SUMOylation of AMPKα1 by PIAS4 specifically regulates mTORC1 signalling

AMP-activated protein kinase (AMPK) inhibits several anabolic pathways such as fatty acid and protein synthesis, and identification of AMPK substrate specificity would be useful to understand its role in particular cellular processes and develop strategies to modulate AMPK activity in a substrate-specific manner. Here we show that SUMOylation of AMPKα1 attenuates AMPK activation specifically towards mTORC1 signalling. SUMOylation is also important for rapid inactivation of AMPK, to allow prompt restoration of mTORC1 signalling. PIAS4 and its SUMO E3 ligase activity are specifically required for the AMPKα1 SUMOylation and the inhibition of AMPKα1 activity towards mTORC1 signalling. The activity of a SUMOylation-deficient AMPKα1 mutant is higher than the wild type towards mTORC1 signalling when reconstituted in AMPKα-deficient cells. PIAS4 depletion reduced growth of breast cancer cells, specifically when combined with direct AMPK activator A769662, suggesting that inhibiting AMPKα1 SUMOylation can be explored to modulate AMPK activation and thereby suppress cancer cell growth.

A MP-activated protein kinase (AMPK) is an evolutionarily conserved energy rheostat sensing lowered energy levels and adjusting multiple metabolic pathways to mount an appropriate reaction 1 . In general, AMPK activation stimulates ATP-producing catabolic pathways and inhibits ATP-consuming anabolic reactions by direct phosphorylation of downstream targets. Thereby, AMPK inhibits several cellular processes also important for tumour development such as fatty acid and protein synthesis 2 , and several AMPK activators including the 5-aminoimidazole-4carboxamide-1-b-D-ribofuranoside (AICAR) 3,4 , metformin 5 and A769662 (ref. 6) have been reported to inhibit cancer cell growth.
AMPK inhibits fatty acid synthesis by phosphorylating CoA carboxylase (ACC) 7 and protein synthesis through suppression of mammalian target of rapamycin complex 1 (mTORC1) 8 . Inhibition of mTORC1 is mediated through an activating phosphorylation of tuberous sclerosis 2 (TSC2) on S1345 (ref. 9) and inhibitory phosphorylation of the regulatory-associated protein of mTOR (Raptor) on S722 and S792 (ref. 10). Although AMPK phosphorylates numerous substrates and regulates many cellular processes 2 , regulation of AMPK activity towards any one of these substrates has not been described. AMPK has also been reported to shuttle between the cytoplasm and the nucleus 11,12 , providing a potential mechanism for substrate selectivity.
AMPK functions as a heterotrimeric complex comprising a catalytic (AMPKa) and two regulatory subunits (AMPKb and AMPKg) 1 . In mammalian cells, all three subunits are encoded by more than one gene: AMPKa1 and a2; AMPKb1 and b2; AMPKg1, g2 and g3 (ref. 1). AMPKa1 and AMPKa2 demonstrate some specificity in tissue distribution 13 , subcellular localization 14 and substrate selection 15 .
There are several ways in which mammalian AMPK is regulated. On energy deprivation, AMPK is activated in several stages 1 : increased cellular levels of AMP and ADP bind to the AMPKg subunit, leading to stabilization of T-loop phosphorylation of AMPKa subunit provided AMPKb is myristoylated 16 , and further allosteric activation following additional AMP binding. The mammalian AMPK complex has been reported to be inhibited by mutant p53 (ref. 17), adaptor protein a-SNAP 18 , glycogen synthase kinase 3-mediated phosphorylation of S479 inhibiting T-loop phosphorylation 19 , E3 ubiquitin ligase Wwp1 (ref. 20) or CIDE family protein Cidea 21 , and activated by scaffold protein KSR2 (ref. 22) or p53 targets Sestrin1 and Sestrin2 (ref. 23). With the exception of S479 phosphorylation 19 , the inhibitory mechanisms have not been characterized beyond protein-protein interactions with AMPK.
In budding yeast, the AMPK orthologue SNF1 is regulated by several posttranslational modifications. Acetylation of Sip2 (b-subunit) inhibits SNF1 kinase activity and prolongs lifespan 24 . SUMOylation of Snf1 (a-subunit) on a residue not conserved in mammalian AMPKa inhibits the kinase by internal SUMOinteracting motif interaction and by targeting Snf1 for degradation 25 , which possibly involves Snf1 ubiquitination 26 . Mammalian AMPKa has been reported to be ubiquitinated 27 and targeted for degradation by ubiquitination in some cancers overexpressing the MAGE-A3/6-TRIM28 ubiquitin ligase 28 , whereas other posttranslational modifications of AMPKa have not been identified.
