasm, with some visible overlap with each of APPL1, EEA1, and LAMP1 (Fig. 6, A-C, left panels). To determine whether the overlap observed between GSK3␤ and each marker was specific, we used quantification by Manders' coefficient to compare overlap between real pairs of image channels, as well as between pairs of images with scrambled channel spatial position. This revealed specific GSK3␤ recruitment to each membrane compartment (Fig. 6, A-C). We performed a similar colocalization analysis of the image data using Pearson's coefficient and obtained similar results (Fig. S4B). This indicates that GSK3␤ indeed exhibits partial yet specific localization to several distinct endomembrane compartments, including APPL1 and EEA1 early endosomes and late endosomes/lysosomes demarked by LAMP1.
To further examine how GSK3␤ may localize to lysosomes, we employed structured illumination microscopy (SIM). Using this method, we were able to resolve the limiting membrane of lysosomes demarked by LAMP1 fluorescence staining (Fig.  6D). Importantly, GSK3␤ fluorescence staining was readily observed in punctate structures, some of which were associated with lysosomes. Using automated segmentation of the lysosomes in SIM images, GSK3␤ puncta were preferentially detected on the limiting membrane versus the interior of lysosomes (Fig. 6D). These results indicate that a subset of GSK3␤ in the cytoplasm exhibits association with the lysosome, either restricted to subdomains of the lysosomal surface (58) or in structures associated with the lysosome, such as within membrane contact sites (59).

Control of GSK3␤ nuclear localization and c-Myc levels requires normal lysosomal membrane traffic
Given the localization of mTORC1 (57) and partial localization of GSK3␤ (Fig. 6, C and D) to or near the lysosome, we next sought to determine the role of late endosome/lysosome membrane traffic in mTORC1-dependent control of GSK3␤ nuclear localization. To do so, we used a Rab7 mutant that is constitutively GDP-bound (T22N), which disrupts membrane traffic at the late endosome/lysosome (60). Cells expressing Rab7 T22N exhibited a significant increase in nuclear GSK3␤, even in the absence of rapamycin treatment, compared with cells expressing Rab7 WT (Fig. 7A). Furthermore, cells expressing Rab7 T22N exhibited a depletion of GSK3␤ from lysosomes, observed by overlap of endogenous GSK3␤ and LAMP1, quan-tified by Manders' coefficient (Fig. S4D). In contrast to the nuclear accumulation of GSK3␤ in cells expressing Rab7 T22N, silencing of APPL1 (Fig. S4C) to disrupt early endosome membrane traffic did not impact GSK3␤ nuclear localization (Fig.  7B). These results indicate that membrane traffic at the late endosome/lysosome may be important to organize mTORC1 signals leading to control of GSK3␤ nuclear localization.
To determine the consequence of Rab7-dependent control of GSK3␤ nuclear localization, we examined the effect of expression of Rab7 T22N on GSK3␤-dependent c-Myc levels. Cells expressing Rab7 T22N exhibited a stark reduction in c-Myc levels relative to cells expressing Rab7 WT (Fig. 7C). Importantly, treatment of cells expressing Rab7 T22N with the GSK3␤ inhibitor CHIR99021 restored c-Myc levels to that observed in cells expressing Rab7 WT (Fig. 7C). Taken together, these results indicate that control of GSK3␤ nuclear localization requires Rab7-dependent late endosome/lysosomal membrane traffic, reflecting perhaps the role of lysosomes as platforms for mTORC1 signaling required to negatively regulate GSK3␤ nuclear translocation.

Discussion
We identified that the nuclear localization of GSK3␤ is regulated by mTORC1, such that conditions that reduce mTORC1 activity result in increased nuclear localization of GSK3␤ and increased GSK3␤-dependent degradation of nuclear substrates, such as c-Myc and Snail (Fig. 8). Furthermore, GSK3␤ was partly but specifically localized to the surface of late endosomes/lysosomes, and perturbation of membrane traffic at the late endosomes/lysosomes disrupted GSK3␤ nucleocytoplasmic shuttling and regulation of c-Myc levels.

