Aldolase is a sensor for both low and high glucose, linking to AMPK and mTORC1

Dear Editor, As well as glucose being the major carbon nutrient for most cells, its availability also acts as a gate-keeper exerting a switch between anabolic and catabolic metabolism, with the protein kinases mTORC1 and AMP-activated protein kinase (AMPK) being the two master controllers. In low glucose, AMPK is activated and phosphorylates a wide range of downstream targets to maintain energy homeostasis, by switching on catabolic pathways while switching off ATP-consuming processes. In high glucose, mTORC1 is activated and shifts the metabolic program of the cell towards anabolic metabolism. It is already known that low glucose, and hence low levels of the glycolytic intermediate fructose-1,6-bisphosphate (FBP), can be sensed by aldolase to trigger AMPK activation via the lysosomal pathway. This occurs independently of increases in AMP and ADP, the classical activators of AMPK. Although it is clear that mTORC1 is also regulated by glucose availability at multiple levels, the underlying mechanisms have not been fully elucidated. To maintain its activity, mTORC1 has to be located on the surface of the lysosome, where it is activated by binding to two classes of lysosome-localized small G proteins, i.e., Rheb and Rags. It has been shown that the mTORC1:Rheb interaction is regulated by GAPDH: when unoccupied by glyceraldehyde-3-phosphate (G3P), a glycolytic metabolite downstream of FBP, GAPDH inhibits the ability of Rheb to activate mTORC1. Rags are regulated by the Ragulator complex, which is associated with the vacuolar ATPase (v-ATPase), a multisubunit proton pump that hydrolyses ATP to provide the energy to acidify the lumen of the lysosome. In high glucose, the Ragulator converts RagA or RagB to their active GTP-bound forms, triggering translocation of mTORC1 to the lysosome. In low glucose, v-ATPase activity is inhibited, which in turn inhibits the Ragulator. It is important to note that an active v-ATPase is required to maintain the Ragulator activity, thus allowing the Rags to activate mTORC1. However, how glucose is sensed and relayed to the RAGs for mTORC1 activation has remained elusive. We have previously shown that the lack of occupancy of aldolase by FBP in low glucose triggers AMPK activation, which prompted us to examine whether aldolase is also involved in the activation of mTORC1 in high glucose. However, knocking down all aldolases (ALDOA/ALDOB/ALDOC) in MEFs directly led to a strong inhibition of the v-ATPase, as evidenced by a decreased signal from LysoSensor Green DND-189 dye (Supplementary information, Fig. S1a), indicating that lysosomal pH was raised. This is consistent with previous findings that aldolase, as well as being a glycolytic enzyme, is also an integral component of the vATPase complex that is required for activity of the latter. The intrinsic requirement of aldolase for the integrity of the v-ATPase therefore precludes the use of ALDO knockdown or knockout approaches to study regulation of mTORC1 by aldolase. We utilized instead the D34S mutant of ALDOA, a mutation that does not significantly affect initial Schiff base formation between FBP and K230 of aldolase, but does block the carbon–carbon cleavage that converts FBP to DHAP and G3P, which is mediated by D34 (Fig. 1a). This mutant therefore still binds FBP even in low glucose, so that its expression mimics a high glucose state. We indeed found that AMPK activation, as well as v-ATPase inhibition (Supplementary information, Fig. S1b) was blocked in these cells, regardless of the presence or absence of glucose in the medium. By contrast, the K230A mutant fails to form a Schiff base with FBP, thus preventing FBP binding. Expression of this mutant therefore mimics a low glucose state and leads to constitutive activation of AMPK. Expression of the D34S mutant in MEFs