Here we present evidence that AMPK activation induces SUMOylation of its catalytic subunit AMPKa1. SUMO E3 ligase PIAS4 catalyses the SUMOylation of AMPKa1, which inhibits AMPK activity specifically towards mTORC1 signalling. Our results therefore uncovered a novel regulatory mechanism by which AMPKa1-mTORC1 signalling is specifically modulated.

Results
PIAS4 interacts with AMPK and modulates mTORC1 signalling. To identify novel interactors of AMPK, we performed yeast two-hybrid screens using human AMPKa1 and AMPKa2 GAL4-DBD fusion proteins as baits. The screens identified both well-characterized AMPKa interactors such as AMPKg1, TRIP6 (ref. 29) and PPP1R12C 30 , as well as putative novel interactors (Supplementary Table 1) including SUMO E3 ligase PIAS3. Owing to the significant sequence homology of the four mammalian PIAS proteins (PIAS1, PIASx/PIAS2, PIAS3 and PIAS4/PIASy) 31 , all of these were included in subsequent validation. Affinity purification of the glutathione S-transferase (GST)-tagged AMPKa1 or AMPKa2 from HEK293 cells followed by western blotting (WB) indicated that both PIAS3 and PIAS4 associated with AMPKa1 and AMPKa2 ( Supplementary Fig. 1a), whereas PIAS1 or PIASx were not detected. In reciprocal experiments, affinity purification of GST-PIAS3 or GST-PIAS4 demonstrated co-purification of all three endogenous AMPK subunits ( Supplementary Fig. 1b).
To investigate whether PIAS E3 ligases regulate AMPK signalling, we analysed AMPK activation on AICAR 32 treatment following small interfering RNA (siRNA)-mediated downregulation of Pias1-4 in immortalized mouse embryonic fibroblasts (MEFs; Fig. 1a and Supplementary Fig. 2a). In knockdown controls, by WB PIAS4 was detected as several bands and are indicated as brackets. It was also noted that Pias2 and Pias3 knockdown concomitantly reduced PIAS4 levels to some extent following AICAR treatment, consistent with the observation that Pias4 messenger RNA level was reduced in the testis of Pias2 À / À mice 33 . Depletion of Pias1, Pias2 and also Pias3 using pooled siRNAs did not result in any noticeable effects (Fig. 1a). By contrast, depletion of Pias4 enhanced suppression of mTORC1 signalling reflected by enhanced dephosphorylation of mTORC1 substrates p70 S6 kinase (S6K), eIF4E binding protein 1 (4EBP1) and ribosomal protein S6 (S6). The result was verified by quantification of p-S6K/S6K ( Supplementary Fig. 2b) and by using two independent siRNAs against Pias4 ( Supplementary  Fig. 2c). Effects of Pias4 depletion were noted at varying AICAR concentrations: the maximal suppression of mTORC1 in control cells was already reached with 0.5 mM AICAR, whereas with the same concentration in Pias4-depleted cells the ability of AICAR in suppressing mTORC1 was higher and the suppression of mTORC1 was further increased up to 2 mM AICAR ( Supplementary Fig. 2d). The result suggests that a larger pool of AMPKa is available for AICAR activation in Pias4-depleted cells. The enhanced AMPK signalling following Pias4 depletion towards mTORC1 was not associated with changes in AMPK activation as measured by phosphorylation of AMPKa-T172 ( Fig. 1a; p-AMPKa; phosphorylation associated with slightly slower migration of AMPKa in these conditions) or by changes in AMPK signalling towards ACC and Raptor ( Fig. 1a; p-ACC and p-Raptor).