Localization of GSK3␤ to multiple membrane compartments within the cytoplasm
Separate studies have reported that GSK3␤ may localize to a number of distinct cellular compartments, including endomembranes, mitochondria, and the nucleus (reviewed by Beurel et al. (1)). By a systematic, unbiased approach, we find that endogenous GSK3␤ localizes to several distinct endomembrane compartments, including APPL1 endosomes, EEA1-positive early endosomes, and LAMP1-positive late endosomes/ lysosomes (Fig. 6). In each case, the overlap of GSK3␤ immunofluorescence signal and that of each compartment marker is clearly limited and partial, with substantial proportions of each signal not exhibiting overlap (Fig. 6, A-C). However, systematic and unbiased analysis of colocalization performed by Manders' (Fig. 6, A-C) or Pearson's ( Fig. S4B) coefficient analysis indicates that GSK3␤ overlap with each compartment is specific and nonrandom. Moreover, the specific recruitment of GSK3␤ to the limiting membrane of

mTORC1 regulates GSK3␤
LAMP1-positive late endosomes/lysosomes is supported by images obtained by SIM and unbiased analysis of the localization of GSK3␤ within the lysosome (Fig. 6D), as well as by the observation that perturbation of late endosome/lysosome membrane traffic by expression of a dominant interfering mutant of Rab7 abolishes the overlap of GSK3␤ with LAMP1 signals (Fig. S4D). Our observations are thus consistent with the notion that GSK3␤ is localized to a number of distinct cellular compartments, with a minor pool that in some cases is Ͻ10% of total cellular GSK3␤, recruited to each such compartment at steady state.
Our observations are also consistent with previous studies showing GSK3␤ localization to APPL1 endosomes (27). APPL1 recruitment to a subset of internalized membranes formed by clathrin-mediated endocytosis precedes the acquisition of markers of the EEA1 early endosome (61). This pool of GSK3␤ within APPL1 endosomes may be specifically targeted by phosphorylation on Ser-9 by Akt, as silencing of APPL1 abolishes Akt-dependent GSK3␤ phosphorylation (27,28). Notably, perturbation of APPL1 by silencing did not impact mTORC1-dependent control of GSK3␤ nuclear localization (Fig. 7B), suggesting that the APPL1-localized pool of GSK3␤ does not directly participate in the regulation of GSK3␤ nuclear localization by mTORC1.
As mTORC1 localizes to the surface of late endosomes and lysosomes, the pool of GSK3␤ on these membranes may be under direct regulation by mTORC1 to control GSK3␤ nucleocytoplasmic shuttling. Indeed, a previous report had observed some overlap of GSK3␤ and the lysosome (29). However, GSK3␤ may also be sequestered within intraluminal vesicles of multivesicular bodies in response to Wnt signaling (62), raising the possibility that the overlap that we observed by spinning disc confocal microscopy between LAMP1-positive structures and GSK3␤ (Fig. 6C) could reflect GSK3␤ within intraluminal vesicles. However, quantification of SIM images suggests that GSK3␤ at lysosomes is preferentially associated with the limiting membrane of these organelles, and not the lumen (Fig. 6D). Moreover, perturbation of Rab7 disrupts the localization of GSK3␤ and LAMP1 (Fig. S4C), yet Rab7 disruption does not impact the sequestration of material into intraluminal vesicles (63). The molecular mechanism(s) by which GSK3␤ is recruited to the surface of lysosomes remains unknown and is beyond the scope of this study. Our results thus add systematic analysis and quantification to indicate that a pool of GSK3␤ is present on the limiting membrane of the lysosome, suggesting that this pool may be subject to regulation by mTORC1, resulting in control of GSK3␤ nuclear localization.

Mechanism of control of GSK3␤ nuclear localization by mTORC1
We found that direct inhibition of any component of the PI3K-Akt-mTORC1 axis or activation of AMPK to trigger mTORC1 inhibition results in an increase in GSK3␤ nuclear localization. Moreover, perturbation of Rab7-dependent membrane traffic also resulted in an increase in GSK3␤ nuclear localization, suggesting that in addition to mTORC1 signals, lysosomal traffic and/or organization is also required to control