maintained the activity of mTORC1 even in low glucose, while expression of K230A caused inhibition of mTORC1 even in high glucose (Fig. 1b); similar results were obtained in HEK293T cells (Fig. 1c). Note that activated AMPK can inhibit mTORC1 by directly phosphorylating its Raptor subunit, or TSC2, an upstream inhibitor of Rheb, or ULK1 to inhibit Rags. Importantly, to exclude the influence of AMPK, we also used AMPK-α1/2 DKO (doubleknockout) MEFs and AMPK-α1/2 DKD (double-knockdown) HEK293T cells; the results confirmed that FBP-binding by aldolase is able to control mTORC1 independently of AMPK (Fig. 1b, c). We next examined whether the lysosomal localization of mTORC1 could be regulated by aldolase, and found that expression of the D34S aldolase mutant retained the lysosomal localization of mTORC1 in low glucose, while expression of the K230A mutant caused dissociation from the lysosome even in high glucose (Fig. 1d, e; Supplementary information, Fig. S2a, b). Similar results were obtained in AMPK-α DKO/DKD cells (Fig. 1d, e; Supplementary information, Fig. S2a, b). We have previously shown that the shortage of glucose (or more precisely FBP, after sensing by aldolase) is relayed to the v-ATPase by inhibiting TRPV channels acting upstream of it. Inhibition of TRPVs would also prevent high glucose from maintaining mTORC1 activity. Indeed, TRPV inhibitors AMG-9810 and BCTC, mimicking a low glucose state, impaired the activity as well as the lysosomal localization of mTORC1 in high glucose, in both WT and AMPKα1/2-DKO cells (Fig. 1f, h; Supplementary information, Fig. S3a). Conversely, activation of TRPV channels by agonists such as capsaicin (for TRPV1) or GSK101 and RN1747 (for TRPV4) restored the activation of mTORC1 in low glucose (Fig. 1g, h; Supplementary information, Fig. S3a). In conclusion, we propose that aldolase acts as a sensor for high glucose signalling to mTORC1. It has been suggested that inactivation of v-ATPase by inactive TRPV channels in low glucose leads to conformational changes of the v-ATPase-Ragulator complex, providing a platform for AXIN docking, where AXIN in turns undergoes conformational changes for interacting with lysosomally localized AMPK. The conformational change of AXIN, an intrinsically disordered scaffold protein, is supported by the results from the FRET–FLIM experiment (Fig. 1i), which showed an increased fluorescence lifetime of the donor (GFP fused to the N-terminus of AXIN), as a result of decreased FRET between Nand C-termini of AXIN. Indeed, Nterminally truncated AXIN-NT, which lacks intramolecular autoinhibition provided by the C-terminal domain, exhibits higher affinity towards the v-ATPase and Ragulator than full-length

Wild-type and AMPKα-DKO MEFs (a) or AMPKα-DKD HEK293T cells (b) expressing wild-type ALDOA, the D34S or K230A mutants were incubated in DMEM with 8 mM glucose (Nor), or starved for glucose (GS) for 2 h. mTOR and the lysosome marker LAMP2 (a) or LAMP1 (b) were immunostained. Representative images are shown, and the areas defined by dashed boxes are enlarged as insets on the right. After superimposition of the red and green images, the yellow colour indicates overlap of the two proteins. See statistical analysis in Fig. 1d and 1e, respectively.
Experiments in this figure were performed at least twice. a Wild-type and AMPKα-DKO MEFs were treated with 5 μM AMG-9810 or 10 μM BCTC for 30 min in DMEM medium containing 8 mM glucose, or glucose starved for 2 h, followed by addition of 50 nM GSK101, 100 nM capsaicin, or 0.7 μM RN1747 for another 15 min. mTOR and LAMP2 were immunostained and representative images are shown. See statistical analysis in Fig. 1h.
b Wild-type and AMPKα-DKO MEFs with AXIN knockdown were infected with lentiviruses expressing HA-tagged AXIN and AXIN-NT. mTOR and LAMP2 were stained and representative images are shown. See statistical analysis in Fig. 1k.