In yeast, the E3 ligase Mms21 is required for glucose-induced inactivation of Snf1 and degradation of sensor/receptor-repressor  (a) Cell lysates from immortalized MEFs transfected with indicated siRNA pools (NT, non-targeting) and 72 h later treated with vehicle ( À ) or 2 mM AICAR ( þ ) for 2 h were analysed by SDS-PAGE and WB using indicated antibodies. Brackets in PIAS4 blot denotes specific signal. (b) Lysates from AMPKa1 þ / À ;a2 þ / þ or AMPKa1 À / À ;a2 À / À immortalized MEFs were analysed directly (Input) or following IP using control IgG or antibodies against AMPKa1 (a-a1) or AMPKa2 (a-a2) and protein G-Sepharose by WB analysis with the antibodies. The bracket of PIAS4 blot denotes specific signals. (c) A scheme of the AICAR addition and withdrawal. Immortalized MEFs were analysed either at indicated times (red arrows) after AICAR treatment (AICAR addition) or at indicated times (small black arrows) following AICAR washout by PBS and changing into media without AICAR (AICAR withdrawal). (d) Upper panels: immortalized MEFs transfected with siNT or siPias4 and 72 h later were subjected for AICAR addition, and withdrawal as depicted in c and analysed as in a. The bracket of PIAS4 blot denotes specific signals.
Lower panels: p-S6K/S6K levels from each time points were quantified (mean±s.e.m., n ¼ 3) and p-S6K inhibition (%) is shown as relative to initial time point (green dotted line: siNT; purple dotted line: siPias4). The speed of p-S6K inhibition after AICAR addition and p-S6K restoration following AICAR withdrawal were calculated from the slopes of least-squares fit to linear regressions of p-S6K inhibition levels versus the first three time points after AICAR addition (0, 0 (SRR) pathway component Mth1 (ref. 25). We therefore investigated whether AMPK inactivation following removal of AICAR is regulated by PIAS4 in mammalian cells by analysing the phosphorylation state of AMPKa, ACC and S6K (Fig. 1d) after indicated times following addition (AICAR addition) and following withdrawal of AICAR (AICAR withdrawal) as depicted in Fig. 1c. Based on the phosphorylation state of AMPKa, ACC and S6K (Fig. 1d), Pias4 depletion led to a significantly faster inhibition of p-S6K following AICAR addition and delayed restoration of p-S6K following AICAR withdrawal (Fig. 1d, AICAR withdrawal), indicating that PIAS4 is essential for both activation and inactivation of AMPK and suggesting conservation in the manner SUMO E3 ligase regulates the inactivation of AMPK/SNF1 kinase.
PIAS4 inhibits AMPKa1 activity towards TSC2. We next investigated whether inhibition of mTORC1 following PIAS4 depletion is due to enhanced kinase activity of AMPKa1 towards TSC2. To this end, we compared the ability of immunoprecipitated AMPKa1 from control (siNT) or PIAS4-depleted (siPias4) MEFs, to phosphorylate a recombinant GST-TSC2 fragment (1,300-1,367) containing the AMPK phosphorylation site S1345 (ref. 9). As expected, no AMPKa1 kinase activity was detected in precipitates from AMPKa1 À / À ;a2 À / À MEFs ( Fig. 3a and Supplementary Fig. 5a). Notably, AMPKa1 activity was significantly higher in AICAR-treated lysates from Pias4-depleted cells ( Fig. 3a and Supplementary Fig. 5a). This result suggests PIAS4 inhibits AMPKa1 activity towards TSC2. The specificity of Pias4 inhibition noted in cells could not be recapitulated in the reconstituted system using the SAMS peptide as a surrogate for ACC 38 (Supplementary Fig. 5b).
To further explore the possible role of SUMOylation in the regulation of mTORC1, we depleted Ubc9, the sole SUMO E2 enzyme 31 , with two independent siRNAs (siUbc9-1 and siUbc9-2) in immortalized MEFs. Depletion of Ubc9 enhanced AICAR-induced suppression of mTORC1 signalling ( Fig. 3d and Supplementary Fig. 5e), suggesting that SUMOylation is involved in inhibiting AMPKa1 activity towards TSC2.
Given the requirement of AMPKa1 in mediating the effect of PIAS4 on mTORC1 signalling (Fig. 2a), we chose to further analyse AMPKa1 SUMOylation in more detail. Knockdown of UBC9 (siUBC9) or PIAS4 (siPIAS4) inhibited SUMOylation of AMPKa1 compared with siNT ( Fig. 4b, lanes NEM þ ). Following nickel affinity purification of 6His-SUMO3, SUMOylated AMPKa1 in PIAS4 untransfected cells was greatly enhanced following AICAR treatment (Fig. 4c, lane AICAR þ ). The results suggest that PIAS4 is required for the SUMOylation of AMPKa1 in cells. We subsequently attempted to reconstitute SUMOylation of AMPKa1 in vitro using SUMO3/E1/E2 and bacterially produced AMPK complex (a1, b1 and g1) and GST-PIAS4. As a positive control for the assay, SUMOylation of p53 was detected in the presence of SUMO3/E1/E2/ATP and its SUMOylation was markedly enhanced by GST-PIAS4 ( Supplementary Fig. 6c, right panels), confirming the previous report 39 . In the same setting, several slower migrating bands were detected with AMPKa1 antibody in the presence of SUMO3/E1/E2/ATP and addition of GST-PIAS4 led to an increase of high-molecular-weight bands ( Supplementary Fig. 6c, left panels, bracket). The results demonstrate that AMPKa1 can be SUMOylated in a PIAS4dependent manner.