. GSK3␤ Ser-9 phosphorylation is not required for GSK3␤ nuclear localization induced by inhibition of PI3K-Akt-mTORC1 signals.
A, RPE cells were treated with either 10 M LY294002, 5 M Akti-1/2, or 1 M rapamycin for 1 h. Shown are representative immunoblots of whole-cell lysates probed with anti-pSer-9 GSK3␤ or anti-total GSK3␤ antibodies. Also shown are mean anti-pSer-9 GSK3␤ levels (normalized to total GSK3␤) Ϯ S.E. (error bars) (n ϭ 3); *, p Ͻ 0.05 relative to that in the control conditions (absence of LY294002, Akti1/2, and rapamycin). B, RPE cells were transfected with plasmids encoding HA-tagged WT or S9A GSK3␤ and then treated with 5 M Akti-1/2 for 1 h, followed by detection of exogenous HA-GSK3␤ proteins. Shown (top) are micrographs obtained by wide-field epifluorescence microscopy representative of three independent experiments. Scale bar, 20 m. Also shown for each condition as HA-GSK3␤ overlay are sample cellular and nuclear outlines and a box corresponding to a magnified image of a single cell. Also shown (bottom) is the mean HA-GSK3␤ nuclear localization index Ϯ S.E. (n ϭ 3, Ͼ30 cells/condition/experiment); *, p Ͻ 0.05 relative to control conditions (absence of Akti1/2 treatment). All Western blotting quantifications shown have been normalized to loading controls.

mTORC1 regulates GSK3␤
GSK3␤ nuclear import. Interestingly, we also observed that inhibition of PI3K-Akt-mTORC1 also increased nuclear localization of GSK3␣. Hence, it is likely that mTORC1 signals similarly gate GSK3␣ and GSK3␤ nuclear localization. GSK3␣ nuclear localization was proposed to be uniquely regulated by a calcium-and calpain-dependent mechanism, dependent on its N terminus (64). However, both GSK3␣ (64) and GSK3␤ (15) were reported to undergo nuclear translocation in response to serum withdrawal, consistent with our observations of a central role for mTORC1 in control of GSK3␣/␤ nuclear localization, which we report here. Taken together, we propose that mTORC1 establishes a form of "molecular licensing" for retention within the cytoplasm for GSK3␣ and GSK3␤, resulting in nuclear exclusion under conditions of elevated mTORC1 activity. This molecular licensing could take the form of a post-translational modification of GSK3␣ or GSK3␤ or of regulation of protein complex formation at specific subcellular locale(s).
GSK3␤ undergoes nucleocytoplasmic shuttling, due to nuclear import in balance with FRAT-1-mediated nuclear export (65). Nuclear import of some (but not all) proteins is controlled by a gradient of GTP-bound and GDP-bound