Experiments in this figure were performed at least twice.  a pHILIC columns can be applied to measure FBP levels in MEFs, but not in HEK293T cells. MEFs were incubated in DMEM medium containing glucose at the indicated concentrations for 2 h, followed by determining intracellular FBP levels via pHILIC column chromatography coupled with mass spectrometry. Independent results from the Hardie (left) and Lin (right) laboratories are shown.
b Glucose starvation leads to a large decrease of intracellular FBP levels in HEK293T cells measured by CE-MS. FBP levels were determined from HEK293T cells incubated in DMEM containing 25 mM glucose, or in glucose-free DMEM for 2 h. Values are presented as mean ± s.d., n = 3-4 for each treatment.
c Certain FBP isomers in HEK293T cells that behave similarly on the pHILIC column overlap the peak of FBP, because FBP should be entirely derived from glucose when glucose is the unique precursor for glycolysis. HEK293T cells were glucose starved for 2 h. The medium was then added with 25 mM [U-13 C]glucose. After another 15 min of incubation, the labelled and unlabelled FBP levels in cells were measured by LC-MS using a pHILIC column. Data were plotted as mean ± s.d., n = 4. Data from CE-MS were plotted from the results shown in a previously study 1 .
d The pHILIC column fails to separate FBP from its isomer, fructose-2, 6-bisphosphate (F-2,6-BP). FBP and F-2,6-BP standards were subjected to chromatography on pHILIC column, followed by determining their MS/MS fragmentation patterns on a triple-Q mass spectrometer (SCIEX, QTRAP 5500). The spectrograms of an MS/MS fragment with m/z 97 of each isomer were shown, indicating the same retention time.
e The D34S mutant of aldolase is able to bind FBP as well as DHAP in low glucose. HEK-293T cells expressing HA-tagged ALDOA-D34S mutant, or ALDOA-K230A mutant as a control, were glucose starved for 2 h and lysed with buffer containing 200 mM NaBH 4 to trap Schiff-base intermediates. Aldolases were then immunoprecipitated. The potential phosphoglucitolylated (six-carbon) and a phosphoglycerolyated (threecarbon) K230 residue were determined by mass spectrometry. Typical spectrograms of phosphoglucitolylated and phosphoglycerolyated K230 residues (left), and their ratios on aldolase purified from starved HEK293T cells are shown (right).
Experiments in this figure were performed at least twice.

Supplementary information, Fig. S5 A simplified model depicting current understandings on the roles of glycolytic intermediary metabolites and their respective enzymes in controlling mTORC1 activity.
In high glucose, glucose-6-phosphate (G6P) converted by hexokinases (HKs), prevents HKII from binding and inhibiting mTORC1 2 . F-2,6-BP generated by phosphofructokinase 2 (PFK2), binds to and activates PFK1, promoting the lysosomal translocation of mTORC1 3 . FBP, perhaps also DHAP, binds to aldolase and maintains the activity of v-ATPase for mTORC1 activation. G3P binds GAPDH and prevents it from interfering with mTORC1-Rheb interaction 4 . These glucose metabolites and glycolytic enzymes thus regulate mTORC1 at multiple levels depending on glucose availability.

Methods
Plasmids. Point mutations of ALDOA and AXIN were performed by PCR-based sitedirected mutagenesis using PrimeSTAR HS polymerase (Takara). Expression plasmids for various proteins were constructed in pBOBI vector for lentivirus packaging (stably expression). PCR products were verified by sequencing (Invitrogen, China). The lentivirus-based vector pLL3.7 was used for expression of siRNA in HEK293T cells  interfaced with a UPLC system (Waters, ACQUITY UPLC system). Some 2 μl of each sample was loaded onto a SeQuant ZIC-pHILIC column (5 μm, 100 X 2.1 mm, Merck) equipped with a 5 mm pre-guard column to prevent from clogging, in a column oven at 40 °C. Mobile phase buffer A was 15 mM ammonium acetate in water (pH adjusted to 9.7 with ammonium hydroxide), and mobile phase buffer B 90% acetonitrile (in water).
Before analysis, the column was equilibrated with 95% buffer B for 10 min at a flow rate 0.2 ml/min. The gradient was: 95% B for 2 min, then to 45% B within 13 min (linear gradient), then maintained for 3 min, then to 95% B directly, and maintained for 3 min. The flow rate was 0.2 ml/min. One blank was run between each sample to eliminate carry-over. The QTRAP mass spectrometer using an Turbo V ion source. The  proceeded. Similar procedures were followed when Student's t test was performed. No samples or animals were excluded from the analysis. Tests were performed with Graphpad Prism 6, and P < 0.05 was considered statistically significant.