The most probably predicted SUMOylation sites on AMPKa1 are lysines in the context of LK 152 PE and FK 280 QD. However, mutating either lysine to arginine did not noticeably affect the SUMOylation of AMPKa1 ( Supplementary Fig. 6d, K152R and K280R). Subsequent mutagenesis and analysis of all remaining lysines in AMPKa1 revealed that substitution of lysine 118 to arginine greatly attenuated the SUMOylation of AMPKa1 (Fig. 4d, K118R), suggesting K118 represents a major SUMOylation site on AMPKa1.
Our earlier results provide evidence that SUMOylation is involved in inhibiting AMPKa1 activity towards TSC2 and thus predict that the SUMOylation-deficient mutant AMPKa1-K118R would be resistant to this inhibition. To test this we reconstituted AMPKa-null MEFs with AMPKa1-WT or AMPKa1-K118R and analysed their ability to suppress mTORC1 signalling following AMPK activation. The results indicate that AMPKa1-K118R reconstituted cells demonstrate an elevated ability to suppress mTORC1 signalling ( Fig. 4e and Supplementary Fig. 6e), whereas the activity towards ACC was comparable to the WT control (Fig. 4e, p-ACC). The results provide further evidence that SUMOylation of AMPKa1 specifically regulates mTORC1 signalling.
We subsequently investigated in which subcellular compartment AMPKa1 SUMOylation occurs using a proximity ligation assay (PLA) 40 in AMPKa-null MEFs reconstituted with WT or SUMOylation-deficient GST-AMPKa1. Although no detectable PLA signal was noted from control MEFs, cells expressing WT AMPKa1 (AMPKa1-WT) demonstrated a strong and predominantly nuclear (83.4%) PLA signal ( Supplementary  Fig. 7b). PLA signal in the AMPKa1-K118R was dramatically reduced to 3% of that in AMPKa1-WT cells consistent with the earlier results, indicating K118 represents a major SUMOylation site. In summary, the localization analyses indicate that SUMOylated AMPKa1 is mostly nuclear, which could be due, for example to reduced nuclear-cytoplasmic shuttling 11,12 .
PIAS4 depletion reduces breast cancer cell proliferation. As PIAS4 depletion was found to potentiate AMPK activity towards mTORC1 (Fig. 1a) and AMPK activation can inhibit cancer cell growth, we next investigated whether PIAS4 depletion affect cancer cell growth in MDA-MB-231 breast cancer cells reported to be sensitive to AMPK activation 4,41 . To this end, we depleted PIAS4 using lentiviral transduction of two independent short hairpin RNAs (Supplementary Fig. 8, shPIAS4-5 or shPIAS4-8) and thereafter activated AMPK using A769662 (ref. 42), which stabilizes the interaction between AMPKa and AMPKb subunit 43  phosphorylation 44 . As expected, administration of A769662 for 12 h led to phosphorylation of ACC and dephosphorylation of mTORC1 substrate S6K (Fig. 5a). In A769662-treated cells, PIAS4 depletion markedly enhanced the dephosphorylation of S6K without altering ACC phosphorylation (Fig. 5a), concordant with observations in MEFs following AICAR treatment (Fig. 1a). Next, we analysed proliferation of control (shNontargeting clone (shNT)) or PIAS4-depleted (shPIAS4-5 or shPIAS-8) MDA-MB-231 cells treated with A769662, rapamycin or both for 72 h. As previously reported, both A769662 (ref. 41) and rapamycin 45 treatment alone modestly suppressed cell growth in control shNT cells. A769662 and rapamycin together had additive effects (Fig. 5b), suggesting that AMPK activation suppress growth partly through inhibition of lipogenesis as noted in glioblastoma cells 46 . Importantly, PIAS4 downregulation specifically potentiated the A769662-induced inhibition of proliferation, whereas no significant changes were noted in vehicle control or A769662 þ rapamycin-treated cells (Fig. 5b).