mTORC1 regulates GSK3␤
Ran that spans the nuclear membrane (66). By expression of mutants of Ran (Fig. 3), we show that the nucleocytoplasmic shuttling of GSK3␤ is Ran-dependent. Nuclear import of GSK3␤ resulting from mTORC1 inhibition by rapamycin was prevented in cells expressing Ran G19V mutant defective in GTP hydrolysis and thus defective in nuclear import. Hence, nuclear import of GSK3␤ regulated by mTORC1 is Ran-dependent.
We examined whether the phosphorylation of Ser-9 on GSK3␤ could control its mTORC1-regulated nuclear localization; however, two observations strongly suggest that this is not the case: (i) inhibition of mTORC1 by rapamycin did not alter Ser-9 phosphorylation of GSK3␤ (Fig. 5A), and (ii) a mutant of GSK3␤ that cannot be phosphorylated at this position (S9A) behaved similarly to WT with respect to mTORC1-dependent nuclear localization (Fig. 5B), similar to a previous report with this GSK3␤ mutant (14). GSK3␤ can also be phosphorylated on a number of other residues, including Tyr-216, which may result from autophosphorylation at the time of GSK3␤ synthesis (1). Further, GSK3␤ can be phosphorylated at Thr-43 (67) and Ser-389 (13) by Erk and p38 MAPK, respectively, each of which lead to reduction in GSK3␤ activity. Notably, using a phos-tag gel electrophoresis approach, a technique that exacerbates the apparent molecular weight increase caused by phosphorylation, we were only able to resolve two bands for GSK3␤ that probably correspond to Ser-9 -phosphorylated and non-Ser-9 -phosphorylated forms (Fig. S4A). It will be interesting to determine in future studies whether and how phosphorylation at sites other than Ser-9 is regulated by mTORC1 to control GSK3␤ nuclear localization.
Other than phosphorylation, other modifications reported for GSK3␤ include citrullination (68), ADP-ribosylation (69), and calpain cleavage (70). Indeed, citrullination of Arg-3 and Arg-5 residues within GSK3␤ is important for nuclear localization (68). However, we observed that mTORC1 controls both GSK3␣ and GSK3␤ nuclear localization, and these two GSK3 paralogs differ at their N terminus within the region of GSK3␤ that undergoes citrullination. Hence, it appears unlikely that mTORC1 controls citrullination of GSK3␤ as a mechanism of control of its nucleocytoplasmic shuttling. Although beyond the scope of this study, it will be interesting to note how future work may resolve whether mTORC1-dependent regulation of post-translational modification of GSK3␤ underlies the regulation of its nuclear localization by mTORC1. mTORC1-dependent control of GSK3␤ nuclear localization may occur as a result of regulation of GSK3␤ interaction with other proteins in various endomembrane compartments. It is worth noting that the vast majority of cytoplasmic, but not nuclear, GSK3␤ is associated with another protein(s) (14). Thus, it is possible that control of GSK3␤ nucleocytoplasmic shuttling involves regulation of protein-protein interactions that serve to occlude the bipartite NLS of GSK3␤ (residues 85-103) (14), thus limiting GSK3␤ nuclear localization when these interactions are present.
We also found that Rab7 is required to retain GSK3␤ in the cytoplasm under conditions when mTORC1 is otherwise active. Importantly, disruption of late endosome/lysosome membrane traffic by perturbations of Rab7 or other proteins does not impact mTORC1 activity (71). This indicates that the ability of mTORC1 to limit the nuclear localization of GSK3␤ requires active traffic to the late endosome/lysosome. This in turn suggests that the protein interactions engaged by GSK3␤ that occlude its NLS and thus limit nuclear localization may occur on the lysosome, consistent with our observed localization of GSK3␤ to the lysosome. Indeed, GSK3␣ and GSK3␤ have nearly identical kinase domains (in which the NLS is found), consistent with the ability of mTORC1 to gate nuclear access for both GSK3 paralogs.
Furthermore, our observations that mTORC1 controls GSK3␤ nuclear localization add to previous reports that GSK3␤ activates mTORC1 signals (42) and suggest the existence of reciprocal regulation of mTORC1 and GSK3␤. Overall, we propose that mTORC1 signals limit the ability of GSK3␤ to localize to the nucleus and that this may result from mTORC1dependent control of GSK3␤ interactions with other proteins in a manner that regulates occlusion of the NLS of GSK3␤ at the lysosome. Figure 8. Model depicting signals and processes regulating GSK3␤ nucleocytoplasmic shuttling. Shown on the left is an mTORC1-active state, in which GSK3␤ is largely excluded from the nucleus. Shown on the right is an mTORC1-inactive or Rab7-perturbed state, in which GSK3␤ accumulates in the nucleus and is thus able to phosphorylate direct substrates, such as c-Myc and Snail. Also shown are the known targets of various inhibitors or other pharmacological agents used in this study that impact mTORC1 signaling. See "Results" and "Discussion" for more details. NPC, nuclear pore complex mTORC1 regulates GSK3␤