These results indicate PIAS4 depletion enhances A769662mediated inhibition of proliferation of MDA-MB-231 breast cancer cells and this occurs via mTORC1 inhibition.

Discussion
Our study identifies PIAS4-mediated SUMOylation as a novel and specific regulatory mechanism to attenuate AMPKa1 activity towards mTORC1 following AMPK activation. PIAS4 was also noted to be important for rapid inactivation of AMPK following AICAR withdrawal (Fig. 1d). Although AMPK inactivation and its regulation has not been studied in   25 , suggesting that SUMOylation is involved in both Snf1 and AMPKa inactivation. The regulation of SUMOylation however appears to differ between yeast and mammalian cells, as in yeast SUMOylation occurs during Snf1 inactivation and is associated with non-phosphorylated Snf1 (ref. 25), whereas in mammalian cells SUMOylation of AMPKa1 was low when AMPK was inactive (Fig. 4c, lane AICAR À PIAS4 À ) and induced concomitant with AMPK activation. Another difference between yeast and mammalian SUMOylation is that, in yeast SUMOylation leads to Snf1 degradation 25 , whereas stability of AMPKa1 was not detectably altered following PIAS4 depletion (Fig. 3a) or when comparing WT and the SUMOylation-deficient mutant in reconstituted AMPKaknockout MEFs (Fig. 4e). The availability of MEFs deficient for AMPKa1, AMPKa2 or both allowed robust analysis of isoform specificity. Interestingly, the ability of AMPK to regulate mTORC1 signalling was lost in cells only containing AMPKa2 ( Fig. 2a and Supplementary  Fig. 4a), suggesting that TSC2 phosphorylation in vivo is specifically mediated by AMPKa1 in MEFs, and providing an explanation for the inability of PIAS4 to regulate mTORC1 signalling in these cells. An alternative explanation could be provided by the significantly lower levels of total AMPKa in MEFs only expressing AMPKa2 (Fig. 2a) and different sensitivities of AMPK substrates as proposed by Houde et al. 47 In that case, TSC2 phosphorylation would require significantly higher levels of AMPK activity compared with phosphorylation of ACC, which was still detected in cells with only AMPKa2 following AICAR treatment (Fig. 2a).
Activation of AMPK is associated with nuclear translocation 11,48 , providing a mechanism by which the predominantly nuclear PIAS4 is associated with AMPKa1 and consistent with our observation that SUMOylated AMPKa1 was detected primarily in the nucleus (Supplementary Fig. 7b).
The apparent low (o5%) stoichiometry of SUMOylation of AMPKa1 (Fig. 4a,b) together with the 5-to 20-fold activation of AMPK towards mTORC1 in conditions inhibiting SUMOylation suggests SUMOylation represents a licensing event necessary for a subsequent, more stable modification as described for NEMO/ IKKg 49 and SF-1 (ref. 50).
The nuclear SUMOylation of AMPKa1 and cytoplasmic localization of TSC2 suggests that TSC2 phosphorylation is mediated by AMPKa1 that has shuttled from the nucleus to the cytoplasm following SUMOylation and licensing. Such a requirement could provide the specificity observed between AMPKa1 substrates TSC2 and ACC/Raptor, where AMPKa1 phosphorylation of the latter substrates would not require prior shuttling. Although further investigations are required to dissect these mechanisms, our observation that AMPK activity is regulated in a substrate-specific manner is interesting, considering the various AMPK modifications and interacting proteins. AMPK is physiologically activated by AMP/ADP via three mechanisms: inhibition of a-Thr172 dephosphorylation by both AMP and ADP, promotion of a-Thr172 phosphorylation and allosteric activation of already phosphorylated AMPK by AMP 51 . AMPK can be also activated by A769662 independently of AMP/ADP and AMPKa T-loop phosphorylation 44 through stabilization of the interaction between AMPKa and AMPKb subunits 43 . Here, PIAS4 depletion enhanced the ability of both AICAR and A769662, to suppress mTORC1 signalling without altering AMPKa protein levels or a-Thr172 phosphorylation (Figs 1a, 3a and 4e), suggesting that SUMOylation regulates AMPKa1 activity at the level of allosteric activation, but in a more stable manner based on the maintenance of this regulation following immunoprecipitation (IP). Thus inhibition of SUMOylation could be used to further potentiate AMPK activation following increased AMP/ADP level or the stabilization of AMPKa/AMPKb interaction by A769662.