Regulation of GSK3␤ nuclear functions by mTORC1
We identified that various metabolic and mitogenic signals gate nuclear access for GSK3␤. This in turn allows for GSK3␤dependent regulation of nuclear substrates in response to mTORC1 signals. Previous studies reported that nuclear and cytoplasmic pools of GSK3␤ have distinct functions, such as nuclear GSK3␤ facilitating stem cell differentiation over selfrenewal (15) or the cytosolic pool of GSK3␤ being sufficient to mediate GSK3␤-dependent cell survival to tumor necrosis factor ␣ apoptotic signals (14).
One of the nuclear substrates of GSK3␤ is c-Myc, a helixloop-helix leucine zipper transcription factor that has a very short half-life (15-30 min) (23,72). As previously reported, nuclear localization of GSK3␤ is required for phosphorylation of c-Myc on Thr-58, resulting in enhanced c-Myc degradation (16). We show that rapamycin treatment, which promotes nuclear localization of GSK3␤, also results in an acute reduction in c-Myc accumulation (Fig. 1), most likely due to c-Myc degradation. A previous report suggested that rapamycin treatment elicits degradation of c-Myc by induction of autophagy, as a result of regulation of AMBRA-dependent dephosphorylation of c-Myc at Ser-62 (55). However, we show that the degradation of c-Myc induced by rapamycin is insensitive to impairment of autophagy induction elicited by siRNA gene silencing of ULK1 (Fig. 4D). Moreover, we find that the rapamycin-induced reduction in c-Myc levels is countered by either chemical inhibition or silencing of GSK3␤ (Fig. 1, A and B). Hence, our results indicate that mTORC1-dependent control of GSK3␤ nuclear localization regulates c-Myc in a manner that does not require induction of autophagy.
Based on the control of GSK3␤ nuclear localization by mTORC1 leading to control of c-Myc, we propose the existence of a metabolic sensing signaling network that links nutrient availability with biomass production and proliferation. Indeed, c-Myc controls the expression of many genes, generally to promote ribosome production, biomass accumulation, and enhanced cellular bioenergetics, such as through mitochondrial biosynthesis (73). Furthermore, c-Myc promotes epithelial-mesenchymal transition (74). Hence, signals activated during nutrient deficiency can impair the anabolic c-Myc-dependent promotion of biomass accumulation via this novel mTORC1-GSK3␤c-Myc signaling axis involving control of GSK3␤ nuclear localization.
GSK3␤ may also regulate other nuclear substrates selectively during conditions of reduced mTORC1 signaling or other states in which GSK3␤ exhibits nuclear localization. Collectively, regulation of other GSK3␤ substrates, such as Snail (leading to degradation (75)) or c-Jun (leading to impaired DNA binding (76)), is consistent with the effect of GSK3␤-dependent degradation of c-Myc: reduced cell cycle progression, impairment of epithelial-mesenchymal transition, and/or reduced biomass accumulation. Whereas examination of mTORC1-dependent regulation of all known GSK3␤ nuclear targets is beyond the scope of this study, it is perhaps tempting to speculate that metabolic and mitogenic signals broadly control the nuclear profile of GSK3␤ functions, coordinating ener-gy-demanding accumulation of biomass, cell cycle progression, and growth with nutrient availability. As cancer cells exhibit heterogeneity of metabolic cues and signals, it is possible that differences in metabolism between cancer cells that result in distinct GSK3␤ nuclear localization profiles may underlie in part the differences in response to drugs targeting GSK3␤ in cancer, although this remains to be examined.
In conclusion, we identified that GSK3␤ nucleocytoplasmic shuttling is controlled by both mitogenic signals, such as PI3K-Akt and metabolic cues including amino acid or ATP availability as a result of mTORC1-dependent control of GSK3␤ nuclear import. In addition, GSK3␤ localized in part to the late endosome/lysosome, and nuclear localization of GSK3␤ was regulated by Rab7, suggesting that membrane traffic at late endosomes and lysosomes impacts signals leading to control of GSK3␤ nuclear localization. Last, we propose that GSK3␤-dependent control of nuclear proteins by mTORC1 occurs by regulation of GSK3␤ nuclear import, linking nutrient availability to control of energy-dependent transcriptional networks.

Plasmid and siRNA transfections
To perform DNA plasmid transfections, Lipofectamine 2000 (Thermo Fisher Scientific) was used according to the manufacturer's instructions and as described previously (79). Briefly, cells were incubated for 4 h with Lipofectamine 2000 reagent and appropriate plasmid in Opti-MEM (Gibco) at a 3:1 ratio. Subsequently, this transfection solution was removed, and cells were incubated in fresh cell growth medium at 37 ºC and 5% CO 2 for 16 -24 h before experimentation.