Enhancement of AMPK activity towards mTORC1 and associated cytostatic effects in MDA-MB-231 breast cancer cells following PIAS4 depletion indicate that inhibition of AMPKa1 SUMOylation could provide a specific way to inhibit cancers with hyperactive mTORC1 signalling.
All siRNAs were purchased either as pools or individual siRNAs from Dharmacon and are listed in Supplementary Table 3.
Real-time quantitative PCR. Total RNAs from cell lysates were isolated using the RNeasy isolation kit (Qiagen) according to manufacturer's instructions and reverse transcribed using Taqman MULV reverse transcriptase and random hexamers (Applied Biosystems). Real-time quantitative PCR was performed with StepOne-Plus RT-PCR System (Applied Biosystems) using SYBR-green RT-PCR reagent (Applied Biosystems) and primers listed in Supplementary Table 4. Relative mRNA levels were calculated relative to mouse or human GAPDH.
Nuclear/cytoplasmic fractionation. Nuclear/cytoplasmic fractionation was performed using NE-PER Nuclear and Cytoplasmic Extraction Reagents (ThermoFisher) and nuclear fractions were extracted using ELB buffer (50 mM HEPES pH 7.4, 150 mM NaCl, 5 mM EDTA, 0.1% NP-40, 1 mM DTT, 2.5 mg ml À 1 aprotinin, 0.5 mM PMSF, 10 mM b-glycerophosphate and 1 mg ml À 1 leupeptin) instead of NER buffer provided by the kit. The fractions were analysed following purification with GST pulldown or analysed directly by WB. The sample size for each condition is 1 (n ¼ 1) within one experiment.
WB analysis. For total protein analysis, cells were lysed using SDS boiling buffer (2.5% SDS, 250 mM Tris pH 6.8, including 50 mM NaF, 10 mM b-glycerophosphate, 0.5 mM DTT, 0.5 mM PMSF) and lysates were boiled at þ 98°C and needled (25-gauge) ten times. Lysates were then cleared by centrifugation and protein concentration was measured by using Bio-Rad DC protein assay (Bio-Rad). Twenty micrograms of protein lysates were used for WB. For WB analysis, 8-12% SDS-PAGE gels were transferred to nitrocellulose membrane and blotted according to the antibody manufacturer's instructions. Immunoblots were developed using SuperSignal West Femto Maximum Sensitivity Substrate (34096, Thermo). Blots were cropped to include at least one marker position. Uncropped blots for all main figures were included in Supplementary Figs 9-14. The sample size for each condition is 1 (n ¼ 1) within one experiment.
Immunofluorescence analysis and quantification. Cells on coverslips were fixed with 4% paraformaldehyde (PFA) for 15 min and then permeabilized using 0.1% Triton X-100 for 5 min. Cells were then blocked with 5% goat serum in PBS for 30 min, incubated with anti-GST antibody (1:500) for 30 min, washed three times with PBS, incubated with Alexa594 secondary antibody for 30 min, washed three times with PBS, labelled with 4,6-diamidino-2-phenylindole (DAPI) and mounted with Immuno-mount (Thermo Scientific).
Stained coverslips were analysed and imaged using Zeiss AxioImager.M2 fluorescent microscope. Transfected cells were identified by thresholding Alexa594 image using ImageJ 1.49a. Total GST-AMPKa1 levels were measured as the integrated density within the identified transfected cells. Nuclei were identified by DAPI stain. Nuclear levels of GST-AMPKa1 were measured as the integrated density within the nuclear region and expressed as a percentage of total GST-AMPKa1 for each identified transfected cells. The mean nuclear levels of GST-AMPKa1-WT and GST-AMPKa1-K118R cells from three equally weighed independent experiments were compared with Student's t-test. The sample size for each condition is 1 (n ¼ 1) within one experiment. protocol. Briefly, cells were fixed with 4% PFA for 15 min and then permeabilized using 0.1% Triton X-100 for 5 min. After blocking with 5% goat serum in PBS for 30 min, cells were labelled with anti-GST (rabbit, 1:500) and anti-SUMO2/3 (mouse, 1:500) for 30 min. After washing with PBS for three times, cells were labelled with PLA probe plus and minus diluted in PBS for 1 h in a pre-heated humidity chamber. Ligation and amplification were performed as detailed by the manufacturer. Slides were mounted with Duolink Mounting medium with DAPI and were imaged using Zeiss AxioPlan2 with Â 10 objectives.
Forty images per coverslip of two independent experiments were analysed using ImageJ 1.49a. Cell nuclei were identified by DAPI stain (area size between 400-3,000 pixels). Regions of interest (ROIs) for PLA signal measurement of individual cells were created by dilating cell nuclei with 20 iterations followed by the watershed algorithm. GST transfected control was used to correct for background PLA signal by using it as a reference to set a threshold for PLA signal (imaged under Alexa 594 channel), where minimal signal is obtained in GST-transfected control. The same threshold was then applied to other samples. PLA signal was measured as arbitrary unit of integrated density within the ROIs. Integrated density for individual cells within GST-transfected control were measured and the maximum value (GSTmax) was used as cutoff, whereby cells with integrated density greater than GSTmax are identified as positive for PLA signal. Fisher's exact test was used for statistical analysis of percentage of positive cells.
PLA signal is also measured within the nuclear region of each cells and percentage of nuclear PLA signal is expressed as integrated density within the nuclear region divided by integrated density within the ROIs multiplied by 100. Student's t-test was used for statistical analysis.
Transfection efficiency was determined from coverslips (from the same transfection of coverslips used for PLA) stained for anti-GST antibody, five images per coverslips from two biological experiments. Cells that were clearly non-transfected were used as background reference for thresholding GST signal to a black and white binary image, of which pixels with intensity value above the threshold value is white and those below is black. A single threshold value is used for all samples. Each cell, represented by ROIs created as described above, is classified as transfected cells when it has more than 500 pixels of white. The appropriateness of such analysis was confirmed manually by the eye. Fisher's exact test was used for statistical analysis of transfection efficiency. The sample size for each condition is 1 (n ¼ 1) within one experiment.
Kinase assays. Cells to be analysed were replenished with fresh DMEM medium with or without 2 mM AICAR for 2 h before lysing in ice-cold IP lysis buffer. Twenty-five micrograms of cleared lysates were incubated with 1 ml of affinity-purified AMPKa1 antibody (H1) at þ 4°C with rotation overnight followed by addition of 20 ml of a 1:1 slurry of Protein G Sepharose prewashed with IP lysis buffer. Beads were washed three times with cold IP lysis buffer and two times with cold HEPES-Brij buffer (50 mM HEPES pH 7.4, 1 mM DTT, 0.02% Brij-35) followed by a kinase assay in 50 mM HEPES pH 7.4, 1 mM DTT, 0.02% Brij-35, 5 mM MgCl 2 , 5 mCi [g-32P] ATP 3,000 Ci mmol À 1 and 2 mg GST-TSC2 (1,300-1,367) or 2.5 mg SAMS peptide in a final reaction volume of 25 ml at þ 30°C for 30 min. For the kinase assay using GST-TSC2 fragment as substrates, reactions were terminated with 25 ml 2 Â Laemmli sample buffer (65.8 mM Tris-HCl pH 6.8, 2.1% SDS, 26.3% (w/v) glycerol, 0.01% bromophenol blue), boiled at þ 98°C for 5 min and subjected to SDS-PAGE and autoradiography. For the kinase assay using SAMS peptide as substrates, reactions were terminated by spotting 10 ml reaction mixture to the centre of P81 phosphocellulose square. P81 squares were then washed two times with 1% phosphoric acid buffer, one time with H 2 O and one time with acetone. Dried P81 squares were transferred to scintillation vial containing 2 ml scintillation cocktail and radioactivity were measured with scintillation counter from three independent experiments. The sample size for each condition is 1 (n ¼ 1) within one experiment. Student's t-test was used for statistical analysis.