Whole-cell lysates, subcellular fractionation, and Western blotting
Western blotting using whole-cell lysates was performed as described previously (80). Cells were lysed in Laemmli sample buffer (0.5 M Tris, pH 6.8, glycerol, 5% bromphenol blue, 10% ␤-mercaptoethanol, 10% SDS; BioShop) containing phosphatase and protease mixture (1 mM sodium orthovanadate, 10 nM okadaic acid, and 20 nM protease inhibitor, all from BioShop). Cell lysates were then heated to 65 ºC for 15 min and then passed through with a 27.5-gauge syringe. Proteins within whole-cell lysates were resolved by glycine-Tris SDS-PAGE and then transferred onto a polyvinylidene fluoride membrane, which was then incubated with a solution containing specific primary antibodies. Western blotting signal intensity detection corresponding to either phosphorylated proteins (e.g. pGSK3␤ Ser-9) or total proteins (e.g. GSK3␤) and the respective loading controls (e.g. actin) was obtained by signal integration in an area corresponding to the specific lane and band for each condition. The measurement of phosphorylation of a specific protein was obtained by normalization of the signal intensity of a phosphorylated form of a protein to that of its loading control signal and then normalization to the signal intensity similarly obtained for the corresponding total protein.
To examine phosphorylation of proteins for which no specific antibodies were available, we used the phos-tag gel system, which results in exaggeration of differences in the apparent molecular weight of phosphorylated forms of specific proteins (56). The phos-tag reagent was obtained from Wako (Osaka, Japan) and was used in conjugation with SDS-PAGE polymerization as per the manufacturer's instructions. After SDS-PAGE was completed, gel was submerged in MnCl 2 for chelation of remaining phos-tag moieties. Subsequently, protein intensity detection, measurement, and processing were identical to steps mentioned above. Subcellular fractionation to isolate nuclear and cytoplasmic fractions was done using a nuclear extraction kit (catalog no. ab219177, Abcam), as per the manufacturer's instructions.

Immunofluorescence staining
Cells grown on glass coverslips were first subjected to fixation using cold methanol, blocked in 5% BSA (BioShop), and then stained with specific primary antibodies, followed by appropriate fluorophore-conjugated secondary antibody and counterstained with DAPI. Last, cells were then mounted on glass slides in fluorescence mounting medium (DAKO, Carpinteria, CA).

Fluorescence microscopy
Wide-field epifluorescence was performed on an Olympus IX83 inverted microscope with a ϫ100 objective, coupled to a Hamamatsu ORCA-Flash4.0 digital camera (Olympus Canada, Richmond Hill, Canada). Spinning disk confocal microscopy was performed using Quorum (Guelph, Canada) Diskovery combination total internal reflection fluorescence and spinning-disc confocal microscope, operating in spinning disc confocal mode. This instrument is composed of a Leica DMi8 microscope equipped with a ϫ63/1.49 numerical aperture objective with a ϫ1.8 camera relay (total magnification ϫ108). Imaging was done using 488-, 561-, and 637-nm laser illumination and 527/30, 630/75, and 700/75 emission filters and acquired using a Zyla 4.2Plus sCMOS camera (Hamamatsu).
SIM was performed using a Zeiss Elyra PS.1 superresolution inverted microscope, as described previously (81). Samples were imaged at an effective magnification of ϫ101 (ϫ63 objective ϩ ϫ1.6 Optovar tube lens) on an oil immersion objective. Typically, 25-35 slices of 0.110 m were captured for each field of view for an imaging volume of ϳ2.75-3.85 m. 488-, 561-, and 643-nm laser lines were directed into the microscope optical train via a multimode fiber coupler. The lasers were passed through a diffraction grating, and a series of diffraction orders (Ϫ1, 0, ϩ1) were projected onto the back focal plane of the objective. These wave fronts were collimated in the objective to create a three-dimensional sinusoidal illumination pattern on the sample. The diffraction grating was then rotated and translated throughout the acquisition to create patterned offset images containing encoded high spatial frequency information. Five lateral positions were acquired at each of five (72°) diffraction grating rotations for a total of 25 raw images per z-plane. SIM imaging with all lasers was carried out at exposures varying mTORC1 regulates GSK3␤ from 50 to 100 ms, with laser power varying between 3 and 10% (6 -20 milliwatts at the output) and a gain level of 60 -80. Imaging parameters were adjusted iteratively to achieve the best possible equalization of pixel intensity dynamic range across channels.
Raw SIM image stacks were processed in Zen under the Structured Illumination toolbar. A series of parameters were set to generate an optical transfer function used for three-dimensional reconstruction. The noise filter for Wiener deconvolution was set to a value of 1.0 ϫ 10 Ϫ4 to maximize the recovery of high spatial frequency information while minimizing illumination pattern artifacts. The maximum isotropy option was left unselected to recover all available frequency information at exactly the 72°rotation angles. Negative values arising as an artifact of the Wiener filter were clipped to zero using the Baseline Cut option. Processed SIM images were then aligned via an affine transformation matrix of predefined values obtained using 100-nm multicolor Tetraspeck fluorescent microspheres (Thermo Fisher Scientific).

Fluorescence microscopy image analysis
Total cellular signal intensity of specific proteins and GSK3␤ nuclear localization index were measured using ImageJ software (National Institutes of Health, Bethesda, MD). For total cellular measurements of specific protein signal, a region of interest corresponding to the cell outline, identified manually, was used to determine raw mean cellular fluorescence intensity. Final cellular signal intensity was obtained by subtracting background fluorescence (similarly obtained from a region on the coverslip with no cells) from the raw mean cellular fluorescence intensity, as described previously (52).
To determine GSK3␤ nuclear localization index, background-subtracted mean fluorescence intensity of regions of interest within the nucleus and cytoplasm were obtained. The GSK3␤ nuclear localization index is the ratio of these nuclear/ cytosolic intensities. Each measurement was performed in at least three independent experiments, with Ͼ30 cells/condition/experiment. Colocalization analysis was performed by determination of Manders' or Pearson's coefficients, as indicated, using the Just Another Colocalization Plugin (82) within ImageJ, as described previously (79).
Quantification of GSK3␤ localization in LAMP1-positive membrane structures versus lysosome interior was performed (Fig. 6D) on two-color three-dimensional registered SIM images, which were split into single channel images consisting of GSK3␤ (red) or LAMP1 (green). To enhance the edge detection of LAMP1-associated membrane structures, a differenceof-Gaussians operation was performed on the LAMP1 images (83) ( 1 ϭ 1 pixel, 2 ϭ 3 pixels). The resulting bandpass-filtered image was used to trace a selection around the membrane by identifying edges via pixel intensity line profiles in ImageJ. The resulting selection was then superimposed onto the original, unfiltered image, and a binary image mask was generated encompassing only the selected pixels identified as membrane-associated LAMP1. A binary image of the corresponding GSK3␤ channel was created by thresholding via the autothreshold operation in ImageJ. The minimum cross entropy thresholding method (84) was selected, as it minimized the unnecessary removal of data (false negatives) and the inclusion of spurious intensity values (false positives). The area of each feature (GSK3␤ puncta) in the image was measured and recorded. Subsequently, an image subtraction operation was performed by subtracting pixels corresponding to LAMP1-positive membrane structures from the corresponding GSK3␤ channel. The area of the resulting GSK3␤ features was measured once again after the subtraction. Features for which a decrease in area was calculated were found to have pixels co-localizing with membrane-positive LAMP1 and thus were assigned as membrane-associated GSK3␤. Features for which the area remained constant were deemed to be associated with either the cytosol (outside the LAMP1 membrane selection) or the lysosomal interior (inside the LAMP1 membrane selection).

Statistical analysis
Statistical analysis was performed as described previously (79). Measurement of samples involving two experimental conditions (Figs. 4B and 6 and Figs. S2, S3A, and S4B) were analyzed by Student's t test, with p Ͻ 0.05 as a threshold for statistically significant difference between conditions. Measurements of samples involving one experimental parameter and more than two conditions (Figs. 1, 2 (B-D), 4 (A and C), and 5A and Fig. S1 (D and E)) were analyzed by one-way analysis of variance, followed by Bonferonni post-test to compare differences between conditions, with p Ͻ 0.05 as a threshold for statistically significant difference between conditions. Measurements of samples involving two experimental parameters (Figs. 3, 4D, 5B, and 7 and Figs. S3B and S4D) were analyzed by two-way analysis of variance, followed by Bonferonni post-test to compare differences between conditions, with p Ͻ 0.05 as a threshold for statistically significant difference between conditions.