SUMOylation assay. Purification of 6His-tagged SUMO under denaturing condition was performed according to the protocol described previously 53 with slight modification. HEK293 cells (10 cm plate) were transfected with cDNAs and 24 h later cells were washed once with 10 ml ice-cold PBS containing 20 mM NEM, and washed cells were scraped off from the plate with another 5 ml ice-cold PBS containing 20 mM NEM. For the Input sample preparation, 500 ml of cell suspension were transferred to 1.5 ml tube and spinned at 1,000g for 2 min. Cell pellet was re-suspended with 100 ml 1 Â Laemmli sample buffer, needled and boiled at þ 98°C for 5 min before loading to SDS-PAGE. To prepare the sample for nickel affinity purification, rest of the cell suspension were centrifuged at 1,000g and re-suspended with 5 ml freshly prepared Lysis buffer (0.1 M Na 2 HPO 4 /NaH 2 PO 4 pH 8.0, 6 M Guanidinium-HCl, 10 mM Tris pH 8.0, 0.05% Tween). Samples were sonicated for 30 s on low power on ice and cleared by centrifugation at 16,000g, þ 4°C for 10 min. Supernatants were then diluted with equal volume of Equilibration buffer (0.1 M Na 2 HPO 4 /NaH 2 PO 4 pH 8.0, 6 M Guanidinium-HCl, 10 mM Tris pH 8.0, 0.05% Tween, 30 mM Imidazole) and incubated overnight with rotation at þ 4°C with 100 ml HisPur Ni-NTA magnetic beads prewashed with Equilibration buffer. Beads were then washed and collected on a magnetic stand three times with Wash buffer 1 (0.1 M Na 2 HPO 4 /NaH 2 PO 4 pH 8.0, 6 M Guanidinium-HCl, 10 mM Tris pH 8.0, 0.05% Tween, 30 mM Imidazole) and three times with Wash buffer 2 (0.1 M Na 2 HPO 4 /NaH 2 PO 4 pH 8.0, 10 mM Tris pH 8.0, 0.05% Tween, 30 mM Imidazole, 8 M urea). His-tagged proteins were finally eluted with 100 ml Elution buffer (200 mM imidazole, 5% SDS, 150 mM Tris-Cl pH 6.7, 30% glycerol, 720 mM b-ME, 0.0025% bromophenol blue) for 20 min at room temperature and 1 min at þ 98°C, and 20 ml protein elutions were resolved by 8% SDS-PAGE gel.
In vitro SUMOylation was performed according to the manual provided by SUMO3 conjugation kit (K-720, Boston Biochem) with a slight modification. Briefly, 9 ml of glutathione elution buffer (10 mM L-Glutathione Reduced, 50 mM Tris pH 8.0, supplemented with 5 mM DTT) with or without GST-PIAS4, 2 ml 10 Â Reaction buffer, 1 ml p53 protein (285 ng) or 0.5 ml AMPK (a1, b1 and g1) protein (500 ng), 2 ml SUMO E1 enzyme, 2 ml SUMO E2 enzyme, 2 ml SUMO3, 2 ml 10 Â Mg 2 þ -ATP Solution were mixed in a 20-ml volume and the reactions were carried out in þ 37°C for 1 h. The reactions were terminated with addition of 5 Â SDS-PAGE sample buffer and 1 ml 1 M DTT, and boiling at 98°C for 5 min. Five-microlitre reactions were resolved by 10% SDS-PAGE gel and blotted with either p53 or AMPKa1 antibody. The sample size for each condition is 1 (n ¼ 1) within one experiment.
Growth inhibition assay in MDA-MB-231 cells. Selected pools of lentiviraltransduced MDA-MB-231 cells were seeded as triplicates (n ¼ 3) (1.5 Â 10 5 cells per well) into 24-well plates. The following day (day 1) cells were treated with vehicle control (dimethyl dulfoxide), A769662 (150 mM), rapamycin (10 mM) or A769662 (150 mM) þ rapamycin (10 mM). Cell proliferation was assessed by comparing number of viable cells on day 2, 3 and 4 with those of day 1 using Trypan blue exclusion and a TC20 cell counter (Bio-Rad) from three independent experiments. Student's t-test was used for statistical analysis. The sample size for each condition is 3 (n ¼ 3) within one experiment.
Quantification of blots and statistics. All experiments have been done at least twice and representative results are shown. Densitometric analysis of WB and 32P radioautography results was done using Image J 1.49a (NIH). Intensity of each band was measured and subtracted with background intensity from adjacent area. Values in the results section were presented as relative intensity versus control. Statistical analyses were performed either by using the analysis of variance followed by Fisher's least significant difference test (GraphPad Prism version 6.0c for Mac OS X) (for Fig. 1d) or by two-sided Student's t-test (for all the other quantifications). Data were presented as mean þ s.e.m. Statistical significance was set at P-values of NS40.05, *o0.05, **o0.01 and ***o0